[heap] Move to LAB-based allocation for newspace evacuation.

[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Semantical change: Previously we did fall back to regular new space promotion if
we are OOM in old space. With this CL we fall back to new space promotion, which
could fail because of fragmentation, again leading to an old space allocation
that finally bails into OOM.

Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.

BUG=chromium:524425
LOG=N

Committed: https://crrev.com/a4e3a3b6a81ef39835c757145ff55299d713bcba
Cr-Commit-Position: refs/heads/master@{#32970}

[Michael Lippautz, 2015-11-30 21:09:23]

Description was changed from

==========
[heap] Move to LAB-based allocation for newspace evacuation.

BUG=
==========

to

==========
[heap] Move to LAB-based allocation for newspace evacuation.

BUG=chromium:524425
LOG=N
==========

[Michael Lippautz, 2015-12-01 10:18:21]

Description was changed from

==========
[heap] Move to LAB-based allocation for newspace evacuation.

BUG=chromium:524425
LOG=N
==========

to

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.

BUG=chromium:524425
LOG=N
==========

[Michael Lippautz, 2015-12-01 10:39:13]

Patchset #2 (id:20001) has been deleted

[Michael Lippautz, 2015-12-01 12:42:56]

Patchset #4 (id:80001) has been deleted

[Michael Lippautz, 2015-12-01 12:43:00]

Patchset #3 (id:60001) has been deleted

[Michael Lippautz, 2015-12-01 14:46:51]

Patchset #3 (id:100001) has been deleted

[Michael Lippautz, 2015-12-01 14:46:57]

Patchset #2 (id:40001) has been deleted

[Michael Lippautz, 2015-12-01 14:51:37]

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

[Michael Lippautz, 2015-12-01 14:51:37]

PTAL, comments welcomed :)

https://codereview.chromium.org/1487853002/diff/120001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1487853002/diff/120001/src/heap/mark-compact....
src/heap/mark-compact.cc:1588: static const intptr_t kLabSize = 2 * KB;
kLabSize influences NewSpace::Size() as we increment limit_ without potentially
using the memory.

[Michael Lippautz, 2015-12-16 12:44:37]

Patchset #3 (id:140001) has been deleted

[Michael Lippautz, 2015-12-16 12:58:35]

Description was changed from

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.

BUG=chromium:524425
LOG=N
==========

to

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Semantical change: Previously we did fall back to regular new space promotion if
we are OOM in old space. With this CL we fall back to new space promotion, which
could fail because of fragmentation, again leading to an old space allocation
that finally bails into OOM.


Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.



BUG=chromium:524425
LOG=N
==========

[Michael Lippautz, 2015-12-16 12:59:01]

Description was changed from

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Semantical change: Previously we did fall back to regular new space promotion if
we are OOM in old space. With this CL we fall back to new space promotion, which
could fail because of fragmentation, again leading to an old space allocation
that finally bails into OOM.


Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.



BUG=chromium:524425
LOG=N
==========

to

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Semantical change: Previously we did fall back to regular new space promotion if
we are OOM in old space. With this CL we fall back to new space promotion, which
could fail because of fragmentation, again leading to an old space allocation
that finally bails into OOM.

Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.

BUG=chromium:524425
LOG=N
==========

[Hannes Payer (out of office), 2015-12-16 14:51:06]

Overall looking good already, a few comments.

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact....
src/heap/mark-compact.cc:1577: static const intptr_t kMaxLabObjectSize = 256;
How did you identify these parameters? Do you think 256 is small enough to be
the common case?

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact....
src/heap/mark-compact.cc:1657: inline AllocationResult AllocateInOldSpace(int
size_in_bytes,
Are you going to provide LABs in old space in a separate CL?

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact....
src/heap/mark-compact.cc:1679: if (!NewLocalAllocationBuffer()) {
It would be nice to preserve the same behavior we have right now in the single
threaded case, i.e. no fragmentation. When allocating a new LAB, you could merge
the leftover from the old LAB with the new LAB if the are adjacent.

[Michael Lippautz, 2015-12-16 20:05:28]

Patchset #4 (id:180001) has been deleted

[Michael Lippautz, 2015-12-16 20:06:05]

Merging adjacent LABs was easier than expected.

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact....
src/heap/mark-compact.cc:1577: static const intptr_t kMaxLabObjectSize = 256;
On 2015/12/16 14:51:06, Hannes Payer wrote:
> How did you identify these parameters? Do you think 256 is small enough to be
> the common case?

These constant are not backed by real-world experiments, yet. Runs on our perf
bots show no regression using these values.

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact....
src/heap/mark-compact.cc:1657: inline AllocationResult AllocateInOldSpace(int
size_in_bytes,
On 2015/12/16 14:51:06, Hannes Payer wrote:
> Are you going to provide LABs in old space in a separate CL?

LABs in old space are not needed as we have compaction spaces. 

A future CL will pass them as parameters to the new space visitor, which will
use them for allocating in old space. This way, the same memory is used for
evacuating new space pages and old space pages.

https://codereview.chromium.org/1487853002/diff/160001/src/heap/mark-compact....
src/heap/mark-compact.cc:1679: if (!NewLocalAllocationBuffer()) {
On 2015/12/16 14:51:06, Hannes Payer wrote:
> It would be nice to preserve the same behavior we have right now in the single
> threaded case, i.e. no fragmentation. When allocating a new LAB, you could
merge
> the leftover from the old LAB with the new LAB if the are adjacent.

Done, however there's still a slight mismatch in behavior which we have to live
with I think. Objects larger than kMaxLabObjectSize will get allocated between
LABs, even in single-threaded cases, making the LABs impossible to merge.

[Michael Lippautz, 2015-12-16 20:38:45]

The CQ bit was checked by mlippautz@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-16 20:38:53]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1487853002/220001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1487853002/220001

[commit-bot: I haz the power, 2015-12-16 21:17:04]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-16 21:17:05]

Dry run: Try jobs failed on following builders:
  v8_linux_dbg on tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_linux_dbg/builds/12222)

[Michael Lippautz, 2015-12-17 08:48:50]

Patchset #5 (id:220001) has been deleted

[Michael Lippautz, 2015-12-17 09:14:01]

The CQ bit was checked by mlippautz@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-17 09:14:09]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1487853002/230001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1487853002/230001

[commit-bot: I haz the power, 2015-12-17 09:18:43]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-17 09:18:44]

Dry run: Try jobs failed on following builders:
  v8_linux64_avx2_rel on tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_linux64_avx2_rel/builds/...)

[Michael Lippautz, 2015-12-17 09:22:26]

Patchset #5 (id:230001) has been deleted

[Michael Lippautz, 2015-12-17 09:22:44]

The CQ bit was checked by mlippautz@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-17 09:22:50]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1487853002/250001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1487853002/250001

[Hannes Payer (out of office), 2015-12-18 10:29:06]

LGTM, if performance is OK.

https://codereview.chromium.org/1487853002/diff/290001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1487853002/diff/290001/src/heap/spaces.h#newc...
src/heap/spaces.h:1871: class LocalAllocationBuffer {
This class needs a bunch of comments, explaining what is going on.

[Michael Lippautz, 2015-12-18 10:53:05]

Verified locally (yesterday) using Octane. Will run the perf bots once more, but
don't expect any issue right now.

https://codereview.chromium.org/1487853002/diff/290001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1487853002/diff/290001/src/heap/spaces.h#newc...
src/heap/spaces.h:1871: class LocalAllocationBuffer {
On 2015/12/18 10:29:06, Hannes Payer wrote:
> This class needs a bunch of comments, explaining what is going on.

Done.

[Michael Lippautz, 2015-12-18 10:54:16]

The CQ bit was checked by mlippautz@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-18 10:54:26]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1487853002/310001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1487853002/310001

[commit-bot: I haz the power, 2015-12-18 11:20:28]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-18 11:20:29]

Dry run: This issue passed the CQ dry run.

[Michael Lippautz, 2015-12-18 16:56:31]

The CQ bit was checked by mlippautz@chromium.org

[Michael Lippautz, 2015-12-18 16:56:32]

The patchset sent to the CQ was uploaded after l-g-t-m from hpayer@chromium.org
Link to the patchset: https://codereview.chromium.org/1487853002/#ps310001
(title: "Added comments describing LocalAllocationBuffer")

[commit-bot: I haz the power, 2015-12-18 16:56:38]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1487853002/310001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1487853002/310001

[commit-bot: I haz the power, 2015-12-18 18:33:16]

Committed patchset #8 (id:310001)

[commit-bot: I haz the power, 2015-12-18 18:33:56]

Description was changed from

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Semantical change: Previously we did fall back to regular new space promotion if
we are OOM in old space. With this CL we fall back to new space promotion, which
could fail because of fragmentation, again leading to an old space allocation
that finally bails into OOM.

Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.

BUG=chromium:524425
LOG=N
==========

to

==========
[heap] Move to LAB-based allocation for newspace evacuation.

This CL prepare newspace evacuation for parallel execution wrt. to actual
allocations. The priority for allocations is:
* Try to allocate from LAB if objects are below kMaxLabObjectSize
* Allocate directly (synchronized) from newspace for larger objects.
* Fall back to old space allocation (which will be backed by a local compaction
  space in future).

Semantical change: Previously we did fall back to regular new space promotion if
we are OOM in old space. With this CL we fall back to new space promotion, which
could fail because of fragmentation, again leading to an old space allocation
that finally bails into OOM.

Newspace evacuation is still single threaded and requires further changes to
allocation site tracking.

BUG=chromium:524425
LOG=N

Committed: https://crrev.com/a4e3a3b6a81ef39835c757145ff55299d713bcba
Cr-Commit-Position: refs/heads/master@{#32970}
==========

[commit-bot: I haz the power, 2015-12-18 18:33:56]

Patchset 8 (id:??) landed as
https://crrev.com/a4e3a3b6a81ef39835c757145ff55299d713bcba
Cr-Commit-Position: refs/heads/master@{#32970}