Turn overdraw metrics on only when about:tracing is recording.

Turn overdraw metrics on only when about:tracing is recording.

Now that we're in chromium, we can tell if about:tracing is enabled ot not!
Turn on overdraw metrics, and their expensive computations, only when
about:tracing is turned on.

When we do turn them on, we don't want the performance characteristics of
the system to suddenly change, or the tracing is not very meaningful! So, we
track the number of pixels read, instead of written, for overdraw (which
should mostly be the same with the new rasterScale feature). This
computation is very cheap compared to the old one.

R=jamesr
BUG=119126
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=166464

[jamesr, 2012-11-07 04:44:01]

How useful is this even when tracing is on?  I like disabling expensive things
when not tracing, but when we _are_ tracing we want to have an accurate picture
of what's going on so maybe we should just drop this (or at least put it in an
isolated trace so we can see and discount it)

[danakj, 2012-11-07 04:49:36]

On 2012/11/07 04:44:01, jamesr wrote:
> How useful is this even when tracing is on?  I like disabling expensive things
> when not tracing, but when we _are_ tracing we want to have an accurate
picture
> of what's going on so maybe we should just drop this (or at least put it in an
> isolated trace so we can see and discount it)

It's only useful if you want to see how much culling is happening. I don't know
if anyone has ever looked at these stats or not.

We could stick it behind a command line flag instead, if you like, or I can
remove them entirely and we can store something cheap in renderstats in the
future. http://code.google.com/p/chromium/issues/detail?id=147295 ?

[danakj, 2012-11-07 04:50:19]

I'm not sure what you mean by isolated trace. Like, not in the "cc" category?

[jamesr, 2012-11-07 04:52:46]

On 2012/11/07 04:49:36, danakj wrote:
> On 2012/11/07 04:44:01, jamesr wrote:
> > How useful is this even when tracing is on?  I like disabling expensive
things
> > when not tracing, but when we _are_ tracing we want to have an accurate
> picture
> > of what's going on so maybe we should just drop this (or at least put it in
an
> > isolated trace so we can see and discount it)
> 
> It's only useful if you want to see how much culling is happening. I don't
know
> if anyone has ever looked at these stats or not.
> 
> We could stick it behind a command line flag instead, if you like, or I can
> remove them entirely and we can store something cheap in renderstats in the
> future. http://code.google.com/p/chromium/issues/detail?id=147295 ?

I really like the ability to compute overdraw since it's something we need to
look closely at, but this method seems too expensive for now.  I say remove or
put it behind a flag (and the latter is relatively easy to do), but definitely
do one or the other.

[jamesr, 2012-11-07 04:53:17]

On 2012/11/07 04:50:19, danakj wrote:
> I'm not sure what you mean by isolated trace. Like, not in the "cc" category?

Sorry, I meant put the computation in a trace.  But that's clearly not possible.

[danakj, 2012-11-07 04:56:20]

The other option is to stop mapping to the target space.

Currently we count the number of pixels we save in writing. That would instead
count the number of pixels we save in reading. And it makes things very cheap.

When I turn on overdraw metrics always, I see didDraw now at 0.77% of time, and
we can do this only when tracing is on.

WDYT, good compromise?

[danakj, 2012-11-07 05:00:19]

Uploaded cheap metrics version.

[jamesr, 2012-11-07 05:48:30]

lgtm, doesn't look so good to the unit tests (kinda surprised we have unit tests
for a pure metrics thing).

[danakj, 2012-11-07 16:33:09]

On 2012/11/07 05:48:30, jamesr wrote:
> lgtm, doesn't look so good to the unit tests (kinda surprised we have unit
tests
> for a pure metrics thing).

ok cool, thanks. gotta be thorough :) will rebaseline tests.

[commit-bot: I haz the power, 2012-11-07 16:45:27]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/danakj@chromium.org/11293143/4002

[commit-bot: I haz the power, 2012-11-07 16:47:15]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/danakj@chromium.org/11293143/4002