OffTarget

Observed issues › gcc/gcc-c-omp60-unimplemented-rejected

Appears not to supportRefused at compile time · C · OpenMP · Open

OpenMP 6.0 features that GCC 16.2 lists as not implemented are rejected (directive-name modifiers, induction, taskgraph clauses, map and defaultmap self/storage, and more)

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

GCC 16.2 rejects OpenMP 6.0 forms that its implementation status table lists as not implemented: directive-name modifiers on clauses, the induction, local, replayable and transparent clauses, num_threads(strict:), variable categories on default, defaultmap(self/storage), map(self:/storage:/ref_*), firstprivate(saved:), reduction(original:), arguments on requires clauses and nowait, nogroup on target_data, and the underscore spellings of multi-word directives. Most stop with a syntax error; declare target local stops with "Sorry, ... not implemented", nogroup with "'nogroup' is not valid for '#pragma omp target data'", and the ignored declare_mapper leaves its mapper "not found". GCC 16.2 claims OpenMP 5.2 (_OPENMP 202111) and parts of 6.0.

Affected variants

CompilerMI300A 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 24.3.0fails another wayfails another wayunknown
not run
fails another wayfails another wayn/an/a
CCE 19.0.0fails another wayfails another wayunknown
not run
unknown
not run
unknown
not run
n/an/a
CCE 20.0.2fails another wayfails another wayunknown
not run
unknown
not run
unknown
not run
n/an/a
CCE 21.0.3fails another wayfails another wayunknown
not run
unknown
not run
unknown
not run
n/an/a
CCE 21.0.4fails another wayfails another wayunknown
not run
unknown
not run
unknown
not run
n/an/a
NVHPC 25.11n/an/an/an/an/afails another wayfails another way
NVHPC 26.9n/an/an/an/an/afails another wayfails another way
GCC 16.2.0not supportednot supportednot supportednot supportednot supportednot supportedn/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

ResultCompiler and GPUMessage
not supportedGCC 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
site-gcc--gcc-c-omp60-unimplemented-rejected.c: In function ‘main’:
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:57: error: expected an OpenMP clause before ‘induction’
   96 |     #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
      |                                                         ^~~~~~~~~
site-gcc--gcc-c-omp60-unimplemented-rejected.c:98:40: error: ‘x’ not specified in enclosing ‘parallel’
   98 |       b[i - 1] = 2 * a[i - 1] + 1 + (x - (i + 2));
      |                                     ~~~^~~~~~~~~~
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:13: note: enclosing ‘parallel’
fails another wayAFAR 24.3.0, MI300A unified
AFAR 24.3.0, MI300A copy
AFAR 24.3.0, MI300X
AFAR 24.3.0, MI350X
show
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:57: error: unexpected OpenMP clause 'induction' in directive '#pragma omp parallel for'
   96 |     #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
      |                                                         ^
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:57: warning: extra tokens at the end of '#pragma omp parallel for' are ignored [-Wextra-tokens]
site-gcc--gcc-c-omp60-unimplemented-rejected.c:98:38: error: variable 'x' must have explicitly specified data sharing attributes
   98 |       b[i - 1] = 2 * a[i - 1] + 1 + (x - (i + 2));
      |                                      ^
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:38: note: explicit data sharing attribute requested here
fails another wayCCE 19.0.0, MI300A unified
CCE 19.0.0, MI300A copy
CCE 20.0.2, MI300A unified
CCE 20.0.2, MI300A copy
CCE 21.0.3, MI300A unified
CCE 21.0.3, MI300A copy
CCE 21.0.4, MI300A unified
CCE 21.0.4, MI300A copy
show
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:57: warning: extra tokens at the end of '#pragma omp parallel for' are ignored [-Wextra-tokens]
   96 |     #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
      |                                                         ^
site-gcc--gcc-c-omp60-unimplemented-rejected.c:98:38: error: variable 'x' must have explicitly specified data sharing attributes
   98 |       b[i - 1] = 2 * a[i - 1] + 1 + (x - (i + 2));
      |                                      ^
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:38: note: explicit data sharing attribute requested here
   96 |     #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
fails another wayNVHPC 25.11, A100 separate
NVHPC 25.11, A100 managed
NVHPC 26.9, A100 separate
NVHPC 26.9, A100 managed
show
"site-gcc--gcc-c-omp60-unimplemented-rejected.c", line 96: error: invalid text in pragma
      #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
                                                          ^
"site-gcc--gcc-c-omp60-unimplemented-rejected.c", line 96: error: extra text after expected end of preprocessing directive
      #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
                                                                   ^
2 errors detected in the compilation of "site-gcc--gcc-c-omp60-unimplemented-rejected.c".

Reproduce

GCC 16.2.0, MI300A unified

export LD_LIBRARY_PATH=$GCC_16_2/lib64:/opt/rocm-7.2.1/lib
$GCC_16_2/bin/gcc -fopenmp -foffload=amdgcn-amdhsa -foffload-options=amdgcn-amdhsa=-march=gfx942 -O2 repro.c -o repro
HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro

GCC 16.2.0, A100 separate

export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=$GCC_16_2/lib64
$GCC_16_2/bin/gcc -fopenmp -foffload=nvptx-none -foffload-options=nvptx-none=-march=sm_80 -foffload-options=nvptx-none=-latomic -O2 repro.c -o repro
OMP_TARGET_OFFLOAD=mandatory ./repro
Program (108 lines)
// # Finding (GCC 16.2): OpenMP 6.0 forms GCC documents as not implemented are rejected (C); the reproducer is the catalog program
//   omp60-c-clause-induction
//
// - toolchains: gcc_g90 (GCC 16.2.0, AMD Instinct MI250X gfx90a, Tioga) and gcc_sm80 (GCC 16.2.0,
//   NVIDIA A100 sm_80, wingtip); separate memory only; -O2; OMP_TARGET_OFFLOAD=mandatory
// - build: gcc -fopenmp -foffload=amdgcn-amdhsa -foffload-options=amdgcn-amdhsa=-march=gfx90a -O2 (gcc_g90); -foffload=nvptx-none
//   -foffload-options=nvptx-none=-march=sm_80 -foffload-options=nvptx-none=-latomic -O2 (gcc_sm80)
// - failure: rejected: expected an OpenMP clause before
// - observed: a compile-time error for every program of the list (the first: `error: expected an OpenMP clause before 'induction'`;
//   nogroup on target_data: `'nogroup' is not valid for '#pragma omp target data'`; the ignored declare_mapper: `mapper 'mm' not
//   found`); the same on both GPUs.
// - expected: compiles; `RESULT ran_on_host=0 check=pass`
// - GCC 16.2 documentation (libgomp.texi of the GCC 16.2.0 release tarball, "OpenMP Implementation Status", OpenMP 6.0 table) lists
//   each of these as not implemented (N): "directive-name-modifier accepted in all clauses", "Support for inductions", "local
//   clause to declare_target directive", "New taskgraph construct including saved modifier and replayable clause", "New transparent
//   clause for multi-generational task-dependence graphs", "strict modifier keyword to num_threads", "default clause on the target
//   directive and accepting variable categories", "New storage map-type modifier", "self modifier to map and self as defaultmap
//   argument", "ref modifier to the map clause", "Reduction over private variables with reduction clause" (the original modifier),
//   "Boolean argument to nowait and nogroup may be non constant", "target_data as composite construct" (nogroup on target_data),
//   "All inarguable clauses take now an optional Boolean argument" (the requires clauses' argument), "For Fortran, the automap
//   modifier to the enter clause", "workdistribute directive for Fortran" and "Multi-word directives now use underscore by default"
//   (declare_simd, declare_mapper, target_loop). The two 6.0 forms no row covers (requires device_safesync, defaultmap(private))
//   are a finding of their own (gcc-c-omp60-undocumented-rejected).
// - spec: the induction clause (6.0 §7.6.13 p.257), the directive-name-modifier (6.0 §5.4 p.173), the replayable clause (6.0 §14.6
//   p.440), the transparent clause (6.0 §17.9.6 p.510).
// - other families: AFAR, CCE and NVHPC reject many of the same 6.0 forms (their own findings and the catalog's unsupported cells).
// - gate and confirmer: 168 cells (84 gcc_g90, 84 gcc_sm80), each confirmed in the gate's four environments (a-d;
//   results/gcc/gate-tioga/ and gate-wingtip/ verdicts.jsonl); confirmer verdicts (results/gcc/confirm/batch.gcc_*.jsonl):
//   unsupported (R11a-newer-form-unclaimed) 168
// - catalog programs (84): omp60-c-clause-aligned-directive_name, omp60-c-clause-allocate-directive_name,
//   omp60-c-clause-atomic_default_mem_order-directive_name, omp60-c-clause-bind-directive_name,
//   omp60-c-clause-collapse-directive_name, omp60-c-clause-default-aggregate, omp60-c-clause-default-all,
//   omp60-c-clause-default-directive_name, omp60-c-clause-default-pointer, omp60-c-clause-default-scalar,
//   omp60-c-clause-defaultmap-directive_name, omp60-c-clause-defaultmap-self, omp60-c-clause-defaultmap-storage,
//   omp60-c-clause-depend-directive_name, omp60-c-clause-device-directive_name, omp60-c-clause-device_type-directive_name,
//   omp60-c-clause-dist_schedule-directive_name, omp60-c-clause-dynamic_allocators-directive_name,
//   omp60-c-clause-dynamic_allocators-required, omp60-c-clause-enter-directive_name, omp60-c-clause-firstprivate-directive_name,
//   omp60-c-clause-firstprivate-saved, omp60-c-clause-from-directive_name, omp60-c-clause-has_device_addr-directive_name,
//   omp60-c-clause-in_reduction-directive_name, omp60-c-clause-indirect-directive_name, omp60-c-clause-induction,
//   omp60-c-clause-induction-directive_name, omp60-c-clause-induction-plus, omp60-c-clause-induction-relaxed,
//   omp60-c-clause-induction-step, omp60-c-clause-induction-strict, omp60-c-clause-induction-times,
//   omp60-c-clause-is_device_ptr-directive_name, omp60-c-clause-lastprivate-directive_name, omp60-c-clause-linear-directive_name,
//   omp60-c-clause-link-directive_name, omp60-c-clause-local, omp60-c-clause-local-directive_name,
//   omp60-c-clause-map-directive_name, omp60-c-clause-map-mapper, omp60-c-clause-map-self, omp60-c-clause-map-storage,
//   omp60-c-clause-message-directive_name, omp60-c-clause-nogroup, omp60-c-clause-nogroup-directive_name,
//   omp60-c-clause-nogroup-do_not_synchronize, omp60-c-clause-nontemporal-directive_name, omp60-c-clause-nowait-directive_name,
//   omp60-c-clause-nowait-do_not_synchronize, omp60-c-clause-num_teams-directive_name, omp60-c-clause-num_threads-directive_name,
//   omp60-c-clause-num_threads-strict, omp60-c-clause-order-directive_name, omp60-c-clause-ordered-directive_name,
//   omp60-c-clause-private-directive_name, omp60-c-clause-proc_bind-directive_name, omp60-c-clause-reduction-directive_name,
//   omp60-c-clause-reduction-original, omp60-c-clause-replayable, omp60-c-clause-replayable-directive_name,
//   omp60-c-clause-replayable-replayable_expression, omp60-c-clause-reverse_offload-directive_name,
//   omp60-c-clause-reverse_offload-required, omp60-c-clause-safelen-directive_name, omp60-c-clause-schedule-directive_name,
//   omp60-c-clause-self_maps-directive_name, omp60-c-clause-self_maps-required, omp60-c-clause-severity-directive_name,
//   omp60-c-clause-shared-directive_name, omp60-c-clause-simdlen-directive_name, omp60-c-clause-thread_limit-directive_name,
//   omp60-c-clause-to-directive_name, omp60-c-clause-transparent, omp60-c-clause-transparent-directive_name,
//   omp60-c-clause-transparent-impex_type, omp60-c-clause-unified_address-directive_name, omp60-c-clause-unified_address-required,
//   omp60-c-clause-unified_shared_memory-directive_name, omp60-c-clause-unified_shared_memory-required,
//   omp60-c-clause-use_device_addr-directive_name, omp60-c-clause-use_device_ptr-directive_name,
//   omp60-c-clause-uses_allocators-directive_name, omp60-c-directive-target_loop
// - legality: the catalog programs are legal by construction (offload.engine.gen, c_gen and acc_gen
//   check every program against the catalog rules and the offload scope as it is built; no
//   compiler). GCC is a compiler under test only: never a legality witness, here or for another
//   family's case (catalog/machines.yaml under_test_only).
// Generated by compliance.c_gen (OpenMP 6.0, C11): the C port (offload.engine.f2c) of omp60-clause-induction. Do not edit.
// id: omp60-c-clause-induction
// primary atom: omp.clause.induction
// legal: yes
// observable: yes
// ran_on_host: inside the target region of interest
// evidence: construct
// needs: none
// std: c11
#include <stdio.h>
#include <stdbool.h>
#include <omp.h>

enum { n = 64 };

int main(void) {
  int a[n];
  int b[n];
  int ref[n];
  int host_flag;
  bool ok;
  int x;
  for (int i = 1; i <= n; i++) {
    a[i - 1] = i;
    b[i - 1] = 0;
    ref[i - 1] = 2 * i + 1;
  }
  host_flag = 1;
  x = 3;
  #pragma omp target map(to: a) map(from: b) map(tofrom: x) map(tofrom: host_flag)
  {
    host_flag = (omp_is_initial_device() ? 1 : 0);
    #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
    for (int i = 1; i <= n; i++) {
      b[i - 1] = 2 * a[i - 1] + 1 + (x - (i + 2));
      x = x + 1;
    }
  }
  bool t1_ = true;
  for (int k2_ = 0; k2_ < n; k2_++)
    t1_ = t1_ && (b[k2_] == ref[k2_]);
  ok = t1_ && (x == n + 3);
  printf("RESULT ran_on_host=%d check=%s\n", host_flag, ok ? "pass" : "fail");
  return 0;
}

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

check=pass

Observed where it behaves incorrectly

site-gcc--gcc-c-omp60-unimplemented-rejected.c: In function ‘main’:
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:57: error: expected an OpenMP clause before ‘induction’
   96 |     #pragma omp parallel for default(none) shared(a, b) induction(step(1), +: x)
      |                                                         ^~~~~~~~~
site-gcc--gcc-c-omp60-unimplemented-rejected.c:98:40: error: ‘x’ not specified in enclosing ‘parallel’
   98 |       b[i - 1] = 2 * a[i - 1] + 1 + (x - (i + 2));
      |                                     ~~~^~~~~~~~~~
site-gcc--gcc-c-omp60-unimplemented-rejected.c:96:13: note: enclosing ‘parallel’

What the specification requires

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

OpenMP 6.0 §7.6.13 p.257

OpenMP 6.0 §5.4 p.173

OpenMP 6.0 §14.6 p.440

OpenMP 6.0 §17.9.6 p.510

Reproducers of this issue