add random test for Add

unit_tests/Test_AddImpl.cpp

@@ -9,12 +9,15 @@
  *
  ********************************************************************************/
 
+#include <numeric> // std::accumulate
+#include <random>  // std::random_device, std::mt19937, std::uniform_real_distribution
 #include <catch2/catch_test_macros.hpp>
 
 #include "aidge/backend/cpu.hpp"
 #include "aidge/backend/cuda.hpp"
 #include "aidge/data/Tensor.hpp"
 #include "aidge/operator/Add.hpp"
+#include "aidge/utils/TensorUtils.hpp"
 
 using namespace Aidge;
 
@@ -207,4 +210,128 @@ TEST_CASE("[gpu/operator] Add(forward)", "[Add][GPU]") {
 
         delete[] computedOutput;
     }
+
+    SECTION("Random Input") {
+        constexpr std::uint16_t NBTRIALS = 10;
+        // Create a random number generator
+        std::random_device rd;
+        std::mt19937 gen(rd());
+        std::uniform_real_distribution<float> valueDist(
+            0.1f, 1.1f); // Random float distribution between 0 and 1
+        std::uniform_int_distribution<std::size_t> dimSizeDist(std::size_t(1),
+                                                               std::size_t(10));
+        std::uniform_int_distribution<std::size_t> nbDimsDist(std::size_t(4), std::size_t(5));
+        std::uniform_int_distribution<int> boolDist(0,1);
+
+        // To measure execution time of 'forward()'
+        std::chrono::time_point<std::chrono::system_clock> start;
+        std::chrono::time_point<std::chrono::system_clock> end;
+        std::chrono::duration<double, std::micro> duration{};
+        std::size_t number_of_operation = 0;
+        for (std::uint16_t trial = 0; trial < NBTRIALS; ++trial)
+        {
+            // Create BatchNorm Operator CUDA
+            std::shared_ptr<Node> myAddCUDA = Add(2, "myaddcuda");
+            auto op_cuda = std::static_pointer_cast<OperatorTensor>(myAddCUDA -> getOperator());
+
+            // Create BatchNorm Operator CPU
+            std::shared_ptr<Node> myAddCPU = Add(2, "myaddcpu");
+            auto op_cpu = std::static_pointer_cast<OperatorTensor>(myAddCPU -> getOperator());
+            op_cpu->setDataType(DataType::Float32);
+            op_cpu->setBackend("cpu");
+
+            const std::size_t nbDims = nbDimsDist(gen);
+            std::vector<std::size_t> dims0, dims1, dims;
+            for (std::size_t i = 0; i < nbDims; ++i) {
+                const std::size_t dim = dimSizeDist(gen);
+                // To test broadcasting, set some dims to 1
+                if (boolDist(gen)) {
+                    dims0.push_back(1);
+                }else{
+                    dims0.push_back(dim);
+                }
+                if (boolDist(gen)) {
+                    dims1.push_back(1);
+                }else{
+                    dims1.push_back(dim);
+                }
+                dims.push_back(std::max(dims0[i], dims1[i]));
+            }
+            const std::size_t nb_elements0 = std::accumulate(dims0.cbegin(), dims0.cend(), std::size_t(1), std::multiplies<std::size_t>());
+            const std::size_t nb_elements1 = std::accumulate(dims1.cbegin(), dims1.cend(), std::size_t(1), std::multiplies<std::size_t>());
+            const std::size_t nb_elements = std::accumulate(dims.cbegin(), dims.cend(), std::size_t(1), std::multiplies<std::size_t>());
+            number_of_operation += nb_elements;
+
+            float* array0 = new float[nb_elements0];
+            float* array1 = new float[nb_elements1];
+
+            for (std::size_t i = 0; i < nb_elements0; ++i) {
+                array0[i] = valueDist(gen);
+            }
+            for (std::size_t i = 0; i < nb_elements1; ++i) {
+                array1[i] = valueDist(gen);
+            }
+
+            // input0 CUDA
+            float* array0_d, *array1_d;
+            std::shared_ptr<Tensor> T0_cuda = std::make_shared<Tensor>();
+            T0_cuda->setDataType(DataType::Float32);
+            T0_cuda->setBackend("cuda");
+            T0_cuda->resize(dims0);
+            op_cuda->associateInput(0, T0_cuda);
+            cudaMalloc(reinterpret_cast<void **>(&array0_d), sizeof(float) * nb_elements0);
+            cudaMemcpy(array0_d, array0, sizeof(float) * nb_elements0, cudaMemcpyHostToDevice);
+            T0_cuda->getImpl()->setRawPtr(array0_d, nb_elements0);
+
+            // input0 CPU
+            std::shared_ptr<Tensor> T0_cpu = std::make_shared<Tensor>();
+            op_cpu->associateInput(0,T0_cpu);
+            T0_cpu->setDataType(DataType::Float32);
+            T0_cpu->setBackend("cpu");
+            T0_cpu->resize(dims0);
+            T0_cpu -> getImpl() -> setRawPtr(array0, nb_elements0);
+
+            // input1 CUDA
+            std::shared_ptr<Tensor> T1_cuda = std::make_shared<Tensor>();
+            T1_cuda->setDataType(DataType::Float32);
+            T1_cuda->setBackend("cuda");
+            T1_cuda->resize(dims1);
+            op_cuda->associateInput(1, T1_cuda);
+            cudaMalloc(reinterpret_cast<void **>(&array1_d), sizeof(float) * nb_elements1);
+            cudaMemcpy(array1_d, array1, sizeof(float) * nb_elements1, cudaMemcpyHostToDevice);
+            T1_cuda->getImpl()->setRawPtr(array1_d, nb_elements1);
+
+            // input1 CPU
+            std::shared_ptr<Tensor> T1_cpu = std::make_shared<Tensor>();
+            op_cpu->associateInput(1,T1_cpu);
+            T1_cpu->setDataType(DataType::Float32);
+            T1_cpu->setBackend("cpu");
+            T1_cpu->resize(dims1);
+            T1_cpu -> getImpl() -> setRawPtr(array1, nb_elements1);
+
+            // forward CUDA
+            op_cuda->setDataType(DataType::Float32);
+            op_cuda->setBackend("cuda");
+            start = std::chrono::system_clock::now();
+            op_cuda->forward();
+            end = std::chrono::system_clock::now();
+            duration += std::chrono::duration_cast<std::chrono::microseconds>(end - start);
+
+            float *computedOutput = new float[nb_elements]();
+            cudaMemcpy(computedOutput, op_cuda->getOutput(0)->getImpl()->rawPtr(), sizeof(float) * nb_elements, cudaMemcpyDeviceToHost);
+
+            // forward CPU
+            op_cpu->forward();
+            float *computedCPU = static_cast<float*>(op_cpu->getOutput(0)->getImpl()->rawPtr());
+            REQUIRE(approxEq<float>(*computedOutput, *computedCPU));
+
+            delete[] array0;
+            delete[] array1;
+            delete[] computedOutput;
+            cudaFree(array0_d);
+            cudaFree(array1_d);
+
+        }
+    }
+
 }
\ No newline at end of file