Honor --gpu-switching=force_discrete to always stay on discrete GPU.

Honor --gpu-switching=force_discrete to always stay on discrete GPU.

Before we only stays on discrete GPU if any GL contexts are active.

Now we truly stay on discrete GPU, always.

This CL also remove --disable-gpu-switching;
use --gpu-switching=force_integrated/force_discrete instead.

Also, user's --gpu-switching option will overwrite the hardwired
force_discrete decision, i.e., if a user wants to stay on integrated
GPU and save battery, we honor that.

BUG=140114, 131276, 111720
TEST=tree
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=158128

[Zhenyao Mo, 2012-09-21 22:39:57]

kbr: please review.

[Zhenyao Mo, 2012-09-21 22:41:13]

Note that the ForceUseOfDiscreteGPU() was deleted from the previous CL.  This CL
simply get it back without modifying it.

[Zhenyao Mo, 2012-09-21 22:41:43]

jbauman: can you also take a look?

[Zhenyao Mo, 2012-09-21 22:56:04]

FYI: the switches are tested, both force_integrated and force_discrete work as
expected.

[Ken Russell (switch to Gerrit), 2012-09-21 23:00:29]

LGTM, but please think about the comment below.

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_cgl.cc
File ui/gl/gl_context_cgl.cc (right):

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_cgl.cc#new...
ui/gl/gl_context_cgl.cc:229: if (format)
Indentation is off.

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_mac.mm
File ui/gl/gl_context_mac.mm (right):

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_mac.mm#new...
ui/gl/gl_context_mac.mm:156: (model_major <
kMacBookProFirstDualAMDIntelGPUModel)) &&
It would be really nice -- though probably difficult -- to express this logic in
terms of entries in the blacklist.

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gpu_switching_manager.h
File ui/gl/gpu_switching_manager.h (right):

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gpu_switching_manager...
ui/gl/gpu_switching_manager.h:31: //   Otherwise, always return
PreferIntegratedGpu.
I understand why you're returning PreferIntegratedGpu all the time in this case
-- to avoid redundant allocations of pixel formats forcing the system onto the
discrete GPU -- but I think that behavior will be confusing when debugging
things later. I think you should return gpu_switching_option_[0] despite any
inefficiency.

[jbauman, 2012-09-21 23:02:41]

Are you forcing into onto the discrete GPU as soon as the GPU process starts, or
only after a context is created?

[Zhenyao Mo, 2012-09-21 23:08:00]

On 2012/09/21 23:02:41, jbauman wrote:
> Are you forcing into onto the discrete GPU as soon as the GPU process starts,
or
> only after a context is created?

After a context is created.  Will it make a difference?

[Zhenyao Mo, 2012-09-21 23:08:08]

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_cgl.cc
File ui/gl/gl_context_cgl.cc (right):

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_cgl.cc#new...
ui/gl/gl_context_cgl.cc:229: if (format)
On 2012/09/21 23:00:29, kbr wrote:
> Indentation is off.

Done.

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_mac.mm
File ui/gl/gl_context_mac.mm (right):

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gl_context_mac.mm#new...
ui/gl/gl_context_mac.mm:156: (model_major <
kMacBookProFirstDualAMDIntelGPUModel)) &&
On 2012/09/21 23:00:29, kbr wrote:
> It would be really nice -- though probably difficult -- to express this logic
in
> terms of entries in the blacklist.

Yes, I am thinking of supporting Mac Models in blacklist.  Will put it on my
TODO list.

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gpu_switching_manager.h
File ui/gl/gpu_switching_manager.h (right):

http://codereview.chromium.org/10959050/diff/9001/ui/gl/gpu_switching_manager...
ui/gl/gpu_switching_manager.h:31: //   Otherwise, always return
PreferIntegratedGpu.
On 2012/09/21 23:00:29, kbr wrote:
> I understand why you're returning PreferIntegratedGpu all the time in this
case
> -- to avoid redundant allocations of pixel formats forcing the system onto the
> discrete GPU -- but I think that behavior will be confusing when debugging
> things later. I think you should return gpu_switching_option_[0] despite any
> inefficiency.

Done.

[jbauman, 2012-09-21 23:52:26]

lgtm