A Study Of Parallel Acceleration Of GPU In 3-D MPS Method

MPS(Moving Particle Semi-implicit)is very suitable to deal with free surfaces flows of large deformation such as dam-breaking and sloshing flows.However,the computational efficiency of the MPS method becomes a bottleneck for large-scale engineering applications with the increase of the amount of number of particles,especially for 3D flow problems.Based on our in-house solver MLParticle-SJTU,a work about how to use the GPU Paralleling to accelerate three-dimensional MPS methods were studied.Through the study of GPU paralleling and MPS theory,this paper adopts two schemes of GPU paralleling: “partially runned on GPU” and “all runned on GPU”.In the case of “partially runned on GPU”,the present work applies GPU technique in searching neighboring particles and solving Poisson equation,which are two most time-consuming parts in the solver.The acceleration performance is analyzed according to 3D sloshing and 3D dam breaking flows.The results show that a speedup up to 10 x can be achieved by CPU+GPU acceleration compared to the MLParticle-SJTU solver.As for the case of “all runned on GPU”,the present work design a new numerical calculation flow of MPS,with the concerning of GPU programming scheme.A more bigger speedup has been achieved by MPSGPU solver,which the calculation results are in good accord with the experimental results in a case of dam-break with obstacles.Finally,this paper analyzes the advantages and disadvantages of “partially runned on GPU” and “all runned on GPU”,and puts forward some improvement methods.The research of this paper can provide some reference for improving the computational efficiency of 3D MPS.