基于能耗热点的SPM静态分配策略

提出一种基于能耗热点的片上存储器静态分配策略.该分配策略依据带权重扩展控制流图将应用程序划分成不同的逻辑节点,并通过SPM平均访问能耗值计算出逻辑节点被放入SPM后产生的访问能耗的大小,以及各节点的能耗密度.构造SPM空间分配的整数线性规划算法(ILP),转化成以能耗密度为优先权的0-1背包算法来选择要放入SPM的节点.实验结果表明,使用该分配策略来管理SPM空间分配,比不使用SPM时的能耗量平均减少34%左右.

A Static Allocation Strategy for SPM Based on Energy Hotpot

A static allocation strategy for scratchpad memory is proposed based on the energy hotpot of application in this paper,this method transforms the application into a graph consisting of many distinct nodes based on WECFG,then compute the nodes access energy consumption and the nodes energy density,if nodes is placed in SPM,by SPM's average access energy consumption.Then convert the integer linear programming(ILP)problem of SPM static allocation to a 0-1 backpack problem.In the last,experiments show that our approach reduce about 34% energy consumption than that system without SPM.