基于GPU加速的边界面法正则积分的研究-完整实体工程结构分析.PDF

基于 GPU 加速的边界面法正则积分的研究 1＋ 1 2 张见明 ，余列祥 ,刘路平 ( 1. 汽车车身先进设计与制造国家重点实验室 湖南大学 长沙 410082 ; 2. 中国水电顾问集团中南勘测设计 研究院 长沙 410014 ) 摘要：相比传统的微处理器 CPU ，图像处理单元GPU 具备低成本、低功率，高性能等特点。由于 GPU 强大的并行计算能力和快速提升的可编程能力，除了其在图形处理上应用，GPU 越来越广泛的应用于数值 计算上。正则积分是边界面法中计算较为密集的部分, 具有高度并行性。本文首次在 CUDA 编程环境中运 用 GPU 对边界面法正则积分进行加速，并在 NVIDIA GTX680 GPU 和英特尔（R ）酷睿（TM ）酷睿i7-3770K CPU 的计算平台上与传统的正则单元积分进行对比。数值算例表明，在保证相同精度的前提下，加速比可 达到 8.3。 关键字：GPU，并行计算，边界面法，CUDA，正则积分 中图分类号：TH164 文献标识代码：A Research on Regular Integration in boundary face method Based on GPU-acceleration 1＋ 1 2 Jianming Zhang Liexiang Yu Luping Liu ( 1. State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Changsha, China, 410082; 2. Hydrochina Zhongnan Engineering Corporation, Changsha, 410014) Abstract: Graphics processing unit (GPU) is a low-cost, low-power (watts per flop) and very high performance alternative to conventional microprocessors. In addition to GPU’s applications in graphics processing, it was used in numerical computing more and more widely due to its powerful ability of parallel computing and rapidly improved programmability. The regular integration possessing high level of parallelism is a computationally intensive part of the BFM. This paper makes a first try of applying GPU to accelerate computation for boundary face method (BFM) in CUDA (Compute Unified Device Architecture) programming environment. Comparative computations compared to traditional regular element integration are made on both NV