[Bug middle-end/45172] New: [4.6 Regression] internal compiler error: verify_flow_info failed

[Bug middle-end/45172] New: [4.6 Regression] internal compiler error: verify_flow_info failed

[jv244 at cam dot ac dot uk]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1711 bytes --]

recent trunk (4.6.0 20100803 (experimental) [trunk revision 162839]) fails
with:

> gfortran -c  -fopenmp -O1 -fexceptions bug.f90
bug.f90: In function ‘dbcsr_mult_m_e_e’:
bug.f90:1:0: error: verify_flow_info: Duplicate edge 28->29
bug.f90:1:0: internal compiler error: verify_flow_info failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.


> cat bug.f90
  SUBROUTINE dbcsr_mult_m_e_e ( )
    LOGICAL, PARAMETER   :: use_combined_types = .FALSE.
    INTEGER, ALLOCATABLE, DIMENSION(:, :) ::  right_index_sr
    INTEGER, ALLOCATABLE, DIMENSION(:, :, :) ::  my_sizes
    INTEGER, ALLOCATABLE,  DIMENSION(:, :, :, :) :: all_sizes
    ALLOCATE (all_sizes(4, LBOUND(my_sizes,2):UBOUND(my_sizes,2), &
              LBOUND(my_sizes,3):UBOUND(my_sizes,3), 0:numnodes-1))
          IF (use_combined_types) THEN
             CALL mp_waitall (right_index_sr)
          ENDIF
          DO ki = 0, min_nimages-1
!$omp parallel default (none) &
!$omp reduction (+: flop_single, t_all, t_dgemm)
!$omp end parallel
          ENDDO
       checksum = dbcsr_checksum (product_matrix, error)
  END SUBROUTINE dbcsr_mult_m_e_e


older gcc (gcc-4_5-branch revision 162409) is fine.


-- 
           Summary: [4.6 Regression] internal compiler error:
                    verify_flow_info failed
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jv244 at cam dot ac dot uk


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[dominiq at lps dot ens dot fr]

------- Comment #1 from dominiq at lps dot ens dot fr  2010-08-03 15:27 -------
Confirmed, trunk r162490 works.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2010-08-03 15:47 -------
Confirmed.  Triggered by ehcleanup.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-08-03 15:47:57
               date|                            |
   Target Milestone|---                         |4.6.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2010-08-03 15:49 -------
  # finally_tmp.15_17 = PHI <1(28), 0(24), 0(22), 0(28), 0(26)>

two times incoming edge from block 28, once with value 1 once with 0.  oops.

<bb 26>:
  _gfortran_runtime_error_at (&"At line 6 of file t.f90"[1]{lb: 1 sz: 1},
&"Attempting to allocate already allocated variable \'%s\'"[1]{lb: 1 sz: 1},
&"all_sizes"[1]{lb: 1 sz: 1});

so we shouldn't have any edge from that block.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[jakub at gcc dot gnu dot org]

------- Comment #4 from jakub at gcc dot gnu dot org  2010-08-19 13:56 -------
Well, the edge from the block ending with _gfortran_runtime_error_at is an EH
edge, not normal edge.

Not sure what exactly Fortran needs -fexceptions for currently, one reason
could be pthread_cancel, but at least with OpenMP pthread_cancel isn't going to
do very nice things anyway.

The two (28) phi args seem to be because the successor block of the empty EH
block that is being removed in ehcleanup already had another incoming edge from
that block, and ehcleanup doesn't expect that.  Not sure if that is a bug from
before ehcleanup, or if ehcleanup should test for that case and either disallow
the optimization, or resolve it somehow else.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[jv244 at cam dot ac dot uk]

------- Comment #5 from jv244 at cam dot ac dot uk  2010-08-19 14:24 -------
(In reply to comment #4)

> Not sure what exactly Fortran needs -fexceptions for currently, one reason
> could be pthread_cancel, but at least with OpenMP pthread_cancel isn't going to
> do very nice things anyway.

We have a mixed MPI/OMP parallel Fortran code, -fexceptions is being added
automagically by the openmpi compiler driver (mpif90). This might be to allow
mixed C++/C/Fortran codes to work (but I couldn't quite find docs about this on
the openmpi webpage). 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-15 13:18:50 UTC ---
We indeed have before ehcleanup:

  # BLOCK 26
  # PRED: 24 (false,exec) 25 (true,exec)
  dbcsr_checksum (&product_matrix, &error);
  # SUCC: 31 (eh,exec) 27 (fallthru,exec)

  # BLOCK 27
  # PRED: 26 (fallthru,exec) 31 (fallthru,exec)
  # finally_tmp.15_17 = PHI <1(26), 0(31)>
  # all_sizes$data_322 = PHI <D.1562_274(26), all_sizes$data_320(31)>

...

  # BLOCK 31
  # PRED: 21 (eh,exec) 23 (eh,exec) 26 (eh,exec)
  # all_sizes$data_320 = PHI <all_sizes$data_318(21), all_sizes$data_318(23),
D.1562_274(26)>
<L71>:
  goto <bb 27>;
  # SUCC: 27 (fallthru,exec)

and we thread that jump leading to the duplicate edge (which verify_flow_info
isn't happy about).  Maybe that's the only thing to fix.

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-15 13:59:50 UTC ---
Index: gcc/cfghooks.c
===================================================================
--- gcc/cfghooks.c      (revision 166751)
+++ gcc/cfghooks.c      (working copy)
@@ -156,9 +156,21 @@ verify_flow_info (void)
        {
          if (last_visited [e->dest->index] == bb)
            {
-             error ("verify_flow_info: Duplicate edge %i->%i",
-                    e->src->index, e->dest->index);
-             err = 1;
+             edge e2;
+             edge_iterator ei2;
+             /* Check if we really have an edge with the same EH status.  */
+             FOR_EACH_EDGE (e2, ei2, bb->succs)
+               {
+                 if (e2 == e)
+                   break;
+                 if (e2->dest == e->dest
+                     && (e2->flags & EDGE_EH) == (e->flags & EDGE_EH))
+                   {
+                     error ("verify_flow_info: Duplicate edge %i->%i",
+                            e->src->index, e->dest->index);
+                     err = 1;
+                   }
+               }
            }
          if (e->probability < 0 || e->probability > REG_BR_PROB_BASE)
            {


but it's probably not the brightest idea as several places in GCC search
edges by just matching src and dest.

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

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

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

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-15 14:36:10 UTC ---
Mine.

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

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

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

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-16 11:43:07 UTC ---
Fixed.

[Bug middle-end/45172] [4.6 Regression] internal compiler error: verify_flow_info failed

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45172

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-16 11:42:53 UTC ---
Author: rguenth
Date: Tue Nov 16 11:42:50 2010
New Revision: 166794

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166794
Log:
2010-11-16  Richard Guenther  <rguenther@suse.de>

    PR middle-end/45172
    * tree-eh.c (cleanup_empty_eh_unsplit): Avoid creating duplicate
    edges.

    * gfortran.dg/gomp/pr45172.f90: New testcase.

Added:
    trunk/gcc/testsuite/gfortran.dg/gomp/pr45172.f90
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-eh.c