Paint collapsed borders of a table as one display item

Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression (https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208, 664887
TBR=chrishtr@chromium.org (for platform/graphics/paint/DisplayItem.h with changed display item types about collapsed borders)

Review-Url: https://codereview.chromium.org/2430313004
Committed: https://crrev.com/16363da3f8fba57bef3d203bbb8590bf4f60037e
Cr-Original-Commit-Position: refs/heads/master@{#431744}
Cr-Commit-Position: refs/heads/master@{#436426}

[Xianzhu, 2016-11-16 19:55:51]

Description was changed from

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL actually improves performance of one test
(blink_perf.paint.large-table-background-change-with-invisible-collapsed-borders.html),
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression [2].

[1] https://codereview.chromium.org/2430313004/#ps60001
[2] https://ct.skia.org/chromium_perf_runs/, run 1433

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

to

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL actually improves performance of one test
(blink_perf.paint.large-table-background-change-with-invisible-collapsed-borders.html),
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression [2].

[1] https://codereview.chromium.org/2430313004/#ps60001
[2] https://ct.skia.org/chromium_perf_runs/, run 1433

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

[Xianzhu, 2016-11-16 19:57:07]

Description was changed from

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL actually improves performance of one test
(blink_perf.paint.large-table-background-change-with-invisible-collapsed-borders.html),
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression [2].

[1] https://codereview.chromium.org/2430313004/#ps60001
[2] https://ct.skia.org/chromium_perf_runs/, run 1433

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

to

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL actually improves performance of one test
(blink_perf.paint.large-table-background-change-with-invisible-collapsed-borders.html),
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression [2].

[1] https://codereview.chromium.org/2430313004/#ps60001
[2] https://ct.skia.org/chromium_perf_runs/, run 1433

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

[Xianzhu, 2016-12-02 23:08:10]

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

[commit-bot: I haz the power, 2016-12-02 23:09:01]

Dry run: CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-12-03 01:31:25]

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

[commit-bot: I haz the power, 2016-12-03 01:31:26]

Dry run: Try jobs failed on following builders:
  win_chromium_rel_ng on master.tryserver.chromium.win (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_...)

[Xianzhu, 2016-12-03 21:27:27]

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

[commit-bot: I haz the power, 2016-12-03 21:27:39]

Dry run: CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-12-03 23:24:14]

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

[commit-bot: I haz the power, 2016-12-03 23:24:15]

Dry run: This issue passed the CQ dry run.

[Xianzhu, 2016-12-04 05:17:49]

Description was changed from

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL actually improves performance of one test
(blink_perf.paint.large-table-background-change-with-invisible-collapsed-borders.html),
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression [2].

[1] https://codereview.chromium.org/2430313004/#ps60001
[2] https://ct.skia.org/chromium_perf_runs/, run 1433

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

to

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

[Xianzhu, 2016-12-04 05:18:49]

wangxianzhu@chromium.org changed reviewers:
+ chrishtr@chromium.org, wkorman@chromium.org

[wkorman, 2016-12-05 05:49:22]

lgtm

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
File third_party/WebKit/Source/core/layout/LayoutTable.cpp (right):

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTable.cpp:736:
invalidateCollapsedBorders(PaintInvalidationForcedByLayout);
I assume this, and the check for it in line 764, are the new bits?

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTable.cpp:794:
!shouldDoFullPaintInvalidation()) {
Do we potentially reduce a bit of work if we put shouldDoFullPaintInvalidation()
first? collectBorderValues does a bunch of stuff whereas
shouldDoFullPaintInvalidation just checks a bit.

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
File third_party/WebKit/Source/core/layout/LayoutTableCell.cpp (right):

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTableCell.cpp:1297: if
(!newValues.startBorder.isVisible() && !newValues.endBorder.isVisible() &&
Just a thought: could improve readability if we move this to a
CollapsedBorderValues::allBordersVisible() method.

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTableCell.cpp:1308: changed =
!m_collapsedBorderValues->startBorder.visuallyEquals(
Similar just-a-thought: could add a
CollapsedBorderValues::bordersVisuallyEqual(CollapsedBorderValues) method to
coalesce the below.

[Xianzhu, 2016-12-05 17:27:57]

The CQ bit was checked by wangxianzhu@chromium.org

[commit-bot: I haz the power, 2016-12-05 17:28:18]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[Xianzhu, 2016-12-05 17:28:54]

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
File third_party/WebKit/Source/core/layout/LayoutTable.cpp (right):

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTable.cpp:736:
invalidateCollapsedBorders(PaintInvalidationForcedByLayout);
On 2016/12/05 05:49:21, wkorman wrote:
> I assume this, and the check for it in line 764, are the new bits?

Yes. This ensures invalidation of collapsed borders on layout change.

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTable.cpp:794:
!shouldDoFullPaintInvalidation()) {
On 2016/12/05 05:49:21, wkorman wrote:
> Do we potentially reduce a bit of work if we put
shouldDoFullPaintInvalidation()
> first? collectBorderValues does a bunch of stuff whereas
> shouldDoFullPaintInvalidation just checks a bit.

No. We should always do collectBorderValues().

Changed to
  bool cellCollapsedBorderChanged = cell->collectBorderValues(values);
  if (cellCollapsedBorderChanged && !shouldDoFullPaintInvalidation())...

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
File third_party/WebKit/Source/core/layout/LayoutTableCell.cpp (right):

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTableCell.cpp:1297: if
(!newValues.startBorder.isVisible() && !newValues.endBorder.isVisible() &&
On 2016/12/05 05:49:22, wkorman wrote:
> Just a thought: could improve readability if we move this to a
> CollapsedBorderValues::allBordersVisible() method.

Done.

https://codereview.chromium.org/2502353003/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/LayoutTableCell.cpp:1308: changed =
!m_collapsedBorderValues->startBorder.visuallyEquals(
On 2016/12/05 05:49:22, wkorman wrote:
> Similar just-a-thought: could add a
> CollapsedBorderValues::bordersVisuallyEqual(CollapsedBorderValues) method to
> coalesce the below.

Done.

[Xianzhu, 2016-12-05 17:29:16]

The CQ bit was unchecked by wangxianzhu@chromium.org

[Xianzhu, 2016-12-05 17:30:00]

The CQ bit was checked by wangxianzhu@chromium.org

[Xianzhu, 2016-12-05 17:30:01]

The patchset sent to the CQ was uploaded after l-g-t-m from wkorman@chromium.org
Link to the patchset: https://codereview.chromium.org/2502353003/#ps60001
(title: "-")

[commit-bot: I haz the power, 2016-12-05 17:30:18]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-12-05 17:37:59]

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

[commit-bot: I haz the power, 2016-12-05 17:38:01]

Try jobs failed on following builders:
  chromium_presubmit on master.tryserver.chromium.linux (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.linux/builders/chromium_presub...)

[Xianzhu, 2016-12-05 19:03:37]

Description was changed from

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

to

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887
TBR=chrishtr@chromium.org (for platform/graphics/paint/DisplayItem.h with
changed display item types about collapsed borders)

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

[Xianzhu, 2016-12-05 19:03:44]

The CQ bit was checked by wangxianzhu@chromium.org

[commit-bot: I haz the power, 2016-12-05 19:04:09]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-12-05 22:27:04]

CQ is committing da patch.

Bot data: {"patchset_id": 60001, "attempt_start_ts": 1480964624059930,
"parent_rev": "1e27f2189aeb4df6dc1beeb324106bba1f65f290", "commit_rev":
"eba9a68f9ff30d5373146110a57d18c964e61971"}

[commit-bot: I haz the power, 2016-12-05 22:27:33]

Description was changed from

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887
TBR=chrishtr@chromium.org (for platform/graphics/paint/DisplayItem.h with
changed display item types about collapsed borders)

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

to

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887
TBR=chrishtr@chromium.org (for platform/graphics/paint/DisplayItem.h with
changed display item types about collapsed borders)

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

[commit-bot: I haz the power, 2016-12-05 22:27:35]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2016-12-05 22:30:22]

Description was changed from

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887
TBR=chrishtr@chromium.org (for platform/graphics/paint/DisplayItem.h with
changed display item types about collapsed borders)

Review-Url: https://codereview.chromium.org/2430313004
Cr-Commit-Position: refs/heads/master@{#431744}
==========

to

==========
Paint collapsed borders of a table as one display item

This simplifies paint invalidation logic on collapsed border changes.

About performance:
With this CL we create a big display item on the table containing all
collapsed border paintings, instead of many small display items on cells.
This will cause more painting when the display item is invalidated, but
saves time to handle many cached display items. Perf try jobs [1] show
that this CL improves performance of one test
(blink_perf.paint.large-table-background-change-with-visible-collapsed-borders.html),
and degrade performance of one test
(blink_perf.paint.large-table-repaint.html)
and doesn't affect much of performance of other tests.
Cluster-telemetry run showed no change or slight progression
(https://ct.skia.org/chromium_perf_runs/, run 1433)

This is a reland of https://codereview.chromium.org/2430313004, with
crbug.com/664887 fixed by force invalidation of collapsed borders on
layout change.

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
BUG=663208,664887
TBR=chrishtr@chromium.org (for platform/graphics/paint/DisplayItem.h with
changed display item types about collapsed borders)

Review-Url: https://codereview.chromium.org/2430313004
Committed: https://crrev.com/16363da3f8fba57bef3d203bbb8590bf4f60037e
Cr-Original-Commit-Position: refs/heads/master@{#431744}
Cr-Commit-Position: refs/heads/master@{#436426}
==========

[commit-bot: I haz the power, 2016-12-05 22:30:24]

Patchset 4 (id:??) landed as
https://crrev.com/16363da3f8fba57bef3d203bbb8590bf4f60037e
Cr-Commit-Position: refs/heads/master@{#436426}

[esprehn, 2016-12-06 01:31:28]

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

[esprehn, 2016-12-06 01:31:29]

Does this mean appending a new row to a table with collapsed borders will always
need to repaint the entire table?

[Xianzhu, 2016-12-06 02:16:38]

On 2016/12/06 01:31:29, esprehn wrote:
> Does this mean appending a new row to a table with collapsed borders will
always
> need to repaint the entire table?

Yes. This is a great case which is not covered by existing paint perf tests.

With this patch, the performance changes in the following stages:

- painting: though repainting all collapsed borders of the whole table increases
cost of actual painting, the reduced number of display items reduces cost of
caching display items in PaintController. Some
PerformanceTests/Paint/large-table* tests showed performance progression mostly
because of the reduced combined cost of repainting and display item caching.

- rasterization: we will rasterize more tiles than needed, which is a big
regression in this stage. Also rasterization of each tile will replay the big
display item which defeats the benefit of R-tree which can efficiently find
eligible display items for rasterization for a tile.

We need to think about more cases, and try to eliminate any overall regressions.
Here are some thoughts:
- Use rows instead of the table as collapsed border display item client;
- Move collapsed border calculation after paint invalidation of all cells so
that we can know which cells changed layout.

Because of other tasks, I'm unlikely to work on the improvement in this month,
so I'm reverting this CL. Will reland with the improvements and more test cases
when it's ready.

[Xianzhu, 2016-12-06 02:17:57]

A revert of this CL (patchset #4 id:60001) has been created in
https://codereview.chromium.org/2556633002/ by wangxianzhu@chromium.org.

The reason for reverting is: Need to think more about performance and improve..