diff mbox series

[committed] openmp: Add omp_set_num_teams, omp_get_max_teams, omp_[gs]et_teams_thread_limit

Message ID 20211011102159.GT304296@tucnak
State New
Headers show
Series [committed] openmp: Add omp_set_num_teams, omp_get_max_teams, omp_[gs]et_teams_thread_limit | expand

Commit Message

Jakub Jelinek Oct. 11, 2021, 10:21 a.m. UTC
Hi!

OpenMP 5.1 adds env vars and functions to set and query new ICVs used
as fallback if thread_limit or num_teams clauses aren't specified on
teams construct.

The following patch implements those, though further work will be needed:
1) OpenMP 5.1 also changed the num_teams clause, so that it can specify
   both lower and upper limit for how many teams should be created and
   changed the meaning when only one expression is provided, instead of
   num_teams(expr) in 5.0 meaning num_teams(1:expr) in 5.1, it now means
   num_teams(expr:expr), i.e. while previously we could create 1 to expr
   teams, in 5.1 we have some low limit by default equal to the single
   expression provided and may not create fewer teams.
   For host teams (which we don't currently implement efficiently for
   NUMA hosts) we trivially satisfy it now by always honoring what the
   user asked for, but for the offloading teams I think we'll need to
   rethink the APIs; currently teams construct is just a call that returns
   and possibly lowers the number of teams; and whenever possible we try
   to evaluate num_teams/thread_limit already on the target construct
   and the GOMP_teams call just sets the number of teams to the minimum
   of provided and requested teams; for some cases e.g. where target
   is not combined with teams and num_teams expression calls some functions
   etc., we need to call those functions in the target region and so it is
   late to figure number of teams, but also hw could just limit what it
   is willing to create; in that case I'm afraid we need to run the target
   body multiple times and arrange for omp_get_team_num () returning the
   right values
2) we need to finally implement the NUMA handling for GOMP_teams_reg
3) I now realize I haven't added some testcase coverage, will do that
   incrementally
4) libgomp.texi needs updates for these new APIs, but also others like
   the allocator

Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk.

2021-10-11  Jakub Jelinek  <jakub@redhat.com>

gcc/
	* omp-low.c (omp_runtime_api_call): Handle omp_get_max_teams,
	omp_[sg]et_teams_thread_limit and omp_set_num_teams.
libgomp/
	* omp.h.in (omp_set_num_teams, omp_get_max_teams,
	omp_set_teams_thread_limit, omp_get_teams_thread_limit): Declare.
	* omp_lib.f90.in (omp_set_num_teams, omp_get_max_teams,
	omp_set_teams_thread_limit, omp_get_teams_thread_limit): Declare.
	* omp_lib.h.in (omp_set_num_teams, omp_get_max_teams,
	omp_set_teams_thread_limit, omp_get_teams_thread_limit): Declare.
	* libgomp.h (gomp_nteams_var, gomp_teams_thread_limit_var): Declare.
	* libgomp.map (OMP_5.1): Export omp_get_max_teams{,_},
	omp_get_teams_thread_limit{,_}, omp_set_num_teams{,_,_8_} and
	omp_set_teams_thread_limit{,_,_8_}.
	* icv.c (omp_set_num_teams, omp_get_max_teams,
	omp_set_teams_thread_limit, omp_get_teams_thread_limit): New
	functions.
	* env.c (gomp_nteams_var, gomp_teams_thread_limit_var): Define.
	(omp_display_env): Print OMP_NUM_TEAMS and OMP_TEAMS_THREAD_LIMIT.
	(initialize_env): Handle OMP_NUM_TEAMS and OMP_TEAMS_THREAD_LIMIT env
	vars.
	* teams.c (GOMP_teams_reg): If thread_limit is not specified, use
	gomp_teams_thread_limit_var as fallback if not zero.  If num_teams
	is not specified, use gomp_nteams_var.
	* fortran.c (omp_set_num_teams, omp_get_max_teams,
	omp_set_teams_thread_limit, omp_get_teams_thread_limit): Add
	ialias_redirect.
	(omp_set_num_teams_, omp_set_num_teams_8_, omp_get_max_teams_,
	omp_set_teams_thread_limit_, omp_set_teams_thread_limit_8_,
	omp_get_teams_thread_limit_): New functions.


	Jakub
diff mbox series

Patch

--- gcc/omp-low.c.jj	2021-09-30 17:12:15.236586906 +0200
+++ gcc/omp-low.c	2021-10-09 14:34:21.119388958 +0200
@@ -3953,6 +3953,7 @@  omp_runtime_api_call (const_tree fndecl)
       "get_level",
       "get_max_active_levels",
       "get_max_task_priority",
+      "get_max_teams",
       "get_max_threads",
       "get_nested",
       "get_num_devices",
@@ -3965,6 +3966,7 @@  omp_runtime_api_call (const_tree fndecl)
       "get_proc_bind",
       "get_supported_active_levels",
       "get_team_num",
+      "get_teams_thread_limit",
       "get_thread_limit",
       "get_thread_num",
       "get_wtick",
@@ -3998,8 +4000,10 @@  omp_runtime_api_call (const_tree fndecl)
       "set_dynamic",
       "set_max_active_levels",
       "set_nested",
+      "set_num_teams",
       "set_num_threads",
-      "set_schedule"
+      "set_schedule",
+      "set_teams_thread_limit"
     };
 
   int mode = 0;
--- libgomp/omp.h.in.jj	2021-10-01 10:32:03.024954096 +0200
+++ libgomp/omp.h.in	2021-10-09 15:06:38.173661594 +0200
@@ -261,6 +261,11 @@  extern int omp_get_max_task_priority (vo
 
 extern void omp_fulfill_event (omp_event_handle_t) __GOMP_NOTHROW;
 
+extern void omp_set_num_teams (int) __GOMP_NOTHROW;
+extern int omp_get_max_teams (void) __GOMP_NOTHROW;
+extern void omp_set_teams_thread_limit (int) __GOMP_NOTHROW;
+extern int omp_get_teams_thread_limit (void) __GOMP_NOTHROW;
+
 extern void *omp_target_alloc (__SIZE_TYPE__, int) __GOMP_NOTHROW;
 extern void omp_target_free (void *, int) __GOMP_NOTHROW;
 extern int omp_target_is_present (const void *, int) __GOMP_NOTHROW;
--- libgomp/omp_lib.f90.in.jj	2021-09-30 17:12:15.251586697 +0200
+++ libgomp/omp_lib.f90.in	2021-10-09 15:06:38.170661637 +0200
@@ -564,6 +564,36 @@ 
           end function omp_get_max_task_priority
         end interface
 
+        interface omp_set_num_teams
+          subroutine omp_set_num_teams (num_teams)
+            integer (4), intent (in) :: num_teams
+          end subroutine omp_set_num_teams
+          subroutine omp_set_num_teams_8 (num_teams)
+            integer (8), intent (in) :: num_teams
+          end subroutine omp_set_num_teams_8
+        end interface
+
+        interface
+          function omp_get_max_teams ()
+            integer (4) :: omp_get_max_teams
+          end function omp_get_max_teams
+        end interface
+
+        interface omp_set_teams_thread_limit
+          subroutine omp_set_teams_thread_limit (thread_limit)
+            integer (4), intent (in) :: thread_limit
+          end subroutine omp_set_teams_thread_limit
+          subroutine omp_set_teams_thread_limit_8 (thread_limit)
+            integer (8), intent (in) :: thread_limit
+          end subroutine omp_set_teams_thread_limit_8
+        end interface
+
+        interface
+          function omp_get_teams_thread_limit ()
+            integer (4) :: omp_get_teams_thread_limit
+          end function omp_get_teams_thread_limit
+        end interface
+
         interface
           subroutine omp_fulfill_event (event)
             use omp_lib_kinds
--- libgomp/omp_lib.h.in.jj	2021-09-30 17:12:15.257586613 +0200
+++ libgomp/omp_lib.h.in	2021-10-09 15:06:38.167661680 +0200
@@ -252,6 +252,10 @@ 
       external omp_get_max_task_priority
       integer(4) omp_get_max_task_priority
 
+      external omp_set_num_teams, omp_set_teams_thread_limit
+      external omp_get_max_teams, omp_get_teams_thread_limit
+      integer(4) omp_get_max_teams, omp_get_teams_thread_limit
+
       external omp_fulfill_event
 
       external omp_set_affinity_format, omp_get_affinity_format
--- libgomp/libgomp.h.jj	2021-07-28 12:06:00.535928239 +0200
+++ libgomp/libgomp.h	2021-10-09 15:06:38.154661866 +0200
@@ -458,6 +458,8 @@  extern unsigned long gomp_bind_var_list_
 extern void **gomp_places_list;
 extern unsigned long gomp_places_list_len;
 extern unsigned int gomp_num_teams_var;
+extern int gomp_nteams_var;
+extern int gomp_teams_thread_limit_var;
 extern int gomp_debug_var;
 extern bool gomp_display_affinity_var;
 extern char *gomp_affinity_format_var;
--- libgomp/libgomp.map.jj	2021-09-30 09:29:56.739900855 +0200
+++ libgomp/libgomp.map	2021-10-09 15:06:38.151661909 +0200
@@ -214,6 +214,16 @@  OMP_5.1 {
 	omp_display_env;
 	omp_display_env_;
 	omp_display_env_8_;
+	omp_set_num_teams;
+	omp_set_num_teams_;
+	omp_set_num_teams_8_;
+	omp_get_max_teams;
+	omp_get_max_teams_;
+	omp_set_teams_thread_limit;
+	omp_set_teams_thread_limit_;
+	omp_set_teams_thread_limit_8_;
+	omp_get_teams_thread_limit;
+	omp_get_teams_thread_limit_;
 } OMP_5.0.2;
 
 GOMP_1.0 {
--- libgomp/icv.c.jj	2021-10-01 10:42:04.046441605 +0200
+++ libgomp/icv.c	2021-10-09 15:06:38.157661823 +0200
@@ -148,6 +148,32 @@  omp_get_supported_active_levels (void)
   return gomp_supported_active_levels;
 }
 
+void
+omp_set_num_teams (int num_teams)
+{
+  if (num_teams >= 0)
+    gomp_nteams_var = num_teams;
+}
+
+int
+omp_get_max_teams (void)
+{
+  return gomp_nteams_var;
+}
+
+void
+omp_set_teams_thread_limit (int thread_limit)
+{
+  if (thread_limit >= 0)
+    gomp_teams_thread_limit_var = thread_limit;
+}
+
+int
+omp_get_teams_thread_limit (void)
+{
+  return gomp_teams_thread_limit_var;
+}
+
 int
 omp_get_cancellation (void)
 {
@@ -248,6 +274,10 @@  ialias (omp_get_thread_limit)
 ialias (omp_set_max_active_levels)
 ialias (omp_get_max_active_levels)
 ialias (omp_get_supported_active_levels)
+ialias (omp_set_num_teams)
+ialias (omp_get_max_teams)
+ialias (omp_set_teams_thread_limit)
+ialias (omp_get_teams_thread_limit)
 ialias (omp_get_cancellation)
 ialias (omp_get_proc_bind)
 ialias (omp_get_max_task_priority)
--- libgomp/env.c.jj	2021-10-01 10:42:04.011442100 +0200
+++ libgomp/env.c	2021-10-09 15:06:38.160661780 +0200
@@ -90,6 +90,8 @@  unsigned long gomp_places_list_len;
 uintptr_t gomp_def_allocator = omp_default_mem_alloc;
 int gomp_debug_var;
 unsigned int gomp_num_teams_var;
+int gomp_nteams_var;
+int gomp_teams_thread_limit_var;
 bool gomp_display_affinity_var;
 char *gomp_affinity_format_var = "level %L thread %i affinity %A";
 size_t gomp_affinity_format_len;
@@ -1319,6 +1321,9 @@  omp_display_env (int verbose)
 	   gomp_global_icv.thread_limit_var);
   fprintf (stderr, "  OMP_MAX_ACTIVE_LEVELS = '%u'\n",
 	   gomp_global_icv.max_active_levels_var);
+  fprintf (stderr, "  OMP_NUM_TEAMS = '%u'\n", gomp_nteams_var);
+  fprintf (stderr, "  OMP_TEAMS_THREAD_LIMIT = '%u'\n",
+	   gomp_teams_thread_limit_var);
 
   fprintf (stderr, "  OMP_CANCELLATION = '%s'\n",
 	   gomp_cancel_var ? "TRUE" : "FALSE");
@@ -1453,6 +1458,8 @@  initialize_env (void)
 				 &gomp_nthreads_var_list,
 				 &gomp_nthreads_var_list_len))
     gomp_global_icv.nthreads_var = gomp_available_cpus;
+  parse_int ("OMP_NUM_TEAMS", &gomp_nteams_var, false);
+  parse_int ("OMP_TEAMS_THREAD_LIMIT", &gomp_teams_thread_limit_var, false);
   bool ignore = false;
   if (parse_bind_var ("OMP_PROC_BIND",
 		      &gomp_global_icv.bind_var,
--- libgomp/teams.c.jj	2021-01-04 10:25:56.062038735 +0100
+++ libgomp/teams.c	2021-10-09 15:06:38.148661952 +0200
@@ -37,6 +37,8 @@  GOMP_teams_reg (void (*fn) (void *), voi
   (void) flags;
   (void) num_teams;
   unsigned old_thread_limit_var = 0;
+  if (thread_limit == 0)
+    thread_limit = gomp_teams_thread_limit_var;
   if (thread_limit)
     {
       struct gomp_task_icv *icv = gomp_icv (true);
@@ -45,7 +47,7 @@  GOMP_teams_reg (void (*fn) (void *), voi
 	= thread_limit > INT_MAX ? UINT_MAX : thread_limit;
     }
   if (num_teams == 0)
-    num_teams = 3;
+    num_teams = gomp_nteams_var ? gomp_nteams_var : 3;
   gomp_num_teams = num_teams;
   for (gomp_team_num = 0; gomp_team_num < num_teams; gomp_team_num++)
     fn (data);
--- libgomp/fortran.c.jj	2021-10-01 10:42:04.034441774 +0200
+++ libgomp/fortran.c	2021-10-09 15:06:38.163661737 +0200
@@ -67,6 +67,10 @@  ialias_redirect (omp_get_thread_limit)
 ialias_redirect (omp_set_max_active_levels)
 ialias_redirect (omp_get_max_active_levels)
 ialias_redirect (omp_get_supported_active_levels)
+ialias_redirect (omp_set_num_teams)
+ialias_redirect (omp_get_max_teams)
+ialias_redirect (omp_set_teams_thread_limit)
+ialias_redirect (omp_get_teams_thread_limit)
 ialias_redirect (omp_get_level)
 ialias_redirect (omp_get_ancestor_thread_num)
 ialias_redirect (omp_get_team_size)
@@ -478,6 +482,42 @@  omp_in_final_ (void)
   return omp_in_final ();
 }
 
+void
+omp_set_num_teams_ (const int32_t *num_teams)
+{
+  omp_set_num_teams (*num_teams);
+}
+
+void
+omp_set_num_teams_8_ (const int64_t *num_teams)
+{
+  omp_set_max_active_levels (TO_INT (*num_teams));
+}
+
+int32_t
+omp_get_max_teams_ (void)
+{
+  return omp_get_max_teams ();
+}
+
+void
+omp_set_teams_thread_limit_ (const int32_t *thread_limit)
+{
+  omp_set_teams_thread_limit (*thread_limit);
+}
+
+void
+omp_set_teams_thread_limit_8_ (const int64_t *thread_limit)
+{
+  omp_set_teams_thread_limit (TO_INT (*thread_limit));
+}
+
+int32_t
+omp_get_teams_thread_limit_ (void)
+{
+  return omp_get_teams_thread_limit ();
+}
+
 int32_t
 omp_get_cancellation_ (void)
 {