cc: Always try to fill the upload queue.

cc: Always try to fill the upload queue.

This keeps our pipeline better filled by priming the pipeline on
the first upload set as well as filling it back up to a specific target
if we fall behind.

Since draws are left vsync aligned on the Render process, we shouldn't
jank if the queued uploads push the draw out a small amount on the
GPU process.

If the latency between the render and gpu process is to high, however,
we turn off targeted texture throttling because we don't have enough
feedback to make good decisions.

GUG=145825

[reveman, 2012-09-22 00:50:26]

I like this a lot. Just a few things I think we need to fix to get it right.

A suggestion that I think will make things easier is to make
performMoreUploads() look like:

if (!m_firstUpdateAttempt)
    updateMoreTexturesNow();

if (!updateMoreTexturesIfEnoughTimeRemaining())
    m_timer->startOneShot(0);

m_firstUpdateAttempt = false;

https://codereview.chromium.org/10963039/diff/1/cc/CCTextureUpdateController.cpp
File cc/CCTextureUpdateController.cpp (right):

https://codereview.chromium.org/10963039/diff/1/cc/CCTextureUpdateController....
cc/CCTextureUpdateController.cpp:155:
m_timer->startOneShot(uploaderBusyTickRate);
I think you have to include this logic in the loop below or we might return from
this function with a busy uploader but no timer set.

https://codereview.chromium.org/10963039/diff/1/cc/CCTextureUpdateController....
cc/CCTextureUpdateController.cpp:160: return false;
And I think you need to include this in the loop below too. This function should
return false when the uploader isn't busy AND we have no more full uploads.

https://codereview.chromium.org/10963039/diff/1/cc/CCTextureUpdateController....
cc/CCTextureUpdateController.cpp:175: m_timer->startOneShot(
Awkward to call startOneShot() here now that this function is called multiple
times. How about we move this to updateMoreTexturesIfEnoughTimeRemaining() and
make sure updateMoreTexturesIfEnoughTimeRemaining() is always called from
performMoreUpdates()?

[reveman, 2012-09-22 19:05:43]

Btw, we might want to land https://chromiumcodereview.appspot.com/10933095/
before this. It will make this CL more powerful and also less complicated.

[brianderson, 2012-10-11 01:06:23]

This works really well combined with https://codereview.chromium.org/11028129/.

Detecting a high latency mode ended up being very simple.

[reveman, 2012-12-06 15:49:00]

On 2012/10/11 01:06:23, Brian Anderson wrote:
> This works really well combined with
https://codereview.chromium.org/11028129/.
> 
> Detecting a high latency mode ended up being very simple.

can this be closed? part of this has landed.