A New Adaptive Algorithm for Inlining: An Experiment on FDO-Based Transformations
Ricardo Luis de Azevedo da Rocha, Mateus C.M. de Freitas Barbosa · Procedia Computer Science · 2018
This paper describes an empirical research focused on inlining, a compiler transformation that explores the idea of expanding a function’s code to uncover optimization opportunities. Previous work has not addressed the problem of representing and utilizing multi-run profiles. To produce sound results on feedback-directed optimization (FDO), we employ multi-run profiles using Berube’scontribution on Combined Profiling (CP). The FDO inliner (FDI) designed by Berube is already an adaptive one. We devised a new algorithm for inlining, a slight modification on Berube’sproduced a speedup over other inliners, including LLVM.