[Bug lto/54980] New: [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[Bug lto/54980] New: [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

             Bug #: 54980
           Summary: [4.8 regression] gimple check: expected
                    gimple_cond(error_mark), have gimple_call() in
                    gimple_cond_set_lhs, at gimple.h:2578
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: lto
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dimhen@gmail.com


$ g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/usr/local/gcc_current/libexec/gcc/x86_64-unknown-linux-gnu/4.8.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: /home/dimhen/src/gcc-current/configure
--prefix=/usr/local/gcc_current --enable-__cxa_atexit --enable-shared
--enable-checking=yes,df,fold,rtl,tree --enable-gnu-unique-object
--enable-linker-build-id --enable-languages=c,c++,lto --enable-plugin
--enable-version-specific-runtime-libs --with-tune=generic
Thread model: posix
gcc version 4.8.0 20121018 (experimental) [trunk revision 192560] (GCC) 

$ g++ -flto -fpreprocessed -c 1.ii -o 1.o
$ g++ -flto -O1 -fpreprocessed -c 2.ii -o 2.o
$ g++ -Ofast -flto -o t 1.o 2.o
In file included from :0:0:
1.ii: In member function ‘parse’:
1.ii:8:17: internal compiler error: gimple check: expected
gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at
gimple.h:2578
     virtual int parse ()
                 ^
0x6af233 gimple_check_failed(gimple_statement_d const*, char const*, int, char
const*, gimple_code, tree_code)
    /home/dimhen/src/gcc-current/gcc/gimple.c:1160
0x104ec27 gimple_cond_set_lhs
    /home/dimhen/src/gcc-current/gcc/gimple.h:2578
0x104e309 gimple_cond_set_lhs
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:533
0x104e309 gimple_cond_make_true
    /home/dimhen/src/gcc-current/gcc/gimple.h:2674
0x104e309 try_unroll_loop_completely
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:548
0x104e309 canonicalize_loop_induction_variables
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:666
0x104e737 tree_unroll_loops_completely(bool, bool)
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:815
0x98ee31 tree_complete_unroll_inner
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop.c:495
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.
lto-wrapper: g++ returned 1 exit status
/usr/bin/ld: fatal error: lto-wrapper failed
collect2: error: ld returned 1 exit status

$ cat 1.ii
extern "C" class A
{
};
template < class T_gengetopt_args_info, int (*t_parser) (int, char **,
        T_gengetopt_args_info
        *) >class B
{
    virtual int parse ()
    {
        A a;
        t_parser (0, 0, 0);
    }
};

extern "C" int f (int, char **, int *);
class C:B < int, f >
{
};
void
selectCommandHandler ()
{
    new C;
}

$ cat 2.ii
extern "C"
{
    void printf (const char *...);
    int f ();
}
const char *a[] = { "", 0 };

int b;
void f_internal ();
int
f ()
{
    f_internal ();
}

void
f_internal ()
{
    while (1)
        switch (b)
        case 'h':
{
    int i;
    while (a[0])
        printf (a[i++]);
    }
}

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #1 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-18 23:05:38 UTC ---
192502 OK

$ g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/usr/local/gcc_current_192502/bin/../libexec/gcc/x86_64-unknown-linux-gnu/4.8.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: /home/dimhen/src/gcc-current/configure
--prefix=/usr/local/gcc_current --enable-__cxa_atexit --enable-shared
--enable-checking=yes,df,fold,rtl,tree --enable-gnu-unique-object
--enable-linker-build-id --enable-languages=c,c++,lto --enable-plugin
--enable-version-specific-runtime-libs --with-tune=native
Thread model: posix
gcc version 4.8.0 20121016 (experimental) [trunk revision 192502] (GCC)

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #2 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-18 23:51:33 UTC ---
more reduced 

$ cat 1.ii
extern "C" class A
{
};
template < int (*t_parser) () > class B
{
    virtual int parse ()
    {
        A a;
        t_parser ();
    }
};

extern "C" int f ();
class C:B < f >
{
};
void
g ()
{
    new C;
}

$ cat 2.ii
extern "C"
{
    void printf (const char *...);
    int f ();
}
const char *a[] = { "", 0 };

int b;
int
f ()
{
    while (1)
        switch (b)
        case 'h':
{
    int i;
    while (a[0])
        printf (a[i++]);
    }
}

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #3 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-19 00:08:54 UTC ---
and more

$ cat 1.ii
class A
{
};
template < int (*t_parser) () > class B
{
    virtual int parse ()
    {
        A a;
        t_parser ();
    }
};

extern "C" int f ();
class C:B < f >
{
};
void
g ()
{
    new C;
}

$ cat 2.ii
extern "C" int f ();
char *a[] = { 0, 0 };

void bar (char *);
int
f ()
{
    int i;
    while (1)
        bar (a[i++]);
}

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-10-19
                 CC|                            |hubicka at gcc dot gnu.org
   Target Milestone|---                         |4.8.0
     Ever Confirmed|0                           |1

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-19 08:38:24 UTC ---
Honza, this is yours.

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #5 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-19 10:26:35 UTC ---
192517 OK
192548 FAIL

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #6 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-20 14:28:28 UTC ---
192529 OK
192538 FAIL

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> 2012-10-23 14:48:05 UTC ---
This does not seem to reproduce for me. The following patch should however fix
it.
Index: tree-ssa-loop-ivcanon.c
===================================================================
--- tree-ssa-loop-ivcanon.c     (revision 192711)
+++ tree-ssa-loop-ivcanon.c     (working copy)
@@ -364,6 +364,10 @@ loop_edge_to_cancel (struct loop *loop)
        else
          edge_to_cancel = EDGE_SUCC (edge_to_cancel->src, 0);

+      /* We only can handle conditionals.  */
+      if (!(edge_to_cancel->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
+       continue;
+
       /* We should never have conditionals in the loop latch. */
       gcc_assert (edge_to_cancel->dest != loop->header);

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #8 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-24 09:58:31 UTC ---
i re-check 192757

--enable-checking=release FAIL with
$ bash -x tf.sh
+ g++ -flto -fpreprocessed -c 1.ii -o 1.o
+ g++ -flto -O1 -fpreprocessed -c 2.ii -o 2.o
+ g++ -Ofast -flto -o t 1.o 2.o
In file included from :0:0:
1.ii: In member function 'parse':
1.ii:6:17: internal compiler error: Segmentation fault
     virtual int parse ()
                 ^
0x7539bf crash_signal
    /home/dimhen/src/gcc-current/gcc/toplev.c:335
0x5fd754 internal_fn_flags
    /home/dimhen/src/gcc-current/gcc/internal-fn.h:48
0x5fd754 gimple_call_flags(gimple_statement_d const*)
    /home/dimhen/src/gcc-current/gcc/gimple.c:1883
0x80262f gimple_call_noreturn_p
    /home/dimhen/src/gcc-current/gcc/gimple.h:2413
0x80262f update_stmt_operands(gimple_statement_d*)
    /home/dimhen/src/gcc-current/gcc/tree-ssa-operands.c:1082
0xb136b6 update_stmt
    /home/dimhen/src/gcc-current/gcc/gimple.h:1531
0xb136b6 try_unroll_loop_completely
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:552
0xb136b6 canonicalize_loop_induction_variables
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:670
0xb1487e tree_unroll_loops_completely(bool, bool)
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:833


--enable-checking=yes,df,fold,rtl,tree FAIL as

1.ii: In member function 'parse':
1.ii:6:17: internal compiler error: gimple check: expected
gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at
gimple.h:2553
     virtual int parse ()
                 ^
0x6b1063 gimple_check_failed(gimple_statement_d const*, char const*, int, char
const*, gimple_code, tree_code)
    /home/dimhen/src/gcc-current/gcc/gimple.c:1160
0x107fe69 gimple_cond_set_lhs
    /home/dimhen/src/gcc-current/gcc/gimple.h:2553
0x107f448 gimple_cond_set_lhs
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:536
0x107f448 gimple_cond_make_true
    /home/dimhen/src/gcc-current/gcc/gimple.h:2649
0x107f448 try_unroll_loop_completely
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:551
0x107f448 canonicalize_loop_induction_variables
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:670
0x107f982 tree_unroll_loops_completely(bool, bool)
    /home/dimhen/src/gcc-current/gcc/tree-ssa-loop-ivcanon.c:833

$ cat 1.ii
class A
{
};
template < int (*t_parser) () > class B
{
    virtual int parse ()
    {
        A a;
        t_parser ();
    }
};

extern "C" int f ();
class C:B < f >
{
};
void
g ()
{
    new C;
}

$ cat 2.ii
extern "C" int f ();
char *a[] = { 0, 0 };

void bar (char *);
int
f ()
{
    int i;
    while (1)
        bar (a[i++]);
}

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #9 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-24 10:00:18 UTC ---
(In reply to comment #7)
> This does not seem to reproduce for me. The following patch should however fix
> it.
> Index: tree-ssa-loop-ivcanon.c
> ===================================================================
> --- tree-ssa-loop-ivcanon.c     (revision 192711)
> +++ tree-ssa-loop-ivcanon.c     (working copy)
> @@ -364,6 +364,10 @@ loop_edge_to_cancel (struct loop *loop)
>         else
>           edge_to_cancel = EDGE_SUCC (edge_to_cancel->src, 0);
> 
> +      /* We only can handle conditionals.  */
> +      if (!(edge_to_cancel->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
> +       continue;
> +
>        /* We should never have conditionals in the loop latch. */
>        gcc_assert (edge_to_cancel->dest != loop->header);

i'll try patch. Thank You

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #10 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-24 12:50:01 UTC ---
(In reply to comment #7)
> This does not seem to reproduce for me. The following patch should however fix
> it.
> Index: tree-ssa-loop-ivcanon.c
> ===================================================================
> --- tree-ssa-loop-ivcanon.c     (revision 192711)
> +++ tree-ssa-loop-ivcanon.c     (working copy)
> @@ -364,6 +364,10 @@ loop_edge_to_cancel (struct loop *loop)
>         else
>           edge_to_cancel = EDGE_SUCC (edge_to_cancel->src, 0);
> 
> +      /* We only can handle conditionals.  */
> +      if (!(edge_to_cancel->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
> +       continue;
> +
>        /* We should never have conditionals in the loop latch. */
>        gcc_assert (edge_to_cancel->dest != loop->header);

192761 + patch
- pass bootstrap with --enable-checking=yes,df,fold,rtl,tree
- test from PR PASS

Thank You!

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

--- Comment #11 from Jan Hubicka <hubicka at gcc dot gnu.org> 2012-10-25 14:52:37 UTC ---
Author: hubicka
Date: Thu Oct 25 14:52:32 2012
New Revision: 192809

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192809
Log:

    PR tree-optimize/54980
    * tree-ssa-loop-ivcanon.c (constant_after_peeling): Fix
    obvious typo.
    (loop_edge_to_cancel): Be sure that the edge is from an conditional
    so we can cancel it.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-ssa-loop-ivcanon.c

[Bug lto/54980] [4.8 regression] gimple check: expected gimple_cond(error_mark), have gimple_call() in gimple_cond_set_lhs, at gimple.h:2578

[dimhen at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54980

Dmitry G. Dyachenko <dimhen at gmail dot com> changed:

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

--- Comment #12 from Dmitry G. Dyachenko <dimhen at gmail dot com> 2012-10-27 16:49:31 UTC ---
fixed in 192809