block.c

H.264视频编解码的标准测试模型

  int i,j,i1,j1,ilev,m5[4],m6[4],coeff_ctr;
  int level,scan_pos,run;
  int nonzeroAC;
  Macroblock *currMB = &img->mb_data[img->current_mb_nr];
  int   intra = IS_INTRA (currMB);

  int qp_per,qp_rem,q_bits;
  int qp_c;

  int*  ACLevel = img->cofAC[b8][b4][0];
  int*  ACRun   = img->cofAC[b8][b4][1];

  int **levelscale, **leveloffset;
  int **invlevelscale;

  Boolean lossless_qpprime = ((img->qp + img->bitdepth_luma_qp_scale)==0 && img->lossless_qpprime_flag==1);

  qp_c      = currMB->qp + img->chroma_qp_offset[uv];
  qp_c      = (qp_c < 0)? qp_c : QP_SCALE_CR[qp_c - MIN_QP];

  qp_per    = qp_per_matrix[(qp_c + img->bitdepth_chroma_qp_scale)];
  qp_rem    = qp_rem_matrix[(qp_c + img->bitdepth_chroma_qp_scale)];
  q_bits    = Q_BITS+qp_per;

  levelscale = LevelScale4x4Chroma[uv][intra][qp_rem];
  leveloffset = LevelOffset4x4Chroma[uv][intra][qp_per];
  invlevelscale = InvLevelScale4x4Chroma[uv][intra][qp_rem];

  //  Horizontal transform
  if(!lossless_qpprime)
  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < 2; i++)
    {
      i1=3-i;
      m5[i]=img->m7[j][i]+img->m7[j][i1];
      m5[i1]=img->m7[j][i]-img->m7[j][i1];
    }
    img->m7[j][0]=(m5[0]+m5[1]);
    img->m7[j][2]=(m5[0]-m5[1]);
    img->m7[j][1]=m5[3]*2+m5[2];
    img->m7[j][3]=m5[3]-m5[2]*2;
  }

  //  Vertical transform
  if(!lossless_qpprime)
  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < 2; j++)
    {
      j1=3-j;
      m5[j]=img->m7[j][i]+img->m7[j1][i];
      m5[j1]=img->m7[j][i]-img->m7[j1][i];
    }
    img->m7[0][i]=(m5[0]+m5[1]);
    img->m7[2][i]=(m5[0]-m5[1]);
    img->m7[1][i]=m5[3]*2+m5[2];
    img->m7[3][i]=m5[3]-m5[2]*2;
  }

  // Quant

  nonzeroAC=FALSE;

  run=-1;
  scan_pos=0;

  if(lossless_qpprime)
    level = absm(img->m7[0][0]);
  else 
    level =(absm(img->m7[0][0]) * levelscale[0][0] + leveloffset[0][0]) >> q_bits;

  b8 -= 4*(uv+1);
  dc_level_temp[uv][2*(b8 & 0x01)+(b4 & 0x01)][2*(b8 >> 1)+(b4 >> 1)] = sign(level, img->m7[0][0]);

  /* Inverse Quantization */
  if(lossless_qpprime)
  {
    img->m7[0][0] = sign( level, img->m7[0][0]);
  }
  else
  {
    if(qp_per<4)
    {
      img->m7[0][0] = sign( ((level*invlevelscale[0][0]+(1<<(3-qp_per)))>>(4-qp_per)), img->m7[0][0]);
    }
    else
    {
      img->m7[0][0] = sign( ((level*invlevelscale[0][0])<<(qp_per-4)), img->m7[0][0]);
    }
  }

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

    run++;
    ilev=0;

    if(lossless_qpprime)
      level = absm (img->m7[j][i]);
    else 
      level = (absm(img->m7[j][i])*levelscale[i][j]+leveloffset[i][j])>>q_bits;
    
    if (level != 0)
    {
      if(i||j) nonzeroAC=TRUE;
      
      ACLevel[scan_pos] = sign(level,img->m7[j][i]);
      ACRun  [scan_pos] = run;
      ++scan_pos;
      run=-1;                     // reset zero level counter
      
      level=sign(level, img->m7[j][i]);
      if(lossless_qpprime)
      {
        ilev=level;
      }
      else if(qp_per<4)
      {
        ilev=(level*invlevelscale[i][j]+(1<<(3-qp_per)))>>(4-qp_per);
      }
      else
      {
        ilev=(level*invlevelscale[i][j])<<(qp_per-4);
      }
    }
    if(!lossless_qpprime)
      img->m7[j][i]=ilev;
  }
  ACLevel[scan_pos] = 0;

  
  //     IDCT.
  //     horizontal
  if(!lossless_qpprime)
  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < BLOCK_SIZE; i++)
    {
      m5[i]=img->m7[j][i];
    }
    m6[0]=(m5[0]+m5[2]);
    m6[1]=(m5[0]-m5[2]);
    m6[2]=(m5[1]>>1)-m5[3];
    m6[3]=m5[1]+(m5[3]>>1);

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

  //  vertical
  if(!lossless_qpprime)
  for (i=0; i < BLOCK_SIZE; i++)
  {
    for (j=0; j < BLOCK_SIZE; j++)
    {
      m5[j]=img->m7[j][i];
    }
    m6[0]=(m5[0]+m5[2]);
    m6[1]=(m5[0]-m5[2]);
    m6[2]=(m5[1]>>1)-m5[3];
    m6[3]=m5[1]+(m5[3]>>1);

    for (j=0; j < 2; j++)
    {
      j1=3-j;
      img->m7[j][i] =(m6[j]+m6[j1]+DQ_ROUND)>>DQ_BITS;
      img->m7[j1][i]=(m6[j]-m6[j1]+DQ_ROUND)>>DQ_BITS;
    }
  }

  return nonzeroAC;
}

// Residue Color Transform
int dct_chroma_DC(int uv, int cr_cbp)
{
  int run, scan_pos, coeff_ctr, level, i, j;
  int*  DCLevel = img->cofDC[uv+1][0];
  int*  DCRun   = img->cofDC[uv+1][1];

  run=-1;
  scan_pos=0;

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

    run++;

    level = absm(dc_level[uv][i][j]);

    if (level  != 0)
    {
      cr_cbp=max(1,cr_cbp);
      DCLevel[scan_pos] = sign(level ,dc_level[uv][i][j]);
      DCRun  [scan_pos] = run;
      scan_pos++;
      run=-1;
    }
  }
  DCLevel[scan_pos] = 0;

  return cr_cbp;
}


/*!
 ************************************************************************
 * \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.
 *
 *
 ************************************************************************
 */
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;
  int qp_const,level,scan_pos,run;
  int nonzero;

  int predicted_block[BLOCK_SIZE][BLOCK_SIZE],c_err,qp_const2;
  int qp_per,qp_rem,q_bits;
  int qp_per_sp,qp_rem_sp,q_bits_sp;

  int   pos_x   = block_x >> BLOCK_SHIFT;
  int   pos_y   = block_y >> BLOCK_SHIFT;
  int   b8      = 2*(pos_y >> 1) + (pos_x >> 1);
  int   b4      = 2*(pos_y & 0x01) + (pos_x & 0x01);
  int*  ACLevel = img->cofAC[b8][b4][0];
  int*  ACRun   = img->cofAC[b8][b4][1];
  Macroblock *currMB = &img->mb_data[img->current_mb_nr];
  short is_field_mode = (img->field_picture || ( img->MbaffFrameFlag && currMB->mb_field));

  // For encoding optimization
  int c_err1, c_err2, level1, level2;
  double D_dis1, D_dis2;
  int len, info;
  double lambda_mode   = 0.85 * pow (2, (currMB->qp - SHIFT_QP)/3.0) * 4; 

  qp_per    = (currMB->qp-MIN_QP)/6;
  qp_rem    = (currMB->qp-MIN_QP)%6;
  q_bits    = Q_BITS+qp_per;
  qp_per_sp    = (currMB->qpsp-MIN_QP)/6;
  qp_rem_sp    = (currMB->qpsp-MIN_QP)%6;
  q_bits_sp    = Q_BITS+qp_per_sp;

  qp_const=(1<<q_bits)/6;    // inter
  qp_const2=(1<<q_bits_sp)/2;  //sp_pred

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

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

  //  Vertical transform

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

  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]);
    predicted_block[2][j]=(m5[0]-m5[1]);
    predicted_block[1][j]=m5[3]*2+m5[2];
    predicted_block[3][j]=m5[3]-m5[2]*2;
  }

  //  Vertical 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]);
    predicted_block[i][2]=(m5[0]-m5[1]);
    predicted_block[i][1]=m5[3]*2+m5[2];
    predicted_block[i][3]=m5[3]-m5[2]*2;
  }

  // Quant
  nonzero=FALSE;

  run=-1;
  scan_pos=0;
  
  for (coeff_ctr=0;coeff_ctr < 16;coeff_ctr++)     // 8 times if double scan, 16 normal scan
  {

    if (is_field_mode) 
    {  // Alternate scan for field coding
        i=FIELD_SCAN[coeff_ctr][0];
        j=FIELD_SCAN[coeff_ctr][1];
    }
    else 
    {
        i=SNGL_SCAN[coeff_ctr][0];
        j=SNGL_SCAN[coeff_ctr][1];
    }
    
    run++;
    ilev=0;
    
    // decide prediction
    
    // case 1
    level1 = (absm (predicted_block[i][j]) * quant_coef[qp_rem_sp][i][j] + qp_const2) >> q_bits_sp; 
    level1 = (level1 << q_bits_sp) / quant_coef[qp_rem_sp][i][j];                 
    c_err1 = img->m7[j][i]-sign(level1, predicted_block[i][j]);                   
    level1 = (absm (c_err1) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;
    
    // case 2
    c_err2=img->m7[j][i]-predicted_block[i][j];
    level2 = (absm (c_err2) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;
    
    // select prediction
    if ((level1 != level2) && (level1 != 0) && (level2 != 0))
    {
      D_dis1 = img->m7[j][i] - ((sign(level1,c_err1)*dequant_coef[qp_rem][i][j]*A[i][j]<< qp_per) >>6) - predicted_block[i][j]; 
      levrun_linfo_inter(level1, run, &len, &info);
      D_dis1 = D_dis1*D_dis1 + lambda_mode * len;
      
      D_dis2 = img->m7[j][i] - ((sign(level2,c_err2)*dequant_coef[qp_rem][i][j]*A[i][j]<< qp_per) >>6) - predicted_block[i][j]; 
      levrun_linfo_inter(level2, run, &len, &info);
      D_dis2 = D_dis2 * D_dis2 + lambda_mode * len;
      
      if (D_dis1 == D_dis2)
        level = (absm(level1) < absm(level2)) ? level1 : level2;
      else
      {
        if (D_dis1 < D_dis2)
          level = level1;
        else
          level = level2;
      }
      c_err = (level == level1) ? c_err1 : c_err2;
    }
    else if (level1 == level2)
    {
      level = level1;
      c_err = c_err1;
    }
    else
    {
      level = (level1 == 0) ? level1 : level2;
      c_err = (level1 == 0) ? c_err1 : c_err2;
    }
    
    if (level != 0)
    {
      nonzero=TRUE;
      if (level > 1)
        *coeff_cost += MAX_VALUE;                // set high cost, shall not be discarded
      else
        *coeff_cost += COEFF_COST[input->disthres][run];
      ACLevel[scan_pos] = sign(level,c_err);
      ACRun  [scan_pos] = run;
      ++scan_pos;
      run=-1;                     // reset zero level counter
      ilev=((sign(level,c_err)*dequant_coef[qp_rem][i][j]*A[i][j]<< qp_per) >>6);
    }
    ilev+=predicted_block[i][j] ; 
    if(!si_frame_indicator && !sp2_frame_indicator)//stores the SP frame coefficients in lrec, will be useful to encode these and create SI or SP switching frame
    {
      lrec[img->pix_y+block_y+j][img->pix_x+block_x+i]= 
        sign((abs(ilev) * quant_coef[qp_rem_sp][i][j] + qp_const2) >> q_bits_sp, ilev); 
    }
    img->m7[j][i] = sign((absm(ilev) * quant_coef[qp_rem_sp][i][j] + qp_const2)>> q_bits_sp, ilev) * dequant_coef[qp_rem_sp][i][j] << qp_per_sp;
  }
  ACLevel[scan_pos] = 0;
  
    
  //     IDCT.
  //     horizontal

  for (j=0; j < BLOCK_SIZE; j++)
  {
    for (i=0; i < BLOCK_SIZE; i++)
    {
      m5[i]=i