一般教科書都這麼寫，感覺不如STL的好。記下以便查閱。

#include <cuda_runtime.h>
#include <iostream>
#include <stdio.h>


__global__ void vector_add_gpu_2(float*d_a,float*d_b,float*d_c,int n)
{
	int tid = threadIdx.x;
	const int t_n = blockDim.x;
	while(tid < n)
	{
		d_c[tid] = d_a[tid] + d_b[tid];
		tid += t_n;
	}
}

int main()
{
	float* d_a,*d_b,*d_c,*h_a,*h_b,*h_c;
	int n = 0;
	std::cin >> n;
	int nBytes = n * sizeof(float);
	h_a = (float*)malloc(nBytes);
	h_b = (float*)malloc(nBytes);
	h_c = (float*)malloc(nBytes);
	
	for(int i = 0;i < n;i++)
	{
		h_a[i] = (float)i;
		h_b[i] = (float)i;
		std::cout << "h_a[" << i << "]= " << h_a[i] << "\t";
		std::cout << "h_b[" << i << "]= " << h_b[i] << "\n";
	}

	cudaMalloc((float**)&d_a,nBytes);
	cudaMalloc((float**)&d_b,nBytes);
	cudaMalloc((float**)&d_c,nBytes);

	cudaMemcpy(d_a,h_a,nBytes,cudaMemcpyHostToDevice);
	cudaMemcpy(d_b,h_b,nBytes,cudaMemcpyHostToDevice);
	
	vector_add_gpu_2<<<1,10>>>(d_a,d_b,d_c,n);

	cudaMemcpy(h_c,d_c,nBytes,cudaMemcpyDeviceToHost);

	for(int i = 0;i < n;i++)
	{
		std::cout << "h_c[" << i << "]= " << h_c[i] << "\t";
	}
	return 0;
}