cuda编程：不要在一个cpp文件中声明__device__和__global__函数

比如一下代码是正常的：

kernel.h

extern __global__ void kernel();

kernel.cu

#include <stdio.h>
#include "kernel.h"
__global__ void kernel()
{
	printf("hello world!");
}

test.cu

#include "kernel.h"
int main(void)
{
	kernel<<<1,1>>>();
	cudaDeviceSynchronize();
return 0;
}

但如果把test.cu的名字改为test.cpp，就出错了。

make all 
Building file: ../src/test2.cpp
Invoking: NVCC Compiler
/usr/local/cuda-5.5/bin/nvcc -G -g -O0 -gencode arch=compute_35,code=sm_35 -odir "src" -M -o "src/test2.d" "../src/test2.cpp"
/usr/local/cuda-5.5/bin/nvcc -G -g -O0 --compile  -x c++ -o  "src/test2.o" "../src/test2.cpp"
In file included from ../src/test2.cpp:3:0:
../src/kernel.h:5:8: 错误： ‘__global__’不是一个类型名
../src/test2.cpp: 在函数‘int main()’中:
../src/test2.cpp:10:2: 错误： ‘kernel’在此作用域中尚未声明
../src/test2.cpp:10:10: 错误： expected primary-expression before ‘<’ token
../src/test2.cpp:10:16: 错误： expected primary-expression before ‘>’ token
../src/test2.cpp:10:18: 错误： expected primary-expression before ‘)’ token
../src/test2.cpp:11:24: 错误： ‘cudaDeviceSynchronize’在此作用域中尚未声明
make: *** [src/test2.o] 错误 1

在这里找到答案

http://stackoverflow.com/questions/16550031/cuda-device-and-global-error-expected-constructor-destructor-or-type

The errors appeared because you included a header file with device code decorators (__device__ and__global__) in a .cpp file. 

The .cpp file by default gets handled directly by the host compiler, which does not know what those decorators mean.

因为cpp默认是用主机编译器即c++的编译器来处理的。

而在test.cpp中包含了kernel.h，kernel.h中声明了__global__函数。