Creating a .so library?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I have a myfile.cu file containing a global (device) function and a host function.

I use nvcc to create a myfile.o object file.

I would then like to be able to link this object file into a shared library, along with another object file using gcc -shared myfile.o myotherfile.o
However, when I try this, I get:

/usr/bin/ld: myfile.o: relocation R_X86_64_32 against `.nvFatBinSegment’ can not be used when making a shared object; recompile with -fPIC
myfile.o: could not read symbols: Bad value
collect2: ld returned 1 exit status

Is there a way around this (where is .nvFatBinSegment? Probably in some proprietry cuda code which I can’t recompile) - I guess not, and that I really need to use the driver API…?

Thanks…

2

Sorted: I needed a --compiler-options ‘-fPIC’ for nvcc.