I was just experimenting with core.atomic and noticed only a
marginal speed difference when accessing variables using
atomicLoad.
Is this because it isn't implemented or because it can be
reduced to a no-op on x86 hardware?

-- 
Marco

On 5 Jun 2013, at 19:19, Marco Leise wrote:
> I was just experimenting with core.atomic and noticed only a
> marginal speed difference when accessing variables using
> atomicLoad.
> Is this because it isn't implemented or because it can be
> reduced to a no-op on x86 hardware?

atomicLoad/atomicStore are implemented for LDC.

As for what causes the speed difference, the only sensible answer is to have a look at the produced assembly. atomicLoad can't be transformed into a normal load in the general case even on x86.

David

Is this because it isn't implemented or because it can be
reduced to a no-op on x86 hardware?


________________
dhoom

On Wednesday, 5 June 2013 at 17:25:32 UTC, David Nadlinger wrote:
> On 5 Jun 2013, at 19:19, Marco Leise wrote:
>> I was just experimenting with core.atomic and noticed only a
>> marginal speed difference when accessing variables using
>> atomicLoad.
>> Is this because it isn't implemented or because it can be
>> reduced to a no-op on x86 hardware?
>
> atomicLoad/atomicStore are implemented for LDC.
>
> As for what causes the speed difference, the only sensible answer is to have a look at the produced assembly. atomicLoad can't be transformed into a normal load in the general case even on x86.
>
> David

Those who are interested in this topic should watch Herb Sutter's 2012 talk on std::atomic
http://channel9.msdn.com/Shows/Going+Deep/Cpp-and-Beyond-2012-Herb-Sutter-atomic-Weapons-1-of-2