Fix Tensor issues

include/backend.hpp

@@ -2,10 +2,11 @@
 
 #include <cstddef>
 
-#include "tensor.hpp"
-
 namespace CUDANet {
 
+// Forward declaration
+class Tensor;
+
 class Backend {
   public:
     // Memory management

include/tensor.hpp

@@ -34,15 +34,21 @@ public:
     size_t numel() const;
 
     template <typename T>
-    const T* data() const;
+    const T* data() const {
+        return static_cast<T*>(d_ptr);
+    }
 
     template <typename T>
-    T* data();
+    T* data() {
+        return static_cast<T*>(d_ptr);
+    }
 
     void zero();
 
     template <typename T>
-    void set_data(T *data);
+    void set_data(T *data) {
+        backend->copy_to_device(*this, data, total_size);
+    }
 
 private:
     Shape       shape;