On 25/03/15 12:04 -0700, Richard Henderson wrote:
>On 03/25/2015 11:49 AM, Jonathan Wakely wrote:
>> On 25/03/15 11:36 -0700, Richard Henderson wrote:
>>> On 03/25/2015 09:22 AM, Jonathan Wakely wrote:
>> On 25/03/15 11:39 -0700, Richard Henderson wrote:
>>> On 03/25/2015 09:22 AM, Jonathan Wakely wrote:
>>>> +static_assert( alignof(std::atomic<twoints>) > alignof(int),
>>>> +               "std::atomic not suitably aligned" );
>>>
>>> This is only true if int64_t has alignment larger than int32_t,
>>> which is unfortunately not always the case.
>>
>> Huh, didn't realise that. I could change the tests to check it's
>> alignof(std::int64_t) as the next assertion does, but is it safe to
>> assume that struct twoints { int a; int b; } is exactly 64 bits
>> everywhere?
>
>Certainly not.  But if you're going to explicitly use int64_t elsewhere, you
>might as well explicitly use int32_t as well.  Then I believe you can
>reasonably assert
>
>  alignof(twoint32) == alignof(int64_t)

Yes, that makes sense, thanks.  Here's what I plan to commit then.

This includes your fix to avoid decreasing alignment, but I didn't add
a test for that as I couldn't make it fail on any of the targets I
test on.