gpu: Only estimate parent compositor run time when needed

content/renderer/gpu/compositor_output_surface.cc

Index: content/renderer/gpu/compositor_output_surface.cc
diff --git a/content/renderer/gpu/compositor_output_surface.cc b/content/renderer/gpu/compositor_output_surface.cc
index a2929643ac17e9a467fdb1d7d172979db087d510..2a566e6852671bb4c38659df4d5f3047256b6389 100644
--- a/content/renderer/gpu/compositor_output_surface.cc
+++ b/content/renderer/gpu/compositor_output_surface.cc
@@ -76,6 +76,9 @@ CompositorOutputSurface::CompositorOutputSurface(
   DCHECK(message_sender_.get());
   if (OutputSurface::software_device())
     capabilities_.max_frames_pending = 1;
+#if defined(OS_ANDROID)

[piman, 2014/06/13 20:46:43]

Why android only? All platforms but Mac* use delegated.
Can we make this common between platforms?

*Mac may go delegated for 37.

[Sami, 2014/06/16 13:15:04]

Yeah, I was wondering if this should be done more generally. Is this really a
question of delegated vs. non-delegated, though? On Android the browser
compositor runtime is subtracted from the deadline, so we need this to avoid
doing the same adjustment twice. I'm not sure if that happens on other platforms
yet, but I'm thinking it should since the browser is in the best place to do
that adjustment.

+  capabilities_.adjust_deadline_for_parent = false;
+#endif
 }
 
 CompositorOutputSurface::~CompositorOutputSurface() {