mode_decision.c

来自「the newest JM software by h.264 JVT offi」· C语言 代码 · 共 1,390 行 · 第 1/4 页

        img->nz_coeff[currMB->mb_nr][i1 + i][j1 + j] = best_nz_coeff[i][j];
    }
#endif

    if (!transform8x8)
    {
      if (min_cost8x8 != INT_MAX)
        dataTr->mb_p8x8_cost += min_cost8x8;
      else
        dataTr->mb_p8x8_cost = INT_MAX;
    }
    
    //----- 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 >> 1);
        i0   = 8*(block & 0x01);
        for (j = j0; j < j0 + BLOCK_SIZE_8x8; j++)
        {
          memcpy(&enc_picture->imgY[img->pix_y + j][img->pix_x], dataTr->rec_mbY8x8[j], BLOCK_SIZE_8x8 * sizeof(imgpel));
        }

        if (params->rdopt == 3)
        {
          errdo_get_best_block(img, enc_picture->p_dec_img[0], decs->dec_mbY_best8x8[transform8x8], j0, BLOCK_SIZE_8x8);
        }

        if(img->type==SP_SLICE &&(!si_frame_indicator))
        {
          for (j = j0; j < j0 + BLOCK_SIZE_8x8; j++)
          {
            memcpy(&lrec[img->pix_y + j][img->pix_x], dataTr->lrec[j],2*BLOCK_SIZE*sizeof(imgpel)); // reset the coefficients for SP slice
          }
        }

        if(img->P444_joined) 
        {
          for (k=0; k<2; k++)
          {
            for (j = j0; j < j0 + BLOCK_SIZE_8x8; j++)
            {
              memcpy(&enc_picture->imgUV[k][img->pix_y + j][img->pix_x], dataTr->rec_mb8x8_cr[k][j], BLOCK_SIZE_8x8 * sizeof(imgpel));
            }
          }
        }
      } // if (block<3)
    }
    else
    {
      //======= save motion data for 8x8 partition for transform size 8x8 ========    
      StoreNewMotionVectorsBlock8x8(0, block, dataTr->part8x8mode[block], dataTr->part8x8l0ref[block], dataTr->part8x8l1ref[block], dataTr->part8x8pdir[block], dataTr->part8x8bipred[block], bslice);
    }

    //===== set motion vectors and reference frames (prediction) =====
    SetRefAndMotionVectors (&enc_picture->motion, currMB, block, dataTr->part8x8mode[block], dataTr->part8x8pdir[block], dataTr->part8x8l0ref[block], dataTr->part8x8l1ref[block], dataTr->part8x8bipred[block]);

    //===== set the coding state after current block =====
    //if (transform8x8 == 0 || block < 3)
    if (stored_state_8x8 == TRUE)
      reset_coding_state (currSlice, currMB, cs_b8);
    else
    {
      update_adaptive_rounding_8x8(dataTr, fadjust);
    }
  }
}

/*!
*************************************************************************************
* \brief
*    Low Complexity Mode Decision for an 8x8 sub-macroblock
*************************************************************************************
*/
void submacroblock_mode_decision_low(Slice *currSlice,
                                     RD_PARAMS *enc_mb,
                                     RD_8x8DATA *dataTr,
                                     Macroblock *currMB,
                                     int ****cofACtr,
                                     int *have_direct,
                                     short bslice,
                                     int block,
                                     int *cost_direct,
                                     int *cost,
                                     int *cost8x8_direct,
                                     int transform8x8,
                                     int is_cavlc)
{
  int64 curr_cbp_blk;
  double min_rdcost, rdcost = 0.0;
  int j0, i0, j1, i1;
  int i,j, k;
  int min_cost8x8, index;
  int mode;
  int direct4x4_tmp, direct8x8_tmp;
  int bmcost[5] = {INT_MAX};
  int cnt_nonz = 0;
  int dummy;
  int best_cnt_nonz = 0;
  int maxindex =  (transform8x8) ? 2 : 5;
  int block_x, block_y;
  int lambda_mf[3];

  int ****fadjust = transform8x8? img->ARCofAdj8x8 : img->ARCofAdj4x4;
  short pdir, bipred_me = 0;

  char best_pdir = 0;
  char best_ref[2] = {0, -1};
  imgpel  **mb_pred = img->mb_pred[0];
  Boolean valid_8x8 = FALSE;
  Boolean stored_state_8x8 = FALSE;

#ifdef BEST_NZ_COEFF
  int best_nz_coeff[2][2];
#endif

  //--- set coordinates ---
  j0 = ((block>>1)<<3);
  j1 = (j0>>2);
  i0 = ((block&0x01)<<3);
  i1 = (i0>>2);

#ifdef BEST_NZ_COEFF
  for(j = 0; j <= 1; j++)
  {
    for(i = 0; i <= 1; i++)
      best_nz_coeff[i][j] = img->nz_coeff[currMB->mb_nr][i1 + i][j1 + j] = 0;
  }
#endif

  if (transform8x8)
    currMB->luma_transform_size_8x8_flag = 1; //switch to transform size 8x8

  //--- store coding state before coding ---
  store_coding_state (currSlice, currMB, cs_cm);

  //=====  LOOP OVER POSSIBLE CODING MODES FOR 8x8 SUB-PARTITION  =====
  for (min_cost8x8 = INT_MAX, min_rdcost = 1e20, index = (bslice?0:1); index < maxindex; index++)
  {
    mode = b8_mode_table[index];
    *cost = 0;

    if (enc_mb->valid[mode] && (!(transform8x8 == 1 && mode > 4)) && (transform8x8 == 0 || mode != 0 || (mode == 0 && active_sps->direct_8x8_inference_flag)))
    {
      if (transform8x8)
      {
        currMB->valid_8x8 = TRUE;
      }

      valid_8x8 = TRUE;
      curr_cbp_blk = 0;

      if (mode==0)
      {
        //--- Direct Mode ---       
        direct4x4_tmp = 0;
        direct8x8_tmp = 0;
        direct4x4_tmp = GetDirectCost8x8 ( currMB, block, &direct8x8_tmp);

        if ((direct4x4_tmp==INT_MAX)||(*cost_direct==INT_MAX))
        {
          *cost_direct = INT_MAX;
          if (transform8x8)
            *cost8x8_direct = INT_MAX;
        }
        else
        {
          *cost_direct += direct4x4_tmp;
          if (transform8x8)
            *cost8x8_direct += direct8x8_tmp;
        }
        (*have_direct) ++;

        if (transform8x8)
        {
          switch(params->Transform8x8Mode)
          {
          case 1: // Mixture of 8x8 & 4x4 transform
            if((direct8x8_tmp < direct4x4_tmp) || !(enc_mb->valid[5] && enc_mb->valid[6] && enc_mb->valid[7]))
              *cost = direct8x8_tmp;
            else
              *cost = direct4x4_tmp;
            break;
          case 2: // 8x8 Transform only
            *cost = direct8x8_tmp;
            break;
          default: // 4x4 Transform only
            *cost = direct4x4_tmp;
            break;
          }
          if (params->Transform8x8Mode==2)
            *cost = INT_MAX;
        }
        else
        {
          *cost = direct4x4_tmp;
        }

        block_x = img->block_x + (block & 0x01)*2;
        block_y = img->block_y + (block & 0x02);
        best_ref[LIST_0] = direct_ref_idx[LIST_0][block_y][block_x];
        best_ref[LIST_1] = direct_ref_idx[LIST_1][block_y][block_x];
        best_pdir        = direct_pdir[block_y][block_x];
      } // if (mode==0)
      else
      {
        int64 ref_pic_num;
        char b_ref;

        //======= motion estimation for all reference frames ========
        //-----------------------------------------------------------
        memcpy(lambda_mf, enc_mb->lambda_mf, 3 * sizeof(int));
        if (params->CtxAdptLagrangeMult == 1)
        {
          lambda_mf[F_PEL] = (int)(lambda_mf[F_PEL] * lambda_mf_factor);
          lambda_mf[H_PEL] = (int)(lambda_mf[H_PEL] * lambda_mf_factor);
          lambda_mf[Q_PEL] = (int)(lambda_mf[Q_PEL] * lambda_mf_factor);
        }

        PartitionMotionSearch (currMB, mode, block, lambda_mf);

        //--- get cost and reference frame for LIST 0 prediction ---
        bmcost[LIST_0] = INT_MAX;
        list_prediction_cost(currMB, LIST_0, block, mode, enc_mb, bmcost, best_ref);

        //store LIST 0 reference index for every block
        block_x = img->block_x + (block & 0x01)*2;
        block_y = img->block_y + (block & 0x02);
        b_ref = best_ref[LIST_0];
        ref_pic_num = enc_picture->ref_pic_num[enc_mb->list_offset[LIST_0]][(short) b_ref];

        for (j = block_y; j< block_y + 2; j++)
        {
          memset(&enc_picture->motion.ref_idx [LIST_0][j][block_x], b_ref, 2 * sizeof(char));
        }

        for (j = block_y; j< block_y + 2; j++)
        {
          for (i = block_x; i < block_x + 2; i++)
          {
            enc_picture->motion.ref_pic_id[LIST_0][j][i] = ref_pic_num;
          }
        }

        if (bslice)
        {
          //--- get cost and reference frame for LIST 1 prediction ---
          bmcost[LIST_1] = INT_MAX;
          bmcost[BI_PRED] = INT_MAX;
          list_prediction_cost(currMB, LIST_1, block, mode, enc_mb, bmcost, best_ref);

          // Compute bipredictive cost between best list 0 and best list 1 references
          list_prediction_cost(currMB, BI_PRED, block, mode, enc_mb, bmcost, best_ref);

          // currently Bi prediction ME is only supported for modes 1, 2, 3 and only for ref 0 and only for ref 0
          if (is_bipred_enabled(mode))
          {
            list_prediction_cost(currMB, BI_PRED_L0, block, mode, enc_mb, bmcost, 0);
            list_prediction_cost(currMB, BI_PRED_L1, block, mode, enc_mb, bmcost, 0);
          }
          else
          {
            bmcost[BI_PRED_L0] = INT_MAX;
            bmcost[BI_PRED_L1] = INT_MAX;
          }

          //--- get prediction direction ----
          determine_prediction_list(mode, bmcost, best_ref, &best_pdir, cost, &bipred_me);

          //store backward reference index for every block
          for (k = LIST_0; k <= LIST_1; k++)
          {
            for (j = block_y; j< block_y + 2; j++)
            {
              memset(&enc_picture->motion.ref_idx[k][j][block_x], best_ref[k], 2 * sizeof(char));
            }
          }
        } // if (bslice)
        else
        {
          best_pdir = 0;
          *cost     = bmcost[LIST_0];
        }
      } // if (mode!=0)

      if (*cost!=INT_MAX)
        *cost += (REF_COST (enc_mb->lambda_mf[Q_PEL], B8Mode2Value (mode, best_pdir),
        enc_mb->list_offset[(best_pdir < 1 ? LIST_0 : LIST_1)]) - 1);

      //--- set variables if best mode has changed ---
      if (*cost < min_cost8x8)
      {
        min_cost8x8                  = *cost;
        min_rdcost                   = rdcost;
        dataTr->part8x8mode  [block] = mode;
        dataTr->part8x8pdir  [block] = best_pdir;
        dataTr->part8x8l0ref [block] = best_ref[LIST_0];
        dataTr->part8x8l1ref [block] = best_ref[LIST_1];
        dataTr->part8x8bipred[block] = bipred_me;

        currMB->b8mode[block] = mode;

#ifdef BEST_NZ_COEFF
        if (cnt_nonz)
        {
          best_nz_coeff[0][0]= img->nz_coeff[currMB->mb_nr][i1    ][j1    ];
          best_nz_coeff[0][1]= img->nz_coeff[currMB->mb_nr][i1    ][j1 + 1];
          best_nz_coeff[1][0]= img->nz_coeff[currMB->mb_nr][i1 + 1][j1    ];
          best_nz_coeff[1][1]= img->nz_coeff[currMB->mb_nr][i1 + 1][j1 + 1];
        }
        else
        {
          best_nz_coeff[0][0]= 0;
          best_nz_coeff[0][1]= 0;
          best_nz_coeff[1][0]= 0;
          best_nz_coeff[1][1]= 0;
        }
#endif

        //--- store number of nonzero coefficients ---
        best_cnt_nonz  = cnt_nonz;                

        //--- store best 8x8 coding state ---
        if (block < 3)
        {
          store_coding_state (currSlice, currMB, cs_b8);
          stored_state_8x8 = TRUE;
        }
      } // if (rdcost <= min_rdcost)

      //--- re-set coding state as it was before coding with current mode was performed ---
      reset_coding_state (currSlice, currMB, 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++)

  if (valid_8x8 == TRUE)
  {
#ifdef BEST_NZ_COEFF
    for(i = 0; i <= 1; i++)  
    {