http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58125
--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> --- OK, the testcase no longer reproduces, but the dump seems clear. We introduced new aliases but did not resize the summary vector, because we do not need info for these. Then we attempt to free it and ICE. I am testing Index: ipa-inline-analysis.c =================================================================== --- ipa-inline-analysis.c (revision 202199) +++ ipa-inline-analysis.c (working copy) @@ -4019,7 +4019,8 @@ if (!inline_edge_summary_vec.exists ()) return; FOR_EACH_DEFINED_FUNCTION (node) - reset_inline_summary (node); + if (!node->symbol.alias) + reset_inline_summary (node); if (function_insertion_hook_holder) cgraph_remove_function_insertion_hook (function_insertion_hook_holder); function_insertion_hook_holder = NULL; and will commit it along with the PR if that passes.