Research Of Space Plasma Particle Simulation Algorithm Based On Heterogeneous Computing Platforms

Space plasma particle simulation usually requires massive computing resources and brings a strong demand on computing performance. PIC(particle-in-cell) plasma simulation is an important physical simulation algorithm which divides a large number of plasma particles into a grid to calculate the properties of a particle system. During last decade, the heterogeneous high performance parallel computers usually used the graphics processor(GPU) and Intel Many Integrated Core Architecture(MIC Many-Integerated-Core) Coprocessor. And more and more large-scale scientific computing accelerated the calculations based on heterogeneous high performance computers. Thus, how to effectively parallelize PIC simulation is facing a series of new challenges, which is one hot topic of parallel computing research and applications.In this paper, we first achieved parallel PIC plasma simulation. And then extended to the heterogeneous parallel computing platform used of GPU and MIC. For the GPU and MIC computing platforms we were achieved parallel PIC simulation and optimization. On GPU platform it mainly used asynchronous streaming, thread blocks optimization, memory access optimization and etc to optimization. On MIC platform it mainly used thread extended, SIMD instructions and etc to optimization. In order to reflect the effect of parallelized speedup on heterogeneous parallel computing platform, we have compared test results with those from INTEL Ivy-Bridge CPU processor, NVIDIA graphics processors(GPU) and Intel Knight Corner coprocessor. When the grid size of PIC simulations was smaller, GPU can get higher speedup than MIC. Our simulation results suggest that heterogeneous parallel platforms have good computing performance and lower cost. And the GPU platform can be used to space data visualization. The heterogeneous parallel computing research will provide a positive way of thinking on high performance computing systems and its development.