C++AMP vector_add

main.cpp

// export PATH=${PATH}:/opr/rocm/bin
// hcc `hcc-config --cxxflags` -g -O3  -o obj/main.o -c main.cpp
// hcc -o vector_add obj/main.o `hcc-config --ldflags`

#include <iostream>
#include <algorithm>
#include <numeric>
#include <array>
#include <amp.h>

using Device = concurrency::accelerator;

void SetDefaultDevice()
{
        const auto devices = concurrency::accelerator::get_all();
        const auto device = std::find_if(devices.crbegin(), devices.crend(), [](const Device& d)
        {
                return !(d.get_is_emulated());
        });

        if(device == devices.crend())
        {
                throw "Useful device not found";
        }

        std::wcout << device->get_description() << std::endl;
        concurrency::accelerator::set_default(device->get_device_path());
}

int main()
{
        constexpr unsigned int n = 10000;

        std::vector<int> a(n);
        std::vector<int> b(n);
        std::vector<int> c(n);
        std::iota(a.begin(), a.end(), 0);
        std::iota(b.rbegin(), b.rend(), n);

        SetDefaultDevice();

        concurrency::array_view<int> aa(n, a);
        concurrency::array_view<int> bb(n, b);
        concurrency::array_view<int> cc(n, c);

        concurrency::parallel_for_each(
                aa.get_extent(),
                [aa, bb, &cc] (const concurrency::index<1> idx) restrict(amp)
                {
                        const int i = idx[0];
                        cc[i] = aa[i] + bb[i];
                });

        cc.synchronize();

        for(int i = 0; i < n; i++)
        {
                if(c[i] != a[i] + b[i])
                {
                        std::cout << i << ": actual=" << c[i] << ", expected="<<  a[i] + b[i] << std::endl;
                        return -1;
                }
        }
        std::cout << "OK" << std::endl;

        return 0;
}

At line #48, aa, bb, cc should all be captured by value. cc is currently captured by reference, it would cause trouble on Linux with dGPU.