block.c

包含了从MPEG4的视频解码到H.264的视频编码部分的源代码

			paddd mm2,mm3
			paddd mm2,mm5
			psrad mm2,6
			PACKSSDW mm0,mm2
			PACKUSWB mm0,mm7
			PUNPCKLbw mm0,mm7
			
			
			movq [ecx],mm0			
			add eax,32
			add ebx,8
			add ecx,8			
			sub dx,1
			jne loop1_2	 
			//转秩
			mov ebx,pt2	
			mov edi,pt3
			movq mm0,[ebx]
			movq mm1,[ebx+8]
			movq mm2,[ebx+16]
			movq mm3,[ebx+24]
			movq mm4,mm0
			movq mm5,mm1
			movq mm6,mm2
			movq mm7,mm3

			PUNPCKLWD mm4,mm5
			PUNPCKLWD mm6,mm7
			movq mm5,mm4
			PUNPCKLDQ mm4,mm6
			PUNPCKHDQ mm5,mm6

			PUNPCKHWD mm0,mm1
			PUNPCKHWD mm2,mm7
			movq mm7,mm0
			PUNPCKLDQ mm0,mm2
			PUNPCKHDQ mm7,mm2
			movq mm6,mm0
			pxor mm0,mm0
			PACKUSWB mm4,mm0
			PACKUSWB mm5,mm0
			PACKUSWB mm6,mm0
			PACKUSWB mm7,mm0
			movd [edi],mm4
			add edi,imgcw
			movd [edi],mm5
			add edi,imgcw
			movd [edi],mm6
			add edi,imgcw
			movd [edi],mm7			
			emms
		}
	}
  }
      //     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;
  int qp_const,level,scan_pos,run;
  int nonzero;

  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];

  // 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, img->qp/3.0) * 4; 

  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;

  run=-1;
  scan_pos=0;
  
  for (coeff_ctr=0;coeff_ctr < 16;coeff_ctr++)     // 8 times if double scan, 16 normal scan
  {

    if (img->field_picture || ( mb_adaptive && img->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 = (abs (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 = img->m7[i][j]-sign(level1, predicted_block[i][j]);                   
    level1 = (abs (c_err1) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;
    
    // case 2
    c_err2=img->m7[i][j]-predicted_block[i][j];
    level2 = (abs (c_err2) * quant_coef[qp_rem][i][j] + qp_const) >> q_bits;
    
    // select prediction
    if ((level1 != level2) && (level1 != 0) && (level2 != 0))
    {
      D_dis1 = img->m7[i][j] - ((sign(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 = img->m7[i][j] - ((sign(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 = (abs(level1) < abs(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 += COEFF_COST[run];
      ACLevel[scan_pos] = sign(level,c_err);
      ACRun  [scan_pos] = run;
      ++scan_pos;
      run=-1;                     // reset zero level counter
      ilev=((sign(level,c_err)*dequant_coef[qp_rem][i][j]*A[i][j]<< qp_per) >>6);
    }
    ilev+=predicted_block[i][j] ; 
    img->m7[i][j] = sign((abs(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]=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]+DQ_ROUND)>>DQ_BITS));
      img->m7[i][j1]=min(255,max(0,(m6[j]-m6[j1]+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_sp(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[BLOCK_SIZE];
  int coeff_cost;
  int cr_cbp_tmp;
  int predicted_chroma_block[MB_BLOCK_SIZE/2][MB_BLOCK_SIZE/2],qp_const2,mp1[BLOCK_SIZE];
  Macroblock *currMB = &img->mb_data[img->current_mb_nr];

  int qp_per,qp_rem,q_bits;
  int qp_per_sp,qp_rem_sp,q_bits_sp;

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

  int c_err1, c_err2, level1, level2;
  int len, info;
  double D_dis1, D_dis2;
  double lambda_mode   = 0.85 * pow (2, img->qp/3.0) * 4; 

  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;
  qp_const=(1<<q_bits)/6;    // inter
  qp_per_sp    = ((img->qpsp<0?img->qpsp:QP_SCALE_CR[img->qpsp])-MIN_QP)/6;
  qp_rem_sp    = ((img->qpsp<0?img->qpsp:QP_SCALE_CR[img->qpsp])-MIN_QP)%6;
  q_bits_sp    = Q_BITS+qp_per_sp;
  qp_const2=(1<<q_bits_sp)/2;  //sp_pred

  for (j=0; j < MB_BLOCK_SIZE/2; j++)
    for (i=0; i < MB_BLOCK_SIZE/2; i++)
    {
      img->m7[i][j]+=img->mpr[i][j];
      predicted_chroma_block[i][j]=img->mpr[i][j];
    }

  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;
      }
    }
  }
  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]=predicted_chroma_block[i+n1][mb_y]+predicted_chroma_block[i1+n1][mb_y];
          m5[i1]=predicted_chroma_block[i+n1][mb_y]-predicted_chroma_block[i1+n1][mb_y];
        }
        predicted_chroma_block[n1][mb_y]  =(m5[0]+m5[1]);
        predicted_chroma_block[n1+2][mb_y]=(m5[0]-m5[1]);
        predicted_chroma_block[n1+1][mb_y]=m5[3]*2+m5[2];
        predicted_chroma_block[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]=predicted_chroma_block[j1][n2+j]+predicted_chroma_block[j1][n2+j2];
          m5[j2]=predicted_chroma_block[j1][n2+j]-predicted_chroma_block[j1][n2+j2];
        }
        predicted_chroma_block[j1][n2+0]=(m5[0]+m5[1]);
        predicted_chroma_block[j1][n2+2]=(m5[0]-m5[1]);
        predicted_chroma_block[j1][n2+1]=m5[3]*2+m5[2];
        predicted_chroma_block[j1][n2+3]=m5[3]-m5[2]*2;