ROE: Runtime Optimization Environment

Traditional compilers rely on static information about programs to perform optimizations. While such optimizations are effective, the programs can often be further optimized if frequent paths through the code and/or common data values are known to the compiler. Such information can be exploited by profile-based optimizers, but such optimizers can be foiled given sufficiently dynamic behavior of programs. In this paper we propose a software/hardware architecture for continuous profiling and dynamic reoptimization of programs. We discuss the advantages and tradeoffs of such an architecture and describe a prototype implementation. While our implementation should be considered work-in-progress, the preliminary results presented here are encouraging, indicating that dynamic reoptimization techniques may indeed be beneficial and should be explored further.