Compiler-Assisted Value Correlation for Indirect Branch Prediction  

作　　者：TAN Mingxing LIU Xianhua ZHANG Jiyu TONG Dong CHENG Xui 

机构地区：[1]Engineering Research Center of Microprocessor and System Ministry of Education, Peking University, Beijing 100871, China [2]Shenzhen Graduate School of Peking University, Shenzhen 518055, China

出　　处：《Chinese Journal of Electronics》2012年第3期414-418,共5页电子学报（英文版）

摘　　要：Indirect branch prediction is important to boost instruction-level parallelism in modern proces- sors. Previous indirect branch predictions usually cannot achieve high performance for the ineffectiveness of cor- related information. This paper proposes the Compiler- assisted value correlation （CVC）, a hardware/software co- operative indirect branch prediction scheme. The key is to identify effective value correlation based on program sub- structures. A compiler algorithm is introduced to iden- tify the effective value correlation based on three program substructures： virtual function calls, switch-case state- ments and function pointer calls. The compiler-identified value correlation is transferred to the dynamic predictor by extending the instruction set architecture. At runtime, the processor relies on a low-complexity Correlated value buffer （CVB） to maintain the compiler-identified value cor- relation and to guide the target address prediction for those indirect branch instructions. Our evaluations show that CVC prediction can significantly improve the perfor- mance with little extra hardware support over the tradi- tional BTB predictor and the state-of-the-art VBBI pre- diction.

关 键 词：PROCESSOR Indirect branch prediction Compiler-assisted value correlation Program substruc-ture. 

分 类 号：TP302[自动化与计算机技术—计算机系统结构] TP314[自动化与计算机技术—计算机科学与技术]