基于重定位的快速寻址链接优化OA
RELOCATION-BASED LINK OPTIMIZATION FOR FAST ADDRESSING
如今,大多数编译器优化方法主要集中在编译阶段,而链接阶段的优化方法相对较少,这种不平衡的关注导致在一些特殊处理器架构上难以实现最佳性能.为解决这一问题,提出一种基于重定位的快速寻址链接优化方法,包括三种优化策略——模块外全局变量优化、函数跳转优化和动态模式转静态模式优化,旨在减少访存开销、降低指令运行时间和降低线程局部存储块维护复杂度.在 SPEC 2006 测试集上的实验结果表明,该优化方法能显著提高国产申威处理器的性能,SPECint 2006 和 SPECfp 2006 课题性能分别平均提高了 3.66%和1.50%.
Presently,the majority of compiler optimization techniques primarily concentrate on the compilation phase,while optimization methods during the linking stage remain relatively scarce.This imbalance in focus has led to challenges in achieving optimal performance on certain specialized processor architectures.To address this predicament,a relocation-based link optimization for fast addressing approach is proposed in this research,encompassing three distinct optimization strategies:module-external global variable optimization,function jump optimization,and dynamic mode to static mode optimization.These strategies are meticulously designed to reduce memory access overhead,diminish instruction execution time,and alleviate the complexity associated with maintaining thread-local storage blocks.Experimental evaluations conducted on the SPEC 2006 benchmark demonstrate that when applied to the domestic Sunway processor,the proposed method achieves an average performance improvement of 3.66%and 1.50%,respectively on the SPECint 2006 and SPECfp 2006 benchmarks.
姜军;黄亮明;羊瑞;李贺龙;朱肖炜
无锡先进技术研究院 江苏 无锡 214000无锡先进技术研究院 江苏 无锡 214000无锡先进技术研究院 江苏 无锡 214000无锡先进技术研究院 江苏 无锡 214000无锡先进技术研究院 江苏 无锡 214000
信息技术与安全科学
GCC编译器国产处理器链接优化重定位
GCC compilerDomestic processorLink optimizationRelocation
《计算机应用与软件》 2026 (9)
19-25,7
科技部重点支持项目(GG20210701).
评论