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 read_le16(const uint8_t *p) {
  return (p[1] << 8) | p[0];
}

static int read_le32(const uint8_t *p) {
  return (p[3] << 24) | (p[2] << 16) | (p[1] << 8) | p[0];
}

// len == 0 is not allowed
static int read_is_valid(const uint8_t *start, size_t len,
                         const uint8_t *end) {
  return start + len > start && start + len <= end;
}

static TXFM_MODE read_txfm_mode(vp9_reader *r) {
  TXFM_MODE mode = vp9_read_literal(r, 2);
  if (mode == ALLOW_32X32)
    mode += vp9_read_bit(r);
  return mode;
}

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;

  v = merge_index(v, n - 1, modulus);
  if ((m << 1) <= n) {
    return vp9_inv_recenter_nonneg(v + 1, m);
  } else {
    return n - 1 - vp9_inv_recenter_nonneg(v + 1, n - 1 - m);
  }
}

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->y_dequant[q][0] = (int16_t)vp9_dc_quant(q, pc->y1dc_delta_q);
    pc->uv_dequant[q][0] = (int16_t)vp9_dc_uv_quant(q, pc->uvdc_delta_q);

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

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

static int get_qindex(MACROBLOCKD *mb, int segment_id, int base_qindex) {
  // Set the Q baseline allowing for any segment level adjustment
  if (vp9_segfeature_active(mb, segment_id, SEG_LVL_ALT_Q)) {
    const int data = vp9_get_segdata(mb, segment_id, SEG_LVL_ALT_Q);
    return mb->mb_segment_abs_delta == SEGMENT_ABSDATA ?
               data :  // Abs value
               clamp(base_qindex + data, 0, MAXQ);  // Delta value
  } else {
    return base_qindex;
  }
}

static void mb_init_dequantizer(VP9D_COMP *pbi, MACROBLOCKD *mb) {
  int i;

  VP9_COMMON *const pc = &pbi->common;
  const int segment_id = mb->mode_info_context->mbmi.segment_id;
  const int qindex = get_qindex(mb, segment_id, pc->base_qindex);
  mb->q_index = qindex;

  for (i = 0; i < 16; i++)
    mb->block[i].dequant = pc->y_dequant[qindex];

  for (i = 16; i < 24; i++)
    mb->block[i].dequant = pc->uv_dequant[qindex];

  if (mb->lossless) {
    assert(qindex == 0);
    mb->inv_txm4x4_1      = vp9_short_iwalsh4x4_1;
    mb->inv_txm4x4        = vp9_short_iwalsh4x4;
    mb->itxm_add          = vp9_dequant_idct_add_lossless_c;
    mb->itxm_add_y_block  = vp9_dequant_idct_add_y_block_lossless_c;
    mb->itxm_add_uv_block = vp9_dequant_idct_add_uv_block_lossless_c;
  } else {
    mb->inv_txm4x4_1      = vp9_short_idct4x4_1;
    mb->inv_txm4x4        = vp9_short_idct4x4;
    mb->itxm_add          = vp9_dequant_idct_add;
    mb->itxm_add_y_block  = vp9_dequant_idct_add_y_block;
    mb->itxm_add_uv_block = vp9_dequant_idct_add_uv_block;
  }
}

#if CONFIG_CODE_NONZEROCOUNT
static void propagate_nzcs(VP9_COMMON *cm, MACROBLOCKD *xd) {
  MODE_INFO *m = xd->mode_info_context;
  BLOCK_SIZE_TYPE sb_type = m->mbmi.sb_type;
  const int mis = cm->mode_info_stride;
  int n;
  if (sb_type == BLOCK_SIZE_SB64X64) {
    for (n = 0; n < 16; ++n) {
      int i = n >> 2;
      int j = n & 3;
      if (i == 0 && j == 0) continue;
      vpx_memcpy((m + j + mis * i)->mbmi.nzcs, m->mbmi.nzcs,
                 384 * sizeof(m->mbmi.nzcs[0]));
    }
  } else if (sb_type == BLOCK_SIZE_SB32X32) {
    for (n = 0; n < 4; ++n) {
      int i = n >> 1;
      int j = n & 1;
      if (i == 0 && j == 0) continue;
      vpx_memcpy((m + j + mis * i)->mbmi.nzcs, m->mbmi.nzcs,
                 384 * sizeof(m->mbmi.nzcs[0]));
    }
  }
}
#endif

static void skip_recon_sb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                          int mb_row, int mb_col,
                          BLOCK_SIZE_TYPE bsize) {
  MODE_INFO *m = xd->mode_info_context;

  if (m->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sbuv_s(xd, bsize);
    vp9_build_intra_predictors_sby_s(xd, bsize);
  } else {
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
  }
#if CONFIG_CODE_NONZEROCOUNT
  vpx_memset(m->mbmi.nzcs, 0, 384 * sizeof(m->mbmi.nzcs[0]));
  propagate_nzcs(&pbi->common, xd);
#endif
}

static void decode_16x16(VP9D_COMP *pbi, MACROBLOCKD *xd,
                         BOOL_DECODER* const bc) {
  const TX_TYPE tx_type = get_tx_type_16x16(xd, 0);

  if (tx_type != DCT_DCT) {
    vp9_dequant_iht_add_16x16_c(tx_type, xd->plane[0].qcoeff,
                                xd->block[0].dequant, xd->dst.y_buffer,
                                xd->dst.y_buffer, xd->dst.y_stride,
                                xd->dst.y_stride, xd->plane[0].eobs[0]);
  } else {
    vp9_dequant_idct_add_16x16(xd->plane[0].qcoeff, xd->block[0].dequant,
                               xd->dst.y_buffer, xd->dst.y_buffer,
                               xd->dst.y_stride, xd->dst.y_stride,
                               xd->plane[0].eobs[0]);
  }

  vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
                           xd->dst.u_buffer, xd->dst.u_buffer,
                           xd->dst.uv_stride, xd->dst.uv_stride,
                           xd->plane[1].eobs[0]);

  vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[20].dequant,
                           xd->dst.v_buffer, xd->dst.v_buffer,
                           xd->dst.uv_stride, xd->dst.uv_stride,
                           xd->plane[2].eobs[0]);
}

static void decode_8x8(VP9D_COMP *pbi, MACROBLOCKD *xd,
                       BOOL_DECODER* const bc) {
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  // luma
  // if the first one is DCT_DCT assume all the rest are as well
  TX_TYPE tx_type = get_tx_type_8x8(xd, 0);
  if (tx_type != DCT_DCT || 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  = BLOCK_OFFSET(xd->plane[0].qcoeff, idx, 16);
      int16_t *dq = xd->block[0].dequant;
      uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
      int stride = xd->dst.y_stride;
      if (mode == I8X8_PRED) {
        BLOCKD *b = &xd->block[ib];
        int i8x8mode = b->bmi.as_mode.first;
        vp9_intra8x8_predict(xd, b, i8x8mode, dst, stride);
      }
      tx_type = get_tx_type_8x8(xd, ib);
      if (tx_type != DCT_DCT) {
        vp9_dequant_iht_add_8x8_c(tx_type, q, dq, dst, dst, stride, stride,
                                  xd->plane[0].eobs[idx]);
      } else {
        vp9_dequant_idct_add_8x8_c(q, dq, dst, dst, stride, stride,
                                   xd->plane[0].eobs[idx]);
      }
    }
  } else {
    vp9_dequant_idct_add_y_block_8x8(xd->plane[0].qcoeff,
                                     xd->block[0].dequant,
                                     xd->dst.y_buffer,
                                     xd->dst.y_stride,
                                     xd->dst.y_buffer,
                                     xd->dst.y_stride,
                                     xd);
  }

  // chroma
  if (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, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
                   b->dequant, *(b->base_dst) + b->dst,
                   *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                   xd->plane[1].eobs[i]);

      b = &xd->block[20 + i];
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
                   b->dequant, *(b->base_dst) + b->dst,
                   *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                   xd->plane[2].eobs[i]);
    }
  } else if (mode == SPLITMV) {
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
         xd->dst.u_buffer, xd->dst.uv_stride, xd->dst.u_buffer,
         xd->dst.uv_stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
         xd->dst.v_buffer, xd->dst.uv_stride, xd->dst.v_buffer,
         xd->dst.uv_stride, xd->plane[2].eobs);
  } else {
    vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
                             xd->dst.u_buffer, xd->dst.u_buffer,
                             xd->dst.uv_stride, xd->dst.uv_stride,
                             xd->plane[1].eobs[0]);

    vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[16].dequant,
                             xd->dst.v_buffer, xd->dst.v_buffer,
                             xd->dst.uv_stride, xd->dst.uv_stride,
                             xd->plane[2].eobs[0]);
  }
}

static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd,
                       BOOL_DECODER* const bc) {
  TX_TYPE tx_type;
  int i = 0;
  const 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;
      BLOCKD *b = &xd->block[ib];
      int i8x8mode = b->bmi.as_mode.first;
      vp9_intra8x8_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
      for (j = 0; j < 4; j++) {
        b = &xd->block[ib + iblock[j]];
        tx_type = get_tx_type_4x4(xd, ib + iblock[j]);
        if (tx_type != DCT_DCT) {
          vp9_dequant_iht_add_c(tx_type,
              BLOCK_OFFSET(xd->plane[0].qcoeff, ib + iblock[j], 16),
                                    b->dequant, *(b->base_dst) + b->dst,
                                    *(b->base_dst) + b->dst, b->dst_stride,
                                    b->dst_stride,
                                    xd->plane[0].eobs[ib + iblock[j]]);
        } else {
          xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, ib + iblock[j], 16),
                       b->dequant, *(b->base_dst) + b->dst,
                       *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                       xd->plane[0].eobs[ib + iblock[j]]);
        }
      }
      b = &xd->block[16 + i];
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
      xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
                   b->dequant, *(b->base_dst) + b->dst,
                   *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                   xd->plane[1].eobs[i]);
      b = &xd->block[20 + i];
      vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
                              b->dst_stride);
      xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
                   b->dequant, *(b->base_dst) + b->dst,
                   *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                   xd->plane[2].eobs[i]);
    }
  } else if (mode == I4X4_PRED) {
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
      int 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(xd, b);
      if (!xd->mode_info_context->mbmi.mb_skip_coeff)
        vp9_decode_coefs_4x4(pbi, xd, bc, PLANE_TYPE_Y_WITH_DC, i);
#endif
      vp9_intra4x4_predict(xd, b, b_mode, *(b->base_dst) + b->dst,
                           b->dst_stride);
      tx_type = get_tx_type_4x4(xd, i);
      if (tx_type != DCT_DCT) {
        vp9_dequant_iht_add_c(tx_type,
                              BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
                              b->dequant, *(b->base_dst) + b->dst,
                              *(b->base_dst) + b->dst, b->dst_stride,
                              b->dst_stride, xd->plane[0].eobs[i]);
      } else {
        xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
                     b->dequant, *(b->base_dst) + b->dst,
                     *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                     xd->plane[0].eobs[i]);
      }
    }
#if CONFIG_NEWBINTRAMODES
    if (!xd->mode_info_context->mbmi.mb_skip_coeff)
      vp9_decode_mb_tokens_4x4_uv(pbi, xd, bc);
#endif
    vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
         xd->dst.u_buffer, xd->dst.uv_stride, xd->dst.u_buffer,
         xd->dst.uv_stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
         xd->dst.v_buffer, xd->dst.uv_stride, xd->dst.v_buffer,
         xd->dst.uv_stride, xd->plane[2].eobs);
  } else if (mode == SPLITMV || get_tx_type_4x4(xd, 0) == DCT_DCT) {
    xd->itxm_add_y_block(xd->plane[0].qcoeff,
                          xd->block[0].dequant,
                          xd->dst.y_buffer, xd->dst.y_stride,
                          xd->dst.y_buffer,
                          xd->dst.y_stride,
                          xd);
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
         xd->dst.u_buffer, xd->dst.uv_stride, xd->dst.u_buffer,
         xd->dst.uv_stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
         xd->dst.v_buffer, xd->dst.uv_stride, xd->dst.v_buffer,
         xd->dst.uv_stride, xd->plane[2].eobs);
  } else {
    for (i = 0; i < 16; i++) {
      BLOCKD *b = &xd->block[i];
      tx_type = get_tx_type_4x4(xd, i);
      if (tx_type != DCT_DCT) {
        vp9_dequant_iht_add_c(tx_type,
                              BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
                              b->dequant, *(b->base_dst) + b->dst,
                              *(b->base_dst) + b->dst, b->dst_stride,
                              b->dst_stride, xd->plane[0].eobs[i]);
      } else {
        xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, i, 16),
                     b->dequant, *(b->base_dst) + b->dst,
                     *(b->base_dst) + b->dst, b->dst_stride, b->dst_stride,
                     xd->plane[0].eobs[i]);
      }
    }
    xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
                          xd->dst.u_buffer, xd->dst.uv_stride, xd->dst.u_buffer,
                          xd->dst.uv_stride, xd->plane[1].eobs);
    xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
                          xd->dst.v_buffer, xd->dst.uv_stride, xd->dst.v_buffer,
                          xd->dst.uv_stride, xd->plane[2].eobs);
  }
}

static INLINE void decode_sby_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize) - 1, bh = 1 << bhl;
  const int y_count = bw * bh;
  int n;

  for (n = 0; n < y_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
    const int y_offset = (y_idx * 32) * mb->dst.y_stride + (x_idx * 32);
    vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
                               mb->block[0].dequant ,
                               mb->dst.y_buffer + y_offset,
                               mb->dst.y_buffer + y_offset,
                               mb->dst.y_stride, mb->dst.y_stride,
                               mb->plane[0].eobs[n * 64]);
  }
}

static INLINE void decode_sbuv_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize) - 1, bw = (1 << bwl) / 2;
  const int bhl = mb_height_log2(bsize) - 1, bh = (1 << bhl) / 2;
  const int uv_count = bw * bh;
  int n;
  for (n = 0; n < uv_count; n++) {
     const int x_idx = n & (bw - 1);
     const int y_idx = n >> (bwl - 1);
     const int uv_offset = (y_idx * 32) * mb->dst.uv_stride + (x_idx * 32);
     vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
                                mb->block[16].dequant,
                                mb->dst.u_buffer + uv_offset,
                                mb->dst.u_buffer + uv_offset,
                                mb->dst.uv_stride, mb->dst.uv_stride,
                                mb->plane[1].eobs[n * 64]);
     vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 1024),
                                mb->block[20].dequant,
                                mb->dst.v_buffer + uv_offset,
                                mb->dst.v_buffer + uv_offset,
                                mb->dst.uv_stride, mb->dst.uv_stride,
                                mb->plane[2].eobs[n * 64]);
  }
}

static INLINE void decode_sby_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize), bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize), bh = 1 << bhl;
  const int y_count = bw * bh;
  int n;

  for (n = 0; n < y_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
    const int y_offset = (y_idx * 16) * mb->dst.y_stride + (x_idx * 16);
    const TX_TYPE tx_type = get_tx_type_16x16(mb,
                                (y_idx * (4 * bw) + x_idx) * 4);
    if (tx_type == DCT_DCT) {
      vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
                                 mb->block[0].dequant ,
                                 mb->dst.y_buffer + y_offset,
                                 mb->dst.y_buffer + y_offset,
                                 mb->dst.y_stride, mb->dst.y_stride,
                                 mb->plane[0].eobs[n * 16]);
    } else {
      vp9_dequant_iht_add_16x16_c(tx_type,
                                  BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
                                  mb->block[0].dequant,
                                  mb->dst.y_buffer + y_offset,
                                  mb->dst.y_buffer + y_offset,
                                  mb->dst.y_stride, mb->dst.y_stride,
                                  mb->plane[0].eobs[n * 16]);
    }
  }
}

static INLINE void decode_sbuv_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize), bw = (1 << bwl) / 2;
  const int bhl = mb_height_log2(bsize), bh = (1 << bhl) / 2;
  const int uv_count = bw * bh;
  int n;

  assert(bsize >= BLOCK_SIZE_SB32X32);

  for (n = 0; n < uv_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
    const int uv_offset = (y_idx * 16) * mb->dst.uv_stride + (x_idx * 16);
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
                               mb->block[16].dequant,
                               mb->dst.u_buffer + uv_offset,
                               mb->dst.u_buffer + uv_offset,
                               mb->dst.uv_stride, mb->dst.uv_stride,
                               mb->plane[1].eobs[n * 16]);
    vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
                               mb->block[20].dequant,
                               mb->dst.v_buffer + uv_offset,
                               mb->dst.v_buffer + uv_offset,
                               mb->dst.uv_stride, mb->dst.uv_stride,
                               mb->plane[2].eobs[n * 16]);
  }
}

static INLINE void decode_sby_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 1, bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize) + 1, bh = 1 << bhl;
  const int y_count = bw * bh;
  int n;

  // luma
  for (n = 0; n < y_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
    const int y_offset = (y_idx * 8) * xd->dst.y_stride + (x_idx * 8);
    const TX_TYPE tx_type = get_tx_type_8x8(xd,
                                            (y_idx * (2 * bw) + x_idx) * 2);
    if (tx_type == DCT_DCT) {
      vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                                 xd->block[0].dequant,
                                 xd->dst.y_buffer + y_offset,
                                 xd->dst.y_buffer + y_offset,
                                 xd->dst.y_stride, xd->dst.y_stride,
                                 xd->plane[0].eobs[n * 4]);
    } else {
      vp9_dequant_iht_add_8x8_c(tx_type,
                                BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
                                xd->block[0].dequant,
                                xd->dst.y_buffer + y_offset,
                                xd->dst.y_buffer + y_offset,
                                xd->dst.y_stride, xd->dst.y_stride,
                                xd->plane[0].eobs[n * 4]);
    }
  }
}

static INLINE void decode_sbuv_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 1, bw = 1 << (bwl - 1);
  const int bhl = mb_height_log2(bsize) + 1, bh = 1 << (bhl - 1);
  const int uv_count = bw * bh;
  int n;

  // chroma
  for (n = 0; n < uv_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
    const int uv_offset = (y_idx * 8) * xd->dst.uv_stride + (x_idx * 8);
    vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
                               xd->block[16].dequant,
                               xd->dst.u_buffer + uv_offset,
                               xd->dst.u_buffer + uv_offset,
                               xd->dst.uv_stride, xd->dst.uv_stride,
                               xd->plane[1].eobs[n * 4]);
    vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
                               xd->block[20].dequant,
                               xd->dst.v_buffer + uv_offset,
                               xd->dst.v_buffer + uv_offset,
                               xd->dst.uv_stride, xd->dst.uv_stride,
                               xd->plane[2].eobs[n * 4]);
  }
}

static INLINE void decode_sby_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 2, bw = 1 << bwl;
  const int bhl = mb_height_log2(bsize) + 2, bh = 1 << bhl;
  const int y_count = bw * bh;
  int n;

  for (n = 0; n < y_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> bwl;
    const int y_offset = (y_idx * 4) * xd->dst.y_stride + (x_idx * 4);
    const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
    if (tx_type == DCT_DCT) {
      xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                   xd->block[0].dequant,
                   xd->dst.y_buffer + y_offset,
                   xd->dst.y_buffer + y_offset,
                   xd->dst.y_stride, xd->dst.y_stride,
                   xd->plane[0].eobs[n]);
    } else {
      vp9_dequant_iht_add_c(tx_type,
                            BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
                            xd->block[0].dequant,
                            xd->dst.y_buffer + y_offset,
                            xd->dst.y_buffer + y_offset,
                            xd->dst.y_stride,
                            xd->dst.y_stride,
                            xd->plane[0].eobs[n]);
    }
  }
}

static INLINE void decode_sbuv_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize)  + 2, bw = 1 << (bwl - 1);
  const int bhl = mb_height_log2(bsize) + 2, bh = 1 << (bhl - 1);
  const int uv_count = bw * bh;
  int n;

  for (n = 0; n < uv_count; n++) {
    const int x_idx = n & (bw - 1);
    const int y_idx = n >> (bwl - 1);
    const int uv_offset = (y_idx * 4) * xd->dst.uv_stride + (x_idx * 4);
    xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
        xd->block[16].dequant,
        xd->dst.u_buffer + uv_offset,
        xd->dst.u_buffer + uv_offset,
        xd->dst.uv_stride, xd->dst.uv_stride, xd->plane[1].eobs[n]);
    xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 16),
        xd->block[20].dequant,
        xd->dst.v_buffer + uv_offset,
        xd->dst.v_buffer + uv_offset,
        xd->dst.uv_stride, xd->dst.uv_stride, xd->plane[2].eobs[n]);
  }
}

// TODO(jingning): combine luma and chroma dequantization and inverse
// transform into a single function looping over planes.
static void decode_sb_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  decode_sby_32x32(mb, bsize);
  if (bsize == BLOCK_SIZE_SB64X64)
    decode_sbuv_32x32(mb, bsize);
  else
    decode_sbuv_16x16(mb, bsize);
}

static void decode_sb_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
  decode_sby_16x16(mb, bsize);
  if (bsize >= BLOCK_SIZE_SB32X32)
    decode_sbuv_16x16(mb, bsize);
  else
    decode_sbuv_8x8(mb, bsize);
}

static void decode_sb(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
                      BOOL_DECODER* const bc, BLOCK_SIZE_TYPE bsize) {
  const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
  const int bw = 1 << bwl, bh = 1 << bhl;
  int n, eobtotal;
  VP9_COMMON *const pc = &pbi->common;
  MODE_INFO *mi = xd->mode_info_context;
  const int mis = pc->mode_info_stride;

  assert(mi->mbmi.sb_type == bsize);

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

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

  if (mi->mbmi.mb_skip_coeff) {
    vp9_reset_sb_tokens_context(xd, bsize);

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

  // generate prediction
  if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
    vp9_build_intra_predictors_sby_s(xd, bsize);
    vp9_build_intra_predictors_sbuv_s(xd, bsize);
  } else {
    vp9_build_inter_predictors_sb(xd, mb_row, mb_col, bsize);
  }

  // dequantization and idct
  eobtotal = vp9_decode_tokens(pbi, xd, bc, bsize);
  if (eobtotal == 0) {  // skip loopfilter
    for (n = 0; n < bw * bh; n++) {
      const int x_idx = n & (bw - 1), y_idx = n >> bwl;

      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:
        decode_sb_32x32(xd, bsize);
        break;
      case TX_16X16:
        decode_sb_16x16(xd, bsize);
        break;
      case TX_8X8:
        decode_sby_8x8(xd, bsize);
        decode_sbuv_8x8(xd, bsize);
        break;
      case TX_4X4:
        decode_sby_4x4(xd, bsize);
        decode_sbuv_4x4(xd, bsize);
        break;
      default: assert(0);
    }
  }
#if CONFIG_CODE_NONZEROCOUNT
  propagate_nzcs(&pbi->common, xd);
#endif
}

// TODO(jingning): Need to merge SB and MB decoding. The MB decoding currently
// couples special handles on I8x8, B_PRED, and splitmv modes.
static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
                     int mb_row, int mb_col,
                     BOOL_DECODER* const bc) {
  int eobtotal = 0;
  const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
  const int tx_size = xd->mode_info_context->mbmi.txfm_size;

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

  // 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, BLOCK_SIZE_MB16X16);
  } else if (!bool_error(bc)) {
#if CONFIG_NEWBINTRAMODES
    if (mode != I4X4_PRED)
#endif
      eobtotal = vp9_decode_tokens(pbi, xd, bc, BLOCK_SIZE_MB16X16);
  }

  //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 != I4X4_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_sb(pbi, xd, mb_row, mb_col, BLOCK_SIZE_MB16X16);
    return;
  }
#if 0  // def 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_sbuv_s(xd, BLOCK_SIZE_MB16X16);
      if (mode != I4X4_PRED)
        vp9_build_intra_predictors_sby_s(xd, BLOCK_SIZE_MB16X16);
    }
  } else {
#if 0  // def 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_s(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("predictor y\n");
    for (i = 0; i < 16; i++) {
      for (j = 0; j < 16; j++)
        printf("%3d ", xd->predictor[i * 16 + j]);
      printf("\n");
    }
    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);