[Cast] Compute utilization metrics and add performance overlay.

[Cast] Compute utilization metrics and add performance overlay.

Adds computation of two frame-level utilization metrics to the software
VP8 encoder in Cast: deadline utilization and lossy utilization.  The
first is a measure of how long the encoding of each frame takes compared
to the frame duration.  The second is a measure of the complexity of a
frame, in terms of the quality versus encoded size trade-off.

In a future change, these utilization metrics will be sent as feedback
signals to the producer of the video frames, allowing the producer to
adjust data volumes based on the consumer's capability throughout a
session.  See bug for more details.

Also, this change adds an overlay display, where frame-level performance
metrics are rendered in the lower-right corner of each video frame just
before it is sent.  This provides an "on screen display" of end-to-end
system performance.  This is turned on with a command line argument:
--vmodule=performance_metrics_overlay=3

BUG=156767
Committed: https://crrev.com/d9e741512cd6ea56cefd0173b371d0dd29f0cce5
Cr-Commit-Position: refs/heads/master@{#330896}

[miu, 2015-05-20 20:55:46]

Patchset #1 (id:1) has been deleted

[miu, 2015-05-20 20:57:10]

miu@chromium.org changed reviewers:
+ hubbe@chromium.org

[miu, 2015-05-20 20:57:11]

hubbe: Get ready for some ASCII art!  :)

[hubbe, 2015-05-20 21:32:19]

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
File media/cast/sender/performance_metrics_overlay.cc (right):

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:36: const int
kDivergeDownThreshold = 32;
Why not make sure that kDivergeDownThreshold is >= kDivergeDownAmount +
kMinIntensity and get rid of the std::max() call?

Also, why is kDivergeDownThreshold != kDivergeUpThreshold?

Perhaps add some comments to explain why you chose these values?

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:81: // ...now move left to
where line[0] would be rendered...
This seems to assume that the actual image is right-adjusted within |stride|,
which I very much doubt is the case.
Also, if you're going to have comments inside the statement, move the operator
(+/-) to the beginning of the line.

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:86: // Render each character.
Did you explore using ttf fonts or something here?

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:168: // ...follow through...
s/follow/fall/

[miu, 2015-05-20 23:44:01]

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
File media/cast/sender/performance_metrics_overlay.cc (right):

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:36: const int
kDivergeDownThreshold = 32;
On 2015/05/20 21:32:18, hubbe wrote:
> Why not make sure that kDivergeDownThreshold is >= kDivergeDownAmount +
> kMinIntensity and get rid of the std::max() call?
> 
> Also, why is kDivergeDownThreshold != kDivergeUpThreshold?
> 
> Perhaps add some comments to explain why you chose these values?

Yeah, all this was the result of experimenting with a variety of heuristics and
thresholds.  I wrote a comment explaining the reasoning and the mapping of
before-->after values.

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:81: // ...now move left to
where line[0] would be rendered...
On 2015/05/20 21:32:18, hubbe wrote:
> This seems to assume that the actual image is right-adjusted within |stride|,
> which I very much doubt is the case.
> Also, if you're going to have comments inside the statement, move the operator
> (+/-) to the beginning of the line.

Good catch!  Fixed.

https://codereview.chromium.org/1148233002/diff/20001/media/cast/sender/perfo...
media/cast/sender/performance_metrics_overlay.cc:86: // Render each character.
On 2015/05/20 21:32:18, hubbe wrote:
> Did you explore using ttf fonts or something here?

There were too many unknowns when looking at what's available in Skia and/or
ui::gfx classes.  Also, some code in ui::gfx says to do things only on the UI
thread, which is a show-stopper here.

As discussed face-to-face, I'm going to leave this as-is for now.  If we
discover a better way (or need a better way) in the future, we can replace this
pretty easily.

[hubbe, 2015-05-20 23:52:46]

lgtm

[miu, 2015-05-21 00:16:30]

The CQ bit was checked by miu@chromium.org

[commit-bot: I haz the power, 2015-05-21 00:16:48]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1148233002/40001

[commit-bot: I haz the power, 2015-05-21 04:44:27]

Committed patchset #2 (id:40001)

[commit-bot: I haz the power, 2015-05-21 04:46:03]

Patchset 2 (id:??) landed as
https://crrev.com/d9e741512cd6ea56cefd0173b371d0dd29f0cce5
Cr-Commit-Position: refs/heads/master@{#330896}

[battre, 2015-05-21 06:42:44]

A revert of this CL (patchset #2 id:40001) has been created in
https://codereview.chromium.org/1143363005/ by battre@chromium.org.

The reason for reverting is: Reverting due to compile failure on Google Chrome
Mac builder:
../../media/cast/sender/h264_vt_encoder_unittest.cc:282:38: error: no viable
conversion from 'Callback<typename internal::BindState<typename
internal::FunctorTraits<void (MetadataRecorder::*)(scoped_ptr<EncodedFrame,
DefaultDeleter<EncodedFrame> >)>::RunnableType, typename
internal::FunctorTraits<void (MetadataRecorder::*)(scoped_ptr<EncodedFrame,
DefaultDeleter<EncodedFrame> >)>::RunType, internal::TypeList<typename
internal::CallbackParamTraits<MetadataRecorder *>::StorageType>
>::UnboundRunType>' to 'Callback<void
(scoped_ptr<media::cast::SenderEncodedFrame>)>'
  VideoEncoder::FrameEncodedCallback cb = base::Bind(
                                     ^    ~~~~~~~~~~~
../../base/callback.h:358:7: note: candidate constructor (the implicit move
constructor) not viable: no known conversion from 'base::Callback<typename
internal::BindState<typename internal::FunctorTraits<void
(MetadataRecorder::*)(scoped_ptr<EncodedFrame, DefaultDeleter<EncodedFrame>
>)>::RunnableType, typename internal::FunctorTraits<void
(MetadataRecorder::*)(scoped_ptr<EncodedFrame, DefaultDeleter<EncodedFrame>
>)>::RunType, internal::TypeList<typename
internal::CallbackParamTraits<MetadataRecorder *>::StorageType>
>::UnboundRunType>' to 'base::Callback<void
(scoped_ptr<media::cast::SenderEncodedFrame,
base::DefaultDeleter<media::cast::SenderEncodedFrame> >)> &&' for 1st argument
class Callback;

http://build.chromium.org/p/chromium.chrome/buildstatus?builder=Google%20Chro....