module thrust

interface thrustmin

    function min_float(input,N) bind(C,name="min_float_wrapper")
    use iso_c_binding
    real(c_float),device:: input(*)
    integer(c_int),value:: N, min_float
    end function

    function min_double(input,N) bind(C,name="min_double_wrapper")
    use iso_c_binding
    real(c_double),device:: input(*)
    integer(c_int),value:: N, min_double
    end function

!   subroutine min_double(input,N) bind(C,name="min_double_wrapper")
!   use iso_c_binding
!   real(c_double),device:: input(*)
!   integer(c_int),value:: N
!   end subroutine

end interface

end module thrust


program GPU
  use cudafor
  use thrust
  implicit none
  real, allocatable, device :: dev_array(:)
  real, allocatable :: host_array(:)
  integer N / 20 /
  integer :: x

  allocate(dev_array(N))
  allocate(host_array(N))

  call random_seed
  call random_number(host_array)

  print *, "host_array:"
  print *, host_array


  print *, "COPY!"
  dev_array = host_array

  print *, "START CUDA!"
  x = thrustmin(dev_array, size(dev_array))
  print *, "DONE  CUDA!"

  print *, "Min:", x

  deallocate(dev_array)
  deallocate(host_array)

  end program GPU