Subj : Re: [.NET] Volatile Fields
To   : comp.programming.threads
From : Joe Seigh
Date : Sat Sep 17 2005 08:52 am

Cool Guy wrote:
> David Schwartz <davids@webmaster.com> wrote:
> 
> 
>>>Shouldn't Test.result be declared as volatile aswell?  My understanding is
>>>that it should because otherwise the write to it in Thread2 might not be
>>>visible to other threads, since the value may be written to a cache 
>>>instead
>>>of to main memory.
>>
>>    It does not have to be on this platform because, as the page says:
> 
> 
> <snip>
> 
> See, this is where my understanding goes out of the window.
> 
> From reading that page it seems there's only one issue here: that
> reads/writes can be re-ordered and thus just making *one* of those fields
> volatile is sufficient.
> 
> But my understanding is that that there are in fact *two* (separate) issues
> here:
> 
>     a. the above; and
>     b. that reads may come from a cache instead of from main memory (and
>        that writes may go to a cache instead of to main memory).


Cache has no effect on read/write ordering.  It's the memory model that does.
You could have a system without cache and it would still have the same issues
with read/write ordering.  So trying to imagine what is happening with a
hardware entity, which by definition you can't see, only serves to confuse
yourself.

-- 
Joe Seigh

When you get lemons, you make lemonade.
When you get hardware, you make software.

.