[OpenMP] libomp: Add missing ompt callbacks in hier barrier reduction

[t-msn]

Summary

Some code path in hier barrier gather misses ompt callbacks in reduction.

[t-msn]

I noticed this when running check-openmp with hier barrier by using https://reviews.llvm.org/D122645

[protze.joachim]

Can you add extra run lines to openmp/runtime/test/ompt/synchronization/reduction/*_reduce.c in order to trigger the different barrier implementations?

[t-msn]

Can you add extra run lines to openmp/runtime/test/ompt/synchronization/reduction/*_reduce.c in order to trigger the different barrier implementations?

Ok.

This is a bit out of topic, but currently tree_reduce tests fails on dist barrier but I think this is a test problem.
As spec says:

If a task participates in multiple reductions, each reduction may be bracketed by its own pair of reduction-begin/reduction-end events or multiple reductions may be bracketed by a single pair of events.

I read this as reduction-begin & end event may occur 1 to ~ number of reductions. Dist barrier put 1 ompt event for reductions surrounding for-loop while other barriers call ompt event in each reduction
and therefore this results in mismatch in the number of called events. I don't know lit's CHECK pattern matching can handle to check if lines appear 1 ~ X times, let me think. (maybe just easier to fix dist barrier handling anyway)

[protze.joachim]

I agree, the expected output does not fit the general case of reductions.
As a workaround, I would change the check prefix for reduction 2-n from CHECK to MULTIPLE and add --check-prefixes CHECK,MULTIPLE for the RUN-lines, where multiple pairs of reduction callbacks are expected. So the default would be "at least one" pair of callbacks.

[t-msn]

Thanks for the suggestion. I tried if lit's regex can be used but got no luck. So, I decided to follow the advice.
I think it is better to separate patches and here is the patch to run barrier/reduction test for each barrier : D123359

openmp/runtime/src/kmp_barrier.cpp

if (__kmp_dflt_blocktime == KMP_MAX_BLOCKTIME &&
                         "T#%d(%d:%d) "
                         "arrived(%p) == %llu\n",
                         gtid, team->t.t_id, tid,
                         __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                         child_tid, &child_bar->b_arrived, new_state));
           kmp_flag_64<> flag(&child_bar->b_arrived, new_state);
           flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
           if (reduce) {
+            OMPT_REDUCTION_DECL(this_thr, gtid);
+            OMPT_REDUCTION_BEGIN;
             KA_TRACE(100, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) += "
                            "T#%d(%d:%d)\n",
                            gtid, team->t.t_id, tid,
                            __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                            child_tid));
             (*reduce)(this_thr->th.th_local.reduce_data,
                       child_thr->th.th_local.reduce_data);
+            OMPT_REDUCTION_END;
           }
         }
       }
     } else { // Blocktime is not infinite
       for (kmp_uint32 d = 0; d < thr_bar->my_level;
            ++d) { // Gather lowest level threads first
         kmp_uint32 last = tid + thr_bar->skip_per_level[d + 1],
                    skip = thr_bar->skip_per_level[d];
         if (last > nproc)
           last = nproc;
         for (child_tid = tid + skip; child_tid < (int)last; child_tid += skip) {
           kmp_info_t *child_thr = other_threads[child_tid];
           kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
           KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) wait "
                         "T#%d(%d:%d) "
                         "arrived(%p) == %llu\n",
                         gtid, team->t.t_id, tid,
                         __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                         child_tid, &child_bar->b_arrived, new_state));
           kmp_flag_64<> flag(&child_bar->b_arrived, new_state);
           flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
           if (reduce) {
+            OMPT_REDUCTION_DECL(this_thr, gtid);
+            OMPT_REDUCTION_BEGIN;
             KA_TRACE(100, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) += "
                            "T#%d(%d:%d)\n",
                            gtid, team->t.t_id, tid,
                            __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                            child_tid));
             (*reduce)(this_thr->th.th_local.reduce_data,
                       child_thr->th.th_local.reduce_data);
+            OMPT_REDUCTION_END;
           }
         }
       }
     }
   }
   // All subordinates are gathered; now release parent if not primary thread
 
   if (!KMP_MASTER_TID(tid)) { // worker threads release parent in hierarchy