OpenCV mat mult

mult_random.cpp

#include <iostream>
#include <vector>
#include <iomanip>
#include <chrono>

#include "opencv2/highgui/highgui.hpp"
#ifdef USE_CUBLAS
#include "opencv2/gpu/gpu.hpp"
#endif
#ifdef USE_CLBLAS
#include "opencv2/ocl/ocl.hpp"
#endif
#include "opencv2/video/video.hpp"

using namespace cv;
#ifdef USE_CUBLAS
using namespace cv::gpu;
#endif
#ifdef USE_CLBLAS
using namespace cv::ocl;
#endif

const unsigned int NB_POINTS = 2048;

int main(int ac, char** av) {
    try {
        for (int i = 0; i < 10; ++i) {
            Mat mat1(NB_POINTS, 128, CV_32FC1);
            Mat mat2(128, NB_POINTS, CV_32FC1);
            Mat mat3;
            Mat result;
            auto start = std::chrono::system_clock::now();
            gemm(mat1, mat2, 1.0, mat3, 1.0, result);
            auto end = std::chrono::system_clock::now();
            std::chrono::duration<double> total = end - start;
            std::cout << "time (CPU)      : " << total.count() << std::endl;
        }
#ifdef USE_CUBLAS
        for (int i = 0; i < 10; ++i) {
            GpuMat mat1(NB_POINTS, 128, CV_32FC1);
            GpuMat mat2(128, NB_POINTS, CV_32FC1);
            GpuMat mat3;
            GpuMat result;
            auto start = std::chrono::system_clock::now();
            cv::gpu::gemm(mat1, mat2, 1.0, mat3, 1.0, result);
            auto end = std::chrono::system_clock::now();
            std::chrono::duration<double> total = end - start;
            std::cout << "time (CUDA)     : " << total.count() << std::endl;
        }
#endif
#ifdef USE_CLBLAS
        for (int i = 0; i < 10; ++i) {
            oclMat mat1(NB_POINTS, 128, CV_32FC1);
            oclMat mat2(128, NB_POINTS, CV_32FC1);
            oclMat mat3;
            oclMat result;
            auto start = std::chrono::system_clock::now();
            cv::ocl::gemm(mat1, mat2, 1.0, mat3, 1.0, result);
            auto end = std::chrono::system_clock::now();
            std::chrono::duration<double> total = end - start;
            std::cout << "time (OpenCL)   : " << total.count() << std::endl;
        }
#endif
    } catch (Exception& ex) {
        std::cout << "exception(CV)   : " << ex.what() << std::endl;
    }
    return 0;
}

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D CUDA_GENERATION=Kepler -DWITH_CUBLAS=ON -DWITH_OPENCLAMDBLAS=ON ..