Fix more cuda memcheck errors

src/layers/dense.cu

@@ -65,21 +65,15 @@ void Dense::initializeBiases() {
 
 float* Dense::forward(const float* d_input) {
 
-    // CUDANet::Utils::clear(d_output, outputSize);
-
-    // CUDA_CHECK(cudaPeekAtLastError());
-
-    std::cout << "Dense::forward" << std::endl;
-
     Kernels::mat_vec_mul<<<forwardGridSize, BLOCK_SIZE>>>(
         d_weights, d_input, d_output, inputSize, outputSize
     );
-    CUDA_CHECK(cudaPeekAtLastError());
+    CUDA_CHECK(cudaGetLastError());
 
     Kernels::vec_vec_add<<<biasGridSize, BLOCK_SIZE>>>(
         d_biases, d_output, d_output, outputSize
     );
-    CUDA_CHECK(cudaPeekAtLastError());
+    CUDA_CHECK(cudaGetLastError());
 
     activation.activate(d_output);
     CUDA_CHECK(cudaDeviceSynchronize());

test/layers/test_avg_pooling.cu

@@ -66,5 +66,4 @@ TEST(AvgPoolingLayerTest, AvgPoolForwardTest) {
     }
 
     cudaFree(d_input);
-    cudaFree(d_output);
 }

test/layers/test_concat.cu

@@ -4,6 +4,9 @@
 #include <vector>
 
 TEST(ConcatLayerTest, Init) {
+
+    cudaError_t cudaStatus;
+
     std::vector<float> inputA = {0.573f, 0.619f, 0.732f, 0.055f, 0.243f};
     std::vector<float> inputB = {0.123f, 0.321f, 0.456f, 0.789f, 0.654f, 0.123f};
 
@@ -11,27 +14,41 @@ TEST(ConcatLayerTest, Init) {
 
     float* d_inputA;
     float* d_inputB;
-    cudaMalloc((void**)&d_inputA, sizeof(float) * 5);
-    cudaMalloc((void**)&d_inputB, sizeof(float) * 6);
-    cudaMemcpy(
+
+    cudaStatus = cudaMalloc((void**)&d_inputA, sizeof(float) * 5);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
+
+    cudaStatus = cudaMalloc((void**)&d_inputB, sizeof(float) * 6);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
+
+    cudaStatus = cudaMemcpy(
         d_inputA, inputA.data(), sizeof(float) * 5, cudaMemcpyHostToDevice
     );
-    cudaMemcpy(
+    EXPECT_EQ(cudaStatus, cudaSuccess);
+
+    cudaStatus = cudaMemcpy(
         d_inputB, inputB.data(), sizeof(float) * 6, cudaMemcpyHostToDevice
     );
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     float* d_output = concat.forward(d_inputA, d_inputB);
 
     std::vector<float> output(11);
-    cudaMemcpy(
+    cudaStatus = cudaMemcpy(
         output.data(), d_output, sizeof(float) * 11, cudaMemcpyDeviceToHost
     );
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     for (int i = 0; i < 5; ++i) {
         EXPECT_EQ(output[i], inputA[i]);
     }
+
     for (int i = 0; i < 6; ++i) {
         EXPECT_EQ(output[i + 5], inputB[i]);
     }
-    cudaFree(d_output);
+
+    cudaStatus = cudaFree(d_inputA);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
+    cudaStatus = cudaFree(d_inputB);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 }