Re: [PATCH] Fix PR35607, another revenge of invariant addresses (ivopts)

[Diego Novillo <dnovillo at google dot com>]

On 03/17/08 09:27, Richard Guenther wrote:

> + /* Return true if T is a gimple invariant address.  */
> + 
> + bool
> + is_gimple_invariant_address (const_tree t)
> + {
> +   tree op;
> + 
> +   if (TREE_CODE (t) != ADDR_EXPR)
> +     return false;
> + 
> +   op = TREE_OPERAND (t, 0);
> +   while (handled_component_p (t))
> +     {
> +       switch (TREE_CODE (t))
> + 	{
> + 	case ARRAY_REF:
> + 	case ARRAY_RANGE_REF:
> + 	  if (!is_gimple_constant (TREE_OPERAND (t, 1))
> + 	      || TREE_OPERAND (t, 2) != NULL_TREE
> + 	      || TREE_OPERAND (t, 3) != NULL_TREE)
> + 	    return false;
> + 	  break;
> + 
> + 	case COMPONENT_REF:
> + 	  if (TREE_OPERAND (t, 2) != NULL_TREE
> + 	      || TREE_OPERAND (t, 3) != NULL_TREE)
> + 	    return false;
> + 	  break;
> + 
> + 	default:;
> + 	}
> +       op = TREE_OPERAND (t, 0);
> +     }
> + 
> +   return TREE_INVARIANT (op);

This looks wrong.  It will return false for '&a' where 'a' is a local 
variable.  I think I would rather move the logic from 
recompute_tree_invariant_for_addr_expr.  This needs to handle statics 
and TLS objects.

> + }
> + 
> + /* Return true if T is a GIMPLE minimal invariant.  It's a restricted
> +    form of function invariant.  */
> + 
> + bool
> + is_gimple_min_invariant (const_tree t)
> + {
> +   if (TREE_CODE (t) == ADDR_EXPR)
> +     return is_gimple_invariant_address (t);
> + 
> +   return is_gimple_constant (t);

You mean is_gimple_const() here.