[Bug middle-end/70240] [6 Regression] ICE: in gimplify_modify_expr, at gimplify.c:4854 with -ftree-vectorize

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70240

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2016-03-15
   Target Milestone|---                         |6.0
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
1455      return gimplify_build3 (gsi, code, inner_type, unshare_expr (cond),
a, b);
1456    }
1457
1458    /* Expand a vector COND_EXPR to scalars, piecewise.  */
1459    static void
(gdb) p code
$1 = COND_EXPR
(gdb) p cond
$2 = <ge_expr 0x7ffff69a07a8>
(gdb) p debug_generic_expr (a)
BIT_FIELD_REF <_205, 32, 0>
$3 = void
(gdb) p debug_generic_expr (b)
BIT_FIELD_REF <{_189, _187, _186, _184, _183, _182, _181, _180, _179, _178,
_177, _176, _175, _174, _173, _172}, 32, 0>
$4 = void

and we get side-effects somehow during gimplification of the COND_EXPR.  Better
gimplify a and b before building the COND_EXPR.