block.c

h.264 影像壓縮 必須在 .net 的環境 下操作

            //      ? rshift_rnd_sf((level*invlevelscale[j][i]),4-qp_per);
            //      : (level*invlevelscale[j][i])<<(qp_per-4);
          }
          else
          {
            m7[i] = 0;
            if (img->AdaptiveRounding)
              fadjust4x4[n2+j][n1+i] = 0;
          }
        }
        ACLevel[scan_pos] = 0;
      }
    }
  
      // Perform thresholding
    // * reset chroma coeffs
    if(coeff_cost < _CHROMA_COEFF_COST_)
    {
      int64 uv_cbpblk = ((int64)cbpblk_pattern[yuv] << (uv << (1+yuv)));
      cr_cbp_tmp = 0;

      for (b8 = 0; b8 < (img->num_blk8x8_uv >> 1); b8++)
      {
        for (b4 = 0; b4 < 4; b4++)
        {
          n1 = hor_offset[yuv][b8][b4];
          n2 = ver_offset[yuv][b8][b4];
          ACLevel = img->cofAC[4 + b8 + uv_scale][b4][0];
          ACRun   = img->cofAC[4 + b8 + uv_scale][b4][1];

          if (DCcoded == 0)
            currMB->cbp_blk &= ~(uv_cbpblk);  // if no chroma DC's: then reset coded-bits of this chroma subblock

          ACLevel[0] = 0;

          for (coeff_ctr=1; coeff_ctr < 16; coeff_ctr++)// ac coeff
          {
            curr_res[n2 + pos_scan[coeff_ctr][1]][n1 + pos_scan[coeff_ctr][0]] = 0;
            ACLevel[coeff_ctr]  = 0;
          }
        }
      }
    }

    //     IDCT.
    //     Horizontal.
    if(cr_cbp_tmp == 2)
      cr_cbp = 2;

    for (n2=0; n2 < img->mb_cr_size_y; n2 += BLOCK_SIZE)
    {
      for (n1=0; n1 < img->mb_cr_size_x; n1 += BLOCK_SIZE)
      {
        inverse4x4(curr_res, curr_res, n2, n1);
      }
    }

    //  Decoded block moved to memory
    for (j=0; j < img->mb_cr_size_y; j++)
    {
      orig_img = &enc_picture->imgUV[uv][img->pix_c_y + j][img->pix_c_x];
      pred_img = curr_mpr[j];
      m7 = curr_res[j];
      for (i=0; i < img->mb_cr_size_x; i++)
      {
        orig_img[i]= (imgpel) iClip1(max_imgpel_value_uv, rshift_rnd_sf((m7[i]),DQ_BITS) + pred_img[i]);
      }
    }
  }
  else
  {
    for (b8=0; b8 < (img->num_blk8x8_uv >> 1); b8++)
    {
      for (b4=0; b4 < 4; b4++)
      {
        int64 uv_cbpblk = ((int64)1) << cbp_blk_chroma[b8 + uv_scale][b4];
        n1 = hor_offset[yuv][b8][b4];
        n2 = ver_offset[yuv][b8][b4];
        ACLevel = img->cofAC[4 + b8 + uv_scale][b4][0];
        ACRun   = img->cofAC[4 + b8 + uv_scale][b4][1];
        run=-1;
        scan_pos=0;

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

          ++run;

          level = iabs(curr_res[n2+j][n1+i]);

          if (img->AdaptiveRounding)
          {
            fadjust4x4[n2+j][n1+i] = 0;
          }

          if (level  != 0)
          {
            currMB->cbp_blk |= uv_cbpblk;
            coeff_cost += MAX_VALUE;                // set high cost, shall not be discarded

            cr_cbp_tmp=2;
            ACLevel[scan_pos  ] = isignab(level,curr_res[n2+j][n1+i]);
            ACRun  [scan_pos++] = run;
            run=-1;

            level=isignab(level, curr_res[n2+j][n1+i]);
          }
        }
        ACLevel[scan_pos] = 0;
      }
    }

    for (j=0; j < img->mb_cr_size_y; j++)
    {      
      orig_img = &enc_picture->imgUV[uv][img->pix_c_y + j][img->pix_c_x];
      m7 = curr_res[j];
      pred_img = curr_mpr[j];
      for (i=0; i < img->mb_cr_size_x; i++)
      {        
        orig_img[i] = (imgpel) m7[i] + pred_img[i];
      }
    }
  }

  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_4x4_sp(Macroblock *currMB, ColorPlane pl, int block_x,int block_y,int *coeff_cost, int intra)
{
  int i,j,j1,m5[4],m6[4],coeff_ctr;
  int qp_const,ilev, level,scan_pos,run;
  int nonzero;

  imgpel **img_enc = enc_picture->p_curr_img;
  imgpel (*curr_mpr)[16] = img->mpr[pl];
  int    (*curr_res)[16] = img->m7[pl]; 
  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;
  const byte *c_cost = COEFF_COST4x4[input->disthres];

  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];

  // 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    = qp_per_matrix[(currMB->qp - MIN_QP)];
  qp_rem    = qp_rem_matrix[(currMB->qp - MIN_QP)];
  q_bits    = Q_BITS + qp_per;
  qp_per_sp = qp_per_matrix[(currMB->qpsp - MIN_QP)];
  qp_rem_sp = qp_rem_matrix[(currMB->qpsp - MIN_QP)];
  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++)
    {
      curr_res[j][i]+=curr_mpr[j+block_y][i+block_x];
      predicted_block[i][j]=curr_mpr[j+block_y][i+block_x];
    }

  for (j=0; j < BLOCK_SIZE; j++)
  {    
    m5[0] = curr_res[j][0] + curr_res[j][3];
    m5[1] = curr_res[j][1] + curr_res[j][2];
    m5[2] = curr_res[j][1] - curr_res[j][2];      
    m5[3] = curr_res[j][0] - curr_res[j][3];
    
    curr_res[j][0] = (m5[0]   + m5[1]);
    curr_res[j][2] = (m5[0]   - m5[1]);
    curr_res[j][1] =  m5[3]*2 + m5[2];
    curr_res[j][3] =  m5[3]   - m5[2]*2;
  }

  //  Vertical transform

  for (i=0; i < BLOCK_SIZE; i++)
  {    
    m5[0] = curr_res[0][i] + curr_res[3][i];
    m5[1] = curr_res[1][i] + curr_res[2][i];
    m5[2] = curr_res[1][i] - curr_res[2][i];    
    m5[3] = curr_res[0][i] - curr_res[3][i];

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

  for (j=0; j < BLOCK_SIZE; j++)
  {
    m5[0] = predicted_block[0][j] + predicted_block[3][j];
    m5[1] = predicted_block[1][j] + predicted_block[2][j];
    m5[2] = predicted_block[1][j] - predicted_block[2][j];      
    m5[3] = predicted_block[0][j] - predicted_block[3][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 (currMB->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 = (iabs (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 = curr_res[j][i]-isignab(level1, predicted_block[i][j]);
    level1 = (iabs (c_err1) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;

    // case 2
    c_err2=curr_res[j][i]-predicted_block[i][j];
    level2 = (iabs (c_err2) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;

    // select prediction
    if ((level1 != level2) && (level1 != 0) && (level2 != 0))
    {
      D_dis1 = curr_res[j][i] - ((isignab(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 = curr_res[j][i] - ((isignab(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 = (iabs(level1) < iabs(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 += c_cost[run];

      ACLevel[scan_pos] = isignab(level,c_err);
      ACRun  [scan_pos] = run;
      ++scan_pos;
      run=-1;                     // reset zero level counter
      ilev=((isignab(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]=
        isignab((iabs(ilev) * quant_coef[qp_rem_sp][i][j] + qp_const2) >> q_bits_sp, ilev);
    }    
    curr_res[j][i] = isignab((iabs(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]=curr_res[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);

    curr_res[j][0]=m6[0]+m6[3];
    curr_res[j][1]=m6[1]+m6[2];
    curr_res[j][2]=m6[1]-m6[2];    
    curr_res[j][3]=m6[0]-m6[3];
  }

  //  vertical

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

    curr_res[0][i] = iClip1 (img->max_imgpel_value,rshift_rnd_sf(m6[0]+m6[3], DQ_BITS));
    curr_res[1][i] = iClip1 (img->max_imgpel_value,rshift_rnd_sf(m6[1]+m6[2], DQ_BITS));
    curr_res[2][i] = iClip1 (img->max_imgpel_value,rshift_rnd_sf(m6[1]-m6[2], DQ_BITS));
    curr_res[3][i] = iClip1 (img->max_imgpel_value,rshift_rnd_sf(m6[0]-m6[3], DQ_BITS));
  }

  //  Decoded block moved to frame memory

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

  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
 *    once for each of the chroma components.
 *
 * \par Input:
 *    uv    : Make difference between the U and V chroma component               \n
 *    cr_cbp: chroma coded block pattern
 *
 * \par Output:
 *    cr_cbp: Updated chroma coded block pattern.
 ************************************************************************
 */
int dct_chroma_sp(Macroblock *currMB, int uv,int cr_cbp)
{
  int i,j,i1,j2,ilev,n2,n1,j1,mb_y,coeff_ctr,qp_const,c_err,level ,scan_pos,run;
  int m1[BLOCK_SIZE],m5[BLOCK_SIZE],m6[BL