Hi,
this is the second ICE during an LTO bootstrap with Ada enabled: the assertion
at dwarf2out.c:11422 fails:
11422 gcc_assert (AT_ref (a)->die_offset);
It's again a fallout of the DW_TAG_GNU_call_site patch:
(gdb) p debug_dwarf_die(die)
DIE 1022787: DW_TAG_GNU_call_site (0x7ffff3d20640)
abbrev id: 103 offset: 1022787 mark: 1
DW_AT_low_pc: label: *.LVL21357
DW_AT_abstract_origin: die -> 0 (0x7ffff31f62d0)
DW_AT_sibling: die -> 1022817 (0x7ffff3d20730)
DIE 1022804: DW_TAG_GNU_call_site_parameter (0x7ffff3d20690)
abbrev id: 48 offset: 1022804 mark: 1
DW_AT_location: location descriptor
DW_AT_GNU_call_site_value: location descriptor
DIE 1022810: DW_TAG_GNU_call_site_parameter (0x7ffff3d206e0)
abbrev id: 48 offset: 1022810 mark: 1
DW_AT_location: location descriptor
DW_AT_GNU_call_site_value: location descriptor
DW_AT_abstract_origin link:
(gdb) p debug_dwarf_die(0x7ffff31f62d0)
DIE 0: DW_TAG_subprogram (0x7ffff31f62d0)
abbrev id: 0 offset: 0 mark: 2 <---- zero offset!
DW_AT_abstract_origin: die -> 1067604 (0x7ffff13b3820)
DW_AT_low_pc: label: *.LFB4069
DW_AT_high_pc: label: *.LFE4069
DW_AT_frame_base: location list -> label:*.LLST6555
DW_AT_static_link: location descriptor
DW_AT_abstract_origin link:
DIE 1067604: DW_TAG_subprogram (0x7ffff13b3820)
abbrev id: 12 offset: 1067604 mark: 1
DW_AT_name: "sem_ch10__build_limited_views__new_internal_shadow_entity"
DW_AT_decl_file: "/home/eric/svn/gcc/gcc/ada/sem_ch10.adb" (66)
DW_AT_decl_line: 5540
DW_AT_prototyped: 1
DW_AT_type: die -> 194 (0x7ffff15762d0)
DW_AT_inline: 1
In the source file:
Lim_Typ :=
New_Internal_Shadow_Entity
(Kind => Ekind (Comp_Typ),
Sloc_Value => Sloc (Comp_Typ),
Id_Char => 'Z');
In the assembly file:
call atree__ekind
movzbl %al, %edi
movq %rbx, %r10
movl %r13d, %esi
call
sem_ch10__build_limited_views__new_internal_shadow_entity.4857.134033.constprop.66.6913
(gdb) p debug_tree(decl)
<function_decl 0x7ffff6c4dc00
sem_ch10__build_limited_views__new_internal_shadow_entity.4857.134033.constprop.66
whose DECL_CONTEXT is:
<function_decl 0x7ffff6c4da00 sem_ch10__build_limited_views
type <function_type 0x7ffff70e3f18
type <void_type 0x7ffff7ee0d20 void asm_written VOID
align 8 symtab 0 alias set -1 canonical type 0x7ffff7ee0d20
pointer_to_this <pointer_type 0x7ffff7ee0dc8>>
QI
size <integer_cst 0x7ffff7ed0f80 constant 8>
unit size <integer_cst 0x7ffff7ed0fa0 constant 1>
align 8 symtab 0 alias set -1 canonical type 0x7ffff70e3f18
arg-types <tree_list 0x7ffff70f5258 value <integer_type 0x7ffff7ee0540
unsigned int>
chain <tree_list 0x7ffff70f5280 value <void_type 0x7ffff7ee0d20
void>>>
pointer_to_this <pointer_type 0x7ffff6536930>>
addressable nothrow static QI file /home/eric/svn/gcc/gcc/ada/sem_ch10.adb
line 5246 col 4 align 8 initial <error_mark 0x7ffff7ed7990>
This DECL is stream out as DECL_CONTEXT but isn't in the LTRANS partition so
its DECL_INITIAL is ERROR_MARK and its nested functions aren't declared with
the DWARF back-end. As a result, both DW_TAG_subprogram DIEs above end up on
the limbo list, in the above order. So when:
dw_die_ref origin = get_AT_ref (die, DW_AT_abstract_origin);
if (origin)
add_child_die (origin->die_parent, die);
is run, origin->die_parent is NULL for the first DIE; it will only be set when
the second DIE is processed on the limbo list.
Hence the proposed fix: not to attach a DIE to a NULL parent. LTO bootstrapped
with Ada on x86_64-suse-linux, OK for the mainline?
2011-04-18 Eric Botcazou <ebotca...@adacore.com>
PR lto/48492
* dwarf2out.c (dwarf2out_finish): Do not attach a DIE on the limbo list
to a NULL parent.
--
Eric Botcazou
Index: dwarf2out.c
===================================================================
--- dwarf2out.c (revision 172617)
+++ dwarf2out.c (working copy)
@@ -23559,7 +23559,7 @@ dwarf2out_finish (const char *filename)
{
dw_die_ref origin = get_AT_ref (die, DW_AT_abstract_origin);
- if (origin)
+ if (origin && origin->die_parent)
add_child_die (origin->die_parent, die);
else if (is_cu_die (die))
;
