OpenMP优化比顺序优化慢

Question

我是OpenMP的新手，我正致力于代码优化工作。 以下功能

int accelerate_flow(const t_param params, t_speed* cells, int* obstacles)
{
  int ii,jj;     /* generic counters */
  double w1,w2;  /* weighting factors */

  /* compute weighting factors */
  w1 = params.density * params.accel / 9.0;
  w2 = params.density * params.accel / 36.0;

  /* modify the 2nd row of the grid */
  ii=params.ny - 2;
  for(jj=0;jj<params.nx;jj++) {
    /* if the cell is not occupied and
    ** we don't send a density negative */
    if( !obstacles[ii*params.nx + jj] && 
    (cells[ii*params.nx + jj].speeds[3] - w1) > 0.0 &&
    (cells[ii*params.nx + jj].speeds[6] - w2) > 0.0 &&
    (cells[ii*params.nx + jj].speeds[7] - w2) > 0.0 ) {
      /* increase 'east-side' densities */
      cells[ii*params.nx + jj].speeds[1] += w1;
      cells[ii*params.nx + jj].speeds[5] += w2;
      cells[ii*params.nx + jj].speeds[8] += w2;
      /* decrease 'west-side' densities */
      cells[ii*params.nx + jj].speeds[3] -= w1;
      cells[ii*params.nx + jj].speeds[6] -= w2;
      cells[ii*params.nx + jj].speeds[7] -= w2;
    }
  }

  return EXIT_SUCCESS;
}

这是原点，当我想像并行

那样优化for循环时

#pragma omp parallel for private(jj) shared(cells)
for(jj =0<jj<params.nx,jj++)
.....

它变得非常慢。我不知道如何更好地并行化。