Re: [rvm-core] Status of the JMM implementation in Jikes RVM

[Eliot M. <mo...@cs...>]

On 2/24/2013 7:32 PM, Antony Hosking wrote:
>
> On Feb 25, 2013, at 10:23 AM, Eliot Moss <mo...@cs...> wrote:

> Can we not use CMPXCHG8B for double-word volatiles?
> That is available on x86-32.

Yes, but that is a lock-free kind of thing that will require a
loop.  However, a close reading of CMPXCHG8B indicates that there is
not really a need to read the locations first: the compare
might fail (probably will) the first time around the loop,
but the effect will be to load the memory value into the
registers used as the compare value.  Thus it is well suited
to loop with a two-instruction loop, and is compact enough
for an inline sequence.

Eliot