Tutorial: Introduction to CUDA Fortran | GTC 2013

Separate Compilation
module kernel_m
contains
attributes(global) subroutine difference(a,b)
use devFunc_m
integer :: a, b, c
c = b
call negate(c)
a = a+c
end subroutine difference
end module kernel_m
$ pgf90 -Mcuda=rdc -c devFunc.cuf
$ pgf90 -Mcuda=rdc -c kernel.cuf
$ pgf90 -Mcuda=rdc aMinusB.cuf devFunc.o kernel.o
module devFunc_m
contains
attributes(device) subroutine negate(a)
integer :: a
a = -a
end subroutine negate
end module devFunc_m
program aMinusB
use kernel_m
integer :: a
integer, device :: a_d, b_d
a_d = 2; b_d = 3
call difference(a_d, b_d)
a = a_d
write(*,"('2-3=',i0)") a
end program aMinusB