[heap] Non-contiguous young generation

[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps during growing and shrinking.

BUG=chromium:581412
LOG=N

Committed: https://crrev.com/3f92137209ce3c1d46d6498678cce0944c460b4d
Cr-Commit-Position: refs/heads/master@{#35261}

[Michael Lippautz, 2016-04-04 11:54:50]

Description was changed from

==========
not working yet


[heap] Non contiguous new space backing memory

* Add a page pool for kPageSize'd memory chunks.

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

to

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

Furthermore, we add a page pool for kPageSize'd memory chunks.

BUG=chromium:581412
LOG=N
==========

[Michael Lippautz, 2016-04-04 12:42:03]

Description was changed from

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

Furthermore, we add a page pool for kPageSize'd memory chunks.

BUG=chromium:581412
LOG=N
==========

to

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps all the time.

BUG=chromium:581412
LOG=N
==========

[Michael Lippautz, 2016-04-04 20:35:08]

Patchset #6 (id:100001) has been deleted

[Michael Lippautz, 2016-04-05 06:52:51]

Patchset #7 (id:140001) has been deleted

[Michael Lippautz, 2016-04-05 06:52:55]

Patchset #6 (id:120001) has been deleted

[Michael Lippautz, 2016-04-05 07:36:03]

Description was changed from

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps all the time.

BUG=chromium:581412
LOG=N
==========

to

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps during growing and shrinking.

BUG=chromium:581412
LOG=N
==========

[Michael Lippautz, 2016-04-05 07:50:30]

Patchset #7 (id:180001) has been deleted

[Michael Lippautz, 2016-04-05 07:57:42]

mlippautz@chromium.org changed reviewers:
+ hlopko@chromium.org, hpayer@chromium.org, ulan@chromium.org

[Michael Lippautz, 2016-04-05 07:57:43]

PTAL

https://codereview.chromium.org/1853783002/diff/200001/src/heap/heap.cc
File src/heap/heap.cc (left):

https://codereview.chromium.org/1853783002/diff/200001/src/heap/heap.cc#oldco...
src/heap/heap.cc:4926: if (isolate()->snapshot_available()) {
I think removing this whole part should be fine as the serializer calls the heap
API, which uses page headers for the respective checks.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h#newc...
src/heap/spaces.h:1268: inline PageType* AllocatePage(intptr_t size, SpaceType*
owner,
In another CL I would then also merge AllocateLargePage into this one.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h#newc...
src/heap/spaces.h:2646: size_t AllocatedSinceLastGC() {
This is roughly 1:1 translation to non-cont. space of the previous behavior.

If this is too slow, we could try to keep a counter. The reason I did not add
the counter in first place is that maintaining it would be distributed
(AllocatedSinceLastGC, Reset, AddFreshPage) and is likely prone to be
out-of-sync.

[ulan, 2016-04-05 08:46:26]

Looking good, few comments.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc
File src/heap/spaces.cc (right):

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc#new...
src/heap/spaces.cc:759: area_end, NOT_EXECUTABLE, owner, &reservation);
s/NOT_EXECUTABLE/executable?

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc#new...
src/heap/spaces.cc:768: chunk_pool_.Add(chunk);
Where a pooled page gets actually release?

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1660: NewSpacePage::kAllocatableMemory, this, executable());
s/executable()/NOT_EXECUTABLE?

or fix SemiSpace::GrowTo to be consistent.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h#newc...
src/heap/spaces.h:2653: if (top() - age_mark > 0) {
Should be (top() - age_mark >= 0) to preserve the previous behavior.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h#newc...
src/heap/spaces.h:2671: current_page = current_page->next_page();
Did you mean current_page = age_mark_page->next_page()?

[Michael Lippautz, 2016-04-05 09:42:34]

PTAL

I verified that pooled allocation actually makes a difference by benchmarking
Splay without allocation site pretenuring:

Results (average +1.7% for 50 runs):
Splay-octane2.1(Score)
  out-regular-no-pretenuring-splay.log:
    samples:    50
    average:  4362.36 base
    median:   4289.50 base
    stddev:    338.87 base
    minimum:  3701    base
    maximum:  5078    base
    CI 95%:   4266.05 to  4458.67
  out-pooled-no-pretenuring-splay.log:
    samples:    50
    average:  4439.38  +1.77
    median:   4459.00  +3.95
    stddev:    387.34 +14.30
    minimum:  3629     -1.95
    maximum:  5094     +0.32
    CI 95%:   4329.30 to  4549.46

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc
File src/heap/spaces.cc (right):

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc#new...
src/heap/spaces.cc:759: area_end, NOT_EXECUTABLE, owner, &reservation);
On 2016/04/05 08:46:25, ulan wrote:
> s/NOT_EXECUTABLE/executable?

Done.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc#new...
src/heap/spaces.cc:768: chunk_pool_.Add(chunk);
On 2016/04/05 08:46:25, ulan wrote:
> Where a pooled page gets actually release?

This is now done in MemoryAllocator::TearDown.

I also added now DCHECKs for size and executability as we can only use pooled
allocation for non-executable pages of constant kPageSize. (Non-executable,
because we would need handling to differentiate pages allocated from a CodeRange
vs regular reservations.)

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1660: NewSpacePage::kAllocatableMemory, this, executable());
On 2016/04/05 08:46:25, ulan wrote:
> s/executable()/NOT_EXECUTABLE?
> 
> or fix SemiSpace::GrowTo to be consistent.

Fixed SemiSpace::GrowTo.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h#newc...
src/heap/spaces.h:2653: if (top() - age_mark > 0) {
On 2016/04/05 08:46:25, ulan wrote:
> Should be (top() - age_mark >= 0) to preserve the previous behavior. 

Done.

https://codereview.chromium.org/1853783002/diff/200001/src/heap/spaces.h#newc...
src/heap/spaces.h:2671: current_page = current_page->next_page();
On 2016/04/05 08:46:25, ulan wrote:
> Did you mean current_page = age_mark_page->next_page()?

current_page == age_mark_page at this point. Added a DCHECK and changed the
assignment.

[Michael Lippautz, 2016-04-05 10:42:32]

Patchset #9 (id:240001) has been deleted

[ulan, 2016-04-05 10:53:02]

LGTM

[Michael Lippautz, 2016-04-05 10:58:07]

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

[commit-bot: I haz the power, 2016-04-05 10:58:09]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1853783002/260001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1853783002/260001

[commit-bot: I haz the power, 2016-04-05 11:20:12]

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

[commit-bot: I haz the power, 2016-04-05 11:20:14]

Dry run: This issue passed the CQ dry run.

[Hannes Payer (out of office), 2016-04-05 12:38:49]

https://codereview.chromium.org/1853783002/diff/260001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1853783002/diff/260001/src/heap/spaces.h#newc...
src/heap/spaces.h:1283: void Free(MemoryChunk* chunk);
Maybe also templatize Free?

https://codereview.chromium.org/1853783002/diff/260001/src/heap/spaces.h#newc...
src/heap/spaces.h:2648: bool seen_age_mark = false;
Maybe we should fix this function at some point to
a) Always return allocated bytes when allocations happened, i.e., not return 0
when top was reset
b) Don't count fragmentation as allocated for fully used pages. Fragmentation
can be high due to big objects in new space.

[Michael Lippautz, 2016-04-05 12:39:34]

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

[commit-bot: I haz the power, 2016-04-05 12:39:37]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1853783002/280001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1853783002/280001

[Michael Lippautz, 2016-04-05 12:47:59]

https://codereview.chromium.org/1853783002/diff/260001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1853783002/diff/260001/src/heap/spaces.h#newc...
src/heap/spaces.h:1283: void Free(MemoryChunk* chunk);
On 2016/04/05 at 12:38:49, Hannes Payer wrote:
> Maybe also templatize Free?

Done.

https://codereview.chromium.org/1853783002/diff/260001/src/heap/spaces.h#newc...
src/heap/spaces.h:2648: bool seen_age_mark = false;
On 2016/04/05 at 12:38:49, Hannes Payer wrote:
> Maybe we should fix this function at some point to
> a) Always return allocated bytes when allocations happened, i.e., not return 0
when top was reset
> b) Don't count fragmentation as allocated for fully used pages. Fragmentation
can be high due to big objects in new space.

Ack.

[commit-bot: I haz the power, 2016-04-05 13:09:20]

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

[commit-bot: I haz the power, 2016-04-05 13:09:21]

Dry run: This issue passed the CQ dry run.

[Michael Lippautz, 2016-04-05 13:10:06]

The CQ bit was checked by mlippautz@chromium.org

[Michael Lippautz, 2016-04-05 13:10:07]

The patchset sent to the CQ was uploaded after l-g-t-m from ulan@chromium.org
Link to the patchset: https://codereview.chromium.org/1853783002/#ps280001
(title: "Hannes comment")

[commit-bot: I haz the power, 2016-04-05 13:10:10]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1853783002/280001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1853783002/280001

[commit-bot: I haz the power, 2016-04-05 13:11:55]

Description was changed from

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps during growing and shrinking.

BUG=chromium:581412
LOG=N
==========

to

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps during growing and shrinking.

BUG=chromium:581412
LOG=N
==========

[commit-bot: I haz the power, 2016-04-05 13:11:56]

Committed patchset #10 (id:280001)

[commit-bot: I haz the power, 2016-04-05 13:12:36]

Description was changed from

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps during growing and shrinking.

BUG=chromium:581412
LOG=N
==========

to

==========
[heap] Non-contiguous young generation

This change removes the large contiguous backing store from the young generation
and replaces it regular pages.

We keep a pool of pages that are committed/uncommitted to avoid creating virtual
memory maps during growing and shrinking.

BUG=chromium:581412
LOG=N

Committed: https://crrev.com/3f92137209ce3c1d46d6498678cce0944c460b4d
Cr-Commit-Position: refs/heads/master@{#35261}
==========

[commit-bot: I haz the power, 2016-04-05 13:12:37]

Patchset 10 (id:??) landed as
https://crrev.com/3f92137209ce3c1d46d6498678cce0944c460b4d
Cr-Commit-Position: refs/heads/master@{#35261}