[aarch64-port-dev ] RFR: aarch64: minor improvements of atomic operations

[Andrew Haley]

On 11/12/19 12:02 PM, Yangfei (Felix) wrote:
> I also suggest we implement these functions with inline assembly here.  

Please let's not. Long term it would be nice to migrate all of HotSpot
from the current inline hackery to real C++ atomics. There has been a
considerable effort to make C++ and Java memory models compatible, and
we should utilize this.

> For Atomic::PlatformXchg, we may issue two consecutive full memory
> barriers with the current status.

OK, but is this actually important? What uses it?

-- 
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