Don't spam the GPU process with memory usage stats

Round managed memory stats to the nearest 8MB, and don't perform
redundant updates, to avoid frequently descheduling the renderer.

BUG=164947
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=171949

[ccameron, 2012-12-07 21:58:41]

This should throttle things somewhat. I found in some experiments, that the
visible/visible-and-nearby bytes will shuffle around by ~16MB (or more) as we
scroll, so this doesn't eliminate all of the spamming.

Do you have a bug number?

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc
File cc/layer_tree_host_impl.cc (right):

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc#ne...
cc/layer_tree_host_impl.cc:1633: // Round the numbers being sent up to the next
8MB, to throttle the rate
Do we have a ROUND_UP macro, or something like that?

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.h
File cc/layer_tree_host_impl.h (right):

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.h#new...
cc/layer_tree_host_impl.h:256: size_t memoryUseBytes);
I'll get rid of memoryUseBytes soon, replacing it with a GL extension.

[jamesr, 2012-12-07 22:04:42]

On 2012/12/07 21:58:41, ccameron1 wrote:
> This should throttle things somewhat. I found in some experiments, that the
> visible/visible-and-nearby bytes will shuffle around by ~16MB (or more) as we
> scroll, so this doesn't eliminate all of the spamming.
> 
> Do you have a bug number?

Nope

> 
> https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc
> File cc/layer_tree_host_impl.cc (right):
> 
>
https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc#ne...
> cc/layer_tree_host_impl.cc:1633: // Round the numbers being sent up to the
next
> 8MB, to throttle the rate
> Do we have a ROUND_UP macro, or something like that?

v8 has one (that we can't use) and texture_uploader.cc has one. I don't see one
in chromium.  Could you extract the texture_uploader one out to somewhere shared
for cc::, perhaps?

[jamesr, 2012-12-07 22:14:33]

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc
File cc/layer_tree_host_impl.cc (right):

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc#ne...
cc/layer_tree_host_impl.cc:1639: if (m_lastSentMemoryVisibleBytes ==
memoryVisibleBytes &&
alternately, why not just calculate a diff from the last value sent and send
again if it's >= threshold? avoids the rounding math completely

[ccameron, 2012-12-07 22:33:49]

I added cc/util.h, because the alternatives (trying to get something into base
or copy-paste) didn't seem nice.

I'm a fan of having this here as a "going towards base" header (where, if
something becomes useful enough, we can slowly push it to base).

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc
File cc/layer_tree_host_impl.cc (right):

https://codereview.chromium.org/11470024/diff/1/cc/layer_tree_host_impl.cc#ne...
cc/layer_tree_host_impl.cc:1639: if (m_lastSentMemoryVisibleBytes ==
memoryVisibleBytes &&
On 2012/12/07 22:14:34, jamesr wrote:
> alternately, why not just calculate a diff from the last value sent and send
> again if it's >= threshold? avoids the rounding math completely

The is less artifact-prone (it doesn't depend on the previous state). It also
always errors on being high.

[jamesr, 2012-12-07 23:19:11]

lgtm

it amuses me that cc/util.h has a math utility and cc/math_util.h has geometry
stuff :). but cc/math_util.h is slated for eventual nukage, so this is fine IMO

https://codereview.chromium.org/11470024/diff/6001/cc/layer_tree_host_impl.cc
File cc/layer_tree_host_impl.cc (right):

https://codereview.chromium.org/11470024/diff/6001/cc/layer_tree_host_impl.cc...
cc/layer_tree_host_impl.cc:1636: const size_t roundingStep = 8 * 1024 * 1024;
nit: "static const ...", so you don't get another copy on the stack every time
you call this

https://codereview.chromium.org/11470024/diff/6001/cc/util.h
File cc/util.h (right):

https://codereview.chromium.org/11470024/diff/6001/cc/util.h#newcode1
cc/util.h:1: // Copyright 2011 The Chromium Authors. All rights reserved.
2012

[ccameron, 2012-12-08 00:21:15]

Thanks!!

https://codereview.chromium.org/11470024/diff/6001/cc/layer_tree_host_impl.cc
File cc/layer_tree_host_impl.cc (right):

https://codereview.chromium.org/11470024/diff/6001/cc/layer_tree_host_impl.cc...
cc/layer_tree_host_impl.cc:1636: const size_t roundingStep = 8 * 1024 * 1024;
On 2012/12/07 23:19:12, jamesr wrote:
> nit: "static const ...", so you don't get another copy on the stack every time
> you call this

Done.

https://codereview.chromium.org/11470024/diff/6001/cc/util.h
File cc/util.h (right):

https://codereview.chromium.org/11470024/diff/6001/cc/util.h#newcode1
cc/util.h:1: // Copyright 2011 The Chromium Authors. All rights reserved.
On 2012/12/07 23:19:12, jamesr wrote:
> 2012

Done.

[commit-bot: I haz the power, 2012-12-08 00:22:36]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/ccameron@chromium.org/11470024/10001

[commit-bot: I haz the power, 2012-12-08 06:24:29]

Change committed as 171949