[Bug tree-optimization/58794] [4.8/4.9 Regression] ICE in set_lattice_value, at tree-ssa-ccp.c:455 on x86_64-linux-gnu (at -O1, -O2, and -O3)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58794

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue being that &a.f and &a.f are not equal because even with
OEP_CONSTANT_ADDRESS_OF set we get into

        case COMPONENT_REF:
          /* Handle operand 2 the same as for ARRAY_REF.  Operand 0
             may be NULL when we're called to compare MEM_EXPRs.  */
          if (!OP_SAME_WITH_NULL (0))
            return 0;
          flags &= ~OEP_CONSTANT_ADDRESS_OF;
          return OP_SAME (1) && OP_SAME_WITH_NULL (2);

and thus drop it before comparing the two FIELD_DECLs which have
TREE_SIDE_EFFECTS set.  Fixed with

Index: gcc/fold-const.c
===================================================================
--- gcc/fold-const.c    (revision 203886)
+++ gcc/fold-const.c    (working copy)
@@ -2715,10 +2715,11 @@ operand_equal_p (const_tree arg0, const_
        case COMPONENT_REF:
          /* Handle operand 2 the same as for ARRAY_REF.  Operand 0
             may be NULL when we're called to compare MEM_EXPRs.  */
-         if (!OP_SAME_WITH_NULL (0))
+         if (!OP_SAME_WITH_NULL (0)
+             || !OP_SAME (1))
            return 0;
          flags &= ~OEP_CONSTANT_ADDRESS_OF;
-         return OP_SAME (1) && OP_SAME_WITH_NULL (2);
+         return OP_SAME_WITH_NULL (2);

        case BIT_FIELD_REF:
          if (!OP_SAME (0))

I spotted this earlier but for some reason chickeded out to make this
change.  Hmm.  Trying to remember why ...