Implement to_cuda function

include/layers/dense.h

@@ -5,15 +5,17 @@
 
 #include <vector>
 #include <cublas_v2.h>
+#include <ilayer.h>
 
 namespace Layers {
 
-    class Dense {
+    class Dense : public ILayer {
     public:
         Dense(int inputSize, int outputSize, cublasHandle_t cublasHandle);
         ~Dense();
 
         void forward(const float* input, float* output);
+        void to_cuda();
 
     private:
         int inputSize;

include/layers/ilayer.h

@@ -0,0 +1,19 @@
+
+#ifndef I_LAYER_H
+#define I_LAYER_H
+
+#include <cublas_v2.h>
+
+namespace Layers {
+
+    class ILayer {
+    public:
+        virtual ~ILayer() {}
+
+        virtual void forward(const float* input, float* output) = 0;
+        virtual void to_cuda() = 0;
+    };
+
+} // namespace Layers
+
+#endif // I_LAYERH

include/utils/cuda_helper.h

@@ -3,6 +3,8 @@
 
 #include <cuda_runtime.h>
 
+#define IDX2C(i,j,ld) (((j)*(ld))+(i))
+
 // CUDA error checking macro
 #define CUDA_CHECK(call) \
 do { \

src/layers/dense.cpp

@@ -1,5 +1,10 @@
 #include "dense.h"
+#include "cuda_helper.h"
+#include <cstdlib>
 #include <cublas_v2.h>
+#include <cstdio>
+#include <stdexcept>
+	
 
 
 Layers::Dense::Dense(int inputSize, int outputSize, cublasHandle_t cublasHandle)
@@ -13,12 +18,10 @@ Layers::Dense::Dense(int inputSize, int outputSize, cublasHandle_t cublasHandle)
     initializeBiases();
 
     // Allocate GPU memory for weights and biases
-    cudaMalloc((void**)&d_weights, sizeof(float) * inputSize * outputSize);
+    CUDA_CHECK(cudaMalloc((void**)&d_weights, sizeof(float) * inputSize * outputSize));
-    cudaMalloc((void**)&d_biases, sizeof(float) * biases.size());
+    CUDA_CHECK(cudaMalloc((void**)&d_biases, sizeof(float) * biases.size()));
 
-    // Copy weights and biases to GPU
+    to_cuda();
-    cudaMemcpy(d_weights, weights.data(), sizeof(float) * inputSize * outputSize, cudaMemcpyHostToDevice);
-    cudaMemcpy(d_biases, biases.data(), sizeof(float) * biases.size(), cudaMemcpyHostToDevice);
 }
 
 Layers::Dense::~Dense() {
@@ -49,4 +52,9 @@ void Layers::Dense::forward(const float* input, float* output) {
 
     // Add biases
     cublasSaxpy(cublasHandle, outputSize, &alpha, d_biases, 1, output, 1);
+}
+
+void Layers::Dense::to_cuda() {
+    CUDA_CHECK(cudaMemcpy(d_weights, weights.data(), sizeof(float) * inputSize * outputSize, cudaMemcpyHostToDevice));
+    CUDA_CHECK(cudaMemcpy(d_biases, biases.data(), sizeof(float) * biases.size(), cudaMemcpyHostToDevice));
 }