LayoutTable: gracefully degrade in the presence of collspan > 1.

LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up large-table-with-collapsed-borders-and-colspans-wider-than-table
microbenchmark by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
Committed: https://crrev.com/60f47383d88f942923b5ff963023d4547397acc2
Cr-Commit-Position: refs/heads/master@{#417829}

[krasin1, 2016-09-09 07:22:32]

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

[commit-bot: I haz the power, 2016-09-09 07:22:47]

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, 2016-09-09 08:35:22]

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

[commit-bot: I haz the power, 2016-09-09 08:35:23]

Dry run: This issue passed the CQ dry run.

[krasin1, 2016-09-09 18:45:52]

Description was changed from

==========
Gracefully degrade in the presence of collspan > 1.

BUG=643724
==========

to

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path when mapping
absolute columns to effective columns (and back), but the fast path is
completely abandoned, if a single call with colspan > 1 presented.

This change allows gracefully degrade the performance of the fast path by
maintaining the number of first cells with colspan = 1, which allows to skip
them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table micro-benchmark
by approximately 2x.

BUG=643724
==========

[krasin1, 2016-09-09 18:48:23]

Description was changed from

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path when mapping
absolute columns to effective columns (and back), but the fast path is
completely abandoned, if a single call with colspan > 1 presented.

This change allows gracefully degrade the performance of the fast path by
maintaining the number of first cells with colspan = 1, which allows to skip
them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table micro-benchmark
by approximately 2x.

BUG=643724
==========

to

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table micro-benchmark
by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
==========

[krasin1, 2016-09-09 18:48:46]

Description was changed from

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table micro-benchmark
by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
==========

to

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table
microbenchmark by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
==========

[krasin1, 2016-09-09 18:52:22]

krasin@chromium.org changed reviewers:
+ eae@chromium.org

[krasin1, 2016-09-09 18:52:23]

Hi Emil,

while investigating perf regression reported in https://crbug.com/643724, I have
come to two findings:

1. Clang codegen regressed a bit by inserting an unnecessary mov into a hot
loop. That's irrelevant to the CL.
2. The majority of the time is spent in the slow path for mapping absolute
columns to effective columns. That's what I partially fix here by degrading the
fast path gracefully.

Please, take a look!

[Nico, 2016-09-09 19:21:04]

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

[commit-bot: I haz the power, 2016-09-09 19:21:38]

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, 2016-09-09 20:46:39]

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

[commit-bot: I haz the power, 2016-09-09 20:46:40]

Dry run: This issue passed the CQ dry run.

[krasin1, 2016-09-09 21:02:26]

CL for src perf tryjob to run blink_perf.layout benchmark on linux platform(s)

[Nico, 2016-09-09 23:38:10]

eae should review this, he actually knows this code.

But since we seem to care about this metric (we have a benchmark for it), this
looks like a great change :-)

[krasin1, 2016-09-09 23:57:38]

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

[commit-bot: I haz the power, 2016-09-09 23:58:01]

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, 2016-09-10 01:33:28]

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

[commit-bot: I haz the power, 2016-09-10 01:33:29]

Dry run: This issue passed the CQ dry run.

[eae, 2016-09-10 10:32:00]

This is great and seems well worth the slight increase in memory usage for
RenderTable. Thanks for working on this and the detailed explanation!

LGTM

[eae, 2016-09-10 10:32:30]

The CQ bit was checked by eae@chromium.org

[commit-bot: I haz the power, 2016-09-10 10:32:34]

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, 2016-09-10 10:35:22]

Description was changed from

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table
microbenchmark by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
==========

to

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table
microbenchmark by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
==========

[commit-bot: I haz the power, 2016-09-10 10:35:23]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2016-09-10 10:37:19]

Description was changed from

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table
microbenchmark by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724
==========

to

==========
LayoutTable: gracefully degrade in the presence of collspan > 1.

Currently, there's a boolean field that allows to take a fast path
when mapping absolute columns to effective columns (and back), but
the fast path is completely abandoned, if a single call with
colspan > 1 presented.

This change allows gracefully degrade the performance of the fast
path by maintaining the number of first cells with colspan = 1,
which allows to skip them, even if some cells have colspan > 1.

This change speeds up
large-table-with-collapsed-borders-and-colspans-wider-than-table
microbenchmark by approximately 2x, see:
https://storage.googleapis.com/cfi-stats/2016-09-09/results.html

BUG=643724

Committed: https://crrev.com/60f47383d88f942923b5ff963023d4547397acc2
Cr-Commit-Position: refs/heads/master@{#417829}
==========

[commit-bot: I haz the power, 2016-09-10 10:37:20]

Patchset 4 (id:??) landed as
https://crrev.com/60f47383d88f942923b5ff963023d4547397acc2
Cr-Commit-Position: refs/heads/master@{#417829}