https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84041
--- Comment #2 from Tom de Vries <vries at gcc dot gnu.org> ---
More minimal, using num_teams and num_threads:
...
int
main (void)
{
#pragma omp target teams num_teams(1)
{
unsigned int i;
#pragma omp distribute parallel for schedule(dynamic,1) num_threads (2)
for (i = 0;
i < 1;
i += 1)
;
}
return 0;
}
...
I'm not sure how the openmp mapping goes, but with num_threads (2) we still
seem to be launching with 8 threads:
...
GOMP_OFFLOAD_run: kernel main$_omp_fn$0: launch [(teams: 1), 1, 1] [32,
(threads: 8), 1]
...
Using num_threads(1) results in the same launch dimentions, but the testcase
does not hang.
