Observed issues › fresh/cce-list-directed-print-in-target-copy-crash

Appears to behave incorrectlyCrashes or aborts at run time · Fortran · OpenMP · Fixed in 21.0.3

ftn: a list-directed PRINT in a target region aborts the run in copy mode

Legality: contested. An independent adjudicating reading finds the result not required: Both readings agree the program is conforming: a print statement is base-language code and nothing forbids it in a target region. 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	correct	correct	correct	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	behaves incorrectly	behaves incorrectly	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	show Memory access fault by GPU node-6 (Agent handle: 0x4c7c50) on address 0x154f25c2a000. Reason: Unknown.
behaves incorrectly	CCE 19.0.0, MI250X	show Memory access fault by GPU node-7 (Agent handle: 0x4bf200) on address 0x15543ae14000. Reason: Unknown. Failed to fetch queues snapshot. GPU core dump failed
behaves incorrectly	CCE 20.0.2, MI300A copy	show Memory access fault by GPU node-7 (Agent handle: 0x4cd0e0) on address 0x154f35a2a000. Reason: Unknown.
behaves incorrectly	CCE 20.0.2, MI250X	show Memory access fault by GPU node-11 (Agent handle: 0x4c5510) on address 0x15543ae14000. Reason: Unknown.

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 -fopenmp -O2 repro.f90 -o repro
HSA_XNACK=0 OMP_TARGET_OFFLOAD=mandatory ./repro

CCE 20.0.2, MI300A copy

module load cce/20.0.2 craype-accel-amd-gfx942 rocm/6.4.3
module unload cray-mpich
ftn -fopenmp -O2 repro.f90 -o repro
HSA_XNACK=0 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 (35 lines)

! # Finding (from the data/bench/fresh draft fresh-022): CCE 19/20 abort on a list-directed
!   PRINT in a target region in copy mode
!
! - source: data/bench/fresh/fresh-022.f90 (llvm-project issue 166338, the reporter's program unchanged)
! - compilers: CCE 19.0.0 and 20.0.2 ftn, -fopenmp -O2, OMP_TARGET_OFFLOAD=mandatory; MI300A (gfx942)
!   at HSA_XNACK=0 and MI250X (gfx90a, Tioga) at HSA_XNACK=0
! - expected: "Hello, world!" from the device. PRINT is a base-language statement; OpenMP 5.2 places
!   no restriction on I/O statements in a target region (the restrictions on structured blocks,
!   §4.3 p.76-77, concern branches into and out of the block), and an implementation is compliant
!   only if it "compiles and executes all other conforming programs" (§1.6 p.34 l.22-24). HPE
!   documents the case: "Fortran PRINT
!   statements are supported, with limitations, when called from OpenMP offload regions compiled for
!   AMD GPU targets. The current implementation supports PRINT statements with a single scalar value
!   of type character, integer, real, or complex" (intro_openmp(7), CCE 19.0.0 and 20.0.2,
!   "Printing from GPU kernels"; data/index/claims/)
! - actual: CCE 19 and 20 at HSA_XNACK=0: "Memory access fault by GPU node-4 ... Reason: Unknown",
!   SIGABRT, nothing printed (MI300A three runs each; MI250X). At HSA_XNACK=1 correct. The same with
!   `print *, 42` (data/reports/fresh-drafts/d022-list-int.f90) and with two list-directed PRINTs
!   (d022-two-lists.f90)
! - control (same compiler): the formatted `print '(a)', "Hello, world!"`
!   (data/reports/fresh-drafts/d022-fmt-literal.f90) prints the line on CCE 19/20 at XNACK 0 and 1.
!   With a mapped scalar written in the same region the list-directed PRINT also works
!   (d022-map-only.f90, d022-list-literal-dev.f90): the abort needs a region without other data
! - other families: AFAR, CCE 21.0.3 and NVHPC 25.11/26.9 (A100, separate and managed memory) print
!   the line
! - legality: legal by reading (no OpenMP construct but target; no data)
! - evidence: results/fresh-drafts/tuo*-*.jsonl, tioga/tioga-a.jsonl, wingtip/nv.jsonl (runs of the review);
!   the data/bench/fresh ground truth (data/bench/fresh/outcomes.csv: cce19/cce20
!   copy crash on gfx942 and gfx90a)
program reproducer
  implicit none
  !$omp target
  print *, "Hello, world!"
  !$omp end target
end program reproducer

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

Hello, world!

Observed where it behaves incorrectly

Memory access fault by GPU node-6 (Agent handle: 0x4c7c50) on address 0x154f25c2a000. Reason: Unknown.

What the specification requires

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

OpenMP 5.2 §4.3 p.76-77

OpenMP 5.2 §1.6 p.34 l.22-24

compiles and executes all other conforming programs

Evidence

Same compiler, one change

a formatted PRINT, print '(a)', instead of list-directed print *: 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
Control program
```
program reproducer
  implicit none
  !$omp target
  print '(a)', "Hello, world!"
  !$omp end target
end program reproducer
```

Other compilers run it correctly

AFAR 24.3.0, NVHPC 25.11, NVHPC 26.9, GCC 16.2.0