Add flag to use virtual contexts to WebGraphicsContext3D::Attributes

Add flag to use virtual contexts to WebGraphicsContext3D::Attributes.

The Mac will specify this flag for its compositor contexts.

BUG=180463

[ccameron, 2013-04-04 20:27:07]

This is the WebKit half of

[ccameron, 2013-04-04 20:28:41]

(mouse had a spasm)

This is the WebKit half of
https://codereview.chromium.org/13647014/
which is the mechanism by which I plan to make compositors use virtualization
(in particular, in RenderWidget::CreateOutputSurface -- that code isn't there
yet cause it's regression-prone).

[jamesr, 2013-04-04 20:29:23]

Why is this a context attribute?

[jamesr, 2013-04-04 20:29:49]

Compositor contexts are already special (they're the only onscreen contexts),
can't we just set those to use virtual contexts on mac?

[ccameron, 2013-04-04 20:31:09]

On 2013/04/04 20:29:23, jamesr wrote:
> Why is this a context attribute?

Some contexts (compositor) will set this, while others (WebGL, canvas, etc)
won't.

Lmk if there is a better way to do accomplish this.

[jamesr, 2013-04-04 20:32:53]

On 2013/04/04 20:31:09, ccameron1 wrote:
> On 2013/04/04 20:29:23, jamesr wrote:
> > Why is this a context attribute?
> 
> Some contexts (compositor) will set this, while others (WebGL, canvas, etc)
> won't.
> 
> Lmk if there is a better way to do accomplish this.

Yes.  Just set this for compositor (onscreen) contexts only.

[ccameron, 2013-04-04 20:35:34]

On 2013/04/04 20:32:53, jamesr wrote:
> On 2013/04/04 20:31:09, ccameron1 wrote:
> > On 2013/04/04 20:29:23, jamesr wrote:
> > > Why is this a context attribute?
> > 
> > Some contexts (compositor) will set this, while others (WebGL, canvas, etc)
> > won't.
> > 
> > Lmk if there is a better way to do accomplish this.
> 
> Yes.  Just set this for compositor (onscreen) contexts only.

(ack, we're hitting send at the same time)

Sure -- in  GpuCommandBufferStub::OnInitialize, I can just do  "if this stub has
a surface_id, and this is mac, then virtualize" -- it feels less intentional
(and more like relying on coincidences), but if that's preferable, then sure!!

[jamesr, 2013-04-04 20:36:18]

I'm sure you can figure out a way to do it cleanly.

[ccameron, 2013-04-05 00:05:06]

On 2013/04/04 20:36:18, jamesr wrote:
> I'm sure you can figure out a way to do it cleanly.

Looking at this some more, how is this different from
WebGraphicsContext3D::Attributes::preferDiscreteGPU? Shouldn't both of those be
set in the same place?

[jamesr, 2013-04-05 00:05:54]

On 2013/04/05 00:05:06, ccameron1 wrote:
> On 2013/04/04 20:36:18, jamesr wrote:
> > I'm sure you can figure out a way to do it cleanly.
> 
> Looking at this some more, how is this different from
> WebGraphicsContext3D::Attributes::preferDiscreteGPU? Shouldn't both of those
be
> set in the same place?

Probably - that one's pretty unusual.

[Ken Russell (switch to Gerrit), 2013-04-05 00:40:47]

On 2013/04/05 00:05:54, jamesr wrote:
> On 2013/04/05 00:05:06, ccameron1 wrote:
> > On 2013/04/04 20:36:18, jamesr wrote:
> > > I'm sure you can figure out a way to do it cleanly.
> > 
> > Looking at this some more, how is this different from
> > WebGraphicsContext3D::Attributes::preferDiscreteGPU? Shouldn't both of those
> be
> > set in the same place?
> 
> Probably - that one's pretty unusual.

I don't know any more where the compositor's context is created. Could one of
you point me to the code?

This patch and the other one look fine to me. I'd prefer the creation mechanism
to be flexible, rather than having some magic hacked in somewhere. WIth the
plumbing in the other patch, it other parts of the system (Pepper, WebGL) could
individually opt in to context virtualization if or when it would be beneficial.

[ccameron, 2013-04-05 01:04:13]

On 2013/04/05 00:40:47, kbr wrote:
> On 2013/04/05 00:05:54, jamesr wrote:
> > On 2013/04/05 00:05:06, ccameron1 wrote:
> > > On 2013/04/04 20:36:18, jamesr wrote:
> > > > I'm sure you can figure out a way to do it cleanly.
> > > 
> > > Looking at this some more, how is this different from
> > > WebGraphicsContext3D::Attributes::preferDiscreteGPU? Shouldn't both of
those
> > be
> > > set in the same place?
> > 
> > Probably - that one's pretty unusual.
> 
> I don't know any more where the compositor's context is created. Could one of
> you point me to the code?

It's in GpuCommandBufferStub::OnInitialize (that's also where virtualization
hooks in)

I'll put in the other plumbing patch first, then we can return to this part.

Also of note is that I need to update virtualization to have different
virtualized contexts for iGPU and dGPU.