[Bug c/51481] New: ice: dead STMT in EH table

[Bug c/51481] New: ice: dead STMT in EH table

[dcb314 at hotmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

             Bug #: 51481
           Summary: ice: dead STMT in EH table
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dcb314@hotmail.com


Created attachment 26031
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26031
C source code

I just tried to compile the package abcm2ps-5.9.21-3
on latest trunk dated 20111208 on an AMD x86_64 box.

The compiler said

subs.c: In function 'error.constprop.3':
subs.c:217:6: error: dead STMT in EH table
# .MEM_37 = VDEF <.MEM_31>
__fprintf_chk (stderr.73_17, 1, "Error ");

subs.c:217:6: internal compiler error: verify_gimple failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Preprocessed source code attached. Flags -O3 -fexceptions required.

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-12-09
          Component|c                           |tree-optimization
      Known to work|                            |4.6.2
   Target Milestone|---                         |4.7.0
            Summary|ice: dead STMT in EH table  |[4.7 Regression] ice: dead
                   |                            |STMT in EH table
     Ever Confirmed|0                           |1

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-09 08:54:30 UTC ---
Catched at IPA inline transform time.  Confirmed, regression from 4.6,
reducing.

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

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

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

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-09 09:01:29 UTC ---
The following fails with -O -fexceptions -fipa-cp -fipa-cp-clone:

typedef __builtin_va_list __gnuc_va_list;
typedef __gnuc_va_list va_list;
extern struct _IO_FILE *stderr;
extern __const unsigned short int **__ctype_b_loc (void) 
     __attribute__ ((__nothrow__)) __attribute__ ((__const));
struct abcsym {
    unsigned short colnum;
    int linenum;
};
struct SYMBOL {
    struct abcsym as;
};
typedef struct SYMBOL *INFO[26];
extern INFO info;
void error(int sev,     struct SYMBOL *s,     char *fmt, ...);
void bug(char *msg, int fatal)
{
  error(1, 0, "Internal error: %s.", msg);
}
void error(int sev,     struct SYMBOL *s,     char *fmt, ...)
{
  va_list args;
  static struct SYMBOL *t;
  if (t != info['T' - 'A'])
    {
      char *p;
      t = info['T' - 'A'];
      while ((*__ctype_b_loc ())[(int) (((unsigned char) *p))])
        p++;
    }
  __builtin_fprintf(stderr, sev == 0 ? "Warning " : "Error ");
  if (s != 0)
      __builtin_fprintf(stderr, "in line %d.%d", s->as.linenum, s->as.colnum);
  __builtin_va_end(args);
}

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-12-12 09:22:06 UTC ---
Started with http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=181172

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |jakub at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-12-12 10:24:49 UTC ---
Created attachment 26053
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26053
gcc47-pr51481.patch

maybe_cleanup_or_replace_eh_stmt wasn't called, the patch fixes that.
Unfortunately, as it is deep inside of fold_stmt called from deep from various
passes, including cfg cleanup, I'm afraid handling eh edge dropping as opposed
to just replacing would be hard.  Let's just hope possibly throwing builtins
aren't optimized away.

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #26053|0                           |1
        is obsolete|                            |

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-12-12 10:48:22 UTC ---
Created attachment 26054
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26054
gcc47-pr51481.patch

Updated patch that won't fold if gimple_purge_dead_eh_edges would be needed.

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-12-12 20:59:31 UTC ---
Author: jakub
Date: Mon Dec 12 20:59:28 2011
New Revision: 182264

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=182264
Log:
    PR tree-optimization/51481
    * gimple-fold.c (gimple_fold_call): Call
    maybe_clean_or_replace_eh_stmt.  Avoid optimization if stmt has EH
    edges, but gimple_fold_builtin result can't throw.

    * gcc.dg/pr51481.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/pr51481.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/gimple-fold.c
    trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-12-12 21:02:38 UTC ---
Fixed.

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-13 11:24:44 UTC ---
The following is a real fix I'm going to test.

@@ -1601,6 +1605,7 @@ replace_uses_by (tree name, tree val)
       if (gimple_code (stmt) != GIMPLE_PHI)
        {
          gimple_stmt_iterator gsi = gsi_for_stmt (stmt);
+         gimple orig_stmt = stmt;
          size_t i;

          fold_stmt (&gsi);
@@ -1619,7 +1624,7 @@ replace_uses_by (tree name, tree val)
                recompute_tree_invariant_for_addr_expr (op);
            }

-         maybe_clean_or_replace_eh_stmt (stmt, stmt);
+         maybe_clean_or_replace_eh_stmt (orig_stmt, stmt);
          update_stmt (stmt);
        }
     }

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-13 12:56:13 UTC ---
Author: rguenth
Date: Tue Dec 13 12:56:09 2011
New Revision: 182280

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=182280
Log:
2011-12-13  Richard Guenther  <rguenther@suse.de>

    PR middle-end/51481
    * tree-cfg.c (replace_uses_by): Pass proper arguments to
    maybe_clean_or_replace_eh_stmt.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-cfg.c

[Bug tree-optimization/51481] [4.7 Regression] ice: dead STMT in EH table

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51481

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-12-13 14:01:04 UTC ---
Author: jakub
Date: Tue Dec 13 14:00:58 2011
New Revision: 182284

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=182284
Log:
    Revert
    2011-12-12  Jakub Jelinek  <jakub@redhat.com>

    PR tree-optimization/51481
    * gimple-fold.c (gimple_fold_call): Call
    maybe_clean_or_replace_eh_stmt.  Avoid optimization if stmt has EH
    edges, but gimple_fold_builtin result can't throw.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/gimple-fold.c