Implemented lazy sweeping of new space.

Implemented lazy sweeping of new space.

BUG=
R=titzer@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=15616

[titzer, 2013-07-10 17:06:35]

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.cc
File src/store-buffer.cc (right):

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.cc#newcod...
src/store-buffer.cc:426: Address& map_field =
Memory::Address_at(object->address());
I'm a little dedup, short and stout. Rip me over and factor me out!

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.cc#newcod...
src/store-buffer.cc:596: }
BTW, do you still want to call the slot callback if the object is dead?

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.h
File src/store-buffer.h (right):

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.h#newcode86
src/store-buffer.h:86: void IteratePointersToNewSpace(ObjectSlotCallback
callback, bool clear_maps);
Boolean flags of doom be gone! Recommend to have two separate methods, one which
clears and one which does not clear. The private implementation can use internal
booleans to avoid duplicating the guts, but the public interface should be more
well-defined.

Either that, or you could have the callback itself doing the clearing...

[Hannes Payer (out of office), 2013-07-11 07:30:40]

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.cc
File src/store-buffer.cc (right):

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.cc#newcod...
src/store-buffer.cc:426: Address& map_field =
Memory::Address_at(object->address());
On 2013/07/10 17:06:35, titzer wrote:
> I'm a little dedup, short and stout. Rip me over and factor me out!

Done.

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.cc#newcod...
src/store-buffer.cc:596: }
On 2013/07/10 17:06:35, titzer wrote:
> BTW, do you still want to call the slot callback if the object is dead?

As soon as we clear the todo in the callback, we can make that call conditional.

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.h
File src/store-buffer.h (right):

https://codereview.chromium.org/18998004/diff/2001/src/store-buffer.h#newcode86
src/store-buffer.h:86: void IteratePointersToNewSpace(ObjectSlotCallback
callback, bool clear_maps);
On 2013/07/10 17:06:35, titzer wrote:
> Boolean flags of doom be gone! Recommend to have two separate methods, one
which
> clears and one which does not clear. The private implementation can use
internal
> booleans to avoid duplicating the guts, but the public interface should be
more
> well-defined.
> 
> Either that, or you could have the callback itself doing the clearing...

Done.

[titzer, 2013-07-11 08:22:22]

LGTM other than moving a method to the StoreBuffer class.

https://codereview.chromium.org/18998004/diff/9001/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/18998004/diff/9001/src/objects.h#newcode1459
src/objects.h:1459: inline void ClearDeadObject();
I don't think this method should live on the HeapObject itself. Can you make it
private to the class where it is used?

[Hannes Payer (out of office), 2013-07-11 08:43:09]

https://codereview.chromium.org/18998004/diff/9001/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/18998004/diff/9001/src/objects.h#newcode1459
src/objects.h:1459: inline void ClearDeadObject();
On 2013/07/11 08:22:23, titzer wrote:
> I don't think this method should live on the HeapObject itself. Can you make
it
> private to the class where it is used?

Done.

[Hannes Payer (out of office), 2013-07-11 09:17:09]

Committed patchset #5 manually as r15616 (presubmit successful).