public inbox for gcc-bugs@sourceware.org help / color / mirror / Atom feed
From: "burnus at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org> To: gcc-bugs@gcc.gnu.org Subject: [Bug libgomp/109875] New: [OpenMP] nteams-var / OMP_NUM_TEAMS → ICV not passed to the device / default value Date: Tue, 16 May 2023 17:39:28 +0000 [thread overview] Message-ID: <bug-109875-4@http.gcc.gnu.org/bugzilla/> (raw) https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109875 Bug ID: 109875 Summary: [OpenMP] nteams-var / OMP_NUM_TEAMS → ICV not passed to the device / default value Product: gcc Version: 13.0 Status: UNCONFIRMED Keywords: openmp Severity: normal Priority: P3 Component: libgomp Assignee: unassigned at gcc dot gnu.org Reporter: burnus at gcc dot gnu.org CC: jakub at gcc dot gnu.org Target Milestone: --- Related: https://github.com/SOLLVE/sollve_vv/pull/728 and OpenMP Spec Issue 3619. The following is with host fallback - and I might miss something because of this. However, I get only a single team with the following code; even though max_num_teams == 3. Replacing the set_num_teams by OMP_NUM_TEAMS=7 OMP_NUM_TEAMS_DEV=8 will yield the expected 7 for max_teams. But it will still execute with a single team. — Using num_teams(3) as clause will have an effect, however. TODO: * Check whether TEAMS = 1 makes sense for host fallback and what the result is with actual offloading. EXPECTED: * The value of the ICV is honored on the device side * Possibly, the initial value should be not 0 but a different value, but that depends on the OpenMP Spec Issue 3619. NOTE: While the init value is now changed, as omp_get_max_teams() is confusingly written, just changing the default is undetectable inside the program as only omp_get_max_teams() or display-env can be used to determine the value - and the values returned by the latter is not really available for consumption inside of the program. * On the host side, the default seems to be 3, given that there is: libgomp/teams.c: num_teams = gomp_nteams_var ? gomp_nteams_var : 3; It seems as if get_gomp_offload_icvs sets the ICV for the device but it does not seem to get actually get used. – At least not for host fallback. #include <omp.h> int main () { int num_teams; #if 1 // set to 0 to test the environment variables omp_set_num_teams(2); #pragma omp target omp_set_num_teams(3); #endif __builtin_printf("max_teams: %d\n", omp_get_max_teams()); #pragma omp target teams map(tofrom: num_teams) //num_teams(4) { if (omp_get_team_num() == 0) { num_teams = omp_get_num_teams(); } } __builtin_printf("num_teams: %d\n", num_teams); return 0; }
next reply other threads:[~2023-05-16 17:39 UTC|newest] Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top 2023-05-16 17:39 burnus at gcc dot gnu.org [this message] 2023-05-17 9:47 ` [Bug libgomp/109875] " burnus at gcc dot gnu.org 2023-05-17 10:11 ` burnus at gcc dot gnu.org 2023-05-21 18:44 ` cvs-commit at gcc dot gnu.org 2023-05-21 18:56 ` burnus at gcc dot gnu.org
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=bug-109875-4@http.gcc.gnu.org/bugzilla/ \ --to=gcc-bugzilla@gcc.gnu.org \ --cc=gcc-bugs@gcc.gnu.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).