[OMPT] Add missing ompt_get_num_procs function

[protze.joachim]

Summary

This function is defined in OpenMP-TR6 section 4.1.5.1.6
The functions was not implemented yet.

Since ompt-functions can only be called after the runtime was initialized and has loaded a tool, it can assume the runtime to be initialized. In contrast to omp_get_num_procs which needs to check whether the runtime is initialized.

[Hahnfeld]

LGTM with one change to the test

runtime/test/ompt/misc/api_calls.c

47–48

Maybe check that these two return the same result as for omp{,t}_get_num_places?

// CHECK: {{^}}[[MASTER_ID]]: omp_get_num_procs()=[[NUM_PROCS:[0-9]+]]
// CHECK: {{^}}[[MASTER_ID]]: ompt_get_num_procs()=[[NUM_PROCS]]

[protze.joachim]

Thanks

[Hahnfeld]

runtime/test/ompt/misc/api_calls.c

67

Can we get a negative result?!?

[protze.joachim]

runtime/test/ompt/misc/api_calls.c

67

The return value is int, thus could be negative.

If a negative value is invalid for omp_get_num_procs, it should be tested somewhere else.

[Hahnfeld]

runtime/test/ompt/misc/api_calls.c

67

omp_get_num_places didn't have it, so I was confused. Anyway I think you are right, this is fine for OMPT as long as the two functions return the same value.

runtime/src/include/50/ompt.h.var

macro (ompt_get_callback)               \
                                             \
     macro (ompt_get_state)                  \
                                             \
     macro (ompt_get_parallel_info)          \
     macro (ompt_get_task_info)              \
     macro (ompt_get_thread_data)            \
     macro (ompt_get_unique_id)              \
                                             \
+    macro(ompt_get_num_procs)               \
     macro(ompt_get_num_places)              \
     macro(ompt_get_place_proc_ids)          \
     macro(ompt_get_place_num)               \
     macro(ompt_get_partition_place_nums)    \
     macro(ompt_get_proc_id)                 \
                                             \
     macro(ompt_get_target_info)             \
     macro(ompt_get_num_devices)
OMPT_API_FUNCTION(int, ompt_get_task_info, (
     int ancestor_level,
     int *type,
     ompt_data_t **task_data,
     ompt_frame_t **task_frame,
     ompt_data_t **parallel_data,
     int *thread_num
 ));
 
+/* procs */
+OMPT_API_FUNCTION(int, ompt_get_num_procs, (void));
+
 /* places */
 OMPT_API_FUNCTION(int, ompt_get_num_places, (void));
 
 OMPT_API_FUNCTION(int, ompt_get_place_proc_ids, (
     int place_num,
     int ids_size,
     int *ids
 ));

runtime/src/ompt-general.cpp

OMPT_API_ROUTINE int ompt_get_task_info(int ancestor_level, int *type,
                                         ompt_frame_t **task_frame,
                                         ompt_data_t **parallel_data,
                                         int *thread_num) {
   return __ompt_get_task_info_internal(ancestor_level, type, task_data,
                                        task_frame, parallel_data, thread_num);
 }
 
 /*****************************************************************************
+ * num_procs
+ ****************************************************************************/
+
+OMPT_API_ROUTINE int ompt_get_num_procs(void) {
+// copied from kmp_ftn_entry.h (but modified: OMPT can only be called when runtime is initialized)
+  return __kmp_avail_proc;
+}
+
+/*****************************************************************************
  * places
  ****************************************************************************/
 
 OMPT_API_ROUTINE int ompt_get_num_places(void) {
 // copied from kmp_ftn_entry.h (but modified)
 #if !KMP_AFFINITY_SUPPORTED
   return 0;
 #else

runtime/test/ompt/callback.h

static const char* ompt_cancel_flag_t_values[] = {
   "ompt_cancel_discarded_task"
 };
 
 static ompt_set_callback_t ompt_set_callback;
 static ompt_get_task_info_t ompt_get_task_info;
 static ompt_get_thread_data_t ompt_get_thread_data;
 static ompt_get_parallel_info_t ompt_get_parallel_info;
 static ompt_get_unique_id_t ompt_get_unique_id;
+static ompt_get_num_procs_t ompt_get_num_procs;
 static ompt_get_num_places_t ompt_get_num_places;
 static ompt_get_place_proc_ids_t ompt_get_place_proc_ids;
 static ompt_get_place_num_t ompt_get_place_num;
 static ompt_get_partition_place_nums_t ompt_get_partition_place_nums;
 static ompt_get_proc_id_t ompt_get_proc_id;
 static ompt_enumerate_states_t ompt_enumerate_states;
 static ompt_enumerate_mutex_impls_t ompt_enumerate_mutex_impls;
 
int ompt_initialize(
   ompt_data_t *tool_data)
 {
   ompt_set_callback = (ompt_set_callback_t) lookup("ompt_set_callback");
   ompt_get_task_info = (ompt_get_task_info_t) lookup("ompt_get_task_info");
   ompt_get_thread_data = (ompt_get_thread_data_t) lookup("ompt_get_thread_data");
   ompt_get_parallel_info = (ompt_get_parallel_info_t) lookup("ompt_get_parallel_info");
   ompt_get_unique_id = (ompt_get_unique_id_t) lookup("ompt_get_unique_id");
 
+  ompt_get_num_procs = (ompt_get_num_procs_t) lookup("ompt_get_num_procs");
   ompt_get_num_places = (ompt_get_num_places_t) lookup("ompt_get_num_places");
   ompt_get_place_proc_ids = (ompt_get_place_proc_ids_t) lookup("ompt_get_place_proc_ids");
   ompt_get_place_num = (ompt_get_place_num_t) lookup("ompt_get_place_num");
   ompt_get_partition_place_nums = (ompt_get_partition_place_nums_t) lookup("ompt_get_partition_place_nums");
   ompt_get_proc_id = (ompt_get_proc_id_t) lookup("ompt_get_proc_id");
   ompt_enumerate_states = (ompt_enumerate_states_t) lookup("ompt_enumerate_states");
   ompt_enumerate_mutex_impls = (ompt_enumerate_mutex_impls_t) lookup("ompt_enumerate_mutex_impls");
 

runtime/test/ompt/misc/api_calls.c

#pragma omp parallel num_threads(1)
     omp_get_partition_place_nums(omp_nums);
     print_list("omp_get_partition_place_nums" ,omp_nums);
     int ompt_nums[omp_get_partition_num_places()];
     ompt_get_partition_place_nums(omp_get_partition_num_places(), ompt_nums);
     print_list("ompt_get_partition_place_nums", ompt_nums);
 
     printf("%" PRIu64 ": sched_getcpu()=%d\n", ompt_get_thread_data()->value, sched_getcpu());
     printf("%" PRIu64 ": ompt_get_proc_id()=%d\n", ompt_get_thread_data()->value, ompt_get_proc_id());
+
+    printf("%" PRIu64 ": omp_get_num_procs()=%d\n", ompt_get_thread_data()->value, omp_get_num_procs());
+    printf("%" PRIu64 ": ompt_get_num_procs()=%d\n", ompt_get_thread_data()->value, ompt_get_num_procs());
   }
 
   // Check if libomp supports the callbacks for this test.
 
   // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
 
   // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: omp_get_num_places()=[[NUM_PLACES:[0-9]+]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_get_num_places()=[[NUM_PLACES]]
 
   // CHECK: {{^}}[[MASTER_ID]]: omp_get_place_proc_ids(0)=([[PROC_IDS:[0-9\,]+]])
   // CHECK: {{^}}[[MASTER_ID]]: ompt_get_place_proc_ids(0)=([[PROC_IDS]])
 
   // CHECK: {{^}}[[MASTER_ID]]: omp_get_place_num()=[[PLACE_NUM:[-]?[0-9]+]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_get_place_num()=[[PLACE_NUM]]
 
   // CHECK: {{^}}[[MASTER_ID]]: sched_getcpu()=[[CPU_ID:[0-9]+]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_get_proc_id()=[[CPU_ID]]
 
+  // CHECK: {{^}}[[MASTER_ID]]: omp_get_num_procs()=[[NUM_PROCS:[-]?[0-9]+]]
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_get_num_procs()=[[NUM_PROCS]]
+
 
   return 0;
 }