当前位置: 首页>>代码示例>>C++>>正文

C++ profiler::tic_cpu方法代码示例

本文整理汇总了C++中vex::profiler::tic_cpu方法的典型用法代码示例。如果您正苦于以下问题：C++ profiler::tic_cpu方法的具体用法？C++ profiler::tic_cpu怎么用？C++ profiler::tic_cpu使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类vex::profiler的用法示例。


在下文中一共展示了profiler::tic_cpu方法的6个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的C++代码示例。

示例1: make_pair

std::pair<double, double> benchmark_reductor(
        const vex::Context &ctx, vex::profiler<> &prof
        )
{
    const size_t N = 16 * 1024 * 1024;
    const size_t M = 1024 / 16;
    double time_elapsed;

    std::vector<real> A = random_vector<real>(N);
    std::vector<real> B = random_vector<real>(N);

    vex::vector<real> a(ctx, A);
    vex::vector<real> b(ctx, B);

    vex::Reductor<real, vex::SUM> sum(ctx);

    real sum_cl = sum(a * b);
    sum_cl = 0;

    prof.tic_cpu("OpenCL");
    for(size_t i = 0; i < M; i++)
        sum_cl += sum(a * b);
    ctx.finish();
    time_elapsed = prof.toc("OpenCL");

    double gflops = 2.0 * N * M / time_elapsed / 1e9;
    double bwidth = 2.0 * N * M * sizeof(real) / time_elapsed / 1e9;

    std::cout
        << "Reduction (" << vex::type_name<real>() << ")\n"
        << "  OpenCL"
        << "\n    GFLOPS:    " << gflops
        << "\n    Bandwidth: " << bwidth
        << std::endl;

    if (options.bm_cpu) {
        real sum_cpp = 0;
        prof.tic_cpu("C++");
        for(size_t i = 0; i < M; i++)
            sum_cpp += std::inner_product(A.begin(), A.end(), B.begin(), static_cast<real>(0));
        time_elapsed = prof.toc("C++");

        {
            double gflops = 2.0 * N * M / time_elapsed / 1e9;
            double bwidth = 2.0 * N * M * sizeof(real) / time_elapsed / 1e9;

            std::cout
                << "  C++"
                << "\n    GFLOPS:    " << gflops
                << "\n    Bandwidth: " << bwidth
                << std::endl;
        }

        std::cout << "  res = " << fabs( (sum_cl - sum_cpp) / sum_cpp )
                  << std::endl << std::endl;
    }

    return std::make_pair(gflops, bwidth);
}
开发者ID:odellus，项目名称:vexcl，代码行数:59，代码来源:benchmark.cpp

示例2: benchmark_rng

void benchmark_rng(
        const vex::Context &ctx, vex::profiler<> &prof
        )
{
    const size_t N = 16 * 1024 * 1024;
    const size_t M = 1024;

    prof.tic_cpu("OpenCL (threefry)");
    double rps = rng_throughput<real, vex::random::threefry>(ctx, N, M);
    prof.toc("OpenCL (threefry)");

    std::cout
        << "Random numbers per second (" << vex::type_name<real>() << ")\n"
        << "    OpenCL (threefry): " << rps << std::endl;

    prof.tic_cpu("OpenCL (philox)");
    rps = rng_throughput<real, vex::random::philox>(ctx, N, M);
    prof.toc("OpenCL (philox)");

    std::cout
        << "    OpenCL (philox):   " << rps << std::endl;

    if (options.bm_cpu) {
        std::mt19937 rng( std::rand() );
        std::uniform_real_distribution<real> rnd(0.0, 1.0);

        prof.tic_cpu("C++ (mt19937)");
        real s = 0;
        for(size_t j = 0; j < N; j++)
            s = std::max(s, rnd(rng));
        double time_elapsed = prof.toc("C++ (mt19937)");

        std::cout
            << "    C++    (mt19937):  " << N / time_elapsed << std::endl;
    }

    std::cout << std::endl;
}
开发者ID:sd-omkar，项目名称:hpc_benchmarks，代码行数:38，代码来源:vector.cpp

示例3: run_tests

void run_tests(const vex::Context &ctx, vex::profiler<> &prof, int n_size)
{
    std::cout
        << "----------------------------------------------------------\n"
        << "Profiling \"" << vex::type_name<real>() << "\" performance\n"
        << "----------------------------------------------------------\n"
        << ctx << std::endl;

    std::ostringstream fname;
    fname << "profile_" << vex::type_name<real>() << ".dat";
    std::ofstream log(fname.str().c_str(), std::ios::app);

    log << ctx.size() << " ";

    double gflops, bwidth;

    prof.tic_cpu( vex::type_name<real>() );

    if (options.bm_saxpy) {
        prof.tic_cpu("Vector SAXPY");
        std::tie(gflops, bwidth) = benchmark_saxpy<real>(ctx, prof);
        prof.toc("Vector SAXPY");

        log << gflops << " " << bwidth << " ";
    }

    if (options.bm_vector) {
        prof.tic_cpu("Vector arithmetic");
        std::tie(gflops, bwidth) = benchmark_vector<real>(ctx, prof);
        prof.toc("Vector arithmetic");

        log << gflops << " " << bwidth << " ";
    }

    if (options.bm_reductor) {
        prof.tic_cpu("Reduction");
        std::tie(gflops, bwidth) = benchmark_reductor<real>(ctx, prof, n_size);
        prof.toc("Reduction");

        log << gflops << " " << bwidth << " ";
    }

    if (options.bm_stencil) {
        prof.tic_cpu("Stencil");
        std::tie(gflops, bwidth) = benchmark_stencil<real>(ctx, prof);
        prof.toc("Stencil");

        log << gflops << " " << bwidth << " ";
    }

    if (options.bm_spmv) {
        prof.tic_cpu("SpMV");
        std::tie(gflops, bwidth) = benchmark_spmv<real>(ctx, prof);
        prof.toc("SpMV");

        log << gflops << " " << bwidth << std::endl;

        prof.tic_cpu("SpMV (CCSR)");
        std::tie(gflops, bwidth) = benchmark_spmv_ccsr<real>(ctx, prof);
        prof.toc("SpMV (CCSR)");
    }

    if (options.bm_rng) {
        prof.tic_cpu("Random number generation");
        benchmark_rng<real>(ctx, prof);
        prof.toc("Random number generation");
    }

    if (options.bm_sort) {
        prof.tic_cpu("Sorting");
        benchmark_sort<real>(ctx, prof);
        prof.toc("Sorting");
    }

    /*
    if (options.bm_scan) {
        prof.tic_cpu("Scanning");
        benchmark_scan<real>(ctx, prof);
        prof.toc("Scanning");
    }
    */

    prof.toc( vex::type_name<real>() );

    std::cout << std::endl << std::endl;
}
开发者ID:sd-omkar，项目名称:hpc_benchmarks，代码行数:86，代码来源:vector.cpp

示例4: make_pair

std::pair<double,double> benchmark_saxpy(
        const vex::Context &ctx, vex::profiler<> &prof
        )
{
    const size_t N = 1024 * 1024;
    const size_t M = 1024;
    double time_elapsed;

    std::vector<real> A(N, 0);
    std::vector<real> B = random_vector<real>(N);
    std::vector<real> alphavec = random_vector<real>(1);
    real alpha = alphavec[0];

    vex::vector<real> a(ctx, A);
    vex::vector<real> b(ctx, B);

    auto ta = vex::tag<1>(a);

    ta = alpha * ta + b;
    ta = 0;

    prof.tic_cpu("OpenCL");
    for(size_t i = 0; i < M; i++)
        ta = alpha * ta + b;
    ctx.finish();
    time_elapsed = prof.toc("OpenCL");

    double gflops = (2.0 * N * M) / time_elapsed / 1e9;
    double bwidth = (3.0 * N * M * sizeof(real)) / time_elapsed / 1e9;

    std::cout
        << "Vector SAXPY (" << vex::type_name<real>() << ")\n"
        << "  OCL"
        << "\n    GFLOPS:    " << gflops
        << "\n    Bandwidth: " << bwidth
        << std::endl;

    if (options.bm_cpu) {
        prof.tic_cpu("C++");
        for(size_t i = 0; i < M; i++)
            for(size_t j = 0; j < N; j++)
                A[j] = alpha * A[j] + B[j];
        time_elapsed = prof.toc("C++");

        {
            double gflops = (2.0 * N * M) / time_elapsed / 1e9;
            double bwidth = (3.0 * N * M * sizeof(real)) / time_elapsed / 1e9;

            std::cout
                << "  C++"
                << "\n    GFLOPS:    " << gflops
                << "\n    Bandwidth: " << bwidth
                << std::endl;
        }

        vex::copy(A, b);
        vex::Reductor<real, vex::SUM> sum(ctx);

        a -= b;
        std::cout << "  res = " << sum(a * a)
                  << std::endl << std::endl;
    }

    return std::make_pair(gflops, bwidth);
}
开发者ID:sd-omkar，项目名称:hpc_benchmarks，代码行数:65，代码来源:vector.cpp

示例5: benchmark_scan

void benchmark_scan(
        const vex::Context &ctx, vex::profiler<> &prof
        )
{
    const size_t N = 16 * 1024 * 1024;
    const size_t M = 16;

    typedef typename std::conditional<
        std::is_same<float, real>::value, cl_uint, cl_ulong
        >::type key_type;

    std::default_random_engine rng( std::rand() );
    std::uniform_int_distribution<key_type> rnd;

    std::vector<key_type> x0(N);
    std::vector<key_type> x1(N);

    std::generate(x0.begin(), x0.end(), [&]() { return rnd(rng); });

    vex::vector<key_type> X0(ctx, x0);
    vex::vector<key_type> X1(ctx, N);

    vex::exclusive_scan(X0, X1);

    ctx.finish();
    prof.tic_cpu("VexCL");

    for(size_t i = 0; i < M; i++)
        vex::exclusive_scan(X0, X1);

    ctx.finish();
    double tot_time = prof.toc("VexCL");

    std::cout
        << "Scan (" << vex::type_name<key_type>() << ")\n"
        << "    VexCL:         " << N * M / tot_time << " keys/sec\n";

#ifdef HAVE_BOOST_COMPUTE
    vex::compute::exclusive_scan(X0, X1);

    ctx.finish();
    prof.tic_cpu("Boost.Compute");

    for(size_t i = 0; i < M; i++)
        vex::compute::exclusive_scan(X0, X1);

    ctx.finish();
    tot_time = prof.toc("Boost.Compute");

    std::cout
        << "    Boost.Compute: " << N * M / tot_time << " keys/sec\n";
#endif

#ifdef HAVE_CLOGS
    vex::clogs::exclusive_scan(X0, X1);

    ctx.finish();
    prof.tic_cpu("CLOGS");

    for(size_t i = 0; i < M; i++)
        vex::clogs::exclusive_scan(X0, X1);

    ctx.finish();
    tot_time = prof.toc("CLOGS");

    std::cout
        << "    CLOGS:         " << N * M / tot_time << " keys/sec\n";
#endif

    if (options.bm_cpu) {
        prof.tic_cpu("CPU");
        for(size_t i = 0; i < M; i++) {
            key_type sum = key_type();
            for(size_t j = 0; j < N; ++j) {
                key_type next = sum + x0[j];
                x1[j] = sum;
                sum = next;
            }
        }
        tot_time = prof.toc("CPU");

        std::cout << "    CPU:           " << N * M / tot_time << " keys/sec\n";
    }

    std::cout << std::endl;
}
开发者ID:sd-omkar，项目名称:hpc_benchmarks，代码行数:86，代码来源:vector.cpp

示例6: benchmark_sort

void benchmark_sort(
        const vex::Context &ctx, vex::profiler<> &prof
        )
{
    const size_t N = 16 * 1024 * 1024;
    const size_t M = 16;

    typedef typename std::conditional<
        std::is_same<float, real>::value, cl_uint, cl_ulong
        >::type key_type;

    std::default_random_engine rng( std::rand() );
    std::uniform_int_distribution<key_type> rnd;

    std::vector<key_type> x0(N);
    std::vector<key_type> x1(N);

    std::generate(x0.begin(), x0.end(), [&]() { return rnd(rng); });

    vex::vector<key_type> X0(ctx, x0);
    vex::vector<key_type> X1(ctx, N);

    X1 = X0;
    vex::sort(X1);

    double tot_time = 0;
    for(size_t i = 0; i < M; i++) {
        X1 = X0;
        ctx.finish();
        prof.tic_cpu("VexCL");
        vex::sort(X1);
        ctx.finish();
        tot_time += prof.toc("VexCL");
    }

    std::cout
        << "Sort (" << vex::type_name<key_type>() << ")\n"
        << "    VexCL:         " << N * M / tot_time << " keys/sec\n";

#ifdef HAVE_BOOST_COMPUTE
    X1 = X0;
    vex::compute::sort(X1);

    tot_time = 0;
    for(size_t i = 0; i < M; i++) {
        X1 = X0;
        ctx.finish();
        prof.tic_cpu("Boost.Compute");
        vex::compute::sort(X1);
        ctx.finish();
        tot_time += prof.toc("Boost.Compute");
    }

    std::cout
        << "    Boost.Compute: " << N * M / tot_time << " keys/sec\n";
#endif

#ifdef HAVE_CLOGS
    X1 = X0;
    vex::clogs::sort(X1);

    tot_time = 0;
    for(size_t i = 0; i < M; i++) {
        X1 = X0;
        ctx.finish();
        prof.tic_cpu("CLOGS");
        vex::clogs::sort(X1);
        ctx.finish();
        tot_time += prof.toc("CLOGS");
    }

    std::cout
        << "    CLOGS:         " << N * M / tot_time << " keys/sec\n";
#endif

    if (options.bm_cpu) {
        tot_time = 0;
        for(size_t i = 0; i < M; i++) {
            std::copy(x0.begin(), x0.end(), x1.begin());
            prof.tic_cpu("STL");
            std::sort(x1.begin(), x1.end());
            tot_time += prof.toc("STL");
        }

        std::cout << "    STL:           " << N * M / tot_time << " keys/sec\n";
    }

    std::cout << std::endl;
}
开发者ID:sd-omkar，项目名称:hpc_benchmarks，代码行数:89，代码来源:vector.cpp


注：本文中的vex::profiler::tic_cpu方法示例由纯净天空整理自Github/MSDocs等开源代码及文档管理平台，相关代码片段筛选自各路编程大神贡献的开源项目，源码版权归原作者所有，传播和使用请参考对应项目的License；未经允许，请勿转载。