How to deal with arrays of different sizes in a kernel

Accelerated Computing CUDA CUDA Programming and Performance
1

I am reading sample code of CUDA kernels that deal with arrays, and they usually are like this

#define N 50

__global__ void doThis(int *array){
   int tid= blockIdx.x;
  if (tid<N)
     //do something with a[tid]
}

which is ok when you know that your array has N elements.
But how can I do when the number of elements is not known a priori, but just when you run the code. How the kernel know the number of elements.

My solution would be to pass it as an argument but that implies:

  • calculate the number of elements (in a host variable)
  • copy this value to a device variable
  • pass this variable as an argument to the kernel

Is this the only way?

2

That is pretty much the only way.

Never mind CUDA. What options do you have in C or C++ to have a function call know the size of an array without passing it as an argument to the function? The only thing I can think of is a global variable, and CUDA has something similar, but it is not any more efficient than passing the size as a function argument.