为解决粒子的无序化易引起的图形处理器(GPU)内存访问冲突问题和提高计算效率,通过建立粒子重排序技术提出了一种改进的GPU加速策略.将该加速策略应用于光滑粒子流体动力学(SPH)方法中对三维带障碍物溃坝进行模拟,并与实验结果对比对算法进行验证,获得了较高的计算精度.基于此算例,通过在不同硬件设施上进行模拟分别对粒子重新编号的效果和算法的求解效率比较研究.结果表明,粒子重新编号技术可以保证稳定的单步运行时间,能够有效解决GPU-SPH算法显存访问冲突问题;该GPU加速的SPH并行算法能够大幅提高SPH方法求解效率,随着粒子数量的增加,其大幅缩短计算时间的优势愈发明显,为扩大SPH方法解决三维数值模拟的适用性提供了可能.

In order to solve the problem of graphics processing unit (GPU) memory access conflicts possibly caused by the disorder of particles and enhance the computation efficiency, an improved GPU acceleration strategy is proposed by establishing particle reorder technology. The acceleration strategy is applied to the smoothed particle hydrodynamics (SPH) method to simulate the dam breaking with obstacles in three dimensions, and the algorithm is verified by comparing with the experimental results, which obtained a high calculation accuracy. Based on this benchmark example of the SPH, the studies on the effect of particle renumbering and the solution efficiency of the algorithm are conducted by comparing the simulations of different hardware facilities. The results indicate that the particle reorder technology can ensure a stable single-step running time, and can effectively solve the problem of graphic card memory access conflicts that commonly exist in the GPU-SPH algorithm. Furthermore, the GPU parallel algorithm can greatly improve the solution efficiency of the SPH method, and with the increase of particle number, the advantage of drastically reducing the computation time becomes more obvious. The method proposed in this paper provides the possibility to expand the application of the SPH method to solve 3D numerical simulations.