Provide fine-grain optimization selection and tuning abilities in GCC to be able to tune default optimization heuristic of the compiler or fine optimizations for a given program on a given architecture entirely automatically using statistical and machine learning techniques from the MILEPOST project.

by Yuanjie Huang for GCC

Iterative feedback-directed compilation combined with statistical and machine learning techniques showed very promising results in automating optimization heuristic tuning and compiler design for rapidly evolving hardware to considerably improve program performance and code size. I’d like to extend GCC to be able to select/deselect individual optimizations on a fine-grain level and tune their parameters automatically using machine learning technology from the MILEPOST project.