Implement GpuArcVideoService for arc video accelerator

content/common/gpu/client/gpu_arc_accelerator_host.h

+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef CONTENT_COMMON_GPU_CLIENT_GPU_ARC_ACCELERATOR_HOST_H_
+#define CONTENT_COMMON_GPU_CLIENT_GPU_ARC_ACCELERATOR_HOST_H_
+
+#include <queue>
+
+#include "base/callback.h"
+#include "base/memory/weak_ptr.h"
+#include "base/threading/non_thread_safe.h"
+#include "ipc/ipc_channel_handle.h"
+#include "ipc/ipc_listener.h"
+
+namespace content {
+class GpuChannelHost;
+
+// kcwu: Should this class be folded into GpuProcessHost?

[Owen Lin, 2015/11/26 03:54:41]

I changed my mind. Let's keep this class. But how about changing the name? The
name is kind of misleading. It is not an accelerator at all.

Let's make it a GpuArcVideoService which is a service lived in GPU process to
server video requests from Arc.

[kcwu, 2015/11/26 10:34:42]

Done.
Nice idea. This name is better.

+class GpuArcAcceleratorHost
+    : public IPC::Listener,
+      public base::NonThreadSafe,
+      public base::SupportsWeakPtr<GpuArcAcceleratorHost> {
+ public:
+  using CreatedCallback = base::Callback<void(IPC::ChannelHandle, uint32_t)>;
+  GpuArcAcceleratorHost();
+  ~GpuArcAcceleratorHost() override;
+
+  // IPC::Listener implementation.
+  void OnChannelError() override;
+  bool OnMessageReceived(const IPC::Message& message) override;
+
+  // Creates one arc accelerator with |device_type|. Created channel and
+  // intialization result are passed back by |cb|.
+  void CreateArcAccelerator(uint32_t device_type, const CreatedCallback& cb);

[Owen Lin, 2015/11/26 03:54:40]

I would prefer keep it cleaner for know. If we do have performance issue we can
measure the improvement and justified the change. 

[kcwu, 2015/11/26 10:34:42]

Done.

+
+  // Shutdowns all existing arc accelerators and abort all pending requests.
+  void Shutdown();
+
+ private:
+  void AbortPendingRequests();
+  void GpuChannelEstablished(uint32_t device_type, const CreatedCallback& cb);
+  void CreateArcAcceleratorWithGpuChannel(uint32_t device_type,
+                                          const CreatedCallback& cb);
+
+  void OnArcAcceleratorCreated(IPC::ChannelHandle handle, uint32_t result);
+
+  int32_t route_id_;
+  scoped_refptr<GpuChannelHost> gpu_channel_host_;

[Owen Lin, 2015/11/26 03:54:41]

As we discussed, Can we just use GpuProcessHost? What's the benefit of using
GpuChannel?

[kcwu, 2015/11/26 10:34:42]

GpuChannelHost and BrowserGpuChannelHostFactory do some life-cycle, threading,
and state management.
For example, all major methods of GpuProcessHost need to be invoked on IO thread
while we are on UI thread. And we have to take more care about gpu crash and
relaunch, etc.

I estimate using GpuProcessHost will need slightly more code. I will try and
compare later.

+  std::queue<CreatedCallback> pending_requests_;
+
+  DISALLOW_COPY_AND_ASSIGN(GpuArcAcceleratorHost);
+};
+
+}  // namespace content
+
+#endif  // CONTENT_COMMON_GPU_CLIENT_GPU_ARC_ACCELERATOR_HOST_H_