552
Subj : Re: A question about atomic_ptr
To   : comp.programming.threads
From : Peter Dimov
Date : Sat Apr 16 2005 07:54 pm

Joe Seigh wrote:
> On Sat, 16 Apr 2005 15:52:09 +0300, Peter Dimov <pdimov@gmail.com>
> wrote:
>> But there's one other thing: shared_ptr is two pointers, not one.
>> Without DWCAS, or at least double-wide atomic loads and stores, even
>> hazards aren't enough to make it atomic. :-(
>
> Sounds like it's using separate refcount and object pointers to make
> dereferencing the object less expensive, as opposed to keeping the
> object pointer with the refcount which is what I do with atomic_ptr. It's 
> an implementation issue.

It's also an interface issue; a separate pointer is needed for pointer 
conversions (shared_ptr<derived> to shared_ptr<base> or shared_ptr<void>).

But if I place a limit of at most one writer... would I be able to get away 
with something like:

l1:

load pointer to object
load pointer to count
set hazard pointer

fence

compare pointer to object
compare pointer to count
compare hazard pointer

if mismatch goto l1

?

Then

l2:

load count
if zero goto l3

attempt to store count+1, goto l2 if not successful
return (pobject, pcount)

l3:

set hazard pointer to zero
scan hazards and deallocate count if not referenced
return (NULL, NULL)

?

.

0