Bang Jun-Young wrote:
quoted text
> I'd say nobody would write an atomic op in pure C. :-) The first one was 
> just
> a conceptual implementation with an if omitted (yes, it was a mistake).

Yes, nobody would probably write it in C, for various reasons.
But it wasn't an if that was missing in the concept, it was a lock.

quoted text
> The
> difference is, Jason undestood what I tried to explain, and you didn't, 
> maybe
> because you didn't read Solaris sources before.

If Jason understood you, then I guess we should all be happy. The reason 
I didn't understand you was because your conceptual implementation was 
broken. :-)

If we were to describe conceptually, without worrying about the fact 
that it wouldn't work, then I'd write it like this:

uint32_t compare_and_swap(uint32_t *target, uint32_t cmp, uint32_t new)
{
   uint32_t old = *target;
   if (*target == cmp) *target = new;
   return old;
}

But that also does away with the whole issue about choosing between 
compare_and_swap() and compare_and_store() since that is about wether 
one would be easier in general to implement on machines. Obviously, from 
this conceptual description, either would be just as easy.

But I'm happy to leave it at that. Just call me stupid. :-)

	Johnny