Profile-guided code optimization

点击翻译按钮获取中文摘要

Modern object-oriented programming languages, Java and C# in particular, rely on a technology called 'just-in-time' (JIT) compilation to achieve good performance. JIT compilers generate machine code while the program executes. However a JIT compiler cannot afford to spend much time on extensive analysis of the program to optimize it fully, and thus to improve performance further.The proposed research involves collecting a minimal amount of information while the program runs to identify 'hot' parts of the program, where a large proportion of the execution time is spent. In the light of this information, selected portions of the program can be recompiled. In the recompiled portions, computations have been moved out of hot parts of the program and moved to cold parts, thus improving the program's overall performance.