Matrix编译器If转换算法的实现  

Implement of Matrix Compiler's If-convertion Algorithm

作　　者：刘飞[1] 陈跃跃[1] 孙海燕[1] 阳柳[1] 

机构地区：[1]国防科技大学计算机学院,长沙410073

出　　处：《计算机科学》2013年第4期55-58,77,共5页Computer Science

摘　　要：指令级并行在提高处理器运行速度方面显得越来越重要,if转换技术是一种在处理器支持条件执行的前提下,有助于提高指令级并行度的编译优化技术。在详细分析GCC(GNU Compiler Collection)内部if转换技术的实现机制和算法的基础上,针对matrix体系结构特点,对GCC中现有if转换算法进行了移植与改进,实现了matrix编译器的if转换算法。实验证明,改进后的if转换算法能够更有效地移除分支,减少基本块的数量,扩大单个基本块的范围,有助于编译器生成更加优化的代码。The ILP（instruction-level-parallel） becomes more and more important for improving the microprocessor＇s speed.The if-convertion technology is an effective compiler-optimization method of facilitating exploiting ILP based on the microprocessor which supplies conditional execution.After introducing if-convertion technology and algorithm of GCC and transplanting it into the Matrix compiler successfully,some improvements for if-convertion of Matrix compi-ler were made.Experiment results indicate that the if-convertion achieved in the paper can remove the branches,reduce the number of basic-blocks and broaden the bound of basic-block.It can also facilitate making more predominant codes for compiler.

关 键 词：谓词执行 if转换 条件执行 指令级并行 MATRIX VLIW 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]