av1/encoder/encodetxb.c

/*
 * Copyright (c) 2017, 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 "aom_ports/mem.h"
#include "av1/common/scan.h"
#include "av1/common/blockd.h"
#include "av1/common/idct.h"
#include "av1/common/pred_common.h"
#include "av1/encoder/bitstream.h"
#include "av1/encoder/encodeframe.h"
#include "av1/encoder/cost.h"
#include "av1/encoder/encodetxb.h"
#include "av1/encoder/hash.h"
#include "av1/encoder/rdopt.h"
#include "av1/encoder/tokenize.h"

#define TEST_OPTIMIZE_TXB 0

static int hbt_hash_needs_init = 1;
static CRC_CALCULATOR crc_calculator;
static CRC_CALCULATOR crc_calculator2;
static const int HBT_HASH_EOB = 16;  // also the length in opt_qcoeff

typedef struct OptTxbQcoeff {
  uint32_t hbt_hash_match;
  double hits;
  tran_low_t opt_qcoeff[16];
} OptTxbQcoeff;

OptTxbQcoeff hbt_hash_table[65536][16];

typedef struct LevelDownStats {
  int update;
  tran_low_t low_qc;
  tran_low_t low_dqc;
  int64_t dist0;
  int rate;
  int rate_low;
  int64_t dist;
  int64_t dist_low;
  int64_t rd;
  int64_t rd_low;
  int64_t nz_rd;
  int64_t rd_diff;
  int cost_diff;
  int64_t dist_diff;
  int new_eob;
} LevelDownStats;

void av1_alloc_txb_buf(AV1_COMP *cpi) {
#if 0
  AV1_COMMON *cm = &cpi->common;
  int mi_block_size = 1 << MI_SIZE_LOG2;
  // TODO(angiebird): Make sure cm->subsampling_x/y is set correctly, and then
  // use precise buffer size according to cm->subsampling_x/y
  int pixel_stride = mi_block_size * cm->mi_cols;
  int pixel_height = mi_block_size * cm->mi_rows;
  int i;
  for (i = 0; i < MAX_MB_PLANE; ++i) {
    CHECK_MEM_ERROR(
        cm, cpi->tcoeff_buf[i],
        aom_malloc(sizeof(*cpi->tcoeff_buf[i]) * pixel_stride * pixel_height));
  }
#else
  AV1_COMMON *cm = &cpi->common;
  int size = ((cm->mi_rows >> cm->mib_size_log2) + 1) *
             ((cm->mi_cols >> cm->mib_size_log2) + 1);

  av1_free_txb_buf(cpi);
  // TODO(jingning): This should be further reduced.
  CHECK_MEM_ERROR(cm, cpi->coeff_buffer_base,
                  aom_malloc(sizeof(*cpi->coeff_buffer_base) * size));
#endif
}

void av1_free_txb_buf(AV1_COMP *cpi) {
#if 0
  int i;
  for (i = 0; i < MAX_MB_PLANE; ++i) {
    aom_free(cpi->tcoeff_buf[i]);
  }
#else
  aom_free(cpi->coeff_buffer_base);
#endif
}

void av1_set_coeff_buffer(const AV1_COMP *const cpi, MACROBLOCK *const x,
                          int mi_row, int mi_col) {
  int mib_size_log2 = cpi->common.mib_size_log2;
  int stride = (cpi->common.mi_cols >> mib_size_log2) + 1;
  int offset = (mi_row >> mib_size_log2) * stride + (mi_col >> mib_size_log2);
  CB_COEFF_BUFFER *coeff_buf = &cpi->coeff_buffer_base[offset];
  const int txb_offset = x->cb_offset / (TX_SIZE_W_MIN * TX_SIZE_H_MIN);
  for (int plane = 0; plane < MAX_MB_PLANE; ++plane) {
    x->mbmi_ext->tcoeff[plane] = coeff_buf->tcoeff[plane] + x->cb_offset;
    x->mbmi_ext->eobs[plane] = coeff_buf->eobs[plane] + txb_offset;
    x->mbmi_ext->txb_skip_ctx[plane] =
        coeff_buf->txb_skip_ctx[plane] + txb_offset;
    x->mbmi_ext->dc_sign_ctx[plane] =
        coeff_buf->dc_sign_ctx[plane] + txb_offset;
  }
}

static void write_golomb(aom_writer *w, int level) {
  int x = level + 1;
  int i = x;
  int length = 0;

  while (i) {
    i >>= 1;
    ++length;
  }
  assert(length > 0);

  for (i = 0; i < length - 1; ++i) aom_write_bit(w, 0);

  for (i = length - 1; i >= 0; --i) aom_write_bit(w, (x >> i) & 0x01);
}

static INLINE tran_low_t get_lower_coeff(tran_low_t qc) {
  if (qc == 0) {
    return 0;
  }
  return qc > 0 ? qc - 1 : qc + 1;
}

static INLINE tran_low_t qcoeff_to_dqcoeff(tran_low_t qc,
#if CONFIG_NEW_QUANT
                                           const tran_low_t *nq_dq,
#endif  // CONFIG_NEW_QUANT
                                           int dqv, int shift) {
  int sgn = qc < 0 ? -1 : 1;
#if CONFIG_NEW_QUANT
  int dqcoeff = av1_dequant_coeff_nuq(abs(qc), dqv, nq_dq, shift);
  return sgn * dqcoeff;
#endif  // CONFIG_NEW_QUANT

  return sgn * ((abs(qc) * dqv) >> shift);
}

static INLINE int64_t get_coeff_dist(tran_low_t tcoeff, tran_low_t dqcoeff,
                                     int shift) {
#if CONFIG_DAALA_TX
  int depth_shift = (TX_COEFF_DEPTH - 11) * 2;
  int depth_round = depth_shift > 1 ? (1 << (depth_shift - 1)) : 0;
  const int64_t diff = tcoeff - dqcoeff;
  const int64_t error = (diff * diff + depth_round) >> depth_shift;
  (void)shift;
#else
  const int64_t diff = (tcoeff - dqcoeff) * (1 << shift);
  const int64_t error = diff * diff;
#endif
  return error;
}

void av1_update_eob_context(int eob, int seg_eob, TX_SIZE tx_size,
                            TX_TYPE tx_type, PLANE_TYPE plane,
                            FRAME_CONTEXT *ec_ctx, FRAME_COUNTS *counts,
                            uint8_t allow_update_cdf) {
  int eob_extra, dummy;
  const int eob_pt = get_eob_pos_token(eob, &eob_extra);
  const int max_eob_pt = get_eob_pos_token(seg_eob, &dummy);
  TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);

  (void)max_eob_pt;
  const int eob_multi_size = txsize_log2_minus4[tx_size];
  const int eob_multi_ctx = (tx_type_to_class[tx_type] == TX_CLASS_2D) ? 0 : 1;

  switch (eob_multi_size) {
    case 0:
      ++counts->eob_multi16[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf16[plane][eob_multi_ctx], eob_pt - 1, 5);
      break;
    case 1:
      ++counts->eob_multi32[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf32[plane][eob_multi_ctx], eob_pt - 1, 6);
      break;
    case 2:
      ++counts->eob_multi64[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf64[plane][eob_multi_ctx], eob_pt - 1, 7);
      break;
    case 3:
      ++counts->eob_multi128[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf128[plane][eob_multi_ctx], eob_pt - 1,
                   8);
      break;
    case 4:
      ++counts->eob_multi256[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf256[plane][eob_multi_ctx], eob_pt - 1,
                   9);
      break;
    case 5:
      ++counts->eob_multi512[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf512[plane][eob_multi_ctx], eob_pt - 1,
                   10);
      break;
    case 6:
    default:
      ++counts->eob_multi1024[plane][eob_multi_ctx][eob_pt - 1];
      if (allow_update_cdf)
        update_cdf(ec_ctx->eob_flag_cdf1024[plane][eob_multi_ctx], eob_pt - 1,
                   11);
      break;
  }

  if (k_eob_offset_bits[eob_pt] > 0) {
    int eob_shift = k_eob_offset_bits[eob_pt] - 1;
    int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
    counts->eob_extra[txs_ctx][plane][eob_pt][bit]++;
    if (allow_update_cdf)
      update_cdf(ec_ctx->eob_extra_cdf[txs_ctx][plane][eob_pt], bit, 2);
  }
}

static int get_eob_cost(int eob, int seg_eob,
                        const LV_MAP_EOB_COST *txb_eob_costs,
                        const LV_MAP_COEFF_COST *txb_costs, TX_TYPE tx_type) {
  int eob_extra, dummy;
  const int eob_pt = get_eob_pos_token(eob, &eob_extra);
  const int max_eob_pt = get_eob_pos_token(seg_eob, &dummy);
  int eob_cost = 0;
  (void)max_eob_pt;
  const int eob_multi_ctx = (tx_type_to_class[tx_type] == TX_CLASS_2D) ? 0 : 1;
  eob_cost = txb_eob_costs->eob_cost[eob_multi_ctx][eob_pt - 1];

  if (k_eob_offset_bits[eob_pt] > 0) {
    int eob_shift = k_eob_offset_bits[eob_pt] - 1;
    int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
    eob_cost += txb_costs->eob_extra_cost[eob_pt][bit];
    for (int i = 1; i < k_eob_offset_bits[eob_pt]; i++) {
      eob_shift = k_eob_offset_bits[eob_pt] - 1 - i;
      bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
      eob_cost += av1_cost_bit(128, bit);
    }
  }
  return eob_cost;
}

static int get_coeff_cost(const tran_low_t qc, const int scan_idx,
                          const int is_eob, const TxbInfo *const txb_info,
                          const LV_MAP_COEFF_COST *const txb_costs,
                          const int coeff_ctx);

static void get_dist_cost_stats(LevelDownStats *const stats, const int scan_idx,
                                const int is_eob,
                                const LV_MAP_COEFF_COST *const txb_costs,
                                const TxbInfo *const txb_info,
                                int has_nz_tail) {
  const int16_t *const scan = txb_info->scan_order->scan;
  const int coeff_idx = scan[scan_idx];
  const tran_low_t qc = txb_info->qcoeff[coeff_idx];
  const uint8_t *const levels = txb_info->levels;
  stats->new_eob = -1;
  stats->update = 0;
  stats->rd_low = 0;
  stats->rd = 0;
  // TODO(mfo): explore if there's a better way to prevent compiler init
  // warnings
  stats->nz_rd = 0;
  stats->dist_low = 0;
  stats->rate_low = 0;
  stats->low_qc = 0;

  const tran_low_t tqc = txb_info->tcoeff[coeff_idx];
  const int dqv = txb_info->dequant[coeff_idx != 0];
#if CONFIG_NEW_QUANT
  const tran_low_t *nq_dequant_val = txb_info->nq_dequant_vals[coeff_idx != 0];
#endif  // CONFIG_NEW_QUANT

  const int coeff_ctx =
      get_nz_map_ctx(levels, coeff_idx, txb_info->bwl, txb_info->height,
                     scan_idx, is_eob, txb_info->tx_size, txb_info->tx_type);
  const int qc_cost =
      get_coeff_cost(qc, scan_idx, is_eob, txb_info, txb_costs, coeff_ctx);
  if (qc == 0) {
    stats->dist = 0;
    stats->rate = qc_cost;
    return;
  } else {
    const tran_low_t dqc = qcoeff_to_dqcoeff(qc,
#if CONFIG_NEW_QUANT
                                             nq_dequant_val,
#endif  // CONFIG_NEW_QUANT
                                             dqv, txb_info->shift);
    const int64_t dqc_dist = get_coeff_dist(tqc, dqc, txb_info->shift);

    // distortion difference when coefficient is quantized to 0
    const tran_low_t dqc0 = qcoeff_to_dqcoeff(0,
#if CONFIG_NEW_QUANT
                                              nq_dequant_val,
#endif  // CONFIG_NEW_QUANT
                                              dqv, txb_info->shift);

    stats->dist0 = get_coeff_dist(tqc, dqc0, txb_info->shift);
    stats->dist = dqc_dist - stats->dist0;
    stats->rate = qc_cost;
  }

  stats->rd = RDCOST(txb_info->rdmult, stats->rate, stats->dist);

  stats->low_qc = get_lower_coeff(qc);

  if (is_eob && stats->low_qc == 0) {
    stats->rd_low = stats->rd;  // disable selection of low_qc in this case.
  } else {
    if (stats->low_qc == 0) {
      stats->dist_low = 0;
    } else {
      stats->low_dqc = qcoeff_to_dqcoeff(stats->low_qc,
#if CONFIG_NEW_QUANT
                                         nq_dequant_val,
#endif  // CONFIG_NEW_QUANT
                                         dqv, txb_info->shift);
      const int64_t low_dqc_dist =
          get_coeff_dist(tqc, stats->low_dqc, txb_info->shift);
      stats->dist_low = low_dqc_dist - stats->dist0;
    }
    const int low_qc_cost = get_coeff_cost(stats->low_qc, scan_idx, is_eob,
                                           txb_info, txb_costs, coeff_ctx);
    stats->rate_low = low_qc_cost;
    stats->rd_low = RDCOST(txb_info->rdmult, stats->rate_low, stats->dist_low);
  }
  if ((has_nz_tail < 2) && ((scan_idx == txb_info->eob - 1) || !is_eob)) {
    (void)levels;
    const int coeff_ctx_temp =
        get_nz_map_ctx(levels, coeff_idx, txb_info->bwl, txb_info->height,
                       scan_idx, 1, txb_info->tx_size, txb_info->tx_type);
    const int qc_eob_cost =
        get_coeff_cost(qc, scan_idx, 1, txb_info, txb_costs, coeff_ctx_temp);
    int64_t rd_eob = RDCOST(txb_info->rdmult, qc_eob_cost, stats->dist);
    if (stats->low_qc != 0) {
      const int low_qc_eob_cost = get_coeff_cost(
          stats->low_qc, scan_idx, 1, txb_info, txb_costs, coeff_ctx_temp);
      int64_t rd_eob_low =
          RDCOST(txb_info->rdmult, low_qc_eob_cost, stats->dist_low);
      rd_eob = (rd_eob > rd_eob_low) ? rd_eob_low : rd_eob;
    }

    stats->nz_rd = AOMMIN(stats->rd_low, stats->rd) - rd_eob;
  }
}

static INLINE void update_qcoeff(const int coeff_idx, const tran_low_t qc,
                                 const TxbInfo *const txb_info) {
  txb_info->qcoeff[coeff_idx] = qc;
  txb_info->levels[get_padded_idx(coeff_idx, txb_info->bwl)] =
      (uint8_t)clamp(abs(qc), 0, INT8_MAX);
}

static INLINE void update_coeff(const int coeff_idx, const tran_low_t qc,
                                const TxbInfo *const txb_info) {
  update_qcoeff(coeff_idx, qc, txb_info);
  const int dqv = txb_info->dequant[coeff_idx != 0];
#if CONFIG_NEW_QUANT
  const tran_low_t *nq_dequant_val = txb_info->nq_dequant_vals[coeff_idx != 0];
#endif  // CONFIG_NEW_QUANT
  txb_info->dqcoeff[coeff_idx] = qcoeff_to_dqcoeff(qc,
#if CONFIG_NEW_QUANT
                                                   nq_dequant_val,
#endif  // CONFIG_NEW_QUANT
                                                   dqv, txb_info->shift);
}

static INLINE void av1_txb_init_levels(const tran_low_t *const coeff,
                                       const int width, const int height,
                                       uint8_t *const levels) {
  const int stride = width + TX_PAD_HOR;
  uint8_t *ls = levels;

  memset(levels - TX_PAD_TOP * stride, 0,
         sizeof(*levels) * TX_PAD_TOP * stride);
  memset(levels + stride * height, 0,
         sizeof(*levels) * (TX_PAD_BOTTOM * stride + TX_PAD_END));

  for (int i = 0; i < height; i++) {
    for (int j = 0; j < width; j++) {
      *ls++ = (uint8_t)clamp(abs(coeff[i * width + j]), 0, INT8_MAX);
    }
    for (int j = 0; j < TX_PAD_HOR; j++) {
      *ls++ = 0;
    }
  }
}

void av1_get_nz_map_contexts_c(const uint8_t *const levels,
                               const int16_t *const scan, const uint16_t eob,
                               const TX_SIZE tx_size, const TX_TYPE tx_type,
                               int8_t *const coeff_contexts) {
  const int bwl = get_txb_bwl(tx_size);
  const int height = get_txb_high(tx_size);
  for (int i = 0; i < eob; ++i) {
    const int pos = scan[i];
    coeff_contexts[pos] = get_nz_map_ctx(levels, pos, bwl, height, i,
                                         i == eob - 1, tx_size, tx_type);
  }
}

void av1_write_coeffs_txb(const AV1_COMMON *const cm, MACROBLOCKD *xd,
                          aom_writer *w, int blk_row, int blk_col, int plane,
                          TX_SIZE tx_size, const tran_low_t *tcoeff,
                          uint16_t eob, TXB_CTX *txb_ctx) {
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  const PLANE_TYPE plane_type = get_plane_type(plane);
  const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
  const TX_TYPE tx_type =
      av1_get_tx_type(plane_type, xd, blk_row, blk_col, tx_size);
  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
  const int16_t *const scan = scan_order->scan;
  const int seg_eob = av1_get_max_eob(tx_size);
  int c;
  const int bwl = get_txb_bwl(tx_size);
  const int width = get_txb_wide(tx_size);
  const int height = get_txb_high(tx_size);
  int update_eob = -1;
  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
  uint8_t levels_buf[TX_PAD_2D];
  uint8_t *const levels = set_levels(levels_buf, width);
  DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);

  aom_write_bin(w, eob == 0,
                ec_ctx->txb_skip_cdf[txs_ctx][txb_ctx->txb_skip_ctx], 2);
#if CONFIG_TXK_SEL
  if (plane == 0 && eob == 0) {
    assert(tx_type == DCT_DCT);
  }
#endif
  if (eob == 0) return;

  av1_txb_init_levels(tcoeff, width, height, levels);

#if CONFIG_TXK_SEL
  av1_write_tx_type(cm, xd, blk_row, blk_col, plane, tx_size, w);
#endif

  int eob_extra, dummy;
  const int eob_pt = get_eob_pos_token(eob, &eob_extra);
  const int max_eob_pt = get_eob_pos_token(seg_eob, &dummy);

  (void)max_eob_pt;
  const int eob_multi_size = txsize_log2_minus4[tx_size];
  const int eob_multi_ctx = (tx_type_to_class[tx_type] == TX_CLASS_2D) ? 0 : 1;
  switch (eob_multi_size) {
    case 0:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf16[plane_type][eob_multi_ctx], 5);
      break;
    case 1:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf32[plane_type][eob_multi_ctx], 6);
      break;
    case 2:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf64[plane_type][eob_multi_ctx], 7);
      break;
    case 3:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf128[plane_type][eob_multi_ctx], 8);
      break;
    case 4:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf256[plane_type][eob_multi_ctx], 9);
      break;
    case 5:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf512[plane_type][eob_multi_ctx], 10);
      break;
    default:
      aom_write_symbol(w, eob_pt - 1,
                       ec_ctx->eob_flag_cdf1024[plane_type][eob_multi_ctx], 11);
      break;
  }

  if (k_eob_offset_bits[eob_pt] > 0) {
    int eob_shift = k_eob_offset_bits[eob_pt] - 1;
    int bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
    aom_write_bin(w, bit, ec_ctx->eob_extra_cdf[txs_ctx][plane_type][eob_pt],
                  2);
    for (int i = 1; i < k_eob_offset_bits[eob_pt]; i++) {
      eob_shift = k_eob_offset_bits[eob_pt] - 1 - i;
      bit = (eob_extra & (1 << eob_shift)) ? 1 : 0;
      aom_write_bit(w, bit);
    }
  }

  av1_get_nz_map_contexts(levels, scan, eob, tx_size, tx_type, coeff_contexts);

  for (c = eob - 1; c >= 0; --c) {
    const int pos = scan[c];
    const int coeff_ctx = coeff_contexts[pos];
    const tran_low_t v = tcoeff[pos];

    if (c == eob - 1) {
      aom_write_symbol(
          w, AOMMIN(abs(v), 3) - 1,
          ec_ctx->coeff_base_eob_cdf[txs_ctx][plane_type][coeff_ctx], 3);
    } else {
      aom_write_symbol(w, AOMMIN(abs(v), 3),
                       ec_ctx->coeff_base_cdf[txs_ctx][plane_type][coeff_ctx],
                       4);
    }
  }
  update_eob = eob - 1;

  // Loop to code all signs in the transform block,
  // starting with the sign of DC (if applicable)
  for (c = 0; c < eob; ++c) {
    const tran_low_t v = tcoeff[scan[c]];
    const tran_low_t level = abs(v);
    const int sign = (v < 0) ? 1 : 0;
    if (level == 0) continue;

    if (c == 0) {
      aom_write_bin(w, sign,
                    ec_ctx->dc_sign_cdf[plane_type][txb_ctx->dc_sign_ctx], 2);
    } else {
      aom_write_bit(w, sign);
    }
  }

  if (update_eob >= 0) {
    for (c = update_eob; c >= 0; --c) {
      const int pos = scan[c];
      const tran_low_t level = abs(tcoeff[pos]);
      int idx;
      int ctx;

      if (level <= NUM_BASE_LEVELS) continue;

      // level is above 1.
      const int base_range = level - 1 - NUM_BASE_LEVELS;
#if USE_CAUSAL_BR_CTX
      ctx = get_br_ctx(levels, pos, bwl, level_counts[pos], tx_type);

#else
      ctx = get_br_ctx(levels, pos, bwl, level_counts[pos]);
#endif
      for (idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
        const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
        aom_write_symbol(w, k,
#if 0
            ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_16X16)][plane_type][ctx],
#else
                         ec_ctx->coeff_br_cdf[AOMMIN(txs_ctx, TX_32X32)]
                                             [plane_type][ctx],
#endif
                         BR_CDF_SIZE);
        if (k < BR_CDF_SIZE - 1) break;
      }
      if (base_range < COEFF_BASE_RANGE) continue;
      // use 0-th order Golomb code to handle the residual level.
      write_golomb(w,
                   abs(tcoeff[pos]) - COEFF_BASE_RANGE - 1 - NUM_BASE_LEVELS);
    }
  }
}

typedef struct encode_txb_args {
  const AV1_COMMON *cm;
  MACROBLOCK *x;
  aom_writer *w;
} ENCODE_TXB_ARGS;

static void av1_write_coeffs_txb_wrap(const AV1_COMMON *cm, MACROBLOCK *x,
                                      aom_writer *w, int plane, int block,
                                      int blk_row, int blk_col,
                                      TX_SIZE tx_size) {
  MACROBLOCKD *xd = &x->e_mbd;
  tran_low_t *tcoeff = BLOCK_OFFSET(x->mbmi_ext->tcoeff[plane], block);
  uint16_t eob = x->mbmi_ext->eobs[plane][block];
  TXB_CTX txb_ctx = { x->mbmi_ext->txb_skip_ctx[plane][block],
                      x->mbmi_ext->dc_sign_ctx[plane][block] };
  av1_write_coeffs_txb(cm, xd, w, blk_row, blk_col, plane, tx_size, tcoeff, eob,
                       &txb_ctx);
}

void av1_write_coeffs_mb(const AV1_COMMON *const cm, MACROBLOCK *x, int mi_row,
                         int mi_col, aom_writer *w, int plane,
                         BLOCK_SIZE bsize) {
  MACROBLOCKD *xd = &x->e_mbd;
  const struct macroblockd_plane *const pd = &xd->plane[plane];
  const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
  const int stepr = tx_size_high_unit[tx_size];
  const int stepc = tx_size_wide_unit[tx_size];
  const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
  int row, col;
  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
  const int max_blocks_high = max_block_high(xd, plane_bsize, plane);

  if (!is_chroma_reference(mi_row, mi_col, bsize, pd->subsampling_x,
                           pd->subsampling_y))
    return;

  int blk_row, blk_col;
  int block = 0;
  const int step = stepr * stepc;
  const BLOCK_SIZE max_unit_bsize = get_plane_block_size(BLOCK_64X64, pd);
  int mu_blocks_wide = block_size_wide[max_unit_bsize] >> tx_size_wide_log2[0];
  int mu_blocks_high = block_size_high[max_unit_bsize] >> tx_size_high_log2[0];
  mu_blocks_wide = AOMMIN(max_blocks_wide, mu_blocks_wide);
  mu_blocks_high = AOMMIN(max_blocks_high, mu_blocks_high);

  for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
    const int unit_height = AOMMIN(mu_blocks_high + row, max_blocks_high);
    for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
      const int unit_width = AOMMIN(mu_blocks_wide + col, max_blocks_wide);

      for (blk_row = row; blk_row < unit_height; blk_row += stepr) {
        for (blk_col = col; blk_col < unit_width; blk_col += stepc) {
          av1_write_coeffs_txb_wrap(cm, x, w, plane, block, blk_row, blk_col,
                                    tx_size);
          block += step;
        }
      }
    }
  }
}

static INLINE int get_br_cost(tran_low_t abs_qc, int ctx,
                              const int *coeff_lps) {
  const tran_low_t min_level = 1 + NUM_BASE_LEVELS;
  const tran_low_t max_level = 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE;
  (void)ctx;
  if (abs_qc >= min_level) {
    if (abs_qc >= max_level)
      return coeff_lps[COEFF_BASE_RANGE];  // COEFF_BASE_RANGE * cost0;
    else
      return coeff_lps[(abs_qc - min_level)];  //  * cost0 + cost1;
  } else {
    return 0;
  }
}

// Note: don't call this function when eob is 0.
int av1_cost_coeffs_txb(const AV1_COMMON *const cm, const MACROBLOCK *x,
                        const int plane, const int blk_row, const int blk_col,
                        const int block, const TX_SIZE tx_size,
                        const TXB_CTX *const txb_ctx) {
  const MACROBLOCKD *const xd = &x->e_mbd;
  const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
  const PLANE_TYPE plane_type = get_plane_type(plane);
  const TX_TYPE tx_type =
      av1_get_tx_type(plane_type, xd, blk_row, blk_col, tx_size);
  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  const struct macroblock_plane *p = &x->plane[plane];
  const int eob = p->eobs[block];
  const tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  int c, cost;
  const int txb_skip_ctx = txb_ctx->txb_skip_ctx;
  const int bwl = get_txb_bwl(tx_size);
  const int width = get_txb_wide(tx_size);
  const int height = get_txb_high(tx_size);
  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
  const int16_t *const scan = scan_order->scan;
  uint8_t levels_buf[TX_PAD_2D];
  uint8_t *const levels = set_levels(levels_buf, width);
  DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]);
  DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);
  const LV_MAP_COEFF_COST *const coeff_costs =
      &x->coeff_costs[txs_ctx][plane_type];

  const int eob_multi_size = txsize_log2_minus4[tx_size];
  const LV_MAP_EOB_COST *const eob_costs =
      &x->eob_costs[eob_multi_size][plane_type];
  // eob must be greater than 0 here.
  assert(eob > 0);
  cost = coeff_costs->txb_skip_cost[txb_skip_ctx][0];

  av1_txb_init_levels(qcoeff, width, height, levels);

#if CONFIG_TXK_SEL
  cost += av1_tx_type_cost(cm, x, xd, mbmi->sb_type, plane, tx_size, tx_type);
#endif

  const int seg_eob = av1_get_max_eob(tx_size);
  int eob_cost = get_eob_cost(eob, seg_eob, eob_costs, coeff_costs, tx_type);
  cost += eob_cost;

  av1_get_nz_map_contexts(levels, scan, eob, tx_size, tx_type, coeff_contexts);

  for (c = eob - 1; c >= 0; --c) {
    const int pos = scan[c];
    const tran_low_t v = qcoeff[pos];
    const int is_nz = (v != 0);
    const int level = abs(v);
    const int coeff_ctx = coeff_contexts[pos];
    if (c == eob - 1) {
      cost += coeff_costs->base_eob_cost[coeff_ctx][AOMMIN(level, 3) - 1];
    } else {
      cost += coeff_costs->base_cost[coeff_ctx][AOMMIN(level, 3)];
    }

    if (is_nz) {
      int sign = (v < 0) ? 1 : 0;

      // sign bit cost
      if (c == 0) {
        int dc_sign_ctx = txb_ctx->dc_sign_ctx;
        cost += coeff_costs->dc_sign_cost[dc_sign_ctx][sign];
      } else {
        cost += av1_cost_bit(128, sign);
      }
      if (level > NUM_BASE_LEVELS) {
        int ctx;
#if USE_CAUSAL_BR_CTX
        ctx = get_br_ctx(levels, pos, bwl, level_counts[pos], tx_type);
#else
        ctx = get_br_ctx(levels, pos, bwl, level_counts[pos]);
#endif
        const int base_range = level - 1 - NUM_BASE_LEVELS;
        if (base_range < COEFF_BASE_RANGE) {
          cost += coeff_costs->lps_cost[ctx][base_range];
        } else {
          cost += coeff_costs->lps_cost[ctx][COEFF_BASE_RANGE];
        }

        if (level >= 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
          // residual cost
          int r = level - COEFF_BASE_RANGE - NUM_BASE_LEVELS;
          int ri = r;
          int length = 0;

          while (ri) {
            ri >>= 1;
            ++length;
          }

          for (ri = 0; ri < length - 1; ++ri) cost += av1_cost_bit(128, 0);

          for (ri = length - 1; ri >= 0; --ri)
            cost += av1_cost_bit(128, (r >> ri) & 0x01);
        }
      }
    }
  }
  return cost;
}

static INLINE int has_base(tran_low_t qc, int base_idx) {
  const int level = base_idx + 1;
  return abs(qc) >= level;
}

static INLINE int has_br(tran_low_t qc) {
  return abs(qc) >= 1 + NUM_BASE_LEVELS;
}

static INLINE int get_sign_bit_cost(tran_low_t qc, int coeff_idx,
                                    const int (*dc_sign_cost)[2],
                                    int dc_sign_ctx) {
  const int sign = (qc < 0) ? 1 : 0;
  // sign bit cost
  if (coeff_idx == 0) {
    return dc_sign_cost[dc_sign_ctx][sign];
  } else {
    return av1_cost_bit(128, sign);
  }
}
static INLINE int get_golomb_cost(int abs_qc) {
  if (abs_qc >= 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
    // residual cost
    int r = abs_qc - COEFF_BASE_RANGE - NUM_BASE_LEVELS;
    int ri = r;
    int length = 0;

    while (ri) {
      ri >>= 1;
      ++length;
    }

    return av1_cost_literal(2 * length - 1);
  } else {
    return 0;
  }
}

void gen_txb_cache(TxbCache *txb_cache, TxbInfo *txb_info) {
  // gen_nz_count_arr
  const int16_t *const scan = txb_info->scan_order->scan;
  const int bwl = txb_info->bwl;
  const int height = txb_info->height;
  const tran_low_t *const qcoeff = txb_info->qcoeff;
  const uint8_t *const levels = txb_info->levels;
  const BASE_CTX_TABLE *base_ctx_table =
      txb_info->coeff_ctx_table->base_ctx_table;
  for (int c = 0; c < txb_info->eob; ++c) {
    const int coeff_idx = scan[c];  // raster order
    const int row = coeff_idx >> bwl;
    const int col = coeff_idx - (row << bwl);

    txb_cache->nz_count_arr[coeff_idx] =
        get_nz_count(levels + get_padded_idx(coeff_idx, bwl), bwl,
                     tx_type_to_class[txb_info->tx_type]);

    txb_cache->nz_ctx_arr[coeff_idx] =
        get_nz_map_ctx_from_stats(0, coeff_idx, bwl, txb_info->tx_size,
                                  tx_type_to_class[txb_info->tx_type]);

    // gen_base_count_mag_arr
    if (!has_base(qcoeff[coeff_idx], 0)) continue;
    int *base_mag = txb_cache->base_mag_arr[coeff_idx];
    int count[NUM_BASE_LEVELS];
    get_base_count_mag(base_mag, count, qcoeff, bwl, height, row, col);

    for (int i = 0; i < NUM_BASE_LEVELS; ++i) {
      if (!has_base(qcoeff[coeff_idx], i)) break;
      txb_cache->base_count_arr[i][coeff_idx] = count[i];
      const int level = i + 1;
      txb_cache->base_ctx_arr[i][coeff_idx] =
          base_ctx_table[row != 0][col != 0][base_mag[0] > level][count[i]];
    }

    // gen_br_count_mag_arr
    if (!has_br(qcoeff[coeff_idx])) continue;
    int *br_count = txb_cache->br_count_arr + coeff_idx;
    int *br_mag = txb_cache->br_mag_arr[coeff_idx];
    *br_count = get_br_count_mag(br_mag, qcoeff, bwl, height, row, col,
                                 NUM_BASE_LEVELS);
    txb_cache->br_ctx_arr[coeff_idx] =
        get_br_ctx_from_count_mag(row, col, *br_count, br_mag[0]);
  }
}

static INLINE const int *get_level_prob(int level, int coeff_idx,
                                        const TxbCache *txb_cache,
                                        const LV_MAP_COEFF_COST *txb_costs) {
  if (level < 1 + NUM_BASE_LEVELS) {
    const int ctx = txb_cache->nz_ctx_arr[coeff_idx];
    return &txb_costs->base_cost[ctx][level];
  } else if (level >= 1 + NUM_BASE_LEVELS &&
             level < 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
    const int ctx = txb_cache->br_ctx_arr[coeff_idx];
    return txb_costs->lps_cost[ctx];
  } else if (level >= 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
    // printf("get_level_prob does not support golomb\n");
    assert(0);
    return 0;
  } else {
    assert(0);
    return 0;
  }
}

static INLINE void update_mag_arr(int *mag_arr, int abs_qc) {
  if (mag_arr[0] == abs_qc) {
    mag_arr[1] -= 1;
    assert(mag_arr[1] >= 0);
  }
}

static INLINE int get_mag_from_mag_arr(const int *mag_arr) {
  int mag;
  if (mag_arr[1] > 0) {
    mag = mag_arr[0];
  } else if (mag_arr[0] > 0) {
    mag = mag_arr[0] - 1;
  } else {
    // no neighbor
    assert(mag_arr[0] == 0 && mag_arr[1] == 0);
    mag = 0;
  }
  return mag;
}

static int neighbor_level_down_update(int *new_count, int *new_mag, int count,
                                      const int *mag, int coeff_idx,
                                      tran_low_t abs_nb_coeff, int nb_coeff_idx,
                                      int level, const TxbInfo *txb_info) {
  *new_count = count;
  *new_mag = get_mag_from_mag_arr(mag);

  int update = 0;
  // check if br_count changes
  if (abs_nb_coeff == level) {
    update = 1;
    *new_count -= 1;
    assert(*new_count >= 0);
  }
  const int row = coeff_idx >> txb_info->bwl;
  const int col = coeff_idx - (row << txb_info->bwl);
  const int nb_row = nb_coeff_idx >> txb_info->bwl;
  const int nb_col = nb_coeff_idx - (nb_row << txb_info->bwl);

  // check if mag changes
  if (nb_row >= row && nb_col >= col) {
    if (abs_nb_coeff == mag[0]) {
      assert(mag[1] > 0);
      if (mag[1] == 1) {
        // the nb is the only qc with max mag
        *new_mag -= 1;
        assert(*new_mag >= 0);
        update = 1;
      }
    }
  }
  return update;
}

static int try_neighbor_level_down_br(int coeff_idx, int nb_coeff_idx,
                                      const TxbCache *txb_cache,
                                      const LV_MAP_COEFF_COST *txb_costs,
                                      const TxbInfo *txb_info) {
  const tran_low_t qc = txb_info->qcoeff[coeff_idx];
  const tran_low_t abs_qc = abs(qc);
  const int level = NUM_BASE_LEVELS + 1;
  if (abs_qc < level) return 0;

  const tran_low_t nb_coeff = txb_info->qcoeff[nb_coeff_idx];
  const tran_low_t abs_nb_coeff = abs(nb_coeff);
  const int count = txb_cache->br_count_arr[coeff_idx];
  const int *mag = txb_cache->br_mag_arr[coeff_idx];
  int new_count;
  int new_mag;
  const int update =
      neighbor_level_down_update(&new_count, &new_mag, count, mag, coeff_idx,
                                 abs_nb_coeff, nb_coeff_idx, level, txb_info);
  if (update) {
    const int row = coeff_idx >> txb_info->bwl;
    const int col = coeff_idx - (row << txb_info->bwl);
    const int ctx = txb_cache->br_ctx_arr[coeff_idx];
    const int org_cost = get_br_cost(abs_qc, ctx, txb_costs->lps_cost[ctx]);

    const int new_ctx = get_br_ctx_from_count_mag(row, col, new_count, new_mag);
    const int new_cost =
        get_br_cost(abs_qc, new_ctx, txb_costs->lps_cost[new_ctx]);
    const int cost_diff = -org_cost + new_cost;
    return cost_diff;
  } else {
    return 0;
  }
}

static int try_neighbor_level_down_base(int coeff_idx, int nb_coeff_idx,
                                        const TxbCache *txb_cache,
                                        const LV_MAP_COEFF_COST *txb_costs,
                                        const TxbInfo *txb_info) {
  // TODO(olah): not implemented yet
  (void)coeff_idx;
  (void)nb_coeff_idx;
  (void)txb_cache;
  (void)txb_costs;
  (void)txb_info;
  return 0;
}

static int try_neighbor_level_down_nz(int coeff_idx, int nb_coeff_idx,
                                      const TxbCache *txb_cache,
                                      const LV_MAP_COEFF_COST *txb_costs,
                                      TxbInfo *txb_info) {
  // assume eob doesn't change
  const tran_low_t qc = txb_info->qcoeff[coeff_idx];
  const tran_low_t abs_qc = abs(qc);
  const tran_low_t nb_coeff = txb_info->qcoeff[nb_coeff_idx];
  const tran_low_t abs_nb_coeff = abs(nb_coeff);
  if (abs_nb_coeff != 1) return 0;
  const int16_t *const iscan = txb_info->scan_order->iscan;
  const int scan_idx = iscan[coeff_idx];
  if (scan_idx == txb_info->seg_eob) return 0;
  const int nb_scan_idx = iscan[nb_coeff_idx];
  if (nb_scan_idx < scan_idx) {
    const int count = txb_cache->nz_count_arr[coeff_idx];
    (void)count;
    assert(count > 0);
    update_qcoeff(nb_coeff_idx, get_lower_coeff(nb_coeff), txb_info);
    const int new_ctx = get_nz_map_ctx_from_stats(
        0, coeff_idx, txb_info->bwl, txb_info->tx_size,
        tx_type_to_class[txb_info->tx_type]);
    update_qcoeff(nb_coeff_idx, nb_coeff, txb_info);
    const int ctx = txb_cache->nz_ctx_arr[coeff_idx];
    const int org_cost = txb_costs->base_cost[ctx][AOMMIN(abs_qc, 3)];
    const int new_cost = txb_costs->base_cost[new_ctx][AOMMIN(abs_qc, 3)];
    const int cost_diff = new_cost - org_cost;
    return cost_diff;
  } else {
    return 0;
  }
}

static int try_self_level_down(tran_low_t *low_coeff, int coeff_idx,
                               const TxbCache *txb_cache,
                               const LV_MAP_COEFF_COST *txb_costs,
                               TxbInfo *txb_info) {
  const tran_low_t qc = txb_info->qcoeff[coeff_idx];
  if (qc == 0) {
    *low_coeff = 0;
    return 0;
  }
  const tran_low_t abs_qc = abs(qc);
  *low_coeff = get_lower_coeff(qc);
  int cost_diff;
  if (*low_coeff == 0) {
    const int scan_idx = txb_info->scan_order->iscan[coeff_idx];
    const int *level_cost =
        get_level_prob(abs_qc, coeff_idx, txb_cache, txb_costs);
    const int *low_level_cost =
        get_level_prob(abs(*low_coeff), coeff_idx, txb_cache, txb_costs);

    if (scan_idx < txb_info->eob - 1) {
      // When level-0, we code the binary of abs_qc > level
      // but when level-k k > 0 we code the binary of abs_qc == level
      // That's why wee need this special treatment for level-0 map
      // TODO(angiebird): make leve-0 consistent to other levels
      cost_diff = -level_cost[1] + low_level_cost[0] - low_level_cost[1];
    } else {
      cost_diff = -level_cost[1];
    }

    const int sign_cost = get_sign_bit_cost(
        qc, coeff_idx, txb_costs->dc_sign_cost, txb_info->txb_ctx->dc_sign_ctx);
    cost_diff -= sign_cost;
  } else if (abs_qc <= NUM_BASE_LEVELS) {
    const int *level_cost =
        get_level_prob(abs_qc, coeff_idx, txb_cache, txb_costs);
    const int *low_level_cost =
        get_level_prob(abs(*low_coeff), coeff_idx, txb_cache, txb_costs);
    cost_diff = -level_cost[1] + low_level_cost[1] - low_level_cost[0];
  } else if (abs_qc == NUM_BASE_LEVELS + 1) {
    const int *level_cost =
        get_level_prob(abs_qc, coeff_idx, txb_cache, txb_costs);
    const int *low_level_cost =
        get_level_prob(abs(*low_coeff), coeff_idx, txb_cache, txb_costs);
    cost_diff = -level_cost[0] + low_level_cost[1] - low_level_cost[0];
  } else if (abs_qc < 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
    const int *level_cost =
        get_level_prob(abs_qc, coeff_idx, txb_cache, txb_costs);
    const int *low_level_cost =
        get_level_prob(abs(*low_coeff), coeff_idx, txb_cache, txb_costs);

    cost_diff = -level_cost[abs_qc - 1 - NUM_BASE_LEVELS] +
                low_level_cost[abs(*low_coeff) - 1 - NUM_BASE_LEVELS];
  } else if (abs_qc == 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE) {
    const int *low_level_cost =
        get_level_prob(abs(*low_coeff), coeff_idx, txb_cache, txb_costs);
    cost_diff = -get_golomb_cost(abs_qc) - low_level_cost[COEFF_BASE_RANGE] +
                low_level_cost[COEFF_BASE_RANGE - 1];
  } else {
    assert(abs_qc > 1 + NUM_BASE_LEVELS + COEFF_BASE_RANGE);
    const tran_low_t abs_low_coeff = abs(*low_coeff);
    cost_diff = -get_golomb_cost(abs_qc) + get_golomb_cost(abs_low_coeff);
  }
  return cost_diff;
}

#define COST_MAP_SIZE 5
#define COST_MAP_OFFSET 2

static INLINE int check_nz_neighbor(tran_low_t qc) { return abs(qc) == 1; }

static INLINE int check_base_neighbor(tran_low_t qc) {
  return abs(qc) <= 1 + NUM_BASE_LEVELS;
}

static INLINE int check_br_neighbor(tran_low_t qc) {
  return abs(qc) > BR_MAG_OFFSET;
}

#define FAST_OPTIMIZE_TXB 1

#if FAST_OPTIMIZE_TXB
#define ALNB_REF_OFFSET_NUM 2
static const int alnb_ref_offset[ALNB_REF_OFFSET_NUM][2] = {
  { -1, 0 }, { 0, -1 },
};
#define NB_REF_OFFSET_NUM 4
static const int nb_ref_offset[NB_REF_OFFSET_NUM][2] = {
  { -1, 0 }, { 0, -1 }, { 1, 0 }, { 0, 1 },
};
#endif  // FAST_OPTIMIZE_TXB

// TODO(angiebird): add static to this function once it's called
int try_level_down(int coeff_idx, const TxbCache *txb_cache,
                   const LV_MAP_COEFF_COST *txb_costs, TxbInfo *txb_info,
                   int (*cost_map)[COST_MAP_SIZE], int fast_mode) {
#if !FAST_OPTIMIZE_TXB
  (void)fast_mode;
#endif
  if (cost_map) {
    for (int i = 0; i < COST_MAP_SIZE; ++i) av1_zero(cost_map[i]);
  }

  tran_low_t qc = txb_info->qcoeff[coeff_idx];
  tran_low_t low_coeff;
  if (qc == 0) return 0;
  int accu_cost_diff = 0;

  const int16_t *const iscan = txb_info->scan_order->iscan;
  const int eob = txb_info->eob;
  const int scan_idx = iscan[coeff_idx];
  if (scan_idx < eob) {
    const int cost_diff = try_self_level_down(&low_coeff, coeff_idx, txb_cache,
                                              txb_costs, txb_info);
    if (cost_map)
      cost_map[0 + COST_MAP_OFFSET][0 + COST_MAP_OFFSET] = cost_diff;
    accu_cost_diff += cost_diff;
  }

  const int row = coeff_idx >> txb_info->bwl;
  const int col = coeff_idx - (row << txb_info->bwl);
  if (check_nz_neighbor(qc)) {
#if FAST_OPTIMIZE_TXB
    const int(*ref_offset)[2];
    int ref_num;
    if (fast_mode) {
      ref_offset = alnb_ref_offset;
      ref_num = ALNB_REF_OFFSET_NUM;
    } else {
      ref_offset = sig_ref_offset;
      ref_num = SIG_REF_OFFSET_NUM;
    }
#else
    const int(*ref_offset)[2] = sig_ref_offset;
    const int ref_num = SIG_REF_OFFSET_NUM;
#endif
    for (int i = 0; i < ref_num; ++i) {
      const int nb_row = row - ref_offset[i][0];
      const int nb_col = col - ref_offset[i][1];
      const int nb_coeff_idx = nb_row * txb_info->width + nb_col;

      if (nb_row < 0 || nb_col < 0 || nb_row >= txb_info->height ||
          nb_col >= txb_info->width)
        continue;

      const int nb_scan_idx = iscan[nb_coeff_idx];
      if (nb_scan_idx < eob) {
        const int cost_diff = try_neighbor_level_down_nz(
            nb_coeff_idx, coeff_idx, txb_cache, txb_costs, txb_info);
        if (cost_map)
          cost_map[nb_row - row + COST_MAP_OFFSET]
                  [nb_col - col + COST_MAP_OFFSET] += cost_diff;
        accu_cost_diff += cost_diff;
      }
    }
  }

  if (check_base_neighbor(qc)) {
#if FAST_OPTIMIZE_TXB
    const int(*ref_offset)[2];
    int ref_num;
    if (fast_mode) {
      ref_offset = nb_ref_offset;
      ref_num = NB_REF_OFFSET_NUM;
    } else {
      ref_offset = base_ref_offset;
      ref_num = BASE_CONTEXT_POSITION_NUM;
    }
#else
    const int(*ref_offset)[2] = base_ref_offset;
    int ref_num = BASE_CONTEXT_POSITION_NUM;
#endif
    for (int i = 0; i < ref_num; ++i) {
      const int nb_row = row - ref_offset[i][0];
      const int nb_col = col - ref_offset[i][1];
      const int nb_coeff_idx = nb_row * txb_info->width + nb_col;

      if (nb_row < 0 || nb_col < 0 || nb_row >= txb_info->height ||
          nb_col >= txb_info->width)
        continue;

      const int nb_scan_idx = iscan[nb_coeff_idx];
      if (nb_scan_idx < eob) {
        const int cost_diff = try_neighbor_level_down_base(
            nb_coeff_idx, coeff_idx, txb_cache, txb_costs, txb_info);
        if (cost_map)
          cost_map[nb_row - row + COST_MAP_OFFSET]
                  [nb_col - col + COST_MAP_OFFSET] += cost_diff;
        accu_cost_diff += cost_diff;
      }
    }
  }

  if (check_br_neighbor(qc)) {
#if FAST_OPTIMIZE_TXB
    const int(*ref_offset)[2];
    int ref_num;
    if (fast_mode) {
      ref_offset = nb_ref_offset;
      ref_num = NB_REF_OFFSET_NUM;
    } else {
      ref_offset = br_ref_offset;
      ref_num = BR_CONTEXT_POSITION_NUM;
    }
#else
    const int(*ref_offset)[2] = br_ref_offset;
    const int ref_num = BR_CONTEXT_POSITION_NUM;
#endif
    for (int i = 0; i < ref_num; ++i) {
      const int nb_row = row - ref_offset[i][0];
      const int nb_col = col - ref_offset[i][1];
      const int nb_coeff_idx = nb_row * txb_info->width + nb_col;

      if (nb_row < 0 || nb_col < 0 || nb_row >= txb_info->height ||
          nb_col >= txb_info->width)
        continue;

      const int nb_scan_idx = iscan[nb_coeff_idx];
      if (nb_scan_idx < eob) {
        const int cost_diff = try_neighbor_level_down_br(
            nb_coeff_idx, coeff_idx, txb_cache, txb_costs, txb_info);
        if (cost_map)
          cost_map[nb_row - row + COST_MAP_OFFSET]
                  [nb_col - col + COST_MAP_OFFSET] += cost_diff;
        accu_cost_diff += cost_diff;
      }
    }
  }

  return accu_cost_diff;
}

static INLINE void set_eob(TxbInfo *txb_info, int eob) {
  txb_info->eob = eob;
  txb_info->seg_eob = av1_get_max_eob(txb_info->tx_size);
}

// TODO(angiebird): add static to this function it's called
void update_level_down(const int coeff_idx, TxbCache *const txb_cache,
                       TxbInfo *const txb_info) {
  const tran_low_t qc = txb_info->qcoeff[coeff_idx];
  const int abs_qc = abs(qc);
  if (qc == 0) return;
  const tran_low_t low_coeff = get_lower_coeff(qc);
  update_coeff(coeff_idx, low_coeff, txb_info);

  const int row = coeff_idx >> txb_info->bwl;
  const int col = coeff_idx - (row << txb_info->bwl);
  const int eob = txb_info->eob;
  const int16_t *const iscan = txb_info->scan_order->iscan;
  for (int i = 0; i < SIG_REF_OFFSET_NUM; ++i) {
    const int nb_row = row - sig_ref_offset[i][0];
    const int nb_col = col - sig_ref_offset[i][1];

    if (!(nb_row >= 0 && nb_col >= 0 && nb_row < txb_info->height &&
          nb_col < txb_info->width))
      continue;

    const int nb_coeff_idx = nb_row * txb_info->width + nb_col;
    const int nb_scan_idx = iscan[nb_coeff_idx];
    if (nb_scan_idx < eob) {
      const int scan_idx = iscan[coeff_idx];
      if (scan_idx < nb_scan_idx) {
        const int level = 1;
        if (abs_qc == level) {
          txb_cache->nz_count_arr[nb_coeff_idx] -= 1;
          assert(txb_cache->nz_count_arr[nb_coeff_idx] >= 0);
        }
        txb_cache->nz_ctx_arr[nb_coeff_idx] = get_nz_map_ctx_from_stats(
            0, nb_coeff_idx, txb_info->bwl, txb_info->tx_size,
            tx_type_to_class[txb_info->tx_type]);
      }
    }
  }

  const BASE_CTX_TABLE *base_ctx_table =
      txb_info->coeff_ctx_table->base_ctx_table;
  for (int i = 0; i < BASE_CONTEXT_POSITION_NUM; ++i) {
    const int nb_row = row - base_ref_offset[i][0];
    const int nb_col = col - base_ref_offset[i][1];
    const int nb_coeff_idx = nb_row * txb_info->width + nb_col;

    if (!(nb_row >= 0 && nb_col >= 0 && nb_row < txb_info->height &&
          nb_col < txb_info->width))
      continue;

    const tran_low_t nb_coeff = txb_info->qcoeff[nb_coeff_idx];
    if (!has_base(nb_coeff, 0)) continue;
    const int nb_scan_idx = iscan[nb_coeff_idx];
    if (nb_scan_idx < eob) {
      if (row >= nb_row && col >= nb_col)
        update_mag_arr(txb_cache->base_mag_arr[nb_coeff_idx], abs_qc);
      const int mag =
          get_mag_from_mag_arr(txb_cache->base_mag_arr[nb_coeff_idx]);
      for (int base_idx = 0; base_idx < NUM_BASE_LEVELS; ++base_idx) {
        if (!has_base(nb_coeff, base_idx)) continue;
        const int level = base_idx + 1;
        if (abs_qc == level) {
          txb_cache->base_count_arr[base_idx][nb_coeff_idx] -= 1;
          assert(txb_cache->base_count_arr[base_idx][nb_coeff_idx] >= 0);
        }
        const int count = txb_cache->base_count_arr[base_idx][nb_coeff_idx];
        txb_cache->base_ctx_arr[base_idx][nb_coeff_idx] =
            base_ctx_table[nb_row != 0][nb_col != 0][mag > level][count];
      }
    }
  }

  for (int i = 0; i < BR_CONTEXT_POSITION_NUM; ++i) {
    const int nb_row = row - br_ref_offset[i][0];
    const int nb_col = col - br_ref_offset[i][1];
    const int nb_coeff_idx = nb_row * txb_info->width + nb_col;

    if (!(nb_row >= 0 && nb_col >= 0 && nb_row < txb_info->height &&
          nb_col < txb_info->width))
      continue;

    const int nb_scan_idx = iscan[nb_coeff_idx];
    const tran_low_t nb_coeff = txb_info->qcoeff[nb_coeff_idx];
    if (!has_br(nb_coeff)) continue;
    if (nb_scan_idx < eob) {
      const int level = 1 + NUM_BASE_LEVELS;
      if (abs_qc == level) {
        txb_cache->br_count_arr[nb_coeff_idx] -= 1;
        assert(txb_cache->br_count_arr[nb_coeff_idx] >= 0);
      }
      if (row >= nb_row && col >= nb_col)
        update_mag_arr(txb_cache->br_mag_arr[nb_coeff_idx], abs_qc);
      const int count = txb_cache->br_count_arr[nb_coeff_idx];
      const int mag = get_mag_from_mag_arr(txb_cache->br_mag_arr[nb_coeff_idx]);
      txb_cache->br_ctx_arr[nb_coeff_idx] =
          get_br_ctx_from_count_mag(nb_row, nb_col, count, mag);
    }
  }
}

static int get_coeff_cost(const tran_low_t qc, const int scan_idx,
                          const int is_eob, const TxbInfo *const txb_info,
                          const LV_MAP_COEFF_COST *const txb_costs,
                          const int coeff_ctx) {
  const TXB_CTX *txb_ctx = txb_info->txb_ctx;
  const int is_nz = (qc != 0);
  const tran_low_t abs_qc = abs(qc);
  int cost = 0;
  const int16_t *const scan = txb_info->scan_order->scan;
  const int pos = scan[scan_idx];

  if (is_eob) {
    cost += txb_costs->base_eob_cost[coeff_ctx][AOMMIN(abs_qc, 3) - 1];
  } else {
    cost += txb_costs->base_cost[coeff_ctx][AOMMIN(abs_qc, 3)];
  }
  if (is_nz) {
    cost += get_sign_bit_cost(qc, scan_idx, txb_costs->dc_sign_cost,
                              txb_ctx->dc_sign_ctx);

    if (abs_qc > NUM_BASE_LEVELS) {
      const int row = pos >> txb_info->bwl;
      const int col = pos - (row << txb_info->bwl);

#if USE_CAUSAL_BR_CTX
      (void)col;
      const int count = 0;
      const int ctx = get_br_ctx(txb_info->levels, pos, txb_info->bwl, count,
                                 txb_info->tx_type);
#else
      const int count = get_level_count(
          txb_info->levels, (1 << txb_info->bwl) + TX_PAD_HOR, row, col,
          NUM_BASE_LEVELS, br_ref_offset, BR_CONTEXT_POSITION_NUM);
      const int ctx = get_br_ctx(txb_info->levels, pos, txb_info->bwl, count);
#endif
      cost += get_br_cost(abs_qc, ctx, txb_costs->lps_cost[ctx]);
      cost += get_golomb_cost(abs_qc);
    }
  }
  return cost;
}

#if TEST_OPTIMIZE_TXB
#define ALL_REF_OFFSET_NUM 17
static const int all_ref_offset[ALL_REF_OFFSET_NUM][2] = {
  { 0, 0 },  { -2, -1 }, { -2, 0 }, { -2, 1 }, { -1, -2 }, { -1, -1 },
  { -1, 0 }, { -1, 1 },  { 0, -2 }, { 0, -1 }, { 1, -2 },  { 1, -1 },
  { 1, 0 },  { 2, 0 },   { 0, 1 },  { 0, 2 },  { 1, 1 },
};

static int try_level_down_ref(int coeff_idx, const LV_MAP_COEFF_COST *txb_costs,
                              TxbInfo *txb_info,
                              int (*cost_map)[COST_MAP_SIZE]) {
  if (cost_map) {
    for (int i = 0; i < COST_MAP_SIZE; ++i) av1_zero(cost_map[i]);
  }
  tran_low_t qc = txb_info->qcoeff[coeff_idx];
  if (qc == 0) return 0;
  int row = coeff_idx >> txb_info->bwl;
  int col = coeff_idx - (row << txb_info->bwl);
  int org_cost = 0;
  for (int i = 0; i < ALL_REF_OFFSET_NUM; ++i) {
    int nb_row = row - all_ref_offset[i][0];
    int nb_col = col - all_ref_offset[i][1];
    int nb_coeff_idx = nb_row * txb_info->width + nb_col;
    int nb_scan_idx = txb_info->scan_order->iscan[nb_coeff_idx];
    if (nb_scan_idx < txb_info->eob && nb_row >= 0 && nb_col >= 0 &&
        nb_row < txb_info->height && nb_col < txb_info->width) {
      const tran_low_t nb_coeff = txb_info->qcoeff[nb_coeff_idx];
      const int coeff_ctx = get_nz_map_ctx(
          txb_info->levels, nb_coeff_idx, txb_info->bwl, txb_info->height,
          nb_scan_idx, is_eob, txb_info->tx_size, txb_info->tx_type);
      const int cost = get_coeff_cost(nb_coeff, nb_scan_idx, is_eob, txb_info,
                                      txb_costs, coeff_ctx);
      if (cost_map)
        cost_map[nb_row - row + COST_MAP_OFFSET]
                [nb_col - col + COST_MAP_OFFSET] -= cost;
      org_cost += cost;
    }
  }
  update_qcoeff(coeff_idx, get_lower_coeff(qc), txb_info);
  int new_cost = 0;
  for (int i = 0; i < ALL_REF_OFFSET_NUM; ++i) {
    int nb_row = row - all_ref_offset[i][0];
    int nb_col = col - all_ref_offset[i][1];
    int nb_coeff_idx = nb_row * txb_info->width + nb_col;
    int nb_scan_idx = txb_info->scan_order->iscan[nb_coeff_idx];
    if (nb_scan_idx < txb_info->eob && nb_row >= 0 && nb_col >= 0 &&
        nb_row < txb_info->height && nb_col < txb_info->width) {
      const tran_low_t nb_coeff = txb_info->qcoeff[nb_coeff_idx];
      const int coeff_ctx = get_nz_map_ctx(
          txb_info->levels, nb_coeff_idx, txb_info->bwl, txb_info->height,
          nb_scan_idx, is_eob, txb_info->tx_size, txb_info->tx_type);
      const int cost = get_coeff_cost(nb_coeff, nb_scan_idx, is_eob, txb_info,
                                      txb_costs, coeff_ctx);
      if (cost_map)
        cost_map[nb_row - row + COST_MAP_OFFSET]
                [nb_col - col + COST_MAP_OFFSET] += cost;
      new_cost += cost;
    }
  }
  update_qcoeff(coeff_idx, qc, txb_info);
  return new_cost - org_cost;
}

static void test_level_down(int coeff_idx, const TxbCache *txb_cache,
                            const LV_MAP_COEFF_COST *txb_costs,
                            TxbInfo *txb_info) {
  int cost_map[COST_MAP_SIZE][COST_MAP_SIZE];
  int ref_cost_map[COST_MAP_SIZE][COST_MAP_SIZE];
  const int cost_diff =
      try_level_down(coeff_idx, txb_cache, txb_costs, txb_info, cost_map, 0);
  const int cost_diff_ref =
      try_level_down_ref(coeff_idx, txb_costs, txb_info, ref_cost_map);
  if (cost_diff != cost_diff_ref) {
    printf("qc %d cost_diff %d cost_diff_ref %d\n", txb_info->qcoeff[coeff_idx],
           cost_diff, cost_diff_ref);
    for (int r = 0; r < COST_MAP_SIZE; ++r) {
      for (int c = 0; c < COST_MAP_SIZE; ++c) {
        printf("%d:%d ", cost_map[r][c], ref_cost_map[r][c]);
      }
      printf("\n");
    }
  }
}
#endif

// TODO(angiebird): make this static once it's called
int get_txb_cost(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs) {
  int cost = 0;
  const int txb_skip_ctx = txb_info->txb_ctx->txb_skip_ctx;
  const int16_t *const scan = txb_info->scan_order->scan;
  if (txb_info->eob == 0) {
    cost = txb_costs->txb_skip_cost[txb_skip_ctx][1];
    return cost;
  }
  cost = txb_costs->txb_skip_cost[txb_skip_ctx][0];
  for (int c = 0; c < txb_info->eob; ++c) {
    const int pos = scan[c];
    const tran_low_t qc = txb_info->qcoeff[pos];
    const int coeff_ctx = get_nz_map_ctx(
        txb_info->levels, pos, txb_info->bwl, txb_info->height, c,
        c == txb_info->eob - 1, txb_info->tx_size, txb_info->tx_type);
    const int coeff_cost = get_coeff_cost(qc, c, c == txb_info->eob - 1,
                                          txb_info, txb_costs, coeff_ctx);
    cost += coeff_cost;
  }
  return cost;
}

#if TEST_OPTIMIZE_TXB
void test_try_change_eob(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
                         TxbCache *txb_cache) {
  const int eob = txb_info->eob;
  const int16_t *const scan = txb_info->scan_order->scan;
  if (eob > 0) {
    int last_si = eob - 1;
    int last_ci = scan[last_si];
    int last_coeff = txb_info->qcoeff[last_ci];
    if (abs(last_coeff) == 1) {
      int new_eob;
      int cost_diff =
          try_change_eob(&new_eob, last_ci, txb_cache, txb_costs, txb_info, 0);
      int org_eob = txb_info->eob;
      int cost = get_txb_cost(txb_info, txb_costs);

      update_qcoeff(last_ci, get_lower_coeff(last_coeff), txb_info);
      set_eob(txb_info, new_eob);
      int new_cost = get_txb_cost(txb_info, txb_costs);
      set_eob(txb_info, org_eob);
      update_qcoeff(last_ci, last_coeff, txb_info);

      int ref_cost_diff = -cost + new_cost;
      if (cost_diff != ref_cost_diff)
        printf("org_eob %d new_eob %d cost_diff %d ref_cost_diff %d\n", org_eob,
               new_eob, cost_diff, ref_cost_diff);
    }
  }
}
#endif

#if 1
static int optimize_txb(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
                        const LV_MAP_EOB_COST *txb_eob_costs,
                        TxbCache *txb_cache, int dry_run, int fast_mode) {
  (void)fast_mode;
  (void)txb_cache;
  int update = 0;
  // return update;  // TODO(DKHE): training only.
  if (txb_info->eob == 0) return update;
  const int max_eob = av1_get_max_eob(txb_info->tx_size);

#if TEST_OPTIMIZE_TXB
  int64_t sse;
  int64_t org_dist =
      av1_block_error_c(txb_info->tcoeff, txb_info->dqcoeff, max_eob, &sse) *
      (1 << (2 * txb_info->shift));
  int org_cost = get_txb_cost(txb_info, txb_probs);
#endif

  tran_low_t *org_qcoeff = txb_info->qcoeff;
  tran_low_t *org_dqcoeff = txb_info->dqcoeff;
  uint8_t *const org_levels = txb_info->levels;

  tran_low_t tmp_qcoeff[MAX_TX_SQUARE];
  tran_low_t tmp_dqcoeff[MAX_TX_SQUARE];
  uint8_t tmp_levels_buf[TX_PAD_2D];
  uint8_t *const tmp_levels = set_levels(tmp_levels_buf, txb_info->width);
  const int org_eob = txb_info->eob;
  if (dry_run) {
    const int stride = txb_info->width + TX_PAD_HOR;
    const int levels_size =

        (stride * (txb_info->height + TX_PAD_VER) + TX_PAD_END);
    memcpy(tmp_qcoeff, org_qcoeff, sizeof(org_qcoeff[0]) * max_eob);
    memcpy(tmp_dqcoeff, org_dqcoeff, sizeof(org_dqcoeff[0]) * max_eob);
    memcpy(tmp_levels, org_levels - TX_PAD_TOP * stride,
           sizeof(org_levels[0]) * levels_size);
    txb_info->qcoeff = tmp_qcoeff;
    txb_info->dqcoeff = tmp_dqcoeff;
    txb_info->levels = tmp_levels;
  }

  const int16_t *const scan = txb_info->scan_order->scan;

  // forward optimize the nz_map`
  const int init_eob = txb_info->eob;
  const int seg_eob = txb_info->seg_eob;
  const int eob_cost = get_eob_cost(init_eob, seg_eob, txb_eob_costs, txb_costs,
                                    txb_info->tx_type);

  // backward optimize the level-k map
  int64_t accu_rate = eob_cost;
  int64_t accu_dist = 0;
  int64_t prev_eob_rd_cost = INT64_MAX;
  int64_t cur_eob_rd_cost = 0;
  int8_t has_nz_tail = 0;

  for (int si = init_eob - 1; si >= 0; --si) {
    const int coeff_idx = scan[si];
    tran_low_t qc = txb_info->qcoeff[coeff_idx];

    LevelDownStats stats;
    get_dist_cost_stats(&stats, si, si == init_eob - 1, txb_costs, txb_info,
                        has_nz_tail);

    if (qc == 0) {
      accu_rate += stats.rate;
    } else {
      if (has_nz_tail < 2) {
        // check if it is better to make this the last significant coefficient
        int cur_eob_rate = get_eob_cost(si + 1, seg_eob, txb_eob_costs,
                                        txb_costs, txb_info->tx_type);
        cur_eob_rd_cost = RDCOST(txb_info->rdmult, cur_eob_rate, 0);
        prev_eob_rd_cost =
            RDCOST(txb_info->rdmult, accu_rate, accu_dist) + stats.nz_rd;
        if (cur_eob_rd_cost <= prev_eob_rd_cost) {
          update = 1;
          for (int j = si + 1; j < txb_info->eob; j++) {
            const int coeff_pos_j = scan[j];
            update_coeff(coeff_pos_j, 0, txb_info);
          }
          txb_info->eob = si + 1;

          // rerun cost calculation due to change of eob
          accu_rate = cur_eob_rate;
          accu_dist = 0;
          get_dist_cost_stats(&stats, si, 1, txb_costs, txb_info, has_nz_tail);
          if ((stats.rd_low < stats.rd) && (stats.low_qc != 0)) {
            update = 1;
            update_coeff(coeff_idx, stats.low_qc, txb_info);
            accu_rate += stats.rate_low;
            accu_dist += stats.dist_low;
          } else {
            accu_rate += stats.rate;
            accu_dist += stats.dist;
          }
          continue;
        }
      }

      int bUpdCoeff = 0;
      if (stats.rd_low < stats.rd) {
        if ((si < txb_info->eob - 1)) {
          bUpdCoeff = 1;
          update = 1;
        }
      } else {
        ++has_nz_tail;
      }

      if (bUpdCoeff) {
        update_coeff(coeff_idx, stats.low_qc, txb_info);
        accu_rate += stats.rate_low;
        accu_dist += stats.dist_low;
      } else {
        accu_rate += stats.rate;
        accu_dist += stats.dist;
      }
    }
  }  // for (si)

  int non_zero_blk_rate =
      txb_costs->txb_skip_cost[txb_info->txb_ctx->txb_skip_ctx][0];
  prev_eob_rd_cost =
      RDCOST(txb_info->rdmult, accu_rate + non_zero_blk_rate, accu_dist);

  int zero_blk_rate =
      txb_costs->txb_skip_cost[txb_info->txb_ctx->txb_skip_ctx][1];
  int64_t zero_blk_rd_cost = RDCOST(txb_info->rdmult, zero_blk_rate, 0);
  if (zero_blk_rd_cost <= prev_eob_rd_cost) {
    update = 1;
    for (int j = 0; j < txb_info->eob; j++) {
      const int coeff_pos_j = scan[j];
      update_coeff(coeff_pos_j, 0, txb_info);
    }
    txb_info->eob = 0;
  }

#if TEST_OPTIMIZE_TXB
  int cost_diff = 0;
  int64_t dist_diff = 0;
  int64_t rd_diff = 0;
  int64_t new_dist =
      av1_block_error_c(txb_info->tcoeff, txb_info->dqcoeff, max_eob, &sse) *
      (1 << (2 * txb_info->shift));
  int new_cost = get_txb_cost(txb_info, txb_probs);
  int64_t ref_dist_diff = new_dist - org_dist;
  int ref_cost_diff = new_cost - org_cost;
  if (cost_diff != ref_cost_diff || dist_diff != ref_dist_diff)
    printf(
        "overall rd_diff %ld\ncost_diff %d ref_cost_diff%d\ndist_diff %ld "
        "ref_dist_diff %ld\neob %d new_eob %d\n\n",
        rd_diff, cost_diff, ref_cost_diff, dist_diff, ref_dist_diff, org_eob,
        txb_info->eob);
#endif
  if (dry_run) {
    txb_info->qcoeff = org_qcoeff;
    txb_info->dqcoeff = org_dqcoeff;
    txb_info->levels = org_levels;
    set_eob(txb_info, org_eob);
  }
  return update;
}

#else
static int optimize_txb(TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
                        TxbCache *txb_cache, int dry_run, int fast_mode) {
  int update = 0;
  if (txb_info->eob == 0) return update;
  int cost_diff = 0;
  int64_t dist_diff = 0;
  int64_t rd_diff = 0;
  const int max_eob = av1_get_max_eob(txb_info->tx_size);

#if TEST_OPTIMIZE_TXB
  int64_t sse;
  int64_t org_dist =
      av1_block_error_c(txb_info->tcoeff, txb_info->dqcoeff, max_eob, &sse) *
      (1 << (2 * txb_info->shift));
  int org_cost = get_txb_cost(txb_info, txb_costs);
#endif

  tran_low_t *org_qcoeff = txb_info->qcoeff;
  tran_low_t *org_dqcoeff = txb_info->dqcoeff;
  uint8_t *const org_levels = txb_info->levels;

  tran_low_t tmp_qcoeff[MAX_TX_SQUARE];
  tran_low_t tmp_dqcoeff[MAX_TX_SQUARE];
  uint8_t tmp_levels_buf[TX_PAD_2D];
  uint8_t *const tmp_levels = set_levels(tmp_levels_buf, txb_info->width);
  const int org_eob = txb_info->eob;
  if (dry_run) {
    const int stride = txb_info->width + TX_PAD_HOR;
    const int levels_size =

        (stride * (txb_info->height + TX_PAD_VER) + TX_PAD_END);
    memcpy(tmp_qcoeff, org_qcoeff, sizeof(org_qcoeff[0]) * max_eob);
    memcpy(tmp_dqcoeff, org_dqcoeff, sizeof(org_dqcoeff[0]) * max_eob);
    memcpy(tmp_levels, org_levels - TX_PAD_TOP * stride,
           sizeof(org_levels[0]) * levels_size);
    txb_info->qcoeff = tmp_qcoeff;
    txb_info->dqcoeff = tmp_dqcoeff;
    txb_info->levels = tmp_levels;
  }

  const int16_t *const scan = txb_info->scan_order->scan;

  // forward optimize the nz_map
  const int cur_eob = txb_info->eob;
  for (int si = 0; si < cur_eob; ++si) {
    const int coeff_idx = scan[si];
    tran_low_t qc = txb_info->qcoeff[coeff_idx];
    if (abs(qc) == 1) {
      LevelDownStats stats;
      try_level_down_facade(&stats, si, txb_cache, txb_costs, txb_info,
                            fast_mode);
      if (stats.update) {
        update = 1;
        cost_diff += stats.cost_diff;
        dist_diff += stats.dist_diff;
        rd_diff += stats.rd_diff;
        update_level_down(coeff_idx, txb_cache, txb_info);
        set_eob(txb_info, stats.new_eob);
      }
    }
  }

  // backward optimize the level-k map
  int eob_fix = 0;
  for (int si = txb_info->eob - 1; si >= 0; --si) {
    const int coeff_idx = scan[si];
    if (eob_fix == 1 && txb_info->qcoeff[coeff_idx] == 1) {
      // when eob is fixed, there is not need to optimize again when
      // abs(qc) == 1
      continue;
    }
    LevelDownStats stats;
    try_level_down_facade(&stats, si, txb_cache, txb_costs, txb_info,
                          fast_mode);
    if (stats.update) {
#if TEST_OPTIMIZE_TXB
// printf("si %d low_qc %d cost_diff %d dist_diff %ld rd_diff %ld eob %d new_eob
// %d\n", si, stats.low_qc, stats.cost_diff, stats.dist_diff, stats.rd_diff,
// txb_info->eob, stats.new_eob);
#endif
      update = 1;
      cost_diff += stats.cost_diff;
      dist_diff += stats.dist_diff;
      rd_diff += stats.rd_diff;
      update_level_down(coeff_idx, txb_cache, txb_info);
      set_eob(txb_info, stats.new_eob);
    }
    if (eob_fix == 0 && txb_info->qcoeff[coeff_idx] != 0) eob_fix = 1;
    if (si > txb_info->eob) si = txb_info->eob;
  }
#if TEST_OPTIMIZE_TXB
  int64_t new_dist =
      av1_block_error_c(txb_info->tcoeff, txb_info->dqcoeff, max_eob, &sse) *
      (1 << (2 * txb_info->shift));
  int new_cost = get_txb_cost(txb_info, txb_costs);
  int64_t ref_dist_diff = new_dist - org_dist;
  int ref_cost_diff = new_cost - org_cost;
  if (cost_diff != ref_cost_diff || dist_diff != ref_dist_diff)
    printf(
        "overall rd_diff %ld\ncost_diff %d ref_cost_diff%d\ndist_diff %ld "
        "ref_dist_diff %ld\neob %d new_eob %d\n\n",
        rd_diff, cost_diff, ref_cost_diff, dist_diff, ref_dist_diff, org_eob,
        txb_info->eob);
#endif
  if (dry_run) {
    txb_info->qcoeff = org_qcoeff;
    txb_info->dqcoeff = org_dqcoeff;
    txb_info->levels = org_levels;
    set_eob(txb_info, org_eob);
  }
  return update;
}
#endif

// These numbers are empirically obtained.
static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
  { 17, 13 }, { 16, 10 },
};

void hbt_hash_init() {
  av1_crc_calculator_init(&crc_calculator, 16, 0x5D6DCB);   // ctx 16 bit hash
  av1_crc_calculator_init(&crc_calculator2, 16, 0x5D6DCB);  // qc 16 bit hash
  memset(hbt_hash_table, 0, sizeof(hbt_hash_table[0][0]) * 65536 * 16);
  hbt_hash_needs_init = 0;
}

int hbt_hash_miss(int found_index, uint16_t hbt_hash_index,
                  uint32_t hbt_hash_match, TxbInfo *txb_info,
                  const LV_MAP_COEFF_COST *txb_costs,
                  const LV_MAP_EOB_COST *txb_eob_costs,
                  const struct macroblock_plane *p, int block, int fast_mode) {
  const int16_t *scan = txb_info->scan_order->scan;

  av1_txb_init_levels(txb_info->qcoeff, txb_info->width, txb_info->height,
                      txb_info->levels);
  // The hash_based_trellis speed feature requires lv_map_multi, so always true.
  const int update =
      optimize_txb(txb_info, txb_costs, txb_eob_costs, NULL, 0, fast_mode);

  if (update) {
    // Overwrite old lowest entry
    hbt_hash_table[hbt_hash_index][found_index].hbt_hash_match = hbt_hash_match;
    hbt_hash_table[hbt_hash_index][found_index].hits = 1.0;
    for (int i = 0; i < txb_info->eob; i++) {
      hbt_hash_table[hbt_hash_index][found_index].opt_qcoeff[i] =
          txb_info->qcoeff[scan[i]];
    }
    for (int i = txb_info->eob; i < HBT_HASH_EOB; i++) {
      hbt_hash_table[hbt_hash_index][found_index].opt_qcoeff[i] = 0;
    }

    p->eobs[block] = txb_info->eob;
    p->txb_entropy_ctx[block] = av1_get_txb_entropy_context(
        txb_info->qcoeff, txb_info->scan_order, txb_info->eob);
  }
  return txb_info->eob;
}

int hbt_hash_hit(uint16_t hbt_hash_index, int found_index, TxbInfo *txb_info,
                 const struct macroblock_plane *p, int block) {
  const int16_t *scan = txb_info->scan_order->scan;
  int new_eob = 0;
  int update = 0;

  for (int i = 0; i < txb_info->eob; i++) {
    if (txb_info->qcoeff[scan[i]] !=
        hbt_hash_table[hbt_hash_index][found_index].opt_qcoeff[i]) {
      txb_info->qcoeff[scan[i]] =
          hbt_hash_table[hbt_hash_index][found_index].opt_qcoeff[i];
      update = 1;
      update_coeff(scan[i], txb_info->qcoeff[scan[i]], txb_info);
    }

    if (txb_info->qcoeff[scan[i]]) new_eob = i + 1;
  }

  if (update) {
    txb_info->eob = new_eob;
    p->eobs[block] = txb_info->eob;
    p->txb_entropy_ctx[block] = av1_get_txb_entropy_context(
        txb_info->qcoeff, txb_info->scan_order, txb_info->eob);
  }
  return txb_info->eob;
}

int search_hbt_hash_match(uint16_t hbt_hash_index, uint32_t hbt_hash_match,
                          TxbInfo *txb_info, const LV_MAP_COEFF_COST *txb_costs,
                          const LV_MAP_EOB_COST *txb_eob_costs,
                          const struct macroblock_plane *p, int block,
                          int fast_mode) {
  // Decay all hits
  double lowest_hits = 1.0;
  int lowest_index = 0;

  for (int i = 0; i < 16; i++) {
    hbt_hash_table[hbt_hash_index][i].hits *= 31.0;
    hbt_hash_table[hbt_hash_index][i].hits /= 32.0;

    if (hbt_hash_table[hbt_hash_index][i].hits < lowest_hits) {
      lowest_hits = hbt_hash_table[hbt_hash_index][i].hits;
      lowest_index = i;
    }
  }

  // Search soft hash vector for qcoeff match
  int found_index = -1;
  for (int i = 0; i < 16; i++) {  // OptTxbQcoeff array has fixed size of 16.
    if (hbt_hash_table[hbt_hash_index][i].hbt_hash_match == hbt_hash_match) {
      found_index = i;
      hbt_hash_table[hbt_hash_index][i].hits += 1.0;
      break;  // Found a match and it's at found_index
    }
  }

  if (found_index == -1) {  // Add new OptTxbQcoeff into array.
    return hbt_hash_miss(lowest_index, hbt_hash_index, hbt_hash_match, txb_info,
                         txb_costs, txb_eob_costs, p, block, fast_mode);
  } else {  // Retrieve data from array.
    return hbt_hash_hit(hbt_hash_index, found_index, txb_info, p, block);
  }
}

int hash_based_trellis_mode(TxbInfo *txb_info,
                            const LV_MAP_COEFF_COST *txb_costs,
                            const LV_MAP_EOB_COST *txb_eob_costs,
                            const struct macroblock_plane *p, int block,
                            int fast_mode, TXB_CTX *txb_ctx) {
  // Initialize hash table if needed.
  if (hbt_hash_needs_init) {
    hbt_hash_init();
  }

  //// Hash creation
  // TODO(mfo): use exact length once input finalized
  uint8_t txb_hash_data[256];
  const int16_t *scan = txb_info->scan_order->scan;
  uint8_t chunk = 0;

  uint16_t ctx_hash = 0;
  uint32_t qc_hash = 0;

  int hash_data_index = 0;
  for (int i = 0; i < txb_info->eob; i++) {
    // Data softening: data from -3 -> 3 is left alone,
    // while 'large' data is put into buckets of 16s
    // Consider bucketing less than 16 down to 4 instead of 0
    // if(txb_info->qcoeff[scan[i]] < 4 && txb_info->qcoeff[scan[i]] > -4)
    chunk = (txb_info->qcoeff[scan[i]]) & 0xff;
    /*else if(txb_info->qcoeff[scan[i]] < 16 && txb_info->qcoeff[scan[i]] > -16)
      chunk = (txb_info->qcoeff[scan[i]]) & 0xfc; //
    else
      chunk = (txb_info->qcoeff[scan[i]]) & 0xf0; // greater than 16*/
    txb_hash_data[hash_data_index++] = chunk;

    chunk = ((txb_info->qcoeff[scan[i]]) & 0xff00) >> 8;
    txb_hash_data[hash_data_index++] = chunk;
  }
  assert(hash_data_index <= 256);
  // 16 bit
  qc_hash = av1_get_crc_value(&crc_calculator2, txb_hash_data, hash_data_index);

  hash_data_index = 0;
  // tcoeff
  for (int i = 0; i < txb_info->eob; i++) {
    chunk = (txb_info->tcoeff[scan[i]] - txb_info->dqcoeff[scan[i]]) & 0xff;
    txb_hash_data[hash_data_index++] = chunk;
  }
  // txb_ctx
  chunk = txb_ctx->txb_skip_ctx & 0xff;
  txb_hash_data[hash_data_index++] = chunk;
  chunk = txb_ctx->dc_sign_ctx & 0xff;
  txb_hash_data[hash_data_index++] = chunk;
  // dequant
  chunk = txb_info->dequant[0] & 0xff;
  txb_hash_data[hash_data_index++] = chunk;
  chunk = (txb_info->dequant[0] & 0xff00) >> 8;
  txb_hash_data[hash_data_index++] = chunk;
  chunk = txb_info->dequant[1] & 0xff;
  txb_hash_data[hash_data_index++] = chunk;
  chunk = (txb_info->dequant[1] & 0xff00) >> 8;
  txb_hash_data[hash_data_index++] = chunk;
  // txb_skip_cost
  /*for (int i = 0; i < 2; i++) {
    for (int j = 0; j < TXB_SKIP_CONTEXTS; j++) {
      chunk = (txb_costs->txb_skip_cost[j][i] & 0xff00) >> 8;
      txb_hash_data[hash_data_index++] = chunk;
    }
  }
  // base_eob_cost
  for (int i = 1; i < 3; i++) {  // i = 0 are softened away
    for (int j = 0; j < SIG_COEF_CONTEXTS_EOB; j++) {
      chunk = (txb_costs->base_eob_cost[j][i] & 0xff00) >> 8;
      txb_hash_data[hash_data_index++] = chunk;
    }
  }*/
  assert(hash_data_index <= 256);
  // Gives 16 bit hash for ctx
  ctx_hash = av1_get_crc_value(&crc_calculator, txb_hash_data, hash_data_index);

  uint16_t hbt_hash_index = ctx_hash;  // 16 bit ctx_hash: index to table
  uint32_t hbt_hash_match = qc_hash;   // 16 bit qc_hash: matched in array
  //// End hash creation

  return search_hbt_hash_match(hbt_hash_index, hbt_hash_match, txb_info,
                               txb_costs, txb_eob_costs, p, block, fast_mode);
}

int av1_optimize_txb(const AV1_COMP *const cpi, MACROBLOCK *x, int plane,
                     int blk_row, int blk_col, int block, TX_SIZE tx_size,
                     TXB_CTX *txb_ctx, int fast_mode) {
  const AV1_COMMON *cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  const PLANE_TYPE plane_type = get_plane_type(plane);
  const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
  const TX_TYPE tx_type =
      av1_get_tx_type(plane_type, xd, blk_row, blk_col, tx_size);
  const MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  const struct macroblock_plane *p = &x->plane[plane];
  struct macroblockd_plane *pd = &xd->plane[plane];
  const int eob = p->eobs[block];
  tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
  const tran_low_t *tcoeff = BLOCK_OFFSET(p->coeff, block);
  const int16_t *dequant = p->dequant_QTX;
  const int seg_eob = av1_get_max_eob(tx_size);
  const int bwl = get_txb_bwl(tx_size);
  const int width = get_txb_wide(tx_size);
  const int height = get_txb_high(tx_size);
  const int is_inter = is_inter_block(mbmi);
  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
  const LV_MAP_COEFF_COST txb_costs = x->coeff_costs[txs_ctx][plane_type];
#if CONFIG_NEW_QUANT
  int dq = get_dq_profile(cm->dq_type, x->qindex, is_inter, plane_type);
  const dequant_val_type_nuq *dequant_val = p->dequant_val_nuq_QTX[dq];
#endif  // CONFIG_NEW_QUANT
  const int eob_multi_size = txsize_log2_minus4[tx_size];
  const LV_MAP_EOB_COST txb_eob_costs =
      x->eob_costs[eob_multi_size][plane_type];

#if CONFIG_DAALA_TX
  const int shift = 0;
#else
  const int shift = av1_get_tx_scale(tx_size);
#endif
  const int64_t rdmult =
      ((x->rdmult * plane_rd_mult[is_inter][plane_type] << (2 * (xd->bd - 8))) +
       2) >>
      2;
  uint8_t levels_buf[TX_PAD_2D];
  uint8_t *const levels = set_levels(levels_buf, width);

  assert(width == (1 << bwl));
  TxbInfo txb_info = {
    qcoeff,
    levels,
    dqcoeff,
    tcoeff,
    dequant,
#if CONFIG_NEW_QUANT
    dequant_val,
#endif  // CONFIG_NEW_QUANT
    shift,
    tx_size,
    txs_ctx,
    tx_type,
    bwl,
    width,
    height,
    eob,
    seg_eob,
    scan_order,
    txb_ctx,
    rdmult,
    &cm->coeff_ctx_table
  };

  // Hash based trellis (hbt) speed feature: avoid expensive optimize_txb calls
  // by storing the optimized coefficients in a hash table.
  // Currently disabled in speedfeatures.c
  if (eob <= HBT_HASH_EOB && eob > 0 && cpi->sf.use_hash_based_trellis) {
    return hash_based_trellis_mode(&txb_info, &txb_costs, &txb_eob_costs, p,
                                   block, fast_mode, txb_ctx);
  }

  av1_txb_init_levels(qcoeff, width, height, levels);

  const int update =
      optimize_txb(&txb_info, &txb_costs, &txb_eob_costs, NULL, 0, fast_mode);

  if (update) {
    p->eobs[block] = txb_info.eob;
    p->txb_entropy_ctx[block] =
        av1_get_txb_entropy_context(qcoeff, scan_order, txb_info.eob);
  }
  return txb_info.eob;
}

int av1_get_txb_entropy_context(const tran_low_t *qcoeff,
                                const SCAN_ORDER *scan_order, int eob) {
  const int16_t *const scan = scan_order->scan;
  int cul_level = 0;
  int c;

  if (eob == 0) return 0;
  for (c = 0; c < eob; ++c) {
    cul_level += abs(qcoeff[scan[c]]);
  }

  cul_level = AOMMIN(COEFF_CONTEXT_MASK, cul_level);
  set_dc_sign(&cul_level, qcoeff[0]);

  return cul_level;
}

void av1_update_txb_context_b(int plane, int block, int blk_row, int blk_col,
                              BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                              void *arg) {
  struct tokenize_b_args *const args = arg;
  const AV1_COMP *cpi = args->cpi;
  const AV1_COMMON *cm = &cpi->common;
  ThreadData *const td = args->td;
  MACROBLOCK *const x = &td->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  struct macroblock_plane *p = &x->plane[plane];
  struct macroblockd_plane *pd = &xd->plane[plane];
  const uint16_t eob = p->eobs[block];
  const tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  const PLANE_TYPE plane_type = pd->plane_type;
  const TX_TYPE tx_type =
      av1_get_tx_type(plane_type, xd, blk_row, blk_col, tx_size);
  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
  (void)plane_bsize;

  int cul_level = av1_get_txb_entropy_context(qcoeff, scan_order, eob);
  av1_set_contexts(xd, pd, plane, tx_size, cul_level, blk_col, blk_row);
}

void av1_update_and_record_txb_context(int plane, int block, int blk_row,
                                       int blk_col, BLOCK_SIZE plane_bsize,
                                       TX_SIZE tx_size, void *arg) {
  struct tokenize_b_args *const args = arg;
  const AV1_COMP *cpi = args->cpi;
  const AV1_COMMON *cm = &cpi->common;
  ThreadData *const td = args->td;
  MACROBLOCK *const x = &td->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  struct macroblock_plane *p = &x->plane[plane];
  struct macroblockd_plane *pd = &xd->plane[plane];
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  int eob = p->eobs[block], update_eob = -1;
  const PLANE_TYPE plane_type = pd->plane_type;
  const tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
  tran_low_t *tcoeff = BLOCK_OFFSET(x->mbmi_ext->tcoeff[plane], block);
  const int segment_id = mbmi->segment_id;
  const TX_TYPE tx_type =
      av1_get_tx_type(plane_type, xd, blk_row, blk_col, tx_size);
  const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, mbmi);
  const int16_t *const scan = scan_order->scan;
  const int seg_eob = av1_get_tx_eob(&cpi->common.seg, segment_id, tx_size);
  int c;
  TXB_CTX txb_ctx;
  get_txb_ctx(plane_bsize, tx_size, plane, pd->above_context + blk_col,
              pd->left_context + blk_row, &txb_ctx);
  const int bwl = get_txb_bwl(tx_size);
  const int width = get_txb_wide(tx_size);
  const int height = get_txb_high(tx_size);
  uint8_t levels_buf[TX_PAD_2D];
  uint8_t *const levels = set_levels(levels_buf, width);
  DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]);
  const uint8_t allow_update_cdf = args->allow_update_cdf;

  TX_SIZE txsize_ctx = get_txsize_entropy_ctx(tx_size);
  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
  DECLARE_ALIGNED(16, int8_t, coeff_contexts[MAX_TX_SQUARE]);

  memcpy(tcoeff, qcoeff, sizeof(*tcoeff) * seg_eob);

  ++td->counts->txb_skip[txsize_ctx][txb_ctx.txb_skip_ctx][eob == 0];
  if (allow_update_cdf)
    update_bin(ec_ctx->txb_skip_cdf[txsize_ctx][txb_ctx.txb_skip_ctx], eob == 0,
               2);
  x->mbmi_ext->txb_skip_ctx[plane][block] = txb_ctx.txb_skip_ctx;

  x->mbmi_ext->eobs[plane][block] = eob;

  if (eob == 0) {
    av1_set_contexts(xd, pd, plane, tx_size, 0, blk_col, blk_row);
    return;
  }

  av1_txb_init_levels(tcoeff, width, height, levels);

#if CONFIG_TXK_SEL
  av1_update_tx_type_count(cm, xd, blk_row, blk_col, plane, mbmi->sb_type,
                           tx_size, td->counts, allow_update_cdf);
#endif

  av1_update_eob_context(eob, seg_eob, tx_size, tx_type, plane_type, ec_ctx,
                         td->counts, allow_update_cdf);

  av1_get_nz_map_contexts(levels, scan, eob, tx_size, tx_type, coeff_contexts);

  update_eob = eob - 1;
  for (c = eob - 1; c >= 0; --c) {
    const int pos = scan[c];
    const int coeff_ctx = coeff_contexts[pos];
    const tran_low_t v = qcoeff[pos];
    const int is_nz = (v != 0);

    (void)is_nz;
    if (allow_update_cdf) {
      if (c == eob - 1) {
        assert(coeff_ctx < 4);
        update_cdf(
            ec_ctx->coeff_base_eob_cdf[txsize_ctx][plane_type][coeff_ctx],
            AOMMIN(abs(v), 3) - 1, 3);
      } else {
        update_cdf(ec_ctx->coeff_base_cdf[txsize_ctx][plane_type][coeff_ctx],
                   AOMMIN(abs(v), 3), 4);
      }
    }
    {
      if (c == eob - 1) {
        assert(coeff_ctx < 4);
        ++td->counts->coeff_base_eob_multi[txsize_ctx][plane_type][coeff_ctx]
                                          [AOMMIN(abs(v), 3) - 1];
      } else {
        ++td->counts->coeff_base_multi[txsize_ctx][plane_type][coeff_ctx]
                                      [AOMMIN(abs(v), 3)];
      }
    }
  }

  // Update the context needed to code the DC sign (if applicable)
  const int sign = (tcoeff[0] < 0) ? 1 : 0;
  if (tcoeff[0] != 0) {
    int dc_sign_ctx = txb_ctx.dc_sign_ctx;

    ++td->counts->dc_sign[plane_type][dc_sign_ctx][sign];
    if (allow_update_cdf)
      update_bin(ec_ctx->dc_sign_cdf[plane_type][dc_sign_ctx], sign, 2);
    x->mbmi_ext->dc_sign_ctx[plane][block] = dc_sign_ctx;
  }

  if (update_eob >= 0) {
    for (c = update_eob; c >= 0; --c) {
      const int pos = scan[c];
      const tran_low_t level = abs(tcoeff[pos]);
      int idx;
      int ctx;

      if (level <= NUM_BASE_LEVELS) continue;

      // level is above 1.

      const int base_range = level - 1 - NUM_BASE_LEVELS;
#if USE_CAUSAL_BR_CTX
      ctx = get_br_ctx(levels, pos, bwl, level_counts[pos], tx_type);
#else
      ctx = get_br_ctx(levels, pos, bwl, level_counts[pos]);
#endif
      for (idx = 0; idx < COEFF_BASE_RANGE; idx += BR_CDF_SIZE - 1) {
        const int k = AOMMIN(base_range - idx, BR_CDF_SIZE - 1);
        if (allow_update_cdf) {
          update_cdf(
#if 0
            ec_ctx->coeff_br_cdf[AOMMIN(txsize_ctx, TX_16X16)][plane_type][ctx],
#else
              ec_ctx
                  ->coeff_br_cdf[AOMMIN(txsize_ctx, TX_32X32)][plane_type][ctx],
#endif
              k, BR_CDF_SIZE);
        }
        for (int lps = 0; lps < BR_CDF_SIZE - 1; lps++) {
#if 0
          ++td->counts->coeff_lps[AOMMIN(txsize_ctx, TX_16X16)][plane_type][lps]
                                 [ctx][lps == k];
#else
          ++td->counts->coeff_lps[AOMMIN(txsize_ctx, TX_32X32)][plane_type][lps]
                                 [ctx][lps == k];
#endif
          if (lps == k) break;
        }
        ++td->counts->coeff_lps_multi[AOMMIN(txsize_ctx, TX_32X32)][plane_type]
                                     [ctx][k];

        if (k < BR_CDF_SIZE - 1) break;
      }
      // use 0-th order Golomb code to handle the residual level.
    }
  }

  int cul_level = av1_get_txb_entropy_context(tcoeff, scan_order, eob);
  av1_set_contexts(xd, pd, plane, tx_size, cul_level, blk_col, blk_row);
}

void av1_update_txb_context(const AV1_COMP *cpi, ThreadData *td,
                            RUN_TYPE dry_run, BLOCK_SIZE bsize, int *rate,
                            int mi_row, int mi_col, uint8_t allow_update_cdf) {
  MACROBLOCK *const x = &td->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  struct tokenize_b_args arg = { cpi, td, NULL, 0, allow_update_cdf };
  (void)rate;
  (void)mi_row;
  (void)mi_col;
  if (mbmi->skip) {
    av1_reset_skip_context(xd, mi_row, mi_col, bsize);
    return;
  }

  if (!dry_run) {
    av1_foreach_transformed_block(xd, bsize, mi_row, mi_col,
                                  av1_update_and_record_txb_context, &arg);
  } else if (dry_run == DRY_RUN_NORMAL) {
    av1_foreach_transformed_block(xd, bsize, mi_row, mi_col,
                                  av1_update_txb_context_b, &arg);
  } else {
    printf("DRY_RUN_COSTCOEFFS is not supported yet\n");
    assert(0);
  }
}

#if CONFIG_TXK_SEL
int64_t av1_search_txk_type(const AV1_COMP *cpi, MACROBLOCK *x, int plane,
                            int block, int blk_row, int blk_col,
                            BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
                            const ENTROPY_CONTEXT *a, const ENTROPY_CONTEXT *l,
                            int use_fast_coef_costing, RD_STATS *rd_stats) {
  const AV1_COMMON *cm = &cpi->common;
  MACROBLOCKD *xd = &x->e_mbd;
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
  TX_TYPE txk_start = DCT_DCT;
  TX_TYPE txk_end = TX_TYPES - 1;
  TX_TYPE best_tx_type = txk_start;
  int64_t best_rd = INT64_MAX;
  uint8_t best_txb_ctx = 0;
  uint16_t best_eob = 0;
  RD_STATS best_rd_stats;
  TX_TYPE tx_type;

  av1_invalid_rd_stats(&best_rd_stats);

  for (tx_type = txk_start; tx_type <= txk_end; ++tx_type) {
    if (plane == 0)
      mbmi->txk_type[(blk_row << MAX_MIB_SIZE_LOG2) + blk_col] = tx_type;
    TX_TYPE ref_tx_type =
        av1_get_tx_type(get_plane_type(plane), xd, blk_row, blk_col, tx_size);
    if (tx_type != ref_tx_type) {
      // use av1_get_tx_type() to check if the tx_type is valid for the current
      // mode if it's not, we skip it here.
      continue;
    }

    const int is_inter = is_inter_block(mbmi);
    const TxSetType tx_set_type = get_ext_tx_set_type(
        tx_size, mbmi->sb_type, is_inter, cm->reduced_tx_set_used);
    if (!av1_ext_tx_used[tx_set_type][tx_type]) continue;

    RD_STATS this_rd_stats;
    av1_invalid_rd_stats(&this_rd_stats);
    if (cpi->sf.optimize_coefficients != FULL_TRELLIS_OPT) {
      av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize,
                      tx_size, AV1_XFORM_QUANT_B);
    } else {
      av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize,
                      tx_size, AV1_XFORM_QUANT_FP);
      av1_optimize_b(cpi, x, plane, blk_row, blk_col, block, plane_bsize,
                     tx_size, a, l, 1);
    }
    av1_dist_block(cpi, x, plane, plane_bsize, block, blk_row, blk_col, tx_size,
                   &this_rd_stats.dist, &this_rd_stats.sse,
                   OUTPUT_HAS_PREDICTED_PIXELS);
    const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, mbmi);
    this_rd_stats.rate =
        av1_cost_coeffs(cpi, x, plane, blk_row, blk_col, block, tx_size,
                        scan_order, a, l, use_fast_coef_costing);
    int64_t rd = RDCOST(x->rdmult, this_rd_stats.rate, this_rd_stats.dist);

    if (rd < best_rd) {
      best_rd = rd;
      best_rd_stats = this_rd_stats;
      best_tx_type = tx_type;
      best_txb_ctx = x->plane[plane].txb_entropy_ctx[block];
      best_eob = x->plane[plane].eobs[block];
    }
  }

  av1_merge_rd_stats(rd_stats, &best_rd_stats);

  if (best_eob == 0) best_tx_type = DCT_DCT;

  if (plane == 0)
    mbmi->txk_type[(blk_row << MAX_MIB_SIZE_LOG2) + blk_col] = best_tx_type;
  x->plane[plane].txb_entropy_ctx[block] = best_txb_ctx;
  x->plane[plane].eobs[block] = best_eob;

  if (!is_inter_block(mbmi)) {
    // intra mode needs decoded result such that the next transform block
    // can use it for prediction.
    if (cpi->sf.optimize_coefficients != FULL_TRELLIS_OPT) {
      av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize,
                      tx_size, AV1_XFORM_QUANT_B);
    } else {
      av1_xform_quant(cm, x, plane, block, blk_row, blk_col, plane_bsize,
                      tx_size, AV1_XFORM_QUANT_FP);
      av1_optimize_b(cpi, x, plane, blk_row, blk_col, block, plane_bsize,
                     tx_size, a, l, 1);
    }

    av1_inverse_transform_block_facade(xd, plane, block, blk_row, blk_col,
                                       x->plane[plane].eobs[block],
                                       cm->reduced_tx_set_used);
  }
  return best_rd;
}
#endif  // CONFIG_TXK_SEL