I'm facing with a problem with making .dll from .cu files with Cuda/C++. There is a .o (C++ object file) compiled by nvcc with some parameters:
nvcc -dlink -o kernel.o -rdc=true kernel.cu
and passing options to the g++ compiler:
g++ -shared kernel.o -o kernel.dll -L"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\lib\x64" -lcudart -lcuda -lcudadevrt
However, I got the following error like below:
Warning: corrupt .drectve at end of def file
C:\Program Files\mingw64\x86_64-w64-mingw32\bin\ld.exe: kernel.o:(.text$mn+0x160): undefined reference to `__fatbinwrap_......_kernel_cu_add'
collect2.exe: error: ld returned 1 exit status
My Cuda toolkit version is 12.0 and the g++ one is 12.2.0, which runs on Windows 10.
Here is my code:
kernel.cu
#include <iostream>
#include <cuda.h>
#include <cuda_runtime.h>
#include "kernel.cuh"
extern "C" __global__ void add(int n, float *a, float *b, float *sum)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
sum[i] = a[i] + b[i];
}
extern "C" __host__ void launchKernel()
{
std::cout << "function called" << std::endl; //DEBUG CODE
//Here launch kernel code.
}
kernel.cuh
extern "C" __declspec(dllexport) __global__ void add(int n, float *a, float *b, float *sum);
extern "C" __declspec(dllexport) __host__ void launchKernel();
I tried here and here but there is still a compile error. I'm beginner at Cuda/C++ so my code might looks like funny. To make .dll file with Cuda/C++, what should I do something else? Thank you.
