offload/test/offloading/fortran/target-depend.f90

   1 ! Offloading test checking the use of the depend clause on
   2 ! the target construct
   3 ! REQUIRES: flang, amdgcn-amd-amdhsa
   4 ! UNSUPPORTED: nvptx64-nvidia-cuda
   5 ! UNSUPPORTED: nvptx64-nvidia-cuda-LTO
   6 ! UNSUPPORTED: aarch64-unknown-linux-gnu
   7 ! UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
   8 ! UNSUPPORTED: x86_64-pc-linux-gnu
   9 ! UNSUPPORTED: x86_64-pc-linux-gnu-LTO
  10
  11 ! RUN: %libomptarget-compile-fortran-run-and-check-generic
  12 program main
  13   implicit none
  14   integer :: a = 0
  15   INTERFACE
  16      FUNCTION omp_get_device_num() BIND(C)
  17        USE, INTRINSIC :: iso_c_binding, ONLY: C_INT
  18        integer :: omp_get_device_num
  19      END FUNCTION omp_get_device_num
  20   END INTERFACE
  21
  22   call foo(5, a)
  23   print*, "======= FORTRAN Test passed! ======="
  24   print*, "foo(5) returned ", a, ", expected 6\n"
  25
  26   !       stop 0
  27   contains
  28     subroutine foo(N, r)
  29       integer, intent(in) :: N
  30       integer, intent(out) :: r
  31       integer :: z, i, accumulator
  32       z = 1
  33       accumulator = 0
  34       ! Spawn 3 threads
  35       !$omp parallel num_threads(3)
  36
  37       ! A single thread will then create two tasks - one is the 'producer' and
  38       ! potentially slower task that updates 'z' to 'N'. The second is an
  39       ! offloaded target task that increments 'z'. If the depend clauses work
  40       ! properly, the target task should wait for the 'producer' task to
  41       ! complete before incrementing 'z'. We use 'omp single' here because the
  42       ! depend clause establishes dependencies between sibling tasks only.
  43       ! This is the easiest way of creating two sibling tasks.
  44       !$omp single
  45       !$omp task depend(out: z) shared(z)
  46       do i=1, 32766
  47          ! dumb loop nest to slow down the update of 'z'.
  48          ! Adding a function call slows down the producer to the point
  49          ! that removing the depend clause from the target construct below
  50          ! frequently results in the wrong answer.
  51          accumulator = accumulator + omp_get_device_num()
  52       end do
  53       z = N
  54       !$omp end task
  55
  56       ! z is 5 now. Increment z to 6.
  57       !$omp target map(tofrom: z) depend(in:z)
  58       z = z + 1
  59       !$omp end target
  60       !$omp end single
  61       !$omp end parallel
  62       ! Use 'accumulator' so it is not optimized away by the compiler.
  63       print *, accumulator
  64       r = z
  65     end subroutine foo
  66
  67 !CHECK: ======= FORTRAN Test passed! =======
  68 !CHECK: foo(5) returned 6 , expected 6
  69 end program main