rdopt.c

H264视频编解码程序

  int     dummy, x, y, rate;
  int64   distortion  = 0;
  int     block_x     = 8*(b8 & 0x01)+4*(b4 & 0x01);
  int     block_y     = 8*(b8 >> 1)+4*(b4 >> 1);
  int     pic_pix_x   = img->pix_x+block_x;
  int     pic_pix_y   = img->pix_y+block_y;
  int     pic_opix_y  = img->opix_y+block_y;
  imgpel  **imgY      = enc_picture->imgY;
  
  Slice          *currSlice    =  img->currentSlice;
  Macroblock     *currMB       = &img->mb_data[img->current_mb_nr];
  SyntaxElement  *currSE       = &img->MB_SyntaxElements[currMB->currSEnr];
  const int      *partMap      = assignSE2partition[input->partition_mode];
  DataPartition  *dataPart;
  
  //===== perform DCT, Q, IQ, IDCT, Reconstruction =====
  dummy = 0;
  
  if(img->type!=SP_SLICE)
    *nonzero = dct_luma (block_x, block_y, &dummy, 1);
  else if(!si_frame_indicator && !sp2_frame_indicator)
  {
    *nonzero = dct_luma_sp(block_x, block_y, &dummy);
  }
  else
  {   
    *nonzero = dct_luma_sp2(block_x, block_y, &dummy);
  }

  //===== get distortion (SSD) of 4x4 block =====
  if(!img->residue_transform_flag)
  {
    for (y=0; y<4; y++)
    {
      for (x=pic_pix_x; x<pic_pix_x+4; x++)
      {
        distortion += img->quad [imgY_org[pic_opix_y+y][x] - imgY[pic_pix_y+y][x]];
      }
    }
  }
  
  //===== RATE for INTRA PREDICTION MODE  (SYMBOL MODE MUST BE SET TO UVLC) =====
  currSE->value1 = (mostProbableMode == ipmode) ? -1 : ipmode < mostProbableMode ? ipmode : ipmode-1;
  
  //--- set position and type ---
  currSE->context = 4*b8 + b4;
  currSE->type    = SE_INTRAPREDMODE;
  
  //--- set function pointer ----
  if (input->symbol_mode != UVLC)    
    currSE->writing = writeIntraPredMode_CABAC;
  
  //--- choose data partition ---
  dataPart = &(currSlice->partArr[partMap[SE_INTRAPREDMODE]]);
  //--- encode and update rate ---
  if (input->symbol_mode == UVLC)    writeSyntaxElement_Intra4x4PredictionMode(currSE, dataPart);
  else                               dataPart->writeSyntaxElement (currSE, dataPart);
  
  rate = currSE->len;
  currSE++;
  currMB->currSEnr++;
  
  //===== RATE for LUMINANCE COEFFICIENTS =====
  if (input->symbol_mode == UVLC)
  {
    rate  += writeCoeff4x4_CAVLC (LUMA, b8, b4, 0);
  }
  else
  {
    rate  += writeLumaCoeff4x4_CABAC (b8, b4, 1);
  }
  //reset_coding_state (cs_cm);
  rdcost = (double)distortion + lambda*(double)rate;
  
  if(img->residue_transform_flag)
    return (double)rate;
  else
    return rdcost;
}


// Residue Color Transform
int RDCost_for_4x4Blocks_Chroma (int     b8,
                                 int     b4,
                                 int  chroma)
{
  int     rate=0;
  
  Slice          *currSlice    =  img->currentSlice;
  Macroblock     *currMB       = &img->mb_data[img->current_mb_nr];
  SyntaxElement  *currSE       = &img->MB_SyntaxElements[currMB->currSEnr];
  const int      *partMap      = assignSE2partition[input->partition_mode];
  int uv;
  
  //===== perform DCT, Q, IQ, IDCT, Reconstruction =====
  if(b8 > 7) 
    uv = 1;
  else 
    uv = 0;
  
  cbp_chroma_block_temp[uv][2*((b8-4*(uv+1)) & 0x01)+(b4 & 0x01)][2*((b8-4*(uv+1)) >> 1)+(b4 >> 1)] = dct_chroma4x4 (chroma, b8, b4);
  
  //===== RATE for LUMINANCE COEFFICIENTS =====
  if (input->symbol_mode == UVLC)
  {
    rate  = writeCoeff4x4_CAVLC (CHROMA_AC, b8, b4, ((2*(b8 & 0x01)+ (b4 & 0x01))<<4) | (2*(b8 >> 1)+(b4 >> 1)));
  }
  else
  {
    int * ACLevel, * ACRun;
    int level, run, k;
    DataPartition*  dataPart;
    int*            bitCount  = currMB->bitcounter;
    ACLevel = img->cofAC[b8][b4][0];
    ACRun   = img->cofAC[b8][b4][1];
    
    level=1;
    
    img->subblock_y = b4 >> 1;
    img->subblock_x = b4 & 0x01;
    
    for (k=0; k < 17 && level != 0; k++)
    {
      level = currSE->value1 = ACLevel[k]; // level
      run   = currSE->value2 = ACRun  [k]; // run
      
      if (input->symbol_mode == UVLC)   
        currSE->mapping = levrun_linfo_inter;
      else                              
        currSE->writing = writeRunLevel_CABAC;
      
      currSE->context     = CHROMA_AC;
      currSE->type        = SE_CHR_AC_INTRA;
      
      img->is_intra_block =  IS_INTRA(currMB);
      img->is_v_block     = uv;
      
      // choose the appropriate data partition
      dataPart = &(currSlice->partArr[partMap[currSE->type]]); 
      dataPart->writeSyntaxElement (currSE, dataPart);
      bitCount[BITS_COEFF_UV_MB] += currSE->len;
      rate                       += currSE->len;
      
      // proceed to next SE
      currSE++;
      currMB->currSEnr++;
    }
  }
  reset_coding_state (cs_cm);
  
  return rate;
}


/*!
 *************************************************************************************
 * \brief
 *    Mode Decision for an 4x4 Intra block
 *************************************************************************************
 */
int Mode_Decision_for_4x4IntraBlocks (int  b8,  int  b4,  double  lambda,  int*  min_cost)
{
  int     ipmode, best_ipmode = 0, i, j, k, x, y, cost, dummy;
  int     c_nz, nonzero = 0, diff[16];
  imgpel  rec4x4[4][4];
  double  rdcost;
  int     block_x     = 8*(b8 & 0x01)+4*(b4 & 0x01);
  int     block_y     = 8*(b8 >> 1)+4*(b4 >> 1);
  int     block_x4    = block_x >> 2;
  int     block_y4    = block_y >> 2;
  int     pic_pix_x   = img->pix_x+block_x;
  int     pic_pix_y   = img->pix_y+block_y;
  int     pic_opix_x   = img->opix_x+block_x;
  int     pic_opix_y   = img->opix_y+block_y;
  int     pic_block_x = pic_pix_x >> 2;
  int     pic_block_y = pic_pix_y >> 2;
  double  min_rdcost  = 1e30;
  
  int left_available, up_available, all_available;
  
  char   upMode;
  char   leftMode;
  int     mostProbableMode;
  
  PixelPos left_block;
  PixelPos top_block;
  
  int     lrec4x4[4][4]; 
  int     lrec4x4_c[2][4][4];
  // Residue Color Transform
  int residue_R, residue_G, residue_B;
  int rate, temp;
  int64 distortion;
  int c_ipmode = img->mb_data[img->current_mb_nr].c_ipred_mode;
  int fixedcost = (int) floor(4 * lambda );
  imgpel rec4x4_c[2][4][4];

#ifdef BEST_NZ_COEFF
  int best_nz_coeff = 0;
  int best_coded_block_flag = 0;
  int bit_pos = 1 + ((((b8>>1)<<1)+(b4>>1))<<2) + (((b8&1)<<1)+(b4&1));
  static int64 cbp_bits;

  if (b8==0 && b4==0)
   cbp_bits = 0;
#endif
  
  getLuma4x4Neighbour(img->current_mb_nr, block_x4, block_y4, -1,  0, &left_block);
  getLuma4x4Neighbour(img->current_mb_nr, block_x4, block_y4,  0, -1, &top_block);
  
  // constrained intra pred
  if (input->UseConstrainedIntraPred)
  {
    left_block.available = left_block.available ? img->intra_block[left_block.mb_addr] : 0;
    top_block.available  = top_block.available  ? img->intra_block[top_block.mb_addr]  : 0;
  }
  
  upMode            =  top_block.available ? img->ipredmode[top_block.pos_y ][top_block.pos_x ] : -1;
  leftMode          = left_block.available ? img->ipredmode[left_block.pos_y][left_block.pos_x] : -1;
  
  mostProbableMode  = (upMode < 0 || leftMode < 0) ? DC_PRED : upMode < leftMode ? upMode : leftMode;
  
  *min_cost = INT_MAX;
  
  //===== INTRA PREDICTION FOR 4x4 BLOCK =====
  intrapred_luma (pic_pix_x, pic_pix_y, &left_available, &up_available, &all_available);
  
  //===== LOOP OVER ALL 4x4 INTRA PREDICTION MODES =====
  for (ipmode=0; ipmode<NO_INTRA_PMODE; ipmode++)
  {
    int available_mode =  (ipmode==DC_PRED) ||
      ((ipmode==VERT_PRED||ipmode==VERT_LEFT_PRED||ipmode==DIAG_DOWN_LEFT_PRED) && up_available ) ||
      ((ipmode==HOR_PRED||ipmode==HOR_UP_PRED) && left_available ) ||(all_available);
    
    if (input->IntraDisableInterOnly==0 || img->type != I_SLICE)
    {
      if (input->Intra4x4ParDisable && (ipmode==VERT_PRED||ipmode==HOR_PRED))
        continue;      
      
      if (input->Intra4x4DiagDisable && (ipmode==DIAG_DOWN_LEFT_PRED||ipmode==DIAG_DOWN_RIGHT_PRED))
        continue;      
      
      if (input->Intra4x4DirDisable && ipmode>=VERT_RIGHT_PRED)
        continue;
    }
    
    if( available_mode)
    {
      if (!input->rdopt)
      {
        for (k=j=0; j<4; j++)
        {
          int jj = pic_opix_y+j;
          for (i=0; i<4; i++, k++)
          {
            diff[k] = imgY_org[jj][pic_opix_x+i] - img->mprr[ipmode][j][i];
          }
        }
        //cost  = (ipmode == mostProbableMode) ? 0 : (int)floor(4 * lambda );
        cost  = (ipmode == mostProbableMode) ? 0 : fixedcost;
        cost += SATD (diff, input->hadamard);
        if (cost < *min_cost)
        {
          best_ipmode = ipmode;
          *min_cost   = cost;
        }
      }
      else
      {
        // Residue Color Transform
        if(!img->residue_transform_flag)
        {
          // get prediction and prediction error
          for (j=0; j<4; j++)
          {
            memcpy(&img->mpr[block_y+j][block_x], img->mprr[ipmode][j], BLOCK_SIZE * sizeof(imgpel));
            for (i=0; i<4; i++)
            {
              img->m7[j][i] = (int) (imgY_org[pic_opix_y+j][pic_opix_x+i] - img->mprr[ipmode][j][i]);
            }
          }
          
          //===== store the coding state =====
          //store_coding_state (cs_cm);
          // get and check rate-distortion cost
#ifdef BEST_NZ_COEFF
          img->mb_data[img->current_mb_nr].cbp_bits = cbp_bits;
#endif
          if ((rdcost = RDCost_for_4x4IntraBlocks (&c_nz, b8, b4, ipmode, lambda, min_rdcost, mostProbableMode)) < min_rdcost)
          {
            //--- set coefficients ---
            memcpy(cofAC4x4[0],img->cofAC[b8][b4][0], 18 * sizeof(int));
            memcpy(cofAC4x4[1],img->cofAC[b8][b4][1], 18 * sizeof(int));

            //--- set reconstruction ---
            for (y=0; y<4; y++)
            {
              memcpy(rec4x4[y],&enc_picture->imgY[pic_pix_y+y][pic_pix_x], BLOCK_SIZE * sizeof(imgpel));
              if(img->type==SP_SLICE &&(!si_frame_indicator && !sp2_frame_indicator))
                memcpy(lrec4x4[y],&lrec[pic_pix_y+y][pic_pix_x], BLOCK_SIZE * sizeof(int));// stores the mode coefficients
            } 
            //--- flag if dct-coefficients must be coded ---
            nonzero = c_nz;
            
            //--- set best mode update minimum cost ---
            min_rdcost    = rdcost;
            best_ipmode   = ipmode;
#ifdef BEST_NZ_COEFF
            best_nz_coeff = img->nz_coeff [img->current_mb_nr][block_x4][block_y4];
            best_coded_block_flag = (int)((img->mb_data[img->current_mb_nr].cbp_bits>>bit_pos)&(int64)(1));
#endif            
            //store_coding_state (cs_ib4);
            if (img->AdaptiveRounding)
            {
              for (j=0; j<4; j++)
                memcpy(&fadjust4x4[block_y+j][block_x],&img->fadjust4x4[1][block_y+j][block_x], BLOCK_SIZE * sizeof(int));
            }
          }

#ifndef RESET_STATE
          reset_coding_state (cs_cm);
#endif
        }
        else 
        {
          for (j=0; j<4; j++)
          {
            for (i=0; i<4; i++)
            {
              residue_B = imgUV_org[0][pic_opix_y+j][pic_opix_x+i] - img->mprr_c[0][c_ipmode][block_y+j][block_x+i];
              residue_G = imgY_org[pic_opix_y+j][pic_opix_x+i] - img->mprr[ipmode][j][i];
              residue_R = imgUV_org[1][pic_opix_y+j][pic_opix_x+i] - img->mprr_c[1][c_ipmode][block_y+j][block_x+i];
              
              /* Foward Residue Transform */
              resTrans_R[j][i] = residue_R-residue_B;
              temp = residue_B+(resTrans_R[j][i]>>1);
              resTrans_B[j][i] = residue_G-temp;
              resTrans_G[j][i] = temp+(resTrans_B[j][i]>>1);
            }
          }
          
          for (j=0; j<4; j++)
          {
            for (i=0; i<4; i++)
            {
              img->m7[j][i]  = resTrans_G[j][i];
            }
          }
          
          store_coding_state (cs_cm);
          // yuwen 2005.11.17 => should be an error here, requiring process for img->mb_data[img->current_mb_nr].cbp_bits
          rate = (int) RDCost_for_4x4IntraBlocks (&c_nz, b8, b4, ipmode, lambda, min_rdcost, mostProbableMode);
          reset_coding_state (cs_cm);
          
          for (j=0; j<4; j++)
          {
            for (i=0; i<4; i++)
            {
              rec_resG[j][i] = img->m7[j][i];
              img->m7[j][i]  = resTrans_B[j][i];
            }
          }
          //store_coding_state (cs_cm);
          // yuwen 2005.11.17 => should be an error here, requiring process for img->mb_data[img->current_mb_nr].cbp_bits
          rate += RDCost_for_4x4Blocks_Chroma (b8+4, b4, 0);
          for (j=0; j<4; j++)
          {
            for (i=0; i<4; i++)
            {
              rec_resB[j][i] = img->m7[j][i];
              img->m7[j][i]  = resTrans_R[j][i];
            }
          }
          // yuwen 2005.11.17 => should be an error here, requiring process for img->mb_data[img->current_mb_nr].cbp_bits
          rate += RDCost_for_4x4Blocks_Chroma (b8+8, b4, 1);
          
          reset_coding_state (cs_cm);
          for (j=0; j<4; j++)
          {
            for (i=0; i<4; i++)
            {
              rec_resR[j][i] = img->m7[j][i];
            }
          }
          
          for (j=0; j<4; j++)
          {
            for (i=0; i<4; i++)
            {
              /* Inverse Residue Transform */
              temp      = rec_resG[j][i]-(rec_resB[j][i]>>1);
              residue_G = rec_resB[j][i]+temp;