macroblock.c

JM 11.0 KTA 2.1 Source Code

                rec_resG[j+block_y][i+block_x] = img->m7[j][i];
              
              if (nonzero)
              {
                (*cbp_blk) |= (int64)1 << cbp_blk_mask;  // one bit for every 4x4 block
                (*cbp)     |= cbp_mask;           // one bit for the 4x4 blocks of an 8x8 block
              }
          }
      }
    }
  }
  
  if(need_8x8_transform)
  {
    if (img->NoResidueDirect != 1 && !skipped)
    {
      if (img->type!=SP_SLICE)
#ifdef USE_INTRA_MDDT
        nonzero = dct_luma8x8   (block8x8, &coeff_cost, 0, -1);
#else
      nonzero = dct_luma8x8   (block8x8, &coeff_cost, 0);
#endif
      
      // Residue Color Transform
      if(img->residue_transform_flag)
      {
        for (j=0; j<8; j++)
          for (i=0; i<8; i++)
            rec_resG[mb_y+j][mb_x+i] = img->m7[j][i];
      }
      
      if (nonzero)
      {
        (*cbp_blk) |= 51 << (4*block8x8-2*(block8x8 & 0x01)); // corresponds to 110011, as if all four 4x4 blocks contain coeff, shifted to block position
        (*cbp)     |= cbp_mask;           // one bit for the 4x4 blocks of an 8x8 block
      }
    }
  }
  
  /*
  The purpose of the action below is to prevent that single or 'expensive' coefficients are coded.
  With 4x4 transform there is larger chance that a single coefficient in a 8x8 or 16x16 block may be nonzero.
  A single small (level=1) coefficient in a 8x8 block will cost: 3 or more bits for the coefficient,
  4 bits for EOBs for the 4x4 blocks,possibly also more bits for CBP.  Hence the total 'cost' of that single
  coefficient will typically be 10-12 bits which in a RD consideration is too much to justify the distortion improvement.
  The action below is to watch such 'single' coefficients and set the reconstructed block equal to the prediction according
  to a given criterium.  The action is taken only for inter luma blocks.
  
    Notice that this is a pure encoder issue and hence does not have any implication on the standard.
    coeff_cost is a parameter set in dct_luma() and accumulated for each 8x8 block.  If level=1 for a coefficient,
    coeff_cost is increased by a number depending on RUN for that coefficient.The numbers are (see also dct_luma()): 3,2,2,1,1,1,0,0,...
    when RUN equals 0,1,2,3,4,5,6, etc.
    If level >1 coeff_cost is increased by 9 (or any number above 3). The threshold is set to 3. This means for example:
    1: If there is one coefficient with (RUN,level)=(0,1) in a 8x8 block this coefficient is discarded.
    2: If there are two coefficients with (RUN,level)=(1,1) and (4,1) the coefficients are also discarded
    sum_cnt_nonz is the accumulation of coeff_cost over a whole macro block.  If sum_cnt_nonz is 5 or less for the whole MB,
    all nonzero coefficients are discarded for the MB and the reconstructed block is set equal to the prediction.
  */
  
  if (img->NoResidueDirect != 1 && !skipped && coeff_cost <= _LUMA_COEFF_COST_ &&
    ((img->qp + img->bitdepth_luma_qp_scale)!=0 || img->lossless_qpprime_flag==0)&&
    !(img->type==SP_SLICE && (si_frame_indicator==1 || sp2_frame_indicator==1 )))// last set of conditions
    // cannot skip when perfect reconstruction is as in switching pictures or SI pictures
  {
    coeff_cost  = 0;
    (*cbp)     &=  (63 - cbp_mask);
    (*cbp_blk) &= ~(51 << (4*block8x8-2*(block8x8 & 0x01)));
    
#ifdef ADAPTIVE_FD_SD_CODING
    currMB->SD_or_FD     [block8x8/2][block8x8%2]=0;
    set_bit (&(currMB->SD_or_FD_t8x8),block8x8,0 );
    
    for (j=0;j<4;j++)
    {
      memset(img->cofAC[block8x8][j][0],0,65 * sizeof(int));
      memset(img->cofAC[block8x8][j][1],0,65 * sizeof(int));
    }
#endif
    
    if(!img->residue_transform_flag)
    {
      for (j=mb_y; j<mb_y+8; j++)
        memcpy(&enc_picture->imgY[img->pix_y + j][img->pix_x + mb_x], &img->mpr[j][mb_x], 2 * BLOCK_SIZE * sizeof(imgpel));
    }
    else        // Residue Color Transform
    {
      for (i=mb_x; i<mb_x+8; i++)
        for (j=mb_y; j<mb_y+8; j++)
        {
          rec_resG[j][i] = 0;
        }
    }
    if (img->type==SP_SLICE)
    {
      for (i=mb_x; i < mb_x+BLOCK_SIZE*2; i+=BLOCK_SIZE)
        for (j=mb_y; j < mb_y+BLOCK_SIZE*2; j+=BLOCK_SIZE)
          copyblock_sp(i,j);
    }
  }
  
#ifdef ADAPTIVE_FD_SD_CODING
  if (img->APEC_in_FD_and_SD && currMB->SD_Coding_on_off==1 && !skipped && currMB->b8mode[block8x8] != IBLOCK)
  {
    if (!need_8x8_transform)
    {
      encode_in_spatial_domain(block8x8,cbp_blk, cbp, &coeff_cost);
    }
    else
    {
      encode_in_spatial_domain_8x8(block8x8,cbp_blk, cbp, &coeff_cost);
    }
  }
  
  if (img->NoResidueDirect != 1 && !skipped && coeff_cost <= _LUMA_COEFF_COST_ &&
    ((img->qp + img->bitdepth_luma_qp_scale)!=0 || img->lossless_qpprime_flag==0)&&
    !(img->type==SP_SLICE && (si_frame_indicator==1 || sp2_frame_indicator==1 )))// last set of conditions
    // cannot skip when perfect reconstruction is as in switching pictures or SI pictures
  {
    coeff_cost  = 0;
    (*cbp)     &=  (63 - cbp_mask);
    (*cbp_blk) &= ~(51 << (4*block8x8-2*(block8x8 & 0x01)));
    
    currMB->SD_or_FD     [block8x8/2][block8x8%2]=0;
    set_bit (&(currMB->SD_or_FD_t8x8),block8x8,0 );
    
    for (j=0;j<4;j++)
    {
      memset(img->cofAC[block8x8][j][0],0,65 * sizeof(int));
      memset(img->cofAC[block8x8][j][1],0,65 * sizeof(int));
    }
    
    if(!img->residue_transform_flag)
    {
      
      
      for (j=mb_y; j<mb_y+8; j++)
        memcpy(&enc_picture->imgY[img->pix_y + j][img->pix_x + mb_x], &img->mpr[j][mb_x], 2 * BLOCK_SIZE * sizeof(imgpel));
    }
    else        // Residue Color Transform
    {
      for (i=mb_x; i<mb_x+8; i++)
        for (j=mb_y; j<mb_y+8; j++)
        {
          rec_resG[j][i] = 0;
        }
    }
    if (img->type==SP_SLICE)
    {
      for (i=mb_x; i < mb_x+BLOCK_SIZE*2; i+=BLOCK_SIZE)
        for (j=mb_y; j < mb_y+BLOCK_SIZE*2; j+=BLOCK_SIZE)
          copyblock_sp(i,j);
    }
  }
#endif
  return coeff_cost;
}


/*!
************************************************************************
* \brief
*    Set mode parameters and reference frames for an 8x8 block
************************************************************************
*/
void SetModesAndRefframe (int b8, short* p_dir, int* fw_mode, int* bw_mode, short* fw_ref, short* bw_ref)
{
  Macroblock* currMB = &img->mb_data[img->current_mb_nr];
  int         j      = 2*(b8>>1);
  int         i      = 2*(b8 & 0x01);
  
  *fw_mode = *bw_mode = *fw_ref = *bw_ref = -1;
  
  *p_dir  = currMB->b8pdir[b8];
  
  if (img->type!=B_SLICE)
  {
    *fw_ref = enc_picture->ref_idx[LIST_0][img->block_y+j][img->block_x+i];
    *bw_ref = 0;
    *fw_mode  = currMB->b8mode[b8];
    *bw_mode  = 0;
  }
  else
  {
    if (currMB->b8pdir[b8]==-1)
    {
      *fw_ref   = -1;
      *bw_ref   = -1;
      *fw_mode  =  0;
      *bw_mode  =  0;
    }
    else if (currMB->b8pdir[b8]==0)
    {
      *fw_ref   = enc_picture->ref_idx[LIST_0][img->block_y+j][img->block_x+i];
      *bw_ref   = 0;
      *fw_mode  = currMB->b8mode[b8];
      *bw_mode  = 0;
    }
    else if (currMB->b8pdir[b8]==1)
    {
      *fw_ref   = 0;
      *bw_ref   = enc_picture->ref_idx[LIST_1][img->block_y+j][img->block_x+i];
      *fw_mode  = 0;
      *bw_mode  = currMB->b8mode[b8];
    }
    else
    {
      *fw_ref   = enc_picture->ref_idx[LIST_0][img->block_y+j][img->block_x+i];
      *bw_ref   = enc_picture->ref_idx[LIST_1][img->block_y+j][img->block_x+i];
      *fw_mode  = currMB->b8mode[b8];
      *bw_mode  = currMB->b8mode[b8];
    }
  }
}


/*!
************************************************************************
* \brief
*    Residual Coding of a Luma macroblock (not for intra)
************************************************************************
*/
void LumaResidualCoding ()
{
  int i,j,block8x8,b8_x,b8_y;
  int fw_mode, bw_mode;
  short p_dir, refframe;
  int sum_cnt_nonz;
  Macroblock *currMB = &img->mb_data[img->current_mb_nr];
  
  currMB->cbp     = 0 ;
  currMB->cbp_blk = 0 ;
  sum_cnt_nonz    = 0 ;
  
  for (block8x8=0; block8x8<4; block8x8++)
  {
    short bw_ref;
    SetModesAndRefframe (block8x8, &p_dir, &fw_mode, &bw_mode, &refframe, &bw_ref);
    
    sum_cnt_nonz += LumaResidualCoding8x8 (&(currMB->cbp), &(currMB->cbp_blk), block8x8,
      p_dir, fw_mode, bw_mode, refframe, bw_ref);
  }
  
  if (sum_cnt_nonz <= _LUMA_MB_COEFF_COST_ &&
    ((img->qp + img->bitdepth_luma_qp_scale)!=0 || img->lossless_qpprime_flag==0) &&
    !(img->type==SP_SLICE && (si_frame_indicator==1 || sp2_frame_indicator==1)))// modif ES added last set of conditions
    //cannot skip if SI or switching SP frame perfect reconstruction is needed
  {
    currMB->cbp     &= 0xfffff0 ;
    currMB->cbp_blk &= 0xff0000 ;
    if(!img->residue_transform_flag)
    {
      for (j=0; j < MB_BLOCK_SIZE; j++)
        memcpy(&enc_picture->imgY[img->pix_y+j][img->pix_x], img->mpr[j], MB_BLOCK_SIZE * sizeof (imgpel));
    }
    else
    {
      for (i=0; i < MB_BLOCK_SIZE; i++)
        for (j=0; j < MB_BLOCK_SIZE; j++)
        {
          rec_resG[j][i] = 0;
        }
    }
    
#ifdef ADAPTIVE_FD_SD_CODING
    for (i=0;i<4;i++)
    {
      for (j=0;j<64;j++)
      {
        img->cofAC[i][0][0][j]=0;
        img->cofAC[i][0][1][j]=0;
        img->cofAC[i][1][0][j]=0;
        img->cofAC[i][1][1][j]=0;
        img->cofAC[i][2][0][j]=0;
        img->cofAC[i][2][1][j]=0;
        img->cofAC[i][3][0][j]=0;
        img->cofAC[i][3][1][j]=0;
      }
    }
#endif
    if (img->type==SP_SLICE)
    {
      for(block8x8=0;block8x8<4;block8x8++)
      {
        b8_x=(block8x8&1)<<3;
        b8_y=(block8x8&2)<<2;
        for (i=b8_x;i<b8_x+8;i+=4)
          for (j=b8_y;j<b8_y+8;j+=4)
            copyblock_sp(i,j);
      }
    }
  }
}


/*!
************************************************************************
* \brief
*    Makes the decision if 8x8 tranform will be used (for RD-off)
************************************************************************
*/
int TransformDecision (int block_check, int *cost)
{
  int    block_y, block_x, pic_pix_y, pic_pix_x, i, j, k;
  int    mb_y, mb_x, block8x8;
  int    fw_mode, bw_mode;
  short  p_dir, fw_ref, bw_ref;
  int    num_blks;
  int    cost8x8=0, cost4x4=0;
  int    diff[64], *diff_ptr;
  
  if(block_check==-1)
  {
    block8x8=0;
    num_blks=4;
  }
  else
  {
    block8x8=block_check;
    num_blks=block_check+1;
  }
  
  for (; block8x8<num_blks; block8x8++)
  {
    SetModesAndRefframe (block8x8, &p_dir, &fw_mode, &bw_mode, &fw_ref, &bw_ref);
    
    mb_y = (block8x8 >> 1) << 3;
    mb_x = (block8x8 & 0x01) << 3;
    //===== loop over 4x4 blocks =====
    k=0;
    for (block_y=mb_y; block_y<mb_y+8; block_y+=4)
    {
      pic_pix_y = img->opix_y + block_y;
      
      for (block_x=mb_x; block_x<mb_x+8; block_x+=4)
      {
        pic_pix_x = img->opix_x + block_x;
        
        //===== prediction of 4x4 block =====
        LumaPrediction4x4 (block_x, block_y, p_dir, fw_mode, bw_mode, fw_ref, bw_ref);
        
        //===== get displaced frame difference ======
        diff_ptr=&diff[k];
        for (j=0; j<4; j++)
        {
          for (i=0; i<4; i++, k++)            
            diff[k] = imgY_org[pic_pix_y+j][pic_pix_x+i] - img->mpr[j+block_y][i+block_x];
        }
        cost4x4 += SATD (diff_ptr, input->hadamard);
      }
    }
    cost8x8 += SATD8X8 (diff, input->hadamard);
  }
  
  if(input->Transform8x8Mode==2) //always allow 8x8 transform
    return 1;
  else if(cost8x8<cost4x4)
    return 1;
  else
  {
    *cost = (*cost-cost8x8+cost4x4);
    return 0;
  }
}

/*!
************************************************************************
* \brief
*    Predict one component of a chroma 4x4 block
************************************************************************
*/
void OneComponentChromaPrediction4x4 (int*        mpred,      //!< array to store prediction values
                                      int         block_c_x,  //!< horizontal pixel coordinate of 4x4 block
                                      int         block_c_y,  //!< vertical   pixel coordinate of 4x4 block
                                      short****** mv,         //!< motion vector array
                                      int         list_idx,   //!< reference picture list