Hi,
I am trying to call several C functions from my Fortran code which allocate device memory and then call CUDA C kernels. In order to pass the data between C functions I tried to declare pointers in Fortran of type c_devptr. I seem to have got myself in a real pointer mess with this and was wondering if anyone has done this before? If so could they share a quick example of how they did it?
Cheers,
Mark
I think you are making it a lot harder than it needs to be:
Fortran will pass assumed size arrays (using F77 calling conventions) just like C expects them. You can also pass arguments by value:
%> cat t1.cu
// Kernel that executes on the CUDA device
extern “C” global void square_array(float *a, int N)
{
int idx = blockIdx.x * blockDim.x + threadIdx.x;
if (idx<N)
a[idx] = a[idx]a[idx];
}
%> cat z1.cuf
program t
use cudafor
interface
attributes(global) subroutine square_array(a, n) bind(c)
real, device :: a()
integer, value :: n
end subroutine
end interface
real, allocatable, device :: a(:)
allocate(a(10))
a = 2.0
call square_array<<<1,10>>> (a, 10)
print *,sum(a)
end
%> nvcc -c t1.cu
%> pgf90 z1.cuf t1.o
z1.cuf:
%> ./a.out
40.00000