[OMPT] Fix return address handling in a few GOMP interface methods

[sconvent]

Summary

Fix return address handling in a few GOMP interface methods

[sconvent]

Include another return address handling fix

[Hahnfeld]

LG but please describe which test cases are fixed by this change (and possibly why it is needed). If no test case has been failing before, this is a sign that we need a new one!

runtime/src/kmp_gsupport.cpp

1212 ↗

(On Diff #126753)

Please add a comment why we need to do this again. I think that the return address is cleared at some point in the function calls above?

openmp/trunk/runtime/src/kmp_gsupport.cpp

#endif /* KMP_DEBUG */
     __kmp_GOMP_parallel_microtask_wrapper(int *gtid, int *npr,
                                           void (*task)(void *), void *data,
                                           unsigned num_threads, ident_t *loc,
                                           enum sched_type schedule, long start,
                                           long end, long incr,
                                           long chunk_size) {
 // Intialize the loop worksharing construct.
 
-#if OMPT_SUPPORT
-  if (ompt_enabled.enabled)
-    OMPT_STORE_RETURN_ADDRESS(*gtid);
-#endif
   KMP_DISPATCH_INIT(loc, *gtid, schedule, start, end, incr, chunk_size,
                     schedule != kmp_sch_static);
 
 #if OMPT_SUPPORT
   kmp_info_t *thr;
   ompt_frame_t *ompt_frame;
   omp_state_t enclosing_state;
 
void func(void (*task)(void *), void *data, unsigned num_threads, long lb,   \
     if (__kmpc_ok_to_fork(&loc) && (num_threads != 1)) {                       \
       if (num_threads != 0) {                                                  \
         __kmp_push_num_threads(&loc, gtid, num_threads);                       \
       }                                                                        \
       __kmp_GOMP_fork_call(&loc, gtid, task,                                   \
                            (microtask_t)__kmp_GOMP_parallel_microtask_wrapper, \
                            9, task, data, num_threads, &loc, (schedule), lb,   \
                            (str > 0) ? (ub - 1) : (ub + 1), str, chunk_sz);    \
+      IF_OMPT_SUPPORT(OMPT_STORE_RETURN_ADDRESS(gtid));                        \
     } else {                                                                   \
       __kmp_GOMP_serialized_parallel(&loc, gtid, task);                        \
+      IF_OMPT_SUPPORT(OMPT_STORE_RETURN_ADDRESS(gtid));                        \
     }                                                                          \
                                                                                \
     KMP_DISPATCH_INIT(&loc, gtid, (schedule), lb,                              \
                       (str > 0) ? (ub - 1) : (ub + 1), str, chunk_sz,          \
                       (schedule) != kmp_sch_static);                           \
                                                                                \
     ompt_post();                                                               \
                                                                                \

 
 unsigned KMP_EXPAND_NAME(KMP_API_NAME_GOMP_SECTIONS_NEXT)(void) {
   int status;
   kmp_int lb, ub, stride;
   int gtid = __kmp_get_gtid();
   MKLOC(loc, "GOMP_sections_next");
   KA_TRACE(20, ("GOMP_sections_next: T#%d\n", gtid));
 
+#if OMPT_SUPPORT
+  OMPT_STORE_RETURN_ADDRESS(gtid);
+#endif
+
   status = KMP_DISPATCH_NEXT(&loc, gtid, NULL, &lb, &ub, &stride);
   if (status) {
     KMP_DEBUG_ASSERT(stride == 1);
     KMP_DEBUG_ASSERT(lb > 0);
     KMP_ASSERT(lb == ub);
   } else {
     lb = 0;
   }
if (__kmpc_ok_to_fork(&loc) && (num_threads != 1)) {
     __kmp_GOMP_fork_call(&loc, gtid, task,
                          (microtask_t)__kmp_GOMP_parallel_microtask_wrapper, 9,
                          task, data, num_threads, &loc, kmp_nm_dynamic_chunked,
                          (kmp_int)1, (kmp_int)count, (kmp_int)1, (kmp_int)1);
   } else {
     __kmp_GOMP_serialized_parallel(&loc, gtid, task);
   }
 
+#if OMPT_SUPPORT
+  OMPT_STORE_RETURN_ADDRESS(gtid);
+#endif
+
   KMP_DISPATCH_INIT(&loc, gtid, kmp_nm_dynamic_chunked, 1, count, 1, 1, TRUE);
 
   task(data);
   KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_END)();
   KA_TRACE(20, ("GOMP_parallel_sections exit: T#%d\n", gtid));
 }
 
 #define PARALLEL_LOOP(func, schedule, ompt_pre, ompt_post)                     \

openmp/trunk/runtime/test/ompt/worksharing/for/base.h

int main()
   // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_implicit_task'
   // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_work'
 
 
   // CHECK: 0: NULL_POINTER=[[NULL:.*$]]
   // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id={{[0-9]+}}, parent_task_frame.exit=[[NULL]], parent_task_frame.reenter={{0x[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, codeptr_ra=0x{{[0-f]+}}, invoker={{[0-9]+}}
 
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   // CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID:[0-9]+]]
-  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=0x{{[0-f]+}}
+  // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_begin: parallel_id=[[PARALLEL_ID]], parent_task_id=[[IMPLICIT_TASK_ID]], codeptr_ra=
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_loop_end: parallel_id=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
   // CHECK: {{^}}[[THREAD_ID]]: ompt_event_implicit_task_end: parallel_id={{[0-9]+}}, task_id=[[IMPLICIT_TASK_ID]]
 
   return 0;
 }