nvfortran -mp=gpu: fort2 crashes (signal 11) on a whole-array assignment to a module allocatable in target teams loop

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	correct	correct	unknown not run	correct	correct	n/a	n/a
CCE 19.0.0	correct	correct	unknown not run	unknown not run	unknown not run	n/a	n/a
CCE 20.0.2	correct	correct	unknown not run	unknown not run	unknown not run	n/a	n/a
CCE 21.0.3	correct	correct	unknown not run	unknown not run	unknown not run	n/a	n/a
CCE 21.0.4	correct	correct	unknown not run	unknown not run	unknown not run	n/a	n/a
NVHPC 25.11	n/a	n/a	n/a	n/a	n/a	behaves incorrectly	behaves incorrectly
NVHPC 26.9	n/a	n/a	n/a	n/a	n/a	behaves incorrectly	behaves incorrectly
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	NVHPC 25.11, A100 separate NVHPC 25.11, A100 managed	show nvfortran-Fatal-$NVHPC_25_11/compilers/bin/tools/fort2 TERMINATED by signal 11
behaves incorrectly	NVHPC 26.9, A100 separate NVHPC 26.9, A100 managed	show nvfortran-Fatal-$NVHPC_26_9/compilers/bin/tools/fort2 TERMINATED by signal 11

Reproduce

NVHPC 25.11, A100 separate

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

NVHPC 26.9, A100 separate

export LD_LIBRARY_PATH=$NVHPC_26_9/compilers/lib
nvfortran -mp=gpu -gpu=cc80 -gpu=mem:separate -O2 repro.f90 -o repro
OMP_TARGET_OFFLOAD=mandatory ./repro

AFAR 24.3.0, MI300A unified (runs correctly)

$AFAR/bin/amdflang -fopenmp --offload-arch=gfx942 -O2 repro.f90 -o repro
HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro

Program (66 lines)

! # Finding (from the data/bench/fresh draft fresh-058): nvfortran's fort2 crashes on a
!   whole-array assignment to a module allocatable in a target teams loop
!
! - source: data/bench/fresh/fresh-058.f90 (NVIDIA developer forum topic 294712, the reproducer of
!   the 2024-06-05 reply, TPR 35799). The reporter's loop runs 10000 iterations that each assign
!   all of A, a data race (every iteration writes the same elements); here the loop has one
!   iteration, so the program is race-free, and the crash is the same
! - compilers: NVHPC nvfortran 25.11 and 26.9, -mp=gpu -gpu=cc80 -O2 (separate and managed memory
!   builds)
! - expected: the program compiles and prints 7. A, an allocatable module array referenced in the
!   target region, is not a scalar, so it is treated as map(tofrom) (OpenMP 5.2 §5.8.1 p.149
!   l.4-5); Z, a scalar, is firstprivate; the single iteration assigns 7 to every element
! - actual: "nvfortran-Fatal-.../maintainers/fort2 TERMINATED by signal 11", both releases, both
!   memory-mode builds (the reporter's racy form too)
! - control (same compiler): the same assignment compiles and prints 7 when A is a fixed-size
!   module array (data/reports/fresh-drafts/d058-fixed.f90), a local allocatable (d058-local.f90),
!   in a plain target region (d058-target.f90), in target teams distribute parallel do
!   (d058-tdpd.f90), or as the element assignment A(II) = Z (d058-element.f90)
! - other families: AFAR and CCE 19/20/21 compile this program and print 7 (MI300A, HSA_XNACK 1
!   and 0; MI250X)
! - legality: legal by reading: one iteration, so no two iterations touch A; the loop construct's
!   restrictions (5.2 §11.7 p.257 l.29-34, p.258 l.1-4) and its implicit order(concurrent)
!   (§10.3 p.234 l.14-24) allow the assignment
! - evidence: results/fresh-drafts/wingtip/nv.jsonl, tuo-p4.jsonl, tioga/tioga-d.jsonl (review)
MODULE GLOBALS
   IMPLICIT NONE
   REAL :: Z
   REAL, ALLOCATABLE, DIMENSION(:)::A

END MODULE GLOBALS

MODULE RECON
   USE GLOBALS

   IMPLICIT NONE

CONTAINS

   SUBROUTINE CRASH_REPRO()

      IMPLICIT NONE
      INTEGER::II

      ALLOCATE (A(10))
      Z = 7

      !$OMP target teams loop
      DO II = 1, 1
         A = Z;
      END DO
      !$OMP end target teams loop

      PRINT *, A(1)

   END SUBROUTINE CRASH_REPRO

END MODULE RECON

PROGRAM compilerCrashTest

   USE RECON
   IMPLICIT NONE

   CALL CRASH_REPRO()

END PROGRAM compilerCrashTest

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

Observed where it behaves incorrectly

nvfortran-Fatal-$NVHPC_25_11/compilers/bin/tools/fort2 TERMINATED by signal 11

What the specification requires

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

OpenMP 5.2 §5.8.1 p.149 l.4-5

OpenMP 5.2 §11.7 p.257 l.29-34

OpenMP 5.2 §10.3 p.234 l.14-24

Evidence

Same compiler, one change

the element assignment A(II) = Z instead of the whole-array assignment A = Z: AFAR 24.3.0: correct; CCE 19.0.0: correct; CCE 20.0.2: correct; CCE 21.0.3: correct; CCE 21.0.4: correct; NVHPC 25.11: correct; NVHPC 26.9: correct

Control program

MODULE GLOBALS
   IMPLICIT NONE
   REAL :: Z
   REAL, ALLOCATABLE, DIMENSION(:)::A

END MODULE GLOBALS

MODULE RECON
   USE GLOBALS

   IMPLICIT NONE

CONTAINS

   SUBROUTINE CRASH_REPRO()

      IMPLICIT NONE
      INTEGER::II

      ALLOCATE (A(10))
      Z = 7

      !$OMP target teams loop
      DO II = 1, 1
         A(II) = Z;
      END DO
      !$OMP end target teams loop

      PRINT *, A(1)

   END SUBROUTINE CRASH_REPRO

END MODULE RECON

PROGRAM compilerCrashTest

   USE RECON
   IMPLICIT NONE

   CALL CRASH_REPRO()

END PROGRAM compilerCrashTest

Other compilers run it correctly

AFAR 24.3.0, CCE 19.0.0, CCE 20.0.2, CCE 21.0.3, CCE 21.0.4, GCC 16.2.0