How will the fast multipole method fare in the exascale era?