On Thu, Jun 30, 2016 at 1:17 PM, Manish Goregaokar <man...@mozilla.com> wrote:
> gcc/ChangeLog:
> PR c/71699
> * fold-const.c (tree_binary_nonzero_warnv_p): Allow
> pointer addition to also be considered nonzero.
> ---
> gcc/fold-const.c | 20 +++++++++++++-------
> 1 file changed, 13 insertions(+), 7 deletions(-)
>
> diff --git a/gcc/fold-const.c b/gcc/fold-const.c
> index 3b9500d..eda713e 100644
> --- a/gcc/fold-const.c
> +++ b/gcc/fold-const.c
> @@ -13200,16 +13200,22 @@ tree_binary_nonzero_warnv_p (enum tree_code code,
> {
> case POINTER_PLUS_EXPR:
> case PLUS_EXPR:
> - if (ANY_INTEGRAL_TYPE_P (type) && TYPE_OVERFLOW_UNDEFINED (type))
> + if ((ANY_INTEGRAL_TYPE_P (type) && TYPE_OVERFLOW_UNDEFINED (type))
> + || POINTER_TYPE_P (type))
> {
> + /* Pointers are always nonnegative, check integers. */
> + if (ANY_INTEGRAL_TYPE_P (type))
> + {
> + sub_strict_overflow_p = false;
> + if (!tree_expr_nonnegative_warnv_p (op0,
> + &sub_strict_overflow_p)
> + || !tree_expr_nonnegative_warnv_p (op1,
> + &sub_strict_overflow_p))
> + return false;
> + }
> /* With the presence of negative values it is hard
> to say something. */
> - sub_strict_overflow_p = false;
> - if (!tree_expr_nonnegative_warnv_p (op0,
> - &sub_strict_overflow_p)
> - || !tree_expr_nonnegative_warnv_p (op1,
> - &sub_strict_overflow_p))
> - return false;
> +
Hmm, note that op1 of POINTER_PLUS_EXPR _is_ an integer that needs to be treated
as signed.
POINTER_PLUS_EXPR is special in other ways in that iff
flag_delete_null_pointer_checks
is set we can assume that ptr + non-zero is never zero. Thus simply do
case POINTER_PLUS_EXPR:
return flag_delete_null_pointer_checks
&& (tree_expr_nonzero_warnv_p (op0, strict_overflow_p)
|| tree_expr_nonzero_warnv_p (op1, strict_overflow_p));
OTOH ptr + -(uintptr_t)ptr is valid from a POINTER_PLUS_EXPR
perspective as GCC does
not have a special tree code for pointer subtraction (but IIRC it uses
MINUS_EXPR on integers
for this).
Richard.
> /* One of operands must be positive and the other non-negative. */
> /* We don't set *STRICT_OVERFLOW_P here: even if this value
> overflows, on a twos-complement machine the sum of two
> --
> 2.8.3
