Merging r285703:

------------------------------------------------------------------------
r285703 | achurbanov | 2016-11-01 11:19:04 -0500 (Tue, 01 Nov 2016) | 9 lines

Fixed problem introduced by part of https://reviews.llvm.org/D21196.

Check Task Scheduling Constraint (TSC) on stealing of untied task.
This is needed because the untied task can produce tied children
those can break TSC if untied is not a descendant of current task.
This can cause live lock on complex tyasking tests
(e.g. kastors/strassen-task-dep).

Differential Revision: https://reviews.llvm.org/D26182
------------------------------------------------------------------------


git-svn-id: https://llvm.org/svn/llvm-project/openmp/branches/release_39@287561 91177308-0d34-0410-b5e6-96231b3b80d8

runtime/src/kmp.h

@@ -3327,6 +3327,8 @@ extern kmp_task_t* __kmp_task_alloc( ident_t *loc_ref, kmp_int32 gtid,
   kmp_routine_entry_t task_entry );
 extern void __kmp_init_implicit_task( ident_t *loc_ref, kmp_info_t *this_thr,
                   kmp_team_t *team, int tid, int set_curr_task );
+extern void __kmp_finish_implicit_task(kmp_info_t *this_thr);
+extern void __kmp_free_implicit_task(kmp_info_t *this_thr);
 
 int __kmp_execute_tasks_32(kmp_info_t *thread, kmp_int32 gtid, kmp_flag_32 *flag, int final_spin,
                            int *thread_finished,
@@ -3496,12 +3498,18 @@ void __kmpc_omp_task_complete( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *tas
 KMP_EXPORT void __kmpc_taskgroup( ident_t * loc, int gtid );
 KMP_EXPORT void __kmpc_end_taskgroup( ident_t * loc, int gtid );
 
-KMP_EXPORT kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task,
-                                                 kmp_int32 ndeps, kmp_depend_info_t *dep_list,
-                                                 kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list );
-KMP_EXPORT void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list,
-                                          kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list );
-extern void __kmp_release_deps ( kmp_int32 gtid, kmp_taskdata_t *task );
+KMP_EXPORT kmp_int32 __kmpc_omp_task_with_deps(
+    ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *new_task, kmp_int32 ndeps,
+    kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias,
+    kmp_depend_info_t *noalias_dep_list);
+KMP_EXPORT void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32 gtid,
+                                     kmp_int32 ndeps,
+                                     kmp_depend_info_t *dep_list,
+                                     kmp_int32 ndeps_noalias,
+                                     kmp_depend_info_t *noalias_dep_list);
+extern void __kmp_release_deps(kmp_int32 gtid, kmp_taskdata_t *task);
+extern void __kmp_dephash_free_entries(kmp_info_t *thread, kmp_dephash_t *h);
+extern void __kmp_dephash_free(kmp_info_t *thread, kmp_dephash_t *h);
 
 extern kmp_int32 __kmp_omp_task( kmp_int32 gtid, kmp_task_t * new_task, bool serialize_immediate );
 

runtime/src/kmp_runtime.c

@@ -4848,6 +4848,19 @@ __kmp_allocate_team( kmp_root_t *root, int new_nproc, int max_nproc,
                 }
 #if KMP_NESTED_HOT_TEAMS
             } // (__kmp_hot_teams_mode == 0)
+            else {
+                // When keeping extra threads in team, switch threads to wait on own b_go flag
+                for (f=new_nproc; f<team->t.t_nproc; ++f) {
+                    KMP_DEBUG_ASSERT(team->t.t_threads[f]);
+                    kmp_balign_t *balign = team->t.t_threads[f]->th.th_bar;
+                    for (int b=0; b<bs_last_barrier; ++b) {
+                        if (balign[b].bb.wait_flag == KMP_BARRIER_PARENT_FLAG) {
+                            balign[b].bb.wait_flag = KMP_BARRIER_SWITCH_TO_OWN_FLAG;
+                        }
+                        KMP_CHECK_UPDATE(balign[b].bb.leaf_kids, 0);
+                    }
+                }
+            }
 #endif // KMP_NESTED_HOT_TEAMS
             team->t.t_nproc =  new_nproc;
             // TODO???: team->t.t_max_active_levels = new_max_active_levels;
@@ -5331,6 +5344,7 @@ __kmp_free_thread( kmp_info_t *this_th )
         if (balign[b].bb.wait_flag == KMP_BARRIER_PARENT_FLAG)
             balign[b].bb.wait_flag = KMP_BARRIER_SWITCH_TO_OWN_FLAG;
         balign[b].bb.team = NULL;
+        balign[b].bb.leaf_kids = 0;
     }
     this_th->th.th_task_state = 0;
 
@@ -5674,6 +5688,8 @@ __kmp_reap_thread(
         --__kmp_thread_pool_nth;
     }; // if
 
+    __kmp_free_implicit_task(thread);
+
     // Free the fast memory for tasking
     #if USE_FAST_MEMORY
         __kmp_free_fast_memory( thread );
@@ -6770,6 +6786,8 @@ __kmp_run_after_invoked_task( int gtid, int tid, kmp_info_t *this_thr,
 {
     if( __kmp_env_consistency_check )
         __kmp_pop_parallel( gtid, team->t.t_ident );
+
+    __kmp_finish_implicit_task(this_thr);
 }
 
 int

runtime/src/kmp_taskdeps.cpp

@@ -97,7 +97,8 @@ __kmp_dephash_create ( kmp_info_t *thread, kmp_taskdata_t *current_task )
     else
         h_size = KMP_DEPHASH_OTHER_SIZE;
 
-    kmp_int32 size = h_size * sizeof(kmp_dephash_entry_t) + sizeof(kmp_dephash_t);
+    kmp_int32 size =
+        h_size * sizeof(kmp_dephash_entry_t *) + sizeof(kmp_dephash_t);
 
 #if USE_FAST_MEMORY
     h = (kmp_dephash_t *) __kmp_fast_allocate( thread, size );
@@ -118,13 +119,13 @@ __kmp_dephash_create ( kmp_info_t *thread, kmp_taskdata_t *current_task )
     return h;
 }
 
-static void
-__kmp_dephash_free ( kmp_info_t *thread, kmp_dephash_t *h )
+void
+__kmp_dephash_free_entries(kmp_info_t *thread, kmp_dephash_t *h)
 {
-    for ( size_t i=0; i < h->size; i++ ) {
-        if ( h->buckets[i] ) {
+    for (size_t i = 0; i < h->size; i++) {
+        if (h->buckets[i]) {
             kmp_dephash_entry_t *next;
-            for ( kmp_dephash_entry_t *entry = h->buckets[i]; entry; entry = next ) {
+            for (kmp_dephash_entry_t *entry = h->buckets[i]; entry; entry = next) {
                 next = entry->next_in_bucket;
                 __kmp_depnode_list_free(thread,entry->last_ins);
                 __kmp_node_deref(thread,entry->last_out);
@@ -134,8 +135,15 @@ __kmp_dephash_free ( kmp_info_t *thread, kmp_dephash_t *h )
                 __kmp_thread_free(thread,entry);
 #endif
             }
+            h->buckets[i] = 0;
         }
     }
+}
+
+void
+__kmp_dephash_free(kmp_info_t *thread, kmp_dephash_t *h)
+{
+    __kmp_dephash_free_entries(thread, h);
 #if USE_FAST_MEMORY
     __kmp_fast_free(thread,h);
 #else

runtime/src/kmp_tasking.c

@@ -852,7 +852,6 @@ __kmp_init_implicit_task( ident_t *loc_ref, kmp_info_t *this_thr, kmp_team_t *te
     task->td_flags.freed       = 0;
 
 #if OMP_40_ENABLED
-    task->td_dephash = NULL;
     task->td_depnode = NULL;
 #endif
 
@@ -861,6 +860,7 @@ __kmp_init_implicit_task( ident_t *loc_ref, kmp_info_t *this_thr, kmp_team_t *te
         task->td_allocated_child_tasks  = 0; // Not used because do not need to deallocate implicit task
 #if OMP_40_ENABLED
         task->td_taskgroup = NULL;           // An implicit task does not have taskgroup
+        task->td_dephash = NULL;
 #endif
         __kmp_push_current_task_to_thread( this_thr, team, tid );
     } else {
@@ -876,6 +876,39 @@ __kmp_init_implicit_task( ident_t *loc_ref, kmp_info_t *this_thr, kmp_team_t *te
                   tid, team, task ) );
 }
 
+
+//-----------------------------------------------------------------------------
+//// __kmp_finish_implicit_task: Release resources associated to implicit tasks
+//// at the end of parallel regions. Some resources are kept for reuse in the
+//// next parallel region.
+////
+//// thread:  thread data structure corresponding to implicit task
+//
+void
+__kmp_finish_implicit_task(kmp_info_t *thread)
+{
+    kmp_taskdata_t *task = thread->th.th_current_task;
+    if (task->td_dephash)
+        __kmp_dephash_free_entries(thread, task->td_dephash);
+}
+
+
+//-----------------------------------------------------------------------------
+//// __kmp_free_implicit_task: Release resources associated to implicit tasks
+//// when these are destroyed regions
+////
+//// thread:  thread data structure corresponding to implicit task
+//
+void
+__kmp_free_implicit_task(kmp_info_t *thread)
+{
+    kmp_taskdata_t *task = thread->th.th_current_task;
+    if (task->td_dephash)
+        __kmp_dephash_free(thread, task->td_dephash);
+    task->td_dephash = NULL;
+}
+
+
 // Round up a size to a power of two specified by val
 // Used to insert padding between structures co-allocated using a single malloc() call
 static size_t
@@ -1755,8 +1788,8 @@ __kmp_steal_task( kmp_info_t *victim, kmp_int32 gtid, kmp_task_team_t *task_team
             parent = parent->td_parent;  // check generation up to the level of the current task
             KMP_DEBUG_ASSERT(parent != NULL);
         }
-        if ( parent != current && (taskdata->td_flags.tiedness == TASK_TIED) ) { // untied is always allowed to be stolen
-            // If the tail task is not a child, then no other childs can appear in the deque (?).
+        if ( parent != current ) {
+            // If the tail task is not a descendant of the current task then do not steal it.
             __kmp_release_bootstrap_lock( & victim_td -> td.td_deque_lock );
             KA_TRACE(10, ("__kmp_steal_task(exit #2): T#%d could not steal from T#%d: task_team=%p "
                           "ntasks=%d head=%u tail=%u\n",

runtime/test/misc_bugs/many-microtask-args.c

@@ -4,6 +4,7 @@
 int main()
 {
 
+  int i;
   int i1 = 0;
   int i2 = 1;
   int i3 = 2;
@@ -23,7 +24,7 @@ int main()
  
   int r = 0; 
   #pragma omp parallel for firstprivate(i1, i2, i3, i4, i5, i6, i7, i8, i9, i10, i11, i12, i13, i14, i15, i16) reduction(+:r)
-  for (int i = 0; i < i16; i++) {
+  for (i = 0; i < i16; i++) {
     r += i + i1 + i2 + i3 + i4 + i5 + i6 + i7 + i8 + i9 + i10 + i11 + i12 + i13 + i14 + i15 + i16;
   }
 

runtime/test/misc_bugs/omp_foreign_thread_team_reuse.c

@@ -1,4 +1,4 @@
-// RUN: %libomp-compile-and-run
+// RUN: %libomp-compile -lpthread && %libomp-run
 #include <stdio.h>
 #include "omp_testsuite.h"