[concurrency-interest] Java Memory Model and ParallelStream

Andrew Haley aph at redhat.com
Fri Mar 6 09:44:52 EST 2020


On 3/6/20 2:35 PM, Luke Hutchison wrote:
> On Fri, Mar 6, 2020, 7:15 AM Andrew Haley <aph at redhat.com> wrote:
> 
>> The CPU *hardware* does this all the time. Few processors have a
>> total store order: x86, which does, is the exception here.
> 
> Well that gets at my core question: whether a computational barrier always
> strictly enforces memory happens-before ordering across the barrier, i.e.
> whether a computational barrier

To answer that I'd need to be told what a "computational barrier" is. I
have some guesses, but I think you should spell it out.

> is always also a memory ordering barrier.
> 
> If the CPU does not have a total store order, I could imagine cases where a
> computational barrier does not enforce memory ordering. What am I missing?
-- 
Andrew Haley  (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671



More information about the Concurrency-interest mailing list