摘要
在稀疏分块矩阵的双向正交链表存储结构的基础上,采用矩阵间的快速算法和基于稀疏分块矩阵的带宽优化技术,减少了结构有限元分析的计算量和存储容量。采用高速缓存和循环展开技术,降低了存储复杂性,提高了浮点运算的平均时间。数值算例表明:采用该存储方案和快速算法,既提高了运算速度,又减少了存储空间。
Based on the two-way cross list for storing sparse block matrices^(\), fast algorithm between block matrices and bandwidth optimization based on node reorder in sparse block matrices were conducted to reduce calculation complexity and memory capacity of the finite element analysis. By using high-speed buffers and do-loop unrolling technique, it reduced memory complexity and improved average time of floating-point calculation. Numerical examples show that the proposed data structure and fast solution algorithm can markedly improve the efficiency in speed and need less memory.
出处
《吉林大学学报(工学版)》
EI
CAS
CSCD
北大核心
2005年第2期195-198,共4页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金资助项目(10202006)
关键词
工程力学
有限元法
稀疏分块矩阵
高性能计算
拓扑重分析
面向对象
engineering mechanics
finite element method
sparse block matrix
high performance computing
topological reanalysis
object oriented