Problem compiling project

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, I’m trying to compile a project I’m working on, it has both .cu and .c files.
When I was using only .cu files, there was no problem, but when I’ve started adding .c files, I simply cannot compile.
Here’s the output of nsight compile line:

**** Build of configuration Debug for project cuSimpleDCT ****

make all 
Building file: ../src/ca.cu
Invoking: NVCC Compiler
nvcc -G -g -O0 -gencode arch=compute_20,code=sm_20 -odir "src" -M -o "src/ca.d" "../src/ca.cu"
nvcc --compile -G -O0 -g -gencode arch=compute_20,code=compute_20 -gencode arch=compute_20,code=sm_20  -x cu -o  "src/ca.o" "../src/ca.cu"
Finished building: ../src/ca.cu
 
Building file: ../src/ga.c
Invoking: NVCC Compiler
nvcc -G -g -O0 -gencode arch=compute_20,code=sm_20 -odir "src" -M -o "src/ga.d" "../src/ga.c"
nvcc -G -g -O0 --compile  -x c -o  "src/ga.o" "../src/ga.c"
In file included from ../src/ga.h:20,
                 from ../src/ga.c:17:
../src/ca.h:59: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'void'
../src/ca.h:68: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'void'
../src/ca.h:75: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'void'
../src/ca.h:83: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'void'
make: *** [src/ga.o] Error 1

**** Build Finished ****

Can anyone help me out?

Thanks.

2

And there is no problem when You simply change the ga.c to ga.cu, right? Tried doing it this way?

3

The problem continues, just tried that.
I’ve realised that, if for instance I have these two files

//utils.h
int test(void);

//utils.c
int test(void){
  printf("Hello
");
  return 0;
}

and try to use the function test() in another file, say:

//main.c
#include "utils.h"
int main(int argc, char *argv[]){
  test();
  return 0;
}

I have similar problems, which shouldn’t happen right?

4

Have You tried to compile the .c files separately, with diffrent compiler (e.g. GCC)?

5

Yep, and it worked like a charm…strange thing…

6

NVCC must have been lacking something. I think it needs all the CUDA includes to even compile the source given, but this I am not sure. NVCC manual can be helpful in such matter :)

Regards,
MK

7

So, I think I’ve got it sorta working out, I had some static functions giving me trouble.
However, now I can only compile .cu files, not .c, is this supposed to be like this?

Thanks

8

I could build your sample code by nvcc. What is the toolkit version you use? Is it CUDA 5.0 Production?

9

sjiagc, yes it is.
But I’ve fixed the sample problem, the issue now is with .c and .cu…

10

The cu file is actually a cpp file. So you need to add extern “C” before C function.

11

I see…would it be bether to make the code always in C++?
I mean for performance and build issues…

12

Can you check …/src/ca.h:59? It looks like you have CUDA C-specific construct there (e.g. “device” specifier) or something C+±specific.

Note that Nsight specifically sets code language to C in your case (see “-x c”). It may happen that GCC invoked from the command line without such switch autodetects your code as C++ and hence is able to compile it.