基于人工神经网络的线程数据加速划分
|
摘要
论文提出一种基于人工神经网络的线程数据加速划分模型。利用该模型学习样本集中隐含的划分知识,来指导新程序的划分。样本中的知识用样本的特征和划分方案组成,它们分别作为网络的输入和输出来训练网络,直至达到指定的精度。在测试阶段,利用测试程序预执行获取的特征信息作为模型输入,运行模型得出该程序的划分方案,作为预测结果。基于Prophet实验平台,该文提出的模型有效地预测出待划分程序的划分方案,平均预测精度约为0.7。而且依据该划分方案指导程序划分后,加速比性能比原有划分方案获得的加速比最大提升了11.8%。实验证明了论文提出的模型在对未知程序划分方案预测方面是有效的。
This paper presents an artificial neural network based on thread data acceleration partition model. This model is used to learn the partitioning knowledge implicit in the sample set to guide the division of new procedures. The knowledge in the sample is composed of sample characteristics and division schemes. They train the network as the input and output of the network,respectively,until the specified accuracy is reached. In the test phase,the test program is used to pre-execute the acquired characteristic information as a model input. The operation model is used to obtain a division scheme of the program as a prediction result.Based on the Prophet experimental platform,the model proposed in this paper effectively predicts the division scheme of the program to be divided,and the average prediction accuracy is about 0.7. Moreover,according to the division scheme guiding program,the acceleration ratio obtained by the acceleration ratio ratio is higher by 11.8% than the original division scheme. The experiment proves that the model presented in this paper is effective in predicting the unknown program partitioning scheme.
This paper presents an artificial neural network based on thread data acceleration partition model. This model is used to learn the partitioning knowledge implicit in the sample set to guide the division of new procedures. The knowledge in the sample is composed of sample characteristics and division schemes. They train the network as the input and output of the network,respectively,until the specified accuracy is reached. In the test phase,the test program is used to pre-execute the acquired characteristic information as a model input. The operation model is used to obtain a division scheme of the program as a prediction result.Based on the Prophet experimental platform,the model proposed in this paper effectively predicts the division scheme of the program to be divided,and the average prediction accuracy is about 0.7. Moreover,according to the division scheme guiding program,the acceleration ratio obtained by the acceleration ratio ratio is higher by 11.8% than the original division scheme. The experiment proves that the model presented in this paper is effective in predicting the unknown program partitioning scheme.
引文
[1]袁一丹,肖先勇,刘亚梅,等.考虑误差影响的两阶段谐波源定位方法[J].电力系统及其自动化学报,2017,29(12):91-96.
[2]意法半导体推出高动态范围音频处理器[J].世界电子元器件,2014(7):48.
[3]黄乐天,别丽华.同构与异构片上多核系统的演进过程[J].电子技术应用,2017,43(3):6-11,20.
[4]刘斌,赵银亮,韩博,等.基于性能预测的推测多线程循环选择方法[J].电子与信息学报,2014,36(11):2768-2774.
[5]杜延宁,赵银亮,韩博,等.一种数据结构制导的线程划分方法与执行模型[J].软件学报,2013,24(10):2432-2459.
[6]Yu Z,Yu Z,Yu X,et al. Low-power multicore processor design with reconfigurable same-instruction multiple process[J]. IEEE Transactions on Circuits and Systems II:Express Briefs,2014,61(6):423-427.
[7]Young-S L E,Muruganandam P,Adhikari S K,et al.OpenMP GNU and Intel Fortran programs for solving the time-dependent Gross-Pitaevskii equation[J]. Computer Physics Communications,2017,220:503-506.
[8]Kiefer M,Klebanov V,Ulbrich M. Relational program reasoning using compiler IR[J]. Journal of Automated Reasoning,2018,60(3):337-363.
[9]王晓,赵丽.基于双层结构的加速K-NN分类方法[J].计算机工程与设计,2018,39(4):1071-1077.
[10]Johnson T L. Religious Heretic,Political Prophet:Malcolm X,Democracy,and Abolition Ethics[J]. Journal of Africana Religions,2015,3(1):62-82.
[11]Liu B,Zhao Y,Li Y,et al. A thread partitioning approach for speculative multithreading[J]. The Journal of Supercomputing,2014,67(3):778-805.
[12]李钟,赵银亮,杜延宁. Prophet推测多线程系统设计与实现[J].计算机科学,2011,38(2):296-301.
[13]Giam X,Olden J D. A new R2-based metric to shed greater insight on variable importance in artificial neural networks[J]. Ecological Modelling,2015,313:307-313.
[14]李山山,刘敬晗.利用Tomasulo算法处理数据相关的流水线CPU设计[J].实验室研究与探索,2014,33(12):90-95.
[15]李远成,阴培培,赵银亮.基于模糊聚类的推测多线程划分算法[J].计算机学报,2014,37(3):580-592.
[16]李远成,赵银亮,李美蓉,等.一种基于路径优化的推测多线程划分算法[J].软件学报,2012,23(8):1950-1964.
[2]意法半导体推出高动态范围音频处理器[J].世界电子元器件,2014(7):48.
[3]黄乐天,别丽华.同构与异构片上多核系统的演进过程[J].电子技术应用,2017,43(3):6-11,20.
[4]刘斌,赵银亮,韩博,等.基于性能预测的推测多线程循环选择方法[J].电子与信息学报,2014,36(11):2768-2774.
[5]杜延宁,赵银亮,韩博,等.一种数据结构制导的线程划分方法与执行模型[J].软件学报,2013,24(10):2432-2459.
[6]Yu Z,Yu Z,Yu X,et al. Low-power multicore processor design with reconfigurable same-instruction multiple process[J]. IEEE Transactions on Circuits and Systems II:Express Briefs,2014,61(6):423-427.
[7]Young-S L E,Muruganandam P,Adhikari S K,et al.OpenMP GNU and Intel Fortran programs for solving the time-dependent Gross-Pitaevskii equation[J]. Computer Physics Communications,2017,220:503-506.
[8]Kiefer M,Klebanov V,Ulbrich M. Relational program reasoning using compiler IR[J]. Journal of Automated Reasoning,2018,60(3):337-363.
[9]王晓,赵丽.基于双层结构的加速K-NN分类方法[J].计算机工程与设计,2018,39(4):1071-1077.
[10]Johnson T L. Religious Heretic,Political Prophet:Malcolm X,Democracy,and Abolition Ethics[J]. Journal of Africana Religions,2015,3(1):62-82.
[11]Liu B,Zhao Y,Li Y,et al. A thread partitioning approach for speculative multithreading[J]. The Journal of Supercomputing,2014,67(3):778-805.
[12]李钟,赵银亮,杜延宁. Prophet推测多线程系统设计与实现[J].计算机科学,2011,38(2):296-301.
[13]Giam X,Olden J D. A new R2-based metric to shed greater insight on variable importance in artificial neural networks[J]. Ecological Modelling,2015,313:307-313.
[14]李山山,刘敬晗.利用Tomasulo算法处理数据相关的流水线CPU设计[J].实验室研究与探索,2014,33(12):90-95.
[15]李远成,阴培培,赵银亮.基于模糊聚类的推测多线程划分算法[J].计算机学报,2014,37(3):580-592.
[16]李远成,赵银亮,李美蓉,等.一种基于路径优化的推测多线程划分算法[J].软件学报,2012,23(8):1950-1964.