Fix incremental marking speed accelleration and start incremental marking when old space is almost …

Fix incremental marking speed accelleration and start incremental marking when old space is almost full when pretenuring objects.


BUG=

[Michael Starzinger, 2013-05-06 12:35:25]

https://codereview.chromium.org/14649014/diff/2001/src/runtime.cc
File src/runtime.cc (right):

https://codereview.chromium.org/14649014/diff/2001/src/runtime.cc#newcode9162
src/runtime.cc:9162: heap->incremental_marking()->Start();
Starting the incremental marker here scares me _a lot_. The incremental marker
should be started due to heuristics inside the GC and not by random hooks
scattered throughout the code-base.

[Michael Starzinger, 2013-05-06 12:49:56]

https://codereview.chromium.org/14649014/diff/2001/src/runtime.cc
File src/runtime.cc (right):

https://codereview.chromium.org/14649014/diff/2001/src/runtime.cc#newcode9162
src/runtime.cc:9162: heap->incremental_marking()->Start();
On 2013/05/06 12:35:25, Michael Starzinger wrote:
> Starting the incremental marker here scares me _a lot_. The incremental marker
> should be started due to heuristics inside the GC and not by random hooks
> scattered throughout the code-base.

I could be convinced to have this hack in IncrementalMarking::OldSpaceStep so
that old-space steps are capable of starting the incremental marker. But for now
I would limit this to when we globally pretenure, because that is what's causing
the starvation of new-space steps.

Also as discussed offline, the MarkCompactCollector::abort_incremental_marking()
makes no sense here, as it will always be false.

[Hannes Payer (out of office), 2013-05-06 13:01:48]

Thanks for the suggestions, true we must not add GC code outside the GC or the
memory allocator. The decision to start incremental marking fits into
IncrementalMarking::OldSpaceStep. 

However, I do not check there for high promotion mode since old space allocation
should in principle always be able to start incremental marking if needed. This
should be independent of pretenuring.

[Michael Starzinger, 2013-05-06 13:05:12]

LGTM.

https://codereview.chromium.org/14649014/diff/6001/src/incremental-marking.cc
File src/incremental-marking.cc (right):

https://codereview.chromium.org/14649014/diff/6001/src/incremental-marking.cc...
src/incremental-marking.cc:864: if (IsStopped() &&
nit: Condition should fit into one line now.

https://codereview.chromium.org/14649014/diff/6001/src/incremental-marking.cc...
src/incremental-marking.cc:869: Step(allocated * kFastMarking /
kInitialMarkingSpeed,
nit: Call to step should fit into one line now.

[Hannes Payer (out of office), 2013-05-06 13:06:46]

https://codereview.chromium.org/14649014/diff/6001/src/incremental-marking.cc
File src/incremental-marking.cc (right):

https://codereview.chromium.org/14649014/diff/6001/src/incremental-marking.cc...
src/incremental-marking.cc:864: if (IsStopped() &&
On 2013/05/06 13:05:12, Michael Starzinger wrote:
> nit: Condition should fit into one line now.

Done.

https://codereview.chromium.org/14649014/diff/6001/src/incremental-marking.cc...
src/incremental-marking.cc:869: Step(allocated * kFastMarking /
kInitialMarkingSpeed,
On 2013/05/06 13:05:12, Michael Starzinger wrote:
> nit: Call to step should fit into one line now.

Done.