Question

#include <cuda.h>
#include <cuda_runtime.h>
#include <iostream>
#include <vector>
using namespace std;

__global__ void kernel(vector <int> *a,vector <int> *b, vector <int> *c)
{
    int index=threadIdx.x;
    c[index]=a[index]+b[index];
};

int main ()
{
    //Paar constanten initializeren
    int N=3;
    vector <int> a (3,1);
    vector <int> b (3,1);
    vector <int> c (3,0);
    int size=N*sizeof(vector <int>);
    //Device variabelen
    vector <int> d_a;
    vector <int> d_b;
    vector <int> d_c;

    //<size> bytes device-geheugen vrijmaken op locatie &a
    cudaMalloc( (void **) &a, size);
    cudaMalloc ( (void **) &b, size);
    cudaMalloc ( (void **) &c, size);

    //Kopieer inputs van &a naar &d_a
    cudaMemcpy(&d_a,&a,size,cudaMemcpyHostToDevice);
    cudaMemcpy(&d_b,&b,size,cudaMemcpyHostToDevice);

    printf( (char *) a[1]);
    printf( (char *) c[1]);
    //Voer kernel uit

    kernel<<<N,1>>>(&d_a,&d_b,&d_c);

    //Kopieer resultaat terug naar host 
    cudaMemcpy(&c,&d_c,size,cudaMemcpyDeviceToHost);

    printf( (char *) c[1]);

    return 0;
};

So this is my basic C++ vector addition program, though it won’t compile because it doesn’t know what to do with the “+” when adding indexes (“no operator “+” matches these operands”). Anyone have an idea why this doesn’t work inside the kernel? I think it may have to do with the fact that it’s a host function or something like that, though I wouldn’t know how to get around that problem.

Greetings and thanks in advance.

Answer 1

You can’t use the STL in device code … you’re going to have to use an array or some other CUDA construct like the Thrust Library.