基于GPU的三维扩散方程在反应堆计算中的应用

Research on three dimensional diffusion equation in the reactor calculation application based on GPU

本文介绍了基于统一计算设备架构(Compute Unified Device Architecture,CUDA)的图形处理器(Graphic Processing Unit,GPU)计算环境在钍基熔盐堆(Thorium Molten Salt Reactor,TMSR)设计平台的建立,并将反应堆球场计算软件SRAC(Structure Research and Analysic Corporation)的中子三维扩散计算模块移植到GPU上进行测试及结果验证。采用中心点差分方法推导出三维扩散计算的差分方程,并用超松弛迭代法(Successive Over Relaxation Method,SOR)求解φ,研究了SOR迭代算法的并行实现过程。结果表明,移植的GPU模块部分计算正确,计算速度得到有效提升,验证了TMSR设计平台在GPU计算环境下可正常工作。

Background： With the development of the current nuclear reactor system, some characteristics such as complex structure, material variety and diverse spectrum of reactor cores need more and more computation consumption for reactor physics. Graphic Processing Unit （GPU） computing technology is widely used on the reactor calculation due to its rapid development. Purpose： This paper illustrates the GPU computing environment based on Compute Unified Device Architecture （CUDA） architecture, which is established on the design platform of Thorium Molten Salt Reactor （TMSR）. And the neutron three-dimensional diffusion calculation module of the reactor would be transplanted into the GPU for testing and verification. Methods： First of all, center difference method was used to deduce the difference equations of three dimensional diffusion calculations, make it suitable for GPU. Then, the over relaxation iteration method was applied to solve these difference equations. Finally, the over relaxation iteration algorithm was studied in the parallel implementation of GPU processes. Results： The results show that the transplantation of GPU computing module section can be used to obtain correct output, and it can effectively improve the computing speed. Conclusion： The GPU computing environment on the TMSR design platform can work normally by carefully transplanting of several key methods for difference equations based on the CUDA of GPU.