模板分文件编写，CUDA打印

ifndef FUN_HPP

define FUN_HPP

if defined(USE_EXPORT)

define EXPORT export

else

define EXPORT

endif

EXPORT
template
void print_typeof(T const&);

if !defined(USE_EXPORT)

include"fun.cpp"

endif // !ined(USE_EXPORT)

endif // FUN_HPP

可以灵活使用，不过有的编译器不支持。
export不能和inline同时使用
一个.hpp，另一个.hpp实例化即可
一个.hpp,一个.cpp，包含.cpp即可
一个.h,一个.hpp,一个.cpp初始化，主函数调用；
//.cu

include <stdio.h>

include <time.h>

include

global void HelloFromGPU(void)
{
printf("Hello from GPU\n");
}
void HelloCpu()
{

for (int i = 0; i < 10; i++)
{
    printf("i = %d\n", i);
}

}
int main(void)
{

pragma omp parallel for

for (int i = 0; i < 2; i++)
{
    HelloCpu();
}
clock_t start, start2, finish, finish2;
double totaltime;
double totaltime2;
start = clock();

HelloCpu();
finish = clock();
totaltime = (double)(finish - start) / CLOCKS_PER_SEC;
std::cout << "\n此程序的运行时间为" << totaltime << "秒！" << std::endl;
start2 = clock();

HelloFromGPU << <1, 5 >> > ();
finish2 = clock();
cudaDeviceReset();

totaltime2 = (double)(finish2 - start2) / CLOCKS_PER_SEC;
std::cout << "\n此程序的运行时间为" << totaltime2 << "秒！" << std::endl;
return 0;

}