[Isolates] Merge 6500:6700 from bleeding_edge to isolates.

[Isolates] Merge 6500:6700 from bleeding_edge to isolates.

Committed: http://code.google.com/p/v8/source/detail?r=7031

[Dmitry Titov, 2011-03-02 18:14:59]

Except regular merge, I pulled LAllocator::Setup() into a new
Isolate::RunStaticInitializers(). The initializing of shared caches can be done
otherwise on multiple threads and it is not thread-safe, even though the end
result of initializing is the same. So it makes sense to only do it once.

One of isolate tests failed because of that.

[Vitaly Repeshko, 2011-03-02 18:27:27]

Rubberstamp LGTM. I only looked into isolate.{h,cc}.

http://codereview.chromium.org/6606006/diff/1/src/isolate.cc
File src/isolate.cc (right):

http://codereview.chromium.org/6606006/diff/1/src/isolate.cc#newcode188
src/isolate.cc:188: Isolate::RunStaticInitializers();
Consider doing this before the default isolate is pre-inited.

http://codereview.chromium.org/6606006/diff/1/src/isolate.h
File src/isolate.h (right):

http://codereview.chromium.org/6606006/diff/1/src/isolate.h#newcode942
src/isolate.h:942: // Used to init the structures that are "const after they are
niitialized
Typo "niitialized".

[Dmitry Titov, 2011-03-02 19:25:30]

Turns out there is static initialization order issue. The static arrays which
host operands can get initialized with default constructors after the static
initializer in Isolate puts all the cahced values in them.

Changed the logic a bit - the SetupCache is now called from the same file which
defines the static arrays so the initialization order is fixed and correct. See
modified DEFINE_OPERAND_CACHE macro.

[Dmitry Titov, 2011-03-02 19:49:13]

Incorporated Vitaly's feedback on Mad's patch
run tests on ia32, x64, arm in release and debug
run lint
landed