Run VDA::Decode on GPU IO thread if VDA supports it.

content/common/gpu/media/gpu_video_decode_accelerator.cc

Index: content/common/gpu/media/gpu_video_decode_accelerator.cc
diff --git a/content/common/gpu/media/gpu_video_decode_accelerator.cc b/content/common/gpu/media/gpu_video_decode_accelerator.cc
index 2b0bb848b69fba9b2acf8d68839441932af28917..fced412477db43e16277b4e96a7867eeca6fdf81 100644
--- a/content/common/gpu/media/gpu_video_decode_accelerator.cc
+++ b/content/common/gpu/media/gpu_video_decode_accelerator.cc
@@ -9,6 +9,7 @@
 #include "base/bind.h"
 #include "base/command_line.h"
 #include "base/logging.h"
+#include "base/message_loop/message_loop_proxy.h"
 #include "base/stl_util.h"
 
 #include "content/common/gpu/gpu_channel.h"
@@ -54,6 +55,62 @@ static bool MakeDecoderContextCurrent(
   return true;
 }
 
+// A message filter to run VDA::Decode in IO thread to reduce decode latency.
+class GpuVideoDecodeAccelerator::GvdaMessageFilter
+    : public IPC::ChannelProxy::MessageFilter {
+ public:
+  GvdaMessageFilter(GpuVideoDecodeAccelerator* gvda,

[Ami GONE FROM CHROMIUM, 2013/08/26 17:20:03]

virtual dtor please

[Ami GONE FROM CHROMIUM, 2013/08/26 17:20:03]

s/gvda/owner/ here and in the member declaration at l.70 would make the comment
unnecessary. 

BUT, it is also misleading, because while GVDA "owns" the filter, it is actually
the filter's OnFilterRemoved method that deletes the GVDA.  The comment in
OnDestroy is good, but I think a further comment to this effect on (the newly
renamed) owner_ would also be good.

[wuchengli, 2013/08/27 12:32:24]

Done.

[wuchengli, 2013/08/27 12:32:24]

Done.

+                    int32 host_route_id,
+                    scoped_refptr<base::MessageLoopProxy> message_loop);

[Ami GONE FROM CHROMIUM, 2013/08/26 17:20:03]

"message_loop" is an uninformative name.

[wuchengli, 2013/08/27 12:32:24]

Changed to child_message_loop.

+  virtual void OnFilterRemoved() OVERRIDE;
+  virtual bool OnMessageReceived(const IPC::Message& msg) OVERRIDE;
+  void OnDecode(base::SharedMemoryHandle handle, int32 id, uint32 size);

[Ami GONE FROM CHROMIUM, 2013/08/26 17:20:03]

private?

[wuchengli, 2013/08/27 12:32:24]

Done.

+
+ private:
+  GpuVideoDecodeAccelerator* gvda_;  // GpuVideoDecodeAccelerator owns this.
+  int32 host_route_id_;
+  scoped_refptr<base::MessageLoopProxy> message_loop_;
+};
+
+GpuVideoDecodeAccelerator::GvdaMessageFilter::GvdaMessageFilter(
+    GpuVideoDecodeAccelerator* gvda,
+    int32 host_route_id,
+    scoped_refptr<base::MessageLoopProxy> message_loop)
+    : gvda_(gvda), host_route_id_(host_route_id), message_loop_(message_loop) {}
+
+void GpuVideoDecodeAccelerator::GvdaMessageFilter::OnFilterRemoved() {
+  message_loop_->DeleteSoon(FROM_HERE, gvda_);
+}
+
+bool GpuVideoDecodeAccelerator::GvdaMessageFilter::OnMessageReceived(
+    const IPC::Message& msg) {
+  if (msg.routing_id() != host_route_id_)
+    return false;
+
+  bool handled = true;
+  IPC_BEGIN_MESSAGE_MAP(GvdaMessageFilter, msg)
+    IPC_MESSAGE_HANDLER(AcceleratedVideoDecoderMsg_Decode, OnDecode)
+    IPC_MESSAGE_UNHANDLED(handled = false)
+  IPC_END_MESSAGE_MAP()
+  return handled;
+}
+
+void GpuVideoDecodeAccelerator::GvdaMessageFilter::OnDecode(
+    base::SharedMemoryHandle handle,
+    int32 id,
+    uint32 size) {
+  if (id < 0) {
+    DLOG(FATAL) << "BitstreamBuffer id " << id << " out of range";
+    message_loop_->PostTask(
+        FROM_HERE,
+        base::Bind(&GpuVideoDecodeAccelerator::NotifyError,
+                   base::Unretained(gvda_),
+                   media::VideoDecodeAccelerator::INVALID_ARGUMENT));
+    return;
+  }
+  gvda_->OnDecode(handle, id, size);
+}
+
 GpuVideoDecodeAccelerator::GpuVideoDecodeAccelerator(int32 host_route_id,
                                                      GpuCommandBufferStub* stub)
     : init_done_msg_(NULL),
@@ -63,17 +120,16 @@ GpuVideoDecodeAccelerator::GpuVideoDecodeAccelerator(int32 host_route_id,
   DCHECK(stub_);
   stub_->AddDestructionObserver(this);
   stub_->channel()->AddRoute(host_route_id_, this);
+  filter_.reset(new GvdaMessageFilter(
+      this, host_route_id, base::MessageLoopProxy::current()));
+  stub_->channel()->AddFilter(filter_.get());
   make_context_current_ =
       base::Bind(&MakeDecoderContextCurrent, stub_->AsWeakPtr());
 }
 
 GpuVideoDecodeAccelerator::~GpuVideoDecodeAccelerator() {
-  DCHECK(stub_);
   if (video_decode_accelerator_)
     video_decode_accelerator_.release()->Destroy();
-
-  stub_->channel()->RemoveRoute(host_route_id_);
-  stub_->RemoveDestructionObserver(this);
 }
 
 bool GpuVideoDecodeAccelerator::OnMessageReceived(const IPC::Message& msg) {
@@ -82,7 +138,6 @@ bool GpuVideoDecodeAccelerator::OnMessageReceived(const IPC::Message& msg) {
     return false;
   bool handled = true;
   IPC_BEGIN_MESSAGE_MAP(GpuVideoDecodeAccelerator, msg)
-    IPC_MESSAGE_HANDLER(AcceleratedVideoDecoderMsg_Decode, OnDecode)
     IPC_MESSAGE_HANDLER(AcceleratedVideoDecoderMsg_AssignPictureBuffers,
                         OnAssignPictureBuffers)
     IPC_MESSAGE_HANDLER(AcceleratedVideoDecoderMsg_ReusePictureBuffer,
@@ -149,7 +204,8 @@ void GpuVideoDecodeAccelerator::NotifyError(
 
 void GpuVideoDecodeAccelerator::Initialize(
     const media::VideoCodecProfile profile,
-    IPC::Message* init_done_msg) {
+    IPC::Message* init_done_msg,
+    const scoped_refptr<base::MessageLoopProxy>& io_message_loop) {
   DCHECK(stub_);
   DCHECK(!video_decode_accelerator_.get());
   DCHECK(!init_done_msg_);
@@ -179,7 +235,8 @@ void GpuVideoDecodeAccelerator::Initialize(
       gfx::GLSurfaceEGL::GetHardwareDisplay(),
       stub_->decoder()->GetGLContext()->GetHandle(),
       this,
-      make_context_current_));
+      make_context_current_,
+      io_message_loop));
 #elif defined(OS_CHROMEOS) && defined(ARCH_CPU_X86_FAMILY) && defined(USE_X11)
   gfx::GLContextGLX* glx_context =
       static_cast<gfx::GLContextGLX*>(stub_->decoder()->GetGLContext());
@@ -187,7 +244,7 @@ void GpuVideoDecodeAccelerator::Initialize(
       static_cast<GLXContext>(glx_context->GetHandle());
   video_decode_accelerator_.reset(new VaapiVideoDecodeAccelerator(
       glx_context->display(), glx_context_handle, this,
-      make_context_current_));
+      make_context_current_, io_message_loop));
 #elif defined(OS_ANDROID)
   video_decode_accelerator_.reset(new AndroidVideoDecodeAccelerator(
       this,
@@ -206,11 +263,6 @@ void GpuVideoDecodeAccelerator::Initialize(
 void GpuVideoDecodeAccelerator::OnDecode(
     base::SharedMemoryHandle handle, int32 id, uint32 size) {
   DCHECK(video_decode_accelerator_.get());
-  if (id < 0) {
-    DLOG(FATAL) << "BitstreamBuffer id " << id << " out of range";
-    NotifyError(media::VideoDecodeAccelerator::INVALID_ARGUMENT);
-    return;
-  }
   video_decode_accelerator_->Decode(media::BitstreamBuffer(id, handle, size));
 }
 
@@ -297,7 +349,12 @@ void GpuVideoDecodeAccelerator::OnReset() {
 
 void GpuVideoDecodeAccelerator::OnDestroy() {
   DCHECK(video_decode_accelerator_.get());
-  delete this;
+  DCHECK(stub_);
+  stub_->channel()->RemoveRoute(host_route_id_);
+  stub_->RemoveDestructionObserver(this);
+  // Remove the filter first because it can access the member variables on IO
+  // thread. When filter is removed, OnFilterRemoved will delete |this|.
+  stub_->channel()->RemoveFilter(filter_.get());
 }
 
 void GpuVideoDecodeAccelerator::NotifyEndOfBitstreamBuffer(
@@ -328,9 +385,7 @@ void GpuVideoDecodeAccelerator::NotifyResetDone() {
     DLOG(ERROR) << "Send(AcceleratedVideoDecoderHostMsg_ResetDone) failed";
 }
 
-void GpuVideoDecodeAccelerator::OnWillDestroyStub() {
-  delete this;
-}
+void GpuVideoDecodeAccelerator::OnWillDestroyStub() { OnDestroy(); }
 
 bool GpuVideoDecodeAccelerator::Send(IPC::Message* message) {
   DCHECK(stub_);