[concurrency-interest] when is safe publication safe?

Doug Lea dl at cs.oswego.edu
Tue Apr 27 19:26:38 EDT 2010


A little follow-up to tie some of these discussions together a bit:

On 04/27/10 18:46, Doug Lea wrote:
> For volatile writes:
> * All of the above need at least a trailing store-load or
> use of atomics that normally costs in the dozens of cycles, but
> less so on some recent x86 and Azul.

Note: This is the fence you don't get with lazySet or
the Fences draft on release-fences. It is the most expensive
of all of fences on all platforms, so is the one that some people
would like to avoid when they are sure they don't need it.
But as always, offering this in a cheap and solidily spec'ed
form that is attractive to use only when it is correct remains
an unsolved problem.

-Doug




More information about the Concurrency-interest mailing list