片上网络良率评估的GPU加速  

GPU acceleration for network-on-chip yield evaluation

作　　者：蓝帆[1] 潘赟[2] 严晓浪[2] 宦若虹[3] CHENG Kwang-ting LAN Fan PAN Yun YAN Xiao-lang HUAN Ruo-hong CHENG Kwang-ting(College of Klectrical Engineering , Zhejiang University , Hangzhou S10027 9 China College of Information Science and Electronic Engineering , Zhejiang University, Hangzhou S10027, China College of Computer Science and Technology , Zhejiang University of Technology , Hangzhou 310023 , China Klectrical Computer Engineering , University of California , Santa Barbara, 93106, USA)

机构地区：[1]浙江大学电气工程学院,浙江杭州310027 [2]浙江大学信息与电子工程学院,浙江杭州310027 [3]浙江工业大学计算机科学与技术学院,浙江杭州310023 [4]University of California,Electrical Computer Engineering

出　　处：《浙江大学学报（工学版）》2017年第1期160-167,共8页Journal of Zhejiang University：Engineering Science

基　　金：浙江省自然科学基金资助项目(LY15F020008);国家自然科学基金资助项目(61204030;61302129);浙江省科技厅公益性技术应用研究计划资助项目(2014C31045)

摘　　要：针对片上网络良率评估速度较慢、效率较低的问题,研究片上网络良率评估的GPU加速,提高评估算法的执行效率.将良率评估中的样本分析算法移植到GPU平台;在分析、比较了不同平台,随机样本生成算法优劣的基础上,发现GPU平台不适合生成样本;进一步优化CPU平台上的样本生成算法,使之能与GPU一起,实现异构并行;提出CPU生成样本、GPU执行样本分析的异构并行方案.与仅使用CPU的评估算法相比,采用提出的异构并行算法实现了10倍的运行效率提升.A speedup method based on GPU platform was presented in order to improve the efficiency of the time-consuming NoC yield evaluation algorithm.The runtime efficiency was improved.The evaluation algorithm was ported to GPU platform.GPU was not suitable for generating samples based on the random number generation comparison between GPU and CPU platform.The sample generation algorithm was optimized on CPU,making it more suitable to cooperate with GPU.A heterogeneous parallel algorithm was proposed,in which CPU generates the random samples and GPU analyzes the generated samples.The proposed algorithm achieved 10 xspeedup compared to the algorithm running on purely CPU.

关 键 词：片上网络 良率 蒙特卡洛 GPU加速 

分 类 号：TN47[电子电信—微电子学与固体电子学]