[Bug fortran/24424] New: Equivalence broken since 20051013

["paul dot richard dot thomas at cea dot fr" <gcc-bugzilla at gcc dot gnu dot org>]

This test, which is reduced from a test that Jakub reduced from the Polyhedron
test_fpu.f90, segfaults in execution.

  integer, dimension (8388608) :: i, j, k
  equivalence (i(1), j(10))
  i = 5
  k = i
end

$ /irun/bin/gfortran -fmax-stack-var-size=1000000  JAKUBS_TEST.f90

THOMASP@PC-THOMAS-P /cygdrive/d/gfortran/pb04/lin/f90
$ ./a
Segmentation fault (core dumped)

It does this for any optimization/options that I have tried.

The code produced is
MAIN__ ()
{
  static union
  {
    int4 j[8388608];
    int4 i[8388608];
  } equiv.0;
  static int4 j[8388608];
  static int4 i[8388608];
  static int4 k[8388608];

  {
    int4 S.0;

    S.0 = 1;
    while (1)
      {
        if (S.0 > 8388608) goto L.1; else (void) 0;
        i[NON_LVALUE_EXPR <S.0> + -1] = 5;
        S.0 = S.0 + 1;
      }
    L.1:;
  }
  {
    int4 S.1;

    S.1 = 1;
    while (1)
      {
        if (S.1 > 8388608) goto L.2; else (void) 0;
        k[NON_LVALUE_EXPR <S.1> + -1] = i[NON_LVALUE_EXPR <S.1> + -1];
        S.1 = S.1 + 1;
      }
    L.2:;
  }
}

whereas a working version of gfortran produces

MAIN__ ()
{
  static union
  {
    int4 j[8388608];
    int4 i[8388608];
  } equiv.0;
  static int4 k[8388608];

  {
    int4[8388608] * D.481;

    D.481 = &equiv.0.i;
    {
      int4 S.0;

      S.0 = 1;
      while (1)
        {
          if (S.0 > 8388608) goto L.1; else (void) 0;
          (*D.481)[NON_LVALUE_EXPR <S.0> + -1] = 5;
          S.0 = S.0 + 1;
        }
      L.1:;
    }
  }
  {
    int4[8388608] * D.484;

    D.484 = &equiv.0.i;
    {
      int4 S.1;

      S.1 = 1;
      while (1)
        {
          if (S.1 > 8388608) goto L.2; else (void) 0;
          k[NON_LVALUE_EXPR <S.1> + -1] = (*D.484)[NON_LVALUE_EXPR <S.1> + -1];
          S.1 = S.1 + 1;
        }
      L.2:;
    }
  }
}

Notice the appearance in the broken version of non-union static arrays for i
and j, which are what the subsequent code refers to.  I understand from the
list that this is not what it appears to be but.... looks, walks and squits
like a duck to me!

Paul T


-- 
           Summary: Equivalence broken since 20051013
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: paul dot richard dot thomas at cea dot fr


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24424