数字信号处理器代码生成的地址分配自适应遗传算法  

Adaptive Genetic Algorithm for Address Assignment in Digital Signal Processor Code Generation

作　　者：张丹[1] 李增智[1] 宋红[2] 刘吉[1] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049 [2]西安石油大学机械工程学院,西安710065

出　　处：《西安交通大学学报》2005年第12期1315-1318,共4页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(60173059)

摘　　要：为了高质量地生成对数字信号处理器自增/减间接寻址模式的目标代码,提出了地址分配自适应遗传算法.该算法的核心思想为:对地址分配方案进行二进制编码,适应度函数通过地址寄存器变化轨迹的非邻接标记之和的倒数来计算,再通过自适应策略对适应度高的个体以较低概率进行顺序交叉和交换变异操作,使良好基因得以保护进入下一代,并对适应度低的个体赋予较高概率,使较差基因被淘汰掉,从而能实现逃离局部最优并能快速收敛.通过对大量随机变量访问序列进行仿真试验,结果表明该算法运行效率较高,地址计算指令代价比Liao的算法和Leupers的算法分别减少11%和8%.An adaptive genetic algorithm is proposed in order to perfectly generate the target code of digital signal processors with auto-increment/auto-decrement indirect addressing modes. The core of the proposed algorithm is as follows. The address assignment solutions are encoded as binary gene schemes. The fitness function is calculated according to the reciprocal of sum of non-neighbor of address register trace. Then the adaptive strategy is applied to operate ordered crossover and swap mutation operations on individuals with higher fitness using lower probability to preserve better genes for next generations, and for those with lower fitness using higher probability to eliminate the worse genes. This strategy can avoid local optimal and speed up convergence. The simulation experiments with large numbers of stochastic variable＇s access sequences show that the proposed algorithm has high execution efficiency and reduces the cost of address computation instructions by 11% and 8% respectively in comparison with Liao＇ s algorithm and Leupers＇ algorithm.

关 键 词：地址分配 自适应遗传算法 自增/减 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]