面向龙芯3A体系结构的Libjpeg库优化  

OPTIMIZATION OF LIBJPEG FOR LOONGSON-3A ARCHITECTURE

作　　者：何颂颂 彭飞 林传文 陶剑文 He Songsong;Peng Fei;Lin Chuanwen;Tao Jianwen(College of Electronical and Information Engineering,Ningbo Polytechnic,Ningbo 315800,Zhejiang,China;Loongson Technology Corporation Limited,Hefei 230088,Anhui,China;School of Artificial Intelligence and Big Data,Hefei University,Hefei 230601,Anhui,China)

机构地区：[1]宁波职业技术学院电子信息工程学院,浙江宁波315800 [2]龙芯中科(合肥)技术有限公司,安徽合肥230088 [3]合肥学院人工智能与大数据学院,安徽合肥230601

出　　处：《计算机应用与软件》2022年第9期28-32,49,共6页Computer Applications and Software

基　　金：浙江省基础公益研究计划项目(LGG20F020013);浙江省自然科学基金项目(LY19F020012);浙江省教育厅科研项目(Y201941140)。

摘　　要：基于龙芯处理器的体系结构和指令系统,提出代码级和指令级优化方案,提升龙芯系列计算机的图像编码和解码速度。分析Libjpeg编解码算法的并行特性,重复使用龙芯处理器的SIMD加速硬件资源进行数据级并行优化。实验结果表明,JPEG编码速率提高70%左右,解码速率提高200%以上。工作成果已经提交到Libjpeg-turbo的主分支,极大改善了国产龙芯系列计算机中多媒体相关应用的用户体验。Based on the architecture and instruction system of Loongson processor,a code level and instruction level optimization scheme is proposed to improve the image coding and decoding speed of Loongson computers.This paper analyzed the parallel characteristics of libjpeg encoding and decoding algorithm,and repeatedly used SIMD of Loongson processor to speed up hardware resources for data level parallel optimization.The experimental results show that the JPEG encoding speed is increased by about 70%and the decoding speed is increased by more than 200%.The work results have been submitted to the main branch of Libjpeg-turbo,which greatly improves the user experience of multi-media related applications in domestic Loongson series computers.

关 键 词：Libjpeg 龙芯3A 代码优化 SIMD 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]