block.c

h.264/avc 视频编码程序,实现分数像素匹配功能,非原创.

    scan_mode=DOUBLE_SCAN;
    loop_rep=2;
    idx=1;
  }
  else
  {
    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;

      level = (abs (img->m7[i][j]) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;

      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];
        ACLevel[scan_pos] = sign(level,img->m7[i][j]);
        ACRun  [scan_pos] = run;
        ++scan_pos;
        run=-1;                     // reset zero level counter
        ilev=level*dequant_coef[qp_rem][i][j]<<qp_per;
      }
      img->m7[i][j]=sign(ilev,img->m7[i][j]);
    }
    ACLevel[scan_pos] = 0;
  }


  //     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]);
    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[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]);
    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[i][j] =min(255,max(0,(m6[j]+m6[j1]+(img->mpr[i+block_x][j+block_y] <<DQ_BITS)+DQ_ROUND)>>DQ_BITS));
      img->m7[i][j1]=min(255,max(0,(m6[j]-m6[j1]+(img->mpr[i+block_x][j1+block_y]<<DQ_BITS)+DQ_ROUND)>>DQ_BITS));
    }
  }

  //  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;
}



/*!
 ************************************************************************
 * \brief
 *    Transform,quantization,inverse transform for chroma.
 *    The main reason why this is done in a separate routine is the
 *    additional 2x2 transform of DC-coeffs. This routine is called
 *    ones for each of the chroma components.
 *
 * \para Input:
 *    uv    : Make difference between the U and V chroma component  \n
 *    cr_cbp: chroma coded block pattern
 *
 * \para Output:
 *    cr_cbp: Updated chroma coded block pattern.
 ************************************************************************
 */
int dct_chroma(int uv,int cr_cbp)
{
  int i,j,i1,j2,ilev,n2,n1,j1,mb_y,coeff_ctr,qp_const,level ,scan_pos,run;
  int m1[BLOCK_SIZE],m5[BLOCK_SIZE],m6[BLOCK_SIZE];
  int coeff_cost;
  int cr_cbp_tmp;
  int nn0,nn1;
  int DCcoded=0 ;
  Macroblock *currMB = &img->mb_data[img->current_mb_nr];

  int qp_per,qp_rem,q_bits;

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

  qp_per    = ((img->qp<0?img->qp:QP_SCALE_CR[img->qp])-MIN_QP)/6;
  qp_rem    = ((img->qp<0?img->qp:QP_SCALE_CR[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

  for (n2=0; n2 <= BLOCK_SIZE; n2 += BLOCK_SIZE)
  {
    for (n1=0; n1 <= BLOCK_SIZE; n1 += BLOCK_SIZE)
    {

      //  Horizontal transform.
      for (j=0; j < BLOCK_SIZE; j++)
      {
        mb_y=n2+j;
        for (i=0; i < 2; i++)
        {
          i1=3-i;
          m5[i]=img->m7[i+n1][mb_y]+img->m7[i1+n1][mb_y];
          m5[i1]=img->m7[i+n1][mb_y]-img->m7[i1+n1][mb_y];
        }
        img->m7[n1][mb_y]  =(m5[0]+m5[1]);
        img->m7[n1+2][mb_y]=(m5[0]-m5[1]);
        img->m7[n1+1][mb_y]=m5[3]*2+m5[2];
        img->m7[n1+3][mb_y]=m5[3]-m5[2]*2;
      }

      //  Vertical transform.

      for (i=0; i < BLOCK_SIZE; i++)
      {
        j1=n1+i;
        for (j=0; j < 2; j++)
        {
          j2=3-j;
          m5[j]=img->m7[j1][n2+j]+img->m7[j1][n2+j2];
          m5[j2]=img->m7[j1][n2+j]-img->m7[j1][n2+j2];
        }
        img->m7[j1][n2+0]=(m5[0]+m5[1]);
        img->m7[j1][n2+2]=(m5[0]-m5[1]);
        img->m7[j1][n2+1]=m5[3]*2+m5[2];
        img->m7[j1][n2+3]=m5[3]-m5[2]*2;
      }
    }
  }

  //     2X2 transform of DC coeffs.
  m1[0]=(img->m7[0][0]+img->m7[4][0]+img->m7[0][4]+img->m7[4][4]);
  m1[1]=(img->m7[0][0]-img->m7[4][0]+img->m7[0][4]-img->m7[4][4]);
  m1[2]=(img->m7[0][0]+img->m7[4][0]-img->m7[0][4]-img->m7[4][4]);
  m1[3]=(img->m7[0][0]-img->m7[4][0]-img->m7[0][4]+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;

    level =(abs(m1[coeff_ctr]) * quant_coef[qp_rem][0][0] + 2*qp_const) >> (q_bits+1);

    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=level*dequant_coef[qp_rem][0][0]<<qp_per;
    }
    m1[coeff_ctr]=sign(ilev,m1[coeff_ctr]);
  }
  DCLevel[scan_pos] = 0;

  //  Invers transform of 2x2 DC levels

  img->m7[0][0]=(m1[0]+m1[1]+m1[2]+m1[3])/2;
  img->m7[4][0]=(m1[0]-m1[1]+m1[2]-m1[3])/2;
  img->m7[0][4]=(m1[0]+m1[1]-m1[2]-m1[3])/2;
  img->m7[4][4]=(m1[0]-m1[1]-m1[2]+m1[3])/2;

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

  for (n2=0; n2 <= BLOCK_SIZE; n2 += BLOCK_SIZE)
  {
    for (n1=0; n1 <= BLOCK_SIZE; n1 += BLOCK_SIZE)
    {
      b4      = 2*(n2/4) + (n1/4);
      ACLevel = img->cofAC[uv+4][b4][0];
      ACRun   = img->cofAC[uv+4][b4][1];
      run=-1;
      scan_pos=0;

      for (coeff_ctr=1; coeff_ctr < 16; coeff_ctr++)// start change rd_quant
      {
        i = SNGL_SCAN[coeff_ctr][0];
        j = SNGL_SCAN[coeff_ctr][1];
        ++run;
        ilev=0;

        level=(abs(img->m7[n1+i][n2+j])*quant_coef[qp_rem][i][j]+qp_const)>>q_bits;

        if (level  != 0)
        {
          currMB->cbp_blk |= 1 << (16 + (uv << 2) + ((n2 >> 1) + (n1 >> 2))) ;
          if (level > 1)
            coeff_cost += MAX_VALUE;                // set high cost, shall not be discarded
          else
            coeff_cost += COEFF_COST[run];

          cr_cbp_tmp=2;
          ACLevel[scan_pos] = sign(level,img->m7[n1+i][n2+j]);
          ACRun  [scan_pos] = run;
          ++scan_pos;
          run=-1;
          ilev=level*dequant_coef[qp_rem][i][j]<<qp_per;
        }
        img->m7[n1+i][n2+j]=sign(ilev,img->m7[n1+i][n2+j]); // for use in IDCT
      }
      ACLevel[scan_pos] = 0;
    }
  }

  // * reset chroma coeffs
  if(coeff_cost<7)
  {
    cr_cbp_tmp = 0 ;
    for (n2=0; n2 <= BLOCK_SIZE; n2 += BLOCK_SIZE)
    {
      for (n1=0; n1 <= BLOCK_SIZE; n1 += BLOCK_SIZE)
      {
        b4      = 2*(n2/4) + (n1/4);
        ACLevel = img->cofAC[uv+4][b4][0];
        ACRun   = img->cofAC[uv+4][b4][1];
        if( DCcoded == 0) currMB->cbp_blk &= ~(0xf0000 << (uv << 2));  // if no chroma DC's: then reset coded-bits of this chroma subblock
        nn0 = (n1>>2) + (uv<<1);
        nn1 = 4 + (n2>>2) ;
        ACLevel[0] = 0;
        for (coeff_ctr=1; coeff_ctr < 16; coeff_ctr++)// ac coeff
        {
          i=SNGL_SCAN[coeff_ctr][0];
          j=SNGL_SCAN[coeff_ctr][1];
          img->m7[n1+i][n2+j]=0;
          ACLevel[coeff_ctr] = 0;
        }
      }
    }
  }
  if(cr_cbp_tmp==2)
    cr_cbp = 2;
  //     IDCT.

      //     Horizontal.
  for (n2=0; n2 <= BLOCK_SIZE; n2 += BLOCK_SIZE)
  {
    for (n1=0; n1 <= BLOCK_SIZE; n1 += BLOCK_SIZE)
    {
      for (j=0; j < BLOCK_SIZE; j++)
      {
        for (i=0; i < BLOCK_SIZE; i++)
        {
          m5[i]=img->m7[n1+i][n2+j];
        }
        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[n1+i][n2+j]=m6[i]+m6[i1];
          img->m7[n1+i1][n2+j]=m6[i]-m6[i1];
        }
      }

      //     Vertical.
      for (i=0; i < BLOCK_SIZE; i++)
      {
        for (j=0; j < BLOCK_SIZE; j++)
        {
          m5[j]=img->m7[n1+i][n2+j];
        }
        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++)
        {
          j2=3-j;
          img->m7[n1+i][n2+j] =min(255,max(0,(m6[j]+m6[j2]+(img->mpr[n1+i][n2+j] <<DQ_BITS)+DQ_ROUND)>>DQ_BITS));
          img->m7[n1+i][n2+j2]=min(255,max(0,(m6[j]-m6[j2]+(img->mpr[n1+i][n2+j2]<<DQ_BITS)+DQ_ROUND)>>DQ_BITS));
        }
      }
    }
  }

  //  Decoded block moved to memory
  for (j=0; j < BLOCK_SIZE*2; j++)
    for (i=0; i < BLOCK_SIZE*2; i++)
      imgUV[uv][img->pix_c_y+j][img->pix_c_x+i]= img->m7[i][j];

  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.
 *
 * \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_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,level,scan_pos,run;
  int nonzero;
  int idx;

  int scan_mode;
  int loop_rep;
  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_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;
  qp_per_sp    = (img->qpsp-MIN_QP)/6;
  qp_rem_sp    = (img->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[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]);
    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;
  }

  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;
  }

  //  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]);
    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;
  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)
      {