基于GPU的稀疏深度神经网络性能优化  

Performance optimization of sparse deep neural network based on GPU

作　　者：石于诚 黄建强[1] 边浩东 吴利[1] 贾金芳 王晓英[1] Shi Yucheng;Huang Jianqiang;Bian Haodong;Wu Li;Jia Jinfang;Wang Xiaoying(Department of Computer Technology and Application,Qinghai University,Xining 810016,China)

机构地区：[1]青海大学计算机技术与应用系,青海西宁810016

出　　处：《电子技术应用》2023年第12期14-19,共6页Application of Electronic Technique

基　　金：国家自然科学基金项目(62062059,62162053,62166032);青海省科技计划项目-应用基础研究计划(2022-ZJ-701)。

摘　　要：随着神经网络层数不断加深,稀疏深度神经网络在计算与存储空间上更具优势,但稀疏深度神经网络的性能仍然有待优化。为此提出基于GPU的稀疏深度神经网络性能优化方法,对于计算顺序进行调整,增强数据重用性,并结合GPU的独特结构与CUDA编程方法,通过预取等方法进一步提升性能。基于GraphChallenge官方提供的数据集,相较于cuSPARSE相关库函数,最高获得了2.5倍的性能加速。With the deepening of neural network layers,the sparse deep neural network has more advantages in computing and storage space,but the performance of the sparse deep neural network still needs to be optimized.Therefore,a performance optimi-zation method based on GPU sparse deep neural network is proposed,which adjusts the order of computation,enhances the reus-ability of data,and combines the unique structure of GPU with CUDA programming method,performance is further improved by prefetching and other methods.According to GraphChallenge's official data set,it achieved up to 2.5 times the performance accel-eration compared to the related cuSPARSE library functions.

关 键 词：深度神经网络 稀疏化 异构平台 稀疏矩阵矩阵乘 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]