[heap] Introduce a new remembered set for typed pointers from old to new.

[heap] Introduce a new remembered set for typed pointers from old to new.

R=ulan@chromium.org, titzer@chromium.org

Committed: https://crrev.com/ff69ca7845a7dedfec3e9c2508c8051641314722
Cr-Commit-Position: refs/heads/master@{#36431}

[ahaas, 2016-05-20 11:45:00]

On 2016/05/20 at 11:25:16, ahaas wrote:
> 

I removed the mode from of the MemoryChunkIterator.

[titzer, 2016-05-20 11:47:37]

https://codereview.chromium.org/2003553002/diff/20001/src/heap/remembered-set.h
File src/heap/remembered-set.h (right):

https://codereview.chromium.org/2003553002/diff/20001/src/heap/remembered-set...
src/heap/remembered-set.h:18: template <PointerDirection direction>
(meta question)
Does this class really need to be templatized? IIUC, the only thing {direction}
is used for now is deciding which of the slots to retrieve for the private
methods below. Is that performance critical?

[ulan, 2016-05-20 12:48:13]

Looking good, two comments

https://codereview.chromium.org/2003553002/diff/20001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/2003553002/diff/20001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:3402: RememberedSet<OLD_TO_OLD>::RemoveRangeTyped(page,
start, end);
RememberedSet<OLD_TO_NEW>::RemoveRangeTyped(page, start, end);

https://codereview.chromium.org/2003553002/diff/20001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:3578: }
Handle OLD_TO_NEW typed slots here.

https://codereview.chromium.org/2003553002/diff/20001/src/heap/remembered-set.h
File src/heap/remembered-set.h (right):

https://codereview.chromium.org/2003553002/diff/20001/src/heap/remembered-set...
src/heap/remembered-set.h:18: template <PointerDirection direction>
On 2016/05/20 11:47:36, titzer wrote:
> (meta question)
> Does this class really need to be templatized? IIUC, the only thing
{direction}
> is used for now is deciding which of the slots to retrieve for the private
> methods below. Is that performance critical?

Insertion is performance critical. Run-time check would add one load dependency
before we can get to slot_set, but maybe that is not a bit deal.

Andreas, could you please add TODO(ulan): Investigate performance of
de-templatizing this class?

[ahaas, 2016-05-20 13:20:16]

https://codereview.chromium.org/2003553002/diff/20001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/2003553002/diff/20001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:3402: RememberedSet<OLD_TO_OLD>::RemoveRangeTyped(page,
start, end);
On 2016/05/20 at 12:48:12, ulan wrote:
> RememberedSet<OLD_TO_NEW>::RemoveRangeTyped(page, start, end);

Done.

https://codereview.chromium.org/2003553002/diff/20001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:3578: }
On 2016/05/20 at 12:48:12, ulan wrote:
> Handle OLD_TO_NEW typed slots here.

Done. PTAL

https://codereview.chromium.org/2003553002/diff/20001/src/heap/remembered-set.h
File src/heap/remembered-set.h (right):

https://codereview.chromium.org/2003553002/diff/20001/src/heap/remembered-set...
src/heap/remembered-set.h:18: template <PointerDirection direction>
On 2016/05/20 at 12:48:12, ulan wrote:
> On 2016/05/20 11:47:36, titzer wrote:
> > (meta question)
> > Does this class really need to be templatized? IIUC, the only thing
{direction}
> > is used for now is deciding which of the slots to retrieve for the private
> > methods below. Is that performance critical?
> 
> Insertion is performance critical. Run-time check would add one load
dependency before we can get to slot_set, but maybe that is not a bit deal.
> 
> Andreas, could you please add TODO(ulan): Investigate performance of
de-templatizing this class?

Done.

[ulan, 2016-05-20 13:24:25]

https://codereview.chromium.org/2003553002/diff/40001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/2003553002/diff/40001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:3576: UpdateSlot);
We need to call CheckAndUpdateOldToNewSlot instead of UpdateSlot for OLD_TO_NEW
slots.

[ahaas, 2016-05-23 08:19:33]

On 2016/05/20 at 13:24:25, ulan wrote:
> https://codereview.chromium.org/2003553002/diff/40001/src/heap/mark-compact.cc
> File src/heap/mark-compact.cc (right):
> 
>
https://codereview.chromium.org/2003553002/diff/40001/src/heap/mark-compact.c...
> src/heap/mark-compact.cc:3576: UpdateSlot);
> We need to call CheckAndUpdateOldToNewSlot instead of UpdateSlot for
OLD_TO_NEW slots.

Done.

[ulan, 2016-05-23 08:40:08]

LGTM

https://codereview.chromium.org/2003553002/diff/60001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/2003553002/diff/60001/src/heap/mark-compact.c...
src/heap/mark-compact.cc:3585: return CheckAndUpdateOldToNewSlot(
For separate CL: We would need to force promotion of all new space objects
referenced by code. Otherwise, we would have to clear old_to_new slots in dying
code objects, which would be expensive.
Let's discuss details offline.

[ahaas, 2016-05-23 08:59:10]

The CQ bit was checked by ahaas@chromium.org

[commit-bot: I haz the power, 2016-05-23 08:59:17]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/2003553002/60001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/2003553002/60001

[commit-bot: I haz the power, 2016-05-23 09:01:24]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2016-05-23 09:01:40]

Description was changed from

==========
[heap] Introduce a new remembered set for typed pointers from old to new.

R=ulan@chromium.org, titzer@chromium.org
==========

to

==========
[heap] Introduce a new remembered set for typed pointers from old to new.

R=ulan@chromium.org, titzer@chromium.org

Committed: https://crrev.com/ff69ca7845a7dedfec3e9c2508c8051641314722
Cr-Commit-Position: refs/heads/master@{#36431}
==========

[commit-bot: I haz the power, 2016-05-23 09:01:42]

Patchset 4 (id:??) landed as
https://crrev.com/ff69ca7845a7dedfec3e9c2508c8051641314722
Cr-Commit-Position: refs/heads/master@{#36431}