Abstract:

This paper describes how to collect and use the invariant information of a variable in GCC. The optimization performance is evaluated using NPB benchmark. It proves that compiler optimizes and improves program performance based on the value profiles, but the effects of value profiling based optimizations are quite different between different programs. It shows that there are further improvements, such as value profiling information or optimization type, etc.

Key words: profile-based optimization, value-profile, GNU Compiler Collection(GCC)

摘要：

介绍在GCC编译器中利用值剖视识别收集变量的不变特征信息并指导代码优化工作的方法。NPB基准测试程序的测试结果表明，GCC基于值剖视的优化引入的开销小，与边剖视一起使用时能获得较好的优化效果，在不同程序间显示出一定的优化针对性和局限性，值剖视信息的类型与数量、优化种类等存在较大的改进空间。

关键词: 基于剖视的优化, 值剖视, GCC编译器

CLC Number: 

• TP314