Migrate Dense layer

include/cudanet.cuh

@@ -8,15 +8,15 @@
 #include "pooling.cuh"
 
 // Layers
-#include "activation.cuh"
+#include "activation.hpp"
 #include "add.cuh"
 #include "avg_pooling.cuh"
 #include "batch_norm.cuh"
 #include "concat.cuh"
 #include "conv2d.cuh"
-#include "dense.cuh"
+#include "dense.hpp"
 #include "input.cuh"
-#include "layer.cuh"
+#include "layer.hpp"
 #include "max_pooling.cuh"
 #include "output.cuh"
 

include/kernels/convolution.cuh

@@ -2,7 +2,7 @@
 #define CUDANET_CONVOLUTION_H
 
 #include <cuda_runtime.h>
-#include "layer.cuh"
+#include "layer.hpp"
 
 namespace CUDANet::Kernels {
 

include/kernels/pooling.cuh

@@ -2,7 +2,7 @@
 #define CUDANET_POOLING_H
 
 #include <cuda_runtime.h>
-#include "layer.cuh"
+#include "layer.hpp"
 
 namespace CUDANet::Kernels {
 

include/layers/activation.hpp

@@ -48,6 +48,8 @@ class Activation {
     ActivationType activationType;
     int length;
 
+    void activateCPU(float* input);
+
 #ifdef USE_CUDA
     int gridSize;
 
@@ -58,10 +60,7 @@ class Activation {
 
     void initCUDA();
     void delCUDA();
-#else
-    void activateCPU(float* input);
 #endif
-
 };
 
 

include/layers/avg_pooling.cuh

@@ -1,8 +1,8 @@
 #ifndef CUDANET_AVG_POOLING_H
 #define CUDANET_AVG_POOLING_H
 
-#include "activation.cuh"
-#include "layer.cuh"
+#include "activation.hpp"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 

include/layers/batch_norm.cuh

@@ -3,8 +3,8 @@
 
 #include <vector>
 
-#include "activation.cuh"
-#include "layer.cuh"
+#include "activation.hpp"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 

include/layers/conv2d.cuh

@@ -3,9 +3,9 @@
 
 #include <vector>
 
-#include "activation.cuh"
+#include "activation.hpp"
 #include "convolution.cuh"
-#include "layer.cuh"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 

include/layers/dense.hpp

@@ -3,8 +3,8 @@
 
 #include <vector>
 
-#include "activation.cuh"
-#include "layer.cuh"
+#include "activation.hpp"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 
@@ -84,20 +84,11 @@ class Dense : public WeightedLayer {
     int inputSize;
     int outputSize;
 
-    float* d_output;
-
-    float* d_weights;
-    float* d_biases;
-
     std::vector<float> weights;
     std::vector<float> biases;
 
     Layers::Activation* activation;
 
-    // Precompute kernel launch parameters
-    int forwardGridSize;
-    int biasGridSize;
-
     /**
      * @brief Initialize the weights to zeros
      *
@@ -110,11 +101,30 @@ class Dense : public WeightedLayer {
      */
     void initializeBiases();
 
+    float* forwardCPU(const float* input);
+
+#ifdef USE_CUDA
+    float* d_output;
+
+    float* d_weights;
+    float* d_biases;
+
+    // Precompute kernel launch parameters
+    int forwardGridSize;
+    int biasGridSize;
+
     /**
      * @brief Copy the weights and biases to the device
      *
      */
     void toCuda();
+
+    void initCUDA();
+    void delCUDA();
+
+    float* forwardCUDA(const float* d_input);
+#endif
+
 };
 
 }  // namespace CUDANet::Layers

include/layers/input.cuh

@@ -1,7 +1,7 @@
 #ifndef CUDANET_INPUT_LAYER_H
 #define CUDANET_INPUT_LAYER_H
 
-#include "layer.cuh"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 

include/layers/layer.hpp

@@ -111,10 +111,12 @@ class WeightedLayer : public SequentialLayer {
      */
     virtual void initializeBiases() = 0;
 
+#ifdef USE_CUDA
     /**
      * @brief Copy the weights and biases to the device
      */
     virtual void toCuda() = 0;
+#endif
 };
 
 }  // namespace CUDANet::Layers

include/layers/max_pooling.cuh

@@ -1,8 +1,8 @@
 #ifndef CUDANET_MAX_POOLING_H
 #define CUDANET_MAX_POOLING_H
 
-#include "activation.cuh"
-#include "layer.cuh"
+#include "activation.hpp"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 

include/layers/output.cuh

@@ -1,7 +1,7 @@
 #ifndef CUDANET_OUTPUT_LAYER_H
 #define CUDANET_OUTPUT_LAYER_H
 
-#include "layer.cuh"
+#include "layer.hpp"
 
 namespace CUDANet::Layers {
 

include/model/model.hpp

@@ -6,7 +6,7 @@
 #include <vector>
 
 #include "input.cuh"
-#include "layer.cuh"
+#include "layer.hpp"
 #include "module.hpp"
 #include "output.cuh"
 

include/model/module.hpp

@@ -5,7 +5,7 @@
 #include <unordered_map>
 #include <vector>
 
-#include "layer.cuh"
+#include "layer.hpp"
 
 namespace CUDANet {