block.c

JM 11.0 KTA 2.1 Source Code

        int masterQP_C;
        masterQP_C = Clip3 ( -img->bitdepth_chroma_qp_scale, 51, img->masterQP + img->chroma_qp_offset[uv] );
        masterQP_C = (masterQP_C < 0 ? masterQP_C : QP_SCALE_CR[masterQP_C]) + img->bitdepth_chroma_qp_scale;

        if(masterQP_C < img->masterQP)
        {
          lambda_md = img->lambda_md[img->type][masterQP_C];
        }
        else
        {
          lambda_md = img->lambda_md[img->type][img->masterQP];
        }
      }
      else
      {
        lambda_md = img->lambda_md[img->type][img->masterQP];
      }
    }
  }
#endif

#ifdef ADAPTIVE_QUANTIZATION
  if(img->slice_fractional_quant_flag)
  {
    levelscale    = LevelScale4x4Chroma_IAQMS   [img->mb_iaqms_idx][uv][intra];
    invlevelscale = InvLevelScale4x4Chroma_IAQMS[img->mb_iaqms_idx][uv][intra];
  }
  else
  {
    levelscale = LevelScale4x4Chroma[uv][intra];
    invlevelscale = InvLevelScale4x4Chroma[uv][intra];
  }
  leveloffset = LevelOffset4x4Chroma[uv][intra];
#else
  levelscale = LevelScale4x4Chroma[uv][intra];
  leveloffset = LevelOffset4x4Chroma[uv][intra];
  invlevelscale = InvLevelScale4x4Chroma[uv][intra];
#endif

  if (img->yuv_format == YUV422)
  {
    //for YUV422 only
    qp_per_dc = (currMB->qpc[uv] + 3 + img->bitdepth_chroma_qp_scale)/6;
    qp_rem_dc = (currMB->qpc[uv] + 3 + img->bitdepth_chroma_qp_scale)%6;
    
    q_bits_422 = Q_BITS+qp_per_dc;  
  }

  
  //============= dct transform ===============  
  for (n2=0; n2 < img->mb_cr_size_y; n2 += BLOCK_SIZE)
  {
    for (n1=0; n1 < img->mb_cr_size_x; n1 += BLOCK_SIZE)
    {

      //  Horizontal transform.
      for (j=0; j < BLOCK_SIZE && !lossless_qpprime; j++)
      {
        mb_y=n2+j;
        
        m5[0]=img->m7[mb_y][n1  ]+img->m7[mb_y][n1+3];
        m5[1]=img->m7[mb_y][n1+1]+img->m7[mb_y][n1+2];
        m5[2]=img->m7[mb_y][n1+1]-img->m7[mb_y][n1+2];
        m5[3]=img->m7[mb_y][n1  ]-img->m7[mb_y][n1+3];
        
        img->m7[mb_y][n1  ] = (m5[0]   + m5[1]);
        img->m7[mb_y][n1+2] = (m5[0]   - m5[1]);
        img->m7[mb_y][n1+1] =  m5[3]*2 + m5[2];
        img->m7[mb_y][n1+3] =  m5[3]   - m5[2]*2;
      }

      //  Vertical transform.

      for (i=0; i < BLOCK_SIZE && !lossless_qpprime; i++)
      {
        j1=n1+i;
        m5[0] = img->m7[n2  ][j1] + img->m7[n2+3][j1];
        m5[1] = img->m7[n2+1][j1] + img->m7[n2+2][j1];
        m5[2] = img->m7[n2+1][j1] - img->m7[n2+2][j1];
        m5[3] = img->m7[n2  ][j1] - img->m7[n2+3][j1];

        img->m7[n2+0][j1] = (m5[0]   + m5[1]);
        img->m7[n2+2][j1] = (m5[0]   - m5[1]);
        img->m7[n2+1][j1] =  m5[3]*2 + m5[2];
        img->m7[n2+3][j1] =  m5[3]   - m5[2]*2;
      }
    }
  }
  
  if (yuv == YUV420)
  {
    //================== CHROMA DC YUV420 ===================
    //     2X2 transform of DC coeffs.
    if(lossless_qpprime)
    {
      m1[0]=img->m7[0][0];
      m1[1]=img->m7[0][4];
      m1[2]=img->m7[4][0];
      m1[3]=img->m7[4][4];
    }
    else 
    {
      m1[0]=(img->m7[0][0] + img->m7[0][4] + img->m7[4][0] + img->m7[4][4]);
      m1[1]=(img->m7[0][0] - img->m7[0][4] + img->m7[4][0] - img->m7[4][4]);
      m1[2]=(img->m7[0][0] + img->m7[0][4] - img->m7[4][0] - img->m7[4][4]);
      m1[3]=(img->m7[0][0] - img->m7[0][4] - img->m7[4][0] + img->m7[4][4]);
    }
    
    //     Quant of chroma 2X2 coeffs.
    run=-1;
    scan_pos=0;
    
    for (coeff_ctr=0; coeff_ctr < 4; coeff_ctr++)
    {
      run++;
      ilev=0;
      
      if(lossless_qpprime)
        level =absm(m1[coeff_ctr]);
      else 
        level =(absm(m1[coeff_ctr]) * levelscale[qp_rem][0][0] + (leveloffset[qp_per][0][0]<<1)) >> (q_bits+1);
      
      if (input->symbol_mode == UVLC && img->qp < 4) 
      {
        if (level > CAVLC_LEVEL_LIMIT) 
          level = CAVLC_LEVEL_LIMIT;
      }
      
      if (level  != 0)
      {
        currMB->cbp_blk |= 0xf0000 << (uv << 2) ;    // if one of the 2x2-DC levels is != 0 set the
        cr_cbp=max(1,cr_cbp);                     // coded-bit all 4 4x4 blocks (bit 16-19 or 20-23)
        DCcoded = 1 ;
        DCLevel[scan_pos] = sign(level ,m1[coeff_ctr]);
        DCRun  [scan_pos] = run;
        scan_pos++;
        run=-1;
        
        ilev=sign(level, m1[coeff_ctr]);
      }
      if(!lossless_qpprime)
        m1[coeff_ctr]=ilev;
    }
    DCLevel[scan_pos] = 0;
    
    //  Inverse transform of 2x2 DC levels
    if(!lossless_qpprime)
    {
      m5[0]=(m1[0] + m1[1] + m1[2] + m1[3]);
      m5[1]=(m1[0] - m1[1] + m1[2] - m1[3]);
      m5[2]=(m1[0] + m1[1] - m1[2] - m1[3]);
      m5[3]=(m1[0] - m1[1] - m1[2] + m1[3]);
      if(qp_per<5)
      {
        for(i=0; i<4; i++)
          m1[i]=(m5[i] * invlevelscale[qp_rem][0][0])>>(5-qp_per);
      }
      else
      {
        for(i=0; i<4; i++)
          m1[i]=(m5[i] * invlevelscale[qp_rem][0][0])<<(qp_per-5);
      }

      img->m7[0][0] = m1[0];
      img->m7[0][4] = m1[1];
      img->m7[4][0] = m1[2];
      img->m7[4][4] = m1[3];
    }
  }
  else if(yuv == YUV422)
  {
    //================== CHROMA DC YUV422 ===================
    //transform DC coeff
    //horizontal
    
    //pick out DC coeff
    for (j=0; j < img->mb_cr_size_y; j+=BLOCK_SIZE)
    {
      for (i=0; i < img->mb_cr_size_x; i+=BLOCK_SIZE)
        m3[i>>2][j>>2]= img->m7[j][i];
    } 
    //horizontal
    if(!lossless_qpprime)
    {
      m4[0][0] = m3[0][0] + m3[1][0];
      m4[0][1] = m3[0][1] + m3[1][1];
      m4[0][2] = m3[0][2] + m3[1][2];
      m4[0][3] = m3[0][3] + m3[1][3];
      
      m4[1][0] = m3[0][0] - m3[1][0];
      m4[1][1] = m3[0][1] - m3[1][1];
      m4[1][2] = m3[0][2] - m3[1][2];
      m4[1][3] = m3[0][3] - m3[1][3];
      
      // vertical
      for (i=0;i<2;i++)
      {
        m5[0] = m4[i][0] + m4[i][3];
        m5[1] = m4[i][1] + m4[i][2];
        m5[2] = m4[i][1] - m4[i][2];
        m5[3] = m4[i][0] - m4[i][3];
        
        m4[i][0] = (m5[0] + m5[1]);
        m4[i][2] = (m5[0] - m5[1]);
        m4[i][1] = (m5[3] + m5[2]);
        m4[i][3] = (m5[3] - m5[2]);
      }
    }
    
    run=-1;
    scan_pos=0;
    
    //quant of chroma DC-coeffs
    for (coeff_ctr=0;coeff_ctr<8;coeff_ctr++)
    {
      i=SCAN_YUV422[coeff_ctr][0];
      j=SCAN_YUV422[coeff_ctr][1];
      
      run++;

      if(lossless_qpprime)
      {
        level = absm(m3[i][j]);
        m4[i][j]=m3[i][j];
      }
      else 
        level =(absm(m4[i][j]) * levelscale[qp_rem_dc][0][0] + (leveloffset[qp_per_dc][0][0]*2)) >> (q_bits_422+1);

      if (level != 0)
      {
        //YUV422
        currMB->cbp_blk |= 0xff0000 << (uv << 3) ;   // if one of the DC levels is != 0 set the
        cr_cbp=max(1,cr_cbp);                           // coded-bit all 4 4x4 blocks (bit 16-31 or 32-47) //YUV444
        DCcoded = 1 ;
        
        DCLevel[scan_pos] = sign(level,m4[i][j]);
        DCRun  [scan_pos] = run;
        ++scan_pos;
        run=-1;
      }
      if(!lossless_qpprime)
        m3[i][j]=sign(level,m4[i][j]);
    }
    DCLevel[scan_pos]=0;

    //inverse DC transform
    //horizontal
    if(!lossless_qpprime)
    {
      m4[0][0] = m3[0][0] + m3[1][0];
      m4[0][1] = m3[0][1] + m3[1][1];
      m4[0][2] = m3[0][2] + m3[1][2];
      m4[0][3] = m3[0][3] + m3[1][3];
      
      m4[1][0] = m3[0][0] - m3[1][0];
      m4[1][1] = m3[0][1] - m3[1][1];
      m4[1][2] = m3[0][2] - m3[1][2];
      m4[1][3] = m3[0][3] - m3[1][3];      
      
      // vertical
      for (i=0;i<2;i++)
      {       
        m6[0]=m4[i][0]+m4[i][2];
        m6[1]=m4[i][0]-m4[i][2];
        m6[2]=m4[i][1]-m4[i][3];
        m6[3]=m4[i][1]+m4[i][3];
        
        if(qp_per_dc<4)
        {          
          img->m7[0 ][i*4]=((((m6[0]+m6[3])*invlevelscale[qp_rem_dc][0][0]+(1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
          img->m7[4 ][i*4]=((((m6[1]+m6[2])*invlevelscale[qp_rem_dc][0][0]+(1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
          img->m7[8 ][i*4]=((((m6[1]-m6[2])*invlevelscale[qp_rem_dc][0][0]+(1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
          img->m7[12][i*4]=((((m6[0]-m6[3])*invlevelscale[qp_rem_dc][0][0]+(1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
        }
        else
        {
          img->m7[0 ][i*4]=((((m6[0]+m6[3])*invlevelscale[qp_rem_dc][0][0])<<(qp_per_dc-4))+2)>>2;
          img->m7[4 ][i*4]=((((m6[1]+m6[2])*invlevelscale[qp_rem_dc][0][0])<<(qp_per_dc-4))+2)>>2;
          img->m7[8 ][i*4]=((((m6[1]-m6[2])*invlevelscale[qp_rem_dc][0][0])<<(qp_per_dc-4))+2)>>2;
          img->m7[12][i*4]=((((m6[0]-m6[3])*invlevelscale[qp_rem_dc][0][0])<<(qp_per_dc-4))+2)>>2;
        }
      }//for (i=0;i<2;i++)    
    }
  }
  else if(yuv == YUV444)
  {
    //================== CHROMA DC YUV444 ===================
    //transform DC coeff
    //pick out DC coeff
    for (j=0; j < img->mb_cr_size_y; j+=BLOCK_SIZE)
    {
      for (i=0; i < img->mb_cr_size_x; i+=BLOCK_SIZE)
        m4[i>>2][j>>2]= img->m7[j][i];
    }
    
    //horizontal
    for (j=0;j<4 && !lossless_qpprime;j++)
    {
      m5[0] = m4[0][j] + m4[3][j];
      m5[1] = m4[1][j] + m4[2][j];
      m5[2] = m4[1][j] - m4[2][j];
      m5[3] = m4[0][j] - m4[3][j];
      
      m4[0][j]=m5[0]+m5[1];
      m4[2][j]=m5[0]-m5[1];
      m4[1][j]=m5[3]+m5[2];
      m4[3][j]=m5[3]-m5[2];
    }
    // vertical
    for (i=0;i<4 && !lossless_qpprime;i++)
    {
      m5[0] = m4[i][0] + m4[i][3];
      m5[1] = m4[i][1] + m4[i][2];
      m5[2] = m4[i][1] - m4[i][2];
      m5[3] = m4[i][0] - m4[i][3];

      m4[i][0]=(m5[0]+m5[1])>>1;
      m4[i][2]=(m5[0]-m5[1])>>1;
      m4[i][1]=(m5[3]+m5[2])>>1;
      m4[i][3]=(m5[3]-m5[2])>>1;
    }

    run=-1;
    scan_pos=0;
    
    //quant of chroma DC-coeffs
    for (coeff_ctr=0;coeff_ctr<16;coeff_ctr++)
    {
      i=SNGL_SCAN[coeff_ctr][0];
      j=SNGL_SCAN[coeff_ctr][1];
      
      run++;
      
      if(lossless_qpprime)
        level = absm(m4[i][j]);
      else 
        level =(absm(m4[i][j]) * levelscale[qp_rem][0][0] + (leveloffset[qp_per][0][0]*2)) >> (q_bits+1);
      
      if (level != 0)
      {
        //YUV444
        currMB->cbp_blk |= ((int64)0xffff0000) << (uv << 4) ;   // if one of the DC levels is != 0 set the
        cr_cbp=max(1,cr_cbp);                           // coded-bit all 4 4x4 blocks (bit 16-31 or 32-47) //YUV444
        DCcoded = 1 ;
        
        DCLevel[scan_pos] = sign(level,m4[i][j]);
        DCRun  [scan_pos] = run;
        ++scan_pos;
        run=-1;
      }
      if(!lossless_qpprime)
        m4[i][j]=sign(level,m4[i][j]);
    }
    DCLevel[scan_pos]=0;

    // inverse DC transform
    //horizontal
    for (j=0;j<4 && !lossless_qpprime;j++)
    {     
      m6[0] = m4[0][j] + m4[2][j];
      m6[1] = m4[0][j] - m4[2][j];
      m6[2] = m4[1][j] - m4[3][j];
      m6[3] = m4[1][j] + m4[3][j];
      
      m4[0][j] = m6[0] + m6[3];
      m4[1][j] = m6[1] + m6[2];
      m4[2][j] = m6[1] - m6[2];
      m4[3][j] = m6[0] - m6[3];
    }
    
    //vertical
    for (i=0;i<4 && !lossless_qpprime;i++)
    {
      m6[0]=m4[i][0]+m4[i][2];
      m6[1]=m4[i][0]-m4[i][2];
      m6[2]=m4[i][1]-m4[i][3];
      m6[3]=m4[i][1]+m4[i][3];

      if(qp_per<4)
      {
        img->m7[0 ][i*4] = ((((m6[0] + m6[3])*invlevelscale[qp_rem][0][0]+(1<<(3-qp_per)))>>(4-qp_per))+2)>>2;
        img->m7[4 ][i*4] = ((((m6[1] + m6[2])*invlevelscale[qp_rem][0][0]+(1<<(3-qp_per)))>>(4-qp_per))+2)>>2;
        img->m7[8 ][i*4] = ((((m6[1] - m6[2])*invlevelscale[qp_rem][0][0]+(1<<(3-qp_per)))>>(4-qp_per))+2)>>2;
        img->m7[12][i*4] = ((((m6[0] - m6[3])*invlevelscale[qp_rem][0][0]+(1<<(3-qp_per)))>>(4-qp_per))+2)>>2;
      }
      else
      {
        img->m7[0 ][i*4] = ((((m6[0]+m6[3])*invlevelscale[qp_rem][0][0])<<(qp_per-4))+2)>>2;
        img->m7[4 ][i*4] = ((((m6[1]+m6[2])*invlevelscale[qp_rem][0][0])<<(qp_per-4))+2)>>2;
        img->m7[8 ][i*4] = ((((m6[1]-m6[2])*invlevelscale[qp_rem][0][0])<<(qp_per-4))+2)>>2;
        img->m7[12][i*4] = ((((m6[0]-m6[3])*invlevelscale[qp_rem][0][0])<<(qp_per-4))+2)>>2;
      }
    }
  }

  //     Quant of chroma AC-coeffs.
  coeff_cost=0;
  cr_cbp_tmp=0;

  for (b8=0; b8 < (img->num_blk8x8_uv >> 1); b8++)
  {
    for (b4=0; b4 < 4; b4++)
    {
      n1 = hor_offset[yuv][b8][b4];
      n2 = ver_offset[yuv][b8][b4];
      ACLevel = img->cofAC[4+b8+uv_scale][b4][0];
      ACRun   = img->cofAC[4+b8+uv_scale][b4][1];
      run=-1;
      scan_pos=0;



#ifdef RDO_Q
//#ifdef TREL_CAVLC
      if(input->UseRDO_Q && active_pps->entropy_coding_mode_flag == UVLC)
      {
        int M4[4][4], n, m;
        for(n=0; n<4; n++)
          for(m=0; m<4; m++)
            M4[n][m] = img->m7[n2+n][n1+m];

        TrellisCAVLC4x4(M4, q_bits, qp_rem, levelscale[qp_rem], leveloffset[qp_per], levelTrellis, CHROMA_AC, 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