先进中子学程序SCAP-N混合并行方法  

Hybrid Parallel Method for Advanced Neutronics Code SCAP-N

作　　者：施骏 彭良辉 杨波[1] 毕光文[1] 杨伟焱[1] SHI Jun;PENG Lianghui;YANG Bo;BI Guangwen;YANG Weiyan(Shanghai Nuclear Engineering Research&Design Institute Co.,Ltd.,Shanghai 200233,China)

机构地区：[1]上海核工程研究设计院股份有限公司,上海200233

出　　处：《现代应用物理》2024年第3期30-35,共6页Modern Applied Physics

基　　金：国家重点研发计划资助项目(2022YFB1902200)。

摘　　要：为提高先进中子学程序SCAP-N的计算效率,提出并实现了1种MPI/OpenMP混合并行方案。采用MPI分布式并行模式,实现各分层2维堆芯计算间的并行及3维堆芯pin-by-pin输运计算的粗粒度区域分解并行。采用OpenMP共享存储式并行模式,实现分层2维堆芯内部非均匀输运计算的并行及3维堆芯pin-by-pin输运计算各MPI分区内部的细粒度并行。数值结果表明,该混合并行方法可有效实现SCAP-N程序的并行化,显著缩短程序计算时间,减少堆芯高精度中子输运计算区域分解数目,缓解源迭代格式退化,提高程序计算稳定性和收敛效率。An MPI/OpenMP hybrid parallel method is proposed and implemented in the advanced neutronics code SCAP-N to improve its computational efficiency.The MPI distributed parallel method is adopted to realize parallelism between the layered two-dimensional core simulations,as well as the coarse-grained region decomposition parallelism of the three-dimensional core pin-by-pin transport calculations.The OpenMP shared storage parallel method is used to achieve parallelism of non-uniform transport calculations within each two-dimensional core,as well as fine-grained parallelism of pin-by-pin transport calculations within each MPI decomposed region of the three-dimensional core.The numerical results show that the MPI/OpenMP hybrid parallel method can be successfully implemented in the SCAP-N code.Compared to the pure MPI method,the hybrid method significantly reduces the computational time,decreases the number of domain decompositions in neutron transport calculations,alleviates the degradation of source iteration format,and improves the program's stability and efficiency of convergence.

关 键 词：非均匀输运 逐棒输运 高性能计算 混合并行 

分 类 号：TL329[核科学技术—核技术及应用] O571[理学—粒子物理与原子核物理]