block.c

H.264视频编解码的标准测试模型

      s1 += imgY_pred[up.pos_y][up.pos_x+i];    // sum hor pix
  }

  if (left_avail)
  {
    for (i=0; i < MB_BLOCK_SIZE; i++)      
      s2 += imgY_pred[left[i+1].pos_y][left[i+1].pos_x];    // sum vert pix
  }

  if (up_avail && left_avail)
    s0=(s1+s2+16)/(2*MB_BLOCK_SIZE);             // no edge
  
  if (!up_avail && left_avail)
    s0=(s2+8)/MB_BLOCK_SIZE;                     // upper edge
  
  if (up_avail && !left_avail)
    s0=(s1+8)/MB_BLOCK_SIZE;                     // left edge
  
  if (!up_avail && !left_avail)
    s0=img->dc_pred_value;                       // top left corner, nothing to predict from

  // vertical prediction
  if (up_avail)
    memcpy(s[0], &imgY_pred[up.pos_y][up.pos_x], MB_BLOCK_SIZE * sizeof(imgpel));
  
  // horizontal prediction
  if (left_avail)
  {
    for (i=0; i < MB_BLOCK_SIZE; i++)
      s[1][i]=imgY_pred[left[i+1].pos_y][left[i+1].pos_x];
  }

  for (j=0; j < MB_BLOCK_SIZE; j++)
  {
    memcpy(img->mprr_2[VERT_PRED_16][j], s[0], MB_BLOCK_SIZE * sizeof(imgpel)); // store vertical prediction
    for (i=0; i < MB_BLOCK_SIZE; i++)
    {
      img->mprr_2[HOR_PRED_16 ][j][i]=s[1][j]; // store horizontal prediction
      img->mprr_2[DC_PRED_16  ][j][i]=s0;      // store DC prediction
    }
  }
  if (!up_avail || !left_avail || !left_up_avail) // edge
    return;

  // 16 bit integer plan pred

  ih=0;
  iv=0;
  for (i=1;i<9;i++)
  {
    if (i<8)
      ih += i*(imgY_pred[up.pos_y][up.pos_x+7+i] - imgY_pred[up.pos_y][up.pos_x+7-i]);
    else
      ih += i*(imgY_pred[up.pos_y][up.pos_x+7+i] - imgY_pred[left[0].pos_y][left[0].pos_x]);
    
    iv += i*(imgY_pred[left[8+i].pos_y][left[8+i].pos_x] - imgY_pred[left[8-i].pos_y][left[8-i].pos_x]);
  }
  ib=(5*ih+32)>>6;
  ic=(5*iv+32)>>6;
  
  iaa=16*(imgY_pred[up.pos_y][up.pos_x+15]+imgY_pred[left[16].pos_y][left[16].pos_x]);

  for (j=0;j< MB_BLOCK_SIZE;j++)
  {
    for (i=0;i< MB_BLOCK_SIZE;i++)
    {
      img->mprr_2[PLANE_16][j][i]=max(0,min((int)img->max_imgpel_value,(iaa+(i-7)*ib +(j-7)*ic + 16)/32));// store plane prediction
    }
  }
}


/*!
 ************************************************************************
 * \brief
 *    For new intra pred routines
 *
 * \par Input:
 *    Image par, 16x16 based intra mode
 *
 * \par Output:
 *    none
 ************************************************************************
 */
int dct_luma_16x16(int new_intra_mode)
{
  //int qp_const;
  int i,j;
  int ii,jj;
  int jdiv, jmod;
  int M1[16][16];
  int M4[4][4];
  int M5[4],M6[4];
  int M0[4][4][4][4];
  int run,scan_pos,coeff_ctr,level;
  int qp_per,qp_rem,q_bits;
  int ac_coef = 0;

  Macroblock *currMB = &img->mb_data[img->current_mb_nr];
  short is_field_mode = (img->field_picture || ( img->MbaffFrameFlag && currMB->mb_field));

  int   b8, b4;
  int*  DCLevel = img->cofDC[0][0];
  int*  DCRun   = img->cofDC[0][1];
  int*  ACLevel;
  int*  ACRun;
  int **levelscale,**leveloffset;
  int **invlevelscale;
  Boolean lossless_qpprime = ((currMB->qp + img->bitdepth_luma_qp_scale)==0 && img->lossless_qpprime_flag==1);
  const byte (*pos_scan)[2] = is_field_mode ? FIELD_SCAN : SNGL_SCAN;

  // Note that we could just use currMB->qp here
  qp_per    = qp_per_matrix[(currMB->qp + img->bitdepth_luma_qp_scale - MIN_QP)];
  qp_rem    = qp_rem_matrix[(currMB->qp + img->bitdepth_luma_qp_scale - MIN_QP)];

  q_bits    = Q_BITS+qp_per;
  levelscale    = LevelScale4x4Luma[1][qp_rem];
  leveloffset   = LevelOffset4x4Luma[1][qp_per];
  invlevelscale = InvLevelScale4x4Luma[1][qp_rem];


  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];

      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++;

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

        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 
        }
      }
    }