blockd.h

/*
 * 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.
 */

#ifndef AV1_COMMON_BLOCKD_H_
#define AV1_COMMON_BLOCKD_H_

#include "./aom_config.h"

#include "aom_dsp/aom_dsp_common.h"
#include "aom_ports/mem.h"
#include "aom_scale/yv12config.h"

#include "av1/common/common_data.h"
#include "av1/common/quant_common.h"
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/mv.h"
#include "av1/common/scale.h"
#include "av1/common/seg_common.h"
#include "av1/common/tile_common.h"

#ifdef __cplusplus
extern "C" {
#endif

#define MAX_MB_PLANE 3

// Set COMPOUND_SEGMENT_TYPE to one of the three
// 0: Uniform
// 1: Difference weighted
#define COMPOUND_SEGMENT_TYPE 1
#define MAX_SEG_MASK_BITS 1

// Disables vartx transform split for chroma
// There are two possible options:
// 1: Always choose chroma tx_size to be the largest allowed for the PU
// 2: If the luma tranform size is split at least one level, split the
// chroma by one level, otherwise use the largest allowed transform size for
// chroma.
#define DISABLE_VARTX_FOR_CHROMA 1

// SEG_MASK_TYPES should not surpass 1 << MAX_SEG_MASK_BITS
typedef enum {
#if COMPOUND_SEGMENT_TYPE == 0
  UNIFORM_45 = 0,
  UNIFORM_45_INV,
#elif COMPOUND_SEGMENT_TYPE == 1
  DIFFWTD_38 = 0,
  DIFFWTD_38_INV,
#endif  // COMPOUND_SEGMENT_TYPE
  SEG_MASK_TYPES,
} SEG_MASK_TYPE;

typedef enum {
  KEY_FRAME = 0,
  INTER_FRAME = 1,
#if CONFIG_OBU
  INTRA_ONLY_FRAME = 2,  // replaces intra-only
  S_FRAME = 3,
#endif
  FRAME_TYPES,
} FRAME_TYPE;

static INLINE int is_comp_ref_allowed(BLOCK_SIZE bsize) {
  (void)bsize;
  return AOMMIN(block_size_wide[bsize], block_size_high[bsize]) >= 8;
}

static INLINE int is_inter_mode(PREDICTION_MODE mode) {
  return mode >= NEARESTMV && mode <= NEW_NEWMV;
}

typedef struct {
  uint8_t *plane[MAX_MB_PLANE];
  int stride[MAX_MB_PLANE];
} BUFFER_SET;

static INLINE int is_inter_singleref_mode(PREDICTION_MODE mode) {
  return mode >= NEARESTMV && mode <= NEWMV;
}
static INLINE int is_inter_compound_mode(PREDICTION_MODE mode) {
  return mode >= NEAREST_NEARESTMV && mode <= NEW_NEWMV;
}

static INLINE PREDICTION_MODE compound_ref0_mode(PREDICTION_MODE mode) {
  static PREDICTION_MODE lut[] = {
    MB_MODE_COUNT,  // DC_PRED
    MB_MODE_COUNT,  // V_PRED
    MB_MODE_COUNT,  // H_PRED
    MB_MODE_COUNT,  // D45_PRED
    MB_MODE_COUNT,  // D135_PRED
    MB_MODE_COUNT,  // D117_PRED
    MB_MODE_COUNT,  // D153_PRED
    MB_MODE_COUNT,  // D207_PRED
    MB_MODE_COUNT,  // D63_PRED
    MB_MODE_COUNT,  // SMOOTH_PRED
    MB_MODE_COUNT,  // SMOOTH_V_PRED
    MB_MODE_COUNT,  // SMOOTH_H_PRED
    MB_MODE_COUNT,  // PAETH_PRED
    MB_MODE_COUNT,  // NEARESTMV
    MB_MODE_COUNT,  // NEARMV
    MB_MODE_COUNT,  // GLOBALMV
    MB_MODE_COUNT,  // NEWMV
    NEARESTMV,      // NEAREST_NEARESTMV
    NEARMV,         // NEAR_NEARMV
    NEARESTMV,      // NEAREST_NEWMV
    NEWMV,          // NEW_NEARESTMV
    NEARMV,         // NEAR_NEWMV
    NEWMV,          // NEW_NEARMV
    GLOBALMV,       // GLOBAL_GLOBALMV
    NEWMV,          // NEW_NEWMV
  };
  assert(NELEMENTS(lut) == MB_MODE_COUNT);
  assert(is_inter_compound_mode(mode));
  return lut[mode];
}

static INLINE PREDICTION_MODE compound_ref1_mode(PREDICTION_MODE mode) {
  static PREDICTION_MODE lut[] = {
    MB_MODE_COUNT,  // DC_PRED
    MB_MODE_COUNT,  // V_PRED
    MB_MODE_COUNT,  // H_PRED
    MB_MODE_COUNT,  // D45_PRED
    MB_MODE_COUNT,  // D135_PRED
    MB_MODE_COUNT,  // D117_PRED
    MB_MODE_COUNT,  // D153_PRED
    MB_MODE_COUNT,  // D207_PRED
    MB_MODE_COUNT,  // D63_PRED
    MB_MODE_COUNT,  // SMOOTH_PRED
    MB_MODE_COUNT,  // SMOOTH_V_PRED
    MB_MODE_COUNT,  // SMOOTH_H_PRED
    MB_MODE_COUNT,  // PAETH_PRED
    MB_MODE_COUNT,  // NEARESTMV
    MB_MODE_COUNT,  // NEARMV
    MB_MODE_COUNT,  // GLOBALMV
    MB_MODE_COUNT,  // NEWMV
    NEARESTMV,      // NEAREST_NEARESTMV
    NEARMV,         // NEAR_NEARMV
    NEWMV,          // NEAREST_NEWMV
    NEARESTMV,      // NEW_NEARESTMV
    NEWMV,          // NEAR_NEWMV
    NEARMV,         // NEW_NEARMV
    GLOBALMV,       // GLOBAL_GLOBALMV
    NEWMV,          // NEW_NEWMV
  };
  assert(NELEMENTS(lut) == MB_MODE_COUNT);
  assert(is_inter_compound_mode(mode));
  return lut[mode];
}

static INLINE int have_nearmv_in_inter_mode(PREDICTION_MODE mode) {
  return (mode == NEARMV || mode == NEAR_NEARMV || mode == NEAR_NEWMV ||
          mode == NEW_NEARMV);
}

static INLINE int have_newmv_in_inter_mode(PREDICTION_MODE mode) {
  return (mode == NEWMV || mode == NEW_NEWMV || mode == NEAREST_NEWMV ||
          mode == NEW_NEARESTMV || mode == NEAR_NEWMV || mode == NEW_NEARMV);
}

static INLINE int use_masked_motion_search(COMPOUND_TYPE type) {
  return (type == COMPOUND_WEDGE);
}

static INLINE int is_masked_compound_type(COMPOUND_TYPE type) {
  return (type == COMPOUND_WEDGE || type == COMPOUND_SEG);
}

/* For keyframes, intra block modes are predicted by the (already decoded)
   modes for the Y blocks to the left and above us; for interframes, there
   is a single probability table. */

typedef struct {
  PREDICTION_MODE as_mode;
  int_mv as_mv[2];  // first, second inter predictor motion vectors
  int_mv pred_mv[2];
  int_mv ref_mv[2];
} b_mode_info;

typedef int8_t MV_REFERENCE_FRAME;

typedef struct {
  // Number of base colors for Y (0) and UV (1)
  uint8_t palette_size[2];
  // Value of base colors for Y, U, and V
  uint16_t palette_colors[3 * PALETTE_MAX_SIZE];
} PALETTE_MODE_INFO;

#if CONFIG_FILTER_INTRA
typedef struct {
  // 1: an ext intra mode is used; 0: otherwise.
  uint8_t use_filter_intra_mode[PLANE_TYPES];
  FILTER_INTRA_MODE filter_intra_mode[PLANE_TYPES];
} FILTER_INTRA_MODE_INFO;

static const PREDICTION_MODE fimode_to_intradir[FILTER_INTRA_MODES] = {
  DC_PRED, V_PRED, H_PRED, D153_PRED, DC_PRED
};
#endif  // CONFIG_FILTER_INTRA

#if CONFIG_RD_DEBUG
#define TXB_COEFF_COST_MAP_SIZE (2 * MAX_MIB_SIZE)
#endif

typedef struct RD_STATS {
  int rate;
  int64_t dist;
  // Please be careful of using rdcost, it's not guaranteed to be set all the
  // time.
  // TODO(angiebird): Create a set of functions to manipulate the RD_STATS. In
  // these functions, make sure rdcost is always up-to-date according to
  // rate/dist.
  int64_t rdcost;
  int64_t sse;
  int skip;  // sse should equal to dist when skip == 1
  int64_t ref_rdcost;
  int zero_rate;
  uint8_t invalid_rate;
#if CONFIG_RD_DEBUG
  int txb_coeff_cost[MAX_MB_PLANE];
  int txb_coeff_cost_map[MAX_MB_PLANE][TXB_COEFF_COST_MAP_SIZE]
                        [TXB_COEFF_COST_MAP_SIZE];
#endif  // CONFIG_RD_DEBUG
} RD_STATS;

// This struct is used to group function args that are commonly
// sent together in functions related to interinter compound modes
typedef struct {
  int wedge_index;
  int wedge_sign;
  SEG_MASK_TYPE mask_type;
  uint8_t *seg_mask;
  COMPOUND_TYPE interinter_compound_type;
} INTERINTER_COMPOUND_DATA;

// This structure now relates to 8x8 block regions.
typedef struct MB_MODE_INFO {
  // Common for both INTER and INTRA blocks
  BLOCK_SIZE sb_type;
  PREDICTION_MODE mode;
  TX_SIZE tx_size;
  // TODO(jingning): This effectively assigned a separate entry for each
  // 8x8 block. Apparently it takes much more space than needed.
  TX_SIZE inter_tx_size[MAX_MIB_SIZE][MAX_MIB_SIZE];
  TX_SIZE min_tx_size;
  int8_t skip;
#if CONFIG_EXT_SKIP
  int8_t skip_mode;
#endif  // CONFIG_EXT_SKIP
  int8_t segment_id;
  int8_t seg_id_predicted;  // valid only when temporal_update is enabled

  // Only for INTRA blocks
  UV_PREDICTION_MODE uv_mode;

  PALETTE_MODE_INFO palette_mode_info;
#if CONFIG_INTRABC
  uint8_t use_intrabc;
#endif  // CONFIG_INTRABC

  // Only for INTER blocks
  InterpFilters interp_filters;
  MV_REFERENCE_FRAME ref_frame[2];
  TX_TYPE tx_type;
#if CONFIG_TXK_SEL
  TX_TYPE txk_type[MAX_SB_SQUARE / (TX_SIZE_W_MIN * TX_SIZE_H_MIN)];
#endif

#if CONFIG_FILTER_INTRA
  FILTER_INTRA_MODE_INFO filter_intra_mode_info;
#endif  // CONFIG_FILTER_INTRA
#if CONFIG_EXT_INTRA
  // The actual prediction angle is the base angle + (angle_delta * step).
  int8_t angle_delta[2];
#endif  // CONFIG_EXT_INTRA

  // interintra members
  INTERINTRA_MODE interintra_mode;
  // TODO(debargha): Consolidate these flags
  int use_wedge_interintra;
  int interintra_wedge_index;
  int interintra_wedge_sign;
  // interinter members
  COMPOUND_TYPE interinter_compound_type;
  int wedge_index;
  int wedge_sign;
  SEG_MASK_TYPE mask_type;
  MOTION_MODE motion_mode;
#if CONFIG_EXT_WARPED_MOTION
  int wm_ctx;
#endif  // CONFIG_EXT_WARPED_MOTION
  int overlappable_neighbors[2];
  int_mv mv[2];
  int_mv pred_mv[2];
  uint8_t ref_mv_idx;
#if CONFIG_EXT_PARTITION_TYPES
  PARTITION_TYPE partition;
#endif
#if CONFIG_NEW_QUANT
  int dq_off_index;
  int send_dq_bit;
#endif  // CONFIG_NEW_QUANT
  /* deringing gain *per-superblock* */
  int8_t cdef_strength;
  int current_q_index;
#if CONFIG_EXT_DELTA_Q
  int current_delta_lf_from_base;
#if CONFIG_LOOPFILTER_LEVEL
  int curr_delta_lf[FRAME_LF_COUNT];
#endif  // CONFIG_LOOPFILTER_LEVEL
#endif
#if CONFIG_RD_DEBUG
  RD_STATS rd_stats;
  int mi_row;
  int mi_col;
#endif
  int num_proj_ref[2];
  WarpedMotionParams wm_params[2];

#if CONFIG_CFL
  // Index of the alpha Cb and alpha Cr combination
  int cfl_alpha_idx;
  // Joint sign of alpha Cb and alpha Cr
  int cfl_alpha_signs;
#endif

  BOUNDARY_TYPE boundary_info;
#if CONFIG_LPF_SB
#if CONFIG_LOOPFILTER_LEVEL
  // 0: y plane vert, 1: y plane horz, 2: u plane, 3: v plane
  uint8_t filt_lvl[4];
  int reuse_sb_lvl[4];
  int sign[4];
  int delta[4];
#else
  uint8_t filt_lvl;
  int reuse_sb_lvl;
  int sign;
  int delta;
#endif  // CONFIG_LOOPFILTER_LEVEL
#endif  // CONFIG_LPF_SB

#if CONFIG_JNT_COMP
  int compound_idx;
  int comp_group_idx;
#endif
} MB_MODE_INFO;

typedef struct MODE_INFO {
  MB_MODE_INFO mbmi;
  b_mode_info bmi[4];
} MODE_INFO;

#if CONFIG_INTRABC
#define NO_FILTER_FOR_IBC 1  // Disable in-loop filters for frame with intrabc

static INLINE int is_intrabc_block(const MB_MODE_INFO *mbmi) {
  return mbmi->use_intrabc;
}
#endif

#if CONFIG_CFL
static INLINE PREDICTION_MODE get_uv_mode(UV_PREDICTION_MODE mode) {
  assert(mode < UV_INTRA_MODES);
  static const PREDICTION_MODE uv2y[UV_INTRA_MODES] = {
    DC_PRED,        // UV_DC_PRED
    V_PRED,         // UV_V_PRED
    H_PRED,         // UV_H_PRED
    D45_PRED,       // UV_D45_PRED
    D135_PRED,      // UV_D135_PRED
    D117_PRED,      // UV_D117_PRED
    D153_PRED,      // UV_D153_PRED
    D207_PRED,      // UV_D207_PRED
    D63_PRED,       // UV_D63_PRED
    SMOOTH_PRED,    // UV_SMOOTH_PRED
    SMOOTH_V_PRED,  // UV_SMOOTH_V_PRED
    SMOOTH_H_PRED,  // UV_SMOOTH_H_PRED
    PAETH_PRED,     // UV_PAETH_PRED
    DC_PRED,        // CFL_PRED
  };
  return uv2y[mode];
}
#else
static INLINE PREDICTION_MODE get_uv_mode(PREDICTION_MODE mode) { return mode; }
#endif  // CONFIG_CFL

static INLINE int is_inter_block(const MB_MODE_INFO *mbmi) {
#if CONFIG_INTRABC
  if (is_intrabc_block(mbmi)) return 1;
#endif
  return mbmi->ref_frame[0] > INTRA_FRAME;
}

static INLINE int has_second_ref(const MB_MODE_INFO *mbmi) {
  return mbmi->ref_frame[1] > INTRA_FRAME;
}

#if CONFIG_EXT_COMP_REFS
static INLINE int has_uni_comp_refs(const MB_MODE_INFO *mbmi) {
  return has_second_ref(mbmi) && (!((mbmi->ref_frame[0] >= BWDREF_FRAME) ^
                                    (mbmi->ref_frame[1] >= BWDREF_FRAME)));
}

static INLINE MV_REFERENCE_FRAME comp_ref0(int ref_idx) {
  static const MV_REFERENCE_FRAME lut[] = {
    LAST_FRAME,    // LAST_LAST2_FRAMES,
    LAST_FRAME,    // LAST_LAST3_FRAMES,
    LAST_FRAME,    // LAST_GOLDEN_FRAMES,
    BWDREF_FRAME,  // BWDREF_ALTREF_FRAMES,
  };
  assert(NELEMENTS(lut) == UNIDIR_COMP_REFS);
  return lut[ref_idx];
}

static INLINE MV_REFERENCE_FRAME comp_ref1(int ref_idx) {
  static const MV_REFERENCE_FRAME lut[] = {
    LAST2_FRAME,   // LAST_LAST2_FRAMES,
    LAST3_FRAME,   // LAST_LAST3_FRAMES,
    GOLDEN_FRAME,  // LAST_GOLDEN_FRAMES,
    ALTREF_FRAME,  // BWDREF_ALTREF_FRAMES,
  };
  assert(NELEMENTS(lut) == UNIDIR_COMP_REFS);
  return lut[ref_idx];
}
#endif  // CONFIG_EXT_COMP_REFS

PREDICTION_MODE av1_left_block_mode(const MODE_INFO *left_mi);

PREDICTION_MODE av1_above_block_mode(const MODE_INFO *above_mi);

static INLINE int is_global_mv_block(const MODE_INFO *mi,
                                     TransformationType type) {
  const MB_MODE_INFO *const mbmi = &mi->mbmi;
  const PREDICTION_MODE mode = mbmi->mode;
#if GLOBAL_SUB8X8_USED
  const int block_size_allowed = 1;
#else
  const BLOCK_SIZE bsize = mbmi->sb_type;
  const int block_size_allowed =
      AOMMIN(block_size_wide[bsize], block_size_high[bsize]) >= 8;
#endif  // GLOBAL_SUB8X8_USED
  return (mode == GLOBALMV || mode == GLOBAL_GLOBALMV) && type > TRANSLATION &&
         block_size_allowed;
}

#if CONFIG_MISMATCH_DEBUG
static INLINE void mi_to_pixel_loc(int *pixel_c, int *pixel_r, int mi_col,
                                   int mi_row, int tx_blk_col, int tx_blk_row,
                                   int subsampling_x, int subsampling_y) {
  *pixel_c = ((mi_col >> subsampling_x) << MI_SIZE_LOG2) +
             (tx_blk_col << tx_size_wide_log2[0]);
  *pixel_r = ((mi_row >> subsampling_y) << MI_SIZE_LOG2) +
             (tx_blk_row << tx_size_high_log2[0]);
}
#endif

enum mv_precision { MV_PRECISION_Q3, MV_PRECISION_Q4 };

struct buf_2d {
  uint8_t *buf;
  uint8_t *buf0;
  int width;
  int height;
  int stride;
};

typedef struct macroblockd_plane {
  tran_low_t *dqcoeff;
  PLANE_TYPE plane_type;
  int subsampling_x;
  int subsampling_y;
  struct buf_2d dst;
  struct buf_2d pre[2];
  ENTROPY_CONTEXT *above_context;
  ENTROPY_CONTEXT *left_context;

  // The dequantizers below are true dequntizers used only in the
  // dequantization process.  They have the same coefficient
  // shift/scale as TX.
  int16_t seg_dequant_QTX[MAX_SEGMENTS][2];
#if CONFIG_NEW_QUANT
  dequant_val_type_nuq seg_dequant_nuq_QTX[MAX_SEGMENTS][QUANT_PROFILES][2];
#endif
  uint8_t *color_index_map;

  // block size in pixels
  uint8_t width, height;

#if CONFIG_AOM_QM
  qm_val_t *seg_iqmatrix[MAX_SEGMENTS][TX_SIZES_ALL];
  qm_val_t *seg_qmatrix[MAX_SEGMENTS][TX_SIZES_ALL];
#endif

  // the 'dequantizers' below are not literal dequantizer values.
  // They're used by encoder RDO to generate ad-hoc lambda values.
  // They use a hardwired Q3 coeff shift and do not necessarily match
  // the TX scale in use.
  const int16_t *dequant_Q3;

#if CONFIG_DIST_8X8
  DECLARE_ALIGNED(16, int16_t, pred[MAX_SB_SQUARE]);
#endif
} MACROBLOCKD_PLANE;

#define BLOCK_OFFSET(x, i) \
  ((x) + (i) * (1 << (tx_size_wide_log2[0] + tx_size_high_log2[0])))

typedef struct RefBuffer {
  int idx;
  YV12_BUFFER_CONFIG *buf;
  struct scale_factors sf;
} RefBuffer;

#if CONFIG_ADAPT_SCAN
typedef int16_t EobThresholdMD[TX_TYPES][EOB_THRESHOLD_NUM];
#endif

#if CONFIG_LOOP_RESTORATION
typedef struct {
  DECLARE_ALIGNED(16, InterpKernel, vfilter);
  DECLARE_ALIGNED(16, InterpKernel, hfilter);
} WienerInfo;

typedef struct {
  int ep;
  int xqd[2];
} SgrprojInfo;
#endif  // CONFIG_LOOP_RESTORATION

#if CONFIG_CFL
#if CONFIG_DEBUG
#define CFL_SUB8X8_VAL_MI_SIZE (4)
#define CFL_SUB8X8_VAL_MI_SQUARE \
  (CFL_SUB8X8_VAL_MI_SIZE * CFL_SUB8X8_VAL_MI_SIZE)
#endif  // CONFIG_DEBUG
#define CFL_MAX_BLOCK_SIZE (BLOCK_32X32)
#define CFL_PRED_BUF_LINE (32)
#define CFL_PRED_BUF_SQUARE (CFL_PRED_BUF_LINE * CFL_PRED_BUF_LINE)
typedef struct cfl_ctx {
  // The CfL prediction buffer is used in two steps:
  //   1. Stores Q3 reconstructed luma pixels
  //      (only Q2 is required, but Q3 is used to avoid shifts)
  //   2. Stores Q3 AC contributions (step1 - tx block avg)
  int16_t pred_buf_q3[CFL_PRED_BUF_SQUARE];

  // Height and width currently used in the CfL prediction buffer.
  int buf_height, buf_width;

  int are_parameters_computed;

  // Chroma subsampling
  int subsampling_x, subsampling_y;

  int mi_row, mi_col;

  // Whether the reconstructed luma pixels need to be stored
  int store_y;

#if CONFIG_DEBUG
  int rate;
#endif  // CONFIG_DEBUG

  int is_chroma_reference;
#if CONFIG_DEBUG
  // Validation buffer is usually 2x2, except for 16x4 and 4x16 in that case it
  // is 4x2 and 2x4 respectively. To simplify accessing the buffer we use a
  // stride of CFL_SUB8X8_VAL_MI_SIZE resulting in a square of 16.
  uint16_t sub8x8_val[CFL_SUB8X8_VAL_MI_SQUARE];
  uint16_t store_counter;
  uint16_t last_compute_counter;
#endif  // CONFIG_DEBUG
} CFL_CTX;
#endif  // CONFIG_CFL

#if CONFIG_JNT_COMP
typedef struct jnt_comp_params {
  int use_jnt_comp_avg;
  int fwd_offset;
  int bck_offset;
} JNT_COMP_PARAMS;
#endif  // CONFIG_JNT_COMP

typedef struct macroblockd {
  struct macroblockd_plane plane[MAX_MB_PLANE];
  uint8_t bmode_blocks_wl;
  uint8_t bmode_blocks_hl;

  FRAME_COUNTS *counts;
  TileInfo tile;

  int mi_stride;

  MODE_INFO **mi;
  MODE_INFO *left_mi;
  MODE_INFO *above_mi;
  MB_MODE_INFO *left_mbmi;
  MB_MODE_INFO *above_mbmi;

  int up_available;
  int left_available;
  int chroma_up_available;
  int chroma_left_available;

  /* Distance of MB away from frame edges in subpixels (1/8th pixel)  */
  int mb_to_left_edge;
  int mb_to_right_edge;
  int mb_to_top_edge;
  int mb_to_bottom_edge;

  FRAME_CONTEXT *fc;

  /* pointers to reference frames */
  const RefBuffer *block_refs[2];

  /* pointer to current frame */
  const YV12_BUFFER_CONFIG *cur_buf;

#if CONFIG_INTRABC
  /* Scale of the current frame with respect to itself */
  struct scale_factors sf_identity;
#endif

  ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
  ENTROPY_CONTEXT left_context[MAX_MB_PLANE][2 * MAX_MIB_SIZE];

  PARTITION_CONTEXT *above_seg_context;
  PARTITION_CONTEXT left_seg_context[MAX_MIB_SIZE];

  TXFM_CONTEXT *above_txfm_context;
  TXFM_CONTEXT *left_txfm_context;
  TXFM_CONTEXT left_txfm_context_buffer[2 * MAX_MIB_SIZE];

#if CONFIG_LOOP_RESTORATION
  WienerInfo wiener_info[MAX_MB_PLANE];
  SgrprojInfo sgrproj_info[MAX_MB_PLANE];
#endif  // CONFIG_LOOP_RESTORATION

  // block dimension in the unit of mode_info.
  uint8_t n8_w, n8_h;

  uint8_t ref_mv_count[MODE_CTX_REF_FRAMES];
  CANDIDATE_MV ref_mv_stack[MODE_CTX_REF_FRAMES][MAX_REF_MV_STACK_SIZE];
  uint8_t is_sec_rect;

  FRAME_CONTEXT *tile_ctx;
  /* Bit depth: 8, 10, 12 */
  int bd;

  int qindex[MAX_SEGMENTS];
  int lossless[MAX_SEGMENTS];
  int corrupted;
#if CONFIG_AMVR
  int cur_frame_force_integer_mv;
// same with that in AV1_COMMON
#endif
  struct aom_internal_error_info *error_info;
  const WarpedMotionParams *global_motion;
  int prev_qindex;
  int delta_qindex;
  int current_qindex;
#if CONFIG_EXT_DELTA_Q
  // Since actual frame level loop filtering level value is not available
  // at the beginning of the tile (only available during actual filtering)
  // at encoder side.we record the delta_lf (against the frame level loop
  // filtering level) and code the delta between previous superblock's delta
  // lf and current delta lf. It is equivalent to the delta between previous
  // superblock's actual lf and current lf.
  int prev_delta_lf_from_base;
  int current_delta_lf_from_base;
#if CONFIG_LOOPFILTER_LEVEL
  // For this experiment, we have four frame filter levels for different plane
  // and direction. So, to support the per superblock update, we need to add
  // a few more params as below.
  // 0: delta loop filter level for y plane vertical
  // 1: delta loop filter level for y plane horizontal
  // 2: delta loop filter level for u plane
  // 3: delta loop filter level for v plane
  // To make it consistent with the reference to each filter level in segment,
  // we need to -1, since
  // SEG_LVL_ALT_LF_Y_V = 1;
  // SEG_LVL_ALT_LF_Y_H = 2;
  // SEG_LVL_ALT_LF_U   = 3;
  // SEG_LVL_ALT_LF_V   = 4;
  int prev_delta_lf[FRAME_LF_COUNT];
  int curr_delta_lf[FRAME_LF_COUNT];
#endif  // CONFIG_LOOPFILTER_LEVEL
#endif
#if CONFIG_ADAPT_SCAN
  const EobThresholdMD *eob_threshold_md;
#endif

  DECLARE_ALIGNED(16, uint8_t, seg_mask[2 * MAX_SB_SQUARE]);

#if CONFIG_CFL
  CFL_CTX cfl;
#endif

#if CONFIG_JNT_COMP
  JNT_COMP_PARAMS jcp_param;
#endif
} MACROBLOCKD;

static INLINE int get_bitdepth_data_path_index(const MACROBLOCKD *xd) {
  return xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH ? 1 : 0;
}

static INLINE BLOCK_SIZE get_subsize(BLOCK_SIZE bsize,
                                     PARTITION_TYPE partition) {
  if (partition == PARTITION_INVALID)
    return BLOCK_INVALID;
  else
    return subsize_lookup[partition][bsize];
}

static const TX_TYPE intra_mode_to_tx_type_context[INTRA_MODES] = {
  DCT_DCT,    // DC
  ADST_DCT,   // V
  DCT_ADST,   // H
  DCT_DCT,    // D45
  ADST_ADST,  // D135
  ADST_DCT,   // D117
  DCT_ADST,   // D153
  DCT_ADST,   // D207
  ADST_DCT,   // D63
  ADST_ADST,  // SMOOTH
  ADST_DCT,   // SMOOTH_V
  DCT_ADST,   // SMOOTH_H
  ADST_ADST,  // PAETH
};

#define USE_TXTYPE_SEARCH_FOR_SUB8X8_IN_CB4X4 1

static INLINE int is_rect_tx(TX_SIZE tx_size) { return tx_size >= TX_SIZES; }

static INLINE int block_signals_txsize(BLOCK_SIZE bsize) {
  return bsize > BLOCK_4X4;
}

#define ALLOW_INTRA_EXT_TX 1

// Number of transform types in each set type
static const int av1_num_ext_tx_set[EXT_TX_SET_TYPES] = {
  1, 2, 5, 7, 7, 10, 12, 16, 16,
};

static const int av1_ext_tx_set_idx_to_type[2][AOMMAX(EXT_TX_SETS_INTRA,
                                                      EXT_TX_SETS_INTER)] = {
  {
      // Intra
      EXT_TX_SET_DCTONLY, EXT_TX_SET_DTT4_IDTX_1DDCT,
#if EXT_TX_16X16_SET == 0
      EXT_TX_SET_DTT4_IDTX_1DDCT_16X16,
#else
      EXT_TX_SET_DTT4_IDTX,
#endif  // EXT_TX_16X16_SET
  },
  {
      // Inter
      EXT_TX_SET_DCTONLY, EXT_TX_SET_ALL16,
#if EXT_TX_16X16_SET == 0
      EXT_TX_SET_ALL16_16X16,
#elif EXT_TX_16X16_SET == 1
      EXT_TX_SET_DTT9_IDTX_1DDCT,
#else
      EXT_TX_SET_DTT9_IDTX,
#endif  // EXT_TX_16X16_SET
      EXT_TX_SET_DCT_IDTX,
  },
};

static const int av1_ext_tx_used[EXT_TX_SET_TYPES][TX_TYPES] = {
  {
      1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  },
  {
      1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,
  },
  {
      1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,
  },
  {
      1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0,
  },
  {
      1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0,
  },
  {
      1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0,
  },
  {
      1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0,
  },
  {
      1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  },
  {
      1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  },
};

static INLINE TxSetType get_ext_tx_set_type(TX_SIZE tx_size, BLOCK_SIZE bs,
                                            int is_inter, int use_reduced_set) {
  const TX_SIZE tx_size_sqr_up = txsize_sqr_up_map[tx_size];
  const TX_SIZE tx_size_sqr = txsize_sqr_map[tx_size];
#if USE_TXTYPE_SEARCH_FOR_SUB8X8_IN_CB4X4
  (void)bs;
  if (tx_size_sqr_up > TX_32X32) return EXT_TX_SET_DCTONLY;
#else
  if (tx_size_sqr_up > TX_32X32 || bs < BLOCK_8X8) return EXT_TX_SET_DCTONLY;
#endif  // USE_TXTYPE_SEARCH_FOR_SUB8X8_IN_CB4X4
  if (use_reduced_set)
    return is_inter ? EXT_TX_SET_DCT_IDTX : EXT_TX_SET_DTT4_IDTX;
#if CONFIG_DAALA_TX_DST32
  if (tx_size_sqr_up > TX_32X32)
    return is_inter ? EXT_TX_SET_DCT_IDTX : EXT_TX_SET_DCTONLY;
  if (is_inter)
    return (tx_size_sqr >= TX_16X16 ? EXT_TX_SET_DTT9_IDTX_1DDCT
                                    : EXT_TX_SET_ALL16);
  else
    return (tx_size_sqr >= TX_16X16 ? EXT_TX_SET_DTT4_IDTX
                                    : EXT_TX_SET_DTT4_IDTX_1DDCT);
#endif
  if (tx_size_sqr_up == TX_32X32)
    return is_inter ? EXT_TX_SET_DCT_IDTX : EXT_TX_SET_DCTONLY;
  if (is_inter)
    return (tx_size_sqr == TX_16X16 ?
#if EXT_TX_16X16_SET == 0
                                    EXT_TX_SET_ALL16_16X16
#elif EXT_TX_16X16_SET == 1
                                    EXT_TX_SET_DTT9_IDTX_1DDCT
#else
                                    EXT_TX_SET_DTT9_IDTX
#endif  // EXT_TX_16X16_SET
                                    : EXT_TX_SET_ALL16);
  else
    return (tx_size_sqr == TX_16X16 ?
#if EXT_TX_16X16_SET == 0
                                    EXT_TX_SET_DTT4_IDTX_1DDCT_16X16
#else
                                    EXT_TX_SET_DTT4_IDTX
#endif  // EXT_TX_16X16_SET
                                    : EXT_TX_SET_DTT4_IDTX_1DDCT);
}

// Maps tx set types to the indices.
static const int ext_tx_set_index[2][EXT_TX_SET_TYPES] = {
  {
      // Intra
      0, -1,
#if EXT_TX_16X16_SET == 0
      -1, 2,
#else
      2, -1,
#endif  // EXT_TX_16X16_SET
      1, -1, -1, -1, -1,
  },
  {
      // Inter
      0, 3, -1, -1, -1,
#if EXT_TX_16X16_SET == 0
      -1, -1, 2,
#elif EXT_TX_16X16_SET == 1
      -1, 2, -1,
#else
      2, -1, -1,
#endif  // EXT_TX_16X16_SET
      1,
  },
};

static INLINE int get_ext_tx_set(TX_SIZE tx_size, BLOCK_SIZE bs, int is_inter,
                                 int use_reduced_set) {
  const TxSetType set_type =
      get_ext_tx_set_type(tx_size, bs, is_inter, use_reduced_set);
  return ext_tx_set_index[is_inter][set_type];
}

static INLINE int get_ext_tx_types(TX_SIZE tx_size, BLOCK_SIZE bs, int is_inter,
                                   int use_reduced_set) {
  const int set_type =
      get_ext_tx_set_type(tx_size, bs, is_inter, use_reduced_set);
  return av1_num_ext_tx_set[set_type];
}

static INLINE int is_rect_tx_allowed_bsize(BLOCK_SIZE bsize) {
  static const char LUT[BLOCK_SIZES_ALL] = {
    0,  // BLOCK_2X2
    0,  // BLOCK_2X4
    0,  // BLOCK_4X2
    0,  // BLOCK_4X4
    1,  // BLOCK_4X8
    1,  // BLOCK_8X4
    0,  // BLOCK_8X8
    1,  // BLOCK_8X16
    1,  // BLOCK_16X8
    0,  // BLOCK_16X16
    1,  // BLOCK_16X32
    1,  // BLOCK_32X16
    0,  // BLOCK_32X32
    1,  // BLOCK_32X64
    1,  // BLOCK_64X32
    0,  // BLOCK_64X64
#if CONFIG_EXT_PARTITION
    0,  // BLOCK_64X128
    0,  // BLOCK_128X64
    0,  // BLOCK_128X128
#endif  // CONFIG_EXT_PARTITION
    1,  // BLOCK_4X16
    1,  // BLOCK_16X4
    1,  // BLOCK_8X32
    1,  // BLOCK_32X8
    1,  // BLOCK_16X64
    1,  // BLOCK_64X16
#if CONFIG_EXT_PARTITION
    1,  // BLOCK_32X128
    1,  // BLOCK_128X32
#endif  // CONFIG_EXT_PARTITION
  };

  return LUT[bsize];
}

static INLINE int is_rect_tx_allowed(const MACROBLOCKD *xd,
                                     const MB_MODE_INFO *mbmi) {
  return is_rect_tx_allowed_bsize(mbmi->sb_type) &&
         !xd->lossless[mbmi->segment_id];
}

static INLINE TX_SIZE get_max_rect_tx_size(BLOCK_SIZE bsize, int is_inter) {
  return max_txsize_rect_lookup[is_inter][bsize];
}

static INLINE TX_SIZE tx_size_from_tx_mode(BLOCK_SIZE bsize, TX_MODE tx_mode,
                                           int is_inter) {
  const TX_SIZE largest_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
  const TX_SIZE max_rect_tx_size = get_max_rect_tx_size(bsize, is_inter);
  (void)is_inter;
  if (bsize == BLOCK_4X4)
    return AOMMIN(max_txsize_lookup[bsize], largest_tx_size);
  if (txsize_sqr_map[max_rect_tx_size] <= largest_tx_size)
    return max_rect_tx_size;
  else
    return largest_tx_size;
}

#if CONFIG_EXT_INTRA
extern const int16_t dr_intra_derivative[90];
static const uint8_t mode_to_angle_map[] = {
  0, 90, 180, 45, 135, 113, 157, 203, 67, 0, 0, 0, 0,
};
#endif  // CONFIG_EXT_INTRA

#if CONFIG_FILTER_INTRA
#define DISABLE_SUB8X8_FILTER_INTRA 0

static INLINE int av1_filter_intra_allowed_bsize(BLOCK_SIZE bs) {
  (void)bs;
#if DISABLE_SUB8X8_FILTER_INTRA
  return block_size_wide[bs] >= 8 && block_size_high[bs] >= 8;
#else
  return 1;
#endif
}

static INLINE int av1_filter_intra_allowed_txsize(TX_SIZE tx) {
  (void)tx;
  if (tx == TX_INVALID) return 0;

#if DISABLE_SUB8X8_FILTER_INTRA
  return tx_size_wide[tx] >= 8 && tx_size_high[tx] >= 8 &&
         tx_size_wide[tx] <= 32 && tx_size_high[tx] <= 32;
#else
  return tx_size_wide[tx] <= 32 && tx_size_high[tx] <= 32;
#endif
}

static INLINE TX_SIZE av1_max_tx_size_for_filter_intra(BLOCK_SIZE bsize,
                                                       TX_MODE tx_mode) {
  const TX_SIZE max_tx_size = tx_size_from_tx_mode(bsize, tx_mode, 0);

  if (tx_mode != TX_MODE_SELECT) return max_tx_size;

  int depth = 0;
  TX_SIZE tx_size = max_tx_size;
  while (depth < MAX_TX_DEPTH && tx_size != TX_4X4) {
    if (av1_filter_intra_allowed_txsize(tx_size)) return tx_size;
    depth++;
    tx_size = sub_tx_size_map[0][tx_size];
  }
  return TX_INVALID;
}
#endif