[Bug ipa/68851] [6 Regression] ICE: in set_comdat_group, at ipa-comdats.c:213

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68851

--- Comment #4 from Jan Hubicka <hubicka at ucw dot cz> ---
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68851
> 
> --- Comment #3 from Martin Liška <marxin at gcc dot gnu.org> ---
> So the mentioned revision is responsible for creation of a new consprop clone:
> 
> IPA decision stage:
> ...
>  - Creating a specialized node of virtual bool
> blink::HTMLFormControlElement::isFormControlElement() const/14820 for all known
> contexts.
> ...
> IPA constant propagation end

This is fine.  At this point we should have the clone and thunks associated
with it
that are local and have no compdat group.
> 
> 
> and the symbol is identified to be set to a comdat group in ipa-comdats.c:365:
> 
>           if (is_a <cgraph_node *> (symbol))
>            dyn_cast <cgraph_node *>(symbol)->call_for_symbol_thunks_and_aliases
>                   (set_comdat_group_1,
>                    *comdat_head_map.get (group),
>                    true);
> 
> as the function calls 'set_comdat_group_1' for the constprop cloned function,
> the original function of the clone is thunk and calls the clone. So that
> set_comdat_group_1
> is called for the original function:
> 
> _ZThn96_NK5blink22HTMLFormControlElement20isFormControlElementEv/14822 (virtual
> bool
> blink::HTMLFormControlElement::_ZThn96_NK5blink22HTMLFormControlElement20isFormControlElementEv()
> const) @0x7fffeb51b8a0
>   Type: function definition analyzed
>   Visibility: externally_visible asm_written public weak comdat
> comdat_group:_ZNK5blink22HTMLFormControlElement20isFormControlElementEv
> one_only
> section:.text._ZNK5blink22HTMLFormControlElement20isFormControlElementEv
> (implicit_section) virtual artificial
>   Same comdat group as:
> _ZNK5blink22HTMLFormControlElement20isFormControlElementEv.constprop.39/31390
>   Aux: @0x1  References: 
>   Referring: 
>   Availability: available
>   First run: 0
>   Function flags: nonfreeing_fn
>   Thunk fixed offset -96 virtual value 0 has virtual offset 0)
>   Called by: 
>   Calls:
> _ZNK5blink22HTMLFormControlElement20isFormControlElementEv.constprop.39/31390
> (1.00 per call) (can throw external) 

When do we put
_ZNK5blink22HTMLFormControlElement20isFormControlElementEv.constprop.39
to the same comdat group?

Perhaps we forget to clear comdat groups after cloning?

Honza
> 
> constprop clone:
> 
> _ZNK5blink22HTMLFormControlElement20isFormControlElementEv.constprop.39/31390
> (<built-in>) @0x7fffe8bc6000
>   Type: function definition analyzed
>   Visibility:
> comdat_group:_ZNK5blink22HTMLFormControlElement20isFormControlElementEv
>   Same comdat group as:
> _ZNK5blink22HTMLFormControlElement20isFormControlElementEv/14820
>   previous sharing asm name: 31394
>   References: 
>   Referring: 
>   Clone of _ZNK5blink22HTMLFormControlElement20isFormControlElementEv/14820
>   Availability: local
>   First run: 0
>   Function flags: local nonfreeing_fn
>   Called by:
> _ZThn96_NK5blink22HTMLFormControlElement20isFormControlElementEv/14822 (1.00
> per call) (can throw external) 
>   Calls: 
> 
> Problem is that the original function has already set comdat_group here:
> 
> #0  symtab_node::set_comdat_group (this=0x7fffeb51b8a0, group=0x7fffeb4dc620)
> at ../../gcc/cgraph.h:220
> #1  0x0000000000a53972 in symtab_node::add_to_same_comdat_group
> (this=0x7fffeb51b8a0, old_node=0x7fffeb51b5c0) at ../../gcc/symtab.c:461
> #2  0x00000000008d05e2 in use_thunk (thunk_fndecl=0x7fffeb4ec7e0, emit_p=true)
> at ../../gcc/cp/method.c:396
> #3  0x00000000008e2979 in emit_associated_thunks (fn=0x7fffeb4dc620) at
> ../../gcc/cp/semantics.c:4080
> #4  0x00000000008e2cb8 in expand_or_defer_fn (fn=0x7fffeb4dc620) at
> ../../gcc/cp/semantics.c:4178
> #5  0x000000000087a103 in cp_parser_function_definition_after_declarator
> (parser=0x7ffff7feeb40, inline_p=true) at ../../gcc/cp/parser.c:25212
> #6  0x000000000087b772 in cp_parser_late_parsing_for_member
> (parser=0x7ffff7feeb40, member_function=0x7fffeb4dc620) at
> ../../gcc/cp/parser.c:26033
> #7  0x00000000008745f7 in cp_parser_class_specifier_1 (parser=0x7ffff7feeb40)
> at ../../gcc/cp/parser.c:21412
> #8  0x00000000008746bb in cp_parser_class_specifier (parser=0x7ffff7feeb40) at
> ../../gcc/cp/parser.c:21438
> #9  0x000000000086b65b in cp_parser_type_specifier (parser=0x7ffff7feeb40,
> flags=1, decl_specs=0x7fffffffd6b0, is_declaration=true,
> declares_class_or_enum=0x7fffffffd634, is_cv_qualifier=0x7fffffffd633) at
> ../../gcc/cp/parser.c:15736
> #10 0x0000000000867008 in cp_parser_decl_specifier_seq (parser=0x7ffff7feeb40,
> flags=1, decl_specs=0x7fffffffd6b0, declares_class_or_enum=0x7fffffffd6ac) at
> ../../gcc/cp/parser.c:12657
> #11 0x0000000000866645 in cp_parser_simple_declaration (parser=0x7ffff7feeb40,
> function_definition_allowed_p=true, maybe_range_for_decl=0x0) at
> ../../gcc/cp/parser.c:12200
> #12 0x00000000008665cd in cp_parser_block_declaration (parser=0x7ffff7feeb40,
> statement_p=false) at ../../gcc/cp/parser.c:12147
> #13 0x000000000086634f in cp_parser_declaration (parser=0x7ffff7feeb40) at
> ../../gcc/cp/parser.c:12044
> #14 0x0000000000865e7b in cp_parser_declaration_seq_opt (parser=0x7ffff7feeb40)
> at ../../gcc/cp/parser.c:11923
> #15 0x000000000086e172 in cp_parser_namespace_body (parser=0x7ffff7feeb40) at
> ../../gcc/cp/parser.c:17505
> #16 0x000000000086e113 in cp_parser_namespace_definition
> (parser=0x7ffff7feeb40) at ../../gcc/cp/parser.c:17482
> #17 0x000000000086624d in cp_parser_declaration (parser=0x7ffff7feeb40) at
> ../../gcc/cp/parser.c:12024
> #18 0x0000000000865e7b in cp_parser_declaration_seq_opt (parser=0x7ffff7feeb40)
> at ../../gcc/cp/parser.c:11923
> #19 0x0000000000857aaa in cp_parser_translation_unit (parser=0x7ffff7feeb40) at
> ../../gcc/cp/parser.c:4246
> #20 0x00000000008993b6 in c_parse_file () at ../../gcc/cp/parser.c:37268
> #21 0x00000000009b64a4 in c_common_parse_file () at
> ../../gcc/c-family/c-opts.c:1064
> #22 0x0000000000ead6a1 in compile_file () at ../../gcc/toplev.c:464
> #23 0x0000000000eafb58 in do_compile () at ../../gcc/toplev.c:1977
> #24 0x0000000000eafdd7 in toplev::main (this=0x7fffffffdb10, argc=16,
> argv=0x7fffffffdc18) at ../../gcc/toplev.c:2084
> #25 0x00000000017d099b in main (argc=16, argv=0x7fffffffdc18) at
> ../../gcc/main.c:39
> 
> Martin
> 
> -- 
> You are receiving this mail because:
> You are on the CC list for the bug.