amdflang: declare reduction on a derived type used in a target loop fails at the device link (flang runtime symbols undefined)

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	not supported	not supported	not supported	correct	correct	n/a	n/a
CCE 19.0.0	fails another way	fails another way	fails another way	unknown not run	unknown not run	n/a	n/a
CCE 20.0.2	fails another way	fails another way	fails another way	unknown not run	unknown not run	n/a	n/a
CCE 21.0.3	fails another way	fails another way	fails another way	unknown not run	unknown not run	n/a	n/a
CCE 21.0.4	fails another way	fails another way	fails another way	unknown not run	unknown not run	n/a	n/a
NVHPC 25.11	n/a	n/a	n/a	n/a	n/a	fails another way	fails another way
NVHPC 26.9	n/a	n/a	n/a	n/a	n/a	fails another way	fails another way
GCC 16.2.0	fails another way	fails another way	fails another way	fails another way	fails another way	fails another way	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
not supported	AFAR 24.3.0, MI300A unified AFAR 24.3.0, MI300A copy	show ld.lld: error: undefined symbol: flang_rt_verbose_abort(char const, ...) >>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(std::__1::__throw_bad_variant_access[abi:nqn240000](char const)) >>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(std::__1::__throw_bad_variant_access[abi:nqn240000](char const*)) ld.lld: error: undefined symbol: Fortran::runtime::io::descr::DescriptorIoTicket<(Fortran::runtime::io::Direction)1>::Begin(Fortran::runtime::WorkQueue&) >>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(_FortranAInitialize) >>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(_FortranAInitialize) ld.lld: error: undefined symbol: Fortran::runtime::io::descr::DescriptorIoTicket<(Fortran::runtime::io::Direction)0>::Begin(Fortran::runtime::WorkQueue&) >>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(_FortranAInitialize)
not supported	AFAR 24.3.0, MI250X	show ld.lld: error: undefined symbol: flang_rt_verbose_abort(char const, ...) >>> referenced by <dir>/repro.amdgpu9.0a.gfx90a-eaacd7.img.lto.o:(std::__1::__throw_bad_variant_access[abi:nqn240000](char const)) >>> referenced by <dir>/repro.amdgpu9.0a.gfx90a-eaacd7.img.lto.o:(std::__1::__throw_bad_variant_access[abi:nqn240000](char const*)) ld.lld: error: undefined symbol: Fortran::runtime::io::descr::DescriptorIoTicket<(Fortran::runtime::io::Direction)1>::Begin(Fortran::runtime::WorkQueue&) >>> referenced by <dir>/repro.amdgpu9.0a.gfx90a-eaacd7.img.lto.o:(_FortranAInitialize) >>> referenced by <dir>/repro.amdgpu9.0a.gfx90a-eaacd7.img.lto.o:(_FortranAInitialize) ld.lld: error: undefined symbol: Fortran::runtime::io::descr::DescriptorIoTicket<(Fortran::runtime::io::Direction)0>::Begin(Fortran::runtime::WorkQueue&) >>> referenced by <dir>/repro.amdgpu9.0a.gfx90a-eaacd7.img.lto.o:(_FortranAInitialize)
fails another way	CCE 19.0.0, MI300A unified CCE 19.0.0, MI300A copy CCE 19.0.0, MI250X	show Warning: Headers and libraries from cray-libsci/25.03.0 will be ignored because depends on PE_MPICH which is unavailable.. !$omp target teams distribute parallel do reduction(padd: rp) map(tofrom: rp) ftn-7032 ftn: ERROR FUZZ_PROG, File = afar-udr-derived-type-link.f90, Line = 17 Unsupported OpenMP construct Calls -- $udr_i_1FUZZ_DMOD
fails another way	CCE 20.0.2, MI300A unified CCE 20.0.2, MI300A copy CCE 20.0.2, MI250X	show Warning: Headers and libraries from cray-libsci/25.09.0 will be ignored because depends on PE_MPICH which is unavailable.. !$omp target teams distribute parallel do reduction(padd: rp) map(tofrom: rp) ftn-7032 ftn: ERROR FUZZ_PROG, File = afar-udr-derived-type-link.f90, Line = 17 Unsupported OpenMP construct Calls -- $udr_i_1FUZZ_DMOD
fails another way	CCE 21.0.3, MI300A unified CCE 21.0.3, MI300A copy CCE 21.0.3, MI250X CCE 21.0.4, MI300A unified CCE 21.0.4, MI300A copy CCE 21.0.4, MI250X	show Warning: Headers and libraries from cray-libsci/26.03.0 will be ignored because depends on PE_MPICH which is unavailable.. !$omp target teams distribute parallel do reduction(padd: rp) map(tofrom: rp) ftn-7032 ftn: ERROR FUZZ_PROG, File = afar-udr-derived-type-link.f90, Line = 17 Unsupported OpenMP construct Calls -- $udr_i_1FUZZ_DMOD
fails another way	NVHPC 25.11, A100 separate NVHPC 25.11, A100 managed NVHPC 26.9, A100 separate NVHPC 26.9, A100 managed	show NVFORTRAN-W-0155-Unimplemented feature - DECLARE REDUCTION (afar-udr-derived-type-link.f90: 6) 0 inform, 1 warnings, 0 severes, 0 fatal for fuzz_dmod NVFORTRAN-S-0155-The reduction intrinsic must be MAX, MIN, IAND, IOR, or IEOR, not padd (afar-udr-derived-type-link.f90: 17) 0 inform, 0 warnings, 1 severes, 0 fatal for fuzz_prog
fails another way	GCC 16.2.0, MI300A unified GCC 16.2.0, MI300A copy GCC 16.2.0, MI250X GCC 16.2.0, MI300X GCC 16.2.0, MI350X GCC 16.2.0, A100 separate	show afar-udr-derived-type-link.f90:17:54: 17 \| !$omp target teams distribute parallel do reduction(padd: rp) map(tofrom: rp) \| 1 Error: !$OMP DECLARE REDUCTION padd not found at (1)

Reproduce

AFAR 24.3.0, MI300A unified

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

AFAR 24.3.0, MI300X (runs correctly)

$AFAR/bin/amdflang -fopenmp --offload-arch=gfx942 -O2 repro.f90 -o repro
HSA_XNACK=0 OMP_TARGET_OFFLOAD=mandatory ./repro

Program (33 lines)

module fuzz_dmod
  implicit none
  type :: pair_t
    integer :: x, y
  end type pair_t
  !$omp declare reduction(padd: pair_t: omp_out = &
  !$omp& pair_t(omp_out%x + omp_in%x, omp_out%y + omp_in%y)) &
  !$omp& initializer(omp_priv = pair_t(0, 0))
end module fuzz_dmod
program fuzz_prog
  use fuzz_dmod
  implicit none
  integer, parameter :: n = 64
  type(pair_t) :: rp
  integer :: i
  rp = pair_t(1, 2)
  !$omp target teams distribute parallel do reduction(padd: rp) map(tofrom: rp)
  do i = 1, n
    rp = pair_t(rp%x + i, rp%y + mod(i, 3))
  end do
  print '(a,i0,a,i0)', 'RESULT checksum=', rp%x * 1000 + rp%y, ' ran_on_host=', 0
end program fuzz_prog
! spec (requirement): A declare reduction directive declares a user-defined reduction identifier usable in a reduction
!   clause.
! "The declare reduction directive declares a reduction-identifier that can be used in a reduction clause as a
!   user-defined reduction." (5.2 §5.5.11 p.139 l.18-19)
! spec (requirement): The identifier can be used in a reduction clause on variables of the types listed in the
!   directive, here a derived type.
! "The reduction-identifier can later be used in a reduction clause that uses variables of the types specified in the
!   declare reduction directive." (5.2 §5.5.11 p.139 l.24-26)
! spec (this program): The module declares reduction(padd: pair_t) and the program uses padd on a target teams
!   distribute parallel do; the compiler must generate the combiner and initializer for the device, and amdflang
!   instead fails at the device link on undefined flang runtime symbols.

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

ld.lld: error: undefined symbol: flang_rt_verbose_abort(char const*, ...)
>>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(std::__1::__throw_bad_variant_access[abi:nqn240000](char const*))
>>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(std::__1::__throw_bad_variant_access[abi:nqn240000](char const*))
ld.lld: error: undefined symbol: Fortran::runtime::io::descr::DescriptorIoTicket<(Fortran::runtime::io::Direction)1>::Begin(Fortran::runtime::WorkQueue&)
>>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(_FortranAInitialize)
>>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(_FortranAInitialize)
ld.lld: error: undefined symbol: Fortran::runtime::io::descr::DescriptorIoTicket<(Fortran::runtime::io::Direction)0>::Begin(Fortran::runtime::WorkQueue&)
>>> referenced by <dir>/repro.amdgpu9.42.gfx942-5a66dd.img.lto.o:(_FortranAInitialize)

What the specification requires

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

OpenMP 5.2 §5.5.11 p.139 l.18-19

The declare reduction directive declares a reduction-identifier that can be used in a reduction clause as a user-defined reduction.

OpenMP 5.2 §5.5.11 p.139 l.24-26

The reduction-identifier can later be used in a reduction clause that uses variables of the types specified in the declare reduction directive.

Evidence

Same compiler, one change

the same reduction on a host parallel do (no target teams distribute): 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

module fuzz_dmod
  implicit none
  type :: pair_t
    integer :: x, y
  end type pair_t
  !$omp declare reduction(padd: pair_t: omp_out = &
  !$omp& pair_t(omp_out%x + omp_in%x, omp_out%y + omp_in%y)) &
  !$omp& initializer(omp_priv = pair_t(0, 0))
end module fuzz_dmod
program fuzz_prog
  use fuzz_dmod
  implicit none
  integer, parameter :: n = 64
  type(pair_t) :: rp
  integer :: i
  rp = pair_t(1, 2)
  !$omp parallel do reduction(padd: rp)
  do i = 1, n
    rp = pair_t(rp%x + i, rp%y + mod(i, 3))
  end do
  print '(a,i0)', 'RESULT checksum=', rp%x * 1000 + rp%y
end program fuzz_prog