Re: Threads: Atomic Counter

From: William Kempf (sirwillard_at_[hidden])
Date: 2000-08-31 13:31:43

--- In boost_at_[hidden], "William Kempf" <sirwillard_at_m...> wrote:
> --- In boost_at_[hidden], "Jeff Paquette" <paquette_at_m...> wrote:

This discussion isn't really on topic, since the exchange() is the 
only portable avenue open to us for reading/writing.  So, if anyone 
wants us to take this offline I'll do so.

> > mov eax, [b]   ;;; pseudo x86 assembly - nits not wanted.
> > mov [a], eax

I've thought a bit more about this.  So what if there's a task switch 
between these two calls?  I don't believe that Interlocked* commands 
insure that there won't be a task switch, only that there won't be 
any side effects caused by such a task switch.  That's going to be 
all that we can gaurantee in a portable library, any way.

If there were a task switch between these two calls the register 
would be preserved.  So the result in 'a' will be equivalent here, 
which is all that you're trying to gaurantee any way.  (Well, 
actually, the InterlockedExchange must also insure that the return 
value is consistent, but that's going beyond the claim made by the 
Win32 documentation about basic read/write being atomic.)

Bill Kempf