Exemplo n.º 1
0
/*!
*************************************************************************************
* \brief
*    Mode Decision for a macroblock with error resilience
*************************************************************************************
*/
void encode_one_macroblock_highloss (Macroblock *currMB)
{
  Slice *currSlice = currMB->p_Slice;
  RDOPTStructure  *p_RDO = currSlice->p_RDO;
  VideoParameters *p_Vid = currMB->p_Vid;
  InputParameters *p_Inp = currMB->p_Inp;
  PicMotionParams **motion = p_Vid->enc_picture->mv_info;

  int         max_index = 9;
  int         rerun, block, index, mode, i, j;
  RD_PARAMS   enc_mb;
  distblk     bmcost[5] = {DISTBLK_MAX};
  distblk     cost=0;
  distblk     min_cost = DISTBLK_MAX;
  int         intra1 = 0;
  int         mb_available[3];

  short       bslice      = (short) (currSlice->slice_type == B_SLICE);
  short       pslice      = (short) ((currSlice->slice_type == P_SLICE) || (currSlice->slice_type == SP_SLICE));
  short       intra       = (short) ((currSlice->slice_type == I_SLICE) || (currSlice->slice_type == SI_SLICE) || (pslice && currMB->mb_y == p_Vid->mb_y_upd && p_Vid->mb_y_upd != p_Vid->mb_y_intra));
  int         lambda_mf[3];
  short       runs        = (short) ((p_Inp->RestrictRef==1 && (pslice  || (bslice && p_Vid->nal_reference_idc>0))) ? 2 : 1);

  imgpel    **mb_pred  = currSlice->mb_pred[0];
  Block8x8Info *b8x8info = p_Vid->b8x8info;

  char       chroma_pred_mode_range[2];
  short       inter_skip = 0;
  BestMode    md_best;
  Info8x8     best;

  init_md_best(&md_best);

  // Init best (need to create simple function)
  best.pdir = 0;
  best.bipred = 0;
  best.ref[LIST_0] = 0;
  best.ref[LIST_1] = -1;

  intra |= RandomIntra (p_Vid, currMB->mbAddrX);    // Forced Pseudo-Random Intra

  //===== Setup Macroblock encoding parameters =====
  init_enc_mb_params(currMB, &enc_mb, intra);

  // Perform multiple encodings if rdopt with losses is enabled
  for (rerun=0; rerun<runs; rerun++)
  {
    if (runs==2)
      p_Inp->rdopt= (rerun==0) ? 1 : 3;

    if (p_Inp->AdaptiveRounding)
    {
      reset_adaptive_rounding(p_Vid);
    }

    if (currSlice->mb_aff_frame_flag)
    {
      reset_mb_nz_coeff(p_Vid, currMB->mbAddrX);
    }

    //=====   S T O R E   C O D I N G   S T A T E   =====
    //---------------------------------------------------
    currSlice->store_coding_state (currMB, currSlice->p_RDO->cs_cm);

    if (!intra)
    {
      //===== set skip/direct motion vectors =====
      if (enc_mb.valid[0])
      {
        if (bslice)
          currSlice->Get_Direct_Motion_Vectors (currMB);
        else 
          FindSkipModeMotionVector (currMB);
      }
      if (p_Inp->CtxAdptLagrangeMult == 1)
      {
        get_initial_mb16x16_cost(currMB);
      }

      //===== MOTION ESTIMATION FOR 16x16, 16x8, 8x16 BLOCKS =====
      for (mode = 1; mode < 4; mode++)
      {
        best.mode = (char) mode;
        best.bipred = 0;
        b8x8info->best[mode][0].bipred = 0;

        if (enc_mb.valid[mode])
        {
          for (cost=0, block=0; block<(mode==1?1:2); block++)
          {
            update_lambda_costs(currMB, &enc_mb, lambda_mf);
            PartitionMotionSearch (currMB, mode, block, lambda_mf);

          //--- set 4x4 block indices (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] = DISTBLK_MAX;
            list_prediction_cost(currMB, LIST_0, block, mode, &enc_mb, bmcost, best.ref);

            if (bslice)
            {
              //--- get cost and reference frame for List 1 prediction ---
              bmcost[LIST_1] = DISTBLK_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 predictive ME is only supported for modes 1, 2, 3 and ref 0
              if (is_bipred_enabled(p_Vid, mode))
              {
                get_bipred_cost(currMB, mode, block, i, j, &best, &enc_mb, bmcost);
              }
              else
              {
                bmcost[BI_PRED_L0] = DISTBLK_MAX;
                bmcost[BI_PRED_L1] = DISTBLK_MAX;
              }

              // Determine prediction list based on mode cost
              determine_prediction_list(bmcost, &best, &cost);
            }
            else // if (bslice)
            {
              best.pdir  = 0;
              cost      += bmcost[LIST_0];
            }

            assign_enc_picture_params(currMB, mode, &best, 2 * block);

            //----- set reference frame and direction parameters -----
            set_block8x8_info(b8x8info, mode, block, &best);

            //--- set reference frames and motion vectors ---
            if (mode>1 && block==0)
              currSlice->set_ref_and_motion_vectors (currMB, motion, &best, block);
          } // for (block=0; block<(mode==1?1:2); block++)
          if (cost < min_cost)
          {
            md_best.mode = (byte) mode;
            md_best.cost = cost;
            currMB->best_mode = (short) mode;
            min_cost  = cost;
            if (p_Inp->CtxAdptLagrangeMult == 1)
            {
              adjust_mb16x16_cost(currMB, cost);
            }
          }
        } // if (enc_mb.valid[mode])
      } // for (mode=1; mode<4; mode++)

      if (enc_mb.valid[P8x8])
      {
        currMB->valid_8x8 = FALSE;

        if (p_Inp->Transform8x8Mode)
        {
          ResetRD8x8Data(p_Vid, p_RDO->tr8x8);
          currMB->luma_transform_size_8x8_flag = TRUE; //switch to 8x8 transform size
          //===========================================================
          // Check 8x8 partition with transform size 8x8
          //===========================================================
          //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
          for (block = 0; block < 4; block++)
          {
            currSlice->submacroblock_mode_decision(currMB, &enc_mb, p_RDO->tr8x8, p_RDO->cofAC8x8ts[block], block, &cost);
            if(!currMB->valid_8x8)
              break;
            set_subblock8x8_info(b8x8info, P8x8, block, p_RDO->tr8x8);
          }
        }// if (p_Inp->Transform8x8Mode)

        currMB->valid_4x4 = FALSE;
        if (p_Inp->Transform8x8Mode != 2)
        {
          currMB->luma_transform_size_8x8_flag = FALSE; //switch to 8x8 transform size
          ResetRD8x8Data(p_Vid, p_RDO->tr4x4);
          //=================================================================
          // Check 8x8, 8x4, 4x8 and 4x4 partitions with transform size 4x4
          //=================================================================
          //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
          for (block = 0; block < 4; block++)
          {
            currSlice->submacroblock_mode_decision(currMB, &enc_mb, p_RDO->tr4x4, p_RDO->coefAC8x8[block], block, &cost);
            if(!currMB->valid_4x4)
              break;
            set_subblock8x8_info(b8x8info, P8x8, block, p_RDO->tr4x4);
          }

        }// if (p_Inp->Transform8x8Mode != 2)

        if (p_Inp->RCEnable)
          rc_store_diff(currSlice->diffy, &p_Vid->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);

        p_Vid->giRDOpt_B8OnlyFlag = FALSE;
      }
    }
    else // if (!intra)
    {
      min_cost = DISTBLK_MAX;
    }

    // Set Chroma mode
    set_chroma_pred_mode(currMB, enc_mb, mb_available, chroma_pred_mode_range);

    //========= C H O O S E   B E S T   M A C R O B L O C K   M O D E =========
    //-------------------------------------------------------------------------
    for (currMB->c_ipred_mode = chroma_pred_mode_range[0]; currMB->c_ipred_mode<=chroma_pred_mode_range[1]; currMB->c_ipred_mode++)
    {
      // bypass if c_ipred_mode is not allowed
      if ( (p_Vid->yuv_format != YUV400) &&
        (  ((!intra || !p_Inp->IntraDisableInterOnly) && p_Inp->ChromaIntraDisable == 1 && currMB->c_ipred_mode!=DC_PRED_8)
        || (currMB->c_ipred_mode == VERT_PRED_8 && !mb_available[0]) 
        || (currMB->c_ipred_mode == HOR_PRED_8  && !mb_available[1]) 
        || (currMB->c_ipred_mode == PLANE_8     && (!mb_available[1] || !mb_available[0] || !mb_available[2]))))
        continue;

      //===== GET BEST MACROBLOCK MODE =====
      for (index=0; index < max_index; index++)
      {
        mode = mb_mode_table[index];
        if (enc_mb.valid[mode])
        {
          if (p_Vid->yuv_format != YUV400)
          {           
            currMB->i16mode = 0; 
          }

          // Skip intra modes in inter slices if best mode is inter <P8x8 with cbp equal to 0    
          if (currSlice->P444_joined)
          { 
            if (p_Inp->SkipIntraInInterSlices && !intra && mode >= I16MB 
              && currMB->best_mode <=3 && currMB->best_cbp == 0 && currSlice->cmp_cbp[1] == 0 && currSlice->cmp_cbp[2] == 0 && (currMB->min_rdcost < weighted_cost(enc_mb.lambda_mdfp,5)))
              continue;
          }
          else
          {
            if (p_Inp->SkipIntraInInterSlices)
            {
              if (!intra && mode >= I4MB)
              {
                if (currMB->best_mode <=3 && currMB->best_cbp == 0 && (currMB->min_rdcost < weighted_cost(enc_mb.lambda_mdfp, 5)))
                {
                  continue;
                }
                else if (currMB->best_mode == 0 && (currMB->min_rdcost < weighted_cost(enc_mb.lambda_mdfp,6)))
                {
                  continue;
                }
              }
            }
          }
          compute_mode_RD_cost(currMB, &enc_mb, (short) mode, &inter_skip);

        }
      }// for (index=0; index<max_index; index++)
    }// for (currMB->c_ipred_mode=DC_PRED_8; currMB->c_ipred_mode<=chroma_pred_mode_range[1]; currMB->c_ipred_mode++)                     

    restore_nz_coeff(currMB);

    if (rerun==0)
      intra1 = is_intra(currMB);
  } // for (rerun=0; rerun<runs; rerun++)

  //=====  S E T   F I N A L   M A C R O B L O C K   P A R A M E T E R S ======
  //---------------------------------------------------------------------------
  update_qp_cbp_tmp(currMB, p_RDO->cbp);
  currSlice->set_stored_mb_parameters (currMB);

  // Rate control
  if(p_Inp->RCEnable && p_Inp->RCUpdateMode <= MAX_RC_MODE)
    rc_store_mad(currMB);

  //===== Decide if this MB will restrict the reference frames =====
  if (p_Inp->RestrictRef)
    update_refresh_map(currMB, intra, intra1);
}
Exemplo n.º 2
0
/*!
*************************************************************************************
* \brief
*    Mode Decision for a macroblock
*************************************************************************************
*/
void encode_one_macroblock_low (Macroblock *currMB)
{
    Slice *currSlice = currMB->p_slice;
    RDOPTStructure  *p_RDO = currSlice->p_RDO;
    ImageParameters *p_Img = currMB->p_Img;
    InputParameters *p_Inp = currMB->p_Inp;
    PicMotionParams *motion = &p_Img->enc_picture->motion;

    imgpel ***temp_img; // to temp store the Y data for 8x8 transform

    int         block, mode, i=0, j;
    RD_PARAMS   enc_mb;
    int         bmcost[5] = {INT_MAX};
    double      rd_cost = 0;
    int         cost = 0;
    int         min_cost = INT_MAX, cost_direct=0, have_direct=0;
    int         intra1 = 0;
    int         temp_cpb = 0;
    Boolean     best_transform_flag = FALSE;
    int         cost8x8_direct = 0;
    short       islice      = (short) (currSlice->slice_type == I_SLICE);
    short       bslice      = (short) (currSlice->slice_type == B_SLICE);
    short       pslice      = (short) ((currSlice->slice_type == P_SLICE) || (currSlice->slice_type == SP_SLICE));
    short       intra       = (short) (islice || (pslice && currMB->mb_y == p_Img->mb_y_upd && p_Img->mb_y_upd!=p_Img->mb_y_intra));
    int         lambda_mf[3];
    Block8x8Info *b8x8info   = p_Img->b8x8info;
    //int         mb_available[3] = { 1, 1, 1};

    char   **ipredmodes = p_Img->ipredmode;
    short   *allmvs = (currSlice->slice_type == I_SLICE) ? NULL: currSlice->all_mv[0][0][0][0][0];
    int     ****i4p;  //for non-RD-opt. mode
    imgpel  **mb_pred = currSlice->mb_pred[0];

    Boolean tmp_8x8_flag, tmp_no_mbpart;

    BestMode    md_best;
    Info8x8 best;

    init_md_best(&md_best);

    // Init best (need to create simple function)
    best.pdir = 0;
    best.bipred = 0;
    best.ref[LIST_0] = 0;
    best.ref[LIST_1] = -1;

    get_mem3Dpel(&temp_img, 3, MB_BLOCK_SIZE, MB_BLOCK_SIZE);

    intra |= RandomIntra (p_Img, currMB->mbAddrX);    // Forced Pseudo-Random Intra

    //===== Setup Macroblock encoding parameters =====
    init_enc_mb_params(currMB, &enc_mb, intra);
    if (p_Inp->AdaptiveRounding)
    {
        reset_adaptive_rounding(p_Img);
    }

    if (currSlice->MbaffFrameFlag)
    {
        reset_mb_nz_coeff(p_Img, currMB->mbAddrX);
    }

    //=====   S T O R E   C O D I N G   S T A T E   =====
    //---------------------------------------------------
    currSlice->store_coding_state (currMB, currSlice->p_RDO->cs_cm);

    if (!intra)
    {
        //===== set direct motion vectors =====
        currMB->best_mode = 10;  // let us set best_mode to an intra mode to avoid possible bug with RDOQ
        if (bslice && enc_mb.valid[0])
        {
            currSlice->Get_Direct_Motion_Vectors (currMB);
        }

        if (p_Inp->CtxAdptLagrangeMult == 1)
        {
            get_initial_mb16x16_cost(currMB);
        }

        //===== MOTION ESTIMATION FOR 16x16, 16x8, 8x16 BLOCKS =====
        for (mode = 1; mode < 4; mode++)
        {
            best.bipred = 0;
            best.mode = (char) mode;
            b8x8info->best[mode][0].bipred = 0;
            if (enc_mb.valid[mode])
            {
                for (cost=0, block=0; block<(mode==1?1:2); block++)
                {
                    update_lambda_costs(currMB, &enc_mb, lambda_mf);
                    PartitionMotionSearch (currMB, mode, block, 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(currMB, 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(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 predictive ME is only supported for modes 1, 2, 3 and ref 0
                        if (is_bipred_enabled(p_Inp, 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;
                        }

                        // Determine prediction list based on mode cost
                        determine_prediction_list(bmcost, &best, &cost);
                    }
                    else // if (bslice)
                    {
                        best.pdir  = 0;
                        cost      += bmcost[LIST_0];
                    }

                    assign_enc_picture_params(currMB, mode, &best, 2 * block);

                    //----- set reference frame and direction parameters -----
                    set_block8x8_info(b8x8info, mode, block, &best);

                    //--- set reference frames and motion vectors ---
                    if (mode>1 && block==0)
                        currSlice->set_ref_and_motion_vectors (currMB, motion, &best, block);
                } // for (block=0; block<(mode==1?1:2); block++)

                currMB->luma_transform_size_8x8_flag = FALSE;
                if (p_Inp->Transform8x8Mode) //for inter rd-off, set 8x8 to do 8x8 transform
                {
                    SetModesAndRefframeForBlocks(currMB, (short) mode);
                    currMB->luma_transform_size_8x8_flag = (byte) TransformDecision(currMB, -1, &cost);
                }

                if (cost < min_cost)
                {
                    currMB->best_mode = (short) mode;
                    min_cost  = cost;
                    best_transform_flag = currMB->luma_transform_size_8x8_flag;

                    if (p_Inp->CtxAdptLagrangeMult == 1)
                    {
                        adjust_mb16x16_cost(currMB, cost);
                    }
                }
            } // if (enc_mb.valid[mode])
        } // for (mode=1; mode<4; mode++)

        if (enc_mb.valid[P8x8])
        {
            //===== store coding state of macroblock =====
            currSlice->store_coding_state (currMB, currSlice->p_RDO->cs_mb);
            memset( currSlice->cofAC[0][0][0], 0, 2080 * sizeof(int)); // 4 * 4 * 2 * 65

            currMB->valid_8x8 = FALSE;

            if (p_Inp->Transform8x8Mode)
            {
                ResetRD8x8Data(p_Img, p_RDO->tr8x8);
                //===========================================================
                // Check 8x8 partition with transform size 8x8
                //===========================================================
                //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
                for (cost_direct =  0, block = 0; block < 4; block++)
                {
                    submacroblock_mode_decision_low(currMB, &enc_mb, p_RDO->tr8x8, p_RDO->cofAC8x8ts[block],
                                                    &have_direct, block, &cost_direct, &cost, &cost8x8_direct, 1);

                    set_subblock8x8_info(b8x8info, P8x8, block, p_RDO->tr8x8);
                }

                currMB->luma_transform_size_8x8_flag = FALSE; //switch to 4x4 transform size
            }// if (p_Inp->Transform8x8Mode)


            if (p_Inp->Transform8x8Mode != 2)
            {
                ResetRD8x8Data(p_Img, p_RDO->tr4x4);
                //=================================================================
                // Check 8x8, 8x4, 4x8 and 4x4 partitions with transform size 4x4
                //=================================================================
                //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
                for (cost_direct = 0, block=0; block<4; block++)
                {
                    submacroblock_mode_decision_low(currMB, &enc_mb, p_RDO->tr4x4, p_RDO->coefAC8x8[block],
                                                    &have_direct, block, &cost_direct, &cost, &cost8x8_direct, 0);

                    set_subblock8x8_info(b8x8info, P8x8, block, p_RDO->tr4x4);
                }
            }// if (p_Inp->Transform8x8Mode != 2)

            if (p_Inp->RCEnable)
                rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);

            //check cost for P8x8 for non-rdopt mode
            if (((p_Inp->Transform8x8Mode < 2) && (p_RDO->tr4x4->mb_p8x8_cost < min_cost)) ||
                    ((p_Inp->Transform8x8Mode >  0) && (p_RDO->tr8x8->mb_p8x8_cost < min_cost)))
            {
                currMB->best_mode = P8x8;
                if (p_Inp->Transform8x8Mode == 2)
                {
                    min_cost = p_RDO->tr8x8->mb_p8x8_cost;
                    currMB->luma_transform_size_8x8_flag = TRUE;
                }
                else if (p_Inp->Transform8x8Mode)
                {
                    if (p_RDO->tr8x8->mb_p8x8_cost < p_RDO->tr4x4->mb_p8x8_cost)
                    {
                        min_cost = p_RDO->tr8x8->mb_p8x8_cost;
                        currMB->luma_transform_size_8x8_flag = TRUE;
                    }
                    else if(p_RDO->tr4x4->mb_p8x8_cost < p_RDO->tr8x8->mb_p8x8_cost)
                    {
                        min_cost = p_RDO->tr4x4->mb_p8x8_cost;
                        currMB->luma_transform_size_8x8_flag = FALSE;
                    }
                    else
                    {
                        if (GetBestTransformP8x8(currMB) == 0)
                        {
                            min_cost = p_RDO->tr4x4->mb_p8x8_cost;
                            currMB->luma_transform_size_8x8_flag = FALSE;
                        }
                        else
                        {
                            min_cost = p_RDO->tr8x8->mb_p8x8_cost;
                            currMB->luma_transform_size_8x8_flag = TRUE;
                        }
                    }
                }
                else
                {
                    min_cost = p_RDO->tr4x4->mb_p8x8_cost;
                    currMB->luma_transform_size_8x8_flag = FALSE;
                }
            }// if ((p_RDO->tr4x4->mb_p8x8_cost < min_cost || p_RDO->tr8x8->mb_p8x8_cost < min_cost))
            p_Img->giRDOpt_B8OnlyFlag = FALSE;
        }

        // Find a motion vector for the Skip mode
        if(pslice)
            FindSkipModeMotionVector (currMB);
    }
    else // if (!intra)
    {
        min_cost = INT_MAX;
    }


    //========= C H O O S E   B E S T   M A C R O B L O C K   M O D E =========
    //-------------------------------------------------------------------------
    tmp_8x8_flag  = currMB->luma_transform_size_8x8_flag;  //save 8x8_flag
    tmp_no_mbpart = currMB->NoMbPartLessThan8x8Flag;      //save no-part-less
    if ((p_Img->yuv_format != YUV400) && (p_Img->yuv_format != YUV444))
        // precompute all chroma intra prediction modes
        intra_chroma_prediction(currMB, NULL, NULL, NULL);

    if (enc_mb.valid[0] && bslice) // check DIRECT MODE
    {
        if(have_direct)
        {
            switch(p_Inp->Transform8x8Mode)
            {
            case 1: // Mixture of 8x8 & 4x4 transform
                cost = ((cost8x8_direct < cost_direct) || !(enc_mb.valid[5] && enc_mb.valid[6] && enc_mb.valid[7]))
                       ? cost8x8_direct : cost_direct;
                break;
            case 2: // 8x8 Transform only
                cost = cost8x8_direct;
                break;
            default: // 4x4 Transform only
                cost = cost_direct;
                break;
            }
        }
        else
        {   //!have_direct
            cost = GetDirectCostMB (currMB);
        }
        if (cost!=INT_MAX)
        {
            cost -= (int)floor(16 * enc_mb.lambda_md + 0.4999);
        }

        if (cost <= min_cost)
        {
            if(p_Img->active_sps->direct_8x8_inference_flag && p_Inp->Transform8x8Mode)
            {
                if(p_Inp->Transform8x8Mode==2)
                    currMB->luma_transform_size_8x8_flag = TRUE;
                else
                {
                    if(cost8x8_direct < cost_direct)
                        currMB->luma_transform_size_8x8_flag = TRUE;
                    else
                        currMB->luma_transform_size_8x8_flag = FALSE;
                }
            }
            else
                currMB->luma_transform_size_8x8_flag = FALSE;

            //Rate control
            if (p_Inp->RCEnable)
                rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);

            min_cost  = cost;
            currMB->best_mode = 0;
            tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;
        }
        else
        {
            currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore if not best
            currMB->NoMbPartLessThan8x8Flag = (byte) tmp_no_mbpart; // restore if not best
        }
    }

    currMB->min_rdcost = (double) min_cost;

    if (enc_mb.valid[I8MB]) // check INTRA8x8
    {
        currMB->luma_transform_size_8x8_flag = TRUE; // at this point cost will ALWAYS be less than min_cost

        currMB->mb_type = currMB->ar_mode = I8MB;
        temp_cpb = Mode_Decision_for_Intra8x8Macroblock (currMB, enc_mb.lambda_md, &rd_cost);


        if (rd_cost <= currMB->min_rdcost) //HYU_NOTE. bug fix. 08/15/07
        {
            currMB->cbp = temp_cpb;
            if (p_Img->P444_joined)
            {
                currSlice->curr_cbp[0] = currSlice->cmp_cbp[1];
                currSlice->curr_cbp[1] = currSlice->cmp_cbp[2];
            }

            if(enc_mb.valid[I4MB])
            {
                //coeffs
                if (p_Inp->Transform8x8Mode != 2)
                {
                    i4p = p_RDO->cofAC;
                    p_RDO->cofAC = currSlice->cofAC;
                    currSlice->cofAC = i4p;
                }
            }

            copy_image_data_16x16(temp_img[0], &p_Img->enc_picture->imgY[currMB->pix_y], 0, currMB->pix_x);

            if (p_Img->P444_joined)
            {
                copy_image_data_16x16(temp_img[1], &p_Img->enc_picture->imgUV[0][currMB->pix_y], 0, currMB->pix_x);
                copy_image_data_16x16(temp_img[2], &p_Img->enc_picture->imgUV[1][currMB->pix_y], 0, currMB->pix_x);
            }

            //Rate control
            if (p_Inp->RCEnable)
                rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);

            currMB->min_rdcost  = rd_cost;
            currMB->best_mode = I8MB;
            tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;
        }
        else
        {
            currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore if not best
            if (p_Img->P444_joined)
            {
                currMB->cbp |= currSlice->curr_cbp[0];
                currMB->cbp |= currSlice->curr_cbp[1];
                currSlice->cmp_cbp[1] = currMB->cbp;
                currSlice->cmp_cbp[2] = currMB->cbp;
            }
        }
    }

    if (enc_mb.valid[I4MB]) // check INTRA4x4
    {
        currMB->luma_transform_size_8x8_flag = FALSE;
        currMB->mb_type = currMB->ar_mode = I4MB;
        temp_cpb = Mode_Decision_for_Intra4x4Macroblock (currMB, enc_mb.lambda_md, &rd_cost);

        if (rd_cost <= currMB->min_rdcost)
        {
            currMB->cbp = temp_cpb;

            //Rate control
            if (p_Inp->RCEnable)
                rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);

            currMB->min_rdcost  = rd_cost;
            currMB->best_mode = I4MB;
            tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;
        }
        else
        {
            currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore if not best
            if (p_Img->P444_joined)
            {
                currMB->cbp |= currSlice->curr_cbp[0];
                currMB->cbp |= currSlice->curr_cbp[1];
                currSlice->cmp_cbp[1] = currMB->cbp;
                currSlice->cmp_cbp[2] = currMB->cbp;
            }
            //coeffs
            i4p = p_RDO->cofAC;
            p_RDO->cofAC = currSlice->cofAC;
            currSlice->cofAC=i4p;
        }
    }
    if (enc_mb.valid[I16MB]) // check INTRA16x16
    {
        currMB->luma_transform_size_8x8_flag = FALSE;
        intrapred_16x16 (currMB, PLANE_Y);
        if (p_Img->P444_joined)
        {
            select_plane(p_Img, PLANE_U);
            intrapred_16x16 (currMB, PLANE_U);
            select_plane(p_Img, PLANE_V);
            intrapred_16x16 (currMB, PLANE_V);
            select_plane(p_Img, PLANE_Y);
        }

        rd_cost = currSlice->find_sad_16x16 (currMB);

        if (rd_cost < currMB->min_rdcost)
        {
            //Rate control
            if (p_Inp->RCEnable)
                rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, currSlice->mpr_16x16[0][(short) currMB->i16mode]);

            currMB->best_mode   = I16MB;
            currMB->min_rdcost  = rd_cost;
            currMB->cbp = currMB->trans_16x16 (currMB, PLANE_Y);

            if (p_Img->P444_joined)
            {
                select_plane(p_Img, PLANE_U);
                currSlice->cmp_cbp[1] = currMB->trans_16x16(currMB, PLANE_U);
                select_plane(p_Img, PLANE_V);
                currSlice->cmp_cbp[2] = currMB->trans_16x16(currMB, PLANE_V);

                select_plane(p_Img, PLANE_Y);
                currMB->cbp |= currSlice->cmp_cbp[1];
                currMB->cbp |= currSlice->cmp_cbp[2];
                currSlice->cmp_cbp[1] = currMB->cbp;
                currSlice->cmp_cbp[2] = currMB->cbp;
            }

        }
        else
        {
            currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore
            currMB->NoMbPartLessThan8x8Flag = (byte) tmp_no_mbpart;     // restore
        }
    }

    intra1 = IS_INTRA(currMB);

    //=====  S E T   F I N A L   M A C R O B L O C K   P A R A M E T E R S ======
    //---------------------------------------------------------------------------
    {
        //===== set parameters for chosen mode =====
        SetModesAndRefframeForBlocks (currMB, currMB->best_mode);

        if (currMB->best_mode == P8x8)
        {
            if (currMB->luma_transform_size_8x8_flag && (p_RDO->tr8x8->cbp8x8 == 0) && p_Inp->Transform8x8Mode != 2)
                currMB->luma_transform_size_8x8_flag = FALSE;

            SetCoeffAndReconstruction8x8 (currMB);

            memset(currMB->intra_pred_modes, DC_PRED, MB_BLOCK_PARTITIONS * sizeof(char));
            for (j = currMB->block_y; j < currMB->block_y + BLOCK_MULTIPLE; j++)
                memset(&ipredmodes[j][currMB->block_x], DC_PRED, BLOCK_MULTIPLE * sizeof(char));
        }
        else
        {
            //===== set parameters for chosen mode =====
            if (currMB->best_mode == I8MB)
            {
                memcpy(currMB->intra_pred_modes,currMB->intra_pred_modes8x8, MB_BLOCK_PARTITIONS * sizeof(char));
                for(j = currMB->block_y; j < currMB->block_y + BLOCK_MULTIPLE; j++)
                    memcpy(&p_Img->ipredmode[j][currMB->block_x],&p_Img->ipredmode8x8[j][currMB->block_x], BLOCK_MULTIPLE * sizeof(char));

                //--- restore reconstruction for 8x8 transform ---
                copy_image_data_16x16(&p_Img->enc_picture->imgY[currMB->pix_y], temp_img[0], currMB->pix_x, 0);

                if (p_Img->P444_joined)
                {
                    copy_image_data_16x16(&p_Img->enc_picture->imgUV[0][currMB->pix_y], temp_img[1], currMB->pix_x, 0);
                    copy_image_data_16x16(&p_Img->enc_picture->imgUV[1][currMB->pix_y], temp_img[2], currMB->pix_x, 0);
                }
            }

            if ((currMB->best_mode!=I4MB)&&(currMB->best_mode != I8MB))
            {
                memset(currMB->intra_pred_modes,DC_PRED, MB_BLOCK_PARTITIONS * sizeof(char));
                for(j = currMB->block_y; j < currMB->block_y + BLOCK_MULTIPLE; j++)
                    memset(&ipredmodes[j][currMB->block_x],DC_PRED, BLOCK_MULTIPLE * sizeof(char));
                currMB->ar_mode = currMB->best_mode;

                if (currMB->best_mode!=I16MB)
                {
                    if((currMB->best_mode >= 1) && (currMB->best_mode <= 3))
                        currMB->luma_transform_size_8x8_flag = (byte) best_transform_flag;

                    if (currSlice->P444_joined)
                    {
                        luma_residual_coding_p444(currMB);
                        if((currMB->cbp==0 && currSlice->cmp_cbp[1] == 0 && currSlice->cmp_cbp[2] == 0) &&(currMB->best_mode == 0))
                            currMB->luma_transform_size_8x8_flag = FALSE;
                    }
                    else
                    {
                        luma_residual_coding(currMB);
                        if((currMB->cbp==0)&&(currMB->best_mode == 0))
                            currMB->luma_transform_size_8x8_flag = FALSE;
                    }

                    //Rate control
                    if (p_Inp->RCEnable)
                        rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);
                }
            }
        }
        //check luma cbp for transform size flag
        if (((currMB->cbp&15) == 0) && currMB->mb_type != I4MB && currMB->mb_type != I8MB)
            currMB->luma_transform_size_8x8_flag = FALSE;

        // precompute all chroma intra prediction modes
        if ((p_Img->yuv_format != YUV400) && (p_Img->yuv_format != YUV444))
            intra_chroma_prediction(currMB, NULL, NULL, NULL);

        currMB->i16offset = 0;

        if ((p_Img->yuv_format != YUV400) && (p_Img->yuv_format != YUV444))
            chroma_residual_coding (currMB);

        if (currMB->best_mode == I16MB)
        {
            currMB->i16offset = I16Offset  (currMB->cbp, currMB->i16mode);
        }

        currSlice->SetMotionVectorsMB (currMB, motion);

        //===== check for SKIP mode =====
        if(p_Img->P444_joined)
        {
            if ((pslice) && currMB->best_mode == 1 && currMB->cbp==0 && currSlice->cmp_cbp[1] == 0 && currSlice->cmp_cbp[2] == 0 &&
                    motion->ref_idx[LIST_0][currMB->block_y][currMB->block_x]    == 0 &&
                    motion->mv     [LIST_0][currMB->block_y][currMB->block_x][0] == allmvs[0] &&
                    motion->mv     [LIST_0][currMB->block_y][currMB->block_x][1] == allmvs[1])
            {
                currMB->mb_type = currMB->b8x8[0].mode = currMB->b8x8[1].mode = currMB->b8x8[2].mode = currMB->b8x8[3].mode = 0;
                currMB->luma_transform_size_8x8_flag = FALSE;
            }
        }
        else if ((pslice) && currMB->best_mode == 1 && currMB->cbp==0 &&
                 motion->ref_idx[LIST_0][currMB->block_y][currMB->block_x]    == 0 &&
                 motion->mv     [LIST_0][currMB->block_y][currMB->block_x][0] == allmvs[0] &&
                 motion->mv     [LIST_0][currMB->block_y][currMB->block_x][1] == allmvs[1])
        {
            currMB->mb_type = currMB->b8x8[0].mode = currMB->b8x8[1].mode = currMB->b8x8[2].mode = currMB->b8x8[3].mode = 0;
            currMB->luma_transform_size_8x8_flag = FALSE;
        }

        if (currSlice->MbaffFrameFlag || (p_Inp->UseRDOQuant && currSlice->RDOQ_QP_Num > 1))
            set_mbaff_parameters(currMB);
    }

    // Rate control
    if(p_Inp->RCEnable && p_Inp->RCUpdateMode <= MAX_RC_MODE)
        rc_store_mad(currMB);

    //===== Decide if this MB will restrict the reference frames =====
    if (p_Inp->RestrictRef)
        update_refresh_map(currMB, intra, intra1);


    /*update adaptive rounding offset p_Inp*/
    if (p_Img->AdaptiveRounding)
    {
        update_offset_params(currMB, currMB->best_mode, currMB->luma_transform_size_8x8_flag);
    }

    free_mem3Dpel(temp_img);
}
Exemplo n.º 3
0
/*!
 ************************************************************************
 * \brief
 *    Predict one Luma block on-the-fly
 ************************************************************************
 */
void luma_prediction_otf ( Macroblock* currMB, //!< Current Macroblock
                     int   block_x,      //!< relative horizontal block coordinate of block
                     int   block_y,      //!< relative vertical   block coordinate of block
                     int   block_size_x, //!< relative horizontal block coordinate of block
                     int   block_size_y, //!< relative vertical   block coordinate of block
                     int   p_dir,        //!< prediction direction (0=list0, 1=list1, 2=bipred)
                     int   list_mode[2], //!< list prediction mode (1-7, 0=DIRECT)
                     char  *ref_idx,     //!< reference pictures
                     short bipred_me     //!< use bi prediction mv (0=no bipred, 1 = use set 1, 2 = use set 2)
                     )
{
  VideoParameters *p_Vid     = currMB->p_Vid;
  Slice           *currSlice = currMB->p_Slice;
  DecodedPictureBuffer *p_Dpb = p_Vid->p_Dpb_layer[p_Vid->dpb_layer_id];
  imgpel l0_pred[MB_PIXELS];
  imgpel l1_pred[MB_PIXELS];
  int    tmp_pred[ (MB_BLOCK_SIZE+5)*(MB_BLOCK_SIZE+5) ]; // temporary pred block to compute on-the-fly interpolation 

  int  pic_opix_x   = ((currMB->pix_x + block_x) << 2);
  int  pic_opix_y   = ((currMB->opix_y + block_y) << 2);
  int  bx           = block_x >> 2;
  int  by           = block_y >> 2;
  MotionVector ***** mv_array = currSlice->all_mv;
  MotionVector *curr_mv = NULL;
  imgpel **mb_pred = currSlice->mb_pred[0];
  int  apply_weights = ( (currSlice->weighted_prediction == 1) || (currSlice->weighted_prediction == 2 && p_dir == 2) );

  if (bipred_me && ref_idx[0] == 0 && ref_idx[1] == 0 && p_dir == 2 && is_bipred_enabled(p_Vid, list_mode[0]) && is_bipred_enabled(p_Vid, list_mode[1]))
    mv_array = currSlice->bipred_mv[bipred_me - 1]; 

  switch (p_dir)
  {
  case 0:
    curr_mv = &mv_array[LIST_0][(short) ref_idx[0]][list_mode[0]][by][bx];
    p_Dpb->pf_get_block_luma (p_Vid, l0_pred, tmp_pred, pic_opix_x + curr_mv->mv_x, pic_opix_y + curr_mv->mv_y, block_size_x, block_size_y, currSlice->listX[LIST_0 + currMB->list_offset][(short) ref_idx[0]], 0 );
    break;
  case 1:
    curr_mv = &mv_array[LIST_1][(short) ref_idx[1]][list_mode[1]][by][bx];
    p_Dpb->pf_get_block_luma (p_Vid, l1_pred, tmp_pred, pic_opix_x + curr_mv->mv_x, pic_opix_y + curr_mv->mv_y, block_size_x, block_size_y, currSlice->listX[LIST_1 + currMB->list_offset][(short)ref_idx[1]], 0 );
    break;
  case 2:
    curr_mv = &mv_array[LIST_0][(short) ref_idx[0]][list_mode[0]][by][bx];
    p_Dpb->pf_get_block_luma (p_Vid, l0_pred, tmp_pred, pic_opix_x + curr_mv->mv_x, pic_opix_y + curr_mv->mv_y, block_size_x, block_size_y, currSlice->listX[LIST_0 + currMB->list_offset][(short)ref_idx[0]], 0);
    curr_mv = &mv_array[LIST_1][(short) ref_idx[1]][list_mode[1]][by][bx];
    p_Dpb->pf_get_block_luma (p_Vid, l1_pred, tmp_pred, pic_opix_x + curr_mv->mv_x, pic_opix_y + curr_mv->mv_y, block_size_x, block_size_y, currSlice->listX[LIST_1 + currMB->list_offset][(short)ref_idx[1]], 0);
    break;
  default:
    break;
  }

  if (apply_weights)
  {
    if (p_dir==2)
    {
      weighted_bi_prediction(&mb_pred[block_y], l0_pred, l1_pred, block_size_y, block_x, block_size_x, 
        p_Vid->max_imgpel_value,
        currSlice->wbp_weight[0][(short)ref_idx[0]][(short)ref_idx[1]][0], currSlice->wbp_weight[1][(short)ref_idx[0]][(short)ref_idx[1]][0],
        (currSlice->wp_offset[0][(short)ref_idx[0]][0] + currSlice->wp_offset[1][(short)ref_idx[1]][0] + 1)>>1, 
         currSlice->luma_log_weight_denom + 1);
    }
    else if (p_dir==0)
/*!
*************************************************************************************
* \brief
*    Mode Decision for a macroblock with error resilience
*************************************************************************************
*/
void encode_one_macroblock_highloss (Macroblock *currMB)
{
  int         max_index = 9;
  int         rerun, block, index, mode, i, j, ctr16x16;
  char        best_pdir;
  RD_PARAMS   enc_mb;
  double      min_rdcost = 1e30;
  double      min_dcost = 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=0;
  int         intra1 = 0;
  int         cost8x8_direct = 0;
  int         mb_available_up;
  int         mb_available_left;
  int         mb_available_up_left;
  int         best8x8l0ref, best8x8l1ref; 
  int         is_cavlc = (img->currentSlice->symbol_mode == CAVLC);

  short       islice      = (short) (img->type==I_SLICE);
  short       bslice      = (short) (img->type==B_SLICE);
  short       pslice      = (short) ((img->type==P_SLICE) || (img->type==SP_SLICE));
  short       intra       = (short) (islice || (pslice && img->mb_y==img->mb_y_upd && img->mb_y_upd!=img->mb_y_intra));
  int         lambda_mf[3];
  short       runs        = (short) (params->RestrictRef==1 && (pslice  || (bslice && img->nal_reference_idc>0)) ? 2 : 1);

  int         prev_mb_nr  = FmoGetPreviousMBNr(img->current_mb_nr);
  Macroblock* prevMB      = (prev_mb_nr >= 0) ? &img->mb_data[prev_mb_nr]:NULL ;
  imgpel  (*mb_pred)[16] = img->mb_pred[0];
  Block8x8Info *b8x8info   = img->b8x8info;

  short   min_chroma_pred_mode, max_chroma_pred_mode;

  short   inter_skip = 0;
  short   bipred_me = 0;
  double  min_rate = 0;

  if(params->SearchMode == UM_HEX)
  {
    UMHEX_decide_intrabk_SAD();
  }
  else if (params->SearchMode == UM_HEX_SIMPLE)
  {
    smpUMHEX_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)
      params->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 (currMB, cs_cm);

    if (!intra)
    {
      //===== set direct motion vectors =====
      best_mode = 1;
      if (bslice)
      {
        Get_Direct_Motion_Vectors (currMB);
      }

      if (params->CtxAdptLagrangeMult == 1)
      {
        get_initial_mb16x16_cost(currMB);
      }

      //===== MOTION ESTIMATION FOR 16x16, 16x8, 8x16 BLOCKS =====
      for (min_cost=INT_MAX, mode=1; mode<4; mode++)
      {
        bipred_me = 0;
        b8x8info->bipred8x8me[mode][0] = 0;
        if (enc_mb.valid[mode])
        {
          for (cost=0, block=0; block<(mode==1?1:2); block++)
          {
            update_lambda_costs(&enc_mb, lambda_mf);
            PartitionMotionSearch (currMB, mode, block, 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(currMB, 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(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
              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;
              }

              // Determine prediction list based on mode cost
              determine_prediction_list(mode, bmcost, best_ref, &best_pdir, &cost, &bipred_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, bipred_me);
            //----- set reference frame and direction parameters -----
            set_block8x8_info(b8x8info, mode, block, best_ref, best_pdir, bipred_me);
            
            
            //--- set reference frames and motion vectors ---
            if (mode>1 && block==0)
              SetRefAndMotionVectors (currMB, block, mode, best_pdir, best_ref[LIST_0], best_ref[LIST_1], bipred_me);
          } // for (block=0; block<(mode==1?1:2); block++)

          if (cost < min_cost)
          {
            best_mode = (short) mode;
            min_cost  = cost;
            if (params->CtxAdptLagrangeMult == 1)
            {
              adjust_mb16x16_cost(cost);
            }
          }
        } // if (enc_mb.valid[mode])
      } // for (mode=1; mode<4; mode++)

    if (enc_mb.valid[P8x8])
      {
        giRDOpt_B8OnlyFlag = 1;

        tr8x8.mb_p8x8_cost = INT_MAX;
        tr4x4.mb_p8x8_cost = INT_MAX;
        //===== store coding state of macroblock =====
        store_coding_state (currMB, cs_mb);

        currMB->all_blk_8x8 = -1;

        if (params->Transform8x8Mode)
        {
          tr8x8.mb_p8x8_cost = 0;
          //===========================================================
          // Check 8x8 partition with transform size 8x8
          //===========================================================
          //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
          for (cost_direct=cbp8x8=cbp_blk8x8=cnt_nonz_8x8=0, block=0; block<4; block++)
          {
            submacroblock_mode_decision(&enc_mb, &tr8x8, currMB, cofAC8x8ts[0][block], cofAC8x8ts[1][block], cofAC8x8ts[2][block],
              &have_direct, bslice, block, &cost_direct, &cost, &cost8x8_direct, 1, is_cavlc);
            set_subblock8x8_info(b8x8info, P8x8, block, &tr8x8);
          }

          // following params could be added in RD_8x8DATA structure
          cbp8_8x8ts      = cbp8x8;
          cbp_blk8_8x8ts  = cbp_blk8x8;
          cnt_nonz8_8x8ts = cnt_nonz_8x8;
          currMB->luma_transform_size_8x8_flag = 0; //switch to 4x4 transform size

          //--- re-set coding state (as it was before 8x8 block coding) ---
          //reset_coding_state (currMB, cs_mb);
        }// if (params->Transform8x8Mode)


        if (params->Transform8x8Mode != 2)
        {
          tr4x4.mb_p8x8_cost = 0;
          //=================================================================
          // Check 8x8, 8x4, 4x8 and 4x4 partitions with transform size 4x4
          //=================================================================
          //=====  LOOP OVER 8x8 SUB-PARTITIONS  (Motion Estimation & Mode Decision) =====
          for (cost_direct=cbp8x8=cbp_blk8x8=cnt_nonz_8x8=0, block=0; block<4; block++)
          {
            submacroblock_mode_decision(&enc_mb, &tr4x4, currMB, cofAC8x8[block], cofAC8x8CbCr[0][block], cofAC8x8CbCr[1][block],
              &have_direct, bslice, block, &cost_direct, &cost, &cost8x8_direct, 0, is_cavlc);
            set_subblock8x8_info(b8x8info, P8x8, block, &tr4x4);
          }
          //--- re-set coding state (as it was before 8x8 block coding) ---
          // reset_coding_state (currMB, cs_mb);
        }// if (params->Transform8x8Mode != 2)

        //--- re-set coding state (as it was before 8x8 block coding) ---
        reset_coding_state (currMB, cs_mb);

        // This is not enabled yet since mpr has reverse order.
        if (params->RCEnable)
          rc_store_diff(img->opix_x, img->opix_y, mb_pred);

        //check cost for P8x8 for non-rdopt mode
        giRDOpt_B8OnlyFlag = 0;
      }
      else // if (enc_mb.valid[P8x8])
      {
        tr4x4.mb_p8x8_cost = INT_MAX;
      }

      // Find a motion vector for the Skip mode
    if(pslice)
        FindSkipModeMotionVector (currMB);
    }
    else // if (!intra)
    {
      min_cost = INT_MAX;
    }

    //========= C H O O S E   B E S T   M A C R O B L O C K   M O D E =========
    //-------------------------------------------------------------------------

   {
     if ((img->yuv_format != YUV400) && !IS_INDEPENDENT(params))
     {
       // precompute all new chroma intra prediction modes
       IntraChromaPrediction(currMB, &mb_available_up, &mb_available_left, &mb_available_up_left);

       if (params->FastCrIntraDecision )
       {
         IntraChromaRDDecision(currMB, enc_mb);
         min_chroma_pred_mode = (short) currMB->c_ipred_mode;
         max_chroma_pred_mode = (short) currMB->c_ipred_mode;
       }
       else
       {
         min_chroma_pred_mode = DC_PRED_8;
         max_chroma_pred_mode = PLANE_8;
       }
     }
     else
     {
       min_chroma_pred_mode = DC_PRED_8;
       max_chroma_pred_mode = DC_PRED_8;
     }

     for (currMB->c_ipred_mode=min_chroma_pred_mode; currMB->c_ipred_mode<=max_chroma_pred_mode; currMB->c_ipred_mode++)
     {
       // bypass if c_ipred_mode is not allowed
       if ( (img->yuv_format != YUV400) &&
         (  ((!intra || !params->IntraDisableInterOnly) && params->ChromaIntraDisable == 1 && currMB->c_ipred_mode!=DC_PRED_8)
         || (currMB->c_ipred_mode == VERT_PRED_8 && !mb_available_up)
         || (currMB->c_ipred_mode == HOR_PRED_8  && !mb_available_left)
         || (currMB->c_ipred_mode == PLANE_8     && (!mb_available_left || !mb_available_up || !mb_available_up_left))))
         continue;

       //===== GET BEST MACROBLOCK MODE =====
       for (ctr16x16=0, index=0; index < max_index; index++)
       {
         mode = mb_mode_table[index];

         if (img->yuv_format != YUV400)
         {
           i16mode = 0;
         }
         //--- for INTER16x16 check all prediction directions ---
         if (mode==1 && bslice)
         {
           update_prediction_for_mode16x16(b8x8info, ctr16x16, &index);
           ctr16x16++;
         }

         // Skip intra modes in inter slices if best mode is inter <P8x8 with cbp equal to 0
         if (params->SkipIntraInInterSlices && !intra && mode >= I4MB && best_mode <=3 && currMB->cbp == 0)
           continue;

      // check if weights are in valid range for biprediction.
         if (bslice && active_pps->weighted_bipred_idc == 1 && mode < P8x8)
         {
           int cur_blk, cur_comp;
           int weight_sum;
           Boolean invalid_mode = FALSE;
           for (cur_blk = 0; cur_blk < 4; cur_blk ++)
           {
             if (b8x8info->best8x8pdir[mode][cur_blk] == 2)
             {
               for (cur_comp = 0; cur_comp < (active_sps->chroma_format_idc == YUV400 ? 1 : 3) ; cur_comp ++)
               {
                 best8x8l0ref = (int) b8x8info->best8x8l0ref[mode][cur_blk];
                 best8x8l1ref = (int) b8x8info->best8x8l1ref[mode][cur_blk];
                 weight_sum = wbp_weight[0][best8x8l0ref][best8x8l1ref][cur_comp] + wbp_weight[1][best8x8l0ref][best8x8l1ref][cur_comp];

                 if (weight_sum < -128 ||  weight_sum > 127)
                 {
                   invalid_mode = TRUE;
                   break;
                 }
               }
               if (invalid_mode == TRUE)
                 break;
             }
           }
           if (invalid_mode == TRUE)
             continue;
         }

         if (enc_mb.valid[mode])
           compute_mode_RD_cost(mode, currMB, &enc_mb, &min_rdcost, &min_dcost, &min_rate, i16mode, bslice, &inter_skip, is_cavlc);

       }// for (ctr16x16=0, index=0; index<max_index; index++)
     }// for (currMB->c_ipred_mode=DC_PRED_8; currMB->c_ipred_mode<=max_chroma_pred_mode; currMB->c_ipred_mode++)

#ifdef BEST_NZ_COEFF
     for (j=0;j<4;j++)
       for (i=0; i<(4+img->num_blk8x8_uv); i++)
         img->nz_coeff[img->current_mb_nr][j][i] = gaaiMBAFF_NZCoeff[j][i];
#endif
   }

   if (rerun==0)
     intra1 = IS_INTRA(currMB);
  } // for (rerun=0; rerun<runs; rerun++)

  //=====  S E T   F I N A L   M A C R O B L O C K   P A R A M E T E R S ======
  //---------------------------------------------------------------------------

  update_qp_cbp_tmp(currMB, cbp, best_mode);
  set_stored_macroblock_parameters (currMB);

  // Rate control
  if(params->RCEnable && params->RCUpdateMode <= MAX_RC_MODE)
    rc_store_mad(currMB);
  update_qp_cbp(currMB, best_mode);

  rdopt->min_rdcost = min_rdcost;
  rdopt->min_dcost = min_dcost;

  if ( (img->MbaffFrameFlag)
    && (img->current_mb_nr%2)
    && (currMB->mb_type ? 0:((bslice) ? !currMB->cbp:1))  // bottom is skip
    && (prevMB->mb_type ? 0:((bslice) ? !prevMB->cbp:1))
    && !(field_flag_inference(currMB) == enc_mb.curr_mb_field)) // top is skip
  {
    rdopt->min_rdcost = 1e30;  // don't allow coding of a MB pair as skip if wrong inference
  }

  //===== Decide if this MB will restrict the reference frames =====
  if (params->RestrictRef)
    update_refresh_map(intra, intra1, currMB);

  if(params->SearchMode == UM_HEX)
  {
    UMHEX_skip_intrabk_SAD(best_mode, listXsize[enc_mb.list_offset[LIST_0]]);
  }
  else if(params->SearchMode == UM_HEX_SIMPLE)
  {
    smpUMHEX_skip_intrabk_SAD(best_mode, listXsize[enc_mb.list_offset[LIST_0]]);
  }

  //--- constrain intra prediction ---
  if(params->UseConstrainedIntraPred && (img->type==P_SLICE || img->type==B_SLICE))
  {
    img->intra_block[img->current_mb_nr] = IS_INTRA(currMB);
  }
}