Observed issues › harvest/cce21-internal-procedure-calls-internal-in-target-fault

Appears to behave incorrectlyRun-time error · Fortran · OpenMP · Open · regression

ftn 21: a target region calling an internal function that calls another internal function faults on the GPU

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	behaves incorrectly	behaves incorrectly	unknown not run	unknown not run	unknown not run	n/a	n/a
CCE 21.0.4	behaves incorrectly	behaves incorrectly	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	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 21.0.3, MI300A unified	show :0:rocdevice.cpp :3586: 22644563817 us: Callback: Queue 0x155545200000 aborting with error : HSA_STATUS_ERROR_EXCEPTION: An HSAIL operation resulted in a hardware exception. code: 0x1016 ACC: libcrayacc/acc_hw_amd.c:1927 CRAY_ACC_ERROR - hipStreamSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' from site-harvest--cce21-internal-procedure-calls-internal-in-targ ACC: libcrayacc/acc_hw_amd.c:1476 CRAY_ACC_ERROR - hipDeviceSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' OpenMP Execution Error: site-harvest--cce21-internal-procedure-calls-internal-in-target-fault.f90:65 - unsupported access to Fortran host-associated variable in GPU internal procedure Kernel Name: p_$ck_L47_1 VGPU=0x941070 SWq=0x155555526000, HWq=0x155545200000, id=1 Dispatch Header =0xb02 (type=2, barrier=1, acquire=1, release=1), setup=0 grid=[1, 1, 1], workgroup=[1, 1, 1]
behaves incorrectly	CCE 21.0.3, MI300A copy	show :0:rocdevice.cpp :3586: 22648093883 us: Callback: Queue 0x155545200000 aborting with error : HSA_STATUS_ERROR_EXCEPTION: An HSAIL operation resulted in a hardware exception. code: 0x1016 ACC: libcrayacc/acc_hw_amd.c:1496 CRAY_ACC_ERROR - hipStreamSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' from site-harvest--cce21-internal-procedure-calls-internal-in-targ ACC: libcrayacc/acc_hw_amd.c:1476 CRAY_ACC_ERROR - hipDeviceSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' OpenMP Execution Error: site-harvest--cce21-internal-procedure-calls-internal-in-target-fault.f90:65 - unsupported access to Fortran host-associated variable in GPU internal procedure Kernel Name: p_$ck_L47_1 VGPU=0x941070 SWq=0x155555526000, HWq=0x155545200000, id=1 Dispatch Header =0xb02 (type=2, barrier=1, acquire=1, release=1), setup=0 grid=[1, 1, 1], workgroup=[1, 1, 1]
behaves incorrectly	CCE 21.0.4, MI300A unified	show :0:rocdevice.cpp :3586: 22644554671 us: Callback: Queue 0x155545200000 aborting with error : HSA_STATUS_ERROR_EXCEPTION: An HSAIL operation resulted in a hardware exception. code: 0x1016 ACC: libcrayacc/acc_hw_amd.c:1927 CRAY_ACC_ERROR - hipStreamSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' from site-harvest--cce21-internal-procedure-calls-internal-in-targ ACC: libcrayacc/acc_hw_amd.c:1476 CRAY_ACC_ERROR - hipDeviceSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' OpenMP Execution Error: site-harvest--cce21-internal-procedure-calls-internal-in-target-fault.f90:65 - unsupported access to Fortran host-associated variable in GPU internal procedure Kernel Name: p_$ck_L47_1 VGPU=0x9410b0 SWq=0x155555526000, HWq=0x155545200000, id=1 Dispatch Header =0xb02 (type=2, barrier=1, acquire=1, release=1), setup=0 grid=[1, 1, 1], workgroup=[1, 1, 1]
behaves incorrectly	CCE 21.0.4, MI300A copy	show :0:rocdevice.cpp :3586: 22648115511 us: Callback: Queue 0x155545200000 aborting with error : HSA_STATUS_ERROR_EXCEPTION: An HSAIL operation resulted in a hardware exception. code: 0x1016 ACC: libcrayacc/acc_hw_amd.c:1496 CRAY_ACC_ERROR - hipStreamSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' from site-harvest--cce21-internal-procedure-calls-internal-in-targ ACC: libcrayacc/acc_hw_amd.c:1476 CRAY_ACC_ERROR - hipDeviceSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' OpenMP Execution Error: site-harvest--cce21-internal-procedure-calls-internal-in-target-fault.f90:65 - unsupported access to Fortran host-associated variable in GPU internal procedure Kernel Name: p_$ck_L47_1 VGPU=0x9410b0 SWq=0x155555526000, HWq=0x155545200000, id=1 Dispatch Header =0xb02 (type=2, barrier=1, acquire=1, release=1), setup=0 grid=[1, 1, 1], workgroup=[1, 1, 1]

Reproduce

CCE 21.0.3, MI300A unified

module load cce/21.0.3 craype-accel-amd-gfx942 rocm/7.2.0
module unload cray-mpich
ftn -fopenmp -O2 repro.f90 -o repro
HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro

CCE 21.0.4, MI300A unified

module load cce/21.0.4 craype-accel-amd-gfx942 rocm/7.2.0
module unload cray-mpich
ftn -fopenmp -O2 repro.f90 -o repro
HSA_XNACK=1 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 (68 lines)

! # Finding: CCE 21 ftn: a target region that calls an internal
! # function which calls another internal function of the same host faults
! # on the GPU ("unspecified launch failure"; CCE 19 and 20 correct)
!
! - compilers: CCE ftn 21.0.3 and 21.0.4, -O2: this file on MI300A gfx942
!   (HSA_XNACK 0 and 1; gate confirmed); the libgomp test it comes from also
!   on MI250X gfx90a with 21.0.3.
! - expected: r= 43 44 a= 6 7 dev= T
! - actual: the kernel faults: "HSA_STATUS_ERROR_EXCEPTION: An HSAIL operation
!   resulted in a hardware exception. code: 0x1016", then "CRAY_ACC_ERROR -
!   hipStreamSynchronize returned hipErrorLaunchFailure: 'unspecified launch
!   failure'" at the target construct, exit status 1, every run.
! - rule: "If a procedure is explicitly or implicitly referenced in a target
!   construct that does not specify a device clause in which the ancestor
!   device-modifier appears then that procedure is treated as if its name had
!   appeared in an enter clause on a declare target directive" (5.2 §13.8
!   p.285 l.1-3), and a procedure referenced in such a procedure is treated
!   the same way (5.2 §7.8 p.205 l.3-6). Both rules are in 5.0 (Appendix B:
!   "implicit declare target directives for some functions ... and
!   subroutines (Fortran) were added").
! - regression: CCE 19.0.0 and 20.0.2 are correct on every form (gfx942
!   both modes; gfx90a).
! - conditions (one change each, CCE 21.0.3 and 21.0.4, gfx942 X0 and X1):
!   the fault needs the internal-to-internal call. Correct: wrap does the
!   addition itself (no call); wrap calls a module procedure instead of the
!   internal function. Faults (forms of the libgomp test): the target
!   construct in the main program or in a third internal function; with or
!   without a module procedure call next to the internal call.
! - other families: AFAR amdflang correct on this file and on the libgomp
!   test (gfx942 X0 and X1, gfx90a).
! - harvest: dev program h2-gomp-f-declare-target-3 (data/bench/harvest2; libgomp
!   testsuite libgomp.fortran/declare-target-3.f90, GPL-3.0-or-later; this
!   file is a new program, not a copy), wrong_answer on CCE 21 only, gate
!   confirmed on gfx90a.
! - legality: offload.engine.prerun.check accepts it (the rules with the race
!   rule, and the AFAR flang front end).
! CCE:  ftn -fopenmp -O2 (module load cce/<v> craype-accel-amd-gfx942 rocm/<r>;
!       module unload cray-mpich)
! Run:  OMP_TARGET_OFFLOAD=mandatory ./a.out
program p
  use omp_lib
  implicit none
  integer :: a(2), r(2)
  logical :: dev
  a = [1, 2]
  dev = .false.
  !$omp target map(tofrom: a, r) map(from: dev)
  r = wrap(a)
  dev = .not. omp_is_initial_device()
  !$omp end target
  print '(a,2(1x,i0),a,2(1x,i0),a,l1)', 'r=', r, ' a=', a, ' dev= ', dev
  if (all(r == [43, 44]) .and. all(a == [6, 7])) then
    print '(a,i0,a)', 'RESULT ran_on_host=', merge(0, 1, dev), ' check=pass'
  else
    print '(a,i0,a)', 'RESULT ran_on_host=', merge(0, 1, dev), ' check=fail'
  end if
contains
  integer function add42(y)
    integer :: y
    add42 = y + 42
  end function
  function wrap(y)
    integer :: y(2), wrap(2)
    y = y + 5
    wrap(1) = add42(y(1) - 5)
    wrap(2) = add42(y(2) - 5)
  end function
end program p

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

r= 43 44 a= 6 7 dev= T

Observed where it behaves incorrectly

:0:rocdevice.cpp            :3586: 22644563817 us:  Callback: Queue 0x155545200000 aborting with error : HSA_STATUS_ERROR_EXCEPTION: An HSAIL operation resulted in a hardware exception. code: 0x1016
ACC: libcrayacc/acc_hw_amd.c:1927 CRAY_ACC_ERROR -  hipStreamSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure' from site-harvest--cce21-internal-procedure-calls-internal-in-targ
ACC: libcrayacc/acc_hw_amd.c:1476 CRAY_ACC_ERROR -  hipDeviceSynchronize returned hipErrorLaunchFailure: 'unspecified launch failure'
OpenMP Execution Error: site-harvest--cce21-internal-procedure-calls-internal-in-target-fault.f90:65 - unsupported access to Fortran host-associated variable in GPU internal procedure
Kernel Name: p_$ck_L47_1
VGPU=0x941070 SWq=0x155555526000, HWq=0x155545200000, id=1
	Dispatch Header =0xb02 (type=2, barrier=1, acquire=1, release=1), setup=0
	grid=[1, 1, 1], workgroup=[1, 1, 1]

What the specification requires

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

OpenMP 5.2 §13.8 p.285 l.1-3

If a procedure is explicitly or implicitly referenced in a target construct that does not specify a device clause in which the ancestor device-modifier appears then that procedure is treated as if its name had appeared in an enter clause on a declare target directive

OpenMP 5.2 §7.8 p.205 l.3-6

Evidence

Other compilers run it correctly

AFAR 24.3.0, NVHPC 25.11, NVHPC 26.9, GCC 16.2.0