基础数学库中的MCET寄存器分配方法

摘要

针对基础数学库中的寄存器分配特点,利用最常用情况执行时间MCET(Most-Case Execution Time)模型对经典的线性扫描寄存器分配算法进行了扩展.该算法能够很大程度上减少数学库中的最常用路径上的变量溢出过程,将变量溢出过程分配到非常用路径上,从而减少全局的寄存器溢出开销,提高数学库的性能.对基础数学库中函数的应用此分配算法之后,最常用路径执行时间、平均路径执行时间都得到了不同程度的提高.%Aiming at the characteristic of register allocation in basic mathematical library, in the paper we extend the classic linear-scan register allocation algorithm by making use of the model of most-case execution time ( MCET). The extended algorithm can reduce to a great extent the process of number spilling out on most-used path of the functions in the library; it distributes the spilling out process to the paths rarely used, thereby decreases the overall cost of register allocation spilling out and improves the performance of the library. Applying the extended allocation algorithm to the functions in the library, the most-case execution time and average-case execution time are all improved in different degrees.