block.c

h.264 影像壓縮 必須在 .net 的環境 下操作

/*!
************************************************************************
* \brief
*    Residual DPCM for Intra lossless coding
************************************************************************
*/
//For residual DPCM
int Residual_DPCM_4x4_for_Intra16x16(int ipmode)  
{
  int i,j;
  int temp[4][4];

  if(ipmode==VERT_PRED)
  {
    for (j=0; j<4; j++)
      for (i=1; i<4; i++) 
        temp[i][j] = lossless_res[i][j] - lossless_res[i-1][j];

    for (j=0; j<4; j++)
      for (i=1; i<4; i++)
        lossless_res[i][j] = temp[i][j];
  }
  else  //HOR_PRED
  {
    for (j=1; j<4; j++)
      for (i=0; i<4; i++)
        temp[i][j] = lossless_res[i][j] - lossless_res[i][j-1];

    for (j=1; j<4; j++)
      for (i=0; i<4; i++)
        lossless_res[i][j] = temp[i][j];
  }
  return 0;
}
/*!
************************************************************************
* \brief
*    Inverse residual DPCM for Intra lossless coding
************************************************************************
*/
//For residual DPCM
int Inv_Residual_DPCM_4x4_for_Intra16x16(int ipmode)  
{
  int i;
  int temp[4][4];

  if(ipmode==VERT_PRED)
  {
    for (i=0; i<4; i++) 
    {
      temp[1][i]=lossless_res[0][i]+lossless_res[1][i];
      temp[2][i]=lossless_res[0][i]+lossless_res[1][i]+lossless_res[2][i];
      temp[3][i]=lossless_res[0][i]+lossless_res[1][i]+lossless_res[2][i]+lossless_res[3][i];
    }
    for (i=0; i<4; i++)
    {
      lossless_res[1][i]=temp[1][i];
      lossless_res[2][i]=temp[2][i];
      lossless_res[3][i]=temp[3][i];
    }
  }
  else  //HOR_PRED
  {
    for(i=0; i<4; i++)
    {
      temp[i][1]=lossless_res[i][0]+lossless_res[i][1];
      temp[i][2]=lossless_res[i][0]+lossless_res[i][1]+lossless_res[i][2];
      temp[i][3]=lossless_res[i][0]+lossless_res[i][1]+lossless_res[i][2]+lossless_res[i][3];
    }
    for (i=0; i<4; i++)
    {
      lossless_res[i][1]=temp[i][1];
      lossless_res[i][2]=temp[i][2];
      lossless_res[i][3]=temp[i][3];
    }    
  }
  return 0;
}

/*!
 ************************************************************************
 * \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(Macroblock *currMB, int uv, int cr_cbp)
{
  int i,j,n2,n1,coeff_ctr,level ,scan_pos,run;
  static int m1[BLOCK_SIZE],m5[BLOCK_SIZE],m6[BLOCK_SIZE];
  int coeff_cost;
  int cr_cbp_tmp;
  int DCcoded=0 ;
  static imgpel *orig_img, *pred_img;

  const byte *c_cost = COEFF_COST4x4[input->disthres];

  int   b4;
  int*  DCLevel = img->cofDC[uv+1][0];
  int*  DCRun   = img->cofDC[uv+1][1];
  int*  ACLevel;
  int*  ACRun;
  int   intra = IS_INTRA (currMB);
  int   uv_scale = uv * (img->num_blk8x8_uv >> 1);
  static int scaled_coeff;

  //FRExt
  static const int64 cbpblk_pattern[4]={0, 0xf0000, 0xff0000, 0xffff0000};
  int yuv = img->yuv_format;
  int b8;
  static int *m7;
  static int m3[4][4];
  static int m4[4][4];
  int qp_per_dc = 0;
  int qp_rem_dc = 0;
  int q_bits_422 = 0;
  const byte (*pos_scan)[2] = currMB->is_field_mode ? FIELD_SCAN : SNGL_SCAN;
  int cur_qp = currMB->qpc[uv] + img->bitdepth_chroma_qp_scale;
  int cur_qp_dc = currMB->qpc[uv] + 3 + img->bitdepth_chroma_qp_scale;

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

  int qp_per = qp_per_matrix[cur_qp];
  int qp_rem = qp_rem_matrix[cur_qp];
  int q_bits = Q_BITS + qp_per;
  int max_imgpel_value_uv = img->max_imgpel_value_comp[1];
  int    (*curr_res)[16] = img->m7[uv + 1]; 
  imgpel (*curr_mpr)[16] = img->mpr[uv + 1]; 

  levelscale    = LevelScale4x4Comp   [uv + 1][intra][qp_rem];
  leveloffset   = LevelOffset4x4Comp  [uv + 1][intra][cur_qp];
  invlevelscale = InvLevelScale4x4Comp[uv + 1][intra][qp_rem];
  fadjust4x4    = img->AdaptiveRounding ? img->fadjust4x4Cr[intra][uv] : NULL;

  if (img->yuv_format == YUV422)
  {
    //for YUV422 only
    qp_per_dc = qp_per_matrix[cur_qp_dc];
    qp_rem_dc = qp_rem_matrix[cur_qp_dc];
    
    invlevelscaleDC = InvLevelScale4x4Comp[uv + 1][intra][qp_rem_dc];
    levelscaleDC    = LevelScale4x4Comp   [uv + 1][intra][qp_rem_dc];
    leveloffsetDC   = LevelOffset4x4Comp  [uv + 1][intra][cur_qp_dc];

    q_bits_422 = Q_BITS + qp_per_dc;
  }


  //============= dct transform ===============
  if (!lossless_qpprime)
  {
    for (n2=0; n2 < img->mb_cr_size_y; n2 += BLOCK_SIZE)
    {
      for (n1=0; n1 < img->mb_cr_size_x; n1 += BLOCK_SIZE)
      {
        forward4x4(curr_res, curr_res, n1, n2);
      }
    }
  }

  if (yuv == YUV420)
  {
    //================== CHROMA DC YUV420 ===================
    //     2X2 transform of DC coeffs.
    run=-1;
    scan_pos=0;
    if(!lossless_qpprime)
    {
      m1[0]=(curr_res[0][0] + curr_res[0][4] + curr_res[4][0] + curr_res[4][4]);
      m1[1]=(curr_res[0][0] - curr_res[0][4] + curr_res[4][0] - curr_res[4][4]);
      m1[2]=(curr_res[0][0] + curr_res[0][4] - curr_res[4][0] - curr_res[4][4]);
      m1[3]=(curr_res[0][0] - curr_res[0][4] - curr_res[4][0] + curr_res[4][4]);

      //     Quant of chroma 2X2 coeffs.
      for (coeff_ctr=0; coeff_ctr < 4; coeff_ctr++)
      {
        run++;

        level =(iabs(m1[coeff_ctr]) * levelscale[0][0] + (leveloffset[0][0]<<1)) >> (q_bits+1);

        if (level  != 0)
        {
          if (input->symbol_mode == CAVLC && img->qp < 4)
            level = imin(level, CAVLC_LEVEL_LIMIT);

          currMB->cbp_blk |= 0xf0000 << (uv << 2) ;    // if one of the 2x2-DC levels is != 0 set the
          cr_cbp=imax(1,cr_cbp);                     // coded-bit all 4 4x4 blocks (bit 16-19 or 20-23)
          DCcoded = 1;
          level = isignab(level, m1[coeff_ctr]);
          DCLevel[scan_pos  ] = level;
          DCRun  [scan_pos++] = run;
          run=-1;          
          m1[coeff_ctr] = level;
        }
        else
          m1[coeff_ctr] = 0;
      }

      DCLevel[scan_pos] = 0;
      //  Inverse transform of 2x2 DC levels
      m5[0]=(m1[0] + m1[1] + m1[2] + m1[3]);
      m5[1]=(m1[0] - m1[1] + m1[2] - m1[3]);
      m5[2]=(m1[0] + m1[1] - m1[2] - m1[3]);
      m5[3]=(m1[0] - m1[1] - m1[2] + m1[3]);

      curr_res[0][0] = ((m5[0] * invlevelscale[0][0]) << qp_per) >> 5;
      curr_res[0][4] = ((m5[1] * invlevelscale[0][0]) << qp_per) >> 5;
      curr_res[4][0] = ((m5[2] * invlevelscale[0][0]) << qp_per) >> 5;
      curr_res[4][4] = ((m5[3] * invlevelscale[0][0]) << qp_per) >> 5;
    }
    else // Lossless qpprime
    {
      m1[0]=curr_res[0][0];
      m1[1]=curr_res[0][4];
      m1[2]=curr_res[4][0];
      m1[3]=curr_res[4][4];

      for (coeff_ctr=0; coeff_ctr < 4; coeff_ctr++)
      {
        run++;

        level =iabs(m1[coeff_ctr]);

        if (level  != 0)
        {
          if (input->symbol_mode == CAVLC && img->qp < 4)
            level = imin(level, CAVLC_LEVEL_LIMIT);

          currMB->cbp_blk |= 0xf0000 << (uv << 2) ;    // if one of the 2x2-DC levels is != 0 set the
          cr_cbp=imax(1, cr_cbp);                     // coded-bit all 4 4x4 blocks (bit 16-19 or 20-23)
          DCcoded = 1;
          level = isignab(level, m1[coeff_ctr]);
          DCLevel[scan_pos  ] = level;
          DCRun  [scan_pos++] = run;
          run=-1;
        }
      }
      DCLevel[scan_pos] = 0;
    }
  }
  else if(yuv == YUV422)
  {
    //================== CHROMA DC YUV422 ===================
    //transform DC coeff
    //horizontal

    //pick out DC coeff
    for (j=0; j < img->mb_cr_size_y; j+=BLOCK_SIZE)
    {
      for (i=0; i < img->mb_cr_size_x; i+=BLOCK_SIZE)
        m3[i>>2][j>>2]= curr_res[j][i];
    }

    //horizontal
    if(!lossless_qpprime)
    {
      m4[0][0] = m3[0][0] + m3[1][0];
      m4[0][1] = m3[0][1] + m3[1][1];
      m4[0][2] = m3[0][2] + m3[1][2];
      m4[0][3] = m3[0][3] + m3[1][3];

      m4[1][0] = m3[0][0] - m3[1][0];
      m4[1][1] = m3[0][1] - m3[1][1];
      m4[1][2] = m3[0][2] - m3[1][2];
      m4[1][3] = m3[0][3] - m3[1][3];

      // vertical
      for (i=0;i<2;i++)
      {
        m5[0] = m4[i][0] + m4[i][3];
        m5[1] = m4[i][1] + m4[i][2];
        m5[2] = m4[i][1] - m4[i][2];
        m5[3] = m4[i][0] - m4[i][3];

        m4[i][0] = (m5[0] + m5[1]);
        m4[i][2] = (m5[0] - m5[1]);
        m4[i][1] = (m5[3] + m5[2]);
        m4[i][3] = (m5[3] - m5[2]);
      }
    }

    run=-1;
    scan_pos=0;

    //quant of chroma DC-coeffs
    for (coeff_ctr=0;coeff_ctr<8;coeff_ctr++)
    {
      i=SCAN_YUV422[coeff_ctr][0];
      j=SCAN_YUV422[coeff_ctr][1];

      run++;

      if(lossless_qpprime)
      {
        level = iabs(m3[i][j]);
        m4[i][j]=m3[i][j];
      }
      else
        level =(iabs(m4[i][j]) * levelscale[0][0] + (leveloffsetDC[0][0]*2)) >> (q_bits_422+1);

      if (level != 0)
      {
        //YUV422
        currMB->cbp_blk |= 0xff0000 << (uv << 3) ;   // if one of the DC levels is != 0 set the
        cr_cbp=imax(1,cr_cbp);                       // coded-bit all 4 4x4 blocks (bit 16-31 or 32-47) //YUV444
        DCcoded = 1 ;

        DCLevel[scan_pos  ] = isignab(level,m4[i][j]);
        DCRun  [scan_pos++] = run;
        run=-1;
      }
      if(!lossless_qpprime)
        m3[i][j]=isignab(level,m4[i][j]);
    }
    DCLevel[scan_pos]=0;

    //inverse DC transform
    //horizontal
    if(!lossless_qpprime)
    {
      m4[0][0] = m3[0][0] + m3[1][0];
      m4[0][1] = m3[0][1] + m3[1][1];
      m4[0][2] = m3[0][2] + m3[1][2];
      m4[0][3] = m3[0][3] + m3[1][3];

      m4[1][0] = m3[0][0] - m3[1][0];
      m4[1][1] = m3[0][1] - m3[1][1];
      m4[1][2] = m3[0][2] - m3[1][2];
      m4[1][3] = m3[0][3] - m3[1][3];

      // vertical
      for (i=0;i<2;i++)
      {
        m6[0]=m4[i][0]+m4[i][2];
        m6[1]=m4[i][0]-m4[i][2];
        m6[2]=m4[i][1]-m4[i][3];
        m6[3]=m4[i][1]+m4[i][3];

        if(qp_per_dc < 4)
        {
          curr_res[0 ][i*4] = ((((m6[0]+m6[3])*invlevelscaleDC[0][0] + (1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
          curr_res[4 ][i*4] = ((((m6[1]+m6[2])*invlevelscaleDC[0][0] + (1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
          curr_res[8 ][i*4] = ((((m6[1]-m6[2])*invlevelscaleDC[0][0] + (1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
          curr_res[12][i*4] = ((((m6[0]-m6[3])*invlevelscaleDC[0][0] + (1<<(3-qp_per_dc)))>>(4-qp_per_dc))+2)>>2;
        }
        else
        {
          curr_res[0 ][i*4] = ((((m6[0]+m6[3])*invlevelscaleDC[0][0])<<(qp_per_dc-4))+2)>>2;
          curr_res[4 ][i*4] = ((((m6[1]+m6[2])*invlevelscaleDC[0][0])<<(qp_per_dc-4))+2)>>2;
          curr_res[8 ][i*4] = ((((m6[1]-m6[2])*invlevelscaleDC[0][0])<<(qp_per_dc-4))+2)>>2;
          curr_res[12][i*4] = ((((m6[0]-m6[3])*invlevelscaleDC[0][0])<<(qp_per_dc-4))+2)>>2;
        }
      }//for (i=0;i<2;i++)
    }
  }

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

  if(!lossless_qpprime)
  {
    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];
          m7 = &curr_res[n2+j][n1];

          ++run;
          scaled_coeff = iabs(m7[i])*levelscale[j][i];
          level=(scaled_coeff + leveloffset[j][i]) >> q_bits;

          if (level  != 0)
          {
            if (img->AdaptiveRounding)
              fadjust4x4[n2+j][n1+i] = rshift_rnd_sf((AdaptRndCrWeight * (scaled_coeff - (level << q_bits))), (q_bits + 1));

            currMB->cbp_blk |= uv_cbpblk;
            // if level > 1 set high cost to avoid thresholding
            coeff_cost += (level > 1) ? MAX_VALUE : c_cost[run];

            cr_cbp_tmp=2;
            level=isignab(level, m7[i]);
            ACLevel[scan_pos  ] = level;
            ACRun  [scan_pos++] = run;
            run=-1;          

            m7[i] = rshift_rnd_sf((level * invlevelscale[j][i]) << qp_per, 4);
            // inverse scale can be alternative performed as follows to ensure 16bit
            // arithmetic is satisfied.
            // curr_res[n2+j][n1+i] = (qp_per<4)