Re: [PATCH] Add stdatomic.h for clang>=3.1 and gcc>=4.1

[Joakim Sindholt <opensource@zhasha.com>]

On Sat, 2014-11-22 at 20:43 -0500, Rich Felker wrote:
> On Sun, Nov 23, 2014 at 02:31:35AM +0100, Jens Gustedt wrote:
> > Hi Rich,
> > 
> > Am Samstag, den 22.11.2014, 18:30 -0500 schrieb Rich Felker:
> > > atomic_flag is not viable for this because it does not have a
> > > wait/wake mechanism. You'd be spinning, which means in processes with
> > > different priorities involved, you could easily get deadlock if the
> > > lower-priority thread got suspended while holding the lock. You really
> > > do need mutexes.
> > 
> > I am probably still too much thinking in C11, only, which doesn't have
> > the notion of priorities.
> > 
> > Actually, I think a specially cooked synchronization tool would be
> > better. Something that combines an atomic pointer (to point to the
> > object) with a futex living on it for the waiting. This would probably
> > be a bit more challenging to implement, but here we really have an
> > interest to have the fast path really fast, just one CAS of the
> > pointer.
> 
> I don't get what you mean. To access an atomic object larger than the
> hardware supports, you have to hold a lock for the whole interval of
> reading/writing. This is O(n) in the size of the object. I don't see
> where your ideas about pointers and CAS are coming in.
> 
> > > > What has all of this to do with VLA? I am lost.
> > > 
> > > The operands of __typeof__ and sizeof get evaluated when they have VLA
> > > type. I think this is the problem.
> > 
> > ah, ok
> > 
> > No, this isn't a problem, I think. Arrays aren't allowed to be subject
> > of an _Atomic qualification (arrays are never qualified
> > themselves). For _Atomic type, the standard explicitly excludes
> > arrays. So arrays in general and VLA in particular should never be
> > passed as such into any of these generic functions, only pointers to
> > atomic objects can.
> 
> Is a pointer to a variably modified type considered variably modified?
> If so maybe these are affected too...
> 
> > > > > I have changed it to be an atomic_bool in a struct as both GCC and Clang
> > > > > has it in a struct. Presumably to separate it from the generic _Atomic
> > > > > stuff.
> > > > 
> > > > Again, since we want to have ABI compatibility, it is not your choice
> > > > to make. You'd simply have to stick to the choice that gcc made. So
> > > > you have to copy the declaration of the struct, including all the
> > > > ifdef fuzz.
> > > 
> > > I'd have to look at it again, but IIRC only one case of the #ifdef
> > > mess was actually possible. The others were for hypothetical archs
> > > without real atomics which we can't support anyway.
> > 
> > We should have it as a struct, if the implementations have it like
> > that, I think:
> > 
> >  - It should have same alignment properties for ABI compatibility.
> >  - It should lead to the same typename when included in C++.
> 
> Yes.
> 
> > The ifdef is a single one to switch between _Bool or unsigned char or
> > so.
> 
> Yes, but I think the #ifdef always comes out one way anyway, though I
> don't remember which one and don't have the file in front of me.

GCC 4.9:

typedef _Atomic struct
{
#if __GCC_ATOMIC_TEST_AND_SET_TRUEVAL == 1
  _Bool __val;
#else
  unsigned char __val;
#endif
} atomic_flag;

Clang 3.6:

#ifdef __cplusplus
typedef _Atomic(bool)               atomic_bool;
#else
typedef _Atomic(_Bool)              atomic_bool;
#endif

typedef struct atomic_flag { atomic_bool _Value; } atomic_flag;