#include <stdio.h>
#include <stdlib.h>

__global__ void mykernel (int *a, int *b, int *c)
{
        //printf("Hi, from device.\n");
        *c = *a + *b;
}

int main(void)
{
        printf("Hi, from host.\n");
        int a=3,b=6,c;
        int *d_a,*d_b,*d_c;
        cudaMalloc((void **)&d_a,sizeof(int));
        cudaMalloc((void **)&d_b,sizeof(int));
        cudaMalloc((void **)&d_c,sizeof(int));
        printf("Here 1\n");
        cudaMemcpy(d_a,&a,sizeof(int),cudaMemcpyHostToDevice);
    cudaMemcpy(d_b,&b,sizeof(int),cudaMemcpyHostToDevice);
        printf("d_a = %d, a = %d\n",*d_a,a);
        //mykernel<<<1,1>>>(d_a,d_b,d_c);
        cudaMemcpy(&c,d_c,sizeof(int),cudaMemcpyHostToDevice);

        printf("Yolo %d %d\n", c, *d_c);

        cudaFree(d_a); cudaFree(d_b); cudaFree(d_c);
        return 0;
}