Fix some compilation errors

include/backend.hpp

@@ -2,6 +2,8 @@
 
 #include <cstddef>
 
+#include "shape.hpp"
+
 namespace CUDANet {
 
 // Forward declaration

include/backend/cpu.hpp

@@ -3,7 +3,7 @@
 #include "backend.hpp"
 #include "tensor.hpp"
 
-namespace CUDANet::Backend {
+namespace CUDANet::Backends {
 
 class CPU : public Backend {
 public:

include/backend/cuda.cuh

@@ -24,7 +24,7 @@ do { \
     } \
 } while (0)
 
-namespace CUDANet::Backend {
+namespace CUDANet::Backends {
 
 class CUDA : public Backend {
   public:

include/kernels/activation_functions.cuh

@@ -1,5 +1,4 @@
-#ifndef CUDANET_ACTIVATION_FUNCTIONS_H
-#define CUDANET_ACTIVATION_FUNCTIONS_H
+#pragma once
 
 #include <cuda_runtime.h>
 
@@ -32,5 +31,3 @@ __global__ void relu(
 );
 
 }  // namespace CUDANet::Kernels
-
-#endif  // CUDANET_ACTIVATION_FUNCTIONS_H

include/kernels/matmul.cuh

@@ -1,5 +1,4 @@
-#ifndef CUDANET_MATMUL_H
-#define CUDANET_MATMUL_H
+#pragma once
 
 #include <cuda_runtime.h>
 
@@ -191,5 +190,3 @@ __global__ void sum_reduce(
 );
 
 }  // namespace CUDANet::Kernels
-
-#endif  // CUDANET_MATMUL_H

include/layers/activation.hpp

@@ -25,7 +25,7 @@ class Activation : public Layer {
 
     Activation() = default;
 
-    Activation(CUDANet::Backend* backend, ActivationType activation, const CUDANet::Shape &shape);
+    Activation(ActivationType activation, const CUDANet::Shape &shape, CUDANet::Backend* backend);
 
     ~Activation() = default;
 

include/layers/conv2d.hpp

@@ -18,7 +18,7 @@ class Conv2d : public Layer {
         CUDANet::Backend* backend
     );
 
-    ~Conv2d() {};
+    ~Conv2d();
 
     CUDANet::Tensor& forward(CUDANet::Tensor& input) override;
 

include/utils/imagenet.hpp

@@ -1,5 +1,4 @@
-#ifndef CUDANET_IMAGENET_H
-#define CUDANET_IMAGENET_H
+#pragma once
 
 #include <map>
 #include <string>
@@ -1012,5 +1011,3 @@ const std::map <int, std::string> IMAGENET_CLASS_MAP = {
 // clang-format on
 
 }
-
-#endif // CUDANET_IMAGENET_H

src/backends/cuda/cuda_backend.cu

@@ -25,7 +25,7 @@ cudaDeviceProp initializeCUDA() {
     return deviceProp;
 }
 
-using namespace CUDANet::Backend;
+using namespace CUDANet::Backends;
 
 void* CUDA::allocate(size_t bytes) {
     void* d_ptr = nullptr;

src/backends/cuda/kernels/activation_functions.cu

@@ -1,5 +1,4 @@
 #include "activation_functions.cuh"
-#include "cuda_helper.cuh"
 
 using namespace CUDANet;
 

src/backends/cuda/kernels/matmul.cu

@@ -1,4 +1,4 @@
-#include "cuda_helper.cuh"
+#include "backend/cuda.cuh"
 #include "matmul.cuh"
 
 using namespace CUDANet;

src/backends/cuda/layer_ops.cu

@@ -4,7 +4,7 @@
 #include "kernels/matmul.cuh"
 #include "kernels/pool.cuh"
 
-using namespace CUDANet::Backend;
+using namespace CUDANet::Backends;
 
 void CUDA::relu(Tensor& tensor) {
     int gridSize = (tensor.numel() + BLOCK_SIZE - 1) / BLOCK_SIZE;

src/backends/cuda/tensor_ops.cu

@@ -4,7 +4,7 @@
 #include "backend/cuda.cuh"
 #include "kernels/matmul.cuh"
 
-using namespace CUDANet::Backend;
+using namespace CUDANet::Backends;
 
 void CUDA::print(const CUDANet::Tensor &input) {
     auto length = input.numel();

src/layers/activation.cpp

@@ -7,11 +7,11 @@
 
 using namespace CUDANet::Layers;
 
-Activation::Activation(CUDANet::Backend* backend, ActivationType activation, const CUDANet::Shape &shape)
+Activation::Activation(ActivationType activation, const CUDANet::Shape &shape, CUDANet::Backend* backend)
     : backend(backend), activationType(activation), shape(shape) {
 
     if (shape.size() != 1) {
-        throw std::runtime_error(std::format("Invalid shape. Expected [1], got {}", shape));
+        throw InvalidShapeException("input", 1, shape.size());
     }
 
     auto length = shape[0];