Build CUDA kernel with Meson

Question

I am trying to build an application that makes use of a Cuda kernel. For that, I am using the meson build system but without success. Basically what I am trying to do is the following:

//main.cpp

extern void kernel_function();

int main(int argc, char *argv[]){

  // some logic here...

  kernel_function();
  return 0;
}

//kernel.cu

__global__ void kernel(){
  // some code here...
}

void kernel_function(){

  dim3 threads( 2, 1 );
  dim3 blocks( 1, 1 );

  kernel<<< blocks, threads >>>();
}

I can compile the above code with the following commands:

g++ -c main.cpp
nvcc -c kernel.cu
nvcc -o main main.o kernel.o

How can I replicate this compilation process that I do on the terminal with Meson?

Answer 1

In the simplest case it could be something like:

project('cuda dependency', 'cpp', 'cuda')
executable('main', 'main.cpp', 'kernel.cu')

And run with

$ CXX=g++ meson build
$ ninja -C build

(based on example).

But down the road you might need to add dependencies like

dep = dependency('cuda', version : '>=10', modules : ['cublas'])
executable(..., dependencies: dep)

or set some special compiler flags with the help of cuda module, check this reference on meson.