Create an abstraction (MarkBit) object to hold the location of the...

Create an abstraction (MarkBit) object to hold the location of the
mark bits for an object.  This allows the compiler to avoid
recalculating the location when accessing two bits for incremental
marking or when testing and then setting mark bits.

Committed: http://code.google.com/p/v8/source/detail?r=7295

[Vyacheslav Egorov (Chromium), 2011-03-21 16:54:23]

LGTM

http://codereview.chromium.org/6713075/diff/3002/src/heap.h
File src/heap.h (right):

http://codereview.chromium.org/6713075/diff/3002/src/heap.h#newcode1788
src/heap.h:1788: ASSERT(object->IsHeapObject());
Do we really want this?

Sometimes it's better to have an early detection of mark stack corruption.

http://codereview.chromium.org/6713075/diff/3002/src/spaces.h
File src/spaces.h (right):

http://codereview.chromium.org/6713075/diff/3002/src/spaces.h#newcode121
src/spaces.h:121: class MarkBit {
// TODO(gc) ensure all operations inlined and object is exploded by compiler on
all platforms.

[Vyacheslav Egorov (Chromium), 2011-03-21 17:16:30]

additional comments.

http://codereview.chromium.org/6713075/diff/3002/src/mark-compact.cc
File src/mark-compact.cc (right):

http://codereview.chromium.org/6713075/diff/3002/src/mark-compact.cc#newcode246
src/mark-compact.cc:246: IncrementalMarking::MarkBlack(old_mark_bit);
new_mark_bit

http://codereview.chromium.org/6713075/diff/3002/src/mark-compact.cc#newcode958
src/mark-compact.cc:958: Code* inlined_code =
inlined->unchecked_shared()->unchecked_code();
I think we will need to kill all unchecked_* getters.

add TODO(gc) for that.

[Erik Corry, 2011-03-21 20:48:53]

http://codereview.chromium.org/6713075/diff/3002/src/heap.h
File src/heap.h (right):

http://codereview.chromium.org/6713075/diff/3002/src/heap.h#newcode1788
src/heap.h:1788: ASSERT(object->IsHeapObject());
On 2011/03/21 16:54:23, Vyacheslav Egorov wrote:
> Do we really want this?
> 
> Sometimes it's better to have an early detection of mark stack corruption.

This is inlined in some of our hottest GC functions.  I'm going to follow the
pattern in the entire rest of the VM and leave sanity tests to the debug
version.

http://codereview.chromium.org/6713075/diff/3002/src/mark-compact.cc
File src/mark-compact.cc (right):

http://codereview.chromium.org/6713075/diff/3002/src/mark-compact.cc#newcode246
src/mark-compact.cc:246: IncrementalMarking::MarkBlack(old_mark_bit);
On 2011/03/21 17:16:30, Vyacheslav Egorov wrote:
> new_mark_bit

Good catch.

http://codereview.chromium.org/6713075/diff/3002/src/mark-compact.cc#newcode958
src/mark-compact.cc:958: Code* inlined_code =
inlined->unchecked_shared()->unchecked_code();
On 2011/03/21 17:16:30, Vyacheslav Egorov wrote:
> I think we will need to kill all unchecked_* getters.
> 
> add TODO(gc) for that.

Done.

http://codereview.chromium.org/6713075/diff/3002/src/spaces.h
File src/spaces.h (right):

http://codereview.chromium.org/6713075/diff/3002/src/spaces.h#newcode121
src/spaces.h:121: class MarkBit {
On 2011/03/21 16:54:23, Vyacheslav Egorov wrote:
> // TODO(gc) ensure all operations inlined and object is exploded by compiler
on
> all platforms.

Checked on gcc/x86.

TODO added.

>