The periodic-linear model of program behavior capture

Abstract. Understanding and controlling program behavior is a challenging objective for the design of advanced compilers and critical system development. In this paper, we propose an analysis and modeling strategy of program behavior characteristics by considering traces generated from opportune code instrumentation. The proposed models consist in periodic and linear interpolations separated into adjacent program phases. It is shown that these models exhibit apparent and useful information on program behavior. Moreover they can directly be used to guide static optimizations or to build dynamic optimization processes as it is shown for the implementation of efficient dynamic data prefetching processes for some benchmark programs