block.c

网络MPEG4IP流媒体开发源代码

    break;
  case QUANT_CHROMA_DC:
    quant_set=QP_SCALE_CR[img->qp];
    idx=0;
    no_coeff=4;
    break;
  case QUANT_CHROMA_AC:
    quant_set=QP_SCALE_CR[img->qp];
    idx=2;
    no_coeff=15;
    break;
  default:
    error("rd_quant: unsupported scan_type", 600);
    break;
  }

  dbl_coeff_ctr=0;
  for (coeff_ctr=0;coeff_ctr < no_coeff ;coeff_ctr++)
  {
    k0=coeff[coeff_ctr];
    k1=abs(k0);

    if (dbl_coeff_ctr < MAX_TWO_LEVEL_COEFF)  // limit the number of 'twin' levels
    {
      level0 = (k0*JQ[quant_set][0])/J20;
      level1 = (k1*JQ[quant_set][0]+JQ4)/J20; // make positive summation
      level1 = sign(level1,k0);// set back sign on level
    }
    else
    {
      level0 = (k1*JQ[quant_set][0]+qp_const)/J20;
      level0 = sign(level0,k0);
      level1 = level0;
    }

    if (level0 != level1)
    {
      dbl_coeff = TRUE;     // decision is still open
      dbl_coeff_ctr++;      // count number of coefficients with 2 possible levels
    }
    else
      dbl_coeff = FALSE;    // level is decided


    snr0 = (12+intra_add)*level0*(64*level0 - (JQ[quant_set][0]*coeff[coeff_ctr])/J13); // find SNR improvement

    level_arr[coeff_ctr][MTLC_POW]=0; // indicates that all coefficients are decided

    for (k=0; k< MTLC_POW; k++)
    {
      level_arr[coeff_ctr][k]=level0;
      snr_arr[coeff_ctr][k]=snr0;
    }
    if (dbl_coeff)
    {
      snr1 = (12+intra_add)*level1*(64*level1 - (JQ[quant_set][0]*coeff[coeff_ctr])/J13);
      ilev= (int)pow(2,dbl_coeff_ctr-1);
      for (k1=ilev; k1<MTLC_POW; k1+=ilev*2)
      {
        for (k2=k1; k2<k1+ilev; k2++)
        {
          level_arr[coeff_ctr][k2]=level1;
          snr_arr[coeff_ctr][k2]=snr1;
        }
      }
    }
  }

  rd_best=0;
  best_coeff_comb= MTLC_POW;      // initial setting, used if no double decision coefficients
  for (k=0; k < pow(2,dbl_coeff_ctr);k++) // go through all combinations
  {
    rd_curr=0;
    run=-1;
    for (coeff_ctr=0;coeff_ctr < no_coeff;coeff_ctr++)
    {
      run++;
      level=min(16,absm(level_arr[coeff_ctr][k]));
      if (level != 0)
      {
        rd_curr += 64*COEFF_BIT_COST[idx][run][level-1]+snr_arr[coeff_ctr][k];
        run = -1;
      }
    }
    if (rd_curr < rd_best)
    {
      rd_best=rd_curr;
      best_coeff_comb=k;
    }
  }
  for (coeff_ctr=0;coeff_ctr < no_coeff ;coeff_ctr++)
    coeff[coeff_ctr]=level_arr[coeff_ctr][best_coeff_comb];

  return;
}

/*!
 ************************************************************************
 * \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.
 *
 * \para Input:
 *    block_x,block_y: Block position inside a macro block (0,4,8,12).
 *
 * \para Output:
 *    nonzero: 0 if no levels are nonzero.  1 if there are nonzero levels.              \n
 *    coeff_cost: Counter for nonzero coefficients, used to discard expencive levels.
 *
 *
************************************************************************/
#ifndef NO_RDQUANT
int dct_luma_sp(int block_x,int block_y,int *coeff_cost)
{
  int sign(int a,int b);

  int i,j,i1,j1,ilev,m5[4],m6[4],coeff_ctr,scan_loop_ctr;
  int pos_x,pos_y,quant_set,level,scan_pos,run;
  int nonzero;
  int idx;

  int scan_mode;
  int loop_rep;
  int predicted_block[BLOCK_SIZE][BLOCK_SIZE],alpha,quant_set1,Fq1q2;
  int coeff[16],coeff2[16];

  pos_x=block_x/BLOCK_SIZE;
  pos_y=block_y/BLOCK_SIZE;

  //  Horizontal transform
  for (j=0; j< BLOCK_SIZE; j++)
    for (i=0; i< BLOCK_SIZE; i++)
    {
      img->m7[i][j]+=img->mpr[i+block_x][j+block_y];
      predicted_block[i][j]=img->mpr[i+block_x][j+block_y];
    }

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < 2; i++)
    {
      i1=3-i;
      m5[i]=img->m7[i][j]+img->m7[i1][j];
      m5[i1]=img->m7[i][j]-img->m7[i1][j];
    }
    img->m7[0][j]=(m5[0]+m5[1])*13;
    img->m7[2][j]=(m5[0]-m5[1])*13;
    img->m7[1][j]=m5[3]*17+m5[2]*7;
    img->m7[3][j]=m5[3]*7-m5[2]*17;
  }

  //  Vertival transform

  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < 2; j++)
    {
      j1=3-j;
      m5[j]=img->m7[i][j]+img->m7[i][j1];
      m5[j1]=img->m7[i][j]-img->m7[i][j1];
    }
    img->m7[i][0]=(m5[0]+m5[1])*13;
    img->m7[i][2]=(m5[0]-m5[1])*13;
    img->m7[i][1]=m5[3]*17+m5[2]*7;
    img->m7[i][3]=m5[3]*7-m5[2]*17;
  }

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < 2; i++)
    {
      i1=3-i;
      m5[i]=predicted_block[i][j]+predicted_block[i1][j];
      m5[i1]=predicted_block[i][j]-predicted_block[i1][j];
    }
    predicted_block[0][j]=(m5[0]+m5[1])*13;
    predicted_block[2][j]=(m5[0]-m5[1])*13;
    predicted_block[1][j]=m5[3]*17+m5[2]*7;
    predicted_block[3][j]=m5[3]*7-m5[2]*17;
  }

  //  Vertival transform

  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < 2; j++)
    {
      j1=3-j;
      m5[j]=predicted_block[i][j]+predicted_block[i][j1];
      m5[j1]=predicted_block[i][j]-predicted_block[i][j1];
    }
    predicted_block[i][0]=(m5[0]+m5[1])*13;
    predicted_block[i][2]=(m5[0]-m5[1])*13;
    predicted_block[i][1]=m5[3]*17+m5[2]*7;
    predicted_block[i][3]=m5[3]*7-m5[2]*17;
  }

  // Quant

  quant_set=img->qp;
  quant_set1=img->qpsp;
  alpha=(JQQ1+JQ[quant_set1][0]/2)/JQ[quant_set1][0];
  Fq1q2=(JQQ1*JQ[quant_set1][0]+JQ[quant_set][0]/2)/JQ[quant_set][0];
  nonzero=FALSE;

    scan_mode=SINGLE_SCAN;
    loop_rep=1;
    idx=0;

  for(scan_loop_ctr=0;scan_loop_ctr<loop_rep;scan_loop_ctr++) // 2 times if double scan, 1 normal scan
  {
    for (coeff_ctr=0;coeff_ctr < 16/loop_rep;coeff_ctr++)     // 8 times if double scan, 16 normal scan
    {
      if (scan_mode==DOUBLE_SCAN)
      {
        i=DBL_SCAN[coeff_ctr][0][scan_loop_ctr];
        j=DBL_SCAN[coeff_ctr][1][scan_loop_ctr];
      }
      else
      {
        i=SNGL_SCAN[coeff_ctr][0];
        j=SNGL_SCAN[coeff_ctr][1];
      }
      coeff[coeff_ctr]=img->m7[i][j];
      coeff2[coeff_ctr]=(img->m7[i][j]-sign(((abs (predicted_block[i][j]) * JQ[quant_set1][0] +JQQ2) / JQQ1),predicted_block[i][j])*alpha);
    }
    rd_quant(QUANT_LUMA_SNG,coeff2);

    run=-1;
    scan_pos=scan_loop_ctr*9;   // for double scan; set first or second scan posision
    for (coeff_ctr=0; coeff_ctr<16/loop_rep; coeff_ctr++)
    {
      if (scan_mode==DOUBLE_SCAN)
      {
        i=DBL_SCAN[coeff_ctr][0][scan_loop_ctr];
        j=DBL_SCAN[coeff_ctr][1][scan_loop_ctr];
      }
      else
      {
        i=SNGL_SCAN[coeff_ctr][0];
        j=SNGL_SCAN[coeff_ctr][1];
      }
      run++;
      ilev=0;

      level= absm(coeff2[coeff_ctr]);
      if (level != 0)
      {
        nonzero=TRUE;
        if (level > 1)
          *coeff_cost += MAX_VALUE;                // set high cost, shall not be discarded
        else
          *coeff_cost += COEFF_COST[run];
        img->cof[pos_x][pos_y][scan_pos][0][scan_mode]=sign(level,coeff2[coeff_ctr]);
        img->cof[pos_x][pos_y][scan_pos][1][scan_mode]=run;
        ++scan_pos;
        run=-1;                     // reset zero level counter
        ilev=level;
      }
      ilev=coeff2[coeff_ctr]*Fq1q2+predicted_block[i][j]*JQ[quant_set1][0];
      img->m7[i][j]=sign((abs(ilev)+JQQ2)/ JQQ1,ilev)*JQ[quant_set1][1];
    }
    img->cof[pos_x][pos_y][scan_pos][0][scan_mode]=0;  // end of block
  }

  //     IDCT.
  //     horizontal

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < BLOCK_SIZE; i++)
    {
      m5[i]=img->m7[i][j];
    }
    m6[0]=(m5[0]+m5[2])*13;
    m6[1]=(m5[0]-m5[2])*13;
    m6[2]=m5[1]*7-m5[3]*17;
    m6[3]=m5[1]*17+m5[3]*7;

    for (i=0; i < 2; i++)
    {
      i1=3-i;
      img->m7[i][j]=m6[i]+m6[i1];
      img->m7[i1][j]=m6[i]-m6[i1];
    }
  }

  //  vertical

  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < BLOCK_SIZE; j++)
    {
      m5[j]=img->m7[i][j];
    }
    m6[0]=(m5[0]+m5[2])*13;
    m6[1]=(m5[0]-m5[2])*13;
    m6[2]=m5[1]*7-m5[3]*17;
    m6[3]=m5[1]*17+m5[3]*7;

    for (j=0; j < 2; j++)
    {
      j1=3-j;
      img->m7[i][j] =min(255,max(0,(m6[j]+m6[j1]+JQQ2)/JQQ1));
      img->m7[i][j1]=min(255,max(0,(m6[j]-m6[j1]+JQQ2)/JQQ1));
    }
  }

  //  Decoded block moved to frame memory

  for (j=0; j < BLOCK_SIZE; j++)
    for (i=0; i < BLOCK_SIZE; i++)
      imgY[img->pix_y+block_y+j][img->pix_x+block_x+i]=img->m7[i][j];


  return nonzero;
}
#endif
#ifdef NO_RDQUANT
int dct_luma_sp(int block_x,int block_y,int *coeff_cost)
{
  int sign(int a,int b);

  int i,j,i1,j1,ilev,m5[4],m6[4],coeff_ctr,scan_loop_ctr;
  int qp_const,pos_x,pos_y,quant_set,level,scan_pos,run;
  int nonzero;
  int idx;

  int scan_mode;
  int loop_rep;
  int predicted_block[BLOCK_SIZE][BLOCK_SIZE],alpha,quant_set1,Fq1q2,c_err;

  qp_const=JQQ4;    // inter

  pos_x=block_x/BLOCK_SIZE;
  pos_y=block_y/BLOCK_SIZE;

  //  Horizontal transform
  for (j=0; j< BLOCK_SIZE; j++)
    for (i=0; i< BLOCK_SIZE; i++)
    {
      img->m7[i][j]+=img->mpr[i+block_x][j+block_y];
      predicted_block[i][j]=img->mpr[i+block_x][j+block_y];
    }

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < 2; i++)
    {
      i1=3-i;
      m5[i]=img->m7[i][j]+img->m7[i1][j];
      m5[i1]=img->m7[i][j]-img->m7[i1][j];
    }
    img->m7[0][j]=(m5[0]+m5[1])*13;
    img->m7[2][j]=(m5[0]-m5[1])*13;
    img->m7[1][j]=m5[3]*17+m5[2]*7;
    img->m7[3][j]=m5[3]*7-m5[2]*17;
  }

  //  Vertival transform

  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < 2; j++)
    {
      j1=3-j;
      m5[j]=img->m7[i][j]+img->m7[i][j1];
      m5[j1]=img->m7[i][j]-img->m7[i][j1];
    }
    img->m7[i][0]=(m5[0]+m5[1])*13;
    img->m7[i][2]=(m5[0]-m5[1])*13;
    img->m7[i][1]=m5[3]*17+m5[2]*7;
    img->m7[i][3]=m5[3]*7-m5[2]*17;
  }

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < 2; i++)
    {
      i1=3-i;
      m5[i]=predicted_block[i][j]+predicted_block[i1][j];
      m5[i1]=predicted_block[i][j]-predicted_block[i1][j];
    }
    predicted_block[0][j]=(m5[0]+m5[1])*13;
    predicted_block[2][j]=(m5[0]-m5[1])*13;
    predicted_block[1][j]=m5[3]*17+m5[2]*7;
    predicted_block[3][j]=m5[3]*7-m5[2]*17;
  }

  //  Vertival transform

  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < 2; j++)
    {
      j1=3-j;
      m5[j]=predicted_block[i][j]+predicted_block[i][j1];
      m5[j1]=predicted_block[i][j]-predicted_block[i][j1];
    }
    predicted_block[i][0]=(m5[0]+m5[1])*13;
    predicted_block[i][2]=(m5[0]-m5[1])*13;
    predicted_block[i][1]=m5[3]*17+m5[2]*7;
    predicted_block[i][3]=m5[3]*7-m5[2]*17;
  }

  // Quant

  quant_set=img->qp;
  quant_set1=img->qpsp;
  alpha=(JQQ1+JQ[quant_set1][0]/2)/JQ[quant_set1][0];
  Fq1q2=(JQQ1*JQ[quant_set1][0]+JQ[quant_set][0]/2)/JQ[quant_set][0];
  nonzero=FALSE;

    scan_mode=SINGLE_SCAN;
    loop_rep=1;
    idx=0;

  for(scan_loop_ctr=0;scan_loop_ctr<loop_rep;scan_loop_ctr++) // 2 times if double scan, 1 normal scan
  {
  run=-1;
  scan_pos=scan_loop_ctr*9;

    for (coeff_ctr=0;coeff_ctr < 16/loop_rep;coeff_ctr++)     // 8 times if double scan, 16 normal scan
    {
      if (scan_mode==DOUBLE_SCAN)
      {
        i=DBL_SCAN[coeff_ctr][0][scan_loop_ctr];
        j=DBL_SCAN[coeff_ctr][1][scan_loop_ctr];
      }
      else
      {
        i=SNGL_SCAN[coeff_ctr][0];
        j=SNGL_SCAN[coeff_ctr][1];
      }

      run++;
      ilev=0;

      c_err=img->m7[i][j]-alpha*sign(((abs (predicted_block[i][j]) * JQ[quant_set1][0] +JQQ2) / JQQ1),predicted_block[i][j]);
      level = (abs (c_err) * JQ[quant_set][0] +qp_const) / JQQ1;
      if (level != 0)
      {
        nonzero=TRUE;
        if (level > 1)
          *coeff_cost += MAX_VALUE;                // set high cost, shall not be discarded
        else
          *coeff_cost += COEFF_COST[run];
        img->cof[pos_x][pos_y][scan_pos][0][scan_mode]=sign(level,c_err);
        img->cof[pos_x][pos_y][scan_pos][1][scan_mode]=run;
        ++scan_pos;
        run=-1;                     // reset zero level counter
        ilev=level;
      }
      ilev=sign(ilev,c_err)*Fq1q2+predicted_block[i][j]*JQ[quant_set1][0];
      img->m7[i][j]=sign((abs(ilev)+JQQ2)/ JQQ1*JQ[quant_set1][1],ilev);
    }
    img->cof[pos_x][pos_y][scan_pos][0][scan_mode]=0;  // end of block
  }


  //     IDCT.
  //     horizontal

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < BLOCK_SIZE; i++)
    {
      m5[i]=img->m7[i][j];
    }
    m6[0]=(m5[0]+m5[2])*13;
    m6[1]=(m5[0]-m5[2])*13;
    m6[2]=m5[1]*7-m5[3]*17;
    m6[3]=m5[1]*17+m5[3]*7;

    for (i=0; i < 2; i++)
    {
      i1=3-i;
      img->m7[i][j]=m6[i]+m6[i1];
      img->m7[i1][j]=m6[i]-m6[i1];
    }