GPU加速MOC输运计算性能分析研究  

Performance Analysis on Acceleration of Transport Calculation with Method of Characteristics Based on GPU

作　　者：宋佩涛 张志俭 梁亮 张乾 赵强 SONG Peitao;ZHANG Zhijian;LIANG Liang;ZHANG Qian;ZHAO Qiang(Fundamental Science on Nuclear Safety and Simulation Technology Laboratory,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学核安全与仿真技术国防重点学科实验室

出　　处：《原子能科学技术》2020年第1期103-111,共9页Atomic Energy Science and Technology

基　　金：核反应堆系统设计技术重点实验室运行基金资助项目;黑龙江省青年科学基金资助项目(QC2018003);数字化反应堆工程化关键技术研究资助项目(J121217001)

摘　　要：特征线方法(MOC)在求解堆芯规模中子输运方程时面临计算时间长的问题,加速和并行算法是目前研究的热点。基于MOC在特征线和能群层面的并行特性,采用统一计算设备构架(CUDA)编程规范,实现了基于图形处理器(GPU)的并行二维MOC算法。测试了菱形差分和步特征线法分别在双精度、混合精度及单精度浮点运算下的计算精度、效率及GPU加速效果。采用性能分析工具对GPU程序性能进行了分析,识别了程序性能瓶颈。结果表明:菱形差分和步特征线法在不同浮点运算精度下均表现出良好的计算精度;相比于CPU单线程计算,GPU加速效果在双精度和单精度情况下分别达到35倍和100倍以上。The method of characteristics(MOC) consumes more computing time when solving the neutron transport equation with the configuration of practical reactor cores. As a result, researches are focused on the acceleration techniques and the parallel algorithms. Based on the parallelism of characteristic rays and energy groups, the GPU-accelerated parallel 2 D MOC algorithm was implemented with the compute unified device architecture(CUDA). The code accuracy and efficiency were tested in the diamond difference scheme and the step characteristics scheme with single-precision, mixed-precision and double-precision floating-point operation. Meanwhile, the performance bottleneck of GPU application was analyzed by utilizing the NVIDIA profiling tool. The numerical results demonstrate that the parallel algorithm maintains the desired accuracy for the diamond difference scheme and the step characteristics scheme in all selected floating-point precision conditions. In addition, the GPU-based code is 35 times and 100 times faster than the CPU-based code in double-precision and single-precision, respectively.

关 键 词：GPU加速 特征线方法 中子输运计算 统一计算设备构架 性能分析 

分 类 号：TL329[核科学技术—核技术及应用]