Fix and improve arm64 atomic operations in Base.

Fix and improve arm64 atomic operations in Base.

 - The 'compare and swap' operations should enforce memory ordering even when
   the exchange does not occur.
 - The exclusive monitor does not need to be cleared by CLREX if a LDRX was
   not followed by a matching STREX.
 - Use LDAR and STLR where possible.
 - Use the 'I' and 'J' constraints to hint for constants valid for immediate
   values

Based on V8 changes at https://codereview.chromium.org/220793002

BUG=354405
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=262986

[rmcilroy, 2014-04-09 14:29:18]

PTAL.

[Mark Mentovai, 2014-04-09 16:50:41]

LGTM

https://codereview.chromium.org/230963002/diff/1/base/atomicops_internals_arm...
File base/atomicops_internals_arm64_gcc.h (right):

https://codereview.chromium.org/230963002/diff/1/base/atomicops_internals_arm...
base/atomicops_internals_arm64_gcc.h:103: Atomic32 result;
This change doesn’t seem to help anything. We usually just declare things at the
point of first use (line 106). Atomic32 is just a standard integral type, so
it’s not like putting this here before the barrier runs any constructors.

https://codereview.chromium.org/230963002/diff/1/base/atomicops_internals_arm...
base/atomicops_internals_arm64_gcc.h:117: prev = NoBarrier_CompareAndSwap(ptr,
old_value, new_value);
Likewise, why not declare it on this line?

Also in Barrier_AtomicIncrement, Acquire_CompareAndSwap, and
Release_CompareAndSwap.

https://codereview.chromium.org/230963002/diff/1/base/atomicops_internals_arm...
base/atomicops_internals_arm64_gcc.h:129: prev = NoBarrier_CompareAndSwap(ptr,
old_value, new_value);
And again here. In fact, can’t this function just be MemoryBarrier(); return
NoBarrier_CompareAndSwap(…);?

[rmcilroy, 2014-04-10 12:30:46]

https://codereview.chromium.org/230963002/diff/1/base/atomicops_internals_arm...
File base/atomicops_internals_arm64_gcc.h (right):

https://codereview.chromium.org/230963002/diff/1/base/atomicops_internals_arm...
base/atomicops_internals_arm64_gcc.h:103: Atomic32 result;
On 2014/04/09 16:50:41, Mark Mentovai wrote:
> This change doesn’t seem to help anything. We usually just declare things at
the
> point of first use (line 106). Atomic32 is just a standard integral type, so
> it’s not like putting this here before the barrier runs any constructors.

I think this change was to be consistent with the following functions, but I
agree - it's more usual to declare on first use.  Done.

[rmcilroy, 2014-04-10 12:30:49]

The CQ bit was checked by rmcilroy@chromium.org

[commit-bot: I haz the power, 2014-04-10 12:31:00]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/rmcilroy@chromium.org/230963002/20001

[commit-bot: I haz the power, 2014-04-10 14:38:23]

Change committed as 262986