问cuda统计和的过程中使用for循环导致输出结果全0？

#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <stdio.h>
#include "device_functions.h"
#include <omp.h>		//计时头文件
#include <fstream>		//写入文件的头文件
#include <random>
#include<array>
#include<iostream>
#define Num  10
using namespace std;
__global__   void test(int* a)
{
	int ID_Block = threadIdx.x;
	int temp = 0;
	if (ID_Block < 5)
	{
		for (int i=0; i < 2; i++)
		{
			temp += a[ID_Block + i << 5];
		}
		printf("%d:%d\n", ID_Block, temp);
		a[ID_Block] = temp;
		//a[ID_Block + 5] += a[ID_Block];
	}
	__syncthreads();
	if (ID_Block == 0)
	{
		temp = 0;
	__volatile	for (int i = 0; i < 5; i++)
		{
			temp += a[i];
		}
		a[0] = temp;
		printf("%d\n", a[0]);
	}
	return;
}
int main(int argc, char** argv)
{
	 int* D;
	int h_D[10];
	for (int i = 0; i < 10; i++)
	{
		h_D[i] = i;
	}
	//****************************************
	//静态内存数组变量的传递测试
	cudaMalloc((void**)&D, Num * sizeof(int));
	cudaMemcpy(D, h_D, Num * sizeof(int), cudaMemcpyHostToDevice);
	test << <1, 5 >> > (D);
	cudaMemcpy(h_D, D, Num * sizeof(int), cudaMemcpyDeviceToHost);
	for (int i = 0; i < 10; i++)
	{
		cout << i << ":\t" << h_D[i] << endl;
	}
	return 0;
}