libgomp: Fix up creation of artificial teams

When not in explicit parallel/target/teams construct, we in some cases create
an artificial parallel with a single thread (either to handle target nowait
or for task reduction purposes).  In those cases, it handled again artificially
created implicit task (created by gomp_new_icv for cases where we needed to write
to some ICVs), but as the testcases show, didn't take into account possibility
of this being done from explicit task(s).  The code would destroy/free the previous
task and replace it with the new implicit task.  If task is an explicit task
(when teams is NULL, all explicit tasks behave like if (0)), it is a pointer to
a local stack variable, so freeing it doesn't work, and additionally we shouldn't
lose the explicit tasks - the new implicit task should instead replace the
ancestor task which is the first implicit one.

2022-10-12  Jakub Jelinek  <jakub@redhat.com>

	* task.c (gomp_create_artificial_team): Fix up handling of invocations
	from within explicit task.
	* target.c (GOMP_target_ext): Likewise.
	* testsuite/libgomp.c/task-7.c: New test.
	* testsuite/libgomp.c/task-8.c: New test.
	* testsuite/libgomp.c-c++-common/task-reduction-17.c: New test.
	* testsuite/libgomp.c-c++-common/task-reduction-18.c: New test.

6 files changed, 117 insertions, 6 deletions

diff --git a/libgomp/target.c b/libgomp/target.c
index 5763483..73c99c7c 100644
--- a/libgomp/target.c
+++ b/libgomp/target.c
@@ -2813,6 +2813,7 @@ GOMP_target_ext (int device, void (*fn) (void *), size_t mapnum,
 	{
 	  struct gomp_team *team = gomp_new_team (1);
 	  struct gomp_task *task = thr->task;
+	  struct gomp_task **implicit_task = &task;
 	  struct gomp_task_icv *icv = task ? &task->icv : &gomp_global_icv;
 	  team->prev_ts = thr->ts;
 	  thr->ts.team = team;
@@ -2825,15 +2826,23 @@ GOMP_target_ext (int device, void (*fn) (void *), size_t mapnum,
 	  thr->ts.static_trip = 0;
 	  thr->task = &team->implicit_task[0];
 	  gomp_init_task (thr->task, NULL, icv);
-	  if (task)
+	  while (*implicit_task
+		 && (*implicit_task)->kind != GOMP_TASK_IMPLICIT)
+	    implicit_task = &(*implicit_task)->parent;
+	  if (*implicit_task)
 	    {
-	      thr->task = task;
+	      thr->task = *implicit_task;
 	      gomp_end_task ();
-	      free (task);
+	      free (*implicit_task);
 	      thr->task = &team->implicit_task[0];
 	    }
 	  else
 	    pthread_setspecific (gomp_thread_destructor, thr);
+	  if (implicit_task != &task)
+	    {
+	      *implicit_task = thr->task;
+	      thr->task = task;
+	    }
 	}
       if (thr->ts.team
 	  && !thr->task->final_task)
diff --git a/libgomp/task.c b/libgomp/task.c
index 30cd046..7766535 100644
--- a/libgomp/task.c
+++ b/libgomp/task.c
@@ -2465,6 +2465,7 @@ gomp_create_artificial_team (void)
   struct gomp_task_icv *icv;
   struct gomp_team *team = gomp_new_team (1);
   struct gomp_task *task = thr->task;
+  struct gomp_task **implicit_task = &task;
   icv = task ? &task->icv : &gomp_global_icv;
   team->prev_ts = thr->ts;
   thr->ts.team = team;
@@ -2477,17 +2478,25 @@ gomp_create_artificial_team (void)
   thr->ts.static_trip = 0;
   thr->task = &team->implicit_task[0];
   gomp_init_task (thr->task, NULL, icv);
-  if (task)
+  while (*implicit_task
+	 && (*implicit_task)->kind != GOMP_TASK_IMPLICIT)
+    implicit_task = &(*implicit_task)->parent;
+  if (*implicit_task)
     {
-      thr->task = task;
+      thr->task = *implicit_task;
       gomp_end_task ();
-      free (task);
+      free (*implicit_task);
       thr->task = &team->implicit_task[0];
     }
 #ifdef LIBGOMP_USE_PTHREADS
   else
     pthread_setspecific (gomp_thread_destructor, thr);
 #endif
+  if (implicit_task != &task)
+    {
+      *implicit_task = thr->task;
+      thr->task = task;
+    }
 }
 
 /* The format of data is:
diff --git a/libgomp/testsuite/libgomp.c-c++-common/task-reduction-17.c b/libgomp/testsuite/libgomp.c-c++-common/task-reduction-17.c
new file mode 100644
index 0000000..4a8d1e8
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/task-reduction-17.c
@@ -0,0 +1,36 @@
+/* { dg-do run } */
+
+#include <omp.h>
+#include <stdlib.h>
+
+int a;
+
+int
+main ()
+{
+  #pragma omp task final (1)
+  {
+    if (!omp_in_final ())
+      abort ();
+    #pragma omp task
+    {
+      if (!omp_in_final ())
+	abort ();
+      #pragma omp taskgroup task_reduction (+: a)
+      {
+	if (!omp_in_final ())
+	  abort ();
+	#pragma omp task in_reduction (+: a)
+	{
+	  ++a;
+	  if (!omp_in_final ())
+	    abort ();
+	}
+      }
+      if (!omp_in_final ())
+	abort ();
+      #pragma omp taskwait
+    }
+  }
+  return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c-c++-common/task-reduction-18.c b/libgomp/testsuite/libgomp.c-c++-common/task-reduction-18.c
new file mode 100644
index 0000000..483f440
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/task-reduction-18.c
@@ -0,0 +1,17 @@
+/* { dg-do run } */
+
+int a;
+
+int
+main ()
+{
+  #pragma omp task
+  {
+    #pragma omp taskgroup task_reduction (+: a)
+    {
+      #pragma omp task in_reduction (+: a)
+      ++a;
+    }
+  }
+  return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c/task-7.c b/libgomp/testsuite/libgomp.c/task-7.c
new file mode 100644
index 0000000..0307575
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c/task-7.c
@@ -0,0 +1,26 @@
+/* { dg-do run } */
+
+#include <omp.h>
+#include <stdlib.h>
+
+int
+main ()
+{
+  #pragma omp task final (1)
+  {
+    if (!omp_in_final ())
+      abort ();
+    #pragma omp task
+    {
+      if (!omp_in_final ())
+	abort ();
+      #pragma omp target nowait
+      if (omp_in_final ())
+	abort ();
+      if (!omp_in_final ())
+	abort ();
+      #pragma omp taskwait
+    }
+  }
+  return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c/task-8.c b/libgomp/testsuite/libgomp.c/task-8.c
new file mode 100644
index 0000000..f03aef6
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c/task-8.c
@@ -0,0 +1,14 @@
+/* { dg-do run } */
+
+int
+main ()
+{
+  int i = 0;
+  #pragma omp task
+  {
+    #pragma omp target nowait private (i)
+    i = 1;
+    #pragma omp taskwait
+  }
+  return 0;
+}