Start conv test implementation

include/layers/conv2d.cuh

@@ -11,16 +11,19 @@ namespace Layers {
 class Conv2d {
   public:
     Conv2d(
-        int            inputSize,
+        int         inputSize,
-        int            inputChannels,
+        int         inputChannels,
-        int            kernelSize,
+        int         kernelSize,
-        int            stride,
+        int         stride,
-        std::string    padding,
+        std::string padding,
-        int            numFilters,
+        int         numFilters,
-        Activation     activation
+        Activation  activation
     );
     ~Conv2d();
 
+    // Outputs
+    int outputSize;
+
     void forward(const float* d_input, float* d_output);
 
   private:
@@ -34,15 +37,12 @@ class Conv2d {
     int paddingSize;
     int numFilters;
 
-    // Outputs
-    int outputSize;
-
     // Kernels
     std::vector<float> kernels;
 
     // Cuda
-    float*         d_kernels;
+    float* d_kernels;
-    float*         d_padded;
+    float* d_padded;
 
     // Kernels
     Activation activation;

test/CMakeLists.txt

@@ -3,6 +3,7 @@ include_directories(${GTEST_INCLUDE_DIRS})
 
 add_executable(test_main 
     layers/test_dense.cu
+    layers/test_conv2d.cu
     kernels/test_activations.cu
     kernels/test_padding.cu
 )

test/layers/test_conv2d.cu

@@ -0,0 +1,35 @@
+#include <cuda_runtime_api.h>
+#include <gtest/gtest.h>
+
+#include <iostream>
+
+#include "conv2d.cuh"
+
+TEST(Conv2dTest, ValidPadding) {
+
+    int inputSize = 3;
+    int inputChannels = 1;
+    int kernelSize = 3;
+    int stride = 1;
+    std::string padding = "VALID";
+    int numFilters = 1;
+    Activation activation = LINEAR;
+
+    Layers::Conv2d conv2d(
+        inputSize,
+        inputChannels,
+        kernelSize,
+        stride,
+        padding,
+        numFilters,
+        activation
+    );
+
+    int outputSize = (inputSize - kernelSize) / stride + 1;
+    EXPECT_EQ(outputSize, conv2d.outputSize);
+
+    std::vector<float> input(inputSize * inputSize * inputChannels);
+    std::vector<float> output(outputSize * outputSize * numFilters);
+    std::vector<float> kernels(kernelSize * kernelSize * inputChannels * numFilters);
+
+}