[OpenMP][OMPT] Indicate loop schedule for worksharing-loop events (#9…

…7429)

Use more specific values from `ompt_work_t` to allow the tool identify
the schedule of a worksharing-loop. With this patch, the runtime will
report the schedule chosen by the runtime rather than necessarily the
schedule literally requested by the clause.
E.g., for guided + just one iteration per thread, the runtime would
choose and report static.

Fixes issue #63904

openmp/runtime/src/kmp_csupport.cpp

@@ -2006,13 +2006,13 @@ void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid) {
 
 #if OMPT_SUPPORT && OMPT_OPTIONAL
   if (ompt_enabled.ompt_callback_work) {
-    ompt_work_t ompt_work_type = ompt_work_loop;
+    ompt_work_t ompt_work_type = ompt_work_loop_static;
     ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
     ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
     // Determine workshare type
     if (loc != NULL) {
       if ((loc->flags & KMP_IDENT_WORK_LOOP) != 0) {
-        ompt_work_type = ompt_work_loop;
+        ompt_work_type = ompt_work_loop_static;
       } else if ((loc->flags & KMP_IDENT_WORK_SECTIONS) != 0) {
         ompt_work_type = ompt_work_sections;
       } else if ((loc->flags & KMP_IDENT_WORK_DISTRIBUTE) != 0) {

openmp/runtime/src/kmp_dispatch.cpp

@@ -1164,8 +1164,9 @@ __kmp_dispatch_init(ident_t *loc, int gtid, enum sched_type schedule, T lb,
     ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
     ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
     ompt_callbacks.ompt_callback(ompt_callback_work)(
-        ompt_work_loop, ompt_scope_begin, &(team_info->parallel_data),
-        &(task_info->task_data), pr->u.p.tc, OMPT_LOAD_RETURN_ADDRESS(gtid));
+        ompt_get_work_schedule(pr->schedule), ompt_scope_begin,
+        &(team_info->parallel_data), &(task_info->task_data), pr->u.p.tc,
+        OMPT_LOAD_RETURN_ADDRESS(gtid));
   }
 #endif
   KMP_PUSH_PARTITIONED_TIMER(OMP_loop_dynamic);
@@ -2121,8 +2122,8 @@ int __kmp_dispatch_next_algorithm(int gtid,
       ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);              \
       ompt_task_info_t *task_info = __ompt_get_task_info_object(0);            \
       ompt_callbacks.ompt_callback(ompt_callback_work)(                        \
-          ompt_work_loop, ompt_scope_end, &(team_info->parallel_data),         \
-          &(task_info->task_data), 0, codeptr);                                \
+          ompt_get_work_schedule(pr->schedule), ompt_scope_end,                \
+          &(team_info->parallel_data), &(task_info->task_data), 0, codeptr);   \
     }                                                                          \
   }
 #define OMPT_LOOP_DISPATCH(lb, ub, st, status)                                 \

openmp/runtime/src/kmp_sched.cpp

@@ -103,7 +103,7 @@ static void __kmp_for_static_init(ident_t *loc, kmp_int32 global_tid,
 #if OMPT_SUPPORT && OMPT_OPTIONAL
   ompt_team_info_t *team_info = NULL;
   ompt_task_info_t *task_info = NULL;
-  ompt_work_t ompt_work_type = ompt_work_loop;
+  ompt_work_t ompt_work_type = ompt_work_loop_static;
 
   static kmp_int8 warn = 0;
 
@@ -114,7 +114,7 @@ static void __kmp_for_static_init(ident_t *loc, kmp_int32 global_tid,
     // Determine workshare type
     if (loc != NULL) {
       if ((loc->flags & KMP_IDENT_WORK_LOOP) != 0) {
-        ompt_work_type = ompt_work_loop;
+        ompt_work_type = ompt_work_loop_static;
       } else if ((loc->flags & KMP_IDENT_WORK_SECTIONS) != 0) {
         ompt_work_type = ompt_work_sections;
       } else if ((loc->flags & KMP_IDENT_WORK_DISTRIBUTE) != 0) {

openmp/runtime/src/ompt-specific.h

@@ -130,6 +130,25 @@ inline const char *ompt_get_runtime_version() {
   return &__kmp_version_lib_ver[KMP_VERSION_MAGIC_LEN];
 }
 
+inline ompt_work_t ompt_get_work_schedule(enum sched_type schedule) {
+  switch (SCHEDULE_WITHOUT_MODIFIERS(schedule)) {
+  case kmp_sch_static_chunked:
+  case kmp_sch_static_balanced:
+  case kmp_sch_static_greedy:
+    return ompt_work_loop_static;
+  case kmp_sch_dynamic_chunked:
+  case kmp_sch_static_steal:
+    return ompt_work_loop_dynamic;
+  case kmp_sch_guided_iterative_chunked:
+  case kmp_sch_guided_analytical_chunked:
+  case kmp_sch_guided_chunked:
+  case kmp_sch_guided_simd:
+    return ompt_work_loop_guided;
+  default:
+    return ompt_work_loop_other;
+  }
+}
+
 class OmptReturnAddressGuard {
 private:
   bool SetAddress{false};

openmp/runtime/test/ompt/callback.h

@@ -47,6 +47,36 @@ static const char* ompt_cancel_flag_t_values[] = {
   "ompt_cancel_discarded_task"
 };
 
+static const char *ompt_work_t_values[] = {"undefined",
+                                           "ompt_work_loop",
+                                           "ompt_work_sections",
+                                           "ompt_work_single_executor",
+                                           "ompt_work_single_other",
+                                           "ompt_work_workshare",
+                                           "ompt_work_distribute",
+                                           "ompt_work_taskloop",
+                                           "ompt_work_scope",
+                                           "ompt_work_workdistribute",
+                                           "ompt_work_loop_static",
+                                           "ompt_work_loop_dynamic",
+                                           "ompt_work_loop_guided",
+                                           "ompt_work_loop_other"};
+
+static const char *ompt_work_events_t_values[] = {"undefined",
+                                                  "ompt_event_loop",
+                                                  "ompt_event_sections",
+                                                  "ompt_event_single_in_block",
+                                                  "ompt_event_single_others",
+                                                  "ompt_event_workshare",
+                                                  "ompt_event_distribute",
+                                                  "ompt_event_taskloop",
+                                                  "ompt_event_scope",
+                                                  "ompt_event_workdistribute",
+                                                  "ompt_event_loop_static",
+                                                  "ompt_event_loop_dynamic",
+                                                  "ompt_event_loop_guided",
+                                                  "ompt_event_loop_other"};
+
 static const char *ompt_dependence_type_t_values[36] = {
     "ompt_dependence_type_UNDEFINED",
     "ompt_dependence_type_in", // 1
@@ -852,144 +882,21 @@ on_ompt_callback_work(
 {
   switch(endpoint)
   {
-    case ompt_scope_begin:
-      switch(wstype)
-      {
-        case ompt_work_loop:
-        case ompt_work_loop_static:
-        case ompt_work_loop_dynamic:
-        case ompt_work_loop_guided:
-        case ompt_work_loop_other:
-        // TODO: add schedule attribute for the different work_loop types.
-        // e.g., ", schedule=%s", ..., ompt_schedule_values[wstype]
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_loop_begin: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_sections:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_sections_begin: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_single_executor:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_single_in_block_begin: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_single_other:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_single_others_begin: parallel_id=%" PRIu64
-                 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_workshare:
-          //impl
-          break;
-        case ompt_work_distribute:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_distribute_begin: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_taskloop:
-          //impl
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_taskloop_begin: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_scope:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_scope_begin: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-      }
-      break;
-    case ompt_scope_end:
-      switch(wstype)
-      {
-        case ompt_work_loop:
-        case ompt_work_loop_static:
-        case ompt_work_loop_dynamic:
-        case ompt_work_loop_guided:
-        case ompt_work_loop_other:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_loop_end: parallel_id=%" PRIu64
-                 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_sections:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_sections_end: parallel_id=%" PRIu64
-                 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_single_executor:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_single_in_block_end: parallel_id=%" PRIu64
-                 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_single_other:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_single_others_end: parallel_id=%" PRIu64
-                 ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_workshare:
-          //impl
-          break;
-        case ompt_work_distribute:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_distribute_end: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_taskloop:
-          //impl
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_taskloop_end: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-        case ompt_work_scope:
-          printf("%" PRIu64 ":" _TOOL_PREFIX
-                 " ompt_event_scope_end: parallel_id=%" PRIu64
-                 ", parent_task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64
-                 "\n",
-                 ompt_get_thread_data()->value, parallel_data->value,
-                 task_data->value, codeptr_ra, count);
-          break;
-      }
-      break;
-    case ompt_scope_beginend:
-      printf("ompt_scope_beginend should never be passed to %s\n", __func__);
-      exit(-1);
+  case ompt_scope_begin:
+    printf("%" PRIu64 ":" _TOOL_PREFIX " %s_begin: parallel_id=%" PRIu64
+           ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
+           ompt_get_thread_data()->value, ompt_work_events_t_values[wstype],
+           parallel_data->value, task_data->value, codeptr_ra, count);
+    break;
+  case ompt_scope_end:
+    printf("%" PRIu64 ":" _TOOL_PREFIX " %s_end: parallel_id=%" PRIu64
+           ", task_id=%" PRIu64 ", codeptr_ra=%p, count=%" PRIu64 "\n",
+           ompt_get_thread_data()->value, ompt_work_events_t_values[wstype],
+           parallel_data->value, task_data->value, codeptr_ra, count);
+    break;
+  case ompt_scope_beginend:
+    printf("ompt_scope_beginend should never be passed to %s\n", __func__);
+    exit(-1);
   }
 }
 

openmp/runtime/test/ompt/synchronization/ordered_dependences.c

@@ -7,7 +7,7 @@
 int main() {
   int a[10][10];
 #pragma omp parallel num_threads(2)
-#pragma omp for ordered(2)
+#pragma omp for ordered(2) schedule(static)
   for (int i = 0; i < 2; i++)
     for (int j = 0; j < 2; j++) {
       a[i][j] = i + j + 1;
@@ -23,8 +23,8 @@ int main() {
 }
 // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
 
-// CHECK: {{^}}[[MASTER:[0-9]+]]: ompt_event_loop_begin:
-// CHECK-SAME: parallel_id={{[0-9]+}}, parent_task_id=[[ITASK:[0-9]+]],
+// CHECK: {{^}}[[MASTER:[0-9]+]]: ompt_event_loop_static_begin:
+// CHECK-SAME: parallel_id={{[0-9]+}}, task_id=[[ITASK:[0-9]+]],
 
 // CHECK: {{^}}[[MASTER]]: ompt_event_dependences: task_id=[[ITASK]],
 // CHECK-SAME: deps=[(0, ompt_dependence_type_source), (0,
@@ -38,8 +38,8 @@ int main() {
 // CHECK-SAME: deps=[(0, ompt_dependence_type_source), (1,
 // CHECK-SAME: ompt_dependence_type_source)], ndeps=2
 
-// CHECK: {{^}}[[WORKER:[0-9]+]]: ompt_event_loop_begin:
-// CHECK-SAME: parallel_id={{[0-9]+}}, parent_task_id=[[ITASK:[0-9]+]],
+// CHECK: {{^}}[[WORKER:[0-9]+]]: ompt_event_loop_static_begin:
+// CHECK-SAME: parallel_id={{[0-9]+}}, task_id=[[ITASK:[0-9]+]],
 
 // CHECK: {{^}}[[WORKER]]: ompt_event_dependences: task_id=[[ITASK]],
 // CHECK-SAME: deps=[(0, ompt_dependence_type_sink), (0,

openmp/runtime/test/ompt/tasks/taskloop.c

@@ -37,7 +37,7 @@ int main() {
   // CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID1]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_taskloop_begin:
   // CHECK-SAME: parallel_id=[[PARALLEL_ID]]
-  // CHECK-SAME: parent_task_id=[[IMPLICIT_TASK_ID1]]
+  // CHECK-SAME: task_id=[[IMPLICIT_TASK_ID1]]
   // CHECK-SAME: codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]], count=2
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create:
   // CHECK-SAME: parent_task_id=[[IMPLICIT_TASK_ID1]]
@@ -52,7 +52,7 @@ int main() {
   // CHECK-NOT: {{^}}[[MASTER_ID]]: ompt_event_task_create:
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_taskloop_end:
   // CHECK-SAME: parallel_id=[[PARALLEL_ID]]
-  // CHECK-SAME: parent_task_id=[[IMPLICIT_TASK_ID1]]
+  // CHECK-SAME: task_id=[[IMPLICIT_TASK_ID1]]
   // CHECK-SAME: count=2
   // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_wait_taskgroup_begin:
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_wait_taskgroup_end:

openmp/runtime/test/ompt/tasks/taskloop_dispatch.c

@@ -32,7 +32,7 @@ int main() {
 
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_taskloop_begin:
   // CHECK-SAME: parallel_id=[[PARALLEL_ID]]
-  // CHECK-SAME: parent_task_id=[[IMPLICIT_TASK_ID1]]
+  // CHECK-SAME: task_id=[[IMPLICIT_TASK_ID1]]
   // CHECK-SAME: codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]], count=16
 
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_task_create:

openmp/runtime/test/ompt/teams/distribute_dispatch.c

@@ -24,28 +24,28 @@ int main() {
 
 // CHECK: {{^}}[[THREAD_ID0:[0-9]+]]: ompt_event_distribute_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID0:[0-9]+]]
-// CHECK-SAME: parent_task_id=[[TASK_ID0:[0-9]+]]
+// CHECK-SAME: task_id=[[TASK_ID0:[0-9]+]]
 // CHECK: {{^}}[[THREAD_ID0]]: ompt_event_distribute_chunk_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID0]], task_id=[[TASK_ID0]]
 // CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
 
 // CHECK: {{^}}[[THREAD_ID1:[0-9]+]]: ompt_event_distribute_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID1:[0-9]+]]
-// CHECK-SAME: parent_task_id=[[TASK_ID1:[0-9]+]]
+// CHECK-SAME: task_id=[[TASK_ID1:[0-9]+]]
 // CHECK: {{^}}[[THREAD_ID1]]: ompt_event_distribute_chunk_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID1]], task_id=[[TASK_ID1]]
 // CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
 
 // CHECK: {{^}}[[THREAD_ID2:[0-9]+]]: ompt_event_distribute_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID2:[0-9]+]]
-// CHECK-SAME: parent_task_id=[[TASK_ID2:[0-9]+]]
+// CHECK-SAME: task_id=[[TASK_ID2:[0-9]+]]
 // CHECK: {{^}}[[THREAD_ID2]]: ompt_event_distribute_chunk_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID2]], task_id=[[TASK_ID2]]
 // CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
 
 // CHECK: {{^}}[[THREAD_ID3:[0-9]+]]: ompt_event_distribute_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID3:[0-9]+]]
-// CHECK-SAME: parent_task_id=[[TASK_ID3:[0-9]+]]
+// CHECK-SAME: task_id=[[TASK_ID3:[0-9]+]]
 // CHECK: {{^}}[[THREAD_ID3]]: ompt_event_distribute_chunk_begin:
 // CHECK-SAME: parallel_id=[[PARALLEL_ID3]], task_id=[[TASK_ID3]]
 // CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16

openmp/runtime/test/ompt/worksharing/for/auto.c

@@ -4,4 +4,7 @@
 // XFAIL: gcc
 
 #define SCHEDULE auto
+// The runtime uses guided schedule for auto,
+// which is a reason choice
+#define SCHED_OUTPUT "guided"
 #include "base.h"

openmp/runtime/test/ompt/worksharing/for/auto_serialized.c

@@ -4,4 +4,7 @@
 // XFAIL: gcc
 
 #define SCHEDULE auto
+// The runtime uses static schedule for serialized loop,
+// which is a reason choice
+#define SCHED_OUTPUT "static"
 #include "base_serialized.h"

openmp/runtime/test/ompt/worksharing/for/auto_split.c

@@ -5,4 +5,7 @@
 // XFAIL: gcc
 
 #define SCHEDULE auto
+// The runtime uses guided schedule for auto,
+// which is a reason choice
+#define SCHED_OUTPUT "guided"
 #include "base_split.h"