block.c

JM 11.0 KTA 2.1 Source Code

#else
  levelscale    = LevelScale4x4Luma[1][qp_rem];
  leveloffset   = LevelOffset4x4Luma[1][qp_per];
  invlevelscale = InvLevelScale4x4Luma[1][qp_rem];
#endif

  for (j=0;j<16;j++)
  {
    jdiv = j >> 2;
    jmod = j & 0x03;
    jj = img->opix_y+j;
    for (i=0;i<16;i++)
    {
      // Residue Color Transform
      if(!img->residue_transform_flag)
        M1[j][i]=imgY_org[jj][img->opix_x+i]-img->mprr_2[new_intra_mode][j][i];
      else
        M1[j][i]=img->m7[j][i];

      M0[jdiv][i >> 2][jmod][i & 0x03]=M1[j][i];
    }
  }

  for (jj=0;jj<4 && !lossless_qpprime;jj++)
  {
    for (ii=0;ii<4;ii++)
    {
      for (j=0;j<4;j++)
      {
        M5[0] = M0[jj][ii][j][0] + M0[jj][ii][j][3];
        M5[1] = M0[jj][ii][j][1] + M0[jj][ii][j][2];
        M5[2] = M0[jj][ii][j][1] - M0[jj][ii][j][2];
        M5[3] = M0[jj][ii][j][0] - M0[jj][ii][j][3];

        M4[j][0] = M5[0]   + M5[1];
        M4[j][2] = M5[0]   - M5[1];
        M4[j][1] = M5[3]*2 + M5[2];
        M4[j][3] = M5[3]   - M5[2]*2;
      }
      // vertical
      for (i=0;i<4;i++)
      {
        M5[0] = M4[0][i] + M4[3][i];
        M5[1] = M4[1][i] + M4[2][i];
        M5[2] = M4[1][i] - M4[2][i];
        M5[3] = M4[0][i] - M4[3][i];
        
        M0[jj][ii][0][i] = M5[0]   + M5[1];
        M0[jj][ii][2][i] = M5[0]   - M5[1];
        M0[jj][ii][1][i] = M5[3]*2 + M5[2];
        M0[jj][ii][3][i] = M5[3]   - M5[2]*2;
      }
    }
  }

  // pick out DC coeff

  for (j=0;j<4;j++)
  {
    for (i=0;i<4;i++)
      M4[j][i]= M0[j][i][0][0];
  }

  if (!lossless_qpprime)
  {
    for (j=0;j<4;j++)
    {
      M5[0] = M4[j][0]+M4[j][3];
      M5[1] = M4[j][1]+M4[j][2];
      M5[2] = M4[j][1]-M4[j][2];
      M5[3] = M4[j][0]-M4[j][3];
      
      M4[j][0] = M5[0]+M5[1];
      M4[j][2] = M5[0]-M5[1];
      M4[j][1] = M5[3]+M5[2];
      M4[j][3] = M5[3]-M5[2];
    }
    
    // vertical
    
    for (i=0;i<4;i++)
    {    
      M5[0] = M4[0][i]+M4[3][i];
      M5[1] = M4[1][i]+M4[2][i];
      M5[2] = M4[1][i]-M4[2][i];
      M5[3] = M4[0][i]-M4[3][i];
      
      M4[0][i]=(M5[0]+M5[1])>>1;
      M4[2][i]=(M5[0]-M5[1])>>1;
      M4[1][i]=(M5[3]+M5[2])>>1;
      M4[3][i]=(M5[3]-M5[2])>>1;
    }
  }
  // quant

  run=-1;
  scan_pos=0;

  for (coeff_ctr=0;coeff_ctr<16;coeff_ctr++)
  {
    i=pos_scan[coeff_ctr][0];
    j=pos_scan[coeff_ctr][1];

    run++;

    if(lossless_qpprime)
      level= absm(M4[j][i]);
    else
      level= (absm(M4[j][i]) * levelscale[0][0] + (leveloffset[0][0]<<1)) >> (q_bits+1);

    if (input->symbol_mode == UVLC && img->qp < 10) 
    {
      if (level > CAVLC_LEVEL_LIMIT) 
        level = CAVLC_LEVEL_LIMIT;
    }

    if (level != 0)
    {
      DCLevel[scan_pos] = sign(level,M4[j][i]);
      DCRun  [scan_pos] = run;
      ++scan_pos;
      run=-1;
    }
    if(!lossless_qpprime)
      M4[j][i]=sign(level,M4[j][i]);
  }
  DCLevel[scan_pos]=0;

  // invers DC transform
  for (j=0;j<4 && !lossless_qpprime;j++)
  {
    M6[0]=M4[j][0]+M4[j][2];
    M6[1]=M4[j][0]-M4[j][2];
    M6[2]=M4[j][1]-M4[j][3];
    M6[3]=M4[j][1]+M4[j][3];

    M4[j][0] = M6[0]+M6[3];
    M4[j][1] = M6[1]+M6[2];
    M4[j][2] = M6[1]-M6[2];
    M4[j][3] = M6[0]-M6[3];
  }

  for (i=0;i<4 && !lossless_qpprime;i++)
  {
    
    M6[0]=M4[0][i]+M4[2][i];
    M6[1]=M4[0][i]-M4[2][i];
    M6[2]=M4[1][i]-M4[3][i];
    M6[3]=M4[1][i]+M4[3][i];
    
    if(qp_per<6)
    {
      M0[0][i][0][0] = ((M6[0]+M6[3])*invlevelscale[0][0]+(1<<(5-qp_per)))>>(6-qp_per);
      M0[1][i][0][0] = ((M6[1]+M6[2])*invlevelscale[0][0]+(1<<(5-qp_per)))>>(6-qp_per);
      M0[2][i][0][0] = ((M6[1]-M6[2])*invlevelscale[0][0]+(1<<(5-qp_per)))>>(6-qp_per);
      M0[3][i][0][0] = ((M6[0]-M6[3])*invlevelscale[0][0]+(1<<(5-qp_per)))>>(6-qp_per);
    }
    else
    {
      M0[0][i][0][0] = ((M6[0]+M6[3])*invlevelscale[0][0])<<(qp_per-6);
      M0[1][i][0][0] = ((M6[1]+M6[2])*invlevelscale[0][0])<<(qp_per-6);
      M0[2][i][0][0] = ((M6[1]-M6[2])*invlevelscale[0][0])<<(qp_per-6);
      M0[3][i][0][0] = ((M6[0]-M6[3])*invlevelscale[0][0])<<(qp_per-6);
    }   
  }

  // AC inverse trans/quant for MB
  for (jj=0;jj<4;jj++)
  {
    for (ii=0;ii<4;ii++)
    {
      for (j=0;j<4;j++)
      {
        memcpy(M4[j],M0[jj][ii][j], BLOCK_SIZE * sizeof(int));
      }

      run      = -1;
      scan_pos =  0;
      b8       = 2*(jj >> 1) + (ii >> 1);
      b4       = 2*(jj & 0x01) + (ii & 0x01);
      ACLevel  = img->cofAC [b8][b4][0];
      ACRun    = img->cofAC [b8][b4][1];

#ifdef RDO_Q
//#ifdef TREL_CAVLC
      if(input->UseRDO_Q && active_pps->entropy_coding_mode_flag == UVLC)
        TrellisCAVLC4x4(M4, q_bits, qp_rem, levelscale, leveloffset, levelTrellis, LUMA_INTRA16x16AC, b8, b4, 15, lambda_md);
//#endif
      if(input->UseRDO_Q && active_pps->entropy_coding_mode_flag == CABAC)
      {
        kStart=0; kStop=0; noCoeff=0;
        for (coeff_ctr=0;coeff_ctr < 15;coeff_ctr++)
        {
          i=pos_scan[coeff_ctr+1][0]; // scan is shifted due to DC
          j=pos_scan[coeff_ctr+1][1]; // scan is shifted due to DC

          levelData[coeff_ctr].levelDouble=absm(M4[j][i]*levelscale[i][j]);
          level = (int)(levelData[coeff_ctr].levelDouble >> q_bits);
          lowerInt=(((int)levelData[coeff_ctr].levelDouble-(level<<q_bits))<(1<<(q_bits-1)))? 1 : 0;

          levelData[coeff_ctr].level[0]=0;
          if (level==0 && lowerInt==1)
          {
            levelData[coeff_ctr].noLevels=1;
          }
          else if (level==0 && lowerInt==0)
          {
            levelData[coeff_ctr].level[1] = level+1;
            levelData[coeff_ctr].noLevels=2;
            kStop=coeff_ctr;
            noCoeff++;
          }
          else if (level>0 && lowerInt==1)
          {
            levelData[coeff_ctr].level[1] = level;
            levelData[coeff_ctr].noLevels=2;
            kStop=coeff_ctr;
            noCoeff++;
          }
          else
          {
            levelData[coeff_ctr].level[1] = level;
            levelData[coeff_ctr].level[2] = level+1;
            levelData[coeff_ctr].noLevels=3;
            kStop=coeff_ctr;
            kStart=coeff_ctr;
            noCoeff++;
          }

          for (k=0; k<levelData[coeff_ctr].noLevels; k++)
          {
            err=(double)(levelData[coeff_ctr].level[k]<<q_bits)-(double)levelData[coeff_ctr].levelDouble;
            levelData[coeff_ctr].errLevel[k]=(err*err*(double)estErr4x4[qp_rem][i][j])/normFact; 
          }
        }
        estBits=est_write_and_store_CBP_block_bit(currMB, LUMA_16AC);

        est_writeRunLevel_CABAC(levelData, levelTrellis, LUMA_16AC, lambda_md, kStart, kStop, noCoeff, estBits);
      }
#endif

      for (coeff_ctr=1;coeff_ctr<16;coeff_ctr++) // set in AC coeff
      {

        i=pos_scan[coeff_ctr][0];
        j=pos_scan[coeff_ctr][1];

        run++;
#ifdef RDO_Q
        if(input->UseRDO_Q)
        {
          level=levelTrellis[coeff_ctr-1];
        }
        else
        {
          if(lossless_qpprime)
            level= absm( M4[j][i]);
          else          
            level= ( absm( M4[j][i]) * levelscale[i][j] + leveloffset[i][j]) >> q_bits;
        }
#else
        if(lossless_qpprime)
          level= absm( M4[j][i]);
        else          
          level= ( absm( M4[j][i]) * levelscale[i][j] + leveloffset[i][j]) >> q_bits;
#endif


        if (img->AdaptiveRounding)
        {
          if (lossless_qpprime || level == 0 )
          {
            img->fadjust4x4[2][jj*BLOCK_SIZE+j][ii*BLOCK_SIZE+i] = 0;
          }
          else
          {
            img->fadjust4x4[2][jj*BLOCK_SIZE+j][ii*BLOCK_SIZE+i] = 
              (AdaptRndWeight * (absm(M4[j][i]) * levelscale[i][j] - (level << q_bits)) + (1<< (q_bits))) >> (q_bits + 1);
          }
        }

        if (level != 0)
        {
          ac_coef = 15;
          ACLevel[scan_pos] = sign(level,M4[j][i]);
          ACRun  [scan_pos] = run;
          ++scan_pos;
          run=-1;
        }
        
        if(!lossless_qpprime)
        {
          level=sign(level, M4[j][i]);
          if(qp_per<4)
            M4[j][i]=(level*invlevelscale[i][j]+(1<<(3-qp_per)))>>(4-qp_per);
          else
            M4[j][i]=(level*invlevelscale[i][j])<<(qp_per-4);
        }
      }
      ACLevel[scan_pos] = 0;


      // IDCT horizontal
      for (j=0;j<4 && !lossless_qpprime;j++)
      {
        M6[0] = M4[j][0]     +  M4[j][2];
        M6[1] = M4[j][0]     -  M4[j][2];
        M6[2] =(M4[j][1]>>1) -  M4[j][3];
        M6[3] = M4[j][1]     + (M4[j][3]>>1);
        
        M4[j][0] = M6[0] + M6[3];
        M4[j][1] = M6[1] + M6[2];
        M4[j][2] = M6[1] - M6[2];
        M4[j][3] = M6[0] - M6[3];
      }

      // vert
      for (i=0;i<4 && !lossless_qpprime;i++)
      {
        M6[0]= M4[0][i]     +  M4[2][i];
        M6[1]= M4[0][i]     -  M4[2][i];
        M6[2]=(M4[1][i]>>1) -  M4[3][i];
        M6[3]= M4[1][i]     + (M4[3][i]>>1);
        
        M0[jj][ii][0][i] = M6[0] + M6[3];
        M0[jj][ii][1][i] = M6[1] + M6[2];
        M0[jj][ii][2][i] = M6[1] - M6[2];
        M0[jj][ii][3][i] = M6[0] - M6[3];
      }
    }
  }

  // Residue Color Transform
  if(!img->residue_transform_flag)
  {
    for (jj=0;jj<BLOCK_MULTIPLE; jj++)
      for (ii=0;ii<BLOCK_MULTIPLE; ii++)
        for (j=0;j<BLOCK_SIZE;j++)
        {
          memcpy(&M1[jj*BLOCK_SIZE + j][ii*BLOCK_SIZE], M0[jj][ii][j], BLOCK_SIZE * sizeof(int));
        }
  }
  else
  {
    if(lossless_qpprime)
    {
      for (j=0;j<MB_BLOCK_SIZE;j++)    
      {
        jdiv = j >> 2;
        jmod = j & 0x03;
        for (i=0;i<MB_BLOCK_SIZE;i++)
          img->m7[j][i]=M0[jdiv][i >> 2][jmod][i & 0x03];
      }
    }
    else
    {
      for (j=0;j<MB_BLOCK_SIZE;j++)    
      {
        jdiv = j >> 2;
        jmod = j & 0x03;
        for (i=0;i<MB_BLOCK_SIZE;i++)
          img->m7[j][i]=((M0[jdiv][i >> 2][jmod][i & 0x03]+DQ_ROUND)>>DQ_BITS);
      }
    }
  }

  if(!img->residue_transform_flag)
  {
    if(lossless_qpprime)
    {
      for (j=0;j<16;j++)
      {
        jj = img->pix_y+j;
        for (i=0;i<16;i++)
        {
          enc_picture->imgY[jj][img->pix_x+i]=(imgpel)(M1[j][i]+img->mprr_2[new_intra_mode][j][i]);
          if(img->type==SP_SLICE)
            lrec[jj][img->pix_x+i]=-16; //signals an I16 block in the SP frame 
        }
      }
    }
    else
    {
      for (j=0;j<16;j++)
      {
        jj = img->pix_y+j;
        for (i=0;i<16;i++)
        {
          enc_picture->imgY[jj][img->pix_x+i]=(imgpel)clip1a((M1[j][i]+((long)img->mprr_2[new_intra_mode][j][i]<<DQ_BITS)+DQ_ROUND)>>DQ_BITS);
          if(img->type==SP_SLICE)
            lrec[jj][img->pix_x+i]=-16; //signals an I16 block in the SP frame 
        }
      }
    }
  }
  return ac_coef;
}


/*!
************************************************************************
* \brief
*    The routine performs transform,quantization,inverse transform, adds the diff.
*    to the prediction and writes the result to the decoded luma frame. Includes the
*    RD constrained quantization also.
*
* \par Input:
*    block_x,block_y: Block position inside a macro block (0,4,8,12).
*
* \par Output_
*    nonzero: 0 if no levels are nonzero.  1 if there are nonzero levels.             \n
*    coeff_cost: Counter for nonzero coefficients, used to discard expensive levels.
************************************************************************
*/
#ifdef USE_INTRA_MDDT
int dct_luma(int block_x,int block_y,int *coeff_cost, int intra, int ipmode)
#else
int dct_luma(int block_x,int block_y,int *coeff_cost, int intra)
#endif
{
  int sign(int a,int b);

  int i,j,ilev, m4[4][4], m5[4],m6[4],coeff_ctr;
  int ii;
  //int qp_const;
  int level,scan_pos,run;
  int nonzero;
  int qp_per,qp_rem,q_bits;

  int   pos_x   = block_x >> BLOCK_SHIFT;
  int   pos_y   = block_y >> BLOCK_SHIFT;
  int   b8      = 2*(pos_y >> 1) + (pos_x >> 1);