基于GPU的近场动力学模拟的并行化方法  

Parallel Computing Method of Peridynamic Models Based on GPU

作　　者：刘肃肃 胡祎乐[1] 余音[1] 

机构地区：[1]上海交通大学航空航天学院,上海200240

出　　处：《上海交通大学学报》2016年第9期1362-1367,1375,共7页Journal of Shanghai Jiaotong University

摘　　要：针对近场动力学（PD）模型计算量庞大、计算效率低的问题，研究了基于OPU的PD建模过程的并行化方法．通过将前处理及求解过程中各物质点的计算映射到GPU的单个线程，实现模型的多线程高效并行计算．计算程序在Microsoft Visual Studio2010集合CUDA4．0工具包的开发环境下执行，采用CUDAC进行编写．对复合材料层压板的2种PD模型（键型PD模型和态型PD模型）的计算表明，采用GPU并行运算的模型模拟结果与试验结果吻合良好，并且相比于CPU串行计算，GPU并行计算获得了2．6～10．3倍的加速比，降低了PD模型的计算代价．In order to improve the low computation efficiency of Peridynamic models, a parallel computing method based on GPU was proposed. A high computing efficiency was realized by dispatching the calcula- tion of each node to each GPU thread both in the pre-processing and sovling period. Under the integrated environment of Visual Studio 2010 and CUDA 4.0 toolkit, the corresponding numerical programs based on GPU parallel computing techniques were executed. The programs were coded by CUDA C. The bond- based peridynamic and state-based peridynamic models of composite laminates were calculated. The results show that the simulation results were consistent with the experimental ones. Compared with the CPU seri- al operation mode, an acceleration of 2.6 to 10.3 times was achieved. Moreover, the computation cost of peridynamic models was reduced.

关 键 词：近场动力学 图形处理器并行计算 复合材料 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论] TB332[自动化与计算机技术—计算机科学与技术]