How to use extern for declaring/defining global variable in C++ and CUDA

Question

I have the following code structure composed of one .cpp, one .cu and one .hxx

UTILITIES.hxx

#ifndef UTILITIES_HXX
#define UTILITIES_HXX

namespace B{

extern int doors;
}

FILE2.cu

#include "utilities.hxx"

namespace A {

int foo (){
    switch(B::doors){
    //do something
    }
}
}

FILE3.cxx

#include "utilities.hxx"

namespace B{

int doors=-1;

class Vehicle{
public:
  void operation() {
      doors++;
      A::foo();
      doors++;
      A::foo();
  }
}
}

I am declaring the doors variable as extern in the header and I am defining it in the .cxx file. So after that, the second .cpp should be able to use it. However I am getting the following error when linking:

/usr/bin/ld: ../src/libapp.a(FILE2.cu.o): in function A::foo(void)': /MYPATH/FILE2.cu:19: undefined reference to B::doors'

What am I doing wrong? Actually the foo function in the FILE2.cu is a normal C++ function, no CUDA involved at all.

I have also used the same namespace in the three files, just to verify that it was not related to that, but I got the same linking error. — alvarella, Jun 16 '22 at 23:33
I would guess the library you built is broken, but that’s a guess because there is a lot of hand waving between three code snippets and a linker error — talonmies, Jun 16 '22 at 23:47

score 1 · Accepted Answer · answered Jun 17 '22 at 03:08

missing #endif, missing return statement, no prototype for A::foo(), missing semicolon

These changes seem to work for me:

$ cat utilities.hxx
#ifndef UTILITIES_HXX
#define UTILITIES_HXX

namespace B{

extern int doors;
}
#endif
$ cat file2.h

namespace A {

int foo ();
}
$ cat file2.cu
#include "utilities.hxx"

namespace A {

int foo (){
    switch(B::doors){
    //do something
    }
    return 0;
}
}
$ cat file3.cpp
#include "utilities.hxx"
#include "file2.h"
namespace B{

int doors=-1;

class Vehicle{
public:
  void operation() {
      doors++;
      A::foo();
      doors++;
      A::foo();
  }
};
}

$ nvcc -shared file2.cu file3.cpp -Xcompiler -fPIC
$ nvcc -lib file2.cu file3.cpp 
$

I can only work with what you have shown.

How to use extern for declaring/defining global variable in C++ and CUDA

1 Answers1