OpenMP订购的问题

Question

我不需要并行化该功能

double relax_gauss (double *u, unsigned sizex, unsigned sizey)
{

  double unew, diff, sum=0.0;
  int chunkx = sizex/omp_get_num_threads();
  int chunky = sizey/omp_get_num_threads();
  int howmany=4;
  #pragma omp parallel for reduction(+:sum) ordered(2)
    for(int bx = 0; bx < sizex/chunkx; ++bx){
      for(int by = 0; by < sizey/chunky; by++){
        int i_start = bx*chunkx;
        int i_end = (chunkx*(1+bx));
        int j_start = by*chunky;
        int j_end = (chunky*(1+by));
        #pragma omp ordered depend(sink:bx,by-1) depend(sink:bx-1, by)
        {
          for (int i=max(1, i_start); i<= min(sizex-2, i_end); i++) {
          for (int j=max(1, j_start); j<= min(sizey-2, j_end); j++) {
            unew= 0.25 * ( u[ i*sizey + (j-1) ]+    // left
                           u[ i*sizey + (j+1) ]+    // right
                           u[ (i-1)*sizey + j ]+    // top
                           u[ (i+1)*sizey   + j ]);   // bottom
            diff = unew - u[i*sizey+ j];
            sum += diff * diff; 
            u[i*sizey+j]=unew;
          }
        }
      }
    }
  } 
  return sum;
}

当我执行代码时，生成的时间与顺序版本的时间相同。问题一定在这里，因为这是唯一修改的功能。