基于多种硬件实现方式探索的软硬件划分算法  

Hardware/software partitioning algorithm based on multiple hardware implementation exploration

作　　者：牛晓霞[1] 吴艳霞[1] 朱若平[2] 顾国昌[1] 刘海波[1] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,哈尔滨150001 [2]黑龙江省计算中心,哈尔滨150001

出　　处：《吉林大学学报（工学版）》2014年第4期1088-1093,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(61003036);中央高校基本科研业务费专项基金项目(HEUCF100606;HEUCFT1202);黑龙江省青年科学基金项目(QC2010049);黑龙江省教育厅科学技术研究项目(12513054)

摘　　要：提出了一种改进的遗传算法,该算法在求解划分问题的同时也解决了多种硬件方式探索问题。算法依据硬件实现方式的硬件延时-面积矛盾的特征,结合Q-学习算法和贪婪规则,自适应地选择合适变异方向,减少变异盲目性,增强遗传算法的局部搜索能力。实验结果表明,与BUB算法和标准遗传算法相比,本文方法在搜索质量、收敛性方面具有更好的效果。An improved genetic algorithm is proposed, which can solve the partitioning problem and the hardware implementation exploration simultaneously. In accordance with the characteristic of hardware delay-area contradiction in implementation, the mutation operator is chosen adaptively based on the Q-learning algorithm and greedy algorithm. The objectives are to avoid the blindness in mutation and to enhance the capability of local search of the genetic algorithm. Experiment results show that the improved genetic algorithm is more efficient than the greedy algorithm and the standard genetic algorithm in terms of searching quality and convergence.

关 键 词：计算机工程 可重构系统 现场可编程门阵列 软硬件划分算法 遗传算法 Q-学习算法 

分 类 号：TP30[自动化与计算机技术—计算机系统结构]