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;

src/layers/dense.cpp

@@ -10,10 +10,10 @@ Dense::Dense(CUDANet::Backend* backend, CUDANet::Shape in, CUDANet::Shape out)
       in_shape(in),
       out_shape(out),
       weights(
-          CUDANet::Tensor{{in[0] * out[0]}, CUDANet::DType::FLOAT32, backend}
+          CUDANet::Tensor(Shape{in[0] * out[0]}, CUDANet::DType::FLOAT32, backend)
       ),
-      biases(CUDANet::Tensor({out[0]}, CUDANet::DType::FLOAT32, backend)),
-      output(CUDANet::Tensor({out[0]}, CUDANet::DType::FLOAT32, backend)) {
+      biases(CUDANet::Tensor(Shape{out[0]}, CUDANet::DType::FLOAT32, backend)),
+      output(CUDANet::Tensor(Shape{out[0]}, CUDANet::DType::FLOAT32, backend)) {
     // Allocate memory for weights and biases
 
     if (in.size() != 1) {
@@ -35,6 +35,8 @@ Dense::Dense(CUDANet::Backend* backend, CUDANet::Shape in, CUDANet::Shape out)
     biases.zero();
 }
 
+Dense::~Dense() {}
+
 CUDANet::Tensor& Dense::forward(const CUDANet::Tensor& input) {
     backend->dense(weights, biases, input, output, in_shape[0], out_shape[0]);
     return output;

src/tensor.cpp

@@ -68,8 +68,10 @@ Tensor& Tensor::operator=(Tensor&& other) noexcept {
 }
 
 Tensor::~Tensor() {
-    backend->deallocate(d_ptr);
-    d_ptr = nullptr;
+    if (backend && d_ptr) {
+        backend->deallocate(d_ptr);
+        d_ptr = nullptr;
+    }
 }
 
 size_t Tensor::numel() const {
@@ -80,21 +82,6 @@ size_t Tensor::size() const {
     return total_size;
 }
 
-template <typename T>
-const T* Tensor::data() const {
-    return static_cast<T*>(d_ptr);
-}
-
-template <typename T>
-T* Tensor::data() {
-    return static_cast<T*>(d_ptr);
-}
-
 void Tensor::zero() {
     backend->zero(*this);
 }
-
-template <typename T>
-void Tensor::set_data(T *data) {
-    backend->copy_to_device(*this, data, total_size);
-}