基于龙芯3A的LAPACK函数优化  

Optimization of LAPACK Based on Loongson 3A

作　　者：张斌[1,2,3] 顾乃杰[1,2,3] 何颂颂[1,2,3] 刘斌斌[1,2,3] 

机构地区：[1]中国科学技术大学计算机科学技术学院,合肥230027 [2]安徽省计算与通信软件重点实验室,合肥230027 [3]中国科学技术大学中科院沈阳计算所网络与通信联合实验室,合肥230027

出　　处：《计算机系统应用》2012年第11期63-67,共5页Computer Systems & Applications

基　　金：国家"核高基"重大专项(2009ZX01028-002-003-005);国家自然科学基金(60833004);高等学校学科创新引智计划(B07033)

摘　　要：针对龙芯3A体系结构,通过底层BLAS库的优化、LAPACK分块算法中分块大小的改善以及LAPACK函数的单独优化这三种途径来提升LAPACK函数的性能.用LAPACK自带的性能测试程序进行测试,实验结果表明,有240个LAPACK函数的性能提升达到30%以上,占全部性能测试函数的81%.According to the characteristics of Loongson 3A architecture, this paper shows three ways to improve the performance of LAPACK： optimization of the underlying BLAS library, the selection of the best block size of the block algorithm in LAPACK and optimization of the specific LAPACK functions. By running the LAPACK Timing Programs, experimental results are obtained and it shows that the performance of 240 LAPACK functions, which account for 81% of all the LAPACK Timing Programs, is increased by more than 30%.

关 键 词：LAPACK BLAS 龙芯3A 优化 双单精度 

分 类 号：TP332[自动化与计算机技术—计算机系统结构]