摘要
由申威众核处理器组成的"神威·太湖之光"是我国当前性能最高的超级计算机,可为大规模NSGA-Ⅱ求解提供硬件平台。基于硬件架构特点,设计了分岛/主从增强混合并行NSGA-Ⅱ。在主从模式基础上,利用从核间寄存器通信,实现核组内从核局部数据存储的共享。优化流程,实现更多算法模块在从核上的并行。运用DMA传输、向量化、双缓冲、存储优化等方法显著提高加速比。实验表明,优化的并行NSGA-Ⅱ在申威众核处理器上具有良好的加速比和扩展性。
The Sunway TaihuLight,which is composed of Sunway many-core processors,is currently the highest performance supercomputer in China.It can provide a hardware platform for NSGA-Ⅱto solve the large-scale problems.Considering the architecture of Sunway many-core processor,this paper designed an island combined with enhanced master-slave hybrid parallel NSGA-Ⅱalgorithm.Based on the master-slave mode,it used register communication to realize the sharing of local data memory of CPE in a core group.It optimized the algorithm process and parallelized more algorithm modules on CPE.By means of DMA transmission,vectorization,double buffering and storage optimization,it significantly increased the speedup.The experiments show that the optimized parallel NSGA-Ⅱhas good speedup and scalability on the Sunway many-core processors.
作者
刘垚
郑琳
郑凯
王肃
廖启丹
Liu Yao;Zheng Lin;Zheng Kai;Wang Su;Liao Qidan(School of Data Science&Engineering,East China Normal University,Shanghai 200062,China;State Key Laboratory of Mathematical Engineering&Advanced Computing,Wuxi Jiangsu 214215,China)
出处
《计算机应用研究》
CSCD
北大核心
2020年第1期96-101,共6页
Application Research of Computers
基金
数学工程与先进计算国家重点实验室开放基金资助项目(2016A05).
关键词
申威众核处理器
NSGA-Ⅱ
并行遗传算法
多目标
并行优化
Sunway many-core processor
NSGA-Ⅱ
parallel genetic algorithm
multi-objective
parallel optimization
