AllocatorAttributes device_alloc_attr;
    device_alloc_attr.set_on_host(false);
    device_allocator_ = device_->GetAllocator(device_alloc_attr);
  }

 protected:
  std::unique_ptr<Device> device_;
  Allocator* host_allocator_;
  Allocator* device_allocator_;
};

TEST_F(XlaHostSendRecvDeviceContextTest, CopyDeviceTensorToCPU) {
  SetDevice("GPU");

//
// Example usage:
//
//  Device device;
//  stream_executor::Stream stream(executor);
//  Tensor cpu_tensor(host_allocator, DT_FLOAT, TensorShape({2, 2}));
//  Tensor device_tensor(device_allocator, DT_FLOAT, TensorShape({2, 2}));
//  se::DeviceMemoryBase gpu_dst{device_tensor.data(), 4 * sizeof(float)};
//  xla::Shape shape(xla::F32, {2, 2}, {}, {})
//  tsl::AsyncValueRef<std::unique_ptr<se::Event>> done_event =

// Concrete once transfer is completed.
//
// Example usage:
//
//  Device device;
//  stream_executor::Stream stream(executor);
//  Tensor device_tensor(device_allocator, DT_FLOAT, TensorShape({2, 2}));
//  se::DeviceMemoryBase gpu_dst{device_tensor.data(), 4 * sizeof(float)};
//  xla::Shape shape(xla::F32, {2, 2}, {}, {})
//  tsl::AsyncValueRef<std::unique_ptr<se::Event>> done_event =