Add smoothness test for hover during trackpad scrolling [perf implications]

Add smoothness test for hover during trackpad scrolling.

The results of this test will improve dramatically when
crbug.com/317007 is fixed.

BUG=317007
Committed: https://crrev.com/83caea53296386c8cded579499ee433f6b46bfe3
Cr-Commit-Position: refs/heads/master@{#333044}

[tdresser, 2015-05-26 15:03:48]

tdresser@chromium.org changed reviewers:
+ dtu@chromium.org, kbr@chromium.org, rbyers@chromium.org

[tdresser, 2015-05-26 15:03:50]

rbyers@, can you make sure this looks sane?
kbr@, can you take a look at gpu_benchmarking_extension.cc?
dtu@, can you take a look at tools/perf/page_sets/?

[tdresser, 2015-05-26 17:16:58]

+jdduke, for opinions on the approach here.

[Rick Byers, 2015-05-26 19:10:18]

This seems reasonable to me.  Can you submit a perf try job so we can see the
results?  In particular, are you expecting this to trigger a "regression" in any
existing benchmarks?

[jdduke (slow), 2015-05-26 19:44:10]

On 2015/05/26 19:10:18, Rick Byers wrote:
> This seems reasonable to me.  Can you submit a perf try job so we can see the
> results?  In particular, are you expecting this to trigger a "regression" in
any
> existing benchmarks?

No problems here, but we should check with the owners of tough_scrolling_cases.

[Ken Russell (switch to Gerrit), 2015-05-26 22:58:32]

kbr@chromium.org changed reviewers:
+ eakuefner@chromium.org, nednguyen@google.com

[Ken Russell (switch to Gerrit), 2015-05-26 22:58:32]

The gpu_benchmarking_extension.cc change looks OK to me but one question. Also
CC'ing some other Telemetry folks.

https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
File content/renderer/gpu/gpu_benchmarking_extension.cc (right):

https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
content/renderer/gpu/gpu_benchmarking_extension.cc:300: mouseMove.y =
(contentRect.height - contentRect.y) / 2;
Sorry if this is a stupid question, but did you mean these to be:

  mouseMove.x = contentRect.x + contentRect.width / 2;
  mouseMove.y = contentRect.y + contentRect.height / 2;

?

[nednguyen, 2015-05-27 02:13:39]

On 2015/05/26 22:58:32, Ken Russell wrote:
> The gpu_benchmarking_extension.cc change looks OK to me but one question. Also
> CC'ing some other Telemetry folks.
> 
>
https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
> File content/renderer/gpu/gpu_benchmarking_extension.cc (right):
> 
>
https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
> content/renderer/gpu/gpu_benchmarking_extension.cc:300: mouseMove.y =
> (contentRect.height - contentRect.y) / 2;
> Sorry if this is a stupid question, but did you mean these to be:
> 
>   mouseMove.x = contentRect.x + contentRect.width / 2;
>   mouseMove.y = contentRect.y + contentRect.height / 2;
> 
> ?

Does this work with the ref browser? If not, you need to find a way disable/skip
the test with ref browser.

[tdresser, 2015-05-27 14:14:15]

tdresser@chromium.org changed reviewers:
+ reveman@chromium.org

[tdresser, 2015-05-27 14:14:17]

I just realized that we're only looking at mean_input_event_latency and
mean_input_event_latency_discrepancy for this benchmark.

For this test, it would be nice to also measure
responsive-biggest_jank_thread_time. reveman@, does that seem reasonable?


nednguyen@, what are the requirements for behavior with the ref browser?
If I run the test with --browser=reference, the test does run, though the hover
effects don't trigger, giving a very distinct performance profile. Is that close
enough, or should I disable the test for the reference browser? Are there
examples of other tests which do this?


Perf tryjobs:
There's a lot of noise in this measures, but it does look like this may cause a
noticeable regression in a few metrics. I think this is desirable - I'll send a
note to the perf sheriffs list before landing.

linux tough_scrolling cases: https://codereview.chromium.org/1161693003
http://storage.googleapis.com/chromium-telemetry/html-results/results-2015-05...

android tough_scrolling cases: https://codereview.chromium.org/1155683007
Bot failure: I'll look into this.

linux key_desktop_move_cases: https://codereview.chromium.org/1155373002
http://storage.googleapis.com/chromium-telemetry/html-results/results-2015-05...

https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
File content/renderer/gpu/gpu_benchmarking_extension.cc (right):

https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
content/renderer/gpu/gpu_benchmarking_extension.cc:300: mouseMove.y =
(contentRect.height - contentRect.y) / 2;
On 2015/05/26 22:58:32, Ken Russell wrote:
> Sorry if this is a stupid question, but did you mean these to be:
> 
>   mouseMove.x = contentRect.x + contentRect.width / 2;
>   mouseMove.y = contentRect.y + contentRect.height / 2;
> 
> ?

Uh... Yeah...
Not a stupid question.
Just a stupid mistake...

Done.

[eakuefner, 2015-05-27 15:41:17]

On 2015/05/27 14:14:17, tdresser wrote:
> I just realized that we're only looking at mean_input_event_latency and
> mean_input_event_latency_discrepancy for this benchmark.
> 
> For this test, it would be nice to also measure
> responsive-biggest_jank_thread_time. reveman@, does that seem reasonable?
> 
> 
> nednguyen@, what are the requirements for behavior with the ref browser?
> If I run the test with --browser=reference, the test does run, though the
hover
> effects don't trigger, giving a very distinct performance profile. Is that
close
> enough, or should I disable the test for the reference browser? Are there
> examples of other tests which do this?
> 
> 
> Perf tryjobs:
> There's a lot of noise in this measures, but it does look like this may cause
a
> noticeable regression in a few metrics. I think this is desirable - I'll send
a
> note to the perf sheriffs list before landing.
> 
> linux tough_scrolling cases: https://codereview.chromium.org/1161693003
>
http://storage.googleapis.com/chromium-telemetry/html-results/results-2015-05...
> 
> android tough_scrolling cases: https://codereview.chromium.org/1155683007
> Bot failure: I'll look into this.
> 
> linux key_desktop_move_cases: https://codereview.chromium.org/1155373002
>
http://storage.googleapis.com/chromium-telemetry/html-results/results-2015-05...
> 
>
https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
> File content/renderer/gpu/gpu_benchmarking_extension.cc (right):
> 
>
https://codereview.chromium.org/1157093002/diff/1/content/renderer/gpu/gpu_be...
> content/renderer/gpu/gpu_benchmarking_extension.cc:300: mouseMove.y =
> (contentRect.height - contentRect.y) / 2;
> On 2015/05/26 22:58:32, Ken Russell wrote:
> > Sorry if this is a stupid question, but did you mean these to be:
> > 
> >   mouseMove.x = contentRect.x + contentRect.width / 2;
> >   mouseMove.y = contentRect.y + contentRect.height / 2;
> > 
> > ?
> 
> Uh... Yeah...
> Not a stupid question.
> Just a stupid mistake...
> 
> Done.

As long as it runs on the ref build that's fine -- the thing that we would be
concerned about is if it crashed the browser, because that would turn the perf
waterfall red.

[Ken Russell (switch to Gerrit), 2015-05-27 17:44:54]

gpu_benchmarking_extension.cc lgtm

[nednguyen, 2015-05-27 17:55:28]

On 2015/05/27 17:44:54, Ken Russell wrote:
> gpu_benchmarking_extension.cc lgtm

lgtm if this does not crash the ref browser.

[eakuefner, 2015-05-27 20:35:17]

lgtm

[Rick Byers, 2015-05-27 23:46:21]

lgtm

[reveman, 2015-05-30 01:09:08]

lgtm

[tdresser, 2015-06-05 13:25:10]

The CQ bit was checked by tdresser@chromium.org

[tdresser, 2015-06-05 13:25:11]

The patchset sent to the CQ was uploaded after l-g-t-m from kbr@chromium.org,
nednguyen@google.com, rbyers@chromium.org, eakuefner@chromium.org
Link to the patchset: https://codereview.chromium.org/1157093002/#ps40001
(title: "Pay attention to page_scale_factor.")

[commit-bot: I haz the power, 2015-06-05 13:25:23]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1157093002/40001

[commit-bot: I haz the power, 2015-06-05 14:33:02]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2015-06-05 14:33:58]

Patchset 3 (id:??) landed as
https://crrev.com/83caea53296386c8cded579499ee433f6b46bfe3
Cr-Commit-Position: refs/heads/master@{#333044}