Migrate input layer

include/cudanet.cuh

@@ -16,7 +16,7 @@
 #include "concat.hpp"
 #include "conv2d.cuh"
 #include "dense.hpp"
-#include "input.cuh"
+#include "input.hpp"
 #include "layer.hpp"
 #include "max_pooling.hpp"
 #include "output.cuh"

include/layers/input.hpp

@@ -40,16 +40,25 @@ class Input : public SequentialLayer {
      */
     int getOutputSize();
 
-        /**
+    /**
      * @brief Get input size
-     * 
+     *
      * @return int input size
      */
     int getInputSize();
 
   private:
-    int    inputSize;
+    int inputSize;
+
+    float* forwardCPU(const float* input);
+
+#ifdef USE_CUDA
     float* d_output;
+
+    float* forwardCUDA(const float* input);
+    void   initCUDA();
+    void   delCUDA();
+#endif
 };
 
 }  // namespace CUDANet::Layers

include/model/model.hpp

@@ -5,7 +5,7 @@
 #include <unordered_map>
 #include <vector>
 
-#include "input.cuh"
+#include "input.hpp"
 #include "layer.hpp"
 #include "module.hpp"
 #include "output.cuh"

src/backends/cuda/layers/input.cu

@@ -1,31 +1,22 @@
 #include "cuda_helper.cuh"
-#include "input.cuh"
+#include "input.hpp"
 
 using namespace CUDANet::Layers;
 
-Input::Input(int inputSize) : inputSize(inputSize) {
+void Input::initCUDA() {
     d_output = nullptr;
     CUDA_CHECK(cudaMalloc((void**)&d_output, sizeof(float) * inputSize));
 }
 
-Input::~Input() {
+void Input::delCUDA() {
     cudaFree(d_output);
 }
 
-float* Input::forward(const float* input) {
+float* Input::forwardCUDA(const float* input) {
     CUDA_CHECK(cudaMemcpy(
         d_output, input, sizeof(float) * inputSize, cudaMemcpyHostToDevice
     ));
     CUDA_CHECK(cudaDeviceSynchronize());
 
     return d_output;
-}
-
-int Input::getOutputSize() {
-    return inputSize;
-}
-
-
-int Input::getInputSize() {
-    return inputSize;
 }

src/layers/input.cpp

@@ -0,0 +1,37 @@
+#include <stdexcept>
+
+#include "input.hpp"
+
+using namespace CUDANet::Layers;
+
+Input::Input(int inputSize) : inputSize(inputSize) {
+#ifdef USE_CUDA
+    initCUDA();
+#endif
+}
+
+Input::~Input() {
+#ifdef USE_CUDA
+    delCUDA();
+#endif
+}
+
+float* Input::forwardCPU(const float* input) {
+    throw std::logic_error("Not implemented");
+}
+
+float* Input::forward(const float* input) {
+#ifdef USE_CUDA
+    return forwardCUDA(input);
+#else
+    return forwardCPU(input);
+#endif
+}
+
+int Input::getOutputSize() {
+    return inputSize;
+}
+
+int Input::getInputSize() {
+    return inputSize;
+}

src/model/model.cpp

@@ -7,7 +7,7 @@
 #include <unordered_map>
 #include <vector>
 
-#include "input.cuh"
+#include "input.hpp"
 #include "layer.hpp"
 #include "batch_norm.cuh"
 

test/cuda/layers/test_input.cu

@@ -1,7 +1,7 @@
 #include <cuda_runtime_api.h>
 #include <gtest/gtest.h>
 
-#include "input.cuh"
+#include "input.hpp"
 
 TEST(InputLayerTest, InputForward) {
     std::vector<float> input = {0.573f, 0.619f, 0.732f, 0.055f, 0.243f, 0.316f};