On Sat, Oct 24, 2015 at 5:14 AM, Jason Merrill <[email protected]> wrote:
> On 10/21/2015 12:16 AM, Richard Biener wrote:
>>
>> On Tue, Oct 20, 2015 at 9:10 PM, Jason Merrill <[email protected]> wrote:
>>>
>>> I made this change on the delayed folding branch and then noticed that it
>>> broke pointer-arith-10.c, which you added to the testsuite. The patch
>>> changes the -original dump from
>>>
>>> return (char *) ((sizetype) p + (sizetype) i);
>>>
>>> to
>>>
>>> return (char *) i + (sizetype) p;
>>>
>>> It's not clear to me why the former should be preferred. Any thoughts?
>>
>>
>> We probably regressed for the former and the dump-scanning just didn't
>> notice. We wanted to check for
>>
>> return p + (sizetype) i;
>>
>> at least GCC 4.4.7 produces that and 4.5.2 regressed.
>>
>> Some time ago we had a folding that explicitely swapped pointer-ness
>> of an integer op like the testcase was supposed to test. But I remember
>> I removed this because it's incorrect (pointer arithmetic is more
>> constrained
>> than unsigned integer arithmetic):
>>
>> 2009-01-16 Richard Guenther <[email protected]>
>>
>> PR tree-optimization/38835
>> PR middle-end/36227
>> * fold-const.c (fold_binary): Remove PTR + INT -> (INT)(PTR p+
>> INT)
>> and INT + PTR -> (INT)(PTR p+ INT) folding.
>> * tree-ssa-address.c (create_mem_ref): Properly use
>> POINTER_PLUS_EXPR.
>>
>> so I think the testcase should be simply removed.
>
>
> How about this change to the testcase?
Hmm, I'd say we want to test for the correctness issue as well, thus pleas also
add
/* { dg-final { scan-tree-dump-not "return p +" "original" } } */
as we do not want 'p + (sizetype)i' as that is incorrect.
Thanks,
Richard.
> Jason
>
>
