基于GPU的多项式矩阵特征值并行计算  

Parallel computing for polynomial matrix eigenvalue with GPU

作　　者：杨智诚[1] 黄友钦[1] 傅继阳[1] 

机构地区：[1]广州大学工程结构抗风与结构安全国际科技合作基地,广州510006

出　　处：《中国科技论文》2016年第8期957-960,共4页China Sciencepaper

基　　金：高等学校博士学科点专项科研基金资助项目(20124410110005);广州市属高校青年项目(1201430662)

摘　　要：以多项式矩阵特征值求解问题为背景,推导任意项式特征值计算方法,阐述了基于图形处理单元(GPU)的并行计算机理并建立Matlab快速并行计算平台。以二次特征值计算为例,分别测试基于GPU的单精度类型计算与双精度类型计算的效率与误差,同时验证公式的正确性。数值计算表明,单精度计算与双精度计算能够大幅度降低计算时间,提高计算效率,误差对比分析也表明GPU的计算结果能够满足一定的工程计算要求。According to the problem of polynomial matrix eigenvalue ,the calculation method is derived ,the parallel computing mechanism based on GPU （graphics processing unit ） is introduced ,and a fast parallel computing platform is builted .Taking quadratic eigenvalue problem for example ,the computational efficiency and error is tested under the single precision data and double precision data ,respectively .The correctness of the derived equation is verified at the mean time .Numerical calculation in-dicates that the computing time of single precision case and double precision case is shorter than the serial computation with CPU , which improves the computational efficiency .The analysis also shows that the results of GPU computation can meet the accuracy requirements of engineering projects .

关 键 词：多项式矩阵特征值 二次特征值 图形处理单元 单精度 双精度 

分 类 号：TP338.6[自动化与计算机技术—计算机系统结构] O241.6[自动化与计算机技术—计算机科学与技术]