Implement Add layer

CMakeLists.txt

@@ -12,6 +12,7 @@ set(LIBRARY_SOURCES
     src/kernels/activation_functions.cu
     src/kernels/convolution.cu
     src/kernels/matmul.cu
+    src/layers/add.cu
     src/layers/dense.cu
     src/layers/conv2d.cu
     src/layers/concat.cu

include/layers/add.cuh

@@ -0,0 +1,40 @@
+#ifndef CUDANET_ADD_LAYER_H
+#define CUDANET_ADD_LAYER_H
+
+namespace CUDANet::Layers {
+
+class Add {
+  public:
+    /**
+     * @brief Create a new Add layer
+     *
+     * @param inputSize Size of the input arrays
+     */
+    Add(int inputSize);
+
+    /**
+     * @brief Destroy the Add layer
+     *
+     */
+    ~Add();
+
+    /**
+     * @brief Adds the two inputs
+     *
+     * @param d_inputA Device pointer to the first input
+     * @param d_inputB Device pointer to the second input
+     *
+     * @return Device pointer to the output
+     */
+    float* forward(const float* d_inputA, const float* d_inputB);
+
+  private:
+    int inputSize;
+    int gridSize;
+
+    float* d_output;
+};
+
+}  // namespace CUDANet::Layers
+
+#endif  // CUDANET_ADD_LAYER_H

src/layers/add.cu

@@ -0,0 +1,29 @@
+#include "add.cuh"
+#include "matmul.cuh"
+#include "cuda_helper.cuh"
+
+using namespace CUDANet;
+
+
+Layers::Add::Add(int inputSize)
+    : inputSize(inputSize) {
+
+    d_output = nullptr;
+    CUDA_CHECK(cudaMalloc((void**)&d_output, sizeof(float) * inputSize));
+
+    gridSize = (inputSize + BLOCK_SIZE - 1) / BLOCK_SIZE;
+}
+
+
+Layers::Add::~Add() {
+    cudaFree(d_output);
+}
+
+
+float* Layers::Add::forward(const float* d_inputA, const float* d_inputB) {
+
+    Kernels::vec_vec_add<<<gridSize, BLOCK_SIZE>>>(
+        d_inputA, d_inputB, d_output, inputSize
+    );
+
+}