面向共享内存结构的并行Delaunay网格生成算法  

Parallel Delaunay mesh generation algorithm for shared memory structure

作　　者：张晓蒙[1,2] 陆忠华[1] 张鉴[1] 马文鹏[1,2] 

机构地区：[1]中国科学院计算机网络信息中心超级计算中心,北京100190 [2]中国科学院大学,北京100190

出　　处：《计算机应用研究》2016年第6期1616-1619,共4页Application Research of Computers

基　　金：国家"863"计划资助项目(2012AA01A304);国家自然科学基金资助项目(91130019;91330206;11271350);数学工程与先进计算国家重点实验室开放课题基金资助项目

摘　　要：针对目前大多数并行Delaunay网格生成算法对共享内存结构利用不充分、不能够利用超级计算机多层次体系结构优势的情况,提出了一种充分利用共享内存结构的基于算法并行模式的并行Delaunay网格生成算法。通过对候选点集进行高效划分来实现插点操作的并行,增大了一次选择之后进行并行插点的点集规模。使用Open MP并行模型对所提出算法进行并行实现,并与串行开源软件Triangle进行了对比。实验结果表明算法能够将候选点集划分成互不冲突的子集进行并行处理,在保证网格质量的同时具有较好的并行效率。In order to suit the shared memory structure well and take advantages of the supercomputers with hybrid architecture,this paper proposed a parallel Delaunay mesh generation algorithm based on parallelizing the procedure of inserting points. It took fully advantages of the shared memory structure. The presented algorithm inserted points in parallel by partitioning the candidate point set into independent subsets,thus it could increase the number of points inserted in each iteration of parallel point insertion. It implemented the presented algorithm using Open MP,and compared with open-source code Triangle.The experimental results show that the presented parallel algorithm can successfully divide the candidate point set into independent subsets and handle them in parallel,together with relatively good mesh quality and parallel efficiency.

关 键 词：网格生成 DELAUNAY算法 并行计算 共享内存 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]