Atomic ops: sync with Chromium.

Atomic ops: Sync with Chromium and add unit test.

R=jarin@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=19738

[c.truta, 2014-01-24 13:26:30]

Hi,

I cherry-picked the latest development on atomic ops from Chromium: arm_gcc and
x86_msvc.

In the process, I also noticed something that appeared like a copy-paste error
in atomicops_tsan. I fixed that also.

I run tests with gcc on Linux/ARM. With MSVC and TSan, it was only inspection.

[Benedikt Meurer, 2014-01-27 06:34:29]

I'd prefer to cherry-pick the current Chromium state for all platforms. That way
we guarantee that the platforms are in sync, and it'll make it easier to pin
issues to a specific Chromium revision.

[c.truta, 2014-02-03 05:06:21]

On 2014/01/27 06:34:29, Benedikt Meurer wrote:
> I'd prefer to cherry-pick the current Chromium state for all platforms.

Ok, I integrated all the recent development on atomic ops from Chromium. Here is
a summary of differences between what's in Chromium vs. what would be in V8
after applying my patch.

* Copyright headers, some horizontal and vertical whitespace corrections, and
other similarly-trivial changes.

* Namespace: base::subtle vs. v8::internal.

* Chromium-specific macros (e.g. ARCH_CPU_NNNN) vs. V8-specific macros (e.g.
V8_HOST_ARCH_NNNN).

* The NaCl changes are missing. I wasn't able to test NaCl, and I wasn't sure
those changes are needed.

* There were some other existing (minor, non-semantic) changes between V8 and
Chromium that seemed deliberate in V8, and I haven't touched those, in
atomicops_internals_x86_gcc.{h,cc} and atomicops_internals_tsan.h.

* I integrated the Chromium unit test (atomicops_unittest.cc) into cctest,
following the cctest conventions and restrictions, as well as working around the
more limited overload-resolution choices given by the CHECK_EQ macro.

Tested on Linux (ARM, x64, ia32), Mac (x64, ia32), Windows (ia32).

[Alexander Potapenko, 2014-02-03 08:02:28]

On 2014/02/03 05:06:21, cosmin.truta wrote:
> On 2014/01/27 06:34:29, Benedikt Meurer wrote:
> > I'd prefer to cherry-pick the current Chromium state for all platforms.
> 
Looks like this typo is missing in base/atomicops_internals_tsan.h in Chromium,
this is a V8-only error.

Can you just copy base/atomicops_internals_tsan.h to v8 verbatim (if it differs
from your change)?

[Alexander Potapenko, 2014-02-03 08:05:29]

> Can you just copy base/atomicops_internals_tsan.h to v8 verbatim (if it
differs
> from your change)?

It actually does not (modulo some v8-specific changes), please disregard.

[c.truta, 2014-02-10 05:30:17]

Hi, this is a new update.

Since Patch Set 2, I landed the QNX port changes in Chromium, in r249390;
then I took some of the atomicops changes previously done in V8 (plus some
whitespace formatting fixes of my own) and I landed them back in Chromium,
in r249938.

Now I would like to bring all of them back to V8, wholesale. After this,
the differences between Chromium's and V8's atomic ops shall be minimal.

[c.truta, 2014-02-14 07:02:39]

In Patch Set 4, I resolved a small conflict with the new ARM64 code.

I removed some of the reviewers from the list, as I've noticed that it's
customary not to have too many. (Some of them have already seen the
changes in the areas of their interest, as I landed them in Chromium.)

[Benedikt Meurer, 2014-02-14 07:05:20]

Looks good to me, I think we can land this next week after the Chromium branch.

Adding jarin@ for review.

[Jarin, 2014-02-14 08:46:03]

On 2014/02/14 07:05:20, Benedikt Meurer wrote:
> Looks good to me, I think we can land this next week after the Chromium
branch.
> 
> Adding jarin@ for review.

Looks great to me

[Jarin, 2014-03-10 06:31:49]

lgtm

[Jarin, 2014-03-10 06:43:32]

Committed patchset #4 manually as r19738 (presubmit successful).