macroblock.c

H.264视频编解码的标准测试模型

    }
  }
  else
  {
    if (p_dir==2)
    {
      for   (j=block_y; j<block_y4; j++)
        for (i=block_x; i<block_x4; i++)  
          img->mpr[j][i] = (*fpred++ + *bpred++ + 1) / 2; 
    }
    else if (p_dir==0)
    {
      for   (j=block_y; j<block_y4; j++)
        for (i=block_x; i<block_x4; i++)  
          img->mpr[j][i] = *fpred++;
    }
    else // p_dir==1
    {
      for   (j=block_y; j<block_y4; j++)
        for (i=block_x; i<block_x4; i++)  
          img->mpr[j][i] = *bpred++;
    }
  }
}



/*!
 ************************************************************************
 * \brief
 *    Residual Coding of an 8x8 Luma block (not for intra)
 *
 * \return
 *    coefficient cost
 ************************************************************************
 */
int LumaResidualCoding8x8 ( int   *cbp,        //!< Output: cbp (updated according to processed 8x8 luminance block)
                            int64 *cbp_blk,    //!< Output: block cbp (updated according to processed 8x8 luminance block)
                            int   block8x8,    //!< block number of 8x8 block
                            short p_dir,       //!< prediction direction
                            int   fw_mode,     //!< list0 prediction mode (1-7, 0=DIRECT)
                            int   bw_mode,     //!< list1 prediction mode (1-7, 0=DIRECT)
                            short fw_refframe, //!< reference picture for list0 prediction
                            short bw_refframe  //!< reference picture for list0 prediction
                           )
{
  int    block_y, block_x, pic_pix_y, pic_pix_x, i, j, nonzero = 0, cbp_blk_mask;
  int    coeff_cost = 0;
  int    mb_y       = (block8x8 >> 1) << 3;
  int    mb_x       = (block8x8 & 0x01) << 3;
  int    pix_y;
  int    cbp_mask   = 1 << block8x8;
  int    bxx, byy;                   // indexing curr_blk
  int    scrFlag    = 0; // 0=noSCR, 1=strongSCR, 2=jmSCR
  int    skipped    = (fw_mode == 0 && bw_mode == 0 && (img->type != B_SLICE));
  Macroblock* currMB = &img->mb_data[img->current_mb_nr];
  //set transform size
  int    need_8x8_transform = currMB->luma_transform_size_8x8_flag;
  // Residue Color Transform
  int residue_R, residue_G, residue_B, temp;

  if (img->type==B_SLICE)
    scrFlag = 1;

  //===== loop over 4x4 blocks =====
  for (byy=0, block_y=mb_y; block_y<mb_y+8; byy+=4, block_y+=4)
  {
    pic_pix_y = img->opix_y + block_y;
    
    for (bxx=0, block_x=mb_x; block_x<mb_x+8; bxx+=4, block_x+=4)
    {
      pic_pix_x = img->opix_x + block_x;
      
      cbp_blk_mask = (block_x>>2) + block_y;
      
      // Residue Color Transform
      if(img->residue_transform_flag)
      {
        ChromaPrediction4x4 (0, block_x, block_y, p_dir, fw_mode, bw_mode, fw_refframe, bw_refframe);
        for (j=0; j<4; j++)
          for (i=0; i<4; i++)
            mprRGB[1][j+block_y][i+block_x] = img->mpr[j+block_y][i+block_x];
          
          ChromaPrediction4x4 (1, block_x, block_y, p_dir, fw_mode, bw_mode, fw_refframe, bw_refframe);
          for (j=0; j<4; j++)
            for (i=0; i<4; i++)
              mprRGB[2][j+block_y][i+block_x] = img->mpr[j+block_y][i+block_x];
      }
      
      //===== prediction of 4x4 block =====
      LumaPrediction4x4 (block_x, block_y, p_dir, fw_mode, bw_mode, fw_refframe, bw_refframe);
      
      // Residue Color Transform
      if(img->residue_transform_flag)
      {
        for (j=0; j<4; j++)
          for (i=0; i<4; i++)
            mprRGB[0][j+block_y][i+block_x] = img->mpr[j+block_y][i+block_x];
      }
      
      //===== get displaced frame difference ======
      if(!img->residue_transform_flag)
      {
        if(!need_8x8_transform)
        {
          
          //===== get displaced frame difference ======                
          for (j=0; j<4; j++)
          {
            pix_y = pic_pix_y + j;
            for (i=0; i<4; i++)
            {
              img->m7[j][i] = imgY_org[pix_y][pic_pix_x + i] - img->mpr[j+block_y][i+block_x];
            }
          } 

          //===== DCT, Quantization, inverse Quantization, IDCT, Reconstruction =====      
          if ( (img->NoResidueDirect != 1 && !skipped  ) ||
               ((img->qp + img->bitdepth_luma_qp_scale)==0 && img->lossless_qpprime_flag==1) )
          {
            //===== DCT, Quantization, inverse Quantization, IDCT, Reconstruction =====
            if (img->type!=SP_SLICE)  
              nonzero = dct_luma   (block_x, block_y, &coeff_cost, 0);
            else if(!si_frame_indicator && !sp2_frame_indicator)     
              nonzero = dct_luma_sp(block_x, block_y, &coeff_cost);// SP frame encoding
            else 
              nonzero = dct_luma_sp2(block_x, block_y, &coeff_cost);//switching SP/SI encoding

            if (nonzero)
            {
              (*cbp_blk) |= 1 << cbp_blk_mask;  // one bit for every 4x4 block
              (*cbp)     |= cbp_mask;           // one bit for the 4x4 blocks of an 8x8 block
            }
          }
        }
        else
        {
          for (j=0; j<4; j++)
          {
            pix_y = pic_pix_y + j;
            for (i=0; i<4; i++)
            {
              img->m7[j+byy][i+bxx] = imgY_org[pix_y][pic_pix_x+i] - img->mpr[j+block_y][i+block_x];
            }
          }
        }
      } 
      else 
      {
        /* Forward Residue Transform */
        for (j=0; j<4; j++)
          for (i=0; i<4; i++)
          {
            residue_B = imgUV_org[0][pic_pix_y+j][pic_pix_x+i] - mprRGB[1][j+block_y][i+block_x];
            residue_G = imgY_org[pic_pix_y+j][pic_pix_x+i] - mprRGB[0][j+block_y][i+block_x];
            residue_R = imgUV_org[1][pic_pix_y+j][pic_pix_x+i] - mprRGB[2][j+block_y][i+block_x];
            
            resTrans_R[j+block_y][i+block_x] = residue_R-residue_B;
            temp = residue_B+(resTrans_R[j+block_y][i+block_x]>>1);
            resTrans_B[j+block_y][i+block_x] = residue_G-temp;
            resTrans_G[j+block_y][i+block_x] = temp+(resTrans_B[j+block_y][i+block_x]>>1);
            
            if(!need_8x8_transform)
              img->m7[j][i] = resTrans_G[j+block_y][i+block_x];
            else
              img->m7[j+byy][i+bxx] = resTrans_G[j+block_y][i+block_x];
          }
          
          // Residue Color Transform
          //===== DCT, Quantization, inverse Quantization, IDCT, Reconstruction =====      
          if (img->NoResidueDirect != 1 && !skipped && !need_8x8_transform )
          {
            //===== DCT, Quantization, inverse Quantization, IDCT, Reconstruction =====
            if (img->type!=SP_SLICE)  nonzero = dct_luma   (block_x, block_y, &coeff_cost, 0);
            else if(!si_frame_indicator && !sp2_frame_indicator)     
              nonzero = dct_luma_sp(block_x, block_y, &coeff_cost);//SP frame encoding
            else 
              nonzero = dct_luma_sp2(block_x, block_y, &coeff_cost);//switching SP of SI encoding
            
            for (j=0; j<4; j++)
              for (i=0; i<4; i++)
                rec_resG[j+block_y][i+block_x] = img->m7[j][i];
              
              if (nonzero)
          {
            (*cbp_blk) |= 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)
        nonzero = dct_luma8x8   (block8x8, &coeff_cost, 0);

      // 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)));

    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);
    }
  }

  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 (void)
{
  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;