Use UniqueSet<T> and Unique<T> in HCheckMaps and HCheckValue.

Use UniqueSet<T> and Unique<T> in HCheckMaps and HCheckValue.

BUG=
R=verwaest@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=16814

[Toon Verwaest, 2013-09-17 10:37:24]

https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (left):

https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h#o...
src/hydrogen-instructions.h:2674: object_unique_id_ = UniqueValueId(object_);
So I guess you'll have to do something like this as well...

https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h#n...
src/hydrogen-instructions.h:2628: Unique<JSFunction> target(func);
Allocating this in the middle of graph-building isn't safe. We allocate
heap-objects during graph creation, potentially causing GC. We should only be
able to construct Unique objects in a DisallowHeapAllocation scope; otherwise
the resulting objects can't guarantee anything. (Granted, it does work for
immovable objects, but we should clearly indicate that if we rely on such
information anywhere).

[titzer, 2013-09-18 09:43:13]

On 2013/09/17 10:37:24, Toon Verwaest wrote:
> https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h
> File src/hydrogen-instructions.h (left):
> 
>
https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h#o...
> src/hydrogen-instructions.h:2674: object_unique_id_ = UniqueValueId(object_);
> So I guess you'll have to do something like this as well...
> 
> https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h
> File src/hydrogen-instructions.h (right):
> 
>
https://codereview.chromium.org/23604062/diff/1/src/hydrogen-instructions.h#n...
> src/hydrogen-instructions.h:2628: Unique<JSFunction> target(func);
> Allocating this in the middle of graph-building isn't safe. We allocate
> heap-objects during graph creation, potentially causing GC. We should only be
> able to construct Unique objects in a DisallowHeapAllocation scope; otherwise
> the resulting objects can't guarantee anything. (Granted, it does work for
> immovable objects, but we should clearly indicate that if we rely on such
> information anywhere).

I've added the notion of an "uninitialized" Unique<T>. Now we first initialize
an HCheckValue to be unitialized and then finalize it like as before. This
reestablishes the "all uniqueness properties are computed at the same time" kind
of invariant. Note we don't do that for HCheckMaps, since we assume that maps
don't move.

[Toon Verwaest, 2013-09-18 15:00:15]

lgtm (Although I'd still like to have proper asserts when creating / filling in
the unique values. This is scary!)

[titzer, 2013-09-19 09:07:37]

Committed patchset #2 manually as r16814 (presubmit successful).