Remove cublas dependency

test/CMakeLists.txt

@@ -7,12 +7,6 @@ add_executable(test_main
     kernels/test_padding.cu
 )
 
-add_library(test_utils
-    test_utils/test_cublas_fixture.cu
-)
-
-target_include_directories(test_utils PUBLIC test_utils)
-
-target_link_libraries(test_main ${GTEST_BOTH_LIBRARIES} CUDANet test_utils)
+target_link_libraries(test_main ${GTEST_BOTH_LIBRARIES} CUDANet)
 
 add_test(NAME TestMain COMMAND test_main)

test/kernels/test_activations.cu

@@ -4,15 +4,11 @@
 #include <iostream>
 
 #include "activations.cuh"
-#include "test_cublas_fixture.cuh"
 
-class ActivationsTest : public CublasTestFixture {
-  protected:
-    cudaError_t    cudaStatus;
-    cublasStatus_t cublasStatus;
-};
+TEST(ActivationsTest, SigmoidSanityCheck) {
+
+    cudaError_t cudaStatus;
 
-TEST_F(ActivationsTest, SigmoidSanityCheck) {
     float input[3] = {-100.0f, 0.0f, 100.0f};
 
     std::vector<float> expected_output = {0.0f, 0.5f, 1.0f};
@@ -26,8 +22,8 @@ TEST_F(ActivationsTest, SigmoidSanityCheck) {
     cudaStatus = cudaMalloc((void**)&d_output, sizeof(float) * 3);
     EXPECT_EQ(cudaStatus, cudaSuccess);
 
-    cublasStatus = cublasSetVector(3, sizeof(float), input, 1, d_input, 1);
-    EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+    cudaStatus = cudaMemcpy(d_input, input, sizeof(float) * 3, cudaMemcpyHostToDevice);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     sigmoid_kernel<<<1, 3>>>(d_input, d_output, 3);
     cudaStatus = cudaDeviceSynchronize();
@@ -35,9 +31,8 @@ TEST_F(ActivationsTest, SigmoidSanityCheck) {
 
     std::vector<float> output(3);
 
-    cublasStatus =
-        cublasGetVector(3, sizeof(float), d_output, 1, output.data(), 1);
-    EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+    cudaStatus = cudaMemcpy(output.data(), d_output, sizeof(float) * 3, cudaMemcpyDeviceToHost);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     for (int i = 0; i < 3; i++) {
         EXPECT_NEAR(expected_output[i], output[i], 1e-5);

test/kernels/test_padding.cu

@@ -4,15 +4,10 @@
 #include <iostream>
 
 #include "padding.cuh"
-#include "test_cublas_fixture.cuh"
 
-class PaddingTest : public CublasTestFixture {
-  protected:
-    cudaError_t    cudaStatus;
-    cublasStatus_t cublasStatus;
-};
+TEST(PaddingTest, SimplePaddingTest) {
+    cudaError_t cudaStatus;
 
-TEST_F(PaddingTest, SimplePaddingTest) {
     int w = 2;
     int h = 3;
     int n = 2;
@@ -48,9 +43,10 @@ TEST_F(PaddingTest, SimplePaddingTest) {
     std::vector<float> input = {0.0f, 2.0f, 4.0f,  1.0f, 3.0f, 5.0f,
                                 6.0f, 8.0f, 10.0f, 7.0f, 9.0f, 11.0f};
 
-    cublasStatus =
-        cublasSetVector(inputSize, sizeof(float), input.data(), 1, d_input, 1);
-    EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+    cudaStatus = cudaMemcpy(
+        d_input, input.data(), sizeof(float) * inputSize, cudaMemcpyHostToDevice
+    );
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     int THREADS_PER_BLOCK = 64;
     int BLOCKS            = paddedSize / THREADS_PER_BLOCK + 1;
@@ -69,9 +65,12 @@ TEST_F(PaddingTest, SimplePaddingTest) {
     };
 
     std::vector<float> output(paddedSize);
-    cublasStatus = cublasGetVector(
-        paddedSize, sizeof(float), d_padded, 1, output.data(), 1
+
+    cudaStatus = cudaMemcpy(
+        output.data(), d_padded, sizeof(float) * paddedSize,
+        cudaMemcpyDeviceToHost
     );
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     for (int i = 0; i < paddedSize; i++) {
         EXPECT_NEAR(expectedOutput[i], output[i], 1e-5);

test/layers/test_dense.cu

@@ -5,9 +5,9 @@
 
 #include "activations.cuh"
 #include "dense.cuh"
-#include "test_cublas_fixture.cuh"
 
-class DenseLayerTest : public CublasTestFixture {
+
+class DenseLayerTest : public::testing::Test {
   protected:
     Layers::Dense commonTestSetup(
         int                              inputSize,
@@ -21,7 +21,7 @@ class DenseLayerTest : public CublasTestFixture {
     ) {
         // Create Dense layer
         Layers::Dense denseLayer(
-            inputSize, outputSize, activation, cublasHandle
+            inputSize, outputSize, activation
         );
 
         // Set weights and biases
@@ -36,10 +36,11 @@ class DenseLayerTest : public CublasTestFixture {
         EXPECT_EQ(cudaStatus, cudaSuccess);
 
         // Copy input to device
-        cublasStatus = cublasSetVector(
-            input.size(), sizeof(float), input.data(), 1, d_input, 1
+        cudaStatus = cudaMemcpy(
+            d_input, input.data(), sizeof(float) * input.size(), cudaMemcpyHostToDevice
         );
-        EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+        EXPECT_EQ(cudaStatus, cudaSuccess);
+
 
         return denseLayer;
     }
@@ -51,7 +52,6 @@ class DenseLayerTest : public CublasTestFixture {
     }
 
     cudaError_t    cudaStatus;
-    cublasStatus_t cublasStatus;
 };
 
 TEST_F(DenseLayerTest, Init) {
@@ -60,10 +60,8 @@ TEST_F(DenseLayerTest, Init) {
             int inputSize  = i;
             int outputSize = j;
 
-            // std::cout << "Dense layer: input size = " << inputSize << ",
-            // output size = " << outputSize << std::endl;
             Layers::Dense denseLayer(
-                inputSize, outputSize, SIGMOID, cublasHandle
+                inputSize, outputSize, SIGMOID
             );
         }
     }
@@ -81,7 +79,7 @@ TEST_F(DenseLayerTest, setWeights) {
         {1.3f, 0.5f, 0.0f, 1.7f}
     };
 
-    Layers::Dense denseLayer(inputSize, outputSize, SIGMOID, cublasHandle);
+    Layers::Dense denseLayer(inputSize, outputSize, SIGMOID);
 
     denseLayer.setWeights(weights);
 }
@@ -113,10 +111,10 @@ TEST_F(DenseLayerTest, ForwardUnitWeightMatrixLinear) {
     denseLayer.forward(d_input, d_output);
 
     std::vector<float> output(outputSize);
-    cublasStatus = cublasGetVector(
-        outputSize, sizeof(float), d_output, 1, output.data(), 1
+    cudaStatus = cudaMemcpy(
+        output.data(), d_output, sizeof(float) * outputSize, cudaMemcpyDeviceToHost
     );
-    EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     // Check if the output is a zero vector
     EXPECT_FLOAT_EQ(output[0], 2.0f);
@@ -150,10 +148,10 @@ TEST_F(DenseLayerTest, ForwardRandomWeightMatrixRelu) {
     denseLayer.forward(d_input, d_output);
 
     std::vector<float> output(outputSize);
-    cublasStatus = cublasGetVector(
-        outputSize, sizeof(float), d_output, 1, output.data(), 1
+    cudaStatus = cudaMemcpy(
+        output.data(), d_output, sizeof(float) * outputSize, cudaMemcpyDeviceToHost
     );
-    EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     // weights * inputs = 0.1, 12.5, 8.3, -2.2
     // + biases = 0.3, 13, 9, -3.3
@@ -193,10 +191,10 @@ TEST_F(DenseLayerTest, ForwardRandomWeightMatrixSigmoid) {
     denseLayer.forward(d_input, d_output);
 
     std::vector<float> output(outputSize);
-    cublasStatus = cublasGetVector(
-        outputSize, sizeof(float), d_output, 1, output.data(), 1
+    cudaStatus = cudaMemcpy(
+        output.data(), d_output, sizeof(float) * outputSize, cudaMemcpyDeviceToHost
     );
-    EXPECT_EQ(cublasStatus, CUBLAS_STATUS_SUCCESS);
+    EXPECT_EQ(cudaStatus, cudaSuccess);
 
     // weights * input = 0.95, 0.43, 0.45, 0.93
     // + biases = 1.05, 0.63, 0.75, 1.33

test/test_utils/test_cublas_fixture.cu

@@ -1,14 +0,0 @@
-#include <cublas_v2.h>
-#include <gtest/gtest.h>
-
-#include "test_cublas_fixture.cuh"
-
-cublasHandle_t CublasTestFixture::cublasHandle;
-
-void CublasTestFixture::SetUpTestSuite() {
-    cublasCreate(&cublasHandle);
-}
-
-void CublasTestFixture::TearDownTestSuite() {
-    cublasDestroy(cublasHandle);
-}

test/test_utils/test_cublas_fixture.cuh

@@ -1,10 +0,0 @@
-#include <cublas_v2.h>
-#include <gtest/gtest.h>
-
-class CublasTestFixture : public ::testing::Test {
-  protected:
-    static cublasHandle_t cublasHandle;
-
-    static void SetUpTestSuite();
-    static void TearDownTestSuite();
-};