用于结构模态分析的 GPU 加速自动多级子结构方法

IF 4.4 2区 工程技术 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS
{"title":"用于结构模态分析的 GPU 加速自动多级子结构方法","authors":"","doi":"10.1016/j.compstruc.2024.107516","DOIUrl":null,"url":null,"abstract":"<div><p>In this work, a novel GPU-accelerated heterogeneous method for the automated multilevel substructuring method(HAMLS) is presented for dealing large finite element models in structural dynamics. Different parallel modes based on node, subtree, and eigenpair have been developed in the solution steps of AMLS to achieve a heterogeneous strategy. First, a new data management method is designed during the model transformation phase to eliminate the determinacy race in the parallel strategy of the separator tree. Considering the distribution characteristics of the nodes in the separator tree and the dependence of node tasks, a load balancing heterogeneous parallel strategy is designed to take full advantage of hosts and devices. By developing an adaptive batch processing program for solving eigenvectors during the back transformation phase, the overheads of launching kernels, as well as the GPU memory requirements, can be reduced by several orders of magnitude. Several numerical examples have been employed to validate the efficiency and practicality of the novel GPU-accelerated heterogeneous strategy. The results demonstrate that the computational efficiency of the novel strategy using one GPU can increase to 3.0x that of the original parallel AMLS method when 16 CPU threads are used.</p></div>","PeriodicalId":50626,"journal":{"name":"Computers & Structures","volume":null,"pages":null},"PeriodicalIF":4.4000,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A GPU-Accelerated automated multilevel substructuring method for modal analysis of structures\",\"authors\":\"\",\"doi\":\"10.1016/j.compstruc.2024.107516\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><p>In this work, a novel GPU-accelerated heterogeneous method for the automated multilevel substructuring method(HAMLS) is presented for dealing large finite element models in structural dynamics. Different parallel modes based on node, subtree, and eigenpair have been developed in the solution steps of AMLS to achieve a heterogeneous strategy. First, a new data management method is designed during the model transformation phase to eliminate the determinacy race in the parallel strategy of the separator tree. Considering the distribution characteristics of the nodes in the separator tree and the dependence of node tasks, a load balancing heterogeneous parallel strategy is designed to take full advantage of hosts and devices. By developing an adaptive batch processing program for solving eigenvectors during the back transformation phase, the overheads of launching kernels, as well as the GPU memory requirements, can be reduced by several orders of magnitude. Several numerical examples have been employed to validate the efficiency and practicality of the novel GPU-accelerated heterogeneous strategy. The results demonstrate that the computational efficiency of the novel strategy using one GPU can increase to 3.0x that of the original parallel AMLS method when 16 CPU threads are used.</p></div>\",\"PeriodicalId\":50626,\"journal\":{\"name\":\"Computers & Structures\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":4.4000,\"publicationDate\":\"2024-09-11\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Computers & Structures\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0045794924002451\",\"RegionNum\":2,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computers & Structures","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0045794924002451","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS","Score":null,"Total":0}
引用次数: 0

摘要

在这项工作中,针对结构动力学中的大型有限元模型,提出了一种新型的 GPU 加速异构自动多级子结构法(HAMLS)。在 AMLS 的求解步骤中,开发了基于节点、子树和特征对的不同并行模式,以实现异构策略。首先,在模型转换阶段设计了一种新的数据管理方法,以消除分离树并行策略中的确定性竞赛。考虑到分离树中节点的分布特点和节点任务的依赖性,设计了一种负载均衡的异构并行策略,以充分利用主机和设备的优势。通过开发一种自适应批处理程序,用于在反变换阶段求解特征向量,启动内核的开销以及 GPU 内存需求可降低几个数量级。为了验证新型 GPU 加速异构策略的效率和实用性,我们使用了几个数值示例。结果表明,当使用 16 个 CPU 线程时,使用一个 GPU 的新型策略的计算效率可提高到原始并行 AMLS 方法的 3.0 倍。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
A GPU-Accelerated automated multilevel substructuring method for modal analysis of structures

In this work, a novel GPU-accelerated heterogeneous method for the automated multilevel substructuring method(HAMLS) is presented for dealing large finite element models in structural dynamics. Different parallel modes based on node, subtree, and eigenpair have been developed in the solution steps of AMLS to achieve a heterogeneous strategy. First, a new data management method is designed during the model transformation phase to eliminate the determinacy race in the parallel strategy of the separator tree. Considering the distribution characteristics of the nodes in the separator tree and the dependence of node tasks, a load balancing heterogeneous parallel strategy is designed to take full advantage of hosts and devices. By developing an adaptive batch processing program for solving eigenvectors during the back transformation phase, the overheads of launching kernels, as well as the GPU memory requirements, can be reduced by several orders of magnitude. Several numerical examples have been employed to validate the efficiency and practicality of the novel GPU-accelerated heterogeneous strategy. The results demonstrate that the computational efficiency of the novel strategy using one GPU can increase to 3.0x that of the original parallel AMLS method when 16 CPU threads are used.

求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
Computers & Structures
Computers & Structures 工程技术-工程:土木
CiteScore
8.80
自引率
6.40%
发文量
122
审稿时长
33 days
期刊介绍: Computers & Structures publishes advances in the development and use of computational methods for the solution of problems in engineering and the sciences. The range of appropriate contributions is wide, and includes papers on establishing appropriate mathematical models and their numerical solution in all areas of mechanics. The journal also includes articles that present a substantial review of a field in the topics of the journal.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信