block.c

G729、h263、h264、MPEG4四种最流行的音频和视频标准的压缩和解压算法的源代码.rar

* \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_luma(int block_x,int block_y,int *coeff_cost, int intra)
{
  int sign(int a,int b);

  int i,j,i1,j1,ilev,m5[4],m6[4],coeff_ctr;
  //int qp_const;
  int level,scan_pos,run;
  int nonzero;
  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];

  Macroblock *currMB = &img->mb_data[img->current_mb_nr];
  Boolean lossless_qpprime = ((currMB->qp + img->bitdepth_luma_qp_scale)==0 && img->lossless_qpprime_flag==1);

  qp_per    = (currMB->qp + img->bitdepth_luma_qp_scale - MIN_QP)/6; 
  qp_rem    = (currMB->qp + img->bitdepth_luma_qp_scale - MIN_QP)%6; 
  q_bits    = Q_BITS+qp_per;
/*
  if (img->type == I_SLICE)
    qp_const=(1<<q_bits)/3;    // intra
  else
    qp_const=(1<<q_bits)/6;    // inter
*/
//  printf("q_bits %d %d %d %d\n",q_bits,qp_per,qp_const,LevelOffset4x4Luma_Intra[qp_per][0][0]);
  //  Horizontal transform
  for (j=0; j < BLOCK_SIZE && !lossless_qpprime; 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;
  }

  //  Vertical transform
  for (i=0; i < BLOCK_SIZE && !lossless_qpprime; 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;

  run=-1;
  scan_pos=0;
  
  for (coeff_ctr=0;coeff_ctr < 16;coeff_ctr++)
  {

    if (img->field_picture || ( img->MbaffFrameFlag && currMB->mb_field )) 
    {  // 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;
    
    if(lossless_qpprime)
      level = abs (img->m7[i][j]);
    else if(intra == 1)
      level = (abs (img->m7[i][j]) * LevelScale4x4Luma_Intra[qp_rem][i][j] + LevelOffset4x4Luma_Intra[qp_per][i][j]) >> q_bits;
    //level = (abs (img->m7[i][j]) * LevelScale4x4Luma_Intra[qp_rem][i][j] + qp_const) >> q_bits;    
    else
      level = (abs (img->m7[i][j]) * LevelScale4x4Luma_Inter[qp_rem][i][j] + LevelOffset4x4Luma_Inter[qp_per][i][j]) >> q_bits;
    //level = (abs (img->m7[i][j]) * LevelScale4x4Luma_Inter[qp_rem][i][j] + qp_const) >> q_bits;

    if (level != 0)
    {
      nonzero=TRUE;
      if (level > 1 || lossless_qpprime)
        *coeff_cost += MAX_VALUE;                // set high cost, shall not be discarded
      else
        *coeff_cost += COEFF_COST[input->disthres][run];
      ACLevel[scan_pos] = sign(level,img->m7[i][j]);
      ACRun  [scan_pos] = run;
      ++scan_pos;
      run=-1;                     // reset zero level counter

      level=sign(level, img->m7[i][j]);
      if(lossless_qpprime)
      {
        ilev=level;
      }
      else if(qp_per<4)
      {
        if(intra == 1)
          ilev=(level*InvLevelScale4x4Luma_Intra[qp_rem][i][j]+(1<<(3-qp_per)))>>(4-qp_per);
        else
          ilev=(level*InvLevelScale4x4Luma_Inter[qp_rem][i][j]+(1<<(3-qp_per)))>>(4-qp_per);
      }
      else
      {
        if(intra == 1)
          ilev=(level*InvLevelScale4x4Luma_Intra[qp_rem][i][j])<<(qp_per-4);
        else
          ilev=(level*InvLevelScale4x4Luma_Inter[qp_rem][i][j])<<(qp_per-4);
      }
    }
    if(!lossless_qpprime)
      img->m7[i][j]=ilev;
  }
  ACLevel[scan_pos] = 0;
  
  
  //     IDCT.
  //     horizontal
  for (j=0; j < BLOCK_SIZE && !lossless_qpprime; 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 && !lossless_qpprime; 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;
      // Residue Color Transform
      if (!img->residue_transform_flag)
      {
        img->m7[i][j] =min(img->max_imgpel_value,max(0,(m6[j]+m6[j1]+((long)img->mpr[i+block_x][j+block_y] <<DQ_BITS)+DQ_ROUND)>>DQ_BITS));
        img->m7[i][j1]=min(img->max_imgpel_value,max(0,(m6[j]-m6[j1]+((long)img->mpr[i+block_x][j1+block_y]<<DQ_BITS)+DQ_ROUND)>>DQ_BITS));
      } 
      else 
      {
        if(lossless_qpprime)
        {
          img->m7[i][j] =m6[j]+m6[j1];
          img->m7[i][j1]=m6[j]-m6[j1];
        }
        else
        {
          img->m7[i][j] =(m6[j]+m6[j1]+DQ_ROUND)>>DQ_BITS;
          img->m7[i][j1]=(m6[j]-m6[j1]+DQ_ROUND)>>DQ_BITS;
        }
      }
    }
  }
  
  //  Decoded block moved to frame memory
  if (!img->residue_transform_flag)
  {
    for (j=0; j < BLOCK_SIZE; j++)
    {
      for (i=0; i < BLOCK_SIZE; i++)
      {
        if(lossless_qpprime)
          enc_picture->imgY[img->pix_y+block_y+j][img->pix_x+block_x+i]=img->m7[i][j]+img->mpr[i+block_x][j+block_y];
        else
          enc_picture->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.
 *
 * \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(int uv,int cr_cbp)
{
  int i,j,i1,j2,ilev,n2,n1,j1,mb_y,coeff_ctr,level ,scan_pos,run;
  int m1[BLOCK_SIZE],m5[BLOCK_SIZE],m6[BLOCK_SIZE];
  int coeff_cost;
  int cr_cbp_tmp;
  int DCcoded=0 ;
  Macroblock *currMB = &img->mb_data[img->current_mb_nr];
 
  int qp_per,qp_rem,q_bits;
  int qp_c;

  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/2);

  //FRExt
  int64 cbpblk_pattern[4]={0, 0xf0000, 0xff0000, 0xffff0000};
  int yuv = img->yuv_format;
  int b8;
  int m3[4][4];
  int m4[4][4];
  int qp_per_dc = 0;
  int qp_rem_dc = 0;
  //int qp_const;
  int q_bits_422 = 0;	
  //int qp_const_422 = 0;
  Boolean lossless_qpprime = ((currMB->qp + img->bitdepth_luma_qp_scale)==0 && img->lossless_qpprime_flag==1);

  qp_c      = currMB->qp + img->chroma_qp_offset[uv];
  qp_c      = Clip3(-img->bitdepth_chroma_qp_scale,51,qp_c);
  qp_c      = (qp_c < 0)? qp_c : QP_SCALE_CR[qp_c - MIN_QP];

  qp_per    = (qp_c + img->bitdepth_chroma_qp_scale)/6;              
  qp_rem    = (qp_c + img->bitdepth_chroma_qp_scale)%6;              
  q_bits    = Q_BITS+qp_per;
  /*
  if (img->type == I_SLICE)
    qp_const=(1<<q_bits)/3;    // intra
  else
    qp_const=(1<<q_bits)/6;    // inter
*/
  if (img->yuv_format == YUV422)
  {
    //for YUV422 only
    qp_per_dc = (qp_c + 3 + img->bitdepth_chroma_qp_scale)/6;
    qp_rem_dc = (qp_c + 3 + img->bitdepth_chroma_qp_scale)%6;
    
    q_bits_422 = Q_BITS+qp_per_dc;
    /*
    if (img->type == I_SLICE)		
      qp_const_422=(1<<q_bits_422)/3;    // intra
    else
      qp_const_422=(1<<q_bits_422)/6;    // inter
      */
  }

  
  //============= dct transform ===============	
  for (n2=0; n2 < img->mb_cr_size_y; n2 += BLOCK_SIZE)
  {
    for (n1=0; n1 < img->mb_cr_size_x; n1 += BLOCK_SIZE)
    {

      //  Horizontal transform.
      for (j=0; j < BLOCK_SIZE && !lossless_qpprime; 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 && !lossless_qpprime; 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;
      }
    }
  }

  if (yuv == YUV420)
  {
    //================== CHROMA DC YUV420 ===================
    //     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;
      
      if(intra == 1)
        level =(abs(m1[coeff_ctr]) * LevelScale4x4Chroma_Intra[uv][qp_rem][0][0] + (LevelOffset4x4Chroma_Intra[uv][qp_per][0][0]<<1)) >> (q_bits+1);
        //level =(abs(m1[coeff_ctr]) * LevelScale4x4Chroma_Intra[uv][qp_rem][0][0] + (qp_const<<1)) >> (q_bits+1);
      else
        level =(abs(m1[coeff_ctr]) * LevelScale4x4Chroma_Inter[uv][qp_rem][0][0] + (LevelOffset4x4Chroma_Inter[uv][qp_per][0][0]<<1)) >> (q_bits+1);
        //level =(abs(m1[coeff_ctr]) * LevelScale4x4Chroma_Inter[uv][qp_rem][0][0] + (qp_const<<1)) >> (q_bits+1);
      
      if (input->symbol_mode == UVLC && img->qp < 4) 
      {
        if (level > CAVLC_LEVEL_LIMIT) 
        {
          level = CAVLC_LEVEL_LIMIT;
        }
      }
      
      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=sign(level, m1[coeff_ctr]);
      }
      m1[coeff_ctr]=ilev;
    }
    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]);
    for(i=0; i<4; i++)
    {
      if(qp_per<5)
      {
        if(intra == 1)
          m1[i]=(m5[i]*InvLevelScale4x4Chroma_Intra[uv][qp_rem][0][0])>>(5-qp_per);
        else
          m1[i]=(m5[i]*InvLevelScale4x4Chroma_Inter[uv][qp_rem][0][0])>>(5-qp_per);
      }
      else
      {
        if(intra == 1)
          m1[i]=(m5[i]*InvLevelScale4x4Chroma_Intra[uv][qp_rem][0][0])<<(qp_per-5);
        else
          m1[i]=(m5[i]*InvLevelScale4x4Chroma_Inter[uv][qp_rem][0][0])<<(qp_per-5);
      }
    }
    img->m7[0][0]=m1[0];
    img->m7[4][0]=m1[1];
    img->m7[0][4]=m1[2];
    img->m7[4][4]=m1[3];
  }
  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]= img->m7[i][j];
      
    //horizontal
    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