[Bug libgomp/100573] [OpenMP] 'omp target teams' fails with nvptx and GCN offloading: FAIL libgomp.c-c++-common/for-3.c + for-9.c

[vries at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573

--- Comment #3 from Tom de Vries <vries at gcc dot gnu.org> ---
Hmm, I reproduced the problem on the original test-case:
libgomp.c-c++-common/for-3.c, and minimized from there:
...
$ cat libgomp/testsuite/libgomp.c-c++-common/for-3.c
/* { dg-additional-options "-std=gnu99" { target c } } */

#include <stdio.h>
extern void abort ();

#define N 1500

int a[N];

int
main (void)
{
  int err = 0;

#pragma omp target map(tofrom: err) map (tofrom: a)
#pragma omp teams num_teams (15) reduction(|:err)
  {
    int i;

    for (i = 0; i < N; i++)
      a[i] = 0;

    do {} while (0);

#pragma omp distribute
    for (i = 0; i < N; i++)
      a[i] += 2;

    do {} while (0);

    for (i = 0; i < N; i++)
      if (a[i] != 2)
        err |= 1;
  }

  int i;
  for (i = 0; i < N; i++)
    if (a[i] != 2)
      printf ("a[%d]: %d\n", i, a[i]);

  if (err)
    abort ();

  return 0;
}
...

Again, I think there's a problem with the test-case.

There's no implicit barrier at the end of a distribute construct, so we can
have:
- team 0 updates part of array
- team 0 checks entire array, finds that only it's part is done, sets error
  flag
- team 1 updates part of array ...

I wonder whether there's an implicit barrier at the start of the distribute
construct.  If not, then the same problem exists for the initialization.