[heap] Add page evacuation mode for new->new

[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

Basic idea:
a) Move page within new space
b) Sweep page to make iterable and process ArrayBuffers
c) Finish sweep till next scavenge

Threshold is currently 70% live bytes, i.e., the same threshold we use
to determine fragmented pages.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel

Committed: https://crrev.com/49b23201671b25092a3c22eb85783f39b95a5f87
Cr-Commit-Position: refs/heads/master@{#36990}

[Michael Lippautz, 2016-05-25 13:50:46]

Description was changed from

==========
[heap] Add page evacuation mode for new->new

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

to

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel
==========

[Michael Lippautz, 2016-05-25 13:55:38]

Patchset #5 (id:80001) has been deleted

[Michael Lippautz, 2016-06-08 09:55:46]

Patchset #8 (id:160001) has been deleted

[Michael Lippautz, 2016-06-08 18:56:45]

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

[commit-bot: I haz the power, 2016-06-08 18:56:47]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1957323003/260001

[commit-bot: I haz the power, 2016-06-08 19:30:15]

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

[commit-bot: I haz the power, 2016-06-08 19:30:16]

Dry run: Try jobs failed on following builders:
  v8_linux64_tsan_rel on tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_linux64_tsan_rel/builds/264)

[Michael Lippautz, 2016-06-09 09:20:59]

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

[commit-bot: I haz the power, 2016-06-09 09:21:04]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1957323003/280001

[Michael Lippautz, 2016-06-09 09:21:21]

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

[Michael Lippautz, 2016-06-09 09:21:22]

PTAL

https://codereview.chromium.org/1957323003/diff/280001/src/flag-definitions.h
File src/flag-definitions.h (right):

https://codereview.chromium.org/1957323003/diff/280001/src/flag-definitions.h...
src/flag-definitions.h:273: DEFINE_INT(page_promotion_threshold, 0,
Will change this back to 70 before landing.

https://codereview.chromium.org/1957323003/diff/280001/src/heap/heap.cc
File src/heap/heap.cc (right):

https://codereview.chromium.org/1957323003/diff/280001/src/heap/heap.cc#newco...
src/heap/heap.cc:1616:
mark_compact_collector()->sweeper().EnsureNewSpaceCompleted();
This one requires IsSweepingCompleted() to be correct.

https://codereview.chromium.org/1957323003/diff/280001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1957323003/diff/280001/src/heap/mark-compact....
src/heap/mark-compact.cc:3271: static void FinalizePageSequentially(Heap* heap,
MemoryChunk* chunk,
The plan (followup) is to somehow get rid of the aborted case, so we can
completely get rid of this method.

[commit-bot: I haz the power, 2016-06-09 10:04:31]

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

[commit-bot: I haz the power, 2016-06-09 10:04:32]

Dry run: This issue passed the CQ dry run.

[Michael Lippautz, 2016-06-09 14:57:38]

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

[commit-bot: I haz the power, 2016-06-09 14:57:46]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1957323003/300001

[commit-bot: I haz the power, 2016-06-09 15:39:21]

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

[commit-bot: I haz the power, 2016-06-09 15:39:22]

Dry run: This issue passed the CQ dry run.

[ulan, 2016-06-10 12:58:55]

Looking good.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact....
src/heap/mark-compact.cc:1898: static void MoveToToSpace(Page* page, Space*
owner) {
Nit: the name seems a bit misleading since it hides the fact that we add a
replacement page to the from space.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact....
src/heap/mark-compact.cc:3702: // Slots can be in "to" space after a page has
been moved. Since there is
DCHECK that the page has new_to_new promotion flag?

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact....
src/heap/mark-compact.cc:3917: if (space != NEW_SPACE)
So we are accounting the free space in new->new pages as live objects?

This might lead to corner case GCs where SizeOfObjects() before GC is smaller
than that after GC.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.cc
File src/heap/spaces.cc (right):

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1894: void SemiSpace::PrependPage(Page* page) {
From the name of the function I would expect that it just prepends the page to
the semispace.

But it actually finds a replacement page and swaps it with the given page.

Would it make sense to break this function into two functions?
page2 = semispace2->FindReplacementPage()
new_space->SwapSemispacePages(page1, page2)

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1899: DCHECK_NE(current_page(), replacement_page);
Why this holds?

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.h#newc...
src/heap/spaces.h:2469: void PrependPage(Page* page);
A description of pre/post condition of this function would be great.

[Michael Lippautz, 2016-06-10 15:18:16]

PTAL, I rewrote the part with replacing pages.

I actually want to simplify this a bit in a follow up. Basically, there's no
need to swap them but we should only move the page. After evacuation we could
reestablish the invariants about newspace capacity, i.e., "free" pages that are
too much, and "allocate"  ones that are missing. (Same holds for  moving pages
new->old where we also replace the page.)

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact....
src/heap/mark-compact.cc:1898: static void MoveToToSpace(Page* page, Space*
owner) {
On 2016/06/10 12:58:54, ulan wrote:
> Nit: the name seems a bit misleading since it hides the fact that we add a
> replacement page to the from space.

Done.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact....
src/heap/mark-compact.cc:3702: // Slots can be in "to" space after a page has
been moved. Since there is
On 2016/06/10 12:58:54, ulan wrote:
> DCHECK that the page has new_to_new promotion flag?

Done.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/mark-compact....
src/heap/mark-compact.cc:3917: if (space != NEW_SPACE)
On 2016/06/10 12:58:54, ulan wrote:
> So we are accounting the free space in new->new pages as live objects?
> 
> This might lead to corner case GCs where SizeOfObjects() before GC is smaller
> than that after GC.

Should not be the case, because SizeOfObjects in NewSpace is always computed
using allocatable area and number of pages. As far as I see it, in the worst
case SizeOfObjects would stay constant although before a GC "live bytes <
SizeOfObjects()" did hold. 

Is this a problem?

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.cc
File src/heap/spaces.cc (right):

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1894: void SemiSpace::PrependPage(Page* page) {
On 2016/06/10 12:58:54, ulan wrote:
> From the name of the function I would expect that it just prepends the page to
> the semispace.
> 
> But it actually finds a replacement page and swaps it with the given page.
> 
> Would it make sense to break this function into two functions?
> page2 = semispace2->FindReplacementPage()
> new_space->SwapSemispacePages(page1, page2)

I restructured the page replacement. It's now a specialized one called
SwapFromSpacePageWithToSpacePage that computes the parameters for a generic one
called SwapPages.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1899: DCHECK_NE(current_page(), replacement_page);
On 2016/06/10 12:58:54, ulan wrote:
> Why this holds?

I added a comment (above !OnLastPage() -- because it's essentially the same
check). Basically, the last page should always be a proper replacement page
because in the case where we move a page, this page should still be free at the
end of the space.

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1957323003/diff/300001/src/heap/spaces.h#newc...
src/heap/spaces.h:2469: void PrependPage(Page* page);
On 2016/06/10 12:58:54, ulan wrote:
> A description of pre/post condition of this function would be great.
> 

Added comments to the new functions.

[Michael Lippautz, 2016-06-13 17:21:32]

Patchset #16 (id:340001) has been deleted

[Michael Lippautz, 2016-06-13 17:22:22]

Patchset #16 (id:360001) has been deleted

[Michael Lippautz, 2016-06-13 17:25:27]

As discussed offline: Let's do the rebalancing simplification already in this
CL. This way we avoid the complex replace logic.

PTAL

[Michael Lippautz, 2016-06-13 17:26:53]

Patchset #16 (id:380001) has been deleted

[ulan, 2016-06-14 15:38:51]

lgtm!

https://codereview.chromium.org/1957323003/diff/320001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1957323003/diff/320001/src/heap/spaces.h#newc...
src/heap/spaces.h:428: // |PAGE_NEW_NEW_PROMITION|: A page tagged with this flag
has been moved
typo PAGE_NEW_NEW_PROMOTION

https://codereview.chromium.org/1957323003/diff/400001/src/heap/spaces.cc
File src/heap/spaces.cc (right):

https://codereview.chromium.org/1957323003/diff/400001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1451: to_remove->Unlink();
Please add dcheck that we discussed offline.

[Michael Lippautz, 2016-06-15 06:49:30]

https://codereview.chromium.org/1957323003/diff/320001/src/heap/spaces.h
File src/heap/spaces.h (right):

https://codereview.chromium.org/1957323003/diff/320001/src/heap/spaces.h#newc...
src/heap/spaces.h:428: // |PAGE_NEW_NEW_PROMITION|: A page tagged with this flag
has been moved
On 2016/06/14 15:38:51, ulan wrote:
> typo PAGE_NEW_NEW_PROMOTION

Done.

https://codereview.chromium.org/1957323003/diff/400001/src/heap/spaces.cc
File src/heap/spaces.cc (right):

https://codereview.chromium.org/1957323003/diff/400001/src/heap/spaces.cc#new...
src/heap/spaces.cc:1451: to_remove->Unlink();
On 2016/06/14 15:38:51, ulan wrote:
> Please add dcheck that we discussed offline.

Done.

[Michael Lippautz, 2016-06-15 06:50:41]

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

[commit-bot: I haz the power, 2016-06-15 06:50:52]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1957323003/420001

[Michael Lippautz, 2016-06-15 07:33:28]

Description was changed from

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel
==========

to

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

Basic idea:
a) Move page within new space
b) Sweep page to make iterable and process ArrayBuffers
c) Finish sweep till next scavenge

Threshold is currently 70% live bytes, i.e., the same threshold we use 
to determine fragmented pages.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel
==========

[Michael Lippautz, 2016-06-15 07:34:18]

The CQ bit was checked by mlippautz@chromium.org

[Michael Lippautz, 2016-06-15 07:34:18]

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/1957323003/#ps440001
(title: "Switch threshold flag back to 70")

[commit-bot: I haz the power, 2016-06-15 07:34:23]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1957323003/440001

[commit-bot: I haz the power, 2016-06-15 08:18:06]

Description was changed from

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

Basic idea:
a) Move page within new space
b) Sweep page to make iterable and process ArrayBuffers
c) Finish sweep till next scavenge

Threshold is currently 70% live bytes, i.e., the same threshold we use 
to determine fragmented pages.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel
==========

to

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

Basic idea:
a) Move page within new space
b) Sweep page to make iterable and process ArrayBuffers
c) Finish sweep till next scavenge

Threshold is currently 70% live bytes, i.e., the same threshold we use 
to determine fragmented pages.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel
==========

[commit-bot: I haz the power, 2016-06-15 08:18:07]

Committed patchset #18 (id:440001)

[commit-bot: I haz the power, 2016-06-15 08:19:55]

Description was changed from

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

Basic idea:
a) Move page within new space
b) Sweep page to make iterable and process ArrayBuffers
c) Finish sweep till next scavenge

Threshold is currently 70% live bytes, i.e., the same threshold we use 
to determine fragmented pages.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel
==========

to

==========
[heap] Add page evacuation mode for new->new

Adds an evacuation mode that allows moving pages within new space without
copying objects.

Basic idea:
a) Move page within new space
b) Sweep page to make iterable and process ArrayBuffers
c) Finish sweep till next scavenge

Threshold is currently 70% live bytes, i.e., the same threshold we use
to determine fragmented pages.

BUG=chromium:581412
LOG=N
CQ_EXTRA_TRYBOTS=tryserver.v8:v8_linux_arm64_gc_stress_dbg,v8_linux_gc_stress_dbg,v8_mac_gc_stress_dbg,v8_linux64_tsan_rel,v8_mac64_asan_rel

Committed: https://crrev.com/49b23201671b25092a3c22eb85783f39b95a5f87
Cr-Commit-Position: refs/heads/master@{#36990}
==========

[commit-bot: I haz the power, 2016-06-15 08:19:57]

Patchset 18 (id:??) landed as
https://crrev.com/49b23201671b25092a3c22eb85783f39b95a5f87
Cr-Commit-Position: refs/heads/master@{#36990}

[Michael Lippautz, 2016-06-16 16:24:10]

A revert of this CL (patchset #18 id:440001) has been created in
https://codereview.chromium.org/2063013005/ by mlippautz@chromium.org.

The reason for reverting is: Fragmentation of LABs could result in increasing
memory usage (pages) instead of shrinking.

BUG=chromium:620320
LOG=N.