[Bug c++/46304] [4.5 Regression] g++ crashes with ICE in bitmap_first_set_bit, at bitmap.c:770

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46304

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Status|NEW                         |ASSIGNED
      Known to work|                            |4.4.4, 4.6.0
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |
   Target Milestone|---                         |4.5.2
            Summary|g++ crashes with ICE in     |[4.5 Regression] g++
                   |bitmap_first_set_bit, at    |crashes with ICE in
                   |bitmap.c:770                |bitmap_first_set_bit, at
                   |                            |bitmap.c:770
      Known to fail|                            |4.5.0, 4.5.1

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-05 
10:48:55 UTC ---
Confirmed.  Happens as we go into SSA form.

#2  0x00000000007ed310 in bitmap_first_set_bit (a=0x19c13e8)
    at /space/rguenther/src/svn/gcc-4_5-branch/gcc/bitmap.c:770
770       gcc_assert (elt);
(gdb) 
#3  0x0000000000c61587 in set_livein_block (var=0x7ffff7ef3d20, 
    bb=0x7ffff5b318f0)
    at /space/rguenther/src/svn/gcc-4_5-branch/gcc/tree-into-ssa.c:550
550           int def_block_index = bitmap_first_set_bit (db_p->def_blocks);

when visiting

(gdb) call debug_gimple_stmt (stmt)
D.2447 = COMPLEX_EXPR <D.2444, 2.0e+0>;

and the bitmap is empty.  D.2444 is not set before.

void ClassA<T>::setSize() [with T = __complex__ double] (struct ClassA * const
this)
{
  complex double D.2447;
  complex double * D.2446;

<bb 2>:
  D.2446 = this->data;
  D.2447 = COMPLEX_EXPR <D.2444, 2.0e+0>;
  *D.2446 = D.2447;
  return;

}

it looks like this caller to bitmap_first_set_bit assumes it's ok to
call it on an empty bitmap where it should return -1.

It's interesting though that 4.6 for the above has

void ClassA<T>::setSize() [with T = __complex__ double] (struct ClassA * this)
{
  complex double D.2421;
  double D.2420;
  complex double * D.2419;

<bb 2>:
  D.2419 = this->data;
  D.2420 = 1.0e+0 + 0.0;
  D.2421 = COMPLEX_EXPR <D.2420, 2.0e+0>;
  *D.2419 = D.2421;
  return;

}

which shows it's also really a wrong-code bug.