CODEWITHSUNDEEP
cuda
skandan
skandan

Reputation: 45

How to call a CUDA file from a C++ header file?

I know the method of calling the .cu files from .c files. But now I want to call the .cu files from a C header file. Is it possible to do it ? If so how should I make the settings of my project ? please help.....

Upvotes: 2

Views: 8020

Answers (1)

Robert Crovella
Robert Crovella

Reputation: 151799

Here's a worked example:

file1.h:

int hello();

file2.h:

#include <stdio.h>
#include "file1.h"

int myfunc(){
  hello();
  return 0;
}

file1.cu:

#include <stdio.h>
#include "file1.h"

__global__ void mykernel(){
  printf("Hello from mykernel\n");
}

int hello(){

  mykernel<<<1,1>>>();
  cudaDeviceSynchronize();
  return 0;
}

file2.cpp:

#include "file2.h"

int main(){
  myfunc();
  return 0;
}

build and test:

$ nvcc -arch=sm_20 -c file1.cu
$ g++ -c file2.cpp
$ g++ -o test file1.o file2.o -L/usr/local/cuda/lib64 -lcudart
$ ./test
Hello from mykernel
$

Assuming you are intending to include your file2.h into a cpp file, you cannot call a cuda kernel directly from that header and use it in a cpp file. You must put a wrapper around the cuda kernel, and call the wrapper, as I have indicated. This is because your cpp file will get compiled by the host compiler, which doesn't know anything about cuda syntax (e.g. mykernel<<<1,1>>>();)

Also, as indicated in the comments, it may make more sense to reserve the header file file2.h just for needed prototypes, and put the actual function definition of myfunc into a cpp file somewhere.

Upvotes: 9

Related Questions