[cpp-threads] Failed compare-and-swap

[Howard Hinnant]

On Aug 1, 2007, at 1:17 PM, Herb Sutter wrote:

> It seems like there are two legitimate choices:
>
>  - CAS returns the old value, in which case a failed CAS should at  
> least by default have acquire semantics because it's a read.
>
>  - CAS returns only a success/fail boolean, in which case a failed  
> CAS need not have any ordering semantics.
>
> Does that make sense? It seems we can happily choose either, but the  
> choice affects the return type.

Lawrence recently generously sent me the following semantics for the  
compare_swap function proposed in N2324:

bool atomic_compare_swap( volatile atomic_integral* a, integral*  
old_v, integral new_v)
{
    if (a == x)
    {
        *a = new_v;
        return true;
     }
     *old_v = *a;
     return false;
}

I've been using this to good effect in the mutex work I've been doing:

         ...
         do
         {
             count = old_state & __num_waiting;
             --count;
             new_state = (old_state & ~__num_waiting) | count;
         } while (!compare_swap(&__state_, &old_state, new_state));
         old_state = new_state;
         ...

Having both the "false" and the updated state in "old_state" on a  
failed CAS is extremely convenient.  I really like the N2324 interface  
in this regard.  And if I'm understanding correctly, this means a  
failed CAS should have read/acquire semantics.

-Howard