Please support the _Atomic keyword in C++

[Jonathan Wakely]

On 15 May 2018 at 11:01, Rodrigo V. G. <rodrivg@gmail.com> wrote:
> On Tue, May 15, 2018 at 12:27 AM, Jonathan Wakely <jwakely.gcc@gmail.com> wrote:
>> On 14 May 2018 at 22:32, Rodrigo V. G. <rodrivg@gmail.com> wrote:
>>> In addition to the bug:
>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60932
>>> I wanted to add some comment:
>>>
>>> It would be very useful if the _Atomic keyword would be supported in C++.
>>> This way the <stdatomic.h> header could be included inconditionally in C++ code.
>>> Even if it is not compatible with the C++ <atomic> header, it would be useful.
>>>
>>> Supporting the _Atomic keyword in C++ would benefit at least two cases:
>>>
>>> - When mixing C and C++ code for interoperability (using, in C++, some
>>> variables declared as _Atomic in a C header).
>>>
>>> - When developing operating systems or kernels in C++, in a
>>> freestanding environment (cross compiler), <atomic> is not available,
>>
>> Why not? It's part of a freestanding C++ implementation.
>
> When building a cross compiler as indicated in:
> https://wiki.osdev.org/GCC_Cross-Compiler#GCC
> it does not install the <atomic> header.

That's not a problem with GCC, that's a problem with those instructions.

If you want a freestanding C++ implementation then configure GCC with
--disable-hosted-libstdcxx and build+install libstdc++. Those
instructions do neither of those things, so it's unsurprising you
don't get a proper freestanding C++ implementation.

>>> but <stdatomic.h> is.
>>
>> How? It's not part of any C++ implementation at all, freestanding or not.
>
> As far as I can see the <stdatomic.h> can be included from C++ code
> (also in the cross compiler).
> Only that it complains about _Atomic and _Bool so it does not work.
> So _Atomic and _Bool seem to be the missing pieces.

So you're saying it can be included, it just doesn't work.

Great!

That's true for this header too:

#error "This header doesn't compile"



>>> So to correctly use things like __atomic_fetch_add in C++ in
>>> freestanding mode, this is the only way. Otherwise one cannot use
>>> atomics at all in these conditions.
>>
>> Why can't you use __atomic_fetch_add directly?
>
> I tried to use __atomic_fetch_add in C++ with a volatile (non _Atomic) variable,
> and it seems to generate the same assembler code.

Why use volatile?
http://isvolatileusefulwiththreads.com/

> The only difference that I saw was that with _Atomic
> it generates a "mfence" instruction after initialization but with
> volatile it does not.
> So I think it might not provide the same guarantees.
>
>
> (Sorry, I forgot to cc the list, now I do cc)