叶继红, 王佳. 基于GPU的杆系离散元并行算法在大型工程结构中的应用[J]. 工程力学, 2021, 38(2): 1-7. DOI: 10.6052/j.issn.1000-4750.2020.07.ST03
引用本文: 叶继红, 王佳. 基于GPU的杆系离散元并行算法在大型工程结构中的应用[J]. 工程力学, 2021, 38(2): 1-7. DOI: 10.6052/j.issn.1000-4750.2020.07.ST03
YE Ji-hong, WANG Jia. APPLICATION OF GPU-BASED PARALLEL COMPUTING METHOD FOR DEM IN LARGE ENGINEERING STRUCTURES[J]. Engineering Mechanics, 2021, 38(2): 1-7. DOI: 10.6052/j.issn.1000-4750.2020.07.ST03
Citation: YE Ji-hong, WANG Jia. APPLICATION OF GPU-BASED PARALLEL COMPUTING METHOD FOR DEM IN LARGE ENGINEERING STRUCTURES[J]. Engineering Mechanics, 2021, 38(2): 1-7. DOI: 10.6052/j.issn.1000-4750.2020.07.ST03

基于GPU的杆系离散元并行算法在大型工程结构中的应用

APPLICATION OF GPU-BASED PARALLEL COMPUTING METHOD FOR DEM IN LARGE ENGINEERING STRUCTURES

  • 摘要: 杆系DEM(离散元,discrete element method)是求解结构强非线性问题的有效方法,但随着结构数值计算规模的扩大,杆系DEM所需要的计算时间也随之急剧膨胀。为了提高杆系DEM的计算效率,该研究提出单元级并行、节点级并行的计算方法,基于CPU-GPU异构平台,建构了杆系DEM并行计算框架,编制了相应的几何非线性计算程序,实现了杆系DEM的GPU多线程并行计算。对杆系DEM并行算法的设计主要包括数据存储方式、GPU线程计算模式、节点物理量集成方式以及数据传输优化。最后采用大型三维框架、球壳结构模型分别验证了杆系DEM并行算法的计算精度,并对杆系DEM并行算法进行了计算性能测试,测试结果表明杆系DEM并行算法加速比最高可达12.7倍。

     

    Abstract: The member DEM (discrete element method) is an effective way for the structural analyses concerning with strong nonlinear issues. However, with the expansion of structural numerical calculation model, the member DEM program has been time-consuming dramatically. It proposes element-level parallel and node-level parallel computing methods to improve the calculation efficiency of the member DEM. A parallel computing framework for the member DEM is first developed based on a CPU-GPU heterogeneous platform. Then, the member DEM program that models structural geometric nonlinearity is compiled and embedded into the framework, and finally the GPU-based multi-thread parallel algorithm for the member DEM is established. The design of this parallel algorithm mainly composed of a data storage mode, a GPU thread computing mode, a node physical quantity integration mode and the way of data transmission optimization. The accuracy of the member DEM parallel algorithm proposed in this study is verified by a large-scale space frame model and a spherical shell structure model, and the performance of the algorithm was further tested. The results show that the speedup of the member DEM parallel algorithm can reach 12.7 times at most.

     

/

返回文章
返回