GPU accelerated MPS method for large-scale 3-D violent free surface flows

In this paper, GPU acceleration technique is applied to overcome the low computational efficiency of MPS. Based on modified MPS and GPU technique, an in-house solver MPSGPU-SJTU is developed and used to simulate large-scale three dimensional (3-D) violent flow problems with over two million particles. In order to save the GPU memory, a suitable algorithm of creating neighbor particles list is adopted in GPU code. In addition, a new optimizing strategy for solving pressure Poisson equation (PPE) is proposed to reduce the storage and improve the computational efficiency. The present solver is used to simulate a benchmark test of dam-break flow with an obstacle. The performance of optimized PPE is firstly verified. Comparing the results of three spatial resolutions, the convergence of present solver is also validated. The calculated results like fluid field, impact pressure on obstacle and wave heights are in good agreements with the experimental data and the results of other numerical researches. It shows that the present solver can effectively simulate the large deformation and nonlinear fragmentation of free surface. Then, this solver is used to model another dam-break flow. A more complex violent flow with moving boundary, water entry of rigid circular cylinder, is also simulated. The calculated fluid fields agree well with the experimental images. Good agreements for the movement of cylinder are achieved to show the feasibility of MPSGPU-SJTU solver on the moving boundary problem. Furthermore, the comparisons of computation times between GPU and CPU codes are conducted in all simulations. The results show that the computation time of MPS is significantly reduced by applying GPU acceleration technique.