Research On Hybrid Parallel Framework For SPH Algorithms

Smoothed Particle Hydrodynamics(SPH)is a popular simulation method,which discretizes simulation space with particles.In order to realize large-scale simulation with more details,a large number of particles are necessary.With the increase of particle number,the overhead of algorithm will be increased.This paper focuses on the research of SPH algorithm performance and devises an efficient framework,which can achieve great performance improvement without any compromise of numerical accuracy.GPU based SPH performs better than the equivalent implementation on CPU.The task scheduling method of shared memory breaks the performance bottleneck for parallel SPH algorithm.However,such method involves the overload of neighbor particles.This paper proposes a novel task scheduling strategy,which can efficiently reduce the overload of neighbor particles.Moreover,a new hash coding method is designed to reduce the waste of computational resource.In the process of performing neighbor traversal,uniform grid involves a large number of irrelevant particles.The simple subdivision of simulation space involves the more overhead of loop iterations,which greatly affect the actual efficiency of SPH algorithm.This paper proposes a novel hierarchical grid method associated with a novel hierarchical neighbor search method,which fully considers the continuity of the neighbor particles and avoids the large overhead of loop iterations.Based on the hierarchical strategies of this paper,a novel parallel framework is constructed,and a variety of SPH algorithms are integrated into this framework,so as to perform a variety of physical simulations.The performance tests show that this framework can achieve about 80% improvement in real physical simulations compared with state-of-the-art frameworks.