Implement fallback from WMPI to WMPA.

content/browser/gpu/gpu_data_manager_impl_private.cc

Index: content/browser/gpu/gpu_data_manager_impl_private.cc
diff --git a/content/browser/gpu/gpu_data_manager_impl_private.cc b/content/browser/gpu/gpu_data_manager_impl_private.cc
index 926886d3e8baa457f780b054064c5bb8e481d2a6..577581ba8c42e08d40ff030991922609e1360bb7 100644
--- a/content/browser/gpu/gpu_data_manager_impl_private.cc
+++ b/content/browser/gpu/gpu_data_manager_impl_private.cc
@@ -933,6 +933,19 @@ bool GpuDataManagerImplPrivate::ShouldDisableAcceleratedVideoDecode(
     return true;
   if (group_name == "Disabled")
     return true;
+
+  // Accelerated decode is not available at all with --disable-gpu. It may be
+  // possible to eventually support --single-process and/or --in-process-gpu,
+  // but currently it fails long before trying because the supported profiles
+  // list isn't initialized. Setting the disable flag in these case is important
+  // for Android, as there is no fallback path if the renderer process chooses
+  // wrong.
+  if (command_line->HasSwitch(switches::kDisableGpu) ||

[DaleCurtis, 2016/01/06 21:03:22]

This is going to mean no webview support since they use --single-process AFAIK.
--in-process-gpu is used by Android One variants I think. Others? I don't think
these are reasonable restrictions, so if you want to land this now these need to
be marked with a TODO() fix and have a bug attached.

Why is this necessary at all? Won't we just fallback if possible during
initialize since the factories would be null in the gpu decoding case.

[sandersd (OOO until July 31), 2016/01/06 22:20:38]

I agree, but I don't think this is going to get fixed soon. TODO added.
No, by that time we already picked WMPI and it's too late.

+      command_line->HasSwitch(switches::kSingleProcess) ||
+      command_line->HasSwitch(switches::kInProcessGPU)) {
+    return true;
+  }
+
   return false;
 }