thrust示例ex1: 求多个不同长度数组数据和

可以对多个不同长度数组进行reduce求和，所有数组都放在一个大数组里，用offsets描述每个数组的偏移和大小信息

#include 
#include 
#include 
#include 
#include 
#include 

#include 
#include 
#include 



using offset_t = int;
using float_t = float;
const unsigned int nArray = 4;

float_t my_rand(void)
{
    static thrust::default_random_engine rng;
    //static thrust::uniform_int_distribution dist(0, 9999);
    static thrust::uniform_real_distribution dist(-100.0f, 100.0f);
    return dist(rng);
}


struct MyFunctor
{
    thrust::device_ptr idata_;
    thrust::device_ptr offsets_;
    thrust::device_ptr odata_;

    __host__ __device__ void operator()(int idx)
    {
        offset_t offset = offsets_[idx];
        int n = offsets_[idx + 1] - offsets_[idx];
        odata_[idx] = thrust::reduce(thrust::cuda::par, idata_ + offset, idata_ + offset + n, float_t(0), thrust::plus());
    }
};


template
void print_array(T* data, int n, std::string name = "")
{
    std::cout << name << "n";
    thrust::copy(data, data + n, std::ostream_iterator(std::cout, "n"));
    return;
}

template
void print_array(const thrust::host_vector& v, std::string name = "")
{
    std::cout << name << "n";
    thrust::copy(v.begin(), v.end(), std::ostream_iterator(std::cout, "n"));
    return;
}


void test()
{
    thrust::host_vector h_offsets(nArray+1);
    h_offsets[0] = 4;  //#0 
    h_offsets[1] = 5;  //#1 
    h_offsets[2] = 6;  //#2
    h_offsets[3] = 7;  //#3

    
    thrust::exclusive_scan(h_offsets.begin(), h_offsets.end(),h_offsets.begin(),0);
    print_array(h_offsets,"h_offsets");

    thrust::device_vector d_offsets = h_offsets;
    
    const unsigned int nData = *(h_offsets.end() - 1);
    std::cout << "nData=" << nData << "n";

    
    thrust::host_vector h_idata(nData);
    thrust::generate(h_idata.begin(), h_idata.end(), my_rand);
    print_array(h_idata,"h_idata");

    thrust::device_vector d_idata = h_idata;


    
    thrust::host_vector h_index(nArray);
    thrust::sequence(h_index.begin(), h_index.end());
    print_array(h_index, "h_index");

    thrust::device_vector d_index = h_index;


    thrust::host_vector h_odata(nArray);
    thrust::device_vector d_odata(nArray);

    

    MyFunctor f{ d_idata.data(),d_offsets.data(),d_odata.data() };
    thrust::for_each(d_index.begin(), d_index.end(), f);
    h_odata = d_odata;
    std::cout << "GPU: reduce sumn";
    print_array(h_odata, "h_odata");
}

int main(int argc, char **argv)
{

    test();

    return 0;
}

• how-to-use-thrust-to-sort-the-rows-of-a-matrix
• device_vector获取底层指针
• thrust in kernel
• using-thrust-functions-within-device-code-cudalaunchcooperativekernel
• Add thrust::cuda::par_nosync execution policy
• tutorial_gpu_thrust_intero