Unable to call Device function in Cuda Fortran

Accelerated Computing CUDA CUDA Programming and Performance
2

Hi,
We don’t normally recommend developers use F90 pointers in device code. They can have very high overheads, though in your case, a pointer to a scalar type is pretty efficient. Linked lists are not good parallel data structures, and in your case here, the allocations in device code also have high overhead.
The bug though is we don’t properly support device functions which return pointers, which is what you have used. I can get the code to compile and run by changing the function to a subroutine:

ATTRIBUTES(DEVICE) SUBROUTINE InsertList(head, elem)
    IMPLICIT NONE

    type( ListElem ), pointer :: head, elem

    elem%next => head

  END SUBROUTINE InsertList
  END MODULE

MODULE Test
  CONTAINS
    ATTRIBUTES(GLOBAL) SUBROUTINE KERNEL()
    USE ListModule
    IMPLICIT NONE
    type( ListElem ), pointer :: head
    type( ListElem ), pointer :: newElem, h
    integer :: i,N = 4
    INTEGER(KIND=4),ALLOCATABLE::ND(:)

    allocate( newElem )
    allocate( head )
    newElem%value=1
    head%value=2

    PRINT*,newElem%value
    call InsertList(head, newElem)
    PRINT*,newElem%next%value

Unless you really need every CUDA thread to have its own linked list, I would recommend looking at other ways to build your data structures.