[OpenMP] libomp: add itt notifications for teams construct on host

Add barrier/region notification for parallel inside teams construct
when number of teams is 1, as VTune only shows outer level regions for
simplicity.

Differential Revision: https://reviews.llvm.org/D84024

diff --git a/openmp/runtime/src/kmp_barrier.cpp b/openmp/runtime/src/kmp_barrier.cpp
index 4aa7a084f53af1f19390b4d70a34dc2915cdd442..de661288bab308488e53b508930b43705b33855d 100644 (file)
--- a/openmp/runtime/src/kmp_barrier.cpp
+++ b/openmp/runtime/src/kmp_barrier.cpp
@@ -1453,7 +1453,8 @@ static int __kmp_barrier_template(enum barrier_type bt, int gtid, int is_split,
       // Barrier - report frame end (only if active_level == 1)
       if ((__itt_frame_submit_v3_ptr || KMP_ITT_DEBUG) &&
           __kmp_forkjoin_frames_mode &&
-          this_thr->th.th_teams_microtask == NULL &&
+          (this_thr->th.th_teams_microtask == NULL || // either not in teams
+           this_thr->th.th_teams_size.nteams == 1) && // or inside single team
           team->t.t_active_level == 1) {
         ident_t *loc = __kmp_threads[gtid]->th.th_ident;
         kmp_uint64 cur_time = __itt_get_timestamp();
@@ -1839,7 +1840,9 @@ void __kmp_join_barrier(int gtid) {
 #if USE_ITT_BUILD && USE_ITT_NOTIFY
     // Join barrier - report frame end
     if ((__itt_frame_submit_v3_ptr || KMP_ITT_DEBUG) &&
-        __kmp_forkjoin_frames_mode && this_thr->th.th_teams_microtask == NULL &&
+        __kmp_forkjoin_frames_mode &&
+        (this_thr->th.th_teams_microtask == NULL || // either not in teams
+         this_thr->th.th_teams_size.nteams == 1) && // or inside single team
         team->t.t_active_level == 1) {
       kmp_uint64 cur_time = __itt_get_timestamp();
       ident_t *loc = team->t.t_ident;

diff --git a/openmp/runtime/src/kmp_runtime.cpp b/openmp/runtime/src/kmp_runtime.cpp
index a53920436901f29f1fee0540ea54d184c616b23a..00e0c7db4536f2f799f5aeb9b5082cec83c72b79 100644 (file)
--- a/openmp/runtime/src/kmp_runtime.cpp
+++ b/openmp/runtime/src/kmp_runtime.cpp
@@ -1625,6 +1625,22 @@ int __kmp_fork_call(ident_t *loc, int gtid,
       }
 #endif
 
+#if USE_ITT_BUILD
+      if (((__itt_frame_submit_v3_ptr && __itt_get_timestamp_ptr) ||
+           KMP_ITT_DEBUG) &&
+          __kmp_forkjoin_frames_mode == 3 &&
+          parent_team->t.t_active_level == 1 // only report frames at level 1
+          && master_th->th.th_teams_size.nteams == 1) {
+        kmp_uint64 tmp_time = __itt_get_timestamp();
+        master_th->th.th_frame_time = tmp_time;
+        parent_team->t.t_region_time = tmp_time;
+      }
+      if (__itt_stack_caller_create_ptr) {
+        // create new stack stitching id before entering fork barrier
+        parent_team->t.t_stack_id = __kmp_itt_stack_caller_create();
+      }
+#endif /* USE_ITT_BUILD */
+
       KF_TRACE(10, ("__kmp_fork_call: before internal fork: root=%p, team=%p, "
                     "master_th=%p, gtid=%d\n",
                     root, parent_team, master_th, gtid));
@@ -2367,14 +2383,13 @@ void __kmp_join_call(ident_t *loc, int gtid
 
 #if USE_ITT_BUILD
   if (__itt_stack_caller_create_ptr) {
-    __kmp_itt_stack_caller_destroy(
-        (__itt_caller)team->t
-            .t_stack_id); // destroy the stack stitching id after join barrier
+    // destroy the stack stitching id after join barrier
+    __kmp_itt_stack_caller_destroy((__itt_caller)team->t.t_stack_id);
   }
-
   // Mark end of "parallel" region for Intel(R) VTune(TM) analyzer.
   if (team->t.t_active_level == 1 &&
-      !master_th->th.th_teams_microtask) { /* not in teams construct */
+      (!master_th->th.th_teams_microtask || /* not in teams construct */
+       master_th->th.th_teams_size.nteams == 1)) {
     master_th->th.th_ident = loc;
     // only one notification scheme (either "submit" or "forking/joined", not
     // both)