Natallie Baikevich on Automatic Construction of Inlining Heuristics using ML

PWL Zürich returns from summer vacations with a compilers paper: Automatic Construction of Inlining Heuristics using Machine Learning by S. Kulkarni, J. Cavazos, C. Wimmer, D. Simon (2013) (pdf (https://www.eecis.udel.edu/~skulkarn/papers/cgo-2013.pdf)).

Method inlining is a very important but also dangerous compiler optimization: an inlining decision might lead to significant speedup or performance degradation and has to be constructed carefully. The paper compares various features and inlining techniques, in particular neuro-evolution. We will also discuss how having an idea about inlining traps and benefits might come useful in the "real-world", where not everybody is a compiler developer.