cuda file did not link with function defined in C file

Question

I have a cuda file test.cu that include a file cuda.h.

the cuda.h contains the following function defintion used in test.cu.

extern void check_error(cudaError_t status);

this function is defined in cuda.c as follow:

void check_error(cudaError_t status)
{
    cudaError_t status2 = cudaGetLastError();
    if (status != cudaSuccess)
    {   
        const char *s = cudaGetErrorString(status);
        char buffer[256];
        printf("CUDA Error: %s\n", s);
        assert(0);
        snprintf(buffer, 256, "CUDA Error: %s", s);
        error(buffer);
    } 
    if (status2 != cudaSuccess)
    {   
        const char *s = cudaGetErrorString(status);
        char buffer[256];
        printf("CUDA Error Prev: %s\n", s);
        assert(0);
        snprintf(buffer, 256, "CUDA Error Prev: %s", s);
        error(buffer);
    } 
}

I use Visual studio 2015 for compiling. cuda.c is compiled as a C file.

There is no Compilation errors. But I get the following linkage error:

test.cu.obj : error LNK2001: unresolved external symbol "void __cdecl check_error(enum cudaError)" (?check_error@@YAXW4cudaError@@@Z)

How to solve this error?

this is not a duplicate of Name mangling in CUDA and C++ because it ask about the reverse order. Call a C function from a Cuda code. in the above question it is for calling a cuda fuction from a C file.

Possible duplicate of [Name mangling in CUDA and C++](http://stackoverflow.com/questions/28782467/name-mangling-in-cuda-and-c) — Klas Lindbäck, May 26 '16 at 13:55
Did you try changing the extension to `.cpp`? CUDA source goes through nvcc AND cc (VS compiler), and can be either C or C++. — 3Dave, May 26 '16 at 14:07
@ProEns08 Yes, I did. The questions look similar enough to be dupes. Don't you agree? — Klas Lindbäck, May 26 '16 at 14:20
@KlasLindbäck: I explain in the question the difference (in the last part). — ProEns08, May 26 '16 at 14:28

Klas Lindbäck · Accepted Answer · 2016-05-26T13:44:26.603

2

When compiling C there is no name mangling.

When you compile C++ you need to turn off name mangling for the declaration with extern "C":

#ifdef __cplusplus
extern "C" {
#endif
    extern void check_error(cudaError_t status);
#ifdef __cplusplus
}
#endif

edited May 26 '16 at 13:44

answered May 26 '16 at 13:42

Klas Lindbäck

33,105
5
57
82

It resolve the issue. Apparently, cuda code is compiled as C code. Thanks, +1. – ProEns08 May 26 '16 at 13:53
Why, then they require C name and dont support C++ name mangling? – ProEns08 May 26 '16 at 13:54
In my case, I was compiling code (C++/class/templates etc) in .cu file. I was not mentioning nvcc compiler in visual studio. As a result the .cu files were not compiled and I was getting a linker error. I used custom build option to solve this issue. Right click project -> Build Dependencies -> Build Customizations -> select CUDA targets. – Saleh Aug 22 '18 at 10:41

cuda file did not link with function defined in C file

1 Answers1