OffTarget

Observed issues › catalog/cce20-c-defaultmap-private-storage-rejected

Appears not to supportRefused at compile time · C · OpenMP · Fixed in 21.0.3

craycc rejects defaultmap(private) (OpenMP 6.0)

Legality: independently confirmed. One independent reading was uncertain; 2 adjudicating readings that weighed both find the program legal and the result required: OpenMP 6.0 lists private among the implicit-behavior keywords of defaultmap, with variable-category defaulting to all. The program is therefore conforming and a 6.0 compiler must accept the keyword. a, b and host_flag are explicitly mapped. c and w are implicitly private on the target, so the host w stays 0 and b gets 2a+1, as expected. The refusal is a compiler error. How this is checked

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 24.3.0correctcorrectcorrectcorrectcorrectn/an/a
CCE 19.0.0not supportednot supportednot supportedunknown
not run
unknown
not run
n/an/a
CCE 20.0.2not supportednot supportednot supportedunknown
not run
unknown
not run
n/an/a
CCE 21.0.3correctcorrectcorrectunknown
not run
unknown
not run
n/an/a
CCE 21.0.4correctcorrectcorrectunknown
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 16.2.0fails another wayfails another wayfails another wayfails another wayfails another wayfails another wayn/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 supportedCCE 19.0.0, MI300A unified
CCE 19.0.0, MI300A copy
CCE 19.0.0, MI250X
CCE 20.0.2, MI300A unified
CCE 20.0.2, MI300A copy
CCE 20.0.2, MI250X
show
cce20-c-defaultmap-private-storage-rejected.c:24:80: error: expected 'alloc', 'from', 'to', 'tofrom', 'firstprivate', 'none', 'default', 'present' in OpenMP clause 'defaultmap'
   24 |   #pragma omp target map(to: a) map(from: b) map(tofrom: host_flag) defaultmap(private)
      |                                                                                ^
1 error generated.
fails another wayNVHPC 25.11, A100 separate
NVHPC 25.11, A100 managed
NVHPC 26.9, A100 separate
NVHPC 26.9, A100 managed
show
"cce20-c-defaultmap-private-storage-rejected.c", line 24: error: invalid text in pragma
    #pragma omp target map(to: a) map(from: b) map(tofrom: host_flag) defaultmap(private)
                                                                                 ^
1 error detected in the compilation of "cce20-c-defaultmap-private-storage-rejected.c".
fails another wayGCC 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
cce20-c-defaultmap-private-storage-rejected.c: In function ‘main’:
cce20-c-defaultmap-private-storage-rejected.c:24:80: error: expected ‘alloc’, ‘to’, ‘from’, ‘tofrom’, ‘firstprivate’, ‘none’ or ‘default’ before ‘private’
   24 |   #pragma omp target map(to: a) map(from: b) map(tofrom: host_flag) defaultmap(private)
      |                                                                                ^~~~~~~

Reproduce

CCE 19.0.0, MI300A unified

module load cce/19.0.0 craype-accel-amd-gfx942 rocm/6.4.3
module unload cray-mpich
craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -fopenmp-version=60 -O2 repro.c -o repro
HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro

CCE 20.0.2, MI300A unified

module load cce/20.0.2 craype-accel-amd-gfx942 rocm/6.4.3
module unload cray-mpich
craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -fopenmp-version=60 -O2 repro.c -o repro
HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro

AFAR 24.3.0, MI300A unified (runs correctly)

$AFAR/bin/amdclang -fopenmp --offload-arch=gfx942 -fopenmp-version=60 -fopenmp-version=60 -O2 repro.c -o repro
HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro
Program (80 lines)
// # Draft finding: omp60-c-clause-defaultmap-private on cce20
//
// - compiler: cce20
// - toolchain: cce20; arch: gfx942; machine: tuolumne; memory mode: unified; opt: -O2
// - program: catalog omp60-c-clause-defaultmap-private (c, version 60, sha256 09a36af3c3918990)
// - expected: None (None; )
// - actual: rejected  (diagnostic: <tmpdir>/omp60-c-clause-defaultmap-private.c:34:80: error: expected 'alloc', 'from', 'to', 'tofrom', 'firstprivate', 'none', 'default', 'present' in OpenMP clause 'defaultmap')
// - verdict: compiler_bug by rule R11-compile-rejection: a legal program the compiler rejects with a diagnostic (or whose link fails on symbols the compiler's own code needs, or whose compile does not finish) while another compiler family compiles it
// - reason: 
// - citations: RUL-COMPILE-TIME (rulings.md)
// - legality: legal (legal by construction: offload.engine.gen / c_gen / acc_gen check every program against the catalog rules and the offload scope as it is built (no compiler))
// - witnesses: none
// - withdrawn: none
//
// Generated by compliance.c_gen (OpenMP 6.0, C11): the C port (offload.engine.f2c) of omp60-clause-defaultmap-private. Do not edit.
// id: omp60-c-clause-defaultmap-private
// primary atom: omp.clause.defaultmap.private
// legal: yes
// observable: yes
// ran_on_host: inside the target region of interest
// evidence: construct
// needs: none
// std: c11
// C: the check also reads w after the region: the defaultmap category keeps the region's writes to the aggregate w private; without the clause w is mapped tofrom and they come back (the Fortran program checks b only)
#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 c;
  int w[n];
  for (int i = 1; i <= n; i++) {
    a[i - 1] = i;
    b[i - 1] = 0;
    ref[i - 1] = 2 * i + 1;
  }
  host_flag = 1;
  c = 0;
  for (int k1_ = 0; k1_ < n; k1_++)
    w[k1_] = 0;
  #pragma omp target map(to: a) map(from: b) map(tofrom: host_flag) defaultmap(private)
  {
    host_flag = (omp_is_initial_device() ? 1 : 0);
    c = 1;
    for (int i = 1; i <= n; i++) {
      w[i - 1] = 2 * a[i - 1];
      b[i - 1] = w[i - 1] + c;
    }
  }
  bool wz = true;
  for (int k_ = 0; k_ < n; k_++)
    wz = wz && (w[k_] == 0);
  bool t2_ = true;
  for (int k3_ = 0; k3_ < n; k3_++)
    t2_ = t2_ && (b[k3_] == ref[k3_]);
  ok = t2_ && wz;
  printf("RESULT ran_on_host=%d check=%s\n", host_flag, ok ? "pass" : "fail");
  return 0;
}
// spec (requirement): In OpenMP 6.0 the defaultmap clause sets the implicit data-mapping or data-sharing attribute of
//   the variables referenced in a target construct, and its implicit-behavior keywords include private.
// "The defaultmap clause controls the implicitly determined data-mapping attributes or implicitly determined
//   data-sharing attributes of certain variables that are referenced in a target construct" (6.0 §7.9.9 p.292 l.2-4)
// spec (requirement): A defaultmap clause with an implicit behavior other than default determines the attribute of
//   the variables of its category.
// "If a defaultmap clause is present for the category of the variable and specifies an implicit behavior other than
//   default, the data-mapping attribute or data-sharing attribute is determined by that clause." (6.0 §7.9.3 p.276
//   l.21-23)
// spec (this program): defaultmap(private) is a 6.0 form: the 6.0 argument table lists private among the
//   implicit-behavior keywords (a table row without line numbers, so it is not quoted), but the 6.0 semantics
//   paragraph of the clause does not describe it, so no sentence states what private does and the cited text is the
//   closest normative rule. The compiler rejects the keyword with a diagnostic that lists alloc, from, to, tofrom,
//   firstprivate, none, default and present.

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

cce20-c-defaultmap-private-storage-rejected.c:24:80: error: expected 'alloc', 'from', 'to', 'tofrom', 'firstprivate', 'none', 'default', 'present' in OpenMP clause 'defaultmap'
   24 |   #pragma omp target map(to: a) map(from: b) map(tofrom: host_flag) defaultmap(private)
      |                                                                                ^
1 error generated.

What the specification requires

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

OpenMP 6.0 §7.9.9 p.292 l.2-4

The defaultmap clause controls the implicitly determined data-mapping attributes or implicitly determined data-sharing attributes of certain variables that are referenced in a target construct

OpenMP 6.0 §7.9.3 p.276 l.21-23

If a defaultmap clause is present for the category of the variable and specifies an implicit behavior other than default, the data-mapping attribute or data-sharing attribute is determined by that clause.

Evidence

Same compiler, one change

  • defaultmap(firstprivate) instead of defaultmap(private): 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
    #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 c;
      int w[n];
      for (int i = 1; i <= n; i++) {
        a[i - 1] = i;
        b[i - 1] = 0;
        ref[i - 1] = 2 * i + 1;
      }
      host_flag = 1;
      c = 0;
      for (int k1_ = 0; k1_ < n; k1_++)
        w[k1_] = 0;
      #pragma omp target map(to: a) map(from: b) map(tofrom: host_flag) defaultmap(firstprivate)
      {
        host_flag = (omp_is_initial_device() ? 1 : 0);
        c = 1;
        for (int i = 1; i <= n; i++) {
          w[i - 1] = 2 * a[i - 1];
          b[i - 1] = w[i - 1] + c;
        }
      }
      bool wz = true;
      for (int k_ = 0; k_ < n; k_++)
        wz = wz && (w[k_] == 0);
      bool t2_ = true;
      for (int k3_ = 0; k3_ < n; k3_++)
        t2_ = t2_ && (b[k3_] == ref[k3_]);
      ok = t2_ && wz;
      printf("RESULT ran_on_host=%d check=%s\n", host_flag, ok ? "pass" : "fail");
      return 0;
    }
    

Other compilers run it correctly

AFAR 24.3.0