if ((spiral_hpel_search_x = (short*)calloc(max_search_points, sizeof(short))) == NULL)
    no_mem_exit("Init_Motion_Search_Module: spiral_hpel_search_x");
  if ((spiral_hpel_search_y = (short*)calloc(max_search_points, sizeof(short))) == NULL)
    no_mem_exit("Init_Motion_Search_Module: spiral_hpel_search_y");
  if ((mvbits = (int*)calloc(2*max_mvd+1, sizeof(int))) == NULL)
    no_mem_exit("Init_Motion_Search_Module: mvbits");
  if ((refbits = (int*)calloc(max_ref, sizeof(int))) == NULL)
    no_mem_exit("Init_Motion_Search_Module: refbits");
  if ((byte_abs = (unsigned int*)calloc(byte_abs_range, sizeof(unsigned int))) == NULL)
    no_mem_exit("Init_Motion_Search_Module: byte_abs");
  
  get_mem4Dint (&motion_cost, 8, 2, img->max_num_references, 4);
  
  //--- set array offsets ---
  mvbits   += max_mvd;
  byte_abs += byte_abs_range/2;
  
  
  //=====   INIT ARRAYS   =====
  //---------------------------
  //--- init array: motion vector bits ---
  mvbits[0] = 1;
  for (bits=3; bits<=max_mv_bits; bits+=2)
  {
    imax = 1    << (bits >> 1);
    imin = imax >> 1;
    
    for (i = imin; i < imax; i++)   mvbits[-i] = mvbits[i] = bits;
  }
  //--- init array: reference frame bits ---
  refbits[0] = 1;
  for (bits=3; bits<=max_ref_bits; bits+=2)
  {
    imax = (1   << ((bits >> 1) + 1)) - 1;
    imin = imax >> 1;
    
    for (i = imin; i < imax; i++)   refbits[i] = bits;
  }
  //--- init array: absolute value ---
  byte_abs[0] = 0;
  for (i=1; i<byte_abs_range/2; i++)   byte_abs[i] = byte_abs[-i] = i;
  //--- init array: search pattern ---
  spiral_search_x[0] = spiral_search_y[0] = 0;
  spiral_hpel_search_x[0] = spiral_hpel_search_y[0] = 0;
  for (k=1, l=1; l<=max(1,search_range); l++)
  {
    for (i=-l+1; i< l; i++)
    {
      spiral_search_x[k] =  i;  spiral_search_y[k] = -l;
      spiral_hpel_search_x[k] =  i<<1;  spiral_hpel_search_y[k++] = -l<<1;
      spiral_search_x[k] =  i;  spiral_search_y[k] =  l;      
      spiral_hpel_search_x[k] =  i<<1;  spiral_hpel_search_y[k++] =  l<<1;
    }
    for (i=-l;   i<=l; i++)
    {
      spiral_search_x[k] = -l;  spiral_search_y[k] =  i;
      spiral_hpel_search_x[k] = -l<<1;  spiral_hpel_search_y[k++] = i<<1;
      spiral_search_x[k] =  l;  spiral_search_y[k] =  i;      
      spiral_hpel_search_x[k] =  l<<1;  spiral_hpel_search_y[k++] = i<<1;
    }
  }
  
#ifdef _FAST_FULL_ME_
  if(!input->FMEnable)
    InitializeFastFullIntegerSearch ();
#endif
}


/*!
 ************************************************************************
 * \brief
 *    Free memory used by motion search
 ************************************************************************
 */
void
Clear_Motion_Search_Module ()
{
  //--- correct array offset ---
  mvbits   -= max_mvd;
  byte_abs -= byte_abs_range/2;
  
  //--- delete arrays ---
  free (spiral_search_x);
  free (spiral_search_y);
  free (spiral_hpel_search_x);
  free (spiral_hpel_search_y);
  free (mvbits);
  free (refbits);
  free (byte_abs);
  free_mem4Dint (motion_cost, 8, 2);
  
#ifdef _FAST_FULL_ME_
  if(!input->FMEnable)
    ClearFastFullIntegerSearch ();
#endif
}



/*!
 ***********************************************************************
 * \brief
 *    Full pixel block motion search
 ***********************************************************************
 */
int                                               //  ==> minimum motion cost after search
FullPelBlockMotionSearch (pel_t**   orig_pic,     // <--  original pixel values for the AxB block
                          short     ref,          // <--  reference frame (0... or -1 (backward))
                          int       list,         // <--  current list
                          int       pic_pix_x,    // <--  absolute x-coordinate of regarded AxB block
                          int       pic_pix_y,    // <--  absolute y-coordinate of regarded AxB block
                          int       blocktype,    // <--  block type (1-16x16 ... 7-4x4)
                          short     pred_mv_x,    // <--  motion vector predictor (x) in sub-pel units
                          short     pred_mv_y,    // <--  motion vector predictor (y) in sub-pel units
                          short*    mv_x,         // <--> in: search center (x) / out: motion vector (x) - in pel units
                          short*    mv_y,         // <--> in: search center (y) / out: motion vector (y) - in pel units
                          int       search_range, // <--  1-d search range in pel units
                          int       min_mcost,    // <--  minimum motion cost (cost for center or huge value)
                          int       lambda_factor)       // <--  lagrangian parameter for determining motion cost
{
  int   pos, cand_x, cand_y, y, x4, mcost;
  
  pel_t *orig_line, *ref_line;
  pel_t *(*get_ref_line)(int, pel_t*, int, int, int, int);
  
  int   list_offset   = img->mb_data[img->current_mb_nr].list_offset; 
  pel_t *ref_pic      = listX[list+list_offset][ref]->imgY_11;
  int   img_width     = listX[list+list_offset][ref]->size_x;
  int   img_height    = listX[list+list_offset][ref]->size_y;
  
  int   best_pos      = 0;                                        // position with minimum motion cost
  int   max_pos       = (2*search_range+1)*(2*search_range+1);    // number of search positions
  int   blocksize_y   = input->blc_size[blocktype][1];            // vertical block size
  int   blocksize_x   = input->blc_size[blocktype][0];            // horizontal block size
  int   blocksize_x4  = blocksize_x >> 2;                         // horizontal block size in 4-pel units
  int   pred_x        = (pic_pix_x << 2) + pred_mv_x;       // predicted position x (in sub-pel units)
  int   pred_y        = (pic_pix_y << 2) + pred_mv_y;       // predicted position y (in sub-pel units)
  int   center_x      = pic_pix_x + *mv_x;                        // center position x (in pel units)
  int   center_y      = pic_pix_y + *mv_y;                        // center position y (in pel units)
  int   check_for_00  = (blocktype==1 && !input->rdopt && img->type!=B_SLICE && ref==0);
  
  int  apply_weights = ( (active_pps->weighted_pred_flag && (img->type == P_SLICE || img->type == SP_SLICE)) ||
    (active_pps->weighted_bipred_idc && (img->type == B_SLICE)));  
  
  //===== Use weighted Reference for ME ====
  if (apply_weights && input->UseWeightedReferenceME)
    ref_pic       = listX[list+list_offset][ref]->imgY_11_w;
  else
    ref_pic       = listX[list+list_offset][ref]->imgY_11;
  
  //===== set function for getting reference picture lines =====
  if ((center_x > search_range) && (center_x < img_width -1-search_range-blocksize_x) &&
    (center_y > search_range) && (center_y < img_height-1-search_range-blocksize_y)   )
  {
    get_ref_line = FastLineX;
  }
  else
  {
    get_ref_line = UMVLineX;
  }
  
  
  //===== loop over all search positions =====
  for (pos=0; pos<max_pos; pos++)
  {
    //--- set candidate position (absolute position in pel units) ---
    cand_x = center_x + spiral_search_x[pos];
    cand_y = center_y + spiral_search_y[pos];
    
    //--- initialize motion cost (cost for motion vector) and check ---
    mcost = MV_COST (lambda_factor, 2, cand_x, cand_y, pred_x, pred_y);
    if (check_for_00 && cand_x==pic_pix_x && cand_y==pic_pix_y)
    {
      mcost -= WEIGHTED_COST (lambda_factor, 16);
    }
    if (mcost >= min_mcost)   continue;
    
    //--- add residual cost to motion cost ---
    for (y=0; y<blocksize_y; y++)
    {
      ref_line  = get_ref_line (blocksize_x, ref_pic, cand_y+y, cand_x, img_height, img_width);
      orig_line = orig_pic [y];
      
      for (x4=0; x4<blocksize_x4; x4++)
      {
        mcost += byte_abs[ *orig_line++ - *ref_line++ ];
        mcost += byte_abs[ *orig_line++ - *ref_line++ ];
        mcost += byte_abs[ *orig_line++ - *ref_line++ ];
        mcost += byte_abs[ *orig_line++ - *ref_line++ ];
      }
      
      if (mcost >= min_mcost)
      {
        break;
      }
    }
    
    //--- check if motion cost is less than minimum cost ---
    if (mcost < min_mcost)
    {
      best_pos  = pos;
      min_mcost = mcost;
    }
  }
  
  
  //===== set best motion vector and return minimum motion cost =====
  if (best_pos)
  {
    *mv_x += spiral_search_x[best_pos];
    *mv_y += spiral_search_y[best_pos];
  }
  return min_mcost;
}
                          
                          
#ifdef _FAST_FULL_ME_
/*!
 ***********************************************************************
 * \brief
 *    Fast Full pixel block motion search
 ***********************************************************************
 */
int                                                   //  ==> minimum motion cost after search
FastFullPelBlockMotionSearch (pel_t**   orig_pic,     // <--  not used
                              short     ref,          // <--  reference frame (0... or -1 (backward))
                              int       list,
                              int       pic_pix_x,    // <--  absolute x-coordinate of regarded AxB block
                              int       pic_pix_y,    // <--  absolute y-coordinate of regarded AxB block
                              int       blocktype,    // <--  block type (1-16x16 ... 7-4x4)
                              short     pred_mv_x,    // <--  motion vector predictor (x) in sub-pel units
                              short     pred_mv_y,    // <--  motion vector predictor (y) in sub-pel units
                              short*    mv_x,         //  --> motion vector (x) - in pel units
                              short*    mv_y,         //  --> motion vector (y) - in pel units
                              int       search_range, // <--  1-d search range in pel units
                              int       min_mcost,    // <--  minimum motion cost (cost for center or huge value)
                              int       lambda_factor)       // <--  lagrangian parameter for determining motion cost
{
  int   pos, offset_x, offset_y, cand_x, cand_y, mcost;
  
  int   max_pos       = (2*search_range+1)*(2*search_range+1);              // number of search positions
  int   best_pos      = 0;                                                  // position with minimum motion cost
  int   block_index;                                                        // block index for indexing SAD array
  distpel*  block_sad;                                                          // pointer to SAD array
  
  block_index   = (pic_pix_y-img->opix_y)+((pic_pix_x-img->opix_x)>>2); // block index for indexing SAD array
  block_sad     = BlockSAD[list][ref][blocktype][block_index];         // pointer to SAD array
  
  //===== set up fast full integer search if needed / set search center =====
  if (!search_setup_done[list][ref])
  {
    SetupFastFullPelSearch (ref, list);
  }
  
  offset_x = search_center_x[list][ref] - img->opix_x;
  offset_y = search_center_y[list][ref] - img->opix_y;
  
  //===== cost for (0,0)-vector: it is done before, because MVCost can be negative =====
  if (!input->rdopt)
  {
    mcost = block_sad[pos_00[list][ref]] + MV_COST (lambda_factor, 2, 0, 0, pred_mv_x, pred_mv_y);
    
    if (mcost < min_mcost)
    {
      min_mcost = mcost;
      best_pos  = pos_00[list][ref];
    }
  }
  
  //===== loop over all search positions =====
  for (pos=0; pos<max_pos; pos++, block_sad++)
  {
    //--- check residual cost ---
    if (*block_sad < min_mcost)
    {
      //--- get motion vector cost ---
      cand_x = offset_x + spiral_search_x[pos];
      cand_y = offset_y + spiral_search_y[pos];
      mcost  = *block_sad;
      mcost += MV_COST (lambda_factor, 2, cand_x, cand_y, pred_mv_x, pred_mv_y);
      
      //--- check motion cost ---
      if (mcost < min_mcost)
      {
        min_mcost = mcost;
        best_pos  = pos;
      }
    }
  }
  
  //===== set best motion vector and return minimum motion cost =====
  *mv_x = offset_x + spiral_search_x[best_pos];
  *mv_y = offset_y + spiral_search_y[best_pos];
  return min_mcost;
}
#endif


/*!
 ***********************************************************************
 * \brief
 *    Calculate SA(T)D
 ***********************************************************************
 */
int
SATD (int* diff, int use_hadamard)
{
  int k, satd = 0, m[16], d[16];
  
  if (use_hadamard)
  {
    /*===== hadamard transform =====*/
    m[ 0] = diff[ 0] + diff[12];
    m[ 1] = diff[ 1] + diff[13];
    m[ 2] = diff[ 2] + diff[14];
    m[ 3] = diff[ 3] + diff[15];
    m[ 4] = diff[ 4] + diff[ 8];
    m[ 5] = diff[ 5] + diff[ 9];
    m[ 6] = diff[ 6] + diff[10];
    m[ 7] = diff[ 7] + diff[11];
    m[ 8] = diff[ 4] - diff[ 8];
    m[ 9] = diff[ 5] - diff[ 9];
    m[10] = diff[ 6] - diff[10];
    m[11] = diff[ 7] - diff[11];
    m[12] = diff[ 0] - diff[12];    
    m[13] = diff[ 1] - diff[13];        
    m[14] = diff[ 2] - diff[14];            
    m[15] = diff[ 3] - diff[15];
    
    d[ 0] = m[ 0] + m[ 4];
    d[ 1] = m[ 1] + m[ 5];
    d[ 2] = m[ 2] + m[ 6];
    d[ 3] = m[ 3] + m[ 7];
    d[ 4] = m[ 8] + m[12];
    d[ 5] = m[ 9] + m[13];
    d[ 6] = m[10] + m[14];
    d[ 7] = m[11] + m[15];
    d[ 8] = m[ 0] - m[ 4];
    d[ 9] = m[ 1] - m[ 5];
    d[10] = m[ 2] - m[ 6];
    d[11] = m[ 3] - m[ 7];
    d[12] = m[12] - m[ 8];
    d[13] = m[13] - m[ 9];
    d[14] = m[14] - m[10];
    d[15] = m[15] - m[11];
    
    m[ 0] = d[ 0] + d[ 3];
    m[ 1] = d[ 1] + d[ 2];
    m[ 2] = d[ 1] - d[ 2];
    m[ 3] = d[ 0] - d[ 3];
    m[ 4] = d[ 4] + d[ 7];
    m[ 5] = d[ 5] + d[ 6];
    m[ 6] = d[ 5] - d[ 6];
    m[ 7] = d[ 4] - d[ 7];
    m[ 8] = d[ 8] + d[11];
    m[ 9] = d[ 9] + d[10];
    m[10] = d[ 9] - d[10];
    m[11] = d[ 8] - d[11];
    m[12] = d[12] + d[15];
    m[13] = d[13] + d[14];
    m[14] = d[13] - d[14];
    m[15] = d[12] - d[15];