Simulation Of Eutectic Growth Using Phase Filed Method And Lbm Based On OpenCL

The purpose of studying high-performance computing is to reduce the simulation time and increase the number of trials within a limited time.High-performance computing promotes the development of various other fields including engineering.Phase-field method(PFM)is the most effective method to simulate the microstructure of eutectic alloy.The phase field method adopts a unified control equation not only to avoid distinguishing and tracking the solid / liquid phase and its interface,but also to couple external fields such as flow field and solute field into the phase field.However,the established phase field model is complicated,and the calculation scale and calculation amount are large,which has become one of the bottlenecks in the development of the phase field method.Lattice Boltzmann Method(LBM)is a numerical method that can effectively simulate fluids.It has been proved to be able to simulate the behavior of various fluids including gas and liquid.LBM is suitable for parallel computing and is widely used in various fields and work including simulated multi-phase flow and multi-empty medium flow.In this paper,a phase field-lattice Boltzmann model(PF-LBM)is established based on the coupling of the phase field model and the lattice Boltzmann model of the calculated fluid.The PF-LBM three-dimensional model solves the problem of huge calculation amount and small simulation area,which cannot solve the problem in a reasonable time.GPU computing provides strong support for solving the PF-LBM model.Based on OpenCL's open and unified programming model,it can realize the advantages of accelerated algorithms on heterogeneous systems and give full play to the computing power of GPU.The high-performance parallel computing method based on GPU-OpenCL can solve the PF-LBM three-dimensional model.The main work and conclusions are as follows:(1)The PF-LBM model can be analyzed in parallel.Analyze the factors that affect the performance of the GPU-OpenCL parallel computing method.For the problem of large data transfer overhead between the GPU and CPU,the PF-LBM model is deployed on a single GPU for calculation.On a single high-performance GPU,the result of the GPU-OpenCL parallel calculation method is consistent with the classical theoretical value,and a performance gain of 12.9X is obtained,which proves that the GPU-OpenCL parallel method is effective in solving the PF-LBM model.(2)Due to the limitation of GPU memory for reading data from GPU,the optimization scheme of the kernel with continuous access to GPU memory is designed to improve the calculation rate.The thread block optimization method of Halo Therad is used to obtain the best GPU-OpenCL program and obtain the maximum performance gain.Compared with the serial CPU code implementation,the GPU-OpenCL program has a maximum acceleration ratio of 136.X times,which improves the calculation efficiency by two orders of magnitude.Solve the problem that the numerical simulation of the PF-LBM three-dimensional model cannot be carried out on a large scale due to the limitation of computing power.(3)On a single GPU,the GPU-OpenCL parallel calculation method doubles the calculation scale of the PF-LBM model,and the calculation scale is ??17031474.In the larger-scale PF-LBM model,the GPU-OpenCL calculation model not only has better calculation efficiency,but also obtains new experimental results: the liquid flow drastically changes the eutectic structure morphology,when the eutectic layer spacing is large The liquid annular flow is obvious,and no heterogeneous nucleation is found at the front of the interface with larger volume.The annular liquid flow has a greater influence on the growth rate of the two sides of the ? phase.