Simulation And Optimization Of Multi-phase Field Based On OpenCL Parallelism

The phase field method can effectively combine the microscopic and macroscopic scales by coupling the phase field with the temperature field,the solute field,the flow field,and other external fields,and has become one of the most effective methods for numerically simulating the crystal growth process.With the application of the phase field method,the phase field model is becoming more and more complex,the calculation scale is becoming larger and larger,the calculation precision demand is higher and higher.However,the computation time required is more and more short.The traditional phase field solution method has become one of the bottlenecks for the development of the phase field method.The high performance calculation method can not only effectively improve the calculation efficiency,but also expand the calculation scale.Moreover,it can be used to hide memory latency by the computation of many threads,so it becomes an effective measure to improve the performance of the program.It is a research hotspot in the field of phase field model in recent years.With the continuous improvement of computing power and programmability,the development of GPU is quite rapid.Application acceleration using GPU on heterogeneous computing systems composed of CPU and GPU has become the main mode to improve program performance.CUDA and OpenCL are the most widely used in GPU programming model,but the CUDA programming model can only be to NVIDIA platform,which is impossible to implement parallel algorithm cross-platform processing.OpenCL is not only supported by many hardware vendors,but also can control the reasonable assignment of hardware,so that each processor can perform its maximum its performance.In this paper,the software and hardware architecture of OpenCL+GPU is discussed in order to solve the multi-field coupling multi-phase field model with large computational complexity,limited computing area and low computational efficiency.The solution of 3D eutectic polyphase field model on a large scale is realized by concurrent execution of multi-process and multi-thread.The main work is as follows:(1)Combining the pure diffusion eutectic growth multiphase field model KKSM with the Lattice Boltzmann method for momentum,mass,and energy transmission,a PF-LBM phase-field model for eutectic growth under forced flow conditions was established.(2)In order to solve the problems existing in traditional multi-field coupled multiphase field model,this paper analyzes the key steps to restrict the computational efficiency of serial algorithms,and analyzes the parallelism of these key steps,and explores a high-performance parallel computing based on OpenCL+GPU.The method realizes the parallel solution of three-dimensional eutectic multiphase field model with coupling flow field.(3)The results of parallel computing and traditional solution method under the same conditions were compared and analyzed to verify the reliability and acceleration performance of the parallel algorithm.Optimizing the algorithm to achieve the best performance,and continuously improve the model.Compared with the serial algorithm,the optimized parallel algorithm achieves a speedup of 24.3 times and 21.6times respectively on different heterogeneous platforms.At the same time,the parallel algorithm based on OpenCL+GPU acquires more accurate results with its powerful floating point computing capability.The simulation results,which achieve the double requirements of computational efficiency and portability,solve the problems such as large amount of calculation,low efficiency,and limited qualitative research in the traditional solution phase field model.