decodeframe.c

/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <assert.h>
#include <stdlib.h>  // qsort()

#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
#include "./aom_scale_rtcd.h"
#include "./av1_rtcd.h"

#include "aom/aom_codec.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/binary_codes_reader.h"
#include "aom_dsp/bitreader.h"
#include "aom_dsp/bitreader_buffer.h"
#include "aom_mem/aom_mem.h"
#include "aom_ports/aom_timer.h"
#include "aom_ports/mem.h"
#include "aom_ports/mem_ops.h"
#include "aom_scale/aom_scale.h"
#include "aom_util/aom_thread.h"

#if CONFIG_BITSTREAM_DEBUG
#include "aom_util/debug_util.h"
#endif  // CONFIG_BITSTREAM_DEBUG

#include "av1/common/alloccommon.h"
#include "av1/common/cdef.h"
#if CONFIG_INSPECTION
#include "av1/decoder/inspection.h"
#endif
#include "av1/common/common.h"
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/entropymv.h"
#include "av1/common/idct.h"
#include "av1/common/mvref_common.h"
#include "av1/common/pred_common.h"
#include "av1/common/quant_common.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"
#if CONFIG_FRAME_SUPERRES
#include "av1/common/resize.h"
#endif  // CONFIG_FRAME_SUPERRES
#include "av1/common/seg_common.h"
#include "av1/common/thread_common.h"
#include "av1/common/tile_common.h"

#include "av1/decoder/decodeframe.h"
#include "av1/decoder/decodemv.h"
#include "av1/decoder/decoder.h"
#if CONFIG_LV_MAP
#include "av1/decoder/decodetxb.h"
#endif
#include "av1/decoder/detokenize.h"
#include "av1/decoder/dsubexp.h"
#include "av1/decoder/symbolrate.h"

#include "av1/common/warped_motion.h"

#define MAX_AV1_HEADER_SIZE 80
#define ACCT_STR __func__

#if CONFIG_CFL
#include "av1/common/cfl.h"
#endif

#if CONFIG_STRIPED_LOOP_RESTORATION && !CONFIG_LOOP_RESTORATION
#error "striped_loop_restoration requires loop_restoration"
#endif

#if CONFIG_LOOP_RESTORATION
static void loop_restoration_read_sb_coeffs(const AV1_COMMON *const cm,
                                            MACROBLOCKD *xd,
                                            aom_reader *const r, int plane,
                                            int rtile_idx);
#endif

static struct aom_read_bit_buffer *init_read_bit_buffer(
    AV1Decoder *pbi, struct aom_read_bit_buffer *rb, const uint8_t *data,
    const uint8_t *data_end, uint8_t clear_data[MAX_AV1_HEADER_SIZE]);
static int read_compressed_header(AV1Decoder *pbi, const uint8_t *data,
                                  size_t partition_size);
static size_t read_uncompressed_header(AV1Decoder *pbi,
                                       struct aom_read_bit_buffer *rb);

static int is_compound_reference_allowed(const AV1_COMMON *cm) {
  return !frame_is_intra_only(cm);
}

static void setup_compound_reference_mode(AV1_COMMON *cm) {
  cm->comp_fwd_ref[0] = LAST_FRAME;
  cm->comp_fwd_ref[1] = LAST2_FRAME;
  cm->comp_fwd_ref[2] = LAST3_FRAME;
  cm->comp_fwd_ref[3] = GOLDEN_FRAME;

  cm->comp_bwd_ref[0] = BWDREF_FRAME;
  cm->comp_bwd_ref[1] = ALTREF2_FRAME;
  cm->comp_bwd_ref[2] = ALTREF_FRAME;
}

static int read_is_valid(const uint8_t *start, size_t len, const uint8_t *end) {
  return len != 0 && len <= (size_t)(end - start);
}

static int decode_unsigned_max(struct aom_read_bit_buffer *rb, int max) {
  const int data = aom_rb_read_literal(rb, get_unsigned_bits(max));
  return data > max ? max : data;
}

#if CONFIG_SIMPLIFY_TX_MODE
static TX_MODE read_tx_mode(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
  if (cm->all_lossless) return ONLY_4X4;
#if CONFIG_VAR_TX_NO_TX_MODE
  (void)rb;
  return TX_MODE_SELECT;
#else
  return aom_rb_read_bit(rb) ? TX_MODE_SELECT : TX_MODE_LARGEST;
#endif  // CONFIG_VAR_TX_NO_TX_MODE
}
#else
static TX_MODE read_tx_mode(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
#if CONFIG_TX64X64
  TX_MODE tx_mode;
#endif
  if (cm->all_lossless) return ONLY_4X4;
#if CONFIG_VAR_TX_NO_TX_MODE
  (void)rb;
  return TX_MODE_SELECT;
#else
#if CONFIG_TX64X64
  tx_mode = aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
  if (tx_mode == ALLOW_32X32) tx_mode += aom_rb_read_bit(rb);
  return tx_mode;
#else
  return aom_rb_read_bit(rb) ? TX_MODE_SELECT : aom_rb_read_literal(rb, 2);
#endif  // CONFIG_TX64X64
#endif  // CONFIG_VAR_TX_NO_TX_MODE
}
#endif  // CONFIG_SIMPLIFY_TX_MODE

static REFERENCE_MODE read_frame_reference_mode(
    const AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
  if (is_compound_reference_allowed(cm)) {
#if CONFIG_REF_ADAPT
    return aom_rb_read_bit(rb) ? REFERENCE_MODE_SELECT : SINGLE_REFERENCE;
#else
    return aom_rb_read_bit(rb)
               ? REFERENCE_MODE_SELECT
               : (aom_rb_read_bit(rb) ? COMPOUND_REFERENCE : SINGLE_REFERENCE);
#endif  // CONFIG_REF_ADAPT
  } else {
    return SINGLE_REFERENCE;
  }
}

static void inverse_transform_block(MACROBLOCKD *xd, int plane,
                                    const TX_TYPE tx_type,
                                    const TX_SIZE tx_size, uint8_t *dst,
                                    int stride, int16_t scan_line, int eob) {
  struct macroblockd_plane *const pd = &xd->plane[plane];
  tran_low_t *const dqcoeff = pd->dqcoeff;
  av1_inverse_transform_block(xd, dqcoeff,
#if CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
                              xd->mrc_mask,
#endif  // CONFIG_MRC_TX && SIGNAL_ANY_MRC_MASK
                              plane, tx_type, tx_size, dst, stride, eob);
  memset(dqcoeff, 0, (scan_line + 1) * sizeof(dqcoeff[0]));
}

static int get_block_idx(const MACROBLOCKD *xd, int plane, int row, int col) {
  const int bsize = xd->mi[0]->mbmi.sb_type;
  const struct macroblockd_plane *pd = &xd->plane[plane];
  const BLOCK_SIZE plane_bsize =
      AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
  const TX_SIZE tx_size = av1_get_tx_size(plane, xd);
  const uint8_t txh_unit = tx_size_high_unit[tx_size];
  return row * max_blocks_wide + col * txh_unit;
}

static void predict_and_reconstruct_intra_block(
    AV1_COMMON *cm, MACROBLOCKD *const xd, aom_reader *const r,
    MB_MODE_INFO *const mbmi, int plane, int row, int col, TX_SIZE tx_size) {
  PLANE_TYPE plane_type = get_plane_type(plane);
  const int block_idx = get_block_idx(xd, plane, row, col);
  av1_predict_intra_block_facade(cm, xd, plane, block_idx, col, row, tx_size);

  if (!mbmi->skip) {
    struct macroblockd_plane *const pd = &xd->plane[plane];
#if TXCOEFF_TIMER
    struct aom_usec_timer timer;
    aom_usec_timer_start(&timer);
#endif
#if CONFIG_LV_MAP
    int16_t max_scan_line = 0;
    int eob;
    av1_read_coeffs_txb_facade(cm, xd, r, row, col, block_idx, plane,
                               pd->dqcoeff, tx_size, &max_scan_line, &eob);
    // tx_type will be read out in av1_read_coeffs_txb_facade
    const TX_TYPE tx_type =
        av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
#else   // CONFIG_LV_MAP
    const TX_TYPE tx_type =
        av1_get_tx_type(plane_type, xd, row, col, block_idx, tx_size);
    const SCAN_ORDER *scan_order = get_scan(cm, tx_size, tx_type, mbmi);
    int16_t max_scan_line = 0;
    const int eob =
        av1_decode_block_tokens(cm, xd, plane, scan_order, col, row, tx_size,
                                tx_type, &max_scan_line, r, mbmi->segment_id);
#endif  // CONFIG_LV_MAP

#if TXCOEFF_TIMER
    aom_usec_timer_mark(&timer);
    const int64_t elapsed_time = aom_usec_timer_elapsed(&timer);
    cm->txcoeff_timer += elapsed_time;
    ++cm->txb_count;
#endif
    if (eob) {
      uint8_t *dst =
          &pd->dst.buf[(row * pd->dst.stride + col) << tx_size_wide_log2[0]];
      inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
                              max_scan_line, eob);
    }
  }
#if CONFIG_CFL
  if (plane == AOM_PLANE_Y && xd->cfl->store_y) {
    cfl_store_tx(xd, row, col, tx_size, mbmi->sb_type);
  }
#endif  // CONFIG_CFL
}

static void decode_reconstruct_tx(AV1_COMMON *cm, MACROBLOCKD *const xd,
                                  aom_reader *r, MB_MODE_INFO *const mbmi,
                                  int plane, BLOCK_SIZE plane_bsize,
                                  int blk_row, int blk_col, int block,
                                  TX_SIZE tx_size, int *eob_total) {
  const struct macroblockd_plane *const pd = &xd->plane[plane];
  const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
  const int tx_row = blk_row >> (1 - pd->subsampling_y);
  const int tx_col = blk_col >> (1 - pd->subsampling_x);
  const TX_SIZE plane_tx_size =
      plane ? uv_txsize_lookup[bsize][mbmi->inter_tx_size[tx_row][tx_col]][0][0]
            : mbmi->inter_tx_size[tx_row][tx_col];
  // Scale to match transform block unit.
  const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
  const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);

  if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;

  if (tx_size == plane_tx_size) {
    PLANE_TYPE plane_type = get_plane_type(plane);
#if TXCOEFF_TIMER
    struct aom_usec_timer timer;
    aom_usec_timer_start(&timer);
#endif
#if CONFIG_LV_MAP
    int16_t max_scan_line = 0;
    int eob;
    av1_read_coeffs_txb_facade(cm, xd, r, blk_row, blk_col, block, plane,
                               pd->dqcoeff, tx_size, &max_scan_line, &eob);
    // tx_type will be read out in av1_read_coeffs_txb_facade
    const TX_TYPE tx_type =
        av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, plane_tx_size);
#else   // CONFIG_LV_MAP
    const TX_TYPE tx_type =
        av1_get_tx_type(plane_type, xd, blk_row, blk_col, block, plane_tx_size);
    const SCAN_ORDER *sc = get_scan(cm, plane_tx_size, tx_type, mbmi);
    int16_t max_scan_line = 0;
    const int eob = av1_decode_block_tokens(
        cm, xd, plane, sc, blk_col, blk_row, plane_tx_size, tx_type,
        &max_scan_line, r, mbmi->segment_id);
#endif  // CONFIG_LV_MAP

#if TXCOEFF_TIMER
    aom_usec_timer_mark(&timer);
    const int64_t elapsed_time = aom_usec_timer_elapsed(&timer);
    cm->txcoeff_timer += elapsed_time;
    ++cm->txb_count;
#endif

    inverse_transform_block(xd, plane, tx_type, plane_tx_size,
                            &pd->dst.buf[(blk_row * pd->dst.stride + blk_col)
                                         << tx_size_wide_log2[0]],
                            pd->dst.stride, max_scan_line, eob);
    *eob_total += eob;
  } else {
    const TX_SIZE sub_txs = sub_tx_size_map[tx_size];
    assert(IMPLIES(tx_size <= TX_4X4, sub_txs == tx_size));
    assert(IMPLIES(tx_size > TX_4X4, sub_txs < tx_size));
    const int bsw = tx_size_wide_unit[sub_txs];
    const int bsh = tx_size_high_unit[sub_txs];
    const int sub_step = bsw * bsh;

    assert(bsw > 0 && bsh > 0);

    for (int row = 0; row < tx_size_high_unit[tx_size]; row += bsh) {
      for (int col = 0; col < tx_size_wide_unit[tx_size]; col += bsw) {
        const int offsetr = blk_row + row;
        const int offsetc = blk_col + col;

        if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;

        decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, offsetr,
                              offsetc, block, sub_txs, eob_total);
        block += sub_step;
      }
    }
  }
}

static void set_offsets(AV1_COMMON *const cm, MACROBLOCKD *const xd,
                        BLOCK_SIZE bsize, int mi_row, int mi_col, int bw,
                        int bh, int x_mis, int y_mis) {
  const int offset = mi_row * cm->mi_stride + mi_col;
  int x, y;
  const TileInfo *const tile = &xd->tile;

  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = &cm->mi[offset];
  // TODO(slavarnway): Generate sb_type based on bwl and bhl, instead of
  // passing bsize from decode_partition().
  xd->mi[0]->mbmi.sb_type = bsize;
#if CONFIG_RD_DEBUG
  xd->mi[0]->mbmi.mi_row = mi_row;
  xd->mi[0]->mbmi.mi_col = mi_col;
#endif
#if CONFIG_CFL
  xd->cfl->mi_row = mi_row;
  xd->cfl->mi_col = mi_col;
#endif

  assert(x_mis && y_mis);
  for (x = 1; x < x_mis; ++x) xd->mi[x] = xd->mi[0];
  int idx = cm->mi_stride;
  for (y = 1; y < y_mis; ++y) {
    memcpy(&xd->mi[idx], &xd->mi[0], x_mis * sizeof(xd->mi[0]));
    idx += cm->mi_stride;
  }

  set_plane_n4(xd, bw, bh);
  set_skip_context(xd, mi_row, mi_col);

  // Distance of Mb to the various image edges. These are specified to 8th pel
  // as they are always compared to values that are in 1/8th pel units
  set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw,
#if CONFIG_DEPENDENT_HORZTILES
                 cm->dependent_horz_tiles,
#endif  // CONFIG_DEPENDENT_HORZTILES
                 cm->mi_rows, cm->mi_cols);

  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
                       mi_col);
}

static void decode_mbmi_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
                              int mi_row, int mi_col, aom_reader *r,
#if CONFIG_EXT_PARTITION_TYPES
                              PARTITION_TYPE partition,
#endif  // CONFIG_EXT_PARTITION_TYPES
                              BLOCK_SIZE bsize) {
  AV1_COMMON *const cm = &pbi->common;
  const int bw = mi_size_wide[bsize];
  const int bh = mi_size_high[bsize];
  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);

#if CONFIG_ACCOUNTING
  aom_accounting_set_context(&pbi->accounting, mi_col, mi_row);
#endif
  set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
#if CONFIG_EXT_PARTITION_TYPES
  xd->mi[0]->mbmi.partition = partition;
#endif
  av1_read_mode_info(pbi, xd, mi_row, mi_col, r, x_mis, y_mis);
  if (bsize >= BLOCK_8X8 && (cm->subsampling_x || cm->subsampling_y)) {
    const BLOCK_SIZE uv_subsize =
        ss_size_lookup[bsize][cm->subsampling_x][cm->subsampling_y];
    if (uv_subsize == BLOCK_INVALID)
      aom_internal_error(xd->error_info, AOM_CODEC_CORRUPT_FRAME,
                         "Invalid block size.");
  }

  int reader_corrupted_flag = aom_reader_has_error(r);
  aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
}

// Converts a Q3 quantizer lookup from static configuration to the
// actual TX scaling in use
static int dequant_Q3_to_QTX(int q3, int bd) {
  // Right now, TX scale in use is still Q3
  (void)bd;
  return q3;
}

static void decode_token_and_recon_block(AV1Decoder *const pbi,
                                         MACROBLOCKD *const xd, int mi_row,
                                         int mi_col, aom_reader *r,
                                         BLOCK_SIZE bsize) {
  AV1_COMMON *const cm = &pbi->common;
  const int bw = mi_size_wide[bsize];
  const int bh = mi_size_high[bsize];
  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);

  set_offsets(cm, xd, bsize, mi_row, mi_col, bw, bh, x_mis, y_mis);
  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
#if CONFIG_CFL
  CFL_CTX *const cfl = xd->cfl;
  cfl->is_chroma_reference = is_chroma_reference(
      mi_row, mi_col, bsize, cfl->subsampling_x, cfl->subsampling_y);
#endif  // CONFIG_CFL

  if (cm->delta_q_present_flag) {
    int i;
    for (i = 0; i < MAX_SEGMENTS; i++) {
#if CONFIG_EXT_DELTA_Q
      const int current_qindex =
#if CONFIG_Q_SEGMENTATION
          av1_get_qindex(&cm->seg, i, i, xd->current_qindex);
#else
          av1_get_qindex(&cm->seg, i, xd->current_qindex);
#endif
#else
      const int current_qindex = xd->current_qindex;
#endif  // CONFIG_EXT_DELTA_Q
      for (int j = 0; j < av1_num_planes(cm); ++j) {
        const int dc_delta_q =
            j == 0 ? cm->y_dc_delta_q
                   : (j == 1 ? cm->u_dc_delta_q : cm->v_dc_delta_q);
        const int ac_delta_q =
            j == 0 ? 0 : (j == 1 ? cm->u_ac_delta_q : cm->v_ac_delta_q);
        xd->plane[j].seg_dequant_QTX[i][0] = dequant_Q3_to_QTX(
            av1_dc_quant(current_qindex, dc_delta_q, cm->bit_depth),
            cm->bit_depth);
        xd->plane[j].seg_dequant_QTX[i][1] = dequant_Q3_to_QTX(
            av1_ac_quant(current_qindex, ac_delta_q, cm->bit_depth),
            cm->bit_depth);
      }
    }
  }
  if (mbmi->skip) av1_reset_skip_context(xd, mi_row, mi_col, bsize);

  if (!is_inter_block(mbmi)) {
    const int num_planes = av1_num_planes(cm);
    for (int plane = 0; plane < AOMMIN(2, num_planes); ++plane) {
      if (mbmi->palette_mode_info.palette_size[plane])
        av1_decode_palette_tokens(xd, plane, r);
    }

    for (int plane = 0; plane < num_planes; ++plane) {
      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 =
          AOMMAX(BLOCK_4X4, 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))
        continue;
      int blk_row, blk_col;
      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)
              predict_and_reconstruct_intra_block(cm, xd, r, mbmi, plane,
                                                  blk_row, blk_col, tx_size);
        }
      }
    }
  } else {
    int ref;

    for (ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
      const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
      if (frame < LAST_FRAME) {
#if CONFIG_INTRABC
        assert(is_intrabc_block(mbmi));
        assert(frame == INTRA_FRAME);
        assert(ref == 0);
#else
        assert(0);
#endif  // CONFIG_INTRABC
      } else {
        RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];

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

    av1_build_inter_predictors_sb(cm, xd, mi_row, mi_col, NULL, bsize);

    if (mbmi->motion_mode == OBMC_CAUSAL) {
#if CONFIG_NCOBMC
      av1_build_ncobmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
#else
      av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
#endif
    }
    // Reconstruction
    if (!mbmi->skip) {
      int eobtotal = 0;
      for (int plane = 0; plane < av1_num_planes(cm); ++plane) {
        const struct macroblockd_plane *const pd = &xd->plane[plane];
        const BLOCK_SIZE plane_bsize =
            AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
        const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
        const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
        int row, col;

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

        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);

        const TX_SIZE max_tx_size = get_vartx_max_txsize(
            mbmi, plane_bsize, pd->subsampling_x || pd->subsampling_y);
        const int bh_var_tx = tx_size_high_unit[max_tx_size];
        const int bw_var_tx = tx_size_wide_unit[max_tx_size];
        int block = 0;
        int step =
            tx_size_wide_unit[max_tx_size] * tx_size_high_unit[max_tx_size];

        for (row = 0; row < max_blocks_high; row += mu_blocks_high) {
          for (col = 0; col < max_blocks_wide; col += mu_blocks_wide) {
            int blk_row, blk_col;
            const int unit_height =
                AOMMIN(mu_blocks_high + row, max_blocks_high);
            const int unit_width =
                AOMMIN(mu_blocks_wide + col, max_blocks_wide);
            for (blk_row = row; blk_row < unit_height; blk_row += bh_var_tx) {
              for (blk_col = col; blk_col < unit_width; blk_col += bw_var_tx) {
                decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize,
                                      blk_row, blk_col, block, max_tx_size,
                                      &eobtotal);
                block += step;
              }
            }
          }
        }
      }
    }
  }
#if CONFIG_CFL
  if (mbmi->uv_mode != UV_CFL_PRED) {
    if (!cfl->is_chroma_reference && is_inter_block(mbmi)) {
      cfl_store_block(xd, mbmi->sb_type, mbmi->tx_size);
    }
  }
#endif  // CONFIG_CFL

  int reader_corrupted_flag = aom_reader_has_error(r);
  aom_merge_corrupted_flag(&xd->corrupted, reader_corrupted_flag);
}

#if NC_MODE_INFO
static void detoken_and_recon_sb(AV1Decoder *const pbi, MACROBLOCKD *const xd,
                                 int mi_row, int mi_col, aom_reader *r,
                                 BLOCK_SIZE bsize) {
  AV1_COMMON *const cm = &pbi->common;
  const int hbs = mi_size_wide[bsize] >> 1;
#if CONFIG_EXT_PARTITION_TYPES
  BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
#endif
  PARTITION_TYPE partition;
  BLOCK_SIZE subsize;
  const int has_rows = (mi_row + hbs) < cm->mi_rows;
  const int has_cols = (mi_col + hbs) < cm->mi_cols;

  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;

  partition = get_partition(cm, mi_row, mi_col, bsize);
  subsize = subsize_lookup[partition][bsize];

  switch (partition) {
    case PARTITION_NONE:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
      break;
    case PARTITION_HORZ:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
      if (has_rows)
        decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, subsize);
      break;
    case PARTITION_VERT:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
      if (has_cols)
        decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, subsize);
      break;
    case PARTITION_SPLIT:
      detoken_and_recon_sb(pbi, xd, mi_row, mi_col, r, subsize);
      detoken_and_recon_sb(pbi, xd, mi_row, mi_col + hbs, r, subsize);
      detoken_and_recon_sb(pbi, xd, mi_row + hbs, mi_col, r, subsize);
      detoken_and_recon_sb(pbi, xd, mi_row + hbs, mi_col + hbs, r, subsize);
      break;
#if CONFIG_EXT_PARTITION_TYPES
#if CONFIG_EXT_PARTITION_TYPES_AB
#error NC_MODE_INFO+MOTION_VAR not yet supported for new HORZ/VERT_AB partitions
#endif
    case PARTITION_HORZ_A:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize2);
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, bsize2);
      decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, subsize);
      break;
    case PARTITION_HORZ_B:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
      decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, bsize2);
      decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col + hbs, r,
                                   bsize2);
      break;
    case PARTITION_VERT_A:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize2);
      decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col, r, bsize2);
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, subsize);
      break;
    case PARTITION_VERT_B:
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, subsize);
      decode_token_and_recon_block(pbi, xd, mi_row, mi_col + hbs, r, bsize2);
      decode_token_and_recon_block(pbi, xd, mi_row + hbs, mi_col + hbs, r,
                                   bsize2);
      break;
#endif
    default: assert(0 && "Invalid partition type");
  }
}
#endif

static void decode_block(AV1Decoder *const pbi, MACROBLOCKD *const xd,
                         int mi_row, int mi_col, aom_reader *r,
#if CONFIG_EXT_PARTITION_TYPES
                         PARTITION_TYPE partition,
#endif  // CONFIG_EXT_PARTITION_TYPES
                         BLOCK_SIZE bsize) {
  decode_mbmi_block(pbi, xd, mi_row, mi_col, r,
#if CONFIG_EXT_PARTITION_TYPES
                    partition,
#endif
                    bsize);

#if !(NC_MODE_INFO)
  decode_token_and_recon_block(pbi, xd, mi_row, mi_col, r, bsize);
#endif
}

static PARTITION_TYPE read_partition(AV1_COMMON *cm, MACROBLOCKD *xd,
                                     int mi_row, int mi_col, aom_reader *r,
                                     int has_rows, int has_cols,
                                     BLOCK_SIZE bsize) {
  const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
  PARTITION_TYPE p;
  FRAME_CONTEXT *ec_ctx = xd->tile_ctx;
  (void)cm;

  aom_cdf_prob *partition_cdf = (ctx >= 0) ? ec_ctx->partition_cdf[ctx] : NULL;

  if (has_rows && has_cols) {
#if CONFIG_EXT_PARTITION_TYPES
    const int num_partition_types =
        (mi_width_log2_lookup[bsize] > mi_width_log2_lookup[BLOCK_8X8])
            ? EXT_PARTITION_TYPES
            : PARTITION_TYPES;
#else
    const int num_partition_types = PARTITION_TYPES;
#endif  // CONFIG_EXT_PARTITION_TYPES
    p = (PARTITION_TYPE)aom_read_symbol(r, partition_cdf, num_partition_types,
                                        ACCT_STR);
  } else if (!has_rows && has_cols) {
    assert(bsize > BLOCK_8X8);
    aom_cdf_prob cdf[2];
    partition_gather_vert_alike(cdf, partition_cdf);
    assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP));
    p = aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_HORZ;
    // gather cols
  } else if (has_rows && !has_cols) {
    assert(bsize > BLOCK_8X8);
    aom_cdf_prob cdf[2];
    partition_gather_horz_alike(cdf, partition_cdf);
    assert(cdf[1] == AOM_ICDF(CDF_PROB_TOP));
    p = aom_read_cdf(r, cdf, 2, ACCT_STR) ? PARTITION_SPLIT : PARTITION_VERT;
  } else {
    p = PARTITION_SPLIT;
  }

  return p;
}

// TODO(slavarnway): eliminate bsize and subsize in future commits
static void decode_partition(AV1Decoder *const pbi, MACROBLOCKD *const xd,
                             int mi_row, int mi_col, aom_reader *r,
                             BLOCK_SIZE bsize) {
  AV1_COMMON *const cm = &pbi->common;
  const int num_8x8_wh = mi_size_wide[bsize];
  const int hbs = num_8x8_wh >> 1;
#if CONFIG_EXT_PARTITION_TYPES && CONFIG_EXT_PARTITION_TYPES_AB
  const int qbs = num_8x8_wh >> 2;
#endif
  PARTITION_TYPE partition;
  BLOCK_SIZE subsize;
#if CONFIG_EXT_PARTITION_TYPES
  const int quarter_step = num_8x8_wh / 4;
  int i;
#if !CONFIG_EXT_PARTITION_TYPES_AB
  BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
#endif
#endif
  const int has_rows = (mi_row + hbs) < cm->mi_rows;
  const int has_cols = (mi_col + hbs) < cm->mi_cols;

  if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;

  partition = (bsize < BLOCK_8X8) ? PARTITION_NONE
                                  : read_partition(cm, xd, mi_row, mi_col, r,
                                                   has_rows, has_cols, bsize);
  subsize = subsize_lookup[partition][bsize];  // get_subsize(bsize, partition);

  // Check the bitstream is conformant: if there is subsampling on the
  // chroma planes, subsize must subsample to a valid block size.
  const struct macroblockd_plane *const pd_u = &xd->plane[1];
  if (get_plane_block_size(subsize, pd_u) == BLOCK_INVALID) {
    aom_internal_error(&cm->error, AOM_CODEC_CORRUPT_FRAME,
                       "Block size %dx%d invalid with this subsampling mode",
                       block_size_wide[subsize], block_size_high[subsize]);
  }

#define DEC_BLOCK_STX_ARG
#if CONFIG_EXT_PARTITION_TYPES
#define DEC_BLOCK_EPT_ARG partition,
#else
#define DEC_BLOCK_EPT_ARG
#endif
#define DEC_BLOCK(db_r, db_c, db_subsize)                   \
  decode_block(pbi, xd, DEC_BLOCK_STX_ARG(db_r), (db_c), r, \
               DEC_BLOCK_EPT_ARG(db_subsize))
#define DEC_PARTITION(db_r, db_c, db_subsize) \
  decode_partition(pbi, xd, DEC_BLOCK_STX_ARG(db_r), (db_c), r, (db_subsize))

  switch (partition) {
    case PARTITION_NONE: DEC_BLOCK(mi_row, mi_col, subsize); break;
    case PARTITION_HORZ:
      DEC_BLOCK(mi_row, mi_col, subsize);
      if (has_rows) DEC_BLOCK(mi_row + hbs, mi_col, subsize);
      break;
    case PARTITION_VERT:
      DEC_BLOCK(mi_row, mi_col, subsize);
      if (has_cols) DEC_BLOCK(mi_row, mi_col + hbs, subsize);
      break;
    case PARTITION_SPLIT:
      DEC_PARTITION(mi_row, mi_col, subsize);
      DEC_PARTITION(mi_row, mi_col + hbs, subsize);
      DEC_PARTITION(mi_row + hbs, mi_col, subsize);
      DEC_PARTITION(mi_row + hbs, mi_col + hbs, subsize);
      break;
#if CONFIG_EXT_PARTITION_TYPES
#if CONFIG_EXT_PARTITION_TYPES_AB
    case PARTITION_HORZ_A:
      DEC_BLOCK(mi_row, mi_col, get_subsize(bsize, PARTITION_HORZ_4));
      DEC_BLOCK(mi_row + qbs, mi_col, get_subsize(bsize, PARTITION_HORZ_4));
      DEC_BLOCK(mi_row + hbs, mi_col, subsize);
      break;
    case PARTITION_HORZ_B:
      DEC_BLOCK(mi_row, mi_col, subsize);
      DEC_BLOCK(mi_row + hbs, mi_col, get_subsize(bsize, PARTITION_HORZ_4));
      if (mi_row + 3 * qbs < cm->mi_rows)
        DEC_BLOCK(mi_row + 3 * qbs, mi_col,
                  get_subsize(bsize, PARTITION_HORZ_4));
      break;
    case PARTITION_VERT_A:
      DEC_BLOCK(mi_row, mi_col, get_subsize(bsize, PARTITION_VERT_4));
      DEC_BLOCK(mi_row, mi_col + qbs, get_subsize(bsize, PARTITION_VERT_4));
      DEC_BLOCK(mi_row, mi_col + hbs, subsize);
      break;
    case PARTITION_VERT_B:
      DEC_BLOCK(mi_row, mi_col, subsize);
      DEC_BLOCK(mi_row, mi_col + hbs, get_subsize(bsize, PARTITION_VERT_4));
      if (mi_col + 3 * qbs < cm->mi_cols)
        DEC_BLOCK(mi_row, mi_col + 3 * qbs,
                  get_subsize(bsize, PARTITION_VERT_4));
      break;
#else
    case PARTITION_HORZ_A:
      DEC_BLOCK(mi_row, mi_col, bsize2);
      DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
      DEC_BLOCK(mi_row + hbs, mi_col, subsize);
      break;
    case PARTITION_HORZ_B:
      DEC_BLOCK(mi_row, mi_col, subsize);
      DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
      DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
      break;
    case PARTITION_VERT_A:
      DEC_BLOCK(mi_row, mi_col, bsize2);
      DEC_BLOCK(mi_row + hbs, mi_col, bsize2);
      DEC_BLOCK(mi_row, mi_col + hbs, subsize);
      break;
    case PARTITION_VERT_B:
      DEC_BLOCK(mi_row, mi_col, subsize);
      DEC_BLOCK(mi_row, mi_col + hbs, bsize2);
      DEC_BLOCK(mi_row + hbs, mi_col + hbs, bsize2);
      break;
#endif
    case PARTITION_HORZ_4:
      for (i = 0; i < 4; ++i) {
        int this_mi_row = mi_row + i * quarter_step;
        if (i > 0 && this_mi_row >= cm->mi_rows) break;
        DEC_BLOCK(this_mi_row, mi_col, subsize);
      }
      break;
    case PARTITION_VERT_4:
      for (i = 0; i < 4; ++i) {
        int this_mi_col = mi_col + i * quarter_step;
        if (i > 0 && this_mi_col >= cm->mi_cols) break;
        DEC_BLOCK(mi_row, this_mi_col, subsize);
      }
      break;
#endif  // CONFIG_EXT_PARTITION_TYPES
    default: assert(0 && "Invalid partition type");
  }

#undef DEC_PARTITION
#undef DEC_BLOCK
#undef DEC_BLOCK_EPT_ARG
#undef DEC_BLOCK_STX_ARG

#if CONFIG_EXT_PARTITION_TYPES
  update_ext_partition_context(xd, mi_row, mi_col, subsize, bsize, partition);
#else
  // update partition context
  if (bsize >= BLOCK_8X8 &&
      (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
    update_partition_context(xd, mi_row, mi_col, subsize, bsize);
#endif  // CONFIG_EXT_PARTITION_TYPES

#if CONFIG_LPF_SB
  if (bsize == cm->sb_size) {
    int filt_lvl;
    if (mi_row == 0 && mi_col == 0) {
      filt_lvl = aom_read_literal(r, 6, ACCT_STR);
      cm->mi_grid_visible[0]->mbmi.reuse_sb_lvl = 0;
      cm->mi_grid_visible[0]->mbmi.delta = 0;
      cm->mi_grid_visible[0]->mbmi.sign = 0;
    } else {
      int prev_mi_row, prev_mi_col;
      if (mi_col - MAX_MIB_SIZE < 0) {
        prev_mi_row = mi_row - MAX_MIB_SIZE;
        prev_mi_col = mi_col;
      } else {
        prev_mi_row = mi_row;
        prev_mi_col = mi_col - MAX_MIB_SIZE;
      }

      MB_MODE_INFO *curr_mbmi =
          &cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col]->mbmi;
      MB_MODE_INFO *prev_mbmi =
          &cm->mi_grid_visible[prev_mi_row * cm->mi_stride + prev_mi_col]->mbmi;
      const uint8_t prev_lvl = prev_mbmi->filt_lvl;

      const int reuse_ctx = prev_mbmi->reuse_sb_lvl;
      const int reuse_prev_lvl = aom_read_symbol(
          r, xd->tile_ctx->lpf_reuse_cdf[reuse_ctx], 2, ACCT_STR);
      curr_mbmi->reuse_sb_lvl = reuse_prev_lvl;

      if (reuse_prev_lvl) {
        filt_lvl = prev_lvl;
        curr_mbmi->delta = 0;
        curr_mbmi->sign = 0;
      } else {
        const int delta_ctx = prev_mbmi->delta;
        unsigned int delta = aom_read_symbol(
            r, xd->tile_ctx->lpf_delta_cdf[delta_ctx], DELTA_RANGE, ACCT_STR);
        curr_mbmi->delta = delta;
        delta *= LPF_STEP;

        if (delta) {