mode_decision.c

H.264ITU-T 标准源码JM98 ITU推荐的免费解码器

        dataTr->part8x8fwref[block] = best_ref[LIST_0];
        dataTr->part8x8bwref[block] = best_ref[LIST_1];
        
#ifdef BEST_NZ_COEFF
        for(j = 0; j <= 1; j++)
        {
          for(i = 0; i <= 1; i++)
            best_nz_coeff[i][j]= cnt_nonz 
            ? img->nz_coeff[img->current_mb_nr][i1 + i][j1 + j] : 0;
        }
#endif
        
        //--- store number of nonzero coefficients ---
        best_cnt_nonz  = cnt_nonz;
        
        if (input->rdopt)
        {
          //--- store block cbp ---
          cbp_blk8x8    &= (~(0x33 << (((block>>1)<<3)+((block%2)<<1)))); // delete bits for block
          cbp_blk8x8    |= curr_cbp_blk;
          
          //--- store coefficients ---
          for (k=0; k< 4; k++)
          {
            for (j=0; j< 2; j++)
              for (i=0; i<65; i++)  
                cofACtr[k][j][i] = img->cofAC[block][k][j][i]; // 18->65 for ABT
          }   
          //--- store reconstruction and prediction --- 
          if(!img->residue_transform_flag)
          {
            for (j=j0; j<j0+8; j++)    
            {
              pix_y = img->pix_y + j;
              for (i=i0; i<i0+8; i++)
              {
                pix_x = img->pix_x + i;
                dataTr->rec_mbY8x8[j][i] = enc_picture->imgY[pix_y][pix_x];
                dataTr->mpr8x8[j][i] = img->mpr[j][i];
              }     
            }
          }
          else
          {
            for (j=j0; j<j0+8; j++)                
              for (i=i0; i<i0+8; i++)
              {
                dataTr->rec_resG_8x8  [j][i] = rec_resG  [j][i];
                dataTr->resTrans_R_8x8[j][i] = resTrans_R[j][i];
                dataTr->resTrans_B_8x8[j][i] = resTrans_B[j][i];
                dataTr->mprRGB_8x8 [0][j][i] = mprRGB [0][j][i];
                dataTr->mprRGB_8x8 [1][j][i] = mprRGB [1][j][i];
                dataTr->mprRGB_8x8 [2][j][i] = mprRGB [2][j][i];                  
              }     
          }
        }                  
        if (img->AdaptiveRounding)
        {
          for (j=j0; j<j0+8; j++)
            for (i=i0; i<i0+8; i++)                    
            {
              fadjust     [j][i] = fadjustTransform  [0]   [j][i];
              fadjustCr[0][j][i] = fadjustTransformCr[0][0][j][i];
              fadjustCr[1][j][i] = fadjustTransformCr[0][1][j][i];
            }
        }
        //--- store best 8x8 coding state ---
        if (block < 3)
        store_coding_state (cs_b8);
      } // if (rdcost <= min_rdcost)
      
      //--- re-set coding state as it was before coding with current mode was performed ---

      reset_coding_state (cs_cm);
    } // if ((enc_mb.valid[mode] && (transform8x8 == 0 || mode != 0 || (mode == 0 && active_sps->direct_8x8_inference_flag)))
  } // for (min_rdcost=1e30, index=(bslice?0:1); index<6; index++)  
  
#ifdef BEST_NZ_COEFF
  for(j = 0; j <= 1; j++)
  {
    for(i = 0; i <= 1; i++)
      img->nz_coeff[img->current_mb_nr][i1 + i][j1 + j] = best_nz_coeff[i][j];
  }
#endif              
  
  if (!transform8x8)
    dataTr->cost8x8 += min_cost8x8;
  
  if (!input->rdopt)
  {
    if (transform8x8)
    {
      dataTr->cost8x8 += min_cost8x8;              
      mode = dataTr->part8x8mode[block];
      pdir = dataTr->part8x8pdir[block];
    }
    else
    {
      mode = dataTr->part8x8mode[block];
      pdir = dataTr->part8x8pdir[block];    
    }
    curr_cbp_blk  = 0;
    best_cnt_nonz = LumaResidualCoding8x8 (&dummy, &curr_cbp_blk, block, pdir,
      (pdir==0||pdir==2?mode:0), (pdir==1||pdir==2?mode:0), dataTr->part8x8fwref[block], dataTr->part8x8bwref[block]);    
    
    cbp_blk8x8   &= (~(0x33 << (((block>>1)<<3)+((block%2)<<1)))); // delete bits for block
    cbp_blk8x8   |= curr_cbp_blk;
    
    //--- store coefficients ---
    for (k=0; k< 4; k++)
    {
        for (j=0; j< 2; j++)
          memcpy(cofACtr[k][j],img->cofAC[block][k][j],65 * sizeof(int));
    } 
    
    //--- store reconstruction and prediction ---
    if(!img->residue_transform_flag) // Residue Color Transform
    {
      for (j=j0; j<j0+2* BLOCK_SIZE; j++)
        {       
        memcpy(&dataTr->rec_mbY8x8[j][i0], &enc_picture->imgY[img->pix_y + j][img->pix_x + i0], 2* BLOCK_SIZE * sizeof (imgpel));
        memcpy(&dataTr->mpr8x8[j][i0], &img->mpr[j][i0], 2* BLOCK_SIZE * sizeof (imgpel));
      }
    }
    else 
    {   
      for (j=j0; j<j0+8; j++)
        for (i=i0; i<i0+8; i++)
        { 
          dataTr->rec_resG_8x8  [j][i] = rec_resG  [j][i];
          dataTr->resTrans_R_8x8[j][i] = resTrans_R[j][i];
          dataTr->resTrans_B_8x8[j][i] = resTrans_B[j][i];          
          dataTr->mprRGB_8x8 [0][j][i] = mprRGB [0][j][i];
          dataTr->mprRGB_8x8 [1][j][i] = mprRGB [1][j][i];
          dataTr->mprRGB_8x8 [2][j][i] = mprRGB [2][j][i];
        }
    }              
  }
  
  //----- set cbp and count of nonzero coefficients ---
  if (best_cnt_nonz)
  {
    cbp8x8       |= (1 << block);
    cnt_nonz_8x8 += best_cnt_nonz;
  }
  
  if (!transform8x8)
  {
    if (block<3)
    {
      //===== re-set reconstructed block =====
      j0   = 8*(block/2);
      i0   = 8*(block%2);
      for (j=j0; j<j0 + 2 * BLOCK_SIZE; j++)
        {
        memcpy(&enc_picture->imgY[img->pix_y + j][img->pix_x], dataTr->rec_mbY8x8[j], 2 * BLOCK_SIZE * sizeof(imgpel));
      }
    } // if (block<3)    
  }  
  else
  {  
    //======= save motion data for 8x8 partition for transform size 8x8 ========
    StoreNewMotionVectorsBlock8x8(0, block, dataTr->part8x8mode[block], dataTr->part8x8fwref[block], dataTr->part8x8bwref[block], dataTr->part8x8pdir[block], bslice);
  }  
  //===== set motion vectors and reference frames (prediction) =====
  SetRefAndMotionVectors (block, dataTr->part8x8mode[block], dataTr->part8x8pdir[block], dataTr->part8x8fwref[block], dataTr->part8x8bwref[block]);
  
  //===== set the coding state after current block =====
  //if (transform8x8 == 0 || block < 3)
  if (block < 3)
    reset_coding_state (cs_b8);
  
  if (img->AdaptiveRounding)
  {
    for (j=j0; j<j0+2 * BLOCK_SIZE; j++)
      {
        memcpy(&fadjustTransform  [lumaAdjustIndex  ]   [j][i0], &fadjust     [j][i0], 2 * BLOCK_SIZE * sizeof(int));
        memcpy(&fadjustTransformCr[chromaAdjustIndex][0][j][i0], &fadjustCr[0][j][i0], 2 * BLOCK_SIZE * sizeof(int));
        memcpy(&fadjustTransformCr[chromaAdjustIndex][1][j][i0], &fadjustCr[1][j][i0], 2 * BLOCK_SIZE * sizeof(int));
      }
  }
}

/*!
*************************************************************************************
* \brief
*    Mode Decision for a macroblock
*************************************************************************************
*/
void encode_one_macroblock ()
{  
  int max_index;
  
  int         rerun, block, index, mode, i, j, k, ctr16x16, dummy;
  short       best_pdir;
  RD_PARAMS   enc_mb;
  double      min_rdcost, max_rdcost=1e30;
  char        best_ref[2] = {0, -1};
  int         bmcost[5] = {INT_MAX};
  int         cost=0;
  int         min_cost = INT_MAX, cost_direct=0, have_direct=0, i16mode;
  int         intra1 = 0;
  int         temp_cpb = 0;
  int         best_transform_flag = 0;
  int         cost8x8_direct = 0;  
  short       islice      = (img->type==I_SLICE);
  short       bslice      = (img->type==B_SLICE);
  short       pslice      = (img->type==P_SLICE) || (img->type==SP_SLICE);
  short       intra       = (islice || (pslice && img->mb_y==img->mb_y_upd && img->mb_y_upd!=img->mb_y_intra));
  
  short       runs        = (input->RestrictRef==1 && input->rdopt==3 && (pslice  || (bslice && img->nal_reference_idc>0)) ? 2 : 1);
  
  int         pix_x, pix_y;
  Macroblock* currMB      = &img->mb_data[img->current_mb_nr];
  Macroblock* prevMB      = img->current_mb_nr ? &img->mb_data[img->current_mb_nr-1]:NULL ;
  
  char   **ipredmodes = img->ipredmode;
  short   *allmvs = img->all_mv[0][0][0][0][0];
  short   max_chroma_pred_mode;
  int     ****i4p;  //for non-RD-opt. mode
  
#ifdef BEST_NZ_COEFF
  int best_nz_coeff[2][2];
#endif  
  
  int tmp_8x8_flag, tmp_no_mbpart;  
  // Residue Color Transform
  int residue_R, residue_G, residue_B, temp;
  int cr_cbp = 0;  
  // Fast Mode Decision
  short inter_skip = 0, intra_skip = 0;
  int cost16 = 0, mode16 = 0;
  double min_rate = 0, RDCost16 = DBL_MAX;
  
  
  if(input->FMEnable)
    decide_intrabk_SAD();
  
  intra |= RandomIntra (img->current_mb_nr);    // Forced Pseudo-Random Intra
  
  //===== Setup Macroblock encoding parameters =====
  init_enc_mb_params(currMB, &enc_mb, intra, bslice);
  
  // Perform multiple encodings if rdopt with losses is enabled
  for (rerun=0; rerun<runs; rerun++)
  {
    if (runs==2)
      input->rdopt= (rerun==0) ? 1 : 3;
    
    // reset chroma intra predictor to default
    currMB->c_ipred_mode = DC_PRED_8;

    //=====   S T O R E   C O D I N G   S T A T E   =====
    //---------------------------------------------------
    store_coding_state (cs_cm);
    
    if (!intra)
    {
      //===== set direct motion vectors =====
      best_mode = 1;
      if (bslice)
      {
        Get_Direct_Motion_Vectors ();
        if (input->rdopt == 2 && enc_mb.valid[0])
        {
          best_mode = 0;
          currMB->c_ipred_mode=DC_PRED_8;
          min_rdcost = max_rdcost;
          compute_mode_RD_cost(0, currMB, enc_mb, &min_rdcost, &min_rate, i16mode, bslice, &inter_skip);
        }
      }
      
      //===== MOTION ESTIMATION FOR 16x16, 16x8, 8x16 BLOCKS =====      
      for (min_cost=INT_MAX, mode=1; mode<4; mode++)
      {
        bi_pred_me = 0;
        img->bi_pred_me[mode]=0;
        if (enc_mb.valid[mode] && !inter_skip)
        {
          for (cost=0, block=0; block<(mode==1?1:2); block++)
          {
            PartitionMotionSearch (mode, block, enc_mb.lambda_mf);
            //--- set 4x4 block indizes (for getting MV) ---
            j = (block==1 && mode==2 ? 2 : 0);
            i = (block==1 && mode==3 ? 2 : 0);
            
            //--- get cost and reference frame for List 0 prediction ---
            bmcost[LIST_0] = INT_MAX;
            list_prediction_cost(LIST_0, block, mode, enc_mb, bmcost, best_ref);
            
            if (bslice)
            {
              //--- get cost and reference frame for List 1 prediction ---
              bmcost[LIST_1] = INT_MAX;
              list_prediction_cost(LIST_1, block, mode, enc_mb, bmcost, best_ref);
              
              // Compute bipredictive cost between best list 0 and best list 1 references
              list_prediction_cost(BI_PRED, block, mode, enc_mb, bmcost, best_ref);
              
              // Finally, if mode 16x16, compute cost for bipredictive ME vectore
              if (input->BiPredMotionEstimation && mode == 1)
              {                
                list_prediction_cost(BI_PRED_L0, block, mode, enc_mb, bmcost, 0);
                list_prediction_cost(BI_PRED_L1, block, mode, enc_mb, bmcost, 0);
              }
              else
              {
                bmcost[BI_PRED_L0] = INT_MAX;
                bmcost[BI_PRED_L1] = INT_MAX;
              }
              
              // Determine prediction list based on mode cost
              determine_prediction_list(mode, bmcost, best_ref, &best_pdir, &cost, &bi_pred_me);
            }
            else // if (bslice)
            {
              best_pdir  = 0;
              cost      += bmcost[LIST_0];
            }
            
            assign_enc_picture_params(mode, best_pdir, block, enc_mb.list_offset[LIST_0], best_ref[LIST_0], best_ref[LIST_1], bslice);
            
            //----- set reference frame and direction parameters -----
            if (mode==3)
            {
              best8x8fwref [3][block  ] = best8x8fwref [3][  block+2] = best_ref[LIST_0];
              best8x8pdir  [3][block  ] = best8x8pdir  [3][  block+2] = best_pdir;
              best8x8bwref [3][block  ] = best8x8bwref [3][  block+2] = best_ref[LIST_1];
            }
            else if (mode==2)
            {
              best8x8fwref [2][2*block] = best8x8fwref [2][2*block+1] = best_ref[LIST_0];
              best8x8pdir  [2][2*block] = best8x8pdir  [2][2*block+1] = best_pdir;
              best8x8bwref [2][2*block] = best8x8bwref [2][2*block+1] = best_ref[LIST_1];
            }
            else
            {
              best8x8fwref [1][0] = best8x8fwref [1][1] = best8x8fwref [1][2] = best8x8fwref [1][3] = best_ref[LIST_0];
              best8x8pdir  [1][0] = best8x8pdir  [1][1] = best8x8pdir  [1][2] = best8x8pdir  [1][3] = best_pdir;
              best8x8bwref [1][0] = best8x8bwref [1][1] = best8x8bwref [1][2] = best8x8bwref [1][3] = best_ref[LIST_1];
            }
            
            //--- set reference frames and motion vectors ---
            if (mode>1 && block==0)
              SetRefAndMotionVectors (block, mode, best_pdir, best_ref[LIST_0], best_ref[LIST_1]);            
          } // for (block=0; block<(mode==1?1:2); block++)
          
          if(!input->rdopt)
          {
            currMB->luma_transform_size_8x8_flag = 0;
            if (input->Transform8x8Mode) //for inter rd-off, set 8x8 to do 8x8 transform
            {
              SetModesAndRefframeForBlocks(mode);
              currMB->luma_transform_size_8x8_flag = TransformDecision(-1, &cost);
            }
          }