一种粗粒度可重构体系结构多目标优化映射算法

发布时间：2018-04-06 06:18

本文选题：可重构单元阵列　切入点：时域映射　出处：《电子学报》2015年11期

【摘要】：针对多约束下的行流水粗粒度可重构体系结构的硬件任务划分映射问题,提出了一种多目标优化映射算法.该算法根据运算节点执行时延、依赖度等因素构造了累加概率权值函数,在满足可重构单元面积和互连等约束下,通过该函数值动态调整就绪节点的映射调度次序,当一块可重构单元阵列当前行映射完毕后,就自动换行,当一块阵列被填满,就切换到下一块,当一个数据流图映射完毕后,就自动计算划分块数等参数.实验结果表明,与层贪婪映射算法相比,文中算法平均执行总周期降低了8.4%(RCA4×4)和5.3%(RCA6×6),与分裂压缩内核映射算法相比,文中算法平均执行总周期降低了20.6%(RCA4×4)和21.0%(RCA6×6),从而验证了文中提出算法的有效性.
[Abstract]:A multi-objective optimization mapping algorithm is proposed to solve the problem of hardware task partitioning mapping in the coarse-grained reconfigurable architecture of row income with multiple constraints.In this algorithm, the cumulative probability weight function is constructed according to the execution delay and dependency of the operation node. Under the constraints of reconfigurable unit area and interconnection, the mapping scheduling order of the ready node is dynamically adjusted by the value of the function.When the current row mapping of a reconfigurable cell array is completed, it automatically wraps the line, when one of the arrays is filled, it switches to the next block, and when a data flow graph is mapped, it automatically calculates the block number and other parameters.The experimental results show that compared with the layer greedy mapping algorithm, the average total execution period of the proposed algorithm is reduced by 8.4%(RCA4 脳 4) and 5.3%(RCA6 脳 6, and compared with the split compressed kernel mapping algorithm.The average total execution period of the proposed algorithm is reduced by 20.6%(RCA4 脳 4) and 21.0%(RCA6 脳 6), which verifies the effectiveness of the proposed algorithm.
【作者单位】：安徽工程大学计算机与信息学院;天津大学计算机科学与技术学院;同济大学软件学院;
【基金】：国家863高技术研究发展计划(No.2009AA011705,No.2013AA013204) 国家自然科学基金重点项目(No.61432017) 安徽省自然科学基金(No.1408085MF124) 芜湖市科技计划自然科学基金(No.芜科计字[2012]94号) 安徽工程大学国家自然科学预研基金安徽省高校省级自然科学基金重点项目(No.Kj2015A003)
【分类号】：TP303;TP301.6

【参考文献】