Re: volatile qualifier hurts single-threaded optimized case

[Benjamin Kosnik <bkoz at redhat dot com>]

> >   static inline _Atomic_word
> >   __attribute__ ((__unused__))
> >-  __exchange_and_add_dispatch(volatile _Atomic_word* __mem, int __val)
> >+  __exchange_and_add_dispatch(_Atomic_word* __mem, int __val)
> >  
> >
> Agreed, but doing this means that in the improvements to rope that 
> Richard (and Hans) also want we have to call __exchange_and_add and 
> __atomic_add directly,  because in that case the involved memory 
> locations are volatile and have to remain like that per Hans desiderata, 
> I gather. 

I find this somewhat confusing, but am going to go ahead with this, as
I have other fixes in the pipeline for this file.

Of course, it would be nice if everybody calls the dispatch functions
directly, since this idea removes having to wrap in __GTHREADS etc.
This interface should not penalize single-thread, however. 

If people want to do more specialized accesses, for instance when
threads are already assumed, or memory is already volatile, and want to
call down into __exchange_and_add and __atomic_add that seems
reasonable to me.

I believe Richard says as much in his follow up email.

This is going to be documented, as it's now complicated. (More
complicated?)  That's one of the patches I've got going....

> More generally, the original __exchange_and_add and 
> __atomic_add could be used also together with volatile memory and now 
> calling the "old" functions becomes necessary, we should document that 
> somewhere (if it weren't for this reason, everyone should always go 
> through the dispatchers).

Yep. 

best,
benjamin