Abstract:
By analyzing the unique characteristics of reduction cyclic algorithm for solving tri-diagonal linear system, this paper presents an optimized algorithm specially for Godson 3B processor. The reduction cyclic algorithm is implemented on Godson 3B platform based on its vector instructions and vectorization techniques are used to boost performance of the algorithm. Experimental results show that, comparied with non-vectorization algorithm, vectorization algorithm larger enhances the performance.
Key words:
Godson 3B,
tri-diagonal linear equation,
cyclic reduction algorithm,
vectorization,
Poisson equation,
discretization
摘要: 研究国产CPU龙芯3B的体系结构,分析三对角线性方程组求解中的循环规约算法,并结合算法特性,提出循环规约算法的优化方案。采用向量化级数,利用龙芯3B处理器提供的向量扩展指令对该算法在龙芯3B上进行实现和优化。实验结果表明,与非向量化算法相比,向量化算法的性能提升较大。
关键词:
龙芯3B,
三对角线性方程,
循环规约算法,
向量化,
泊松方程,
离散化
CLC Number:
TUN Xi, HUANG Zhang-Jin, GU Ai-Jie. Vectorization Study for Cyclic Reduction Algorithm Based on Godson 3B[J]. Computer Engineering, 2013, 39(2): 293-298.
吴淅, 黄章进, 顾乃杰. 基于龙芯3B的循环规约算法向量化研究[J]. 计算机工程, 2013, 39(2): 293-298.