Sliced data structure for particle-based simulations on GPUs

Abstract

We present a sliced data structure that is effective for use in neighboring particle search for particle-based simulations. In this method, a grid is dynamically constructed to fit to a particle distribution. Rather than computing the grid to fit perfectly to the particle distribution, it compute a grid with some margin to the distribution. This lowers the computation cost of constructing the data structure. Before storing particle indices on a grid, key values which are used to compute the index of a voxel are calculated. The proposed data structure can be introduced into particle-based simulations that run entirely on the Graphics Processing Unit (GPU) because the construction of this data structure and access to storing values can also be also performed entirely on the GPU. The proposed data structure removes the restriction of a computation region with a fixed grid and makes it possible to simulate particle motions over a larger area. Moreover, the cost of the proposed method is low enough for use in real-time applications. In this paper, we first introduce the sliced data structure and then describe its implementation on the GPU. Finally, we apply the propsed method to particle-based simulations and present its quantitative evaluations.