如何使用extern CUDA设备变量 [英] How to use extern cuda device variables
问题描述
我需要写code分成几个.CU文件。但我应该在哪里定义设备变量,是使用了很多.CU文件。
I need to write the code into several .cu files. But where should I define the device variables which are use for many .cu files.
一个例子
文件COMMON.H
File common.h
__device__ int x;
文件A.cu
__global__ void a()
文件B.cu
__global__ void b()
A(),B()都使用X。我该怎么办?
a(),b() both use x. what should I do?
在C语言中,我喜欢写东西
EXTERN 设备 INT X;
然后,我定义的设备 INT中X另一个地方。但在CUDA我不能这样做。如果我这样做,它告诉我'..........'$ P $这里pviously声明
In C language, I should write something like extern device int x; Then I define device int x in another place. But in CUDA I can not do it. If I do, it tells me ‘..........’ previously declared here
推荐答案
修改:@talonmies是正确的(像往常一样)。所以,我已经删除了我的意见约4.1 CUDA
EDIT : @talonmies was right (as usual). So I've deleted my comment about CUDA 4.1
此外,我给了编译命令是不完全正确。因此,让我有一个demonstrably工作并具有适当的指示代替我的答案。
Furthermore the compiling commands I gave were not quite right. So let me replace my answer with one that demonstrably works and has the proper instructions.
的您需要CUDA 5.0和计算能力2.0或为此工作更多的设备。的
我敢肯定,可能有更好的方法,但是这似乎为我工作:
I'm sure there's probably a better way, but this seems to work for me:
com.h:
#ifndef DEVMAIN
extern __device__ int x;
#endif
a.cu:
a.cu:
#include "com.h"
__global__ void a(){
x = -5;
}
b.cu:
#include <stdio.h>
#define DEVMAIN
#include "com.h"
extern __global__ void a();
__device__ int x;
__global__ void b(){
x = 5;
}
int main() {
int temp=7;
cudaMemcpyToSymbol(x,&temp, sizeof(int));
a<<<1,1>>>();
cudaMemcpyFromSymbol(&temp,x,sizeof(int));
printf("in host : %d\n",temp);
b<<<1,1>>>();
cudaMemcpyFromSymbol(&temp,x,sizeof(int));
printf("in host2 : %d\n",temp);
return 0;
}
编译:
nvcc -arch=sm_20 -dc a.cu
nvcc -arch=sm_20 -dc b.cu
nvcc -arch=sm_20 -o ab a.o b.o
输出:
$ ./ab
in host : -5
in host2 : 5
$
对不起,我的previous错误。
Sorry for my previous errors.
这篇关于如何使用extern CUDA设备变量的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!