[Bug middle-end/50178] New: [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[Bug middle-end/50178] New: [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[toon at moene dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

             Bug #: 50178
           Summary: [4.6 regression] ICE with gfortran -O3, not with
                    gfortran -02
    Classification: Unclassified
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: toon@moene.org


Created attachment 25089
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25089
Source code of the failing compilation.

Compiling the attached source code with gfortran 4.6.1 using optimization -O3
leads to the following Internal Compiler Error:

$ gfortran -O3 -c suedyn.f90 
suedyn.f90: In function ‘suedyn’:
suedyn.f90:10:0: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.

Compiling with -O2 instead succeeds (hence my suspicion that it is a middle-end
failure).

[Bug middle-end/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[toon at moene dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

Toon Moene <toon at moene dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-08-24
     Ever Confirmed|0                           |1

--- Comment #1 from Toon Moene <toon at moene dot org> 2011-08-24 17:21:23 UTC ---
The source code as attached is much reduced from the original, courtesy of
Steve Kargl.

Dominique Dhumieres wrote:

"It started to fail between revisions 158105 and 158252 and it has been
fixed on 4.7 between revisions 174599 and 175148."

in this note:

http://gcc.gnu.org/ml/fortran/2011-08/msg00198.html

[Bug middle-end/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
                 CC|                            |burnus at gcc dot gnu.org

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-08-24 17:42:00 UTC ---
>From http://gcc.gnu.org/ml/fortran/2011-08/msg00197.html

4.6.2-20110818 (Rev. 177878) crashes at:

==27008== Invalid read of size 8
==27008==    at 0x89A58F: vect_is_simple_use_1 (tree-vect-stmts.c:5346)
==27008==    by 0x89B4E5: vectorizable_operation (tree-vect-stmts.c:2440)
==27008==    by 0x89F452: vect_transform_stmt (tree-vect-stmts.c:4841)
==27008==    by 0x8ACB23: vect_transform_loop (tree-vect-loop.c:4872)
==27008==    by 0x8B66E9: vectorize_loops (tree-vectorizer.c:205)

[Bug middle-end/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2011-08-24 18:20:01 UTC ---
The code compiles with gfortran 4.6.1 and -O3 -ffast-math.

[Bug middle-end/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

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

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-24 18:52:07 UTC ---
Fixed by http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=175074 (or perhaps
just made latent, hard to say).

[Bug tree-optimization/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |tree-optimization
   Target Milestone|---                         |4.6.2

[Bug tree-optimization/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

Ira Rosen <irar at il dot ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |irar at il dot ibm.com

--- Comment #5 from Ira Rosen <irar at il dot ibm.com> 2011-09-01 07:01:07 UTC ---
The problem here is that in vectorizable_call we replace the original call with
a dummy stmt and also remove its stmt_vec_info, which causes the segfault when
we try to access it through related pattern stmt. I'm testing the following
patch that updates related pattern stmt to be the dummy stmt:

Index: tree-vect-stmts.c
===================================================================
--- tree-vect-stmts.c   (revision 178373)
+++ tree-vect-stmts.c   (working copy)
@@ -1583,6 +1583,14 @@ vectorizable_call (gimple stmt, gimple_stmt_iterat
   new_stmt = gimple_build_assign (gimple_call_lhs (stmt),
                                  build_zero_cst (type));
   set_vinfo_for_stmt (new_stmt, stmt_info);
+  /* For pattern statements make the related statement to point to
+     NEW_STMT in order to be able to retrieve the original statement
+     information later.  */
+  if (is_pattern_stmt_p (stmt_info))
+    {
+      gimple related = STMT_VINFO_RELATED_STMT (stmt_info);
+      STMT_VINFO_RELATED_STMT (vinfo_for_stmt (related)) = new_stmt;
+    }
   set_vinfo_for_stmt (stmt, NULL);
   STMT_VINFO_STMT (stmt_info) = new_stmt;
   gsi_replace (gsi, new_stmt, false);
@@ -4957,11 +4965,7 @@ vect_transform_stmt (gimple stmt, gimple_stmt_iter
             the stmt_info of ORIG_STMT_IN_PATTERN.  See more details in the
             documentation of vect_pattern_recog.  */
          if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
-           {
-             gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo)
-                           == orig_scalar_stmt);
-             STMT_VINFO_VEC_STMT (stmt_vinfo) = vec_stmt;
-           }
+           STMT_VINFO_VEC_STMT (stmt_vinfo) = vec_stmt;
        }
     }


r175074 really fixed this problem since now we don't need to retrieve the
original def stmt.

[Bug tree-optimization/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[irar at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

--- Comment #6 from irar at gcc dot gnu.org 2011-09-01 08:29:41 UTC ---
Author: irar
Date: Thu Sep  1 08:29:33 2011
New Revision: 178396

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=178396
Log:

        PR tree-optimization/50178
        * tree-vect-stmts.c (vectorizable_call): Update the related
        pattern statement before deleting the original call.
        (vect_transform_stmt): Don't expect the related pattern statement
        to match the original statement after transformation.


Added:
    branches/gcc-4_6-branch/gcc/testsuite/gfortran.dg/vect/pr50178.f90
Modified:
    branches/gcc-4_6-branch/gcc/ChangeLog
    branches/gcc-4_6-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_6-branch/gcc/tree-vect-stmts.c

[Bug tree-optimization/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[irar at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

--- Comment #7 from irar at gcc dot gnu.org 2011-09-01 08:47:05 UTC ---
Author: irar
Date: Thu Sep  1 08:46:59 2011
New Revision: 178397

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=178397
Log:

        PR tree-optimization/50178
        * gfortran.dg/vect/pr50178.f90: New test.


Added:
    trunk/gcc/testsuite/gfortran.dg/vect/pr50178.f90
Modified:
    trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/50178] [4.6 regression] ICE with gfortran -O3, not with gfortran -02

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50178

Ira Rosen <irar at il dot ibm.com> changed:

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

--- Comment #8 from Ira Rosen <irar at il dot ibm.com> 2011-09-01 09:05:01 UTC ---
Fixed.