[Bug c++/58252] New: ice in gimple_get_virt_method_for_binfo with -O2

[Bug c++/58252] New: ice in gimple_get_virt_method_for_binfo with -O2

[dcb314 at hotmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

            Bug ID: 58252
           Summary: ice in gimple_get_virt_method_for_binfo with -O2
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dcb314 at hotmail dot com

Created attachment 30705
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30705&action=edit
gzipped C++ source code

I just tried to compile package omniORB-4.1.6-5.fc20 with gcc 4.9 trunk dated
20130826. It said

In file included from irstub.cc:37:0:
../omniORB4/irSK.cc:13580:59: internal compiler error: in
gimple_get_virt_method_for_binfo, at gimple-fold.c:3142
 POA_CORBA::AbstractInterfaceDef::~AbstractInterfaceDef() {}
                                                           ^
0x92e140 gimple_get_virt_method_for_binfo(long, tree_node*)
    ../../src/trunk/gcc/gimple-fold.c:3141
0x96aed4 record_binfo
    ../../src/trunk/gcc/ipa-devirt.c:439
0x96ae72 record_binfo
    ../../src/trunk/gcc/ipa-devirt.c:454
0x96ae72 record_binfo
    ../../src/trunk/gcc/ipa-devirt.c:454
0x96b0e4 possible_polymorphic_call_targets_1
    ../../src/trunk/gcc/ipa-devirt.c:474
0x96b123 possible_polymorphic_call_targets_1
    ../../src/trunk/gcc/ipa-devirt.c:480
0x96c098 possible_polymorphic_call_targets(tree_node*, long, bool*, void**)
    ../../src/trunk/gcc/ipa-devirt.c:652
0x7de330 possible_polymorphic_call_targets
    ../../src/trunk/gcc/ipa-utils.h:82
0x7de330 analyze_functions
    ../../src/trunk/gcc/cgraphunit.c:936
0x7dfa85 finalize_compilation_unit()
    ../../src/trunk/gcc/cgraphunit.c:2200
0x5f3900 cp_write_global_declarations()
    ../../src/trunk/gcc/cp/decl2.c:4364
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.

Preprocessed source code attached. Flag -O2 required.

[Bug c++/58252] ice in gimple_get_virt_method_for_binfo with -O2

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Paolo Carlini <paolo.carlini at oracle dot com> changed:

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

--- Comment #1 from Paolo Carlini <paolo.carlini at oracle dot com> ---
Honza in CC I would say

[Bug c++/58252] ice in gimple_get_virt_method_for_binfo with -O2

[jamborm at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #2 from Martin Jambor <jamborm at gcc dot gnu.org> ---
Created attachment 30718
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30718&action=edit
Delta reduced testcase

A fair bit smaller multidelta-delta reduced testcase.
Requires -fpermissive in addition to -O2.

[Bug c++/58252] ice in gimple_get_virt_method_for_binfo with -O2

[jamborm at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #3 from Martin Jambor <jamborm at gcc dot gnu.org> ---
We are getting an integer instead of an expected ADDR_EXPR to a
FUNCTIONN_DECL.  Token is 3, known_binfo which is a binfo pertaining to

(gdb) pge known_binfo->typed.type
struct StructDef

has the following virtual methods (according to c++ front-end function
dump_binfo):

virtuals:
_ZN9POA_CORBA9StructDefD1Ev [0 =? 0]
_ZN9POA_CORBA9StructDefD0Ev [1 =? 1]
_ZN11omniServant15_ptrToInterfaceEPKc [2 =? 2]

One with index, zero, the other with 1, the third with 2.  No function
with index 3 there.  The constant we are loading from the initializer
is actually there right behind the address of the third method, so
that is correct.  Moreover, the original OTR has 3 in it.  The code
that actually loads the destination from the vptr also adds 24 bytes,
so it also jumps to the fourth function:

(gdb) up 8
(gdb) call debug_function (edge->caller->symbol.decl, 0)

...

  D.3162 = impl->D.2424.D.2333._vptr.omniServant;
  D.3163 = D.3162 + 24;
  D.3164 = *D.3163;
  D.3165 = tcd->arg_3;
  D.3166 = tcd->arg_2;
  D.3167 = tcd->arg_1;
  D.3168 = tcd->arg_0;
  D.3171 = OBJ_TYPE_REF(D.3164;impl->3) (impl, D.3168, D.3167, D.3166, D.3165);

When I look at what is the OTR recorderd in the edge I get

(gdb) pge edge->indirect_info->otr_type
struct _impl_Container

and its binfo has 4 virtual methods!:

(gdb) call debug_binfo(edge->indirect_info->otr_type->type_non_common.binfo)

...
virtuals:
_ZN5CORBA15_impl_ContainerD1Ev [0 =? 0]
_ZN5CORBA15_impl_ContainerD0Ev [1 =? 1]
_ZN11omniServant15_ptrToInterfaceEPKc [2 =? 2]
_ZN5CORBA15_impl_Container16create_exceptionEPKcS2_S2_RKNS_15StructMemberSeqE
[3 =? 3]

yet, in the 000i.type_inheritance dump, StructDef seems to be a
descendant of _impl_Container:

     type 5: struct _impl_Container
     defined at: a.cc:67
     base odr type ids:  1
     derived types:
       type 6: struct _impl_StructDef
       defined at: a.cc:75
       base odr type ids:  4 5
       derived types:
         type 13: struct StructDef
         defined at: a.cc:93
         base odr type ids:  6 11 12

But that I would say is impossible, because it has fewer virtual
methods than its alleged ancestor.  I would say it if was not clearly
obvious from the source that the dump file is correct.  It seems that
because the fourth virtual function is never implemented in this part
of the hierarchy, it is simply omitted in the virtual table (and
BINFO_VIRTUALs).

I have no clue how to detect this in gimple_get_virt_method_for_binfo
other than gracefully ignore rubbish... and hope that there isn't an
address of a function in that rubbish... :-/

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-09-03
            Summary|ice in                      |[4.9 Regression] ice in
                   |gimple_get_virt_method_for_ |gimple_get_virt_method_for_
                   |binfo with -O2              |binfo with -O2
     Ever confirmed|0                           |1

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |hubicka at gcc dot gnu.org

--- Comment #4 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Jason,
I think I need your help here.  There is OBJ_TYPE_REF calling method of class
_impl_Container with token 3.  
We walk possible targets and while walking StructDef, we find BINFO matghing
_impl_Container and walk its vtable for method at token 3. The vtable however
does not contain 3 methods at all.

I added code to pretty-print to also print obj_type_ref_class_type and I get:
  <bb 2>:
  tcd = cd;
  D.3158 = svnt->_vptr.omniServant;
  D.3159 = D.3158 + 16;
  D.3160 = *D.3159;
  _PD_repoId.0 = _PD_repoId;
  D.3170 = OBJ_TYPE_REF(D.3160;(struct omniServant)svnt->2) (svnt,
_PD_repoId.0);
  impl = D.3170;
  D.3162 = impl->D.2424.D.2333._vptr.omniServant;
  D.3163 = D.3162 + 24;
  D.3164 = *D.3163;
  D.3165 = tcd->arg_3;
  D.3166 = tcd->arg_2;
  D.3167 = tcd->arg_1;
  D.3168 = tcd->arg_0;
  D.3171 = OBJ_TYPE_REF(D.3164;(struct _impl_Container)impl->3) (impl, D.3168,
D.3167, D.3166, D.3165);

So I think the problem i sthat the code really looks into vtable of
omniServant, but OBJ_TYPE_REF is interepretted as _impl_Container.
I take the class type from the THIS pointer type of the method_type of the
call.  Is it posisble for omniServant vtable to contains method accepting
different types?

In that case we will need different implementation of obj_type_ref_class_type?

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.9.0

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

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

--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
This seems like a frontend issue to me.  OBJ_TYPE_REF must link object with
proper vtable.  If we can not safely fetch the type from THIS pointer type of
respective method, we need to improve the representation.

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[trippels at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Markus Trippelsdorf <trippels at gcc dot gnu.org> changed:

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

--- Comment #8 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
FWIW, with the patch from comment 7 I could successfully build Chromium.

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[trippels at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #9 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
Here's a small testcase that ICEs even with the patch from comment 7 applied:

markus@x4 library % < test.ii
typedef enum {} nsresult;
class B {
  void *mMappedMemory;

public:
  virtual int m_fn1();
};
class C : public virtual B {};
class D : C {
  virtual nsresult m_fn2();
};
nsresult D::m_fn2() {
  switch (0)
  case 0:
  m_fn1();
}
markus@x4 library % c++ -r -nostdlib -flto -O2 test.ii
lto1: internal compiler error: in record_target_from_binfo, at ipa-devirt.c:673

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[dcb314 at hotmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #10 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to David Binderman from comment #6)
> (In reply to Jan Hubicka from comment #5)
> > This seems like a frontend issue to me.  
> 
> Still broken in trunk dated 20131016.

Still broken in trunk 20140101

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #11 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
I am still hoping for help from C++ FE maintainers, but I think the bits we are
missing is logic from 
tree
lookup_base (tree t, tree base, base_access access,
             base_kind *kind_ptr, tsubst_flags_t complain)

that seems to be the tool C++ FE looks up base and binfo of given subtype.  I
am trying to get through the (somewhat convoluted at first glance) logic of
this function.

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #12 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Should be fixed by proposed fix for
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58585.
Please help me to test it.

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #13 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Author: hubicka
Date: Fri Jan 10 09:33:24 2014
New Revision: 206516

URL: http://gcc.gnu.org/viewcvs?rev=206516&root=gcc&view=rev
Log:

    PR ipa/58252
    PR ipa/59226
    * ipa-devirt.c record_target_from_binfo): Take as argument
    stack of binfos and lookup matching one for virtual inheritance.
    (possible_polymorphic_call_targets_1): Update.

    * g++.dg/ipa/devirt-20.C: New testcase.    
    * g++.dg/torture/pr58252.C: Likewise.
    * g++.dg/torture/pr59226.C: Likewise.

Added:
    trunk/gcc/testsuite/g++.dg/ipa/devirt-20.C
    trunk/gcc/testsuite/g++.dg/torture/pr58252.C
    trunk/gcc/testsuite/g++.dg/torture/pr59226.C
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa-devirt.c
    trunk/gcc/testsuite/ChangeLog

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

--- Comment #14 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Fixed.

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #15 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Fixed ;)

[Bug c++/58252] [4.9 Regression] ice in gimple_get_virt_method_for_binfo with -O2

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58252

Bug 58252 depends on bug 58585, which changed state.

Bug 58585 Summary: [4.9 Regression] ICE in ipa with virtual inheritance
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58585

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED