Apply gpu workarounds.

Apply gpu workarounds.

Fixes surfaces composition on some qualcomm devices.

Fixes domokit/sky_engine#117.

R=jamesr@chromium.org

Committed: https://chromium.googlesource.com/external/mojo/+/c45263ef3e753767a277459a789abc214c9f5a54

[cdotstout, 2015-09-18 16:21:27]

cstout@chromium.org changed reviewers:
+ jamesr@chromium.org

[cdotstout, 2015-09-18 16:21:27]

James wdyt... too easy? :)

[jamesr, 2015-09-18 16:34:09]

https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
File services/gles2/gpu_state.cc (right):

https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
services/gles2/gpu_state.cc:19:
gpu::ApplyGpuDriverBugWorkarounds(base::CommandLine::ForCurrentProcess());
what populates the workaround list? the way this works in chrome is the browser
proc generates a bunch of stuff into the command line for the gpu process when
then gets populated here. maybe the command line parameter isn't actually used?

[cdotstout, 2015-09-18 17:54:04]

https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
File services/gles2/gpu_state.cc (right):

https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
services/gles2/gpu_state.cc:19:
gpu::ApplyGpuDriverBugWorkarounds(base::CommandLine::ForCurrentProcess());
On 2015/09/18 16:34:09, jamesr wrote:
> what populates the workaround list? the way this works in chrome is the
browser
> proc generates a bunch of stuff into the command line for the gpu process when
> then gets populated here. maybe the command line parameter isn't actually
used?

Some workarounds are set automatically based on the gpu, a good example being
the adreno 420 fix for framebuffer deletion.

The commandline provides additional workarounds (see
GpuDriverBugList::AppendWorkaroundsFromCommandLine).  In this case, the
commandline happens to be empty.  I'm not sure we want to actually use a command
line for this, maybe it would be better to just construct an empty command line
and pass that in.

[jamesr, 2015-09-18 17:54:57]

On 2015/09/18 at 17:54:04, cstout wrote:
>
https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
> File services/gles2/gpu_state.cc (right):
> 
>
https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
> services/gles2/gpu_state.cc:19:
gpu::ApplyGpuDriverBugWorkarounds(base::CommandLine::ForCurrentProcess());
> On 2015/09/18 16:34:09, jamesr wrote:
> > what populates the workaround list? the way this works in chrome is the
browser
> > proc generates a bunch of stuff into the command line for the gpu process
when
> > then gets populated here. maybe the command line parameter isn't actually
used?
> 
> Some workarounds are set automatically based on the gpu, a good example being
the adreno 420 fix for framebuffer deletion.
> 
> The commandline provides additional workarounds (see
GpuDriverBugList::AppendWorkaroundsFromCommandLine).  In this case, the
commandline happens to be empty.  I'm not sure we want to actually use a command
line for this, maybe it would be better to just construct an empty command line
and pass that in.

Yeah, let's do that (accessing the static is thread-dodgy).

lgtm

[cdotstout, 2015-09-21 20:00:39]

On 2015/09/18 17:54:57, jamesr wrote:
> On 2015/09/18 at 17:54:04, cstout wrote:
> >
>
https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
> > File services/gles2/gpu_state.cc (right):
> > 
> >
>
https://codereview.chromium.org/1350723004/diff/20001/services/gles2/gpu_stat...
> > services/gles2/gpu_state.cc:19:
> gpu::ApplyGpuDriverBugWorkarounds(base::CommandLine::ForCurrentProcess());
> > On 2015/09/18 16:34:09, jamesr wrote:
> > > what populates the workaround list? the way this works in chrome is the
> browser
> > > proc generates a bunch of stuff into the command line for the gpu process
> when
> > > then gets populated here. maybe the command line parameter isn't actually
> used?
> > 
> > Some workarounds are set automatically based on the gpu, a good example
being
> the adreno 420 fix for framebuffer deletion.
> > 
> > The commandline provides additional workarounds (see
> GpuDriverBugList::AppendWorkaroundsFromCommandLine).  In this case, the
> commandline happens to be empty.  I'm not sure we want to actually use a
command
> line for this, maybe it would be better to just construct an empty command
line
> and pass that in.
> 
> Yeah, let's do that (accessing the static is thread-dodgy).
> 
> lgtm

The gpu command buffer is actually getting the gpu workarounds from the command
line of the current process each time it creates a FeatureInfo object.

So I moved the workaround init to where the gpu process command line is
initialized.

[jamesr, 2015-09-21 20:14:51]

OK - it's a bit annoying to duplicate across shell impls but there's not much
you can do about that.

still lgtm

[cdotstout, 2015-09-21 20:41:29]

Committed patchset #4 (id:60001) manually as
c45263ef3e753767a277459a789abc214c9f5a54 (presubmit successful).