Observed issues › fresh/cce19-acc-automatic-array-copyout-copy-crash

Appears to behave incorrectlyCrashes or aborts at run time · Fortran · OpenACC · Fixed in 20.0.2

ftn 19 (OpenACC): a two-dimensional automatic array in copyout crashes the run in copy mode

Legality: independently confirmed. 2 independent readings find the program legal and the result required. How this is checked

Affected variants

Compiler	MI300A gfx942 unified memory (XNACK 1)	MI300A gfx942 copy (XNACK 0)	MI250X gfx90a copy (XNACK 0)	MI300X gfx942 copy (XNACK 0)	MI350X gfx950 copy (XNACK 0)	A100 sm_80 separate memory	A100 sm_80 managed memory
AFAR 24.3.0	n/a	n/a	n/a	n/a	n/a	n/a	n/a
CCE 19.0.0	correct	behaves incorrectly	behaves incorrectly	unknown not run	unknown not run	n/a	n/a
CCE 20.0.2	correct	correct	correct	unknown not run	unknown not run	n/a	n/a
CCE 21.0.3	correct	correct	correct	unknown not run	unknown not run	n/a	n/a
CCE 21.0.4	correct	correct	correct	unknown not run	unknown not run	n/a	n/a
NVHPC 25.11	n/a	n/a	n/a	n/a	n/a	correct	correct
NVHPC 26.9	n/a	n/a	n/a	n/a	n/a	correct	correct
GCC 16.2.0	correct	correct	correct	correct	correct	correct	n/a

"behaves incorrectly" / "not supported": this issue, confirmed in up to four environments (another build directory, another node, a repeat). "fails another way": a different failure, not counted as this issue. "unknown": not run there, or the runs disagree.

Messages

Result	Compiler and GPU	Message
behaves incorrectly	CCE 19.0.0, MI300A copy CCE 19.0.0, MI250X	show RUN_FAIL

Reproduce

CCE 19.0.0, MI300A copy

module load cce/19.0.0 craype-accel-amd-gfx942 rocm/6.4.3
module unload cray-mpich
ftn -h acc -O2 repro.f90 -o repro
HSA_XNACK=0 ./repro

NVHPC 25.11, A100 separate (runs correctly)

export LD_LIBRARY_PATH=$NVHPC_25_11/compilers/lib
nvfortran -acc=gpu -gpu=cc80 -gpu=mem:separate -O2 repro.f90 -o repro
./repro

Program (51 lines)

! # Finding (from the data/bench/fresh draft fresh-084): CCE 19 crashes on a two-dimensional
!   automatic array in copyout in copy mode (OpenACC)
!
! - source: data/bench/fresh/fresh-084.f90 (reconstructed from the ICON workaround at
!   src/atm_dyn_iconam/mo_nh_supervise.f90, icon-model 8597da45, line 765: an automatic array in
!   copyout, zeroed in kernels and filled in a parallel loop). Neither the kernels region nor the
!   copyin array matters; the automatic array's rank does (the one-dimensional form,
!   data/reports/fresh-drafts/d084-min.f90, is correct)
! - compilers: CCE 19.0.0 ftn -h acc -O2; MI300A HSA_XNACK=0, MI250X (Tioga) HSA_XNACK=0
! - expected: s 3150.0. copyout(w) on the data construct allocates w on the device at entry and
!   transfers it out at exit when the reference counters reach zero (OpenACC 3.4 §2.7.9 p.56
!   l.1932-1953); the parallel loop writes every element, so w(i, k) = i*k on the host after the
!   region and the sum over i = 1..20, k = 1..5 is 210 * 15 = 3150
! - actual: SIGSEGV (exit -11), nothing printed, three runs of three; at HSA_XNACK=1 correct
! - control (same compiler): w a fixed-size local w(20, 5) instead of the automatic w(nb, nl)
!   prints s 3150.0 at XNACK 0 and 1 (data/reports/fresh-drafts/d084-min2d-fixed.f90); so do an
!   allocatable vn_aux in the draft (d084-allocatable.f90) and the one-dimensional automatic array
!   (d084-min.f90)
! - other families: CCE 20.0.2 and 21.0.3 print s 3150.0 (both modes); NVHPC 25.11 and 26.9 print
!   s 3150.0 on this program and run the draft correctly (A100, separate and managed
!   memory); AFAR has no Fortran OpenACC
! - legality: legal by reading
! - evidence: results/fresh-drafts/tuo-p2.jsonl, tuo2-q2.jsonl, tuo3-r*-b.jsonl, tioga/tioga-b.jsonl,
!   wingtip/nv.jsonl, wingtip/nv4.jsonl (review)
module m
  implicit none
contains
  subroutine fill(nb, nl, s)
    integer, intent(in) :: nb, nl
    real(8), intent(out) :: s
    real(8) :: w(nb, nl)
    integer :: i, k
    !$acc data copyout(w)
    !$acc parallel loop collapse(2)
    do k = 1, nl
      do i = 1, nb
        w(i, k) = i * k
      end do
    end do
    !$acc end data
    s = sum(w)
  end subroutine fill
end module m

program main
  use m
  implicit none
  real(8) :: s
  call fill(20, 5, s)
  print *, 's', s
end program main

Source in the repository

Something to add? Dispute its legality · Report it fixed in a newer release · Report a mistake on this page

Expected and observed

Expected

s 3150.

What the specification requires

The specification text this program relies on and the compiler is expected to support.

OpenACC 3.4 §2.7.9 p.56 l.1932-1953

Evidence

Same compiler, one change

w a fixed-size local array w(20, 5) instead of the automatic w(nb, nl): CCE 19.0.0: correct; CCE 20.0.2: correct; CCE 21.0.3: correct; CCE 21.0.4: correct

Control program

module m
  implicit none
contains
  subroutine fill(nb, nl, s)
    integer, intent(in) :: nb, nl
    real(8), intent(out) :: s
    real(8) :: w(20, 5)
    integer :: i, k
    !$acc data copyout(w)
    !$acc parallel loop collapse(2)
    do k = 1, nl
      do i = 1, nb
        w(i, k) = i * k
      end do
    end do
    !$acc end data
    s = sum(w)
  end subroutine fill
end module m

program main
  use m
  implicit none
  real(8) :: s
  call fill(20, 5, s)
  print *, 's', s
end program main

Other compilers run it correctly

NVHPC 25.11, NVHPC 26.9, GCC 16.2.0