摘要
大规模振子经耦合所形成的网络广泛存在于自然界的众多领域。由于存在高维非线性系统,研究网络的群体行为,是对其进行设计控制和应用的前提。但是,对耦合振子网络群体构造的特定高维系统的研究往往涉及的大规模数值计算,算法如果仅基于CPU的计算平台往往会存在效率低的问题。为解决上述问题,提出了一种基于CPU+GPU平台的异构并行算法。算法首先设计了高效GPU内核程序,然后借助GPU加速,相对基于仅CPU的平台,效率提高了近30倍,实现了耦合振子网络的高效仿真。
It is a wide range of existence of networks formed by coupled large - scale oscillators in the nature. The study of the group behavior of the networks is a premise that carries on designing, controlling and putting into application. However, the problem of massive numerical calculation and the low efficiency of the computing platform based on CPU are often involved in the research of the high - dimension system built by the coupled oscillator network groups. In order to solve the problems mentioned above, the paper first presented a heterogeneous parallel algorism based on CPU&GPU platform and developed a high efficient GPU kernel program. Then by adding GPU into the plat-form, the efficiency of the new platform was increased nearly 30 times compare to the CPU platform. Finally, the high efficient simulation of the coupled oscillator network was implemented.
出处
《计算机仿真》
CSCD
北大核心
2013年第12期219-223,共5页
Computer Simulation
基金
国家自然科学基金资助项目(61104150
10972082)
关键词
耦合振子网络
异构平台
并行计算
Coupled oscillator network
Heterogeneous platform
Parallel computing