vp9_decodeframe.c

/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

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

#include "./vp9_rtcd.h"
#include "./vpx_scale_rtcd.h"

#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem_ops.h"
#include "vpx_scale/vpx_scale.h"

#include "vp9/common/vp9_alloccommon.h"
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_idct.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_quant_common.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_tile_common.h"

#include "vp9/decoder/vp9_decodeframe.h"
#include "vp9/decoder/vp9_detokenize.h"
#include "vp9/decoder/vp9_decodemv.h"
#include "vp9/decoder/vp9_decoder.h"
#include "vp9/decoder/vp9_dsubexp.h"
#include "vp9/decoder/vp9_dthread.h"
#include "vp9/decoder/vp9_read_bit_buffer.h"
#include "vp9/decoder/vp9_reader.h"
#include "vp9/decoder/vp9_thread.h"

#define MAX_VP9_HEADER_SIZE 80

static int is_compound_reference_allowed(const VP9_COMMON *cm) {
  int i;
  for (i = 1; i < REFS_PER_FRAME; ++i)
    if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1])
      return 1;

  return 0;
}

static void setup_compound_reference_mode(VP9_COMMON *cm) {
  if (cm->ref_frame_sign_bias[LAST_FRAME] ==
          cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
    cm->comp_fixed_ref = ALTREF_FRAME;
    cm->comp_var_ref[0] = LAST_FRAME;
    cm->comp_var_ref[1] = GOLDEN_FRAME;
  } else if (cm->ref_frame_sign_bias[LAST_FRAME] ==
                 cm->ref_frame_sign_bias[ALTREF_FRAME]) {
    cm->comp_fixed_ref = GOLDEN_FRAME;
    cm->comp_var_ref[0] = LAST_FRAME;
    cm->comp_var_ref[1] = ALTREF_FRAME;
  } else {
    cm->comp_fixed_ref = LAST_FRAME;
    cm->comp_var_ref[0] = GOLDEN_FRAME;
    cm->comp_var_ref[1] = 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 vp9_read_bit_buffer *rb, int max) {
  const int data = vp9_rb_read_literal(rb, get_unsigned_bits(max));
  return data > max ? max : data;
}

static TX_MODE read_tx_mode(vp9_reader *r) {
  TX_MODE tx_mode = vp9_read_literal(r, 2);
  if (tx_mode == ALLOW_32X32)
    tx_mode += vp9_read_bit(r);
  return tx_mode;
}

static void read_tx_mode_probs(struct tx_probs *tx_probs, vp9_reader *r) {
  int i, j;

  for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
    for (j = 0; j < TX_SIZES - 3; ++j)
      vp9_diff_update_prob(r, &tx_probs->p8x8[i][j]);

  for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
    for (j = 0; j < TX_SIZES - 2; ++j)
      vp9_diff_update_prob(r, &tx_probs->p16x16[i][j]);

  for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
    for (j = 0; j < TX_SIZES - 1; ++j)
      vp9_diff_update_prob(r, &tx_probs->p32x32[i][j]);
}

static void read_switchable_interp_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
  int i, j;
  for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
    for (i = 0; i < SWITCHABLE_FILTERS - 1; ++i)
      vp9_diff_update_prob(r, &fc->switchable_interp_prob[j][i]);
}

static void read_inter_mode_probs(FRAME_CONTEXT *fc, vp9_reader *r) {
  int i, j;
  for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
    for (j = 0; j < INTER_MODES - 1; ++j)
      vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
}

static REFERENCE_MODE read_frame_reference_mode(const VP9_COMMON *cm,
                                                vp9_reader *r) {
  if (is_compound_reference_allowed(cm)) {
    return vp9_read_bit(r) ? (vp9_read_bit(r) ? REFERENCE_MODE_SELECT
                                              : COMPOUND_REFERENCE)
                           : SINGLE_REFERENCE;
  } else {
    return SINGLE_REFERENCE;
  }
}

static void read_frame_reference_mode_probs(VP9_COMMON *cm, vp9_reader *r) {
  FRAME_CONTEXT *const fc = &cm->fc;
  int i;

  if (cm->reference_mode == REFERENCE_MODE_SELECT)
    for (i = 0; i < COMP_INTER_CONTEXTS; ++i)
      vp9_diff_update_prob(r, &fc->comp_inter_prob[i]);

  if (cm->reference_mode != COMPOUND_REFERENCE)
    for (i = 0; i < REF_CONTEXTS; ++i) {
      vp9_diff_update_prob(r, &fc->single_ref_prob[i][0]);
      vp9_diff_update_prob(r, &fc->single_ref_prob[i][1]);
    }

  if (cm->reference_mode != SINGLE_REFERENCE)
    for (i = 0; i < REF_CONTEXTS; ++i)
      vp9_diff_update_prob(r, &fc->comp_ref_prob[i]);
}

static void update_mv_probs(vp9_prob *p, int n, vp9_reader *r) {
  int i;
  for (i = 0; i < n; ++i)
    if (vp9_read(r, MV_UPDATE_PROB))
      p[i] = (vp9_read_literal(r, 7) << 1) | 1;
}

static void read_mv_probs(nmv_context *ctx, int allow_hp, vp9_reader *r) {
  int i, j;

  update_mv_probs(ctx->joints, MV_JOINTS - 1, r);

  for (i = 0; i < 2; ++i) {
    nmv_component *const comp_ctx = &ctx->comps[i];
    update_mv_probs(&comp_ctx->sign, 1, r);
    update_mv_probs(comp_ctx->classes, MV_CLASSES - 1, r);
    update_mv_probs(comp_ctx->class0, CLASS0_SIZE - 1, r);
    update_mv_probs(comp_ctx->bits, MV_OFFSET_BITS, r);
  }

  for (i = 0; i < 2; ++i) {
    nmv_component *const comp_ctx = &ctx->comps[i];
    for (j = 0; j < CLASS0_SIZE; ++j)
      update_mv_probs(comp_ctx->class0_fp[j], MV_FP_SIZE - 1, r);
    update_mv_probs(comp_ctx->fp, 3, r);
  }

  if (allow_hp) {
    for (i = 0; i < 2; ++i) {
      nmv_component *const comp_ctx = &ctx->comps[i];
      update_mv_probs(&comp_ctx->class0_hp, 1, r);
      update_mv_probs(&comp_ctx->hp, 1, r);
    }
  }
}

static void setup_plane_dequants(VP9_COMMON *cm, MACROBLOCKD *xd, int q_index) {
  int i;
  xd->plane[0].dequant = cm->y_dequant[q_index];

  for (i = 1; i < MAX_MB_PLANE; i++)
    xd->plane[i].dequant = cm->uv_dequant[q_index];
}

static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
                                    TX_SIZE tx_size, uint8_t *dst, int stride,
                                    int eob) {
  struct macroblockd_plane *const pd = &xd->plane[plane];
  if (eob > 0) {
    TX_TYPE tx_type;
    int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
    if (xd->lossless) {
      tx_type = DCT_DCT;
      vp9_iwht4x4_add(dqcoeff, dst, stride, eob);
    } else {
      const PLANE_TYPE plane_type = pd->plane_type;
      switch (tx_size) {
        case TX_4X4:
          tx_type = get_tx_type_4x4(plane_type, xd, block);
          vp9_iht4x4_add(tx_type, dqcoeff, dst, stride, eob);
          break;
        case TX_8X8:
          tx_type = get_tx_type(plane_type, xd);
          vp9_iht8x8_add(tx_type, dqcoeff, dst, stride, eob);
          break;
        case TX_16X16:
          tx_type = get_tx_type(plane_type, xd);
          vp9_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
          break;
        case TX_32X32:
          tx_type = DCT_DCT;
          vp9_idct32x32_add(dqcoeff, dst, stride, eob);
          break;
        default:
          assert(0 && "Invalid transform size");
      }
    }

    if (eob == 1) {
      vpx_memset(dqcoeff, 0, 2 * sizeof(dqcoeff[0]));
    } else {
      if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
        vpx_memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0]));
      else if (tx_size == TX_32X32 && eob <= 34)
        vpx_memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
      else
        vpx_memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0]));
    }
  }
}

struct intra_args {
  VP9_COMMON *cm;
  MACROBLOCKD *xd;
  vp9_reader *r;
};

static void predict_and_reconstruct_intra_block(int plane, int block,
                                                BLOCK_SIZE plane_bsize,
                                                TX_SIZE tx_size, void *arg) {
  struct intra_args *const args = (struct intra_args *)arg;
  VP9_COMMON *const cm = args->cm;
  MACROBLOCKD *const xd = args->xd;
  struct macroblockd_plane *const pd = &xd->plane[plane];
  MODE_INFO *const mi = xd->mi[0];
  const PREDICTION_MODE mode = (plane == 0) ? get_y_mode(mi, block)
                                            : mi->mbmi.uv_mode;
  int x, y;
  uint8_t *dst;
  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
  dst = &pd->dst.buf[4 * y * pd->dst.stride + 4 * x];

  vp9_predict_intra_block(xd, block >> (tx_size << 1),
                          b_width_log2(plane_bsize), tx_size, mode,
                          dst, pd->dst.stride, dst, pd->dst.stride,
                          x, y, plane);

  if (!mi->mbmi.skip) {
    const int eob = vp9_decode_block_tokens(cm, xd, plane, block,
                                            plane_bsize, x, y, tx_size,
                                            args->r);
    inverse_transform_block(xd, plane, block, tx_size, dst, pd->dst.stride,
                            eob);
  }
}

struct inter_args {
  VP9_COMMON *cm;
  MACROBLOCKD *xd;
  vp9_reader *r;
  int *eobtotal;
};

static void reconstruct_inter_block(int plane, int block,
                                    BLOCK_SIZE plane_bsize,
                                    TX_SIZE tx_size, void *arg) {
  struct inter_args *args = (struct inter_args *)arg;
  VP9_COMMON *const cm = args->cm;
  MACROBLOCKD *const xd = args->xd;
  struct macroblockd_plane *const pd = &xd->plane[plane];
  int x, y, eob;
  txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
  eob = vp9_decode_block_tokens(cm, xd, plane, block, plane_bsize, x, y,
                                tx_size, args->r);
  inverse_transform_block(xd, plane, block, tx_size,
                          &pd->dst.buf[4 * y * pd->dst.stride + 4 * x],
                          pd->dst.stride, eob);
  *args->eobtotal += eob;
}

static MB_MODE_INFO *set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                                 const TileInfo *const tile,
                                 BLOCK_SIZE bsize, int mi_row, int mi_col) {
  const int bw = num_8x8_blocks_wide_lookup[bsize];
  const int bh = num_8x8_blocks_high_lookup[bsize];
  const int x_mis = MIN(bw, cm->mi_cols - mi_col);
  const int y_mis = MIN(bh, cm->mi_rows - mi_row);
  const int offset = mi_row * cm->mi_stride + mi_col;
  int x, y;

  xd->mi = cm->mi_grid_visible + offset;
  xd->mi[0] = &cm->mi[offset];
  xd->mi[0]->mbmi.sb_type = bsize;
  for (y = 0; y < y_mis; ++y)
    for (x = !y; x < x_mis; ++x)
      xd->mi[y * cm->mi_stride + x] = xd->mi[0];

  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, cm->mi_rows, cm->mi_cols);

  vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
  return &xd->mi[0]->mbmi;
}

static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                    int idx, int mi_row, int mi_col) {
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  RefBuffer *ref_buffer = &cm->frame_refs[mbmi->ref_frame[idx] - LAST_FRAME];
  xd->block_refs[idx] = ref_buffer;
  if (!vp9_is_valid_scale(&ref_buffer->sf))
    vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
                       "Invalid scale factors");
  vp9_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
                       &ref_buffer->sf);
  xd->corrupted |= ref_buffer->buf->corrupted;
}

static void decode_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                         const TileInfo *const tile,
                         int mi_row, int mi_col,
                         vp9_reader *r, BLOCK_SIZE bsize) {
  const int less8x8 = bsize < BLOCK_8X8;
  MB_MODE_INFO *mbmi = set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
  vp9_read_mode_info(cm, xd, tile, mi_row, mi_col, r);

  if (less8x8)
    bsize = BLOCK_8X8;

  if (mbmi->skip) {
    reset_skip_context(xd, bsize);
  } else {
    if (cm->seg.enabled)
      setup_plane_dequants(cm, xd, vp9_get_qindex(&cm->seg, mbmi->segment_id,
                                                  cm->base_qindex));
  }

  if (!is_inter_block(mbmi)) {
    struct intra_args arg = { cm, xd, r };
    vp9_foreach_transformed_block(xd, bsize,
                                  predict_and_reconstruct_intra_block, &arg);
  } else {
    // Setup
    set_ref(cm, xd, 0, mi_row, mi_col);
    if (has_second_ref(mbmi))
      set_ref(cm, xd, 1, mi_row, mi_col);

    // Prediction
    vp9_dec_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);

    // Reconstruction
    if (!mbmi->skip) {
      int eobtotal = 0;
      struct inter_args arg = { cm, xd, r, &eobtotal };
      vp9_foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
      if (!less8x8 && eobtotal == 0)
        mbmi->skip = 1;  // skip loopfilter
    }
  }

  xd->corrupted |= vp9_reader_has_error(r);
}

static PARTITION_TYPE read_partition(VP9_COMMON *cm, MACROBLOCKD *xd, int hbs,
                                     int mi_row, int mi_col, BLOCK_SIZE bsize,
                                     vp9_reader *r) {
  const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
  const vp9_prob *const probs = get_partition_probs(cm, ctx);
  const int has_rows = (mi_row + hbs) < cm->mi_rows;
  const int has_cols = (mi_col + hbs) < cm->mi_cols;
  PARTITION_TYPE p;

  if (has_rows && has_cols)
    p = (PARTITION_TYPE)vp9_read_tree(r, vp9_partition_tree, probs);
  else if (!has_rows && has_cols)
    p = vp9_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ;
  else if (has_rows && !has_cols)
    p = vp9_read(r, probs[2]) ? PARTITION_SPLIT : PARTITION_VERT;
  else
    p = PARTITION_SPLIT;

  if (!cm->frame_parallel_decoding_mode)
    ++cm->counts.partition[ctx][p];

  return p;
}

static void decode_partition(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                             const TileInfo *const tile,
                             int mi_row, int mi_col,
                             vp9_reader* r, BLOCK_SIZE bsize) {
  const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
  PARTITION_TYPE partition;
  BLOCK_SIZE subsize;

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

  partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r);
  subsize = get_subsize(bsize, partition);
  if (subsize < BLOCK_8X8) {
    decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
  } else {
    switch (partition) {
      case PARTITION_NONE:
        decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
        break;
      case PARTITION_HORZ:
        decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
        if (mi_row + hbs < cm->mi_rows)
          decode_block(cm, xd, tile, mi_row + hbs, mi_col, r, subsize);
        break;
      case PARTITION_VERT:
        decode_block(cm, xd, tile, mi_row, mi_col, r, subsize);
        if (mi_col + hbs < cm->mi_cols)
          decode_block(cm, xd, tile, mi_row, mi_col + hbs, r, subsize);
        break;
      case PARTITION_SPLIT:
        decode_partition(cm, xd, tile, mi_row,       mi_col,       r, subsize);
        decode_partition(cm, xd, tile, mi_row,       mi_col + hbs, r, subsize);
        decode_partition(cm, xd, tile, mi_row + hbs, mi_col,       r, subsize);
        decode_partition(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize);
        break;
      default:
        assert(0 && "Invalid partition type");
    }
  }

  // update partition context
  if (bsize >= BLOCK_8X8 &&
      (bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
    update_partition_context(xd, mi_row, mi_col, subsize, bsize);
}

static void setup_token_decoder(const uint8_t *data,
                                const uint8_t *data_end,
                                size_t read_size,
                                struct vpx_internal_error_info *error_info,
                                vp9_reader *r,
                                vpx_decrypt_cb decrypt_cb,
                                void *decrypt_state) {
  // Validate the calculated partition length. If the buffer
  // described by the partition can't be fully read, then restrict
  // it to the portion that can be (for EC mode) or throw an error.
  if (!read_is_valid(data, read_size, data_end))
    vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
                       "Truncated packet or corrupt tile length");

  if (vp9_reader_init(r, data, read_size, decrypt_cb, decrypt_state))
    vpx_internal_error(error_info, VPX_CODEC_MEM_ERROR,
                       "Failed to allocate bool decoder %d", 1);
}

static void read_coef_probs_common(vp9_coeff_probs_model *coef_probs,
                                   vp9_reader *r) {
  int i, j, k, l, m;

  if (vp9_read_bit(r))
    for (i = 0; i < PLANE_TYPES; ++i)
      for (j = 0; j < REF_TYPES; ++j)
        for (k = 0; k < COEF_BANDS; ++k)
          for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
            for (m = 0; m < UNCONSTRAINED_NODES; ++m)
              vp9_diff_update_prob(r, &coef_probs[i][j][k][l][m]);
}

static void read_coef_probs(FRAME_CONTEXT *fc, TX_MODE tx_mode,
                            vp9_reader *r) {
    const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
    TX_SIZE tx_size;
    for (tx_size = TX_4X4; tx_size <= max_tx_size; ++tx_size)
      read_coef_probs_common(fc->coef_probs[tx_size], r);
}

static void setup_segmentation(struct segmentation *seg,
                               struct vp9_read_bit_buffer *rb) {
  int i, j;

  seg->update_map = 0;
  seg->update_data = 0;

  seg->enabled = vp9_rb_read_bit(rb);
  if (!seg->enabled)
    return;

  // Segmentation map update
  seg->update_map = vp9_rb_read_bit(rb);
  if (seg->update_map) {
    for (i = 0; i < SEG_TREE_PROBS; i++)
      seg->tree_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8)
                                               : MAX_PROB;

    seg->temporal_update = vp9_rb_read_bit(rb);
    if (seg->temporal_update) {
      for (i = 0; i < PREDICTION_PROBS; i++)
        seg->pred_probs[i] = vp9_rb_read_bit(rb) ? vp9_rb_read_literal(rb, 8)
                                                 : MAX_PROB;
    } else {
      for (i = 0; i < PREDICTION_PROBS; i++)
        seg->pred_probs[i] = MAX_PROB;
    }
  }

  // Segmentation data update
  seg->update_data = vp9_rb_read_bit(rb);
  if (seg->update_data) {
    seg->abs_delta = vp9_rb_read_bit(rb);

    vp9_clearall_segfeatures(seg);

    for (i = 0; i < MAX_SEGMENTS; i++) {
      for (j = 0; j < SEG_LVL_MAX; j++) {
        int data = 0;
        const int feature_enabled = vp9_rb_read_bit(rb);
        if (feature_enabled) {
          vp9_enable_segfeature(seg, i, j);
          data = decode_unsigned_max(rb, vp9_seg_feature_data_max(j));
          if (vp9_is_segfeature_signed(j))
            data = vp9_rb_read_bit(rb) ? -data : data;
        }
        vp9_set_segdata(seg, i, j, data);
      }
    }
  }
}

static void setup_loopfilter(struct loopfilter *lf,
                             struct vp9_read_bit_buffer *rb) {
  lf->filter_level = vp9_rb_read_literal(rb, 6);
  lf->sharpness_level = vp9_rb_read_literal(rb, 3);

  // Read in loop filter deltas applied at the MB level based on mode or ref
  // frame.
  lf->mode_ref_delta_update = 0;

  lf->mode_ref_delta_enabled = vp9_rb_read_bit(rb);
  if (lf->mode_ref_delta_enabled) {
    lf->mode_ref_delta_update = vp9_rb_read_bit(rb);
    if (lf->mode_ref_delta_update) {
      int i;

      for (i = 0; i < MAX_REF_LF_DELTAS; i++)
        if (vp9_rb_read_bit(rb))
          lf->ref_deltas[i] = vp9_rb_read_signed_literal(rb, 6);

      for (i = 0; i < MAX_MODE_LF_DELTAS; i++)
        if (vp9_rb_read_bit(rb))
          lf->mode_deltas[i] = vp9_rb_read_signed_literal(rb, 6);
    }
  }
}

static int read_delta_q(struct vp9_read_bit_buffer *rb, int *delta_q) {
  const int old = *delta_q;
  *delta_q = vp9_rb_read_bit(rb) ? vp9_rb_read_signed_literal(rb, 4) : 0;
  return old != *delta_q;
}

static void setup_quantization(VP9_COMMON *const cm, MACROBLOCKD *const xd,
                               struct vp9_read_bit_buffer *rb) {
  int update = 0;

  cm->base_qindex = vp9_rb_read_literal(rb, QINDEX_BITS);
  update |= read_delta_q(rb, &cm->y_dc_delta_q);
  update |= read_delta_q(rb, &cm->uv_dc_delta_q);
  update |= read_delta_q(rb, &cm->uv_ac_delta_q);
  if (update)
    vp9_init_dequantizer(cm);

  xd->lossless = cm->base_qindex == 0 &&
                 cm->y_dc_delta_q == 0 &&
                 cm->uv_dc_delta_q == 0 &&
                 cm->uv_ac_delta_q == 0;
}

static INTERP_FILTER read_interp_filter(struct vp9_read_bit_buffer *rb) {
  const INTERP_FILTER literal_to_filter[] = { EIGHTTAP_SMOOTH,
                                              EIGHTTAP,
                                              EIGHTTAP_SHARP,
                                              BILINEAR };
  return vp9_rb_read_bit(rb) ? SWITCHABLE
                             : literal_to_filter[vp9_rb_read_literal(rb, 2)];
}

static void read_frame_size(struct vp9_read_bit_buffer *rb,
                            int *width, int *height) {
  const int w = vp9_rb_read_literal(rb, 16) + 1;
  const int h = vp9_rb_read_literal(rb, 16) + 1;
  *width = w;
  *height = h;
}

static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
  cm->display_width = cm->width;
  cm->display_height = cm->height;
  if (vp9_rb_read_bit(rb))
    read_frame_size(rb, &cm->display_width, &cm->display_height);
}

static void apply_frame_size(VP9_COMMON *cm, int width, int height) {
  if (cm->width != width || cm->height != height) {
    // Change in frame size.
    // TODO(agrange) Don't test width/height, check overall size.
    if (width > cm->width || height > cm->height) {
      // Rescale frame buffers only if they're not big enough already.
      if (vp9_resize_frame_buffers(cm, width, height))
        vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                           "Failed to allocate frame buffers");
    }

    cm->width = width;
    cm->height = height;

    vp9_update_frame_size(cm);
  }

  if (vp9_realloc_frame_buffer(
          get_frame_new_buffer(cm), cm->width, cm->height,
          cm->subsampling_x, cm->subsampling_y, VP9_DEC_BORDER_IN_PIXELS,
          &cm->frame_bufs[cm->new_fb_idx].raw_frame_buffer, cm->get_fb_cb,
          cm->cb_priv)) {
    vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                       "Failed to allocate frame buffer");
  }
}

static void setup_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
  int width, height;
  read_frame_size(rb, &width, &height);
  apply_frame_size(cm, width, height);
  setup_display_size(cm, rb);
}

static void setup_frame_size_with_refs(VP9_COMMON *cm,
                                       struct vp9_read_bit_buffer *rb) {
  int width, height;
  int found = 0, i;
  for (i = 0; i < REFS_PER_FRAME; ++i) {
    if (vp9_rb_read_bit(rb)) {
      YV12_BUFFER_CONFIG *const buf = cm->frame_refs[i].buf;
      width = buf->y_crop_width;
      height = buf->y_crop_height;
      found = 1;
      break;
    }
  }

  if (!found)
    read_frame_size(rb, &width, &height);

  if (width <= 0 || height <= 0)
    vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
                       "Referenced frame with invalid size");

  apply_frame_size(cm, width, height);
  setup_display_size(cm, rb);
}

static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
  int min_log2_tile_cols, max_log2_tile_cols, max_ones;
  vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);

  // columns
  max_ones = max_log2_tile_cols - min_log2_tile_cols;
  cm->log2_tile_cols = min_log2_tile_cols;
  while (max_ones-- && vp9_rb_read_bit(rb))
    cm->log2_tile_cols++;

  // rows
  cm->log2_tile_rows = vp9_rb_read_bit(rb);
  if (cm->log2_tile_rows)
    cm->log2_tile_rows += vp9_rb_read_bit(rb);
}

typedef struct TileBuffer {
  const uint8_t *data;
  size_t size;
  int col;  // only used with multi-threaded decoding
} TileBuffer;

// Reads the next tile returning its size and adjusting '*data' accordingly
// based on 'is_last'.
static void get_tile_buffer(const uint8_t *const data_end,
                            int is_last,
                            struct vpx_internal_error_info *error_info,
                            const uint8_t **data,
                            vpx_decrypt_cb decrypt_cb, void *decrypt_state,
                            TileBuffer *buf) {
  size_t size;

  if (!is_last) {
    if (!read_is_valid(*data, 4, data_end))
      vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
                         "Truncated packet or corrupt tile length");

    if (decrypt_cb) {
      uint8_t be_data[4];
      decrypt_cb(decrypt_state, *data, be_data, 4);
      size = mem_get_be32(be_data);
    } else {
      size = mem_get_be32(*data);
    }
    *data += 4;

    if (size > (size_t)(data_end - *data))
      vpx_internal_error(error_info, VPX_CODEC_CORRUPT_FRAME,
                         "Truncated packet or corrupt tile size");
  } else {
    size = data_end - *data;
  }

  buf->data = *data;
  buf->size = size;

  *data += size;
}

static void get_tile_buffers(VP9Decoder *pbi,
                             const uint8_t *data, const uint8_t *data_end,
                             int tile_cols, int tile_rows,
                             TileBuffer (*tile_buffers)[1 << 6]) {
  int r, c;

  for (r = 0; r < tile_rows; ++r) {
    for (c = 0; c < tile_cols; ++c) {
      const int is_last = (r == tile_rows - 1) && (c == tile_cols - 1);
      TileBuffer *const buf = &tile_buffers[r][c];
      buf->col = c;
      get_tile_buffer(data_end, is_last, &pbi->common.error, &data,
                      pbi->decrypt_cb, pbi->decrypt_state, buf);
    }
  }
}

static const uint8_t *decode_tiles(VP9Decoder *pbi,
                                   const uint8_t *data,
                                   const uint8_t *data_end) {
  VP9_COMMON *const cm = &pbi->common;
  const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols);
  const int tile_cols = 1 << cm->log2_tile_cols;
  const int tile_rows = 1 << cm->log2_tile_rows;
  TileBuffer tile_buffers[4][1 << 6];
  int tile_row, tile_col;
  int mi_row, mi_col;
  TileData *tile_data = NULL;

  if (cm->lf.filter_level && pbi->lf_worker.data1 == NULL) {
    CHECK_MEM_ERROR(cm, pbi->lf_worker.data1,
                    vpx_memalign(32, sizeof(LFWorkerData)));
    pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker;
    if (pbi->max_threads > 1 && !vp9_worker_reset(&pbi->lf_worker)) {
      vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
                         "Loop filter thread creation failed");
    }
  }

  if (cm->lf.filter_level) {
    LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
    lf_data->frame_buffer = get_frame_new_buffer(cm);
    lf_data->cm = cm;
    vp9_copy(lf_data->planes, pbi->mb.plane);
    lf_data->stop = 0;
    lf_data->y_only = 0;
    vp9_loop_filter_frame_init(cm, cm->lf.filter_level);
  }

  assert(tile_rows <= 4);
  assert(tile_cols <= (1 << 6));

  // Note: this memset assumes above_context[0], [1] and [2]
  // are allocated as part of the same buffer.
  vpx_memset(cm->above_context, 0,
             sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_cols);

  vpx_memset(cm->above_seg_context, 0,
             sizeof(*cm->above_seg_context) * aligned_cols);

  get_tile_buffers(pbi, data, data_end, tile_cols, tile_rows, tile_buffers);

  if (pbi->tile_data == NULL ||
      (tile_cols * tile_rows) != pbi->total_tiles) {
    vpx_free(pbi->tile_data);
    CHECK_MEM_ERROR(
        cm,
        pbi->tile_data,
        vpx_memalign(32, tile_cols * tile_rows * (sizeof(*pbi->tile_data))));