一种基于Matrix的QR分解向量化方法

J4 ›› 2016, Vol. 38 ›› Issue (02): 210-216.

一种基于Matrix的QR分解向量化方法

鲁庆男,刘仲

（国防科学技术大学计算机学院,湖南长沙 410073）

收稿日期:2015-02-28 修回日期:2015-05-26 出版日期:2016-02-25 发布日期:2016-02-25
基金资助:
千核级通用微处理器共享存储体系结构研究基金(61472432)

A vectorization method of QR
decomposition based on Matrix

LU Qingnan,LIU Zhong

（College of Computer,National University of Defense Technology,Changsha 410073,China）

Received:2015-02-28 Revised:2015-05-26 Online:2016-02-25 Published:2016-02-25

摘要/Abstract

摘要：

提出一种基于Matrix的Givens旋转的QR分解向量化方法。针对Matrix的体系结构特点,对向量数据访存和计算进行优化,使计算均衡分布到各个向量处理单元；设计双缓冲DMA的数据传输策略,使得内核的计算与DMA数据搬移的时间完全重迭,内核始终处于峰值计算,从而取得最佳的计算效率。实验结果表明,该方法能够取得较高的计算效率和性能加速比。

关键词: QR分解, 向量处理器, Givens旋转, 软件流水

Abstract:

We propose a vectorization method of QR decomposition with Givens rotation on Matrix processors. According to the systematic characteristics of Matrix architecture, the computation tasks are evenly distributed to all vector processing elements by optimizing the memory access to vector data and calculation. We also design a double DMA buffering scheme to smooth the data transfers, which can fully overlap the kernel computation time and the DMA data transfer time so that the kernel computation is always at its peak speed and the best computation efficiency is achieved. Experimental results show that the proposal can achieve higher computation efficiency and performance speedup.

Key words: QR decomposition;vector processor;Givens rotation;software pipeline

鲁庆男,刘仲. 一种基于Matrix的QR分解向量化方法[J]. J4, 2016, 38(02): 210-216.

LU Qingnan,LIU Zhong. A vectorization method of QR
decomposition based on Matrix [J]. J4, 2016, 38(02): 210-216.

参考文献［5］

［1］	Zhou Jie,Chen Xiaoyang,Zhao Jianxun,et al. Design and implementation of FPGA for large matrix QR decomposition［J］.Computer Engineering & Science,2010,32(10):3437.(in Chinese)
［2］	Hamill R,McCanny J V,Walke R I.Online CORDIC algorithm and VLSI architecture for implementing QRarray processors［J］.IEEE Transactions on Signal Processing,2000,48(2):592598.
［3］	Sergyienko A,Maslennikov O.Implementation of Givens QRdecomposition in FPGA［C］∥Proc of International Conference on Parallel Processing and Applied Mathematics,2000:458456.
［4］	Lorenzelli F,Yao K.A systematic folding design proccdure for a1D RLS systolic array［C］∥Proc of IEEE VLSI Signal Processing V,1992:469482.
［5］	Wang Xiufeng,Chen Hui,Fan Dehui,et al.QR decomposition of two special type matrices［J］.Journal of Jinan University(Natural Science),2005,26(3):298306.(in Chinese)
［6］	Hofmann M,Kontoghiorghes E J.Pipeline Givens sequences for computing the QR decomposition as a EREW PRAM［J］.Parallel Computing,2006,32:222230.
	附中文参考文献：
［1］	周杰,陈啸洋,赵建勋,等.大矩阵QR分解的FPGA设计与实现［J］.计算机工程与科学,2010,32(10):3437.
［5］	王秀凤,陈辉,范德辉,等.两种特殊类型矩阵的QR分解［J］.暨南大学学报（自然科学版）,2005,26(3):295306.

[1]	郭艳君，许道云，秦永彬. 基于QR分解重构虚拟样本的人脸识别算法[J]. 计算机工程与科学, 2016, 38(11): 2275-2281.
[2]	周杰1，陈啸洋1，赵建勋2，窦勇1. 大矩阵QR分解的FPGA设计与实现[J]. J4, 2010, 32(10): 34-37.