block.c

h.264/avc 视频编码程序,实现分数像素匹配功能,非原创.

    img->mprr[DIAG_PRED_RL][0][0] = (A+B)/2; // a
    img->mprr[DIAG_PRED_RL][1][0] = B;       // e
    img->mprr[DIAG_PRED_RL][0][1] = img->mprr[DIAG_PRED_RL][2][0] = (B+C)/2; // b i
    img->mprr[DIAG_PRED_RL][1][1] = img->mprr[DIAG_PRED_RL][3][0] = C;       // f m
    img->mprr[DIAG_PRED_RL][0][2] = img->mprr[DIAG_PRED_RL][2][1] = (C+D)/2; // c j
    img->mprr[DIAG_PRED_RL][3][1] =
        img->mprr[DIAG_PRED_RL][1][2] =
            img->mprr[DIAG_PRED_RL][2][2] =
                img->mprr[DIAG_PRED_RL][3][2] =
                    img->mprr[DIAG_PRED_RL][0][3] =
                        img->mprr[DIAG_PRED_RL][1][3] =
                            img->mprr[DIAG_PRED_RL][2][3] =
                                img->mprr[DIAG_PRED_RL][3][3] = D; // d g h k l n o p
  }

  if (block_available_left)
  { // Do prediction 5
    int E = imgY[img_y+0][img_x-1];
    int F = imgY[img_y+1][img_x-1];
    int G = imgY[img_y+2][img_x-1];
    int H = imgY[img_y+3][img_x-1];

    img->mprr[DIAG_PRED_LR][0][0] = (E+F)/2; // a
    img->mprr[DIAG_PRED_LR][0][1] = F;       // b
    img->mprr[DIAG_PRED_LR][1][0] = img->mprr[DIAG_PRED_LR][0][2] = (F+G)/2; // e c
    img->mprr[DIAG_PRED_LR][1][1] = img->mprr[DIAG_PRED_LR][0][3] = G;       // f d
    img->mprr[DIAG_PRED_LR][2][0] = img->mprr[DIAG_PRED_LR][1][2] = (G+H)/2; // i g

    img->mprr[DIAG_PRED_LR][1][3] =
        img->mprr[DIAG_PRED_LR][2][1] =
            img->mprr[DIAG_PRED_LR][2][2] =
                img->mprr[DIAG_PRED_LR][2][3] =
                    img->mprr[DIAG_PRED_LR][3][0] =
                        img->mprr[DIAG_PRED_LR][3][1] =
                            img->mprr[DIAG_PRED_LR][3][2] =
                                img->mprr[DIAG_PRED_LR][3][3] = H;
  }
}

#endif

/*!
 ************************************************************************
 * \brief
 *    16x16 based luma prediction
 *
 * \para Input:
 *    Image parameters
 *
 * \para Output:
 *    none
 ************************************************************************
 */
void intrapred_luma_2()
{
  int s0=0,s1,s2;
  int i,j;
  int s[16][2];

  int ih,iv;
  int ib,ic,iaa;

  int mb_nr = img->current_mb_nr;
  int mb_width = img->width/16;
  int mb_available_up = (img->mb_y == 0) ? 0 : (img->mb_data[mb_nr].slice_nr == img->mb_data[mb_nr-mb_width].slice_nr);
  int mb_available_left = (img->mb_x == 0) ? 0 : (img->mb_data[mb_nr].slice_nr == img->mb_data[mb_nr-1].slice_nr);

  if(input->UseConstrainedIntraPred)
  {
    if (mb_available_up   && (img->intra_block[mb_nr-mb_width][2]==0 || img->intra_block[mb_nr-mb_width][3]==0))
      mb_available_up   = 0;
    if (mb_available_left && (img->intra_block[mb_nr-       1][1]==0 || img->intra_block[mb_nr       -1][3]==0))
      mb_available_left = 0;
  }

  s1=s2=0;
  // make DC prediction
  for (i=0; i < MB_BLOCK_SIZE; i++)
  {
    if (mb_available_up)
      s1 += imgY[img->pix_y-1][img->pix_x+i];    // sum hor pix
    if (mb_available_left)
      s2 += imgY[img->pix_y+i][img->pix_x-1];    // sum vert pix
  }
  if (mb_available_up && mb_available_left)
    s0=(s1+s2+16)/(2*MB_BLOCK_SIZE);             // no edge
  if (!mb_available_up && mb_available_left)
    s0=(s2+8)/MB_BLOCK_SIZE;                     // upper edge
  if (mb_available_up && !mb_available_left)
    s0=(s1+8)/MB_BLOCK_SIZE;                     // left edge
  if (!mb_available_up && !mb_available_left)
    s0=128;                                      // top left corner, nothing to predict from

  for (i=0; i < MB_BLOCK_SIZE; i++)
  {
    // vertical prediction
    if (mb_available_up)
      s[i][0]=imgY[img->pix_y-1][img->pix_x+i];
    // horizontal prediction
    if (mb_available_left)
      s[i][1]=imgY[img->pix_y+i][img->pix_x-1];
  }

  for (j=0; j < MB_BLOCK_SIZE; j++)
  {
    for (i=0; i < MB_BLOCK_SIZE; i++)
    {
      img->mprr_2[VERT_PRED_16][j][i]=s[i][0]; // store vertical prediction
      img->mprr_2[HOR_PRED_16 ][j][i]=s[j][1]; // store horizontal prediction
      img->mprr_2[DC_PRED_16  ][j][i]=s0;      // store DC prediction
    }
  }
  if (!mb_available_up || !mb_available_left) // edge
    return;

  // 16 bit integer plan pred

  ih=0;
  iv=0;
  for (i=1;i<9;i++)
  {
    ih += i*(imgY[img->pix_y-1][img->pix_x+7+i] - imgY[img->pix_y-1][img->pix_x+7-i]);
    iv += i*(imgY[img->pix_y+7+i][img->pix_x-1] - imgY[img->pix_y+7-i][img->pix_x-1]);
  }
  ib=5*(ih/4)/16;
  ic=5*(iv/4)/16;

  iaa=16*(imgY[img->pix_y-1][img->pix_x+15]+imgY[img->pix_y+15][img->pix_x-1]);
  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(255,(iaa+(i-7)*ib +(j-7)*ic + 16)/32));// store plane prediction
    }
  }
}

/*!
 ************************************************************************
 * \brief
 *    For new intra pred routines
 *
 * \para Input:
 *    Image par, 16x16 based intra mode
 *
 * \para Output:
 *    none
 ************************************************************************
 */
int dct_luma2(int new_intra_mode)
{
  int qp_const;
  int i,j;
  int ii,jj;
  int i1,j1;
  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;

  int   b8, b4;
  int*  DCLevel = img->cofDC[0][0];
  int*  DCRun   = img->cofDC[0][1];
  int*  ACLevel;
  int*  ACRun;

  qp_per    = (img->qp-MIN_QP)/6;
  qp_rem    = (img->qp-MIN_QP)%6;
  q_bits    = Q_BITS+qp_per;
  qp_const  = (1<<q_bits)/3;


  for (j=0;j<16;j++)
  {
    for (i=0;i<16;i++)
    {
      M1[i][j]=imgY_org[img->pix_y+j][img->pix_x+i]-img->mprr_2[new_intra_mode][j][i];
      M0[i%4][i/4][j%4][j/4]=M1[i][j];
    }
  }

  for (jj=0;jj<4;jj++)
  {
    for (ii=0;ii<4;ii++)
    {
      for (j=0;j<4;j++)
      {
        for (i=0;i<2;i++)
        {
          i1=3-i;
          M5[i]=  M0[i][ii][j][jj]+M0[i1][ii][j][jj];
          M5[i1]= M0[i][ii][j][jj]-M0[i1][ii][j][jj];
        }
        M0[0][ii][j][jj]=M5[0]+M5[1];
        M0[2][ii][j][jj]=M5[0]-M5[1];
        M0[1][ii][j][jj]=M5[3]*2+M5[2];
        M0[3][ii][j][jj]=M5[3]-M5[2]*2;
      }
      // vertical
      for (i=0;i<4;i++)
      {
        for (j=0;j<2;j++)
        {
          j1=3-j;
          M5[j] = M0[i][ii][j][jj]+M0[i][ii][j1][jj];
          M5[j1]= M0[i][ii][j][jj]-M0[i][ii][j1][jj];
        }
        M0[i][ii][0][jj]=M5[0]+M5[1];
        M0[i][ii][2][jj]=M5[0]-M5[1];
        M0[i][ii][1][jj]=M5[3]*2+M5[2];
        M0[i][ii][3][jj]=M5[3] -M5[2]*2;
      }
    }
  }

  // pick out DC coeff

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

  for (j=0;j<4;j++)
  {
    for (i=0;i<2;i++)
    {
      i1=3-i;
      M5[i]= M4[i][j]+M4[i1][j];
      M5[i1]=M4[i][j]-M4[i1][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;i++)
  {
    for (j=0;j<2;j++)
    {
      j1=3-j;
      M5[j]= M4[i][j]+M4[i][j1];
      M5[j1]=M4[i][j]-M4[i][j1];
    }
    M4[i][0]=(M5[0]+M5[1])/2;
    M4[i][2]=(M5[0]-M5[1])/2;
    M4[i][1]=(M5[3]+M5[2])/2;
    M4[i][3]=(M5[3]-M5[2])/2;
  }

  // quant

  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= (abs(M4[i][j]) * quant_coef[qp_rem][0][0] + 2*qp_const)>>(q_bits+1);

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

  // invers DC transform
  for (j=0;j<4;j++)
  {
    for (i=0;i<4;i++)
      M5[i]=M4[i][j];

    M6[0]=M5[0]+M5[2];
    M6[1]=M5[0]-M5[2];
    M6[2]=M5[1]-M5[3];
    M6[3]=M5[1]+M5[3];

    for (i=0;i<2;i++)
    {
      i1=3-i;
      M4[i][j]= M6[i]+M6[i1];
      M4[i1][j]=M6[i]-M6[i1];
    }
  }

  for (i=0;i<4;i++)
  {
    for (j=0;j<4;j++)
      M5[j]=M4[i][j];

    M6[0]=M5[0]+M5[2];
    M6[1]=M5[0]-M5[2];
    M6[2]=M5[1]-M5[3];
    M6[3]=M5[1]+M5[3];

    for (j=0;j<2;j++)
    {
      j1=3-j;
      M0[0][i][0][j] = ((M6[j]+M6[j1])*dequant_coef[qp_rem][0][0]<<qp_per)/4;
      M0[0][i][0][j1]= ((M6[j]-M6[j1])*dequant_coef[qp_rem][0][0]<<qp_per)/4;
    }
  }

  // AC invers trans/quant for MB
  for (jj=0;jj<4;jj++)
  {
    for (ii=0;ii<4;ii++)
    {
      run      = -1;
      scan_pos =  0;
      b8       = 2*(jj/2) + (ii/2);
      b4       = 2*(jj%2) + (ii%2);
      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=SNGL_SCAN[coeff_ctr][0];
        j=SNGL_SCAN[coeff_ctr][1];
        run++;

        level= ( abs( M0[i][ii][j][jj]) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;

        if (level != 0)
        {
          ac_coef = 15;
          ACLevel[scan_pos] = sign(level,M0[i][ii][j][jj]);
          ACRun  [scan_pos] = run;
          ++scan_pos;
          run=-1;
        }
        M0[i][ii][j][jj]=sign(level*dequant_coef[qp_rem][i][j]<<qp_per,M0[i][ii][j][jj]);
      }
      ACLevel[scan_pos] = 0;


      // IDCT horizontal

      for (j=0;j<4;j++)
      {
        for (i=0;i<4;i++)
        {
          M5[i]=M0[i][ii][j][jj];
        }

        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;
          M0[i][ii][j][jj] =M6[i]+M6[i1];
          M0[i1][ii][j][jj]=M6[i]-M6[i1];
        }
      }

      // vert
      for (i=0;i<4;i++)
      {
        for (j=0;j<4;j++)
          M5[j]=M0[i][ii][j][jj];

        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;
          M0[i][ii][ j][jj]=M6[j]+M6[j1];
          M0[i][ii][j1][jj]=M6[j]-M6[j1];

        }
      }

    }
  }

  for (j=0;j<16;j++)
  {
    for (i=0;i<16;i++)
    {
      M1[i][j]=M0[i%4][i/4][j%4][j/4];
    }
  }

  for (j=0;j<16;j++)
    for (i=0;i<16;i++)
      imgY[img->pix_y+j][img->pix_x+i]=(byte)min(255,max(0,(M1[i][j]+(img->mprr_2[new_intra_mode][j][i]<<DQ_BITS)+DQ_ROUND)>>DQ_BITS));

    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.
 *
 * \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.
 ************************************************************************
 */
int dct_luma(int block_x,int block_y,int *coeff_cost, int old_intra_mode)
{
  int sign(int a,int b);

  int i,j,i1,j1,ilev,m5[4],m6[4],coeff_ctr,scan_loop_ctr;
  int qp_const,level,scan_pos,run;
  int nonzero;
  int idx;
  int scan_mode;
  int loop_rep;
  int qp_per,qp_rem,q_bits;

  int   pos_x   = block_x/BLOCK_SIZE;
  int   pos_y   = block_y/BLOCK_SIZE;
  int   b8      = 2*(pos_y/2) + (pos_x/2);
  int   b4      = 2*(pos_y%2) + (pos_x%2);
  int*  ACLevel = img->cofAC[b8][b4][0];
  int*  ACRun   = img->cofAC[b8][b4][1];

  qp_per    = (img->qp-MIN_QP)/6;
  qp_rem    = (img->qp-MIN_QP)%6;
  q_bits    = Q_BITS+qp_per;

  if (img->type == INTRA_IMG)
    qp_const=(1<<q_bits)/3;    // intra
  else
    qp_const=(1<<q_bits)/6;    // inter

  //  Horizontal transform
  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]);
    img->m7[2][j]=(m5[0]-m5[1]);
    img->m7[1][j]=m5[3]*2+m5[2];
    img->m7[3][j]=m5[3]-m5[2]*2;
  }

  //  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]);
    img->m7[i][2]=(m5[0]-m5[1]);
    img->m7[i][1]=m5[3]*2+m5[2];
    img->m7[i][3]=m5[3]-m5[2]*2;
  }

  // Quant

  nonzero=FALSE;

  if (old_intra_mode && (img->qp<24 || input->symbol_mode == CABAC)) // for CABAC double scan always
  {