A Data-Parallel Adaptive N-body Method

Abstract

We present a data-parallel formulation of an adaptive version of Anderson's method for N-body particle interactions. Our formulation consists of a storage and computationally efficient array-based representation for the nonuniform hierarchy that models arbitrary particle distributions for the computational procedure. We also present data-parallel implementations (in HPF) of several well known partitioning methods. These partitioning methods balance nodal weights (for computation). We present preliminary experimental results for these partitioning schemes and discuss the complete code for adaptive particle simulations.