Migrate concat layer

src/backends/cuda/layer_ops.cu

@@ -211,4 +211,24 @@ CUDANet::Tensor& CUDA::batch_norm(
         );
         CUDA_CHECK(cudaGetLastError());
     }
+}
+
+CUDANet::Tensor& CUDA::concat(
+    CUDANet::Tensor& input_a,
+    CUDANet::Tensor& input_b,
+    CUDANet::Tensor& output
+) {
+    CUDA_CHECK(cudaMemcpy(
+        output.data<float>(), input_a.data<float>(), input_a.size(),
+        cudaMemcpyDeviceToDevice
+    ));
+
+    CUDA_CHECK(cudaMemcpy(
+        output.data<float>() + input_a.numel(), input_b.data<float>(), input_b.size(),
+        cudaMemcpyDeviceToDevice
+    ));
+
+    CUDA_CHECK(cudaDeviceSynchronize());
+
+    return output;
 }

src/backends/cuda/layers/concat.cu

@@ -1,31 +0,0 @@
-#include "concat.hpp"
-#include "cuda_helper.cuh"
-
-using namespace CUDANet::Layers;
-
-void Concat::initCUDA() {
-    d_output = nullptr;
-    CUDA_CHECK(
-        cudaMalloc((void**)&d_output, sizeof(float) * (inputASize + inputBSize))
-    );
-}
-
-void Concat::delCUDA() {
-    cudaFree(d_output);
-}
-
-float* Concat::forwardCUDA(const float* d_input_A, const float* d_input_B) {
-    CUDA_CHECK(cudaMemcpy(
-        d_output, d_input_A, sizeof(float) * inputASize,
-        cudaMemcpyDeviceToDevice
-    ));
-
-    CUDA_CHECK(cudaMemcpy(
-        d_output + inputASize, d_input_B, sizeof(float) * inputBSize,
-        cudaMemcpyDeviceToDevice
-    ));
-
-    CUDA_CHECK(cudaDeviceSynchronize());
-
-    return d_output;
-}