[Bug lto/59332] New: Segmentation fault in inline_summary with LTO + attribute optimize

[Bug lto/59332] New: Segmentation fault in inline_summary with LTO + attribute optimize

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

            Bug ID: 59332
           Summary: Segmentation fault in inline_summary with LTO +
                    attribute optimize
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: lto
          Assignee: unassigned at gcc dot gnu.org
          Reporter: d.g.gorbachev at gmail dot com

$ echo '__attribute__((optimize("no-lto"))) void foo(void) { }' > 1.c
$ gcc -S -flto 1.c
1.c:1:1: internal compiler error: Segmentation fault
 __attribute__((optimize("no-lto"))) void foo(void) { }
 ^
0x86dfff0 crash_signal
    ../../gcc-4.9/gcc/toplev.c:336
0x85230d4 vec<inline_summary, va_gc, vl_embed>::operator[](unsigned int)
    ../../gcc-4.9/gcc/vec.h:718
0x85230d4 inline_summary
    ../../gcc-4.9/gcc/ipa-inline.h:242
0x85230d4 inline_write_summary()
    ../../gcc-4.9/gcc/ipa-inline-analysis.c:4079
0x86172ee ipa_write_summaries_2
    ../../gcc-4.9/gcc/passes.c:2310
0x86173a4 ipa_write_summaries_1
    ../../gcc-4.9/gcc/passes.c:2340
0x8618148 ipa_write_summaries()
    ../../gcc-4.9/gcc/passes.c:2399
0x82f22a0 ipa_passes
    ../../gcc-4.9/gcc/cgraphunit.c:2030
0x82f31b4 compile()
    ../../gcc-4.9/gcc/cgraphunit.c:2126
0x82f3519 finalize_compilation_unit()
    ../../gcc-4.9/gcc/cgraphunit.c:2280
0x8143690 c_write_global_declarations()
    ../../gcc-4.9/gcc/c/c-decl.c:10388

GCC 4.9.0 20131124. Also fails with GCC 4.7, 4.8.

[Bug lto/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

--- Comment #1 from Dmitry Gorbachev <d.g.gorbachev at gmail dot com> ---
In view of LTO shortcomings / bugs, I think that switching LTO *off* on a
per-function basis can be useful.

Perhaps it should be marked as 'enhancement'...

[Bug lto/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Dmitry Gorbachev from comment #1)
> In view of LTO shortcomings / bugs, I think that switching LTO *off* on a
> per-function basis can be useful.
> 
> Perhaps it should be marked as 'enhancement'...

Can you expand on those shortcomings/bugs?

[Bug c/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|ice-on-valid-code, lto      |accepts-invalid,
                   |                            |ice-on-invalid-code
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-11-29
          Component|lto                         |c
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
I don't think that will ever work ;)  Thus the bug is that we accept no-lto
as argument to optimize().

Thus, confirmed.

[Bug c/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

--- Comment #4 from Dmitry Gorbachev <d.g.gorbachev at gmail dot com> ---
> Can you expand on those shortcomings/bugs?

There are LTO-related issues in GCC and LD, such as PR43038, PR56536 and
others. Not all problems will be fixed soon, and more bugs will be discovered
in the future. As a workaround, an attribute (a pragma) to disable LTO would
come in handy.

> I don't think that will ever work ;)

__attribute__((optimize("lto"))) works, surprisingly!

[Bug c/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[mpolacek at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mpolacek at gcc dot gnu.org

--- Comment #5 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
(In reply to Dmitry Gorbachev from comment #4)
> __attribute__((optimize("lto"))) works, surprisingly!

I wonder whether that's intentional or just pure luck ;).

[Bug c/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

--- Comment #6 from Dmitry Gorbachev <d.g.gorbachev at gmail dot com> ---
It was discovered by accident. However, if it does work, why "no-lto" (which is
so much more useful) should not work?

[Bug lto/59332] Segmentation fault in inline_summary with LTO + attribute optimize

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59332

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |7.0
         Resolution|---                         |DUPLICATE
             Status|NEW                         |RESOLVED

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
in GCC 7+ we warning and ignore the attribute now:
<source>:1:1: warning: bad option '-fno-lto' to attribute 'optimize'
[-Wattributes]
 __attribute__((optimize("no-lto"))) void foo(void) { }


Dup of bug 61564.

*** This bug has been marked as a duplicate of bug 61564 ***