基于OpenMP的近场动力学模拟并行实现

Parallel implementation of peridynamic simulation based on OpenMP

针对近场动力学模拟计算量大、计算效率低等问题,研究基于OpenMP的近场动力学(peridynamic,PD)并行化方法,即在多核处理器计算机上利用OpenMP技术对PD程序中耗时占比大的构建邻接节点域、计算物质点表面修正系数、时间积分3个模块进行并行化处理,实现PD模拟的多线程并行化。通过对二维各向同性板的2种键型PD模型(准静态模拟PD模型和动态裂纹扩展模拟PD模型)测试,得出不同线程数下并行计算时间。结果表明,采用OpenMP并行技术可有效提高近场动力学模拟运算中的计算效率,相比于串行计算,并行计算获得了2.0~4.0倍的加速比。

In order to solve the low computation efficiency and large computation of peridynamic(PD)simulation,a parallel computing method based on OpenMP was studied.Using OpenMP on multi-core processor,the three most time-consuming modules in the programming,searching neighbors for each node,calculating surface correction coefficient and time integration,were parallelized.A multi-threading parallelization of PD was therefore implemented.Through the tests of two types of bond-based PD models of two-dimensional isotropic plates,the quasi-static PD model and dynamic crack-growing PD model,the computing time under different threadings was obtained.Experimental results showed that the use of OpenMP parallel technique could effectively improve the time efficiency of PD simulation,as a result,the parallel programming got 2.0 to 4.0 acceleration.