该论文基于共享高速缓存的多核系统，对非规则应用程序的预取技术进行研究。主要针对非规则应用程序的特点、数据预取技术以及控制参数的选取进行了深入分析与探讨。提出了一种改进的基于参数控制的帮助线程预取模型。该模型采用梯度下降算法对控制参数求解最优值，从而有效地控制帮助线程与主线程的访存任务量，使帮助线程领先于主线程，并且梯度学习算法克服了传统的人工枚举方法既复杂又费时的缺点。该论文相关的科研成果有：

　　This paper studies the prefetching technology of irregular applications based on a multi-core system with shared cache. It mainly analyzes and discusses the characteristics of irregular applications, data prefetching techniques and the selection of control parameters. An improved prefetch thread-based help control parameters of the model. The model uses the gradient descent algorithm to solve the optimal value of the control parameters, thus effectively controlling the amount of memory access between the help thread and the main thread, so that the help thread is ahead of the main thread, and the gradient learning algorithm overcomes the traditional manual enumeration methods which are complex and time consuming shortcomings. The research results related to this paper are:

　　1、Papers

一、论文

1.Songwen Pei, Junge Zhang, Naixue Xiong, Myoung-Seo Kim, Jean-Luc Gaudiot. Performance-Energy Efficiency Model of Heterogeneous Parallel Multicore System, Proceedings of IEEE International Green and Sustainable Computing Conference(IGSC),Las Vegas,USA,2015:1-6.

2.Songwen Pei, Junge Zhang, Linhua Jiang, Myoung-Seo Kim, Jean-Luc Gaudiot. Rethinking Gustafson’s Law on Asymmetric Multicore Architecture, Proceedings of International Conference on Electronics, Electrical Engineering, Computer Science (EEECS), Phuket, Thailand,2016:1-6.

3.裴颂文,宁静,张俊格.CPU-GPU异构多核系统的动态任务调度算法，计算机应用研究，2016(33):1-6.  

3.SongWen Pei,Jing Ning,JunGe Zhang.Dynamic task scheduling algorithm for CPU-GPU heterogeneous multi-core systems,Application Research of Computers,016(33):1-6.

本文工作受到上海市自然科学基金（15ZR1428600）和 计算机体系结构国家重点实验室开发基金（CARCH201206）的资助。

The work of this paper was funded by the Shanghai Natural Science Foundation (15ZR1428600) and the National Key Laboratory Development Fund for Computer Architecture (CARCH201206).