1.     #include <stdio.h>
2.     #include <stdlib.h>
3.     #include <string.h>
4.     #include <cuda.h>
5.     #include <cuda_runtime.h>
6.     
7.     
8.     // simple kernel function that adds two vectors
9.     __global__ void vect_add(float *a, float *b, int N)
10.    {
11.       int idx = threadIdx.x;
12.       if (idx<N) a[idx] = a[idx] + b[idx];
13.    }
14.    
15.    // function called from main fortran program
16.    extern "C" void kernel_wrapper_(float *a, float *b, int *Np)
17.    {
18.       float  *a_d, *b_d;  // declare GPU vector copies
19.       
20.       int blocks = 1;     // uses 1 block of
21.       int N = *Np;        // N threads on GPU
22.    
23.       // Allocate memory on GPU
24.       cudaMalloc( (void **)&a_d, sizeof(float) * N );
25.       cudaMalloc( (void **)&b_d, sizeof(float) * N );
26.    
27.       // copy vectors from CPU to GPU
28.       cudaMemcpy( a_d, a, sizeof(float) * N, cudaMemcpyHostToDevice );
29.       cudaMemcpy( b_d, b, sizeof(float) * N, cudaMemcpyHostToDevice );
30.    
31.       // call function on GPU
32.       vect_add<<< blocks, N >>>( a_d, b_d, N);
33.    
34.       // copy vectors back from GPU to CPU
35.       cudaMemcpy( a, a_d, sizeof(float) * N, cudaMemcpyDeviceToHost );
36.       cudaMemcpy( b, b_d, sizeof(float) * N, cudaMemcpyDeviceToHost );
37.    
38.       // free GPU memory
39.       cudaFree(a_d);
40.       cudaFree(b_d);
41.       return;
42.    }