Optimize clearing of map transitions.

Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

Simple transitions are in the weak cell list.

Full transitions are in the transitions array list.

BUG=chromium:554488
LOG=NO

Committed: https://crrev.com/8c376b4635831ff28bb4a04f3859ad7e51c17e8b
Cr-Commit-Position: refs/heads/master@{#32684}

[ulan, 2015-12-01 10:12:00]

Description was changed from

==========
Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

BUG=chromium:554488
LOG=NO
==========

to

==========
Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

Simple transitions are in the weak cell list.

Full transitions are in the transitions array list.

BUG=chromium:554488
LOG=NO
==========

[ulan, 2015-12-01 10:13:15]

ulan@chromium.org changed reviewers:
+ hpayer@chromium.org

[ulan, 2015-12-01 10:13:15]

PTAL

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (left):

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:1997: if (Marking::IsWhite(map_mark)) {
This code moved to ClearWeakCells

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:2321: DependentCode* dependent_code_list =
DependentCodeListFromNonLiveMaps();
ProcessAndClearWeakCells() and 
MarkDependentCodeListForDeoptimization(dependent_code_list) are in
ClearNonLiveReferences now

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:2346: for (HeapObject* obj = map_iterator.Next(); obj
!= NULL;
This is the main change: this loop is replaced by ClearSimpleMapTransitions and
ClearFullMapTransitions.

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:2316: DependentCode* current = list_head;
This code is from MarkDependentCodeListForDeoptimization below

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:2369: // A previously existing simple transition
(stored in a WeakCell) is going
This code is from lines 2426 - 2436

https://codereview.chromium.org/1488593003/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:2392: Map* map = array->GetTarget(0);
This code does the same as lines 2440 - 2493.

[Yang, 2015-12-01 11:21:40]

yangguo@chromium.org changed reviewers:
+ yangguo@chromium.org

[Yang, 2015-12-01 11:21:40]

https://codereview.chromium.org/1488593003/diff/40001/src/snapshot/serialize.cc
File src/snapshot/serialize.cc (right):

https://codereview.chromium.org/1488593003/diff/40001/src/snapshot/serialize....
src/snapshot/serialize.cc:1977:
weak_cell_->clear_next(object->GetHeap()->the_hole_value());
Why does the caller need to know with what sentinel value we clear the weak
cell?

[ulan, 2015-12-01 12:49:52]

https://codereview.chromium.org/1488593003/diff/40001/src/snapshot/serialize.cc
File src/snapshot/serialize.cc (right):

https://codereview.chromium.org/1488593003/diff/40001/src/snapshot/serialize....
src/snapshot/serialize.cc:1977:
weak_cell_->clear_next(object->GetHeap()->the_hole_value());
On 2015/12/01 11:21:40, Yang wrote:
> Why does the caller need to know with what sentinel value we clear the weak
> cell?

For efficiency, this function is used in a hot loop in mark-compact where we can
cache the value.

I think it is OK, since clear_next shouldn't be used by anyone except GC (and
serializer).

[Yang, 2015-12-01 13:13:41]

On 2015/12/01 12:49:52, ulan wrote:
>
https://codereview.chromium.org/1488593003/diff/40001/src/snapshot/serialize.cc
> File src/snapshot/serialize.cc (right):
> 
>
https://codereview.chromium.org/1488593003/diff/40001/src/snapshot/serialize....
> src/snapshot/serialize.cc:1977:
> weak_cell_->clear_next(object->GetHeap()->the_hole_value());
> On 2015/12/01 11:21:40, Yang wrote:
> > Why does the caller need to know with what sentinel value we clear the weak
> > cell?
> 
> For efficiency, this function is used in a hot loop in mark-compact where we
can
> cache the value.
> 
> I think it is OK, since clear_next shouldn't be used by anyone except GC (and
> serializer).

src/serializer lgtm.

[Hannes Payer (out of office), 2015-12-08 14:19:41]

lgtm

https://codereview.chromium.org/1488593003/diff/40001/src/heap/gc-tracer.h
File src/heap/gc-tracer.h (right):

https://codereview.chromium.org/1488593003/diff/40001/src/heap/gc-tracer.h#ne...
src/heap/gc-tracer.h:136: MC_CLEAR_DEPENDENT_CODE,
at one point we should create an alphabetic order there.

https://codereview.chromium.org/1488593003/diff/40001/src/objects-inl.h
File src/objects-inl.h (right):

https://codereview.chromium.org/1488593003/diff/40001/src/objects-inl.h#newco...
src/objects-inl.h:2084: set_next(the_hole_value, SKIP_WRITE_BARRIER);
Let's DCHECK that the_hole_value is the hole value.

https://codereview.chromium.org/1488593003/diff/40001/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/1488593003/diff/40001/src/objects.h#newcode9469
src/objects.h:9469: inline void clear_next(Object* the_hole_value);
+1

[ulan, 2015-12-08 16:25:04]

Thanks for review!

https://codereview.chromium.org/1488593003/diff/40001/src/heap/gc-tracer.h
File src/heap/gc-tracer.h (right):

https://codereview.chromium.org/1488593003/diff/40001/src/heap/gc-tracer.h#ne...
src/heap/gc-tracer.h:136: MC_CLEAR_DEPENDENT_CODE,
On 2015/12/08 14:19:41, Hannes Payer wrote:
> at one point we should create an alphabetic order there.

Acknowledged.

https://codereview.chromium.org/1488593003/diff/40001/src/objects-inl.h
File src/objects-inl.h (right):

https://codereview.chromium.org/1488593003/diff/40001/src/objects-inl.h#newco...
src/objects-inl.h:2084: set_next(the_hole_value, SKIP_WRITE_BARRIER);
On 2015/12/08 14:19:41, Hannes Payer wrote:
> Let's DCHECK that the_hole_value is the hole value.

Done.

[ulan, 2015-12-08 17:06:40]

The CQ bit was checked by ulan@chromium.org

[ulan, 2015-12-08 17:06:42]

The patchset sent to the CQ was uploaded after l-g-t-m from hpayer@chromium.org,
yangguo@chromium.org
Link to the patchset: https://codereview.chromium.org/1488593003/#ps60001
(title: "Address comment")

[commit-bot: I haz the power, 2015-12-08 17:06:49]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1488593003/60001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1488593003/60001

[commit-bot: I haz the power, 2015-12-08 20:51:01]

Description was changed from

==========
Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

Simple transitions are in the weak cell list.

Full transitions are in the transitions array list.

BUG=chromium:554488
LOG=NO
==========

to

==========
Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

Simple transitions are in the weak cell list.

Full transitions are in the transitions array list.

BUG=chromium:554488
LOG=NO
==========

[commit-bot: I haz the power, 2015-12-08 20:51:04]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2015-12-08 20:52:06]

Description was changed from

==========
Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

Simple transitions are in the weak cell list.

Full transitions are in the transitions array list.

BUG=chromium:554488
LOG=NO
==========

to

==========
Optimize clearing of map transitions.

Instead of iterating the whole map space to find dead transitions,
look in weak cell list and transition array list.

Simple transitions are in the weak cell list.

Full transitions are in the transitions array list.

BUG=chromium:554488
LOG=NO

Committed: https://crrev.com/8c376b4635831ff28bb4a04f3859ad7e51c17e8b
Cr-Commit-Position: refs/heads/master@{#32684}
==========

[commit-bot: I haz the power, 2015-12-08 20:52:07]

Patchset 4 (id:??) landed as
https://crrev.com/8c376b4635831ff28bb4a04f3859ad7e51c17e8b
Cr-Commit-Position: refs/heads/master@{#32684}