Why won't nvcc show symbols? (newbie problem)

Accelerated Computing CUDA CUDA Programming and Performance
#1

I have two files:

main.c:

[codebox]void myfunc(void);

int

main(void)

{

myfunc();

return;

}[/codebox]

and cuda.cu:

[codebox]void

myfunc(void)

{

return;

}

[/codebox]

I compile them like so:

nvcc -c cuda.cu

gcc -c main.c

(Those works fine. The next one doesn’t)

gcc -L/usr/local/cuda/lib -lcudart -o program main.o cuda.o

The last line, the linking, gives me this:

main.o: In function `main’:

main.c:(.text+0x12): undefined reference to `myfunc’

collect2: ld returned 1 exit status

Had cuda.cu been a regular c file compiled with gcc, this would’ve worked and main() would’ve seen myfunc, but with nvcc this isn’t the case. What am I doing wrong?

Note that I need to build main.c with gcc and I also need to link with gcc, so replacing gcc with nvcc is not an option right now.

#2

you need to declare myfunc as extern “C”

#3

Be careful.

Intel works with odd address, Nvidia GPU not.

struct {

int32_t i;

int64_t ii;

}

will have different sizes and differen offset of “ii” in your compilation.

nvcc knows about it and calls gcc with -malign-double.

Yes, nvcc calls gcc - it’s secret knowledge ;)

nvcc is just a script.

I’m waiting for someone who rewrite it in shell - for better clearance of process.

Regards,

    Sergiy
#4

Hmmm… but as long as I just use basic types, I should be fine, right?

Thanks for the tip, both of you!