- add support for omp taskgroup

- add testcase for omp taskgroup

include/starpu_openmp.h

@@ -60,6 +60,7 @@ extern void starpu_omp_master(void (*f)(void *arg), void *arg, int nowait) __STA
 extern void starpu_omp_single(void (*f)(void *arg), void *arg, int nowait) __STARPU_OMP_NOTHROW;
 extern void starpu_omp_critical(void (*f)(void *arg), void *arg, const char *name) __STARPU_OMP_NOTHROW;
 extern void starpu_omp_taskwait(void) __STARPU_OMP_NOTHROW;
+extern void starpu_omp_taskgroup(void (*f)(void *arg), void *arg) __STARPU_OMP_NOTHROW;
 
 extern void starpu_omp_set_num_threads(int threads) __STARPU_OMP_NOTHROW;
 extern int starpu_omp_get_num_threads() __STARPU_OMP_NOTHROW;

src/util/openmp_runtime_support.c

@@ -352,6 +352,21 @@ static void starpu_omp_explicit_task_exec(void *buffers[], void *cl_arg)
 		{
 			_starpu_spin_unlock(&parallel_region->lock);
 		}
+		if (task->task_group)
+		{
+			struct starpu_omp_task *leader_task = task->task_group->leader_task;
+			STARPU_ASSERT(leader_task != task);
+			_starpu_spin_lock(&leader_task->lock);
+			if (STARPU_ATOMIC_ADD(&task->task_group->descendent_task_count, -1) == 0)
+			{
+				if (leader_task->wait_on & starpu_omp_task_wait_on_group)
+				{
+					leader_task->wait_on &= ~starpu_omp_task_wait_on_group;
+					_wake_up_locked_task(leader_task);
+				}
+			}
+			_starpu_spin_unlock(&leader_task->lock);
+		}
 	}
 	else if (task->state != starpu_omp_task_state_preempted)
 		_STARPU_ERROR("invalid omp task state");
@@ -1015,13 +1030,9 @@ void starpu_omp_task_region(const struct starpu_codelet * const _task_region_cl,
 		{
 			(void)STARPU_ATOMIC_ADD(&generating_task->child_task_count, 1);
 			(void)STARPU_ATOMIC_ADD(&parallel_region->bound_explicit_task_count, 1);
+			if (generated_task->task_group)
 			{
-				struct starpu_omp_task_group *_task_group = generated_task->task_group;
-				while (_task_group)
-				{
-					(void)STARPU_ATOMIC_ADD(&_task_group->descendent_task_count, 1);
-					_task_group = _task_group->next;
-				}
+				(void)STARPU_ATOMIC_ADD(&generated_task->task_group->descendent_task_count, 1);
 			}
 			if (is_undeferred)
 			{
@@ -1062,6 +1073,36 @@ void starpu_omp_taskwait(void)
 	}
 }
 
+static void group__sleep_callback(void *_task)
+{
+	struct starpu_omp_task *task = _task;
+	_starpu_spin_unlock(&task->lock);
+}
+
+void starpu_omp_taskgroup(void (*f)(void *arg), void *arg)
+{
+	struct starpu_omp_task *task = STARPU_PTHREAD_GETSPECIFIC(omp_task_key);
+	struct starpu_omp_task_group *p_previous_task_group;
+	struct starpu_omp_task_group task_group;
+	p_previous_task_group = task->task_group;
+	task_group.descendent_task_count = 0;
+	task_group.leader_task = task;
+	task->task_group = &task_group;
+	f(arg);
+	_starpu_spin_lock(&task->lock);
+	if (task_group.descendent_task_count > 0)
+	{
+		task->wait_on |= starpu_omp_task_wait_on_group;
+		_starpu_task_prepare_for_continuation_ext(0, group__sleep_callback, task);
+		starpu_omp_task_preempt();
+		STARPU_ASSERT(task_group.descendent_task_count == 0);
+	}
+	else
+	{
+		_starpu_spin_unlock(&task->lock);
+	}
+	task->task_group = p_previous_task_group;
+}
 /*
  * restore deprecated diagnostics (-Wdeprecated-declarations)
  */

src/util/openmp_runtime_support.h

@@ -166,7 +166,7 @@ struct starpu_omp_initial_icv_values
 struct starpu_omp_task_group
 {
 	int descendent_task_count;
-	struct starpu_omp_task_group *next;
+	struct starpu_omp_task *leader_task;
 };
 
 struct starpu_omp_task_link

tests/Makefile.am

@@ -234,6 +234,7 @@ noinst_PROGRAMS =				\
 	openmp/task_01				\
 	openmp/task_01				\
 	openmp/taskwait_01			\
+	openmp/taskgroup_01			\
 	overlap/overlap				\
 	overlap/gpu_concurrency			\
 	parallel_tasks/explicit_combined_worker	\
@@ -486,6 +487,9 @@ openmp_task_01_SOURCES = 	\
 openmp_taskwait_01_SOURCES = 	\
 	openmp/taskwait_01.c
 
+openmp_taskgroup_01_SOURCES = 	\
+	openmp/taskgroup_01.c
+
 ###################
 # Block interface #
 ###################

tests/openmp/taskgroup_01.c

@@ -0,0 +1,108 @@
+/* StarPU --- Runtime system for heterogeneous multicore architectures.
+ *
+ * Copyright (C) 2014  Inria
+ *
+ * StarPU is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU Lesser General Public License as published by
+ * the Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ *
+ * StarPU is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * See the GNU Lesser General Public License in COPYING.LGPL for more details.
+ */
+
+#include <pthread.h>
+#include <starpu.h>
+#include "../helper.h"
+#include <stdio.h>
+
+#if !defined(STARPU_OPENMP)
+int main(int argc, char **argv)
+{
+	return STARPU_TEST_SKIPPED;
+}
+#else
+__attribute__((constructor))
+static void omp_constructor(void)
+{
+	int ret = starpu_omp_init();
+	STARPU_CHECK_RETURN_VALUE(ret, "starpu_omp_init");
+}
+
+__attribute__((destructor))
+static void omp_destructor(void)
+{
+	starpu_omp_shutdown();
+}
+
+void task_region_g(void *buffers[], void *args)
+{
+	(void) buffers;
+	int i = (int)(intptr_t) args;
+	int worker_id;
+	pthread_t tid;
+	tid = pthread_self();
+	worker_id = starpu_worker_get_id();
+	printf("[tid %p] task thread = %d: explicit task \"g[%d]\"\n", (void *)tid, worker_id, i);
+}
+
+static struct starpu_codelet task_region_cl =
+{
+	.cpu_funcs    = { task_region_g, NULL },
+	.where        = STARPU_CPU,
+	.nbuffers     = 0,
+	.model        = NULL
+};
+
+void taskgroup_f(void *arg)
+{
+	int *p_i = (int *)arg;
+	/*
+	 * if_clause: 1
+	 * final_clause: 0
+	 * untied_clause: 1
+	 * mergeable_clause: 0
+	 */
+	starpu_omp_task_region(&task_region_cl, (void *)(intptr_t)(*p_i)++, 1, 0, 1, 0);
+	starpu_omp_task_region(&task_region_cl, (void *)(intptr_t)(*p_i)++, 1, 0, 1, 0);
+}
+
+void parallel_region_f(void *buffers[], void *args)
+{
+	(void) buffers;
+	(void) args;
+	int worker_id;
+	pthread_t tid;
+	int i = 0;
+
+	tid = pthread_self();
+	worker_id = starpu_worker_get_id();
+	printf("[tid %p] task thread = %d: implicit task \"f\"\n", (void *)tid, worker_id);
+	
+	starpu_omp_taskgroup(taskgroup_f, (void *)&i);
+	printf("[tid %p] task thread = %d: implicit task \"f\": taskgroup\n", (void *)tid, worker_id);
+
+	starpu_omp_taskgroup(taskgroup_f, (void *)&i);
+	printf("[tid %p] task thread = %d: implicit task \"f\": taskgroup\n", (void *)tid, worker_id);
+
+	starpu_omp_task_region(&task_region_cl, (void *)(intptr_t)i++, 1, 0, 1, 0);
+	starpu_omp_task_region(&task_region_cl, (void *)(intptr_t)i++, 1, 0, 1, 0);
+}
+
+static struct starpu_codelet parallel_region_cl =
+{
+	.cpu_funcs    = { parallel_region_f, NULL },
+	.where        = STARPU_CPU,
+	.nbuffers     = 0,
+	.model        = NULL
+};
+
+int
+main (int argc, char *argv[]) {
+	starpu_omp_parallel_region(&parallel_region_cl, NULL);
+	return 0;
+}
+#endif