https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115825
--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Georg-Johann Lay from comment #3) > How is the tree-ssa passes quering the costs? Does it bake RTX patterns, or > is it again some implicit assumptions without looking at actual (estimated) > real costs? It is estimate_num_insns in tree-inline.cc with the weight of eni_size_weights (or eni_time_weights). Though I am not 100% sure if -Os uses eni_size_weights for eni_time_weights.
