From: Andrew J Bromage (ajb_at_[hidden])
Date: 2002-08-07 19:28:36
G'day all.
On Thu, Aug 08, 2002 at 09:07:56AM +1000, Andrew J Bromage wrote:
> On some architectures we have an atomic_count operation which would
> avoid the lock:
...actually, now that I think about it, this only works if the
atomic_count operations issue full barriers, which I suppose is not
guaranteed.
Given that double-checked locking is so hard to get correct and
portable, I think that's all the more reason to provide an operation
which _is_ correct and portable, even though it may not be fully
optimised on platforms where we don't have access to the appropriate
barrier operations.
Cheers,
Andrew Bromage