mode_decision.c

H.264编码实现

      // Note that if currMB->cbp is 0 one could choose to skip 8x8 mode
      // although this could be due to deadzoning decisions.
      //if (params->Transform8x8Mode==1 && currMB->cbp!=0)
      if (params->Transform8x8Mode==1)
      {
        //=========== try mb_types 1,2,3 with 8x8 transform ===========
        if ((mode >= 1 && mode <= 3) && currMB->luma_transform_size_8x8_flag == 0)
        {
          //try with 8x8 transform size
          currMB->luma_transform_size_8x8_flag = 1;
          continue;
        }
        //=========== try DIRECT-MODE with 8x8 transform ===========
        else if (mode == 0 && bslice && active_sps->direct_8x8_inference_flag && currMB->luma_transform_size_8x8_flag == 0)
        {
          //try with 8x8 transform size
          currMB->luma_transform_size_8x8_flag = 1;
          continue;
        }
        //=========== try mb_type P8x8 for mode 4 with 4x4/8x8 transform ===========
        else if ((mode == P8x8) && (enc_mb->valid[4]) && (currMB->luma_transform_size_8x8_flag == 0))
        {
          currMB->luma_transform_size_8x8_flag = 1; //check 8x8 partition for transform size 8x8
          continue;
        }
        else
        {
          currMB->luma_transform_size_8x8_flag = 0;
          break;
        }
      }
      else
        break;
    }

    // Encode with no coefficients. Currently only for direct. This could be extended to all other modes as in example.
    //if (mode < P8x8 && (*inter_skip == 0) && enc_mb->valid[mode] && currMB->cbp && (currMB->cbp&15) != 15 && !params->nobskip)
    if ( bslice && mode == 0 && (*inter_skip == 0) && enc_mb->valid[mode] 
    && currMB->cbp && (currMB->cbp&15) != 15 && !params->nobskip
    && !(currMB->qp_scaled[0] == 0 && img->lossless_qpprime_flag==1) )
    {
      img->NoResidueDirect = 1;

      if (RDCost_for_macroblocks (currMB, enc_mb->lambda_md, mode, min_rdcost, min_dcost, min_rate, i16mode, is_cavlc))
      {
        //Rate control
        if (params->RCEnable)
          rc_store_diff(img->opix_x,img->opix_y,pred);

        store_macroblock_parameters (currMB, mode);
      }
    }
  }
}


/*!
*************************************************************************************
* \brief
*    Mode Decision for an 8x8 sub-macroblock
*************************************************************************************
*/
void submacroblock_mode_decision(RD_PARAMS *enc_mb,
                                 RD_8x8DATA *dataTr,
                                 Macroblock *currMB,
                                 int ***cofACtr,
                                 int ***cofACCbCrtr1, 
                                 int ***cofACCbCrtr2,
                                 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];
  static int fadjust[16][16], fadjustCr[2][16][16];
  int ***fadjustTransform = transform8x8? img->fadjust8x8 : img->fadjust4x4;
  int ****fadjustTransformCr   = (params->AdaptRndChroma || (img->P444_joined )) 
    ? (transform8x8 ? img->fadjust8x8Cr : img->fadjust4x4Cr): NULL;
  int lumaAdjustIndex   = transform8x8 ? 2 : 3;
  int chromaAdjustIndex = transform8x8 ? 0 : 2;
  short pdir, bipred_me = 0;


  char best_pdir = 0;
  char best_ref[2] = {0, -1};
  imgpel  (*mb_pred)[16] = img->mb_pred[0];

#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[img->current_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 (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 == 0 || mode != 0 || (mode == 0 && active_sps->direct_8x8_inference_flag)))
    {
      curr_cbp_blk = 0;

      if (mode==0)
      {
        //--- Direct Mode ---
        if (!params->rdopt )
        {
          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 (params->rdopt)
      {
        //--- get and check rate-distortion cost ---
        rdcost = RDCost_for_8x8blocks (currMB, &cnt_nonz, &curr_cbp_blk, enc_mb->lambda_md,
          block, mode, best_pdir, best_ref[LIST_0], best_ref[LIST_1], bipred_me, is_cavlc);
      }
      else
      {
        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 ( ( params->rdopt && rdcost < min_rdcost) || (!params->rdopt && *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)
        {
          for(i = 0; i <= 1; i++)
          {
            for(j = 0; j <= 1; j++)  
              best_nz_coeff[i][j]= img->nz_coeff[img->current_mb_nr][i1 + i][j1 + j];
          }
        }
        else
        {
          for(i = 0; i <= 1; i++)
          {
            best_nz_coeff[i][0]= 0;
            best_nz_coeff[i][1]= 0;
          }
        }
#endif

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

        if (params->rdopt)
        {
          //--- store block cbp ---
          cbp_blk8x8 &= (~(0x33 << (((block>>1)<<3)+((block & 0x01)<<1)))); // delete bits for block
          cbp_blk8x8 |= curr_cbp_blk;

          //--- store coefficients ---
          memcpy(&cofACtr[0][0][0],&img->cofAC[block][0][0][0], 4 * 2 * 65 * sizeof(int));

          if( img->P444_joined ) 
          {
            //--- store coefficients ---
            memcpy(&cofACCbCrtr1[0][0][0],&img->cofAC[block + 4][0][0][0], 4 * 2 * 65 * sizeof(int));
            memcpy(&cofACCbCrtr2[0][0][0],&img->cofAC[block + 8][0][0][0], 4 * 2 * 65 * sizeof(int));
          }
          
          //--- store reconstruction and prediction ---
          for (j=j0; j<j0 + BLOCK_SIZE_8x8; j++)
          {
            memcpy(&dataTr->rec_mbY8x8[j][i0],&enc_picture->imgY[img->pix_y + j][img->pix_x + i0], BLOCK_SIZE_8x8 * sizeof(imgpel));
            memcpy(&dataTr->mpr8x8[j][i0],&mb_pred[j][i0], BLOCK_SIZE_8x8 * sizeof(imgpel));
          }

          if (params->rdopt == 3)
          {
            errdo_store_best_block(img, decs->dec_mbY8x8, enc_picture->p_dec_img[0], i0, j0, BLOCK_SIZE_8x8);  //Store best 8x8 block for each hypothetical decoder
          }

          if(img->type==SP_SLICE && (!si_frame_indicator))
          {
            for (j=j0; j<j0 + BLOCK_SIZE_8x8; j++)
            {
              memcpy(&dataTr->lrec[j][i0],&lrec[img->pix_y + j][img->pix_x + i0], BLOCK_SIZE_8x8 * sizeof(int));
            }
          }
 
          if(img->P444_joined)