GA based inlining optimization in front-end synthesis of embedded software

The front-end optimization of embedded software has attracted a lot of interest in recent years. Since most computation is consumed in small fraction of code called hot path or computation kernel, even a bit performance improvement in hot path can result in great promotion, hence, function inlining become one of the effective techniques because taking away the function calling overhead will speed up the hot path. Most of previous work inlines functions globally, and also causes unwanted code increment in cold path. In our approach to function inlining, Accurate Functions Dependency Graph (AFDG) is presented to accurately modeling the function calling behavior, and the inlining optimization is carried out according to profiling data on AFDG, so that functions are partially inlined, and code increment in cold path is prevented. In order to get a satisfactory solution from the tremendous solution space in short time, a meta-heuristic genetic algorithm (GA) is applied. In experiments, the proposed algorithm shows promising results compared with previous work.