encodeframe.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 <limits.h>
#include <math.h>
#include <stdio.h>

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

#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/binary_codes_writer.h"
#include "aom_ports/mem.h"
#include "aom_ports/aom_timer.h"
#include "aom_ports/system_state.h"

#include "av1/common/common.h"
#include "av1/common/entropy.h"
#include "av1/common/entropymode.h"
#include "av1/common/idct.h"
#include "av1/common/mv.h"
#include "av1/common/mvref_common.h"
#include "av1/common/pred_common.h"
#include "av1/common/quant_common.h"
#include "av1/common/reconintra.h"
#include "av1/common/reconinter.h"
#include "av1/common/seg_common.h"
#include "av1/common/tile_common.h"

#include "av1/encoder/aq_complexity.h"
#include "av1/encoder/aq_cyclicrefresh.h"
#include "av1/encoder/aq_variance.h"
#if CONFIG_SUPERTX
#include "av1/encoder/cost.h"
#endif
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#include "av1/common/warped_motion.h"
#endif  // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#if CONFIG_GLOBAL_MOTION
#include "av1/encoder/global_motion.h"
#endif  // CONFIG_GLOBAL_MOTION
#include "av1/encoder/encodeframe.h"
#include "av1/encoder/encodemb.h"
#include "av1/encoder/encodemv.h"
#if CONFIG_LV_MAP
#include "av1/encoder/encodetxb.h"
#endif
#include "av1/encoder/ethread.h"
#include "av1/encoder/extend.h"
#include "av1/encoder/rd.h"
#include "av1/encoder/rdopt.h"
#include "av1/encoder/segmentation.h"
#include "av1/encoder/tokenize.h"
#if CONFIG_PVQ
#include "av1/common/pvq.h"
#include "av1/encoder/pvq_encoder.h"
#endif
#if CONFIG_HIGHBITDEPTH
#define IF_HBD(...) __VA_ARGS__
#else
#define IF_HBD(...)
#endif  // CONFIG_HIGHBITDEPTH

static void encode_superblock(const AV1_COMP *const cpi, ThreadData *td,
                              TOKENEXTRA **t, RUN_TYPE dry_run, int mi_row,
                              int mi_col, BLOCK_SIZE bsize, int *rate);

#if CONFIG_SUPERTX
static int check_intra_b(PICK_MODE_CONTEXT *ctx);

static int check_intra_sb(const AV1_COMP *cpi, const TileInfo *const tile,
                          int mi_row, int mi_col, BLOCK_SIZE bsize,
                          PC_TREE *pc_tree);
static void predict_superblock(const AV1_COMP *const cpi, ThreadData *td,
#if CONFIG_EXT_INTER
                               int mi_row_ori, int mi_col_ori,
#endif  // CONFIG_EXT_INTER
                               int mi_row_pred, int mi_col_pred, int plane,
                               BLOCK_SIZE bsize_pred, int b_sub8x8, int block);
static int check_supertx_sb(BLOCK_SIZE bsize, TX_SIZE supertx_size,
                            PC_TREE *pc_tree);
static void predict_sb_complex(const AV1_COMP *const cpi, ThreadData *td,
                               const TileInfo *const tile, int mi_row,
                               int mi_col, int mi_row_ori, int mi_col_ori,
                               RUN_TYPE dry_run, BLOCK_SIZE bsize,
                               BLOCK_SIZE top_bsize, uint8_t *dst_buf[3],
                               int dst_stride[3], PC_TREE *pc_tree);
static void update_state_sb_supertx(const AV1_COMP *const cpi, ThreadData *td,
                                    const TileInfo *const tile, int mi_row,
                                    int mi_col, BLOCK_SIZE bsize,
                                    RUN_TYPE dry_run, PC_TREE *pc_tree);
static void rd_supertx_sb(const AV1_COMP *const cpi, ThreadData *td,
                          const TileInfo *const tile, int mi_row, int mi_col,
                          BLOCK_SIZE bsize, int *tmp_rate, int64_t *tmp_dist,
                          TX_TYPE *best_tx, PC_TREE *pc_tree);
#endif  // CONFIG_SUPERTX

// This is used as a reference when computing the source variance for the
//  purposes of activity masking.
// Eventually this should be replaced by custom no-reference routines,
//  which will be faster.
static const uint8_t AV1_VAR_OFFS[MAX_SB_SIZE] = {
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
#if CONFIG_EXT_PARTITION
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128
#endif  // CONFIG_EXT_PARTITION
};

#if CONFIG_HIGHBITDEPTH
static const uint16_t AV1_HIGH_VAR_OFFS_8[MAX_SB_SIZE] = {
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
#if CONFIG_EXT_PARTITION
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128
#endif  // CONFIG_EXT_PARTITION
};

static const uint16_t AV1_HIGH_VAR_OFFS_10[MAX_SB_SIZE] = {
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
#if CONFIG_EXT_PARTITION
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4,
  128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4, 128 * 4
#endif  // CONFIG_EXT_PARTITION
};

static const uint16_t AV1_HIGH_VAR_OFFS_12[MAX_SB_SIZE] = {
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16,
#if CONFIG_EXT_PARTITION
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16, 128 * 16,
  128 * 16
#endif  // CONFIG_EXT_PARTITION
};
#endif  // CONFIG_HIGHBITDEPTH

unsigned int av1_get_sby_perpixel_variance(const AV1_COMP *cpi,
                                           const struct buf_2d *ref,
                                           BLOCK_SIZE bs) {
  unsigned int sse;
  const unsigned int var =
      cpi->fn_ptr[bs].vf(ref->buf, ref->stride, AV1_VAR_OFFS, 0, &sse);
  return ROUND_POWER_OF_TWO(var, num_pels_log2_lookup[bs]);
}

#if CONFIG_HIGHBITDEPTH
unsigned int av1_high_get_sby_perpixel_variance(const AV1_COMP *cpi,
                                                const struct buf_2d *ref,
                                                BLOCK_SIZE bs, int bd) {
  unsigned int var, sse;
  switch (bd) {
    case 10:
      var =
          cpi->fn_ptr[bs].vf(ref->buf, ref->stride,
                             CONVERT_TO_BYTEPTR(AV1_HIGH_VAR_OFFS_10), 0, &sse);
      break;
    case 12:
      var =
          cpi->fn_ptr[bs].vf(ref->buf, ref->stride,
                             CONVERT_TO_BYTEPTR(AV1_HIGH_VAR_OFFS_12), 0, &sse);
      break;
    case 8:
    default:
      var =
          cpi->fn_ptr[bs].vf(ref->buf, ref->stride,
                             CONVERT_TO_BYTEPTR(AV1_HIGH_VAR_OFFS_8), 0, &sse);
      break;
  }
  return ROUND_POWER_OF_TWO(var, num_pels_log2_lookup[bs]);
}
#endif  // CONFIG_HIGHBITDEPTH

static unsigned int get_sby_perpixel_diff_variance(const AV1_COMP *const cpi,
                                                   const struct buf_2d *ref,
                                                   int mi_row, int mi_col,
                                                   BLOCK_SIZE bs) {
  unsigned int sse, var;
  uint8_t *last_y;
  const YV12_BUFFER_CONFIG *last = get_ref_frame_buffer(cpi, LAST_FRAME);

  assert(last != NULL);
  last_y =
      &last->y_buffer[mi_row * MI_SIZE * last->y_stride + mi_col * MI_SIZE];
  var = cpi->fn_ptr[bs].vf(ref->buf, ref->stride, last_y, last->y_stride, &sse);
  return ROUND_POWER_OF_TWO(var, num_pels_log2_lookup[bs]);
}

static BLOCK_SIZE get_rd_var_based_fixed_partition(AV1_COMP *cpi, MACROBLOCK *x,
                                                   int mi_row, int mi_col) {
  unsigned int var = get_sby_perpixel_diff_variance(
      cpi, &x->plane[0].src, mi_row, mi_col, BLOCK_64X64);
  if (var < 8)
    return BLOCK_64X64;
  else if (var < 128)
    return BLOCK_32X32;
  else if (var < 2048)
    return BLOCK_16X16;
  else
    return BLOCK_8X8;
}

// Lighter version of set_offsets that only sets the mode info
// pointers.
static void set_mode_info_offsets(const AV1_COMP *const cpi,
                                  MACROBLOCK *const x, MACROBLOCKD *const xd,
                                  int mi_row, int mi_col) {
  const AV1_COMMON *const cm = &cpi->common;
  const int idx_str = xd->mi_stride * mi_row + mi_col;
  xd->mi = cm->mi_grid_visible + idx_str;
  xd->mi[0] = cm->mi + idx_str;
  x->mbmi_ext = cpi->mbmi_ext_base + (mi_row * cm->mi_cols + mi_col);
}

static void set_offsets_without_segment_id(const AV1_COMP *const cpi,
                                           const TileInfo *const tile,
                                           MACROBLOCK *const x, int mi_row,
                                           int mi_col, BLOCK_SIZE bsize) {
  const AV1_COMMON *const cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  const int mi_width = mi_size_wide[bsize];
  const int mi_height = mi_size_high[bsize];

  set_mode_info_offsets(cpi, x, xd, mi_row, mi_col);
  set_skip_context(xd, mi_row, mi_col);
#if CONFIG_VAR_TX
  xd->above_txfm_context =
      cm->above_txfm_context + (mi_col << TX_UNIT_WIDE_LOG2);
  xd->left_txfm_context = xd->left_txfm_context_buffer +
                          ((mi_row & MAX_MIB_MASK) << TX_UNIT_HIGH_LOG2);
  xd->max_tx_size = max_txsize_lookup[bsize];
#endif

  // Set up destination pointers.
  av1_setup_dst_planes(xd->plane, bsize, get_frame_new_buffer(cm), mi_row,
                       mi_col);

  // Set up limit values for MV components.
  // Mv beyond the range do not produce new/different prediction block.
  x->mv_limits.row_min =
      -(((mi_row + mi_height) * MI_SIZE) + AOM_INTERP_EXTEND);
  x->mv_limits.col_min = -(((mi_col + mi_width) * MI_SIZE) + AOM_INTERP_EXTEND);
  x->mv_limits.row_max = (cm->mi_rows - mi_row) * MI_SIZE + AOM_INTERP_EXTEND;
  x->mv_limits.col_max = (cm->mi_cols - mi_col) * MI_SIZE + AOM_INTERP_EXTEND;

  set_plane_n4(xd, mi_width, mi_height);

  // Set up distance of MB to edge of frame in 1/8th pel units.
  assert(!(mi_col & (mi_width - 1)) && !(mi_row & (mi_height - 1)));
  set_mi_row_col(xd, tile, mi_row, mi_height, mi_col, mi_width,
#if CONFIG_DEPENDENT_HORZTILES
                 cm->dependent_horz_tiles,
#endif  // CONFIG_DEPENDENT_HORZTILES
                 cm->mi_rows, cm->mi_cols);

  // Set up source buffers.
  av1_setup_src_planes(x, cpi->source, mi_row, mi_col);

  // R/D setup.
  x->rdmult = cpi->rd.RDMULT;

  // required by av1_append_sub8x8_mvs_for_idx() and av1_find_best_ref_mvs()
  xd->tile = *tile;
}

static void set_offsets(const AV1_COMP *const cpi, const TileInfo *const tile,
                        MACROBLOCK *const x, int mi_row, int mi_col,
                        BLOCK_SIZE bsize) {
  const AV1_COMMON *const cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *mbmi;
  const struct segmentation *const seg = &cm->seg;

  set_offsets_without_segment_id(cpi, tile, x, mi_row, mi_col, bsize);

  mbmi = &xd->mi[0]->mbmi;
#if CONFIG_CFL
  xd->cfl->mi_row = mi_row;
  xd->cfl->mi_col = mi_col;
#endif

  // Setup segment ID.
  if (seg->enabled) {
    if (!cpi->vaq_refresh) {
      const uint8_t *const map =
          seg->update_map ? cpi->segmentation_map : cm->last_frame_seg_map;
      mbmi->segment_id = get_segment_id(cm, map, bsize, mi_row, mi_col);
    }
    av1_init_plane_quantizers(cpi, x, mbmi->segment_id);
  } else {
    mbmi->segment_id = 0;
  }

#if CONFIG_SUPERTX
  mbmi->segment_id_supertx = MAX_SEGMENTS;
#endif  // CONFIG_SUPERTX
}

#if CONFIG_SUPERTX
static void set_offsets_supertx(const AV1_COMP *const cpi, ThreadData *td,
                                const TileInfo *const tile, int mi_row,
                                int mi_col, BLOCK_SIZE bsize) {
  MACROBLOCK *const x = &td->mb;
  const AV1_COMMON *const cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  const int mi_width = mi_size_wide[bsize];
  const int mi_height = mi_size_high[bsize];
#if CONFIG_DEPENDENT_HORZTILES
  set_mode_info_offsets(cpi, x, xd, mi_row, mi_col, cm->dependent_horz_tiles);
#else
  set_mode_info_offsets(cpi, x, xd, mi_row, mi_col);
#endif

  // Set up distance of MB to edge of frame in 1/8th pel units.
  assert(!(mi_col & (mi_width - 1)) && !(mi_row & (mi_height - 1)));
  set_mi_row_col(xd, tile, mi_row, mi_height, mi_col, mi_width,
#if CONFIG_DEPENDENT_HORZTILES
                 cm->dependent_horz_tiles,
#endif  // CONFIG_DEPENDENT_HORZTILES
                 cm->mi_rows, cm->mi_cols);
}

static void set_offsets_extend(const AV1_COMP *const cpi, ThreadData *td,
                               const TileInfo *const tile, int mi_row_pred,
                               int mi_col_pred, int mi_row_ori, int mi_col_ori,
                               BLOCK_SIZE bsize_pred) {
  // Used in supertx
  // (mi_row_ori, mi_col_ori, bsize_ori): region for mv
  // (mi_row_pred, mi_col_pred, bsize_pred): region to predict
  MACROBLOCK *const x = &td->mb;
  const AV1_COMMON *const cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  const int mi_width = mi_size_wide[bsize_pred];
  const int mi_height = mi_size_high[bsize_pred];

#if CONFIG_DEPENDENT_HORZTILES
  set_mode_info_offsets(cpi, x, xd, mi_row_ori, mi_col_ori,
                        cm->dependent_horz_tiles);
#else
  set_mode_info_offsets(cpi, x, xd, mi_row_ori, mi_col_ori);
#endif

  // Set up limit values for MV components.
  // Mv beyond the range do not produce new/different prediction block.
  x->mv_limits.row_min =
      -(((mi_row_pred + mi_height) * MI_SIZE) + AOM_INTERP_EXTEND);
  x->mv_limits.col_min =
      -(((mi_col_pred + mi_width) * MI_SIZE) + AOM_INTERP_EXTEND);
  x->mv_limits.row_max =
      (cm->mi_rows - mi_row_pred) * MI_SIZE + AOM_INTERP_EXTEND;
  x->mv_limits.col_max =
      (cm->mi_cols - mi_col_pred) * MI_SIZE + AOM_INTERP_EXTEND;

// Set up distance of MB to edge of frame in 1/8th pel units.
#if !CONFIG_CB4X4
  assert(!(mi_col_pred & (mi_width - mi_size_wide[BLOCK_8X8])) &&
         !(mi_row_pred & (mi_height - mi_size_high[BLOCK_8X8])));
#endif
  set_mi_row_col(xd, tile, mi_row_pred, mi_height, mi_col_pred, mi_width,
#if CONFIG_DEPENDENT_HORZTILES
                 cm->dependent_horz_tiles,
#endif  // CONFIG_DEPENDENT_HORZTILES
                 cm->mi_rows, cm->mi_cols);
  xd->up_available = (mi_row_ori > tile->mi_row_start);
  xd->left_available = (mi_col_ori > tile->mi_col_start);

  // R/D setup.
  x->rdmult = cpi->rd.RDMULT;
}

static void set_segment_id_supertx(const AV1_COMP *const cpi,
                                   MACROBLOCK *const x, const int mi_row,
                                   const int mi_col, const BLOCK_SIZE bsize) {
  const AV1_COMMON *cm = &cpi->common;
  const struct segmentation *seg = &cm->seg;
  const int miw = AOMMIN(mi_size_wide[bsize], cm->mi_cols - mi_col);
  const int mih = AOMMIN(mi_size_high[bsize], cm->mi_rows - mi_row);
  const int mi_offset = mi_row * cm->mi_stride + mi_col;
  MODE_INFO **const mip = cm->mi_grid_visible + mi_offset;
  int r, c;
  int seg_id_supertx = MAX_SEGMENTS;

  if (!seg->enabled) {
    seg_id_supertx = 0;
  } else {
    // Find the minimum segment_id
    for (r = 0; r < mih; r++)
      for (c = 0; c < miw; c++)
        seg_id_supertx =
            AOMMIN(mip[r * cm->mi_stride + c]->mbmi.segment_id, seg_id_supertx);
    assert(0 <= seg_id_supertx && seg_id_supertx < MAX_SEGMENTS);

    // Initialize plane quantisers
    av1_init_plane_quantizers(cpi, x, seg_id_supertx);
  }

  // Assign the the segment_id back to segment_id_supertx
  for (r = 0; r < mih; r++)
    for (c = 0; c < miw; c++)
      mip[r * cm->mi_stride + c]->mbmi.segment_id_supertx = seg_id_supertx;
}
#endif  // CONFIG_SUPERTX

#if CONFIG_DUAL_FILTER
static void reset_intmv_filter_type(const AV1_COMMON *const cm, MACROBLOCKD *xd,
                                    MB_MODE_INFO *mbmi) {
  int dir;
  for (dir = 0; dir < 2; ++dir) {
    if (!has_subpel_mv_component(xd->mi[0], xd, dir) &&
        (mbmi->ref_frame[1] == NONE_FRAME ||
         !has_subpel_mv_component(xd->mi[0], xd, dir + 2)))
      mbmi->interp_filter[dir] = (cm->interp_filter == SWITCHABLE)
                                     ? EIGHTTAP_REGULAR
                                     : cm->interp_filter;
    mbmi->interp_filter[dir + 2] = mbmi->interp_filter[dir];
  }
}

static void update_filter_type_count(FRAME_COUNTS *counts,
                                     const MACROBLOCKD *xd,
                                     const MB_MODE_INFO *mbmi) {
  int dir;
  for (dir = 0; dir < 2; ++dir) {
    if (has_subpel_mv_component(xd->mi[0], xd, dir) ||
        (mbmi->ref_frame[1] > INTRA_FRAME &&
         has_subpel_mv_component(xd->mi[0], xd, dir + 2))) {
      const int ctx = av1_get_pred_context_switchable_interp(xd, dir);
      ++counts->switchable_interp[ctx][mbmi->interp_filter[dir]];
    }
  }
}
#endif
#if CONFIG_GLOBAL_MOTION
static void update_global_motion_used(PREDICTION_MODE mode, BLOCK_SIZE bsize,
                                      const MB_MODE_INFO *mbmi,
                                      RD_COUNTS *rdc) {
  if (mode == ZEROMV
#if CONFIG_EXT_INTER
      || mode == ZERO_ZEROMV
#endif
      ) {
    const int num_4x4s =
        num_4x4_blocks_wide_lookup[bsize] * num_4x4_blocks_high_lookup[bsize];
    int ref;
    for (ref = 0; ref < 1 + has_second_ref(mbmi); ++ref) {
      rdc->global_motion_used[mbmi->ref_frame[ref]] += num_4x4s;
    }
  }
}
#endif  // CONFIG_GLOBAL_MOTION

static void reset_tx_size(MACROBLOCKD *xd, MB_MODE_INFO *mbmi,
                          const TX_MODE tx_mode) {
  if (xd->lossless[mbmi->segment_id]) {
    mbmi->tx_size = TX_4X4;
  } else if (tx_mode != TX_MODE_SELECT) {
    mbmi->tx_size =
        tx_size_from_tx_mode(mbmi->sb_type, tx_mode, is_inter_block(mbmi));
  }
}

static void set_ref_and_pred_mvs(MACROBLOCK *const x, int_mv *const mi_pred_mv,
                                 int8_t rf_type) {
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;

  const int bw = xd->n8_w << MI_SIZE_LOG2;
  const int bh = xd->n8_h << MI_SIZE_LOG2;
  int ref_mv_idx = mbmi->ref_mv_idx;
  MB_MODE_INFO_EXT *const mbmi_ext = x->mbmi_ext;
  CANDIDATE_MV *const curr_ref_mv_stack = mbmi_ext->ref_mv_stack[rf_type];

#if CONFIG_EXT_INTER
  if (has_second_ref(mbmi)) {
    // Special case: NEAR_NEWMV and NEW_NEARMV modes use 1 + mbmi->ref_mv_idx
    // (like NEARMV) instead
    if (mbmi->mode == NEAR_NEWMV || mbmi->mode == NEW_NEARMV) ref_mv_idx += 1;

    if (compound_ref0_mode(mbmi->mode) == NEWMV) {
      int_mv this_mv = curr_ref_mv_stack[ref_mv_idx].this_mv;
      clamp_mv_ref(&this_mv.as_mv, bw, bh, xd);
      mbmi_ext->ref_mvs[mbmi->ref_frame[0]][0] = this_mv;
      mbmi->pred_mv[0] = this_mv;
      mi_pred_mv[0] = this_mv;
    }
    if (compound_ref1_mode(mbmi->mode) == NEWMV) {
      int_mv this_mv = curr_ref_mv_stack[ref_mv_idx].comp_mv;
      clamp_mv_ref(&this_mv.as_mv, bw, bh, xd);
      mbmi_ext->ref_mvs[mbmi->ref_frame[1]][0] = this_mv;
      mbmi->pred_mv[1] = this_mv;
      mi_pred_mv[1] = this_mv;
    }
#if CONFIG_COMPOUND_SINGLEREF
  } else if (is_inter_singleref_comp_mode(mbmi->mode)) {
    // Special case: SR_NEAR_NEWMV uses 1 + mbmi->ref_mv_idx
    // (like NEARMV) instead
    if (mbmi->mode == SR_NEAR_NEWMV) ref_mv_idx += 1;

    if (compound_ref0_mode(mbmi->mode) == NEWMV ||
        compound_ref1_mode(mbmi->mode) == NEWMV) {
      int_mv this_mv = curr_ref_mv_stack[ref_mv_idx].this_mv;
      clamp_mv_ref(&this_mv.as_mv, bw, bh, xd);
      mbmi_ext->ref_mvs[mbmi->ref_frame[0]][0] = this_mv;
      mbmi->pred_mv[0] = this_mv;
      mi_pred_mv[0] = this_mv;
    }
#endif  // CONFIG_COMPOUND_SINGLEREF
  } else {
#endif  // CONFIG_EXT_INTER
    if (mbmi->mode == NEWMV) {
      int i;
      for (i = 0; i < 1 + has_second_ref(mbmi); ++i) {
        int_mv this_mv = (i == 0) ? curr_ref_mv_stack[ref_mv_idx].this_mv
                                  : curr_ref_mv_stack[ref_mv_idx].comp_mv;
        clamp_mv_ref(&this_mv.as_mv, bw, bh, xd);
        mbmi_ext->ref_mvs[mbmi->ref_frame[i]][0] = this_mv;
        mbmi->pred_mv[i] = this_mv;
        mi_pred_mv[i] = this_mv;
      }
    }
#if CONFIG_EXT_INTER
  }
#endif  // CONFIG_EXT_INTER
}

static void update_state(const AV1_COMP *const cpi, ThreadData *td,
                         PICK_MODE_CONTEXT *ctx, int mi_row, int mi_col,
                         BLOCK_SIZE bsize, RUN_TYPE dry_run) {
  int i, x_idx, y;
  const AV1_COMMON *const cm = &cpi->common;
  RD_COUNTS *const rdc = &td->rd_counts;
  MACROBLOCK *const x = &td->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  struct macroblock_plane *const p = x->plane;
  struct macroblockd_plane *const pd = xd->plane;
  MODE_INFO *mi = &ctx->mic;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  MODE_INFO *mi_addr = xd->mi[0];
  const struct segmentation *const seg = &cm->seg;
  const int bw = mi_size_wide[mi->mbmi.sb_type];
  const int bh = mi_size_high[mi->mbmi.sb_type];
  const int x_mis = AOMMIN(bw, cm->mi_cols - mi_col);
  const int y_mis = AOMMIN(bh, cm->mi_rows - mi_row);
  MV_REF *const frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
  int w, h;

  const int mis = cm->mi_stride;
  const int mi_width = mi_size_wide[bsize];
  const int mi_height = mi_size_high[bsize];
  const int unify_bsize = CONFIG_CB4X4;

  int8_t rf_type;

#if !CONFIG_SUPERTX
  assert(mi->mbmi.sb_type == bsize);
#endif

  *mi_addr = *mi;
  *x->mbmi_ext = ctx->mbmi_ext;

#if CONFIG_DUAL_FILTER
  reset_intmv_filter_type(cm, xd, mbmi);
#endif

  rf_type = av1_ref_frame_type(mbmi->ref_frame);
  if (x->mbmi_ext->ref_mv_count[rf_type] > 1 &&
      (mbmi->sb_type >= BLOCK_8X8 || unify_bsize)) {
    set_ref_and_pred_mvs(x, mi->mbmi.pred_mv, rf_type);
  }

  // If segmentation in use
  if (seg->enabled) {
    // For in frame complexity AQ copy the segment id from the segment map.
    if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
      const uint8_t *const map =
          seg->update_map ? cpi->segmentation_map : cm->last_frame_seg_map;
      mi_addr->mbmi.segment_id = get_segment_id(cm, map, bsize, mi_row, mi_col);
      reset_tx_size(xd, &mi_addr->mbmi, cm->tx_mode);
    }
    // Else for cyclic refresh mode update the segment map, set the segment id
    // and then update the quantizer.
    if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
      av1_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi, mi_row, mi_col,
                                        bsize, ctx->rate, ctx->dist, x->skip);
      reset_tx_size(xd, &mi_addr->mbmi, cm->tx_mode);
    }
  }

  for (i = 0; i < MAX_MB_PLANE; ++i) {
    p[i].coeff = ctx->coeff[i];
    p[i].qcoeff = ctx->qcoeff[i];
    pd[i].dqcoeff = ctx->dqcoeff[i];
#if CONFIG_PVQ
    pd[i].pvq_ref_coeff = ctx->pvq_ref_coeff[i];
#endif
    p[i].eobs = ctx->eobs[i];
#if CONFIG_LV_MAP
    p[i].txb_entropy_ctx = ctx->txb_entropy_ctx[i];
#endif  // CONFIG_LV_MAP
  }
#if CONFIG_PALETTE
  for (i = 0; i < 2; ++i) pd[i].color_index_map = ctx->color_index_map[i];
#endif  // CONFIG_PALETTE
  // Restore the coding context of the MB to that that was in place
  // when the mode was picked for it
  for (y = 0; y < mi_height; y++)
    for (x_idx = 0; x_idx < mi_width; x_idx++)
      if ((xd->mb_to_right_edge >> (3 + MI_SIZE_LOG2)) + mi_width > x_idx &&
          (xd->mb_to_bottom_edge >> (3 + MI_SIZE_LOG2)) + mi_height > y) {
        xd->mi[x_idx + y * mis] = mi_addr;
      }

#if CONFIG_DELTA_Q && !CONFIG_EXT_DELTA_Q
  if (cpi->oxcf.aq_mode > NO_AQ && cpi->oxcf.aq_mode < DELTA_AQ)
    av1_init_plane_quantizers(cpi, x, xd->mi[0]->mbmi.segment_id);
#else
  if (cpi->oxcf.aq_mode)
    av1_init_plane_quantizers(cpi, x, xd->mi[0]->mbmi.segment_id);
#endif

  if (is_inter_block(mbmi) && mbmi->sb_type < BLOCK_8X8 && !unify_bsize) {
    mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
    mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
  }

  x->skip = ctx->skip;

#if CONFIG_VAR_TX
  for (i = 0; i < 1; ++i)
    memcpy(x->blk_skip[i], ctx->blk_skip[i],
           sizeof(uint8_t) * ctx->num_4x4_blk);
#endif

  if (dry_run) return;

#if CONFIG_INTERNAL_STATS
  {
    unsigned int *const mode_chosen_counts =
        (unsigned int *)cpi->mode_chosen_counts;  // Cast const away.
    if (frame_is_intra_only(cm)) {
      static const int kf_mode_index[] = {
        THR_DC /*DC_PRED*/,
        THR_V_PRED /*V_PRED*/,
        THR_H_PRED /*H_PRED*/,
        THR_D45_PRED /*D45_PRED*/,
        THR_D135_PRED /*D135_PRED*/,
        THR_D117_PRED /*D117_PRED*/,
        THR_D153_PRED /*D153_PRED*/,
        THR_D207_PRED /*D207_PRED*/,
        THR_D63_PRED /*D63_PRED*/,
#if CONFIG_ALT_INTRA
        THR_SMOOTH, /*SMOOTH_PRED*/
#if CONFIG_SMOOTH_HV
        THR_SMOOTH_V, /*SMOOTH_V_PRED*/
        THR_SMOOTH_H, /*SMOOTH_H_PRED*/
#endif                // CONFIG_SMOOTH_HV
#endif                // CONFIG_ALT_INTRA
        THR_TM /*TM_PRED*/,
      };
      ++mode_chosen_counts[kf_mode_index[mbmi->mode]];
    } else {
      // Note how often each mode chosen as best
      ++mode_chosen_counts[ctx->best_mode_index];
    }
  }
#endif
  if (!frame_is_intra_only(cm)) {
    if (is_inter_block(mbmi)) {
      av1_update_mv_count(td);
#if CONFIG_GLOBAL_MOTION
      if (bsize >= BLOCK_8X8) {
        // TODO(sarahparker): global motion stats need to be handled per-tile
        // to be compatible with tile-based threading.
        update_global_motion_used(mbmi->mode, bsize, mbmi, rdc);
      } else {
        const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
        const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];
        int idx, idy;
        for (idy = 0; idy < 2; idy += num_4x4_h) {
          for (idx = 0; idx < 2; idx += num_4x4_w) {
            const int j = idy * 2 + idx;
            update_global_motion_used(mi->bmi[j].as_mode, bsize, mbmi, rdc);
          }
        }
      }
#endif  // CONFIG_GLOBAL_MOTION
      if (cm->interp_filter == SWITCHABLE
#if CONFIG_WARPED_MOTION
          && mbmi->motion_mode != WARPED_CAUSAL
#endif  // CONFIG_WARPED_MOTION
#if CONFIG_GLOBAL_MOTION
          && !is_nontrans_global_motion(xd)
#endif  // CONFIG_GLOBAL_MOTION
              ) {
#if CONFIG_DUAL_FILTER
        update_filter_type_count(td->counts, xd, mbmi);
#else
        const int switchable_ctx = av1_get_pred_context_switchable_interp(xd);
        ++td->counts->switchable_interp[switchable_ctx][mbmi->interp_filter];
#endif
      }
    }

    rdc->comp_pred_diff[SINGLE_REFERENCE] += ctx->single_pred_diff;
    rdc->comp_pred_diff[COMPOUND_REFERENCE] += ctx->comp_pred_diff;
    rdc->comp_pred_diff[REFERENCE_MODE_SELECT] += ctx->hybrid_pred_diff;
  }

  for (h = 0; h < y_mis; ++h) {
    MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols;
    for (w = 0; w < x_mis; ++w) {
      MV_REF *const mv = frame_mv + w;
      mv->ref_frame[0] = mi->mbmi.ref_frame[0];
      mv->ref_frame[1] = mi->mbmi.ref_frame[1];
      mv->mv[0].as_int = mi->mbmi.mv[0].as_int;
      mv->mv[1].as_int = mi->mbmi.mv[1].as_int;
    }
  }
}

#if CONFIG_SUPERTX
static void update_state_supertx(const AV1_COMP *const cpi, ThreadData *td,
                                 PICK_MODE_CONTEXT *ctx, int mi_row, int mi_col,
                                 BLOCK_SIZE bsize, RUN_TYPE dry_run) {
  int y, x_idx;
#if CONFIG_VAR_TX
  int i;
#endif
  const AV1_COMMON *const cm = &cpi->common;
  RD_COUNTS *const rdc = &td->rd_counts;
  MACROBLOCK *const x = &td->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  MODE_INFO *mi = &ctx->mic;
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  MODE_INFO *mi_addr = xd->mi[0];
  const struct segmentation *const seg = &cm->seg;
  const int mis = cm->mi_stride;
  const int mi_width = mi_size_wide[bsize];
  const int mi_height = mi_size_high[bsize];
  const int x_mis = AOMMIN(mi_width, cm->mi_cols - mi_col);
  const int y_mis = AOMMIN(mi_height, cm->mi_rows - mi_row);
  const int unify_bsize = CONFIG_CB4X4;
  MV_REF *const frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
  int w, h;

  int8_t rf_type;

  *mi_addr = *mi;
  *x->mbmi_ext = ctx->mbmi_ext;
  assert(is_inter_block(mbmi));
  assert(mbmi->tx_size == ctx->mic.mbmi.tx_size);

#if CONFIG_DUAL_FILTER
  reset_intmv_filter_type(cm, xd, mbmi);
#endif

  rf_type = av1_ref_frame_type(mbmi->ref_frame);
  if (x->mbmi_ext->ref_mv_count[rf_type] > 1 &&
      (mbmi->sb_type >= BLOCK_8X8 || unify_bsize)) {
    set_ref_and_pred_mvs(x, mi->mbmi.pred_mv, rf_type);
  }

  // If segmentation in use
  if (seg->enabled) {
    if (cpi->vaq_refresh) {
      const int energy =
          bsize <= BLOCK_16X16 ? x->mb_energy : av1_block_energy(cpi, x, bsize);
      mi_addr->mbmi.segment_id = av1_vaq_segment_id(energy);
    } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
      // For cyclic refresh mode, now update the segment map
      // and set the segment id.
      av1_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi, mi_row, mi_col,
                                        bsize, ctx->rate, ctx->dist, 1);
    } else {
      // Otherwise just set the segment id based on the current segment map
      const uint8_t *const map =
          seg->update_map ? cpi->segmentation_map : cm->last_frame_seg_map;
      mi_addr->mbmi.segment_id = get_segment_id(cm, map, bsize, mi_row, mi_col);
    }
    mi_addr->mbmi.segment_id_supertx = MAX_SEGMENTS;
  }
  // Restore the coding context of the MB to that that was in place
  // when the mode was picked for it
  for (y = 0; y < mi_height; y++)
    for (x_idx = 0; x_idx < mi_width; x_idx++)
      if ((xd->mb_to_right_edge >> (3 + MI_SIZE_LOG2)) + mi_width > x_idx &&
          (xd->mb_to_bottom_edge >> (3 + MI_SIZE_LOG2)) + mi_height > y) {
        xd->mi[x_idx + y * mis] = mi_addr;
      }

#if !CONFIG_CB4X4
  if (is_inter_block(mbmi) && mbmi->sb_type < BLOCK_8X8) {
    mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
    mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
  }
#endif

  x->skip = ctx->skip;

#if CONFIG_VAR_TX
  for (i = 0; i < 1; ++i)
    memcpy(x->blk_skip[i], ctx->blk_skip[i],
           sizeof(uint8_t) * ctx->num_4x4_blk);

  if (!is_inter_block(mbmi) || mbmi->skip)
    mbmi->min_tx_size = get_min_tx_size(mbmi->tx_size);
#endif  // CONFIG_VAR_TX

#if CONFIG_VAR_TX
  {
    const TX_SIZE mtx = mbmi->tx_size;
    const int num_4x4_blocks_wide = tx_size_wide_unit[mtx] >> 1;
    const int num_4x4_blocks_high = tx_size_high_unit[mtx] >> 1;
    int idy, idx;
    mbmi->inter_tx_size[0][0] = mtx;
    for (idy = 0; idy < num_4x4_blocks_high; ++idy)
      for (idx = 0; idx < num_4x4_blocks_wide; ++idx)
        mbmi->inter_tx_size[idy][idx] = mtx;
  }
#endif  // CONFIG_VAR_TX
  // Turn motion variation off for supertx
  mbmi->motion_mode = SIMPLE_TRANSLATION;

  if (dry_run) return;

  if (!frame_is_intra_only(cm)) {
    av1_update_mv_count(td);

#if CONFIG_GLOBAL_MOTION
    if (is_inter_block(mbmi)) {
      if (bsize >= BLOCK_8X8) {
        // TODO(sarahparker): global motion stats need to be handled per-tile
        // to be compatible with tile-based threading.
        update_global_motion_used(mbmi->mode, bsize, mbmi, rdc);
      } else {
        const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
        const int num_4x4_h = num_4x4_blocks_high_lookup[bsize];
        int idx, idy;
        for (idy = 0; idy < 2; idy += num_4x4_h) {
          for (idx = 0; idx < 2; idx += num_4x4_w) {
            const int j = idy * 2 + idx;
            update_global_motion_used(mi->bmi[j].as_mode, bsize, mbmi, rdc);
          }
        }
      }
    }
#endif  // CONFIG_GLOBAL_MOTION

    if (cm->interp_filter == SWITCHABLE
#if CONFIG_GLOBAL_MOTION
        && !is_nontrans_global_motion(xd)
#endif  // CONFIG_GLOBAL_MOTION
            ) {
#if CONFIG_DUAL_FILTER
      update_filter_type_count(td->counts, xd, mbmi);
#else
      const int pred_ctx = av1_get_pred_context_switchable_interp(xd);
      ++td->counts->switchable_interp[pred_ctx][mbmi->interp_filter];
#endif
    }

    rdc->comp_pred_diff[SINGLE_REFERENCE] += ctx->single_pred_diff;
    rdc->comp_pred_diff[COMPOUND_REFERENCE] += ctx->comp_pred_diff;
    rdc->comp_pred_diff[REFERENCE_MODE_SELECT] += ctx->hybrid_pred_diff;
  }

  for (h = 0; h < y_mis; ++h) {
    MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols;
    for (w = 0; w < x_mis; ++w) {
      MV_REF *const mv = frame_mv + w;
      mv->ref_frame[0] = mi->mbmi.ref_frame[0];
      mv->ref_frame[1] = mi->mbmi.ref_frame[1];
      mv->mv[0].as_int = mi->mbmi.mv[0].as_int;
      mv->mv[1].as_int = mi->mbmi.mv[1].as_int;
    }
  }
}

static void update_state_sb_supertx(const AV1_COMP *const cpi, ThreadData *td,
                                    const TileInfo *const tile, int mi_row,
                                    int mi_col, BLOCK_SIZE bsize,
                                    RUN_TYPE dry_run, PC_TREE *pc_tree) {
  const AV1_COMMON *const cm = &cpi->common;
  MACROBLOCK *const x = &td->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  struct macroblock_plane *const p = x->plane;
  struct macroblockd_plane *const pd = xd->plane;
  int hbs = mi_size_wide[bsize] / 2;
#if CONFIG_CB4X4
  const int unify_bsize = 1;
#else
  const int unify_bsize = 0;
#endif
  PARTITION_TYPE partition = pc_tree->partitioning;
  BLOCK_SIZE subsize = get_subsize(bsize, partition);
  int i;
#if CONFIG_EXT_PARTITION_TYPES
  BLOCK_SIZE bsize2 = get_subsize(bsize, PARTITION_SPLIT);
#endif
  PICK_MODE_CONTEXT *pmc = NULL;

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