[OpenMP][OMPT][GOMP] task frame support in KMP_API_NAME_GOMP_PARALLEL_SECTIONS

[vladaindjic]

Summary

KMP_API_NAME_GOMP_PARALLEL_SECTIONS function was missing the task frame support.
This patch introduced a fix responsible to set properly the exit_frame of
the innermost implicit task that corresponds to the parallel section construct,
as well as the enter_frame of the task that encloses the mentioned implicit task.

This patch also introduced a simple test case sections_serialized.c that contains
serialized parallel section construct and validates whether the mentioned
task frames are set correctly.

[jlpeyton]

openmp/runtime/test/ompt/worksharing/taskinfo/sections_serialized.c

9

Can we not use print_ids() from callback.h?

31

Use `#include "ompt-signal.h" and the wait and signal functions inside that header.

[vladaindjic]

As @jlpeyton suggessted, the following is done:

• Unnecessary code has been removed from sections_serialized.c test case
• print_ids is used inside the same test case.

[jlpeyton]

LGTM

[vladaindjic]

This comment was removed by vladaindjic.

openmp/runtime/test/ompt/worksharing/taskinfo/sections_serialized.c

31

This is some old code, which I'll remove. It is not necessary for the test case.

[vladaindjic]

@jlpeyton Thank you a lot for the review. Could you please land this patch for me?

openmp/runtime/src/kmp_gsupport.cpp

void KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_SECTIONS)(void (*task)(void *),
                                                           unsigned num_threads,
                                                           unsigned count,
                                                           unsigned flags) {
   int gtid = __kmp_entry_gtid();
   MKLOC(loc, "GOMP_parallel_sections");
   KA_TRACE(20, ("GOMP_parallel_sections: T#%d\n", gtid));
 
 #if OMPT_SUPPORT
+  ompt_frame_t *task_frame;
+  kmp_info_t *thr;
+  if (ompt_enabled.enabled) {
+    thr = __kmp_threads[gtid];
+    task_frame = &(thr->th.th_current_task->ompt_task_info.frame);
+    task_frame->enter_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
+  }
   OMPT_STORE_RETURN_ADDRESS(gtid);
 #endif
 
   __kmp_GOMP_fork_call(&loc, gtid, num_threads, flags, 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);
 
   {
 #if OMPT_SUPPORT
     OMPT_STORE_RETURN_ADDRESS(gtid);
 #endif
 
     KMP_DISPATCH_INIT(&loc, gtid, kmp_nm_dynamic_chunked, 1, count, 1, 1, TRUE);
   }
+
+#if OMPT_SUPPORT
+  ompt_frame_t *child_frame;
+  if (ompt_enabled.enabled) {
+    child_frame = &(thr->th.th_current_task->ompt_task_info.frame);
+    child_frame->exit_frame.ptr = OMPT_GET_FRAME_ADDRESS(0);
+  }
+#endif
+
   task(data);
+
+#if OMPT_SUPPORT
+  if (ompt_enabled.enabled) {
+    child_frame->exit_frame = ompt_data_none;
+  }
+#endif
+
   KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_END)();
   KA_TRACE(20, ("GOMP_parallel_sections exit: T#%d\n", gtid));
+
+#if OMPT_SUPPORT
+  if (ompt_enabled.enabled) {
+    task_frame->enter_frame = ompt_data_none;
+  }
+#endif
 }
 
 #define PARALLEL_LOOP(func, schedule, ompt_pre, ompt_post)                     \
   void func(void (*task)(void *), void *data, unsigned num_threads, long lb,   \
             long ub, long str, long chunk_sz, unsigned flags) {                \
     int gtid = __kmp_entry_gtid();                                             \
     MKLOC(loc, KMP_STR(func));                                                 \
     KA_TRACE(                                                                  \

openmp/runtime/test/ompt/worksharing/taskinfo/sections_serialized.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
+// REQUIRES: ompt
+
+#include "callback.h"
+#include <omp.h>
+
+
+__attribute__ ((noinline)) // workaround for bug in icc
+void print_task_info_at(int ancestor_level)
+{
+#pragma omp critical
+  {
+    int task_type;
+    char buffer[2048];
+    ompt_data_t *parallel_data;
+    ompt_data_t *task_data;
+    ompt_frame_t *task_frame;
+    ompt_get_task_info(ancestor_level, &task_type, &task_data, &task_frame,
+                       &parallel_data, NULL);
+    format_task_type(task_type, buffer);
+    printf("%" PRIu64 ": ancestor_level=%d task_type=%s=%d parallel_id=%" PRIu64
+           " task_id=%" PRIu64 " task_frame.exit=%p task_frame.enter=%p\n",
+        ompt_get_thread_data()->value, ancestor_level, buffer,
+        task_type, parallel_data->value, task_data->value,
+        task_frame->exit_frame.ptr, task_frame->enter_frame.ptr);
+  }
+};
+
+volatile int wait = 0;
+
+void spin_wait() {
+  while (wait == 0);
+}
+
+void signal() {
+  wait = 1;
+}
+
+int main()
+{
+#pragma omp parallel sections num_threads(1)
+  {
+#pragma omp section
+    {
+      // implicit task info
+      print_task_info_at(0);
+      // initial task info
+      print_task_info_at(1);
+    }
+  }
+
+  // Check if libomp supports the callbacks for this test.
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
+  // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_implicit_task'
+
+
+  // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_initial_task_begin: parallel_id=[[INITIAL_PARALLEL_ID:[0-9]+]], task_id=[[INITIAL_TASK_ID:[0-9]+]], actual_parallelism=1, index=1, flags=1
+
+  // region 0
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin
+  // CHECK-SAME: parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[INITIAL_TASK_FRAME_ENTER:0x[0-f]+]],
+  // CHECK-SAME: parallel_id=[[PARALLEL_ID:[0-9]+]]
+
+  // CHECK: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID:[0-9]+]]
+
+  // information about implicit task (exit frame should be set, while enter should be NULL)
+  // CHECK: {{^}}[[MASTER_ID]]: ancestor_level=0 task_type=ompt_task_implicit
+  // CHECK-SAME: parallel_id=[[PARALLEL_ID]] task_id=[[TASK_ID]]
+  // CHECK-SAME: task_frame.exit={{0x[0-f]+}}
+  // CHECK-SAME: task_frame.enter=[[NULL]]
+
+  // information about initial task (exit frame should be NULL, while enter frame shoule be set)
+  // CHECK: {{^}}[[MASTER_ID]]: ancestor_level=1 task_type=ompt_task_initial
+  // CHECK-SAME: parallel_id=[[INITIAL_PARALLEL_ID]] task_id=[[INITIAL_TASK_ID]]
+  // CHECK-SAME: task_frame.exit=[[NULL]]
+  // CHECK-SAME: task_frame.enter=[[INITIAL_TASK_FRAME_ENTER]]
+
+
+  return 0;
+}