http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55683
--- Comment #11 from Jan Hubicka <hubicka at gcc dot gnu.org> 2012-12-18 17:15:51 UTC --- Created attachment 29001 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29001 proposed patch OK, we know the argument is constant <addr_expr 0x7ffff7585700 type <pointer_type 0x7ffff75741f8 type <record_type 0x7ffff7560f18 C sizes-gimplified addressable needs-constructing type_1 type_5 BLK size <integer_cst 0x7ffff73f7e40 constant 128> unit size <integer_cst 0x7ffff73f7e60 constant 16> align 64 symtab 0 alias set 4 canonical type 0x7ffff7560f18 fields <field_decl 0x7ffff755ca18 c2> context <translation_unit_decl 0x7ffff7410170 D.1> full-name "struct C" needs-constructor X() X(constX&) this=(X&) n_parents=0 use_template=0 interface-unknown pointer_to_this <pointer_type 0x7ffff75741f8> chain <type_decl 0x7ffff755aac8 C>> public unsigned DI size <integer_cst 0x7ffff73f7dc0 constant 64> unit size <integer_cst 0x7ffff73f7de0 constant 8> align 64 symtab 0 alias set -1 canonical type 0x7ffff75741f8> constant arg 0 <var_decl 0x7ffff755cda8 c type <record_type 0x7ffff7560f18 C> addressable used static tree_1 tree_3 decl_5 decl_6 BLK file t.C line 25 col 12 size <integer_cst 0x7ffff73f7e40 128> unit size <integer_cst 0x7ffff73f7e60 16> align 128 context <function_decl 0x7ffff7576900 bar>>> ipa_get_indirect_edge_target used when we estimate effect of inlining goes into path looking up the binfo based on constant if (TREE_CODE (t) != TREE_BINFO) { tree binfo; binfo = gimple_extract_devirt_binfo_from_cst (t); if (!binfo) return NULL_TREE; binfo = get_binfo_at_offset (binfo, anc_offset, otr_type); if (!binfo) return NULL_TREE; return gimple_get_virt_method_for_binfo (token, binfo); } this code is missing in try_make_edge_direct_virtual_call that is actually responsible for devirtualizing. I am testing the attached patch. If I read this right, we should get the problem every time we devirtualize based on static object. I am surprised this do not trigger in testsuite/bootstrap. Honza
