Don't invalidate cell when collapsed border changes

Don't invalidate cell when collapsed border changes

We paint collapsed borders within ScopeRecorder which disables display
item caching. We no more need to invalidate individual cells.

BUG=509847
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=198981

[Xianzhu, 2015-07-15 17:44:39]

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

[Xianzhu, 2015-07-15 17:44:40]

Perf try jobs show that this CL improves performance of
large-table-with-collapsed-borders-and-no-colspans and
large-table-with-collapsed-borders-and-colspans by about 6%.
http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisec...

It's weird that this CL failed to show improvement on win perf try bots.
However, disabling slimming-paint also didn't show difference
(http://build.chromium.org/p/tryserver.chromium.perf/builders/win_perf_bisect/...),
so I guess there are some issues about perf try bots.

Would like to see results on real perf bots.

[chrishtr, 2015-07-15 17:50:40]

The CQ bit was checked by chrishtr@chromium.org

[chrishtr, 2015-07-15 17:50:41]

lgtm

[commit-bot: I haz the power, 2015-07-15 17:51:11]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1234463005/1

[wkorman, 2015-07-15 18:05:20]

On 2015/07/15 at 17:44:40, wangxianzhu wrote:
> Perf try jobs show that this CL improves performance of
large-table-with-collapsed-borders-and-no-colspans and
large-table-with-collapsed-borders-and-colspans by about 6%.
>
http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisec...
> 
> It's weird that this CL failed to show improvement on win perf try bots.
However, disabling slimming-paint also didn't show difference
(http://build.chromium.org/p/tryserver.chromium.perf/builders/win_perf_bisect/...),
so I guess there are some issues about perf try bots.
> 
> Would like to see results on real perf bots.

Noogler Q -- the perf try job labels read as backwards, "Patch" value is higher
than "ToT" value for the noted tests so I assume the confusing naming is just an
artifact of the way the patch is applied as a "perf bisect" or something?

There's a slight perf difference in the disabled-SP try job but it's not as
dramatic as one would expect. Did you run with
--extra-browser-args="--disable-slimming-paint"?

I can try this on a local Windows machine just as another data point to see.

[Xianzhu, 2015-07-15 18:29:32]

The CQ bit was unchecked by wangxianzhu@chromium.org

[Xianzhu, 2015-07-15 18:32:58]

On 2015/07/15 18:05:20, wkorman wrote:
> On 2015/07/15 at 17:44:40, wangxianzhu wrote:
> > Perf try jobs show that this CL improves performance of
> large-table-with-collapsed-borders-and-no-colspans and
> large-table-with-collapsed-borders-and-colspans by about 6%.
> >
>
http://build.chromium.org/p/tryserver.chromium.perf/builders/linux_perf_bisec...
> > 
> > It's weird that this CL failed to show improvement on win perf try bots.
> However, disabling slimming-paint also didn't show difference
>
(http://build.chromium.org/p/tryserver.chromium.perf/builders/win_perf_bisect/...),
> so I guess there are some issues about perf try bots.
> > 
> > Would like to see results on real perf bots.
> 
> Noogler Q -- the perf try job labels read as backwards, "Patch" value is
higher
> than "ToT" value for the noted tests so I assume the confusing naming is just
an
> artifact of the way the patch is applied as a "perf bisect" or something?
> 

I found I actually made a mistake when interpreting the perf try job results.
According to the result, there were actually regressions with this patch. This
should not happen in theory because this patch just reduce work for the cpu. The
differences shown in the results seem just noise.

> There's a slight perf difference in the disabled-SP try job but it's not as
> dramatic as one would expect. Did you run with
> --extra-browser-args="--disable-slimming-paint"?

No. With it we can't compare results sp and non-sp results in the same try job.
I tried a patch (https://codereview.chromium.org/1226163015/) to disable
slimming-paint by default instead. I'm wondering if the try bot actually applied
the patch.

> I can try this on a local Windows machine just as another data point to see.

Thanks. I'll also try locally on Linux.

[Xianzhu, 2015-07-15 18:38:35]

On 2015/07/15 18:32:58, Xianzhu wrote:

> I'll also try locally on Linux.

Just remembered that I have tried locally. There seemed improvements, but there
was too much noise (e.g. unrelated tests having 20+% progression or regression).

[Xianzhu, 2015-07-15 18:44:43]

The CQ bit was checked by wangxianzhu@chromium.org

[Xianzhu, 2015-07-15 18:44:43]

The patchset sent to the CQ was uploaded after l-g-t-m from
chrishtr@chromium.org
Link to the patchset: https://codereview.chromium.org/1234463005/#ps20001
(title: "NeedsRebaseline")

[commit-bot: I haz the power, 2015-07-15 18:45:09]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1234463005/20001

[commit-bot: I haz the power, 2015-07-15 20:15:11]

Committed patchset #2 (id:20001) as
https://src.chromium.org/viewvc/blink?view=rev&revision=198981