[OMPT] Fix parallel_id and task_id in loop_end with schedule static

[Hahnfeld]

Summary

For serialized parallel regions, wrong ids were reported. Now the same code is used as in kmp_dispatch.cpp which emits the correct ids.

[omalyshe]

LGTM.

openmp/trunk/runtime/src/kmp_csupport.c

 void
 __kmpc_for_static_fini( ident_t *loc, kmp_int32 global_tid )
 {
     KE_TRACE( 10, ("__kmpc_for_static_fini called T#%d\n", global_tid));
 
 #if OMPT_SUPPORT && OMPT_TRACE
     if (ompt_enabled &&
         ompt_callbacks.ompt_callback(ompt_event_loop_end)) {
-        kmp_info_t *this_thr = __kmp_threads[ global_tid ];
-        kmp_team_t *team     = this_thr -> th.th_team;
-        int tid = __kmp_tid_from_gtid( global_tid );
-
+        ompt_team_info_t *team_info = __ompt_get_teaminfo(0, NULL);
+        ompt_task_info_t *task_info = __ompt_get_taskinfo(0);
         ompt_callbacks.ompt_callback(ompt_event_loop_end)(
-            team->t.ompt_team_info.parallel_id,
-            team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_id);
+            team_info->parallel_id, task_info->task_id);
     }
 #endif
 
     if ( __kmp_env_consistency_check )
      __kmp_pop_workshare( global_tid, ct_pdo, loc );
 }
 
 /*

openmp/trunk/runtime/test/lit.cfg

 # -*- Python -*- vim: set ft=python ts=4 sw=4 expandtab tw=79:
 # Configuration file for the 'lit' test runner.
 
 import os
+import re
 import subprocess
 import lit.formats
 
 # Tell pylint that we know config and lit_config exist somewhere.
 if 'PYLINT_IMPORT' in os.environ:
     config = object()
     lit_config = object()
 

 
 # compiler flags
 config.test_cflags = config.test_openmp_flag + \
     " -I " + config.test_source_root + \
     " -I " + config.omp_header_directory + \
     " -L " + config.library_dir + \
     " " + config.test_extra_cflags
 
+# Allow XFAIL to work
+config.target_triple = [ ]
+if re.search('gcc', config.test_compiler) is not None:
+    config.available_features.add('gcc')
+
 # Setup environment to find dynamic library at runtime
 append_dynamic_library_path(config.library_dir)
 if config.using_hwloc:
     append_dynamic_library_path(config.hwloc_library_dir)
 
 # Rpath modifications for Darwin
 if config.operating_system == 'Darwin':
     config.test_cflags += " -Wl,-rpath," + config.library_dir

openmp/trunk/runtime/test/ompt/callback.h

 on_ompt_event_implicit_task_end(
   ompt_parallel_id_t parallel_id,
   ompt_task_id_t task_id)
 {
   printf("%" PRIu64 ": ompt_event_implicit_task_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
 }
 
 static void
+on_ompt_event_loop_begin(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t parent_task_id,
+  void *workshare_function)
+{
+  printf("%" PRIu64 ": ompt_event_loop_begin: parallel_id=%" PRIu64 ", parent_task_id=%" PRIu64 ", workshare_function=%p\n", ompt_get_thread_id(), parallel_id, parent_task_id, workshare_function);
+}
+
+static void
+on_ompt_event_loop_end(
+  ompt_parallel_id_t parallel_id,
+  ompt_task_id_t task_id)
+{
+  printf("%" PRIu64 ": ompt_event_loop_end: parallel_id=%" PRIu64 ", task_id=%" PRIu64 "\n", ompt_get_thread_id(), parallel_id, task_id);
+}
+
+static void
 on_ompt_event_parallel_begin(
   ompt_task_id_t parent_task_id,
   ompt_frame_t *parent_task_frame,
   ompt_parallel_id_t parallel_id,
   uint32_t requested_team_size,
   void *parallel_function,
   ompt_invoker_t invoker)
 {
void ompt_initialize(
   ompt_get_task_id = (ompt_get_task_id_t) lookup("ompt_get_task_id");
   ompt_get_thread_id = (ompt_get_thread_id_t) lookup("ompt_get_thread_id");
   ompt_get_parallel_id = (ompt_get_parallel_id_t) lookup("ompt_get_parallel_id");
 
   ompt_set_callback(ompt_event_barrier_begin, (ompt_callback_t) &on_ompt_event_barrier_begin);
   ompt_set_callback(ompt_event_barrier_end, (ompt_callback_t) &on_ompt_event_barrier_end);
   ompt_set_callback(ompt_event_implicit_task_begin, (ompt_callback_t) &on_ompt_event_implicit_task_begin);
   ompt_set_callback(ompt_event_implicit_task_end, (ompt_callback_t) &on_ompt_event_implicit_task_end);
+  ompt_set_callback(ompt_event_loop_begin, (ompt_callback_t) &on_ompt_event_loop_begin);
+  ompt_set_callback(ompt_event_loop_end, (ompt_callback_t) &on_ompt_event_loop_end);
   ompt_set_callback(ompt_event_parallel_begin, (ompt_callback_t) &on_ompt_event_parallel_begin);
   ompt_set_callback(ompt_event_parallel_end, (ompt_callback_t) &on_ompt_event_parallel_end);
 }
 
 ompt_initialize_t ompt_tool()
 {
   return &ompt_initialize;
 }

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

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base.h
+// REQUIRES: ompt
+// GCC doesn't call runtime for auto = static schedule
+// XFAIL: gcc
+
+#define SCHEDULE auto
+#include "base.h"

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

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_serialized.h
+// REQUIRES: ompt
+// GCC doesn't call runtime for auto = static schedule
+// XFAIL: gcc
+
+#define SCHEDULE auto
+#include "base_serialized.h"

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

+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  unsigned int i;
+
+  #pragma omp parallel for num_threads(4) schedule(SCHEDULE)
+  for (i = 0; i < 4; i++) {
+  }
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id={{[0-9]+}}, parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=4, parallel_function=0x{{[0-f]+}}, invoker={{.*}}
+
+  // 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]], workshare_function=0x{{[0-f]+}}
+  // 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=[[PARALLEL_ID]], 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]], workshare_function=0x{{[0-f]+}}
+  // 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=[[PARALLEL_ID]], 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]], workshare_function=0x{{[0-f]+}}
+  // 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=[[PARALLEL_ID]], 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]], workshare_function=0x{{[0-f]+}}
+  // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  return 0;
+}

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

+#include "callback.h"
+#include <omp.h>
+
+int main()
+{
+  unsigned int i;
+
+  #pragma omp parallel for num_threads(1) schedule(SCHEDULE)
+  for (i = 0; i < 1; i++) {
+  }
+
+  // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_parallel_begin: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame=0x{{[0-f]+}}, parallel_id=[[PARALLEL_ID:[0-9]+]], requested_team_size=1, parallel_function=0x{{[0-f]+}}, invoker={{.+}}
+
+  // 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]], workshare_function=0x{{[0-f]+}}
+  // 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=[[PARALLEL_ID]], task_id=[[IMPLICIT_TASK_ID]]
+
+  return 0;
+}

openmp/trunk/runtime/test/ompt/worksharing/for/dynamic.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base.h
+// REQUIRES: ompt
+
+#define SCHEDULE dynamic
+#include "base.h"

openmp/trunk/runtime/test/ompt/worksharing/for/dynamic_serialized.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_serialized.h
+// REQUIRES: ompt
+
+#define SCHEDULE dynamic
+#include "base_serialized.h"

openmp/trunk/runtime/test/ompt/worksharing/for/guided.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base.h
+// REQUIRES: ompt
+
+#define SCHEDULE guided
+#include "base.h"

openmp/trunk/runtime/test/ompt/worksharing/for/guided_serialized.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_serialized.h
+// REQUIRES: ompt
+
+#define SCHEDULE guided
+#include "base_serialized.h"

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

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base.h
+// REQUIRES: ompt
+
+#define SCHEDULE runtime
+#include "base.h"

openmp/trunk/runtime/test/ompt/worksharing/for/runtime_serialized.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_serialized.h
+// REQUIRES: ompt
+
+#define SCHEDULE runtime
+#include "base_serialized.h"

openmp/trunk/runtime/test/ompt/worksharing/for/static.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base.h
+// REQUIRES: ompt
+// GCC doesn't call runtime for static schedule
+// XFAIL: gcc
+
+#define SCHEDULE static
+#include "base.h"

openmp/trunk/runtime/test/ompt/worksharing/for/static_serialized.c

+// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %S/base_serialized.h
+// REQUIRES: ompt
+// GCC doesn't call runtime for static schedule
+// XFAIL: gcc
+
+#define SCHEDULE static
+#include "base_serialized.h"