main.cpp

#include <amp.h>
#include <vector>
#include <iostream>
#include <chrono>

int main()
{
    using namespace std;
    using namespace std::chrono;
    using namespace concurrency;

    constexpr int N = 1024;
    vector<float> vecA(N*N), vecB(N*N), vecC(N*N);

    concurrency::extent<2> size(N, N);

    array_view<const float, 2> A(size, vecA.data());
    array_view<const float, 2> B(size, vecB.data());
    array_view<float, 2> C(size, vecC.data());

    auto before = high_resolution_clock::now();

    parallel_for_each(C.extent, [=](index<2> idx) restrict(amp)
    {
        const int y = idx[0]; 
        const int x = idx[1];
        float sum = 0;
        for (int k = 0; k < N; ++k)
#if 1
            sum += A(y, k) * B(k, x);
#else
            sum += A(y, k) * B(x, k);
#endif
        C[idx] = sum;
    });

    auto mid = high_resolution_clock::now();

    C.synchronize();
    
    auto after = high_resolution_clock::now();

    cout << duration_cast<duration<float, milli>>(mid-before).count() << "ms" << endl;
    cout << duration_cast<duration<float, milli>>(after-mid).count() << "ms" << endl;
}