vp9_decodframe.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 "vp9/decoder/vp9_onyxd_int.h"
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_header.h"
#include "vp9/common/vp9_reconintra.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/decoder/vp9_decodframe.h"
#include "vp9/decoder/vp9_detokenize.h"
#include "vp9/common/vp9_invtrans.h"
#include "vp9/common/vp9_alloccommon.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_quant_common.h"
#include "vpx_scale/vpx_scale.h"
#include "vp9/common/vp9_setupintrarecon.h"

#include "vp9/decoder/vp9_decodemv.h"
#include "vp9/common/vp9_extend.h"
#include "vp9/common/vp9_modecont.h"
#include "vpx_mem/vpx_mem.h"
#include "vp9/decoder/vp9_dboolhuff.h"

#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_tile_common.h"
#include "vp9_rtcd.h"

#include <assert.h>
#include <stdio.h>

#define COEFCOUNT_TESTING

//#define DEC_DEBUG
#ifdef DEC_DEBUG
int dec_debug = 0;
#endif

static int merge_index(int v, int n, int modulus) {
  int max1 = (n - 1 - modulus / 2) / modulus + 1;
  if (v < max1) v = v * modulus + modulus / 2;
  else {
    int w;
    v -= max1;
    w = v;
    v += (v + modulus - modulus / 2) / modulus;
    while (v % modulus == modulus / 2 ||
           w != v - (v + modulus - modulus / 2) / modulus) v++;
  }
  return v;
}

static int inv_remap_prob(int v, int m) {
  const int n = 256;
  const int modulus = MODULUS_PARAM;
  int i;
  v = merge_index(v, n - 1, modulus);
  if ((m << 1) <= n) {
    i = vp9_inv_recenter_nonneg(v + 1, m);
  } else {
    i = n - 1 - vp9_inv_recenter_nonneg(v + 1, n - 1 - m);
  }
  return i;
}

static vp9_prob read_prob_diff_update(vp9_reader *const bc, int oldp) {
  int delp = vp9_decode_term_subexp(bc, SUBEXP_PARAM, 255);
  return (vp9_prob)inv_remap_prob(delp, oldp);
}

void vp9_init_de_quantizer(VP9D_COMP *pbi) {
  int i;
  int q;
  VP9_COMMON *const pc = &pbi->common;

  for (q = 0; q < QINDEX_RANGE; q++) {
    pc->Y1dequant[q][0] = (int16_t)vp9_dc_quant(q, pc->y1dc_delta_q);
    pc->UVdequant[q][0] = (int16_t)vp9_dc_uv_quant(q, pc->uvdc_delta_q);

    /* all the ac values =; */
    for (i = 1; i < 16; i++) {
      int rc = vp9_default_zig_zag1d_4x4[i];

      pc->Y1dequant[q][rc] = (int16_t)vp9_ac_yquant(q);
      pc->UVdequant[q][rc] = (int16_t)vp9_ac_uv_quant(q, pc->uvac_delta_q);
    }
  }
}

static void mb_init_dequantizer(VP9D_COMP *pbi, MACROBLOCKD *xd) {
  int i;
  int qindex;
  VP9_COMMON *const pc = &pbi->common;
  int segment_id = xd->mode_info_context->mbmi.segment_id;

  // Set the Q baseline allowing for any segment level adjustment
  if (vp9_segfeature_active(xd, segment_id, SEG_LVL_ALT_Q)) {
    if (xd->mb_segment_abs_delta == SEGMENT_ABSDATA)
      /* Abs Value */
      qindex = vp9_get_segdata(xd, segment_id, SEG_LVL_ALT_Q);
    else {
      /* Delta Value */
      qindex = pc->base_qindex +
               vp9_get_segdata(xd, segment_id, SEG_LVL_ALT_Q);
      /* Clamp to valid range */
      qindex = (qindex >= 0) ? ((qindex <= MAXQ) ? qindex : MAXQ) : 0;
    }
  } else
    qindex = pc->base_qindex;

  xd->q_index = qindex;

  /* Set up the block level dequant pointers */
  for (i = 0; i < 16; i++) {
    xd->block[i].dequant = pc->Y1dequant[qindex];
  }

  xd->inv_txm4x4_1      = vp9_short_idct4x4llm_1;
  xd->inv_txm4x4        = vp9_short_idct4x4llm;
  xd->itxm_add          = vp9_dequant_idct_add;
  xd->itxm_add_y_block  = vp9_dequant_idct_add_y_block;
  xd->itxm_add_uv_block = vp9_dequant_idct_add_uv_block;
  if (xd->lossless) {
    assert(qindex == 0);
    xd->inv_txm4x4_1      = vp9_short_inv_walsh4x4_1_x8;
    xd->inv_txm4x4        = vp9_short_inv_walsh4x4_x8;
    xd->itxm_add          = vp9_dequant_idct_add_lossless_c;
    xd->itxm_add_y_block  = vp9_dequant_idct_add_y_block_lossless_c;
    xd->itxm_add_uv_block = vp9_dequant_idct_add_uv_block_lossless_c;
  }

  for (i = 16; i < 24; i++) {
    xd->block[i].dequant = pc->UVdequant[qindex];
  }
}

/* skip_recon_mb() is Modified: Instead of writing the result to predictor buffer and then copying it
 *  to dst buffer, we can write the result directly to dst buffer. This eliminates unnecessary copy.
 */
static void skip_recon_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                          int mb_row, int mb_col) {
  BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;

  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    if (sb_type == BLOCK_SIZE_SB64X64) {
      vp9_build_intra_predictors_sb64uv_s(xd);
      vp9_build_intra_predictors_sb64y_s(xd);
    } else if (sb_type == BLOCK_SIZE_SB32X32) {
      vp9_build_intra_predictors_sbuv_s(xd);
      vp9_build_intra_predictors_sby_s(xd);
    } else {
      vp9_build_intra_predictors_mbuv_s(xd);
      vp9_build_intra_predictors_mby_s(xd);
    }
  } else {
    if (sb_type == BLOCK_SIZE_SB64X64) {
      vp9_build_inter64x64_predictors_sb(xd,
                                         xd->dst.y_buffer,
                                         xd->dst.u_buffer,
                                         xd->dst.v_buffer,
                                         xd->dst.y_stride,
                                         xd->dst.uv_stride,
                                         mb_row, mb_col);
    } else if (sb_type == BLOCK_SIZE_SB32X32) {
      vp9_build_inter32x32_predictors_sb(xd,
                                         xd->dst.y_buffer,
                                         xd->dst.u_buffer,
                                         xd->dst.v_buffer,
                                         xd->dst.y_stride,
                                         xd->dst.uv_stride,
                                         mb_row, mb_col);
    } else {
      vp9_build_inter16x16_predictors_mb(xd,
                                         xd->dst.y_buffer,
                                         xd->dst.u_buffer,
                                         xd->dst.v_buffer,
                                         xd->dst.y_stride,
                                         xd->dst.uv_stride,
                                         mb_row, mb_col);
#if CONFIG_COMP_INTERINTRA_PRED
      if (xd->mode_info_context->mbmi.second_ref_frame == INTRA_FRAME) {
        vp9_build_interintra_16x16_predictors_mb(xd,
                                                 xd->dst.y_buffer,
                                                 xd->dst.u_buffer,
                                                 xd->dst.v_buffer,
                                                 xd->dst.y_stride,
                                                 xd->dst.uv_stride);
      }
#endif
    }
  }
}

static void decode_16x16(VP9D_COMP *pbi, MACROBLOCKD *xd,
                         BOOL_DECODER* const bc) {
  BLOCKD *bd = &xd->block[0];
  TX_TYPE tx_type = get_tx_type_16x16(xd, bd);
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i;
    printf("\n");
    printf("qcoeff 16x16\n");
    for (i = 0; i < 400; i++) {
      printf("%3d ", xd->qcoeff[i]);
      if (i % 16 == 15) printf("\n");
    }
    printf("\n");
    printf("predictor\n");
    for (i = 0; i < 400; i++) {
      printf("%3d ", xd->predictor[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
  if (tx_type != DCT_DCT) {
    vp9_ht_dequant_idct_add_16x16_c(tx_type, xd->qcoeff,
                                    xd->block[0].dequant, xd->predictor,
                                    xd->dst.y_buffer, 16, xd->dst.y_stride,
                                    xd->eobs[0]);
  } else {
    vp9_dequant_idct_add_16x16(xd->qcoeff, xd->block[0].dequant,
                               xd->predictor, xd->dst.y_buffer,
                               16, xd->dst.y_stride, xd->eobs[0]);
  }
  vp9_dequant_idct_add_uv_block_8x8(
      xd->qcoeff + 16 * 16, xd->block[16].dequant,
      xd->predictor + 16 * 16, xd->dst.u_buffer, xd->dst.v_buffer,
      xd->dst.uv_stride, xd);
}

static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd,
                       BOOL_DECODER* const bc) {
  // First do Y
  // if the first one is DCT_DCT assume all the rest are as well
  TX_TYPE tx_type = get_tx_type_8x8(xd, &xd->block[0]);
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i;
    printf("\n");
    printf("qcoeff 8x8\n");
    for (i = 0; i < 400; i++) {
      printf("%3d ", xd->qcoeff[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
  if (tx_type != DCT_DCT || xd->mode_info_context->mbmi.mode == I8X8_PRED) {
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      int idx = (ib & 0x02) ? (ib + 2) : ib;
      int16_t *q  = xd->block[idx].qcoeff;
      int16_t *dq = xd->block[0].dequant;
      uint8_t *pre = xd->block[ib].predictor;
      uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
      int stride = xd->dst.y_stride;
      BLOCKD *b = &xd->block[ib];
      if (xd->mode_info_context->mbmi.mode == I8X8_PRED) {
        int i8x8mode = b->bmi.as_mode.first;
        vp9_intra8x8_predict(xd, b, i8x8mode, b->predictor);
      }
      tx_type = get_tx_type_8x8(xd, &xd->block[ib]);
      if (tx_type != DCT_DCT) {
        vp9_ht_dequant_idct_add_8x8_c(tx_type, q, dq, pre, dst, 16, stride,
                                      xd->eobs[idx]);
      } else {
        vp9_dequant_idct_add_8x8_c(q, dq, pre, dst, 16, stride,
                                   xd->eobs[idx]);
      }
    }
  } else {
    vp9_dequant_idct_add_y_block_8x8(xd->qcoeff,
                                     xd->block[0].dequant,
                                     xd->predictor,
                                     xd->dst.y_buffer,
                                     xd->dst.y_stride,
                                     xd);
  }

  // Now do UV
  if (xd->mode_info_context->mbmi.mode == I8X8_PRED) {
    int i;
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
      b = &xd->block[16 + i];
      vp9_intra_uv4x4_predict(xd, &xd->block[16 + i], i8x8mode, b->predictor);
      xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[16 + i]);
      b = &xd->block[20 + i];
      vp9_intra_uv4x4_predict(xd, &xd->block[20 + i], i8x8mode, b->predictor);
      xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[20 + i]);
    }
  } else if (xd->mode_info_context->mbmi.mode == SPLITMV) {
    xd->itxm_add_uv_block(xd->qcoeff + 16 * 16, xd->block[16].dequant,
         xd->predictor + 16 * 16, xd->dst.u_buffer, xd->dst.v_buffer,
         xd->dst.uv_stride, xd);
  } else {
    vp9_dequant_idct_add_uv_block_8x8
        (xd->qcoeff + 16 * 16, xd->block[16].dequant,
         xd->predictor + 16 * 16, xd->dst.u_buffer, xd->dst.v_buffer,
         xd->dst.uv_stride, xd);
  }
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i;
    printf("\n");
    printf("predictor\n");
    for (i = 0; i < 384; i++) {
      printf("%3d ", xd->predictor[i]);
      if (i % 16 == 15) printf("\n");
    }
  }
#endif
}

static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd,
                       BOOL_DECODER* const bc) {
  TX_TYPE tx_type;
  int i, eobtotal = 0;
  MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  if (mode == I8X8_PRED) {
    for (i = 0; i < 4; i++) {
      int ib = vp9_i8x8_block[i];
      const int iblock[4] = {0, 1, 4, 5};
      int j;
      int i8x8mode;
      BLOCKD *b;
      b = &xd->block[ib];
      i8x8mode = b->bmi.as_mode.first;
      vp9_intra8x8_predict(xd, b, i8x8mode, b->predictor);
      for (j = 0; j < 4; j++) {
        b = &xd->block[ib + iblock[j]];
        tx_type = get_tx_type_4x4(xd, b);
        if (tx_type != DCT_DCT) {
          vp9_ht_dequant_idct_add_c(tx_type, b->qcoeff,
                                    b->dequant, b->predictor,
                                    *(b->base_dst) + b->dst, 16,
                                    b->dst_stride, xd->eobs[ib + iblock[j]]);
        } else {
          xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                       *(b->base_dst) + b->dst, 16, b->dst_stride,
                       xd->eobs[ib + iblock[j]]);
        }
      }
      b = &xd->block[16 + i];
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, b->predictor);
      xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[16 + i]);
      b = &xd->block[20 + i];
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, b->predictor);
      xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                   *(b->base_dst) + b->dst, 8, b->dst_stride, xd->eobs[20 + i]);
    }
  } else if (mode == B_PRED) {
    for (i = 0; i < 16; i++) {
      int b_mode;
      BLOCKD *b = &xd->block[i];
      b_mode = xd->mode_info_context->bmi[i].as_mode.first;
#if CONFIG_NEWBINTRAMODES
      xd->mode_info_context->bmi[i].as_mode.context = b->bmi.as_mode.context =
          vp9_find_bpred_context(b);
#endif
      if (!xd->mode_info_context->mbmi.mb_skip_coeff)
        eobtotal += vp9_decode_coefs_4x4(pbi, xd, bc, PLANE_TYPE_Y_WITH_DC, i);

      vp9_intra4x4_predict(xd, b, b_mode, b->predictor);
      tx_type = get_tx_type_4x4(xd, b);
      if (tx_type != DCT_DCT) {
        vp9_ht_dequant_idct_add_c(tx_type, b->qcoeff,
                                  b->dequant, b->predictor,
                                  *(b->base_dst) + b->dst, 16, b->dst_stride,
                                  xd->eobs[i]);
      } else {
        xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                      *(b->base_dst) + b->dst, 16, b->dst_stride, xd->eobs[i]);
      }
    }
    if (!xd->mode_info_context->mbmi.mb_skip_coeff) {
      vp9_decode_mb_tokens_4x4_uv(pbi, xd, bc);
    }
    vp9_build_intra_predictors_mbuv(xd);
    xd->itxm_add_uv_block(xd->qcoeff + 16 * 16,
                           xd->block[16].dequant,
                           xd->predictor + 16 * 16,
                           xd->dst.u_buffer,
                           xd->dst.v_buffer,
                           xd->dst.uv_stride,
                           xd);
  } else if (mode == SPLITMV || get_tx_type_4x4(xd, &xd->block[0]) == DCT_DCT) {
    xd->itxm_add_y_block(xd->qcoeff,
                          xd->block[0].dequant,
                          xd->predictor,
                          xd->dst.y_buffer,
                          xd->dst.y_stride,
                          xd);
    xd->itxm_add_uv_block(xd->qcoeff + 16 * 16,
                           xd->block[16].dequant,
                           xd->predictor + 16 * 16,
                           xd->dst.u_buffer,
                           xd->dst.v_buffer,
                           xd->dst.uv_stride,
                           xd);
  } else {
#ifdef DEC_DEBUG
    if (dec_debug) {
      int i;
      printf("\n");
      printf("qcoeff 4x4\n");
      for (i = 0; i < 400; i++) {
        printf("%3d ", xd->qcoeff[i]);
        if (i % 16 == 15) printf("\n");
      }
      printf("\n");
      printf("predictor\n");
      for (i = 0; i < 400; i++) {
        printf("%3d ", xd->predictor[i]);
        if (i % 16 == 15) printf("\n");
      }
    }
#endif
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
      tx_type = get_tx_type_4x4(xd, b);
      if (tx_type != DCT_DCT) {
        vp9_ht_dequant_idct_add_c(tx_type, b->qcoeff,
                                  b->dequant, b->predictor,
                                  *(b->base_dst) + b->dst, 16,
                                  b->dst_stride, xd->eobs[i]);
      } else {
        xd->itxm_add(b->qcoeff, b->dequant, b->predictor,
                      *(b->base_dst) + b->dst, 16, b->dst_stride, xd->eobs[i]);
      }
    }
    xd->itxm_add_uv_block(xd->qcoeff + 16 * 16,
                           xd->block[16].dequant,
                           xd->predictor + 16 * 16,
                           xd->dst.u_buffer,
                           xd->dst.v_buffer,
                           xd->dst.uv_stride,
                           xd);
  }
}

static void decode_superblock64(VP9D_COMP *pbi, MACROBLOCKD *xd,
                                int mb_row, int mb_col,
                                BOOL_DECODER* const bc) {
  int n, eobtotal;
  VP9_COMMON *const pc = &pbi->common;
  MODE_INFO *mi = xd->mode_info_context;
  const int mis = pc->mode_info_stride;

  assert(xd->mode_info_context->mbmi.sb_type == BLOCK_SIZE_SB64X64);

  if (pbi->common.frame_type != KEY_FRAME)
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter, pc);

  // re-initialize macroblock dequantizer before detokenization
  if (xd->segmentation_enabled)
    mb_init_dequantizer(pbi, xd);

  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
    vp9_reset_sb64_tokens_context(xd);

    /* Special case:  Force the loopfilter to skip when eobtotal and
     * mb_skip_coeff are zero.
     */
    skip_recon_mb(pbi, xd, mb_row, mb_col);
    return;
  }

  /* do prediction */
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sb64y_s(xd);
    vp9_build_intra_predictors_sb64uv_s(xd);
  } else {
    vp9_build_inter64x64_predictors_sb(xd, xd->dst.y_buffer,
                                       xd->dst.u_buffer, xd->dst.v_buffer,
                                       xd->dst.y_stride, xd->dst.uv_stride,
                                       mb_row, mb_col);
  }

  /* dequantization and idct */
  eobtotal = vp9_decode_sb64_tokens(pbi, xd, bc);
  if (eobtotal == 0) {  // skip loopfilter
    for (n = 0; n < 16; n++) {
      const int x_idx = n & 3, y_idx = n >> 2;

      if (mb_col + x_idx < pc->mb_cols && mb_row + y_idx < pc->mb_rows)
        mi[y_idx * mis + x_idx].mbmi.mb_skip_coeff = mi->mbmi.mb_skip_coeff;
    }
  } else {
    switch (xd->mode_info_context->mbmi.txfm_size) {
      case TX_32X32:
        for (n = 0; n < 4; n++) {
          const int x_idx = n & 1, y_idx = n >> 1;
          vp9_dequant_idct_add_32x32(xd->qcoeff + n * 1024,
              xd->block[0].dequant,
              xd->dst.y_buffer + x_idx * 32 + y_idx * xd->dst.y_stride * 32,
              xd->dst.y_buffer + x_idx * 32 + y_idx * xd->dst.y_stride * 32,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n * 64]);
        }
        vp9_dequant_idct_add_32x32(xd->qcoeff + 4096,
            xd->block[16].dequant, xd->dst.u_buffer, xd->dst.u_buffer,
            xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[256]);
        vp9_dequant_idct_add_32x32(xd->qcoeff + 4096 + 1024,
            xd->block[20].dequant, xd->dst.v_buffer, xd->dst.v_buffer,
            xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[320]);
        break;
      case TX_16X16:  // FIXME(rbultje): adst
        for (n = 0; n < 16; n++) {
          const int x_idx = n & 3, y_idx = n >> 2;
          vp9_dequant_idct_add_16x16(xd->qcoeff + n * 256, xd->block[0].dequant,
              xd->dst.y_buffer + y_idx * 16 * xd->dst.y_stride + x_idx * 16,
              xd->dst.y_buffer + y_idx * 16 * xd->dst.y_stride + x_idx * 16,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n * 16]);
        }
        for (n = 0; n < 4; n++) {
          const int x_idx = n & 1, y_idx = n >> 1;
          vp9_dequant_idct_add_16x16(xd->qcoeff + 4096 + n * 256,
              xd->block[16].dequant,
              xd->dst.u_buffer + y_idx * 16 * xd->dst.uv_stride + x_idx * 16,
              xd->dst.u_buffer + y_idx * 16 * xd->dst.uv_stride + x_idx * 16,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[256 + n * 16]);
          vp9_dequant_idct_add_16x16(xd->qcoeff + 4096 + 1024 + n * 256,
              xd->block[20].dequant,
              xd->dst.v_buffer + y_idx * 16 * xd->dst.uv_stride + x_idx * 16,
              xd->dst.v_buffer + y_idx * 16 * xd->dst.uv_stride + x_idx * 16,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[320 + n * 16]);
        }
        break;
      case TX_8X8:  // FIXME(rbultje): adst
        for (n = 0; n < 64; n++) {
          const int x_idx = n & 7, y_idx = n >> 3;
          vp9_dequant_idct_add_8x8_c(xd->qcoeff + n * 64, xd->block[0].dequant,
              xd->dst.y_buffer + y_idx * 8 * xd->dst.y_stride + x_idx * 8,
              xd->dst.y_buffer + y_idx * 8 * xd->dst.y_stride + x_idx * 8,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n * 4]);
        }
        for (n = 0; n < 16; n++) {
          const int x_idx = n & 3, y_idx = n >> 2;
          vp9_dequant_idct_add_8x8_c(xd->qcoeff + n * 64 + 4096,
              xd->block[16].dequant,
              xd->dst.u_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.u_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[256 + n * 4]);
          vp9_dequant_idct_add_8x8_c(xd->qcoeff + n * 64 + 4096 + 1024,
              xd->block[20].dequant,
              xd->dst.v_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.v_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[320 + n * 4]);
        }
        break;
      case TX_4X4:  // FIXME(rbultje): adst
        for (n = 0; n < 256; n++) {
          const int x_idx = n & 15, y_idx = n >> 4;
          xd->itxm_add(xd->qcoeff + n * 16, xd->block[0].dequant,
              xd->dst.y_buffer + y_idx * 4 * xd->dst.y_stride + x_idx * 4,
              xd->dst.y_buffer + y_idx * 4 * xd->dst.y_stride + x_idx * 4,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n]);
        }
        for (n = 0; n < 64; n++) {
          const int x_idx = n & 7, y_idx = n >> 3;
          xd->itxm_add(xd->qcoeff + 4096 + n * 16,
              xd->block[16].dequant,
              xd->dst.u_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.u_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[256 + n]);
          xd->itxm_add(xd->qcoeff + 4096 + 1024 + n * 16,
              xd->block[20].dequant,
              xd->dst.v_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.v_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[320 + n]);
        }
        break;
      default: assert(0);
    }
  }
}

static void decode_superblock32(VP9D_COMP *pbi, MACROBLOCKD *xd,
                                int mb_row, int mb_col,
                                BOOL_DECODER* const bc) {
  int n, eobtotal;
  VP9_COMMON *const pc = &pbi->common;
  const int mis = pc->mode_info_stride;

  assert(xd->mode_info_context->mbmi.sb_type == BLOCK_SIZE_SB32X32);

  if (pbi->common.frame_type != KEY_FRAME)
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter, pc);

  // re-initialize macroblock dequantizer before detokenization
  if (xd->segmentation_enabled)
    mb_init_dequantizer(pbi, xd);

  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd);

    /* Special case:  Force the loopfilter to skip when eobtotal and
     * mb_skip_coeff are zero.
     */
    skip_recon_mb(pbi, xd, mb_row, mb_col);
    return;
  }

  /* do prediction */
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sby_s(xd);
    vp9_build_intra_predictors_sbuv_s(xd);
  } else {
    vp9_build_inter32x32_predictors_sb(xd, xd->dst.y_buffer,
                                       xd->dst.u_buffer, xd->dst.v_buffer,
                                       xd->dst.y_stride, xd->dst.uv_stride,
                                       mb_row, mb_col);
  }

  /* dequantization and idct */
  eobtotal = vp9_decode_sb_tokens(pbi, xd, bc);
  if (eobtotal == 0) {  // skip loopfilter
    xd->mode_info_context->mbmi.mb_skip_coeff = 1;
    if (mb_col + 1 < pc->mb_cols)
      xd->mode_info_context[1].mbmi.mb_skip_coeff = 1;
    if (mb_row + 1 < pc->mb_rows) {
      xd->mode_info_context[mis].mbmi.mb_skip_coeff = 1;
      if (mb_col + 1 < pc->mb_cols)
        xd->mode_info_context[mis + 1].mbmi.mb_skip_coeff = 1;
    }
  } else {
    switch (xd->mode_info_context->mbmi.txfm_size) {
      case TX_32X32:
        vp9_dequant_idct_add_32x32(xd->qcoeff, xd->block[0].dequant,
                                   xd->dst.y_buffer, xd->dst.y_buffer,
                                   xd->dst.y_stride, xd->dst.y_stride,
                                   xd->eobs[0]);
        vp9_dequant_idct_add_uv_block_16x16_c(xd->qcoeff + 1024,
                                              xd->block[16].dequant,
                                              xd->dst.u_buffer,
                                              xd->dst.v_buffer,
                                              xd->dst.uv_stride, xd);
        break;
      case TX_16X16:  // FIXME(rbultje): adst
        for (n = 0; n < 4; n++) {
          const int x_idx = n & 1, y_idx = n >> 1;
          vp9_dequant_idct_add_16x16(
              xd->qcoeff + n * 256, xd->block[0].dequant,
              xd->dst.y_buffer + y_idx * 16 * xd->dst.y_stride + x_idx * 16,
              xd->dst.y_buffer + y_idx * 16 * xd->dst.y_stride + x_idx * 16,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n * 16]);
        }
        vp9_dequant_idct_add_uv_block_16x16_c(xd->qcoeff + 1024,
                                              xd->block[16].dequant,
                                              xd->dst.u_buffer,
                                              xd->dst.v_buffer,
                                              xd->dst.uv_stride, xd);
        break;
      case TX_8X8:  // FIXME(rbultje): adst
        for (n = 0; n < 16; n++) {
          const int x_idx = n & 3, y_idx = n >> 2;
          vp9_dequant_idct_add_8x8_c(xd->qcoeff + n * 64, xd->block[0].dequant,
              xd->dst.y_buffer + y_idx * 8 * xd->dst.y_stride + x_idx * 8,
              xd->dst.y_buffer + y_idx * 8 * xd->dst.y_stride + x_idx * 8,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n * 4]);
        }
        for (n = 0; n < 4; n++) {
          const int x_idx = n & 1, y_idx = n >> 1;
          vp9_dequant_idct_add_8x8_c(xd->qcoeff + n * 64 + 1024,
              xd->block[16].dequant,
              xd->dst.u_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.u_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[64 + n * 4]);
          vp9_dequant_idct_add_8x8_c(xd->qcoeff + n * 64 + 1280,
              xd->block[20].dequant,
              xd->dst.v_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.v_buffer + y_idx * 8 * xd->dst.uv_stride + x_idx * 8,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[80 + n * 4]);
        }
        break;
      case TX_4X4:  // FIXME(rbultje): adst
        for (n = 0; n < 64; n++) {
          const int x_idx = n & 7, y_idx = n >> 3;
          xd->itxm_add(xd->qcoeff + n * 16, xd->block[0].dequant,
              xd->dst.y_buffer + y_idx * 4 * xd->dst.y_stride + x_idx * 4,
              xd->dst.y_buffer + y_idx * 4 * xd->dst.y_stride + x_idx * 4,
              xd->dst.y_stride, xd->dst.y_stride, xd->eobs[n]);
        }
        for (n = 0; n < 16; n++) {
          const int x_idx = n & 3, y_idx = n >> 2;
          xd->itxm_add(xd->qcoeff + 1024 + n * 16,
              xd->block[16].dequant,
              xd->dst.u_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.u_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[64 + n]);
          xd->itxm_add(xd->qcoeff + 1280 + n * 16,
              xd->block[20].dequant,
              xd->dst.v_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.v_buffer + y_idx * 4 * xd->dst.uv_stride + x_idx * 4,
              xd->dst.uv_stride, xd->dst.uv_stride, xd->eobs[80 + n]);
        }
        break;
      default: assert(0);
    }
  }
}

static void decode_macroblock(VP9D_COMP *pbi, MACROBLOCKD *xd,
                              int mb_row, unsigned int mb_col,
                              BOOL_DECODER* const bc) {
  int eobtotal = 0;
  MB_PREDICTION_MODE mode;
  int tx_size;

  assert(!xd->mode_info_context->mbmi.sb_type);

  // re-initialize macroblock dequantizer before detokenization
  if (xd->segmentation_enabled)
    mb_init_dequantizer(pbi, xd);

  tx_size = xd->mode_info_context->mbmi.txfm_size;
  mode = xd->mode_info_context->mbmi.mode;

  if (xd->mode_info_context->mbmi.mb_skip_coeff) {
    vp9_reset_mb_tokens_context(xd);
  } else if (!bool_error(bc)) {
    if (mode != B_PRED) {
      eobtotal = vp9_decode_mb_tokens(pbi, xd, bc);
    }
  }

  //mode = xd->mode_info_context->mbmi.mode;
  if (pbi->common.frame_type != KEY_FRAME)
    vp9_setup_interp_filters(xd, xd->mode_info_context->mbmi.interp_filter,
                             &pbi->common);

  if (eobtotal == 0 &&
      mode != B_PRED &&
      mode != SPLITMV &&
      mode != I8X8_PRED &&
      !bool_error(bc)) {
    /* Special case:  Force the loopfilter to skip when eobtotal and
       mb_skip_coeff are zero. */
    xd->mode_info_context->mbmi.mb_skip_coeff = 1;
    skip_recon_mb(pbi, xd, mb_row, mb_col);
    return;
  }
#ifdef DEC_DEBUG
  if (dec_debug)
    printf("Decoding mb:  %d %d\n", xd->mode_info_context->mbmi.mode, tx_size);
#endif

  // moved to be performed before detokenization
//  if (xd->segmentation_enabled)
//    mb_init_dequantizer(pbi, xd);

  /* do prediction */
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    if (mode != I8X8_PRED) {
      vp9_build_intra_predictors_mbuv(xd);
      if (mode != B_PRED) {
        vp9_build_intra_predictors_mby(xd);
      }
    }
  } else {
#ifdef DEC_DEBUG
  if (dec_debug)
    printf("Decoding mb:  %d %d interp %d\n",
           xd->mode_info_context->mbmi.mode, tx_size,
           xd->mode_info_context->mbmi.interp_filter);
#endif
    vp9_build_inter_predictors_mb(xd, mb_row, mb_col);
  }

  if (tx_size == TX_16X16) {
    decode_16x16(pbi, xd, bc);
  } else if (tx_size == TX_8X8) {
    decode_8x8(pbi, xd, bc);
  } else {
    decode_4x4(pbi, xd, bc);
  }
#ifdef DEC_DEBUG
  if (dec_debug) {
    int i, j;
    printf("\n");
    printf("final y\n");
    for (i = 0; i < 16; i++) {
      for (j = 0; j < 16; j++)
        printf("%3d ", xd->dst.y_buffer[i * xd->dst.y_stride + j]);
      printf("\n");
    }
    printf("\n");
    printf("final u\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
        printf("%3d ", xd->dst.u_buffer[i * xd->dst.uv_stride + j]);
      printf("\n");
    }
    printf("\n");
    printf("final v\n");
    for (i = 0; i < 8; i++) {
      for (j = 0; j < 8; j++)
        printf("%3d ", xd->dst.v_buffer[i * xd->dst.uv_stride + j]);
      printf("\n");
    }
    fflush(stdout);
  }
#endif
}


static int get_delta_q(vp9_reader *bc, int prev, int *q_update) {
  int ret_val = 0;

  if (vp9_read_bit(bc)) {
    ret_val = vp9_read_literal(bc, 4);

    if (vp9_read_bit(bc))
      ret_val = -ret_val;
  }

  /* Trigger a quantizer update if the delta-q value has changed */
  if (ret_val != prev)
    *q_update = 1;

  return ret_val;
}

#ifdef PACKET_TESTING
#include <stdio.h>
FILE *vpxlog = 0;
#endif

static void set_offsets(VP9D_COMP *pbi, int block_size,
                        int mb_row, int mb_col) {
  VP9_COMMON *const cm = &pbi->common;
  MACROBLOCKD *const xd = &pbi->mb;
  const int mis = cm->mode_info_stride;
  const int idx = mis * mb_row + mb_col;
  const int dst_fb_idx = cm->new_fb_idx;
  const int recon_y_stride = cm->yv12_fb[dst_fb_idx].y_stride;
  const int recon_uv_stride = cm->yv12_fb[dst_fb_idx].uv_stride;
  const int recon_yoffset = mb_row * 16 * recon_y_stride + 16 * mb_col;
  const int recon_uvoffset = mb_row * 8 * recon_uv_stride + 8 * mb_col;

  xd->mode_info_context = cm->mi + idx;
  xd->mode_info_context->mbmi.sb_type = block_size >> 5;
  xd->prev_mode_info_context = cm->prev_mi + idx;
  xd->above_context = cm->above_context + mb_col;
  xd->left_context = cm->left_context + (mb_row & 3);

  /* 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
   */
  block_size >>= 4;  // in mb units
  xd->mb_to_top_edge = -((mb_row * 16)) << 3;
  xd->mb_to_left_edge = -((mb_col * 16) << 3);
  xd->mb_to_bottom_edge = ((cm->mb_rows - block_size - mb_row) * 16) << 3;
  xd->mb_to_right_edge = ((cm->mb_cols - block_size - mb_col) * 16) << 3;

  xd->up_available    = (mb_row != 0);
  xd->left_available  = (mb_col > cm->cur_tile_mb_col_start);
  xd->right_available = (mb_col + block_size < cm->cur_tile_mb_col_end);

  xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset;
  xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset;
  xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset;
}

static void set_refs(VP9D_COMP *pbi, int block_size,
                     int mb_row, int mb_col) {
  VP9_COMMON *const cm = &pbi->common;
  MACROBLOCKD *const xd = &pbi->mb;
  MODE_INFO *mi = xd->mode_info_context;
  MB_MODE_INFO *const mbmi = &mi->mbmi;

  if (mbmi->ref_frame > INTRA_FRAME) {
    int ref_fb_idx;

    /* Select the appropriate reference frame for this MB */
    ref_fb_idx = cm->active_ref_idx[mbmi->ref_frame - 1];
    xd->scale_factor[0] = cm->active_ref_scale[mbmi->ref_frame - 1];
    xd->scale_factor_uv[0] = cm->active_ref_scale[mbmi->ref_frame - 1];
    setup_pred_block(&xd->pre, &cm->yv12_fb[ref_fb_idx], mb_row, mb_col,
                     &xd->scale_factor[0], &xd->scale_factor_uv[0]);

    /* propagate errors from reference frames */
    xd->corrupted |= cm->yv12_fb[ref_fb_idx].corrupted;

    if (mbmi->second_ref_frame > INTRA_FRAME) {
      int second_ref_fb_idx;

      /* Select the appropriate reference frame for this MB */
      second_ref_fb_idx = cm->active_ref_idx[mbmi->second_ref_frame - 1];

      setup_pred_block(&xd->second_pre, &cm->yv12_fb[second_ref_fb_idx],
                       mb_row, mb_col,
                       &xd->scale_factor[1], &xd->scale_factor_uv[1]);

      /* propagate errors from reference frames */
      xd->corrupted |= cm->yv12_fb[second_ref_fb_idx].corrupted;
    }
  }

  if (mbmi->sb_type) {
    const int n_mbs = 1 << mbmi->sb_type;
    const int y_mbs = MIN(n_mbs, cm->mb_rows - mb_row);
    const int x_mbs = MIN(n_mbs, cm->mb_cols - mb_col);
    const int mis = cm->mode_info_stride;
    int x, y;

    for (y = 0; y < y_mbs; y++) {
      for (x = !y; x < x_mbs; x++) {
        mi[y * mis + x] = *mi;
      }
    }
  }
}

/* Decode a row of Superblocks (2x2 region of MBs) */
static void decode_sb_row(VP9D_COMP *pbi, VP9_COMMON *pc,
                          int mb_row, MACROBLOCKD *xd,
                          BOOL_DECODER* const bc) {
  int mb_col;

  // For a SB there are 2 left contexts, each pertaining to a MB row within
  vpx_memset(pc->left_context, 0, sizeof(pc->left_context));

  for (mb_col = pc->cur_tile_mb_col_start;
       mb_col < pc->cur_tile_mb_col_end; mb_col += 4) {
    if (vp9_read(bc, pc->sb64_coded)) {
      set_offsets(pbi, 64, mb_row, mb_col);
      vp9_decode_mb_mode_mv(pbi, xd, mb_row, mb_col, bc);
      set_refs(pbi, 64, mb_row, mb_col);
      decode_superblock64(pbi, xd, mb_row, mb_col, bc);
      xd->corrupted |= bool_error(bc);
    } else {
      int j;

      for (j = 0; j < 4; j++) {
        const int x_idx_sb = (j & 1) << 1, y_idx_sb = j & 2;

        if (mb_row + y_idx_sb >= pc->mb_rows ||
            mb_col + x_idx_sb >= pc->mb_cols) {
          // MB lies outside frame, skip on to next
          continue;
        }

        xd->sb_index = j;

        if (vp9_read(bc, pc->sb32_coded)) {
          set_offsets(pbi, 32, mb_row + y_idx_sb, mb_col + x_idx_sb);
          vp9_decode_mb_mode_mv(pbi,
                                xd, mb_row + y_idx_sb, mb_col + x_idx_sb, bc);
          set_refs(pbi, 32, mb_row + y_idx_sb, mb_col + x_idx_sb);
          decode_superblock32(pbi,
                              xd, mb_row + y_idx_sb, mb_col + x_idx_sb, bc);
          xd->corrupted |= bool_error(bc);
        } else {
          int i;

          // Process the 4 MBs within the SB in the order:
          // top-left, top-right, bottom-left, bottom-right
          for (i = 0; i < 4; i++) {
            const int x_idx = x_idx_sb + (i & 1), y_idx = y_idx_sb + (i >> 1);

            if (mb_row + y_idx >= pc->mb_rows ||
                mb_col + x_idx >= pc->mb_cols) {
              // MB lies outside frame, skip on to next
              continue;
            }

            set_offsets(pbi, 16, mb_row + y_idx, mb_col + x_idx);
            xd->mb_index = i;
            vp9_decode_mb_mode_mv(pbi, xd, mb_row + y_idx, mb_col + x_idx, bc);
            update_blockd_bmi(xd);
            set_refs(pbi, 16, mb_row + y_idx, mb_col + x_idx);
            decode_macroblock(pbi, xd, mb_row + y_idx, mb_col + x_idx, bc);

            /* check if the boolean decoder has suffered an error */
            xd->corrupted |= bool_error(bc);
          }
        }
      }
    }
  }
}

static unsigned int read_partition_size(const unsigned char *cx_size) {
  const unsigned int size =
    cx_size[0] + (cx_size[1] << 8) + (cx_size[2] << 16);