Use compiler barrier instead of memory barrier for release/acquire atomic operations on mac.

Use compiler barrier instead of memory barrier for release/acquire atomic operations on mac.

BUG=
R=jarin@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=25147

[Hannes Payer (out of office), 2014-11-05 10:50:37]

hpayer@chromium.org changed reviewers:
+ jarin@chromium.org

[Jarin, 2014-11-05 11:54:53]

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
File src/base/atomicops_internals_mac.h (right):

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
src/base/atomicops_internals_mac.h:83: ATOMICOPS_COMPILER_BARRIER();
I think it is better not to touch anything else than AcquireLoad because the
semantics of that is not completely clear.

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
src/base/atomicops_internals_mac.h:101: ATOMICOPS_COMPILER_BARRIER();
How about just saying here

// On x86 processors, loads already have acquire semantics, so
// there is no need to put a full barrier here.
#if V8_HOST_ARCH_IA32
  ATOMICOPS_COMPILER_BARRIER();
#else  
  MemoryBarrier();
#endif

Or we can introduce an AcquireMemoryBarrier function that would have the #if
ARCH.

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
src/base/atomicops_internals_mac.h:191: ATOMICOPS_COMPILER_BARRIER();
As above, maybe we want

// On x86 processors, loads already have acquire semantics, so
// there is no need to put a full barrier here.
#if V8_HOST_ARCH_X64
  ATOMICOPS_COMPILER_BARRIER();
#else  
  MemoryBarrier();
#endif

[Hannes Payer (out of office), 2014-11-05 12:16:23]

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
File src/base/atomicops_internals_mac.h (right):

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
src/base/atomicops_internals_mac.h:83: ATOMICOPS_COMPILER_BARRIER();
On 2014/11/05 11:54:53, jarin wrote:
> I think it is better not to touch anything else than AcquireLoad because the
> semantics of that is not completely clear.

Done. However, I would also change the release stores. WDYT?

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
src/base/atomicops_internals_mac.h:101: ATOMICOPS_COMPILER_BARRIER();
On 2014/11/05 11:54:53, jarin wrote:
> How about just saying here
> 
> // On x86 processors, loads already have acquire semantics, so
> // there is no need to put a full barrier here.
> #if V8_HOST_ARCH_IA32
>   ATOMICOPS_COMPILER_BARRIER();
> #else  
>   MemoryBarrier();
> #endif
> 
> Or we can introduce an AcquireMemoryBarrier function that would have the #if
> ARCH.

Done.

https://codereview.chromium.org/694703003/diff/20001/src/base/atomicops_inter...
src/base/atomicops_internals_mac.h:191: ATOMICOPS_COMPILER_BARRIER();
On 2014/11/05 11:54:53, jarin wrote:
> As above, maybe we want
> 
> // On x86 processors, loads already have acquire semantics, so
> // there is no need to put a full barrier here.
> #if V8_HOST_ARCH_X64
>   ATOMICOPS_COMPILER_BARRIER();
> #else  
>   MemoryBarrier();
> #endif

Done.

[Jarin, 2014-11-05 12:31:16]

lgtm. Thanks!

[Hannes Payer (out of office), 2014-11-05 12:35:30]

Committed patchset #3 (id:40001) manually as 25147 (presubmit successful).