Delegated renderer: Put textures on those quads.

A few changes are necessary to allow fetching textures provided by
the render processes through IPC and bound to their respective GL
context in the GPU process and use them in the QtQuick scene graph.

- Remove the plain color test textures.
- Allow setting the QtQuick QOpenGLContext's handle as the share
  context for all context set as shared in the Chromium GPU process.
  We do this by letting the GpuChannelManager ask us for a
  ShareGroup instance responsible for returning a sharing GL context
  handle.
- Fetch texture IDs from the MailboxManager used by the GPU process
  using the Mailbox given to us in the DelegatedFrameData.
  This is the same mechanism used by Chromium to share textures
  between "client" GL contexts.
- Keep the QtQuick scene graph threads and Chromium in-process GPU
  thread separate. The complicated part of merging those two
  rendering pipelines on the same thread is that it would force Qt
  to also use only one thread for rendering. For the moment we will
  try to synchronize those threads together instead.
- Lock the Qt SG thread while waiting for resource sync points.
  Do so by posting a callback to the Chromium GPU thread and wait
  until the sync point of every resource has been retired by the
  producing contexts.
- Acknowledge the delegated from once QtQuick swapped the GL buffers
  instead of right after we added the frame to the scene graph.
  This fixes some issues where the textures for the previous frame
  would already be released as Chromium was producing the new frame.

There are still a few issues regarding synchronization that have to
be fixed, especially when Qt triggers the rendering of a new frame
while Chromium is starting to produce the next frame.

Note: To enable it we still need to pass the following command switches:
--enable-delegated-renderer --enable-threaded-compositing --in-process-gpu

Change-Id: I2d4f7fac603b1808ed1495a8d689cb48e9ed41b9
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>

1 files changed, 84 insertions, 0 deletions

diff --git a/patches/0001-Add-accessors-for-the-Qt-delegated-renderer-integrat.patch b/patches/0001-Add-accessors-for-the-Qt-delegated-renderer-integrat.patch
new file mode 100644
index 000000000..fe63dfc0f
--- /dev/null
+++ b/patches/0001-Add-accessors-for-the-Qt-delegated-renderer-integrat.patch
@@ -0,0 +1,84 @@
+From e40d3482d46ff88755aa2bd76048de698ae03915 Mon Sep 17 00:00:00 2001
+From: Jocelyn Turcotte <jocelyn.turcotte@digia.com>
+Date: Fri, 20 Sep 2013 15:00:47 +0200
+Subject: [PATCH] Add accessors for the Qt delegated renderer integration.
+
+This is needed to fetch the MessageLoop, the MailboxManager and the
+SyncPointManager of the GPU in-process host.
+---
+ content/common/gpu/gpu_channel_manager.h | 2 ++
+ content/gpu/gpu_child_thread.cc          | 6 ++++++
+ content/gpu/gpu_child_thread.h           | 6 ++++++
+ 3 files changed, 14 insertions(+)
+
+diff --git a/content/common/gpu/gpu_channel_manager.h b/content/common/gpu/gpu_channel_manager.h
+index f01a1c5..e41a548 100644
+--- a/content/common/gpu/gpu_channel_manager.h
++++ b/content/common/gpu/gpu_channel_manager.h
+@@ -89,6 +89,8 @@ class GpuChannelManager : public IPC::Listener,
+ 
+   gpu::gles2::ProgramCache* program_cache();
+ 
++  gpu::gles2::MailboxManager* mailbox_manager() { return mailbox_manager_.get(); }
++
+   GpuMemoryManager* gpu_memory_manager() { return &gpu_memory_manager_; }
+ 
+   GpuChannel* LookupChannel(int32 client_id);
+diff --git a/content/gpu/gpu_child_thread.cc b/content/gpu/gpu_child_thread.cc
+index 170b364..a84a9e0 100644
+--- a/content/gpu/gpu_child_thread.cc
++++ b/content/gpu/gpu_child_thread.cc
+@@ -40,6 +40,8 @@ bool GpuProcessLogMessageHandler(int severity,
+ 
+ }  // namespace
+ 
++GpuChildThread* GpuChildThread::instance_ = 0;
++
+ GpuChildThread::GpuChildThread(GpuWatchdogThread* watchdog_thread,
+                                bool dead_on_arrival,
+                                const gpu::GPUInfo& gpu_info,
+@@ -53,6 +55,8 @@ GpuChildThread::GpuChildThread(GpuWatchdogThread* watchdog_thread,
+   target_services_ = NULL;
+ #endif
+   g_thread_safe_sender.Get() = thread_safe_sender();
++
++  instance_ = this;
+ }
+ 
+ GpuChildThread::GpuChildThread(const std::string& channel_id)
+@@ -71,6 +75,8 @@ GpuChildThread::GpuChildThread(const std::string& channel_id)
+     VLOG(1) << "gfx::GLSurface::InitializeOneOff()";
+   }
+   g_thread_safe_sender.Get() = thread_safe_sender();
++
++  instance_ = this;
+ }
+ 
+ GpuChildThread::~GpuChildThread() {
+diff --git a/content/gpu/gpu_child_thread.h b/content/gpu/gpu_child_thread.h
+index ff0c925..82bb32a 100644
+--- a/content/gpu/gpu_child_thread.h
++++ b/content/gpu/gpu_child_thread.h
+@@ -56,6 +56,10 @@ class GpuChildThread : public ChildThread {
+   virtual bool Send(IPC::Message* msg) OVERRIDE;
+   virtual bool OnControlMessageReceived(const IPC::Message& msg) OVERRIDE;
+ 
++  GpuChannelManager* ChannelManager() const { return gpu_channel_manager_.get(); }
++
++  static GpuChildThread* instance() { return instance_; }
++
+  private:
+   // Message handlers.
+   void OnInitialize();
+@@ -94,6 +98,8 @@ class GpuChildThread : public ChildThread {
+   // Whether the GPU thread is running in the browser process.
+   bool in_browser_process_;
+ 
++  static GpuChildThread* instance_;
++
+   DISALLOW_COPY_AND_ASSIGN(GpuChildThread);
+ };
+ 
+-- 
+1.8.4.2
+