Merge branch 'master' of https://gitlab.inria.fr/starpu/starpu

configure.ac

@@ -159,18 +159,18 @@ AC_MSG_RESULT($enable_prof_tool)
 
 ###############################################################################
 #                                                                             #
-#                            Hierarchical dags support                        #
+#                            Recursive tasks support                          #
 #                                                                             #
 ###############################################################################
 
 AC_ARG_ENABLE(bubble, [AS_HELP_STRING([--enable-bubble],
-			[build the hierarchical dags (a.k.a bubble) support])],
+			[build the recursive tasks (a.k.a bubble) support])],
 			enable_bubble=$enableval, enable_bubble=no)
 
-AC_MSG_CHECKING([for hierarchical dags - a.k.a bubble - support])
+AC_MSG_CHECKING([for recursive tasks - a.k.a bubble - support])
 
 if test x$enable_bubble = xyes; then
-	AC_DEFINE(STARPU_BUBBLE, [1], [Define this to enable hierarchical dags support])
+	AC_DEFINE(STARPU_BUBBLE, [1], [Define this to enable recursive tasks support])
 fi
 
 AM_CONDITIONAL([STARPU_BUBBLE], [test "x$enable_bubble" = "xyes"])
@@ -4548,7 +4548,7 @@ AC_MSG_NOTICE([
 	       Native fortran support:                        $enable_build_fortran
 	       Native MPI fortran support:                    $use_mpi_fort
 	       Support for multiple linear regression models: $support_mlr
-	       Hierarchical dags support:                     $enable_bubble
+	       Recursive tasks support:                       $enable_bubble
 	       JULIA enabled:                                 $enable_julia
 ])
 

doc/doxygen/chapters/api/bubble_support.doxy

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2017-2023  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2017-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), 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
@@ -18,8 +18,8 @@
 * The file is empty but necessary to define the group API_Bubble
 */
 
-/*! \defgroup API_Bubble Hierarchical Dags
+/*! \defgroup API_Bubble Recursive Tasks
 
-\brief API for Hierarchical DAGS
+\brief API for Recursive Tasks
 
 */

doc/doxygen/chapters/starpu_extensions/bubble.doxy

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2017-2023  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2017-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), 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
@@ -14,7 +14,7 @@
  * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  */
 
-/*! \page HierarchicalDAGS Hierarchical DAGS
+/*! \page RecursiveTasks Recursive Tasks
 
 The STF model has the intrinsic limitation of supporting static task
 graphs only, which leads to potential submission overhead and to a
@@ -23,11 +23,11 @@ heterogeneous systems.
 
 To address these problems, we have extended the STF model to enable
 tasks subgraphs at runtime. We refer to these tasks as
-<em>hierarchical tasks</em>. This approach allows for a more dynamic
+<em>recursive tasks</em>. This approach allows for a more dynamic
 task graph. This allows to dynamically adapt the granularity to meet
 the optimal size of the targeted computing resource.
 
-<em>Hierarchical tasks</em> are tasks that can transform themselves into
+<em>Recursive tasks</em> are tasks that can transform themselves into
 a new task-graph dynamically at runtime. Programmers submit a coarse
 version of the DAG, called the bubbles graph, which represents the
 general shape of the application tasks graph. The execution of this
@@ -42,7 +42,7 @@ tasks.
 
 \section BubblesExamples An Example
 
-In order to understand the hierarchical tasks model, an example of
+In order to understand the recursive tasks model, an example of
 "bubblification" is showed here. We start from a simple example,
 multiplying the elements of a vector.
 
@@ -168,7 +168,7 @@ int vector_bubble()
 
 The full example is available in the file <c>bubble/tests/vector/vector.c</c>.
 
-To define a hierarchical task, one needs to define the fields
+To define a recursive task, one needs to define the fields
 starpu_codelet::bubble_func and starpu_codelet::bubble_gen_dag_func.
 
 The field starpu_codelet::bubble_func is a pointer function which will
@@ -188,7 +188,7 @@ as parameter the task being checked, and the value specified with
 ::STARPU_BUBBLE_FUNC_ARG.
 
 When executed, the function starpu_codelet::bubble_gen_dag_func will be
-given as parameter the task being turned into a hierarchical task and
+given as parameter the task being turned into a recursive task and
 the value specified with ::STARPU_BUBBLE_GEN_DAG_FUNC_ARG.
 
 An example involving these functions is in <c>bubble/tests/basic/brec.c</c>. And more examples are available in <c>bubble/tests/basic/*.c</c>.

doc/doxygen/chapters/starpu_extensions/extensions_intro.doxy

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2023  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2009-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), 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
@@ -85,8 +85,8 @@ can transparently be run using StarPU, by givings unified access
 to every available OpenCL device.
 </li>
 <li>
-We propose a hierarchical tasks model in Chapter \ref
-HierarchicalDAGS to enable tasks subgraphs at runtime for a more
+We propose a recursive tasks model in Chapter \ref
+RecursiveTasks to enable tasks subgraphs at runtime for a more
 dynamic task graph.
 </li>
 <li>

doc/doxygen/chapters/starpu_installation/configure_options.doxy

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2023  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2009-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  * Copyright (C) 2020       Federal University of Rio Grande do Sul (UFRGS)
  *
  * StarPU is free software; you can redistribute it and/or modify
@@ -559,7 +559,7 @@ Enable LLVM OpenMP Support (\ref OMPLLVM)
 <dd>
 \anchor enable-bubble
 \addindex __configure__--enable-bubble
-Enable Hierarchical dags support (\ref HierarchicalDAGS)
+Enable recursive tasks support (\ref RecursiveTasks)
 
 <dt>--enable-parallel-worker</dt>
 <dd>

doc/doxygen/chapters/starpu_introduction/doc_organization.doxy

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2023  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2009-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), 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
@@ -76,7 +76,7 @@ The documentation chapters include
     <li> \ref FaultTolerance
     <li> \ref FFTSupport
     <li> \ref SOCLOpenclExtensions
-    <li> \ref HierarchicalDAGS
+    <li> \ref RecursiveTasks
     <li> \ref ParallelWorker
     <li> \ref InteroperabilitySupport
     <li> \ref SimGridSupport

doc/doxygen/refman.tex

@@ -273,10 +273,10 @@ \chapter{SOCL OpenCL Extensions}
 \hypertarget{SOCLOpenclExtensions}{}
 \input{SOCLOpenclExtensions}
 
-\chapter{Hierarchical DAGS}
-\label{HierarchicalDAGS}
-\hypertarget{HierarchicalDAGS}{}
-\input{HierarchicalDAGS}
+\chapter{Recursive Tasks}
+\label{RecursiveTasks}
+\hypertarget{RecursiveTasks}{}
+\input{RecursiveTasks}
 
 \chapter{Parallel Workers}
 \label{ParallelWorker}

doc/doxygen_web_extensions/refman.tex

@@ -1,6 +1,6 @@
 % StarPU --- Runtime system for heterogeneous multicore architectures.
 %
-% Copyright (C) 2013-2023  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+% Copyright (C) 2013-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
 % Copyright (C) 2013       Simon Archipoff
 %
 % StarPU is free software; you can redistribute it and/or modify
@@ -103,10 +103,10 @@ \chapter{SOCL OpenCL Extensions}
 \hypertarget{SOCLOpenclExtensions}{}
 \input{SOCLOpenclExtensions}
 
-\chapter{Hierarchical DAGS}
-\label{HierarchicalDAGS}
-\hypertarget{HierarchicalDAGS}{}
-\input{HierarchicalDAGS}
+\chapter{Recursive Tasks}
+\label{RecursiveTasks}
+\hypertarget{RecursiveTasks}{}
+\input{RecursiveTasks}
 
 \chapter{Parallel Workers}
 \label{ParallelWorker}

include/starpu_task.h

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2023 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2009-2024 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  * Copyright (C) 2011       Télécom-SudParis
  * Copyright (C) 2016       Uppsala University
  *
@@ -210,13 +210,13 @@ typedef void (*starpu_opencl_func_t)(void **, void *);
 typedef void (*starpu_max_fpga_func_t)(void **, void *);
 
 /**
-   @ingroup API_Bubble Hierarchical Dags
+   @ingroup API_Bubble Recursive Tasks
    Bubble decision function
 */
-typedef int (*starpu_bubble_func_t)(struct starpu_task *t, void *arg);
+typedef int (*starpu_bubble_func_t)(struct starpu_task *, void *);
 
 /**
-   @ingroup API_Bubble Hierarchical Dags
+   @ingroup API_Bubble Recursive Tasks
    Bubble DAG generation function
 */
 typedef void (*starpu_bubble_gen_dag_func_t)(struct starpu_task *t, void *arg);
@@ -1423,31 +1423,31 @@ struct starpu_task
 #endif
 
 	/**
-	   When using hierarchical dags, the job identifier of the
+	   When using recursive tasks, the job identifier of the
 	   bubble task which created the current task
 	*/
 	unsigned long bubble_parent;
 
 	/**
-	   When using hierarchical dags, a pointer to the bubble
+	   When using recursive tasks, a pointer to the bubble
 	   decision function
 	*/
 	starpu_bubble_func_t bubble_func;
 
 	/**
-	   When using hierarchical dags, a pointer to an argument to
+	   When using recursive tasks, a pointer to an argument to
 	   be given when calling the bubble decision function
 	*/
 	void *bubble_func_arg;
 
 	/**
-	   When using hierarchical dags, a pointer to the bubble
+	   When using recursive tasks, a pointer to the bubble
 	   DAG generation function
 	*/
 	starpu_bubble_gen_dag_func_t bubble_gen_dag_func;
 
 	/**
-	   When using hierarchical dags, a pointer to an argument to
+	   When using recursive tasks, a pointer to an argument to
 	   be given when calling the bubble DAG generation function
 	 */
 	void *bubble_gen_dag_func_arg;
@@ -1538,7 +1538,7 @@ struct starpu_task
    starpu_codelet::nbuffers, or starpu_task::nbuffers if the former is
    \ref STARPU_VARIABLE_NBUFFERS.
 */
-#define STARPU_TASK_GET_NBUFFERS(task) ((unsigned)((task)->cl->nbuffers == STARPU_VARIABLE_NBUFFERS ? ((task)->nbuffers) : ((task)->cl->nbuffers)))
+#define STARPU_TASK_GET_NBUFFERS(task) ((unsigned) ( ((task)->cl) ? (((task)->cl->nbuffers == STARPU_VARIABLE_NBUFFERS) ? ((task)->nbuffers) : ((task)->cl->nbuffers)) : (task)->nbuffers))
 
 /**
    Return the \p i -th data handle of \p task. If \p task is defined

mpi/tests/mpi_data_cpy.c

@@ -46,12 +46,19 @@ int main(int argc, char **argv)
         int ret;
         int value = 0;
         starpu_data_handle_t *data;
+	struct starpu_conf conf;
         int mpi_init;
 	int i;
 
         MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
 
-        ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
+	starpu_conf_init(&conf);
+	starpu_conf_noworker(&conf);
+	conf.ncpus = -1;
+	conf.nmpi_ms = -1;
+	conf.ntcpip_ms = -1;
+
+        ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, &conf);
         STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
 
 	starpu_mpi_comm_rank(MPI_COMM_WORLD, &rank);
@@ -71,7 +78,8 @@ int main(int argc, char **argv)
 	{
 		int j;
 
-                starpu_mpi_task_insert(MPI_COMM_WORLD, &mycodelet, STARPU_RW, data[i%size], STARPU_VALUE, &rank, sizeof(rank), 0);
+                ret = starpu_mpi_task_insert(MPI_COMM_WORLD, &mycodelet, STARPU_RW, data[i%size], STARPU_VALUE, &rank, sizeof(rank), 0);
+		if (ret == -ENODEV) goto enodev;
 
                 for(j = 0; j<size; j++)
 		{
@@ -80,13 +88,18 @@ int main(int argc, char **argv)
         }
 
         starpu_task_wait_for_all();
+
+ enodev:
 	for(i=0; i<size; i++)
 	{
                 starpu_data_unregister(data[i]);
 	}
 
-	FPRINTF_MPI(stderr, "value after calculation: %d (expected %d)\n", value, INC_COUNT);
-        STARPU_ASSERT_MSG(value == INC_COUNT, "[rank %d] value %d is not the expected value %d\n", rank, value, INC_COUNT);
+	if (ret == 0)
+	{
+		FPRINTF_MPI(stderr, "value after calculation: %d (expected %d)\n", value, INC_COUNT);
+		STARPU_ASSERT_MSG(value == INC_COUNT, "[rank %d] value %d is not the expected value %d\n", rank, value, INC_COUNT);
+	}
 
         starpu_mpi_shutdown();
 

src/core/perfmodel/energy_model.c

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2008-2022  Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
+ * Copyright (C) 2008-2024  Université de Bordeaux, CNRS (LaBRI UMR 5800), 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
@@ -146,15 +146,18 @@ int starpu_energy_start(int workerid STARPU_ATTRIBUTE_UNUSED, enum starpu_worker
 #ifdef HAVE_NVMLDEVICEGETTOTALENERGYCONSUMPTION
 	case STARPU_CUDA_WORKER:
 	{
+		if (!_starpu_nvmlDeviceGetHandleByIndex || !_starpu_nvmlDeviceGetTotalEnergyConsumption)
+			return -1;
+
 		STARPU_ASSERT_MSG(workerid != -1, "For CUDA GPUs we measure each GPU separately, please specify a worker\n");
 		int devid = starpu_worker_get_devid(workerid);
-		int ret = nvmlDeviceGetHandleByIndex_v2(devid,  &device);
+		int ret = _starpu_nvmlDeviceGetHandleByIndex(devid,  &device);
 		if (ret != NVML_SUCCESS)
 		{
 			_STARPU_DISP("Could not get CUDA device %d from nvml\n", devid);
 			return -1;
 		}
-		ret = nvmlDeviceGetTotalEnergyConsumption(device, &energy_begin);
+		ret = _starpu_nvmlDeviceGetTotalEnergyConsumption(device, &energy_begin);
 		if (ret != NVML_SUCCESS)
 		{
 			_STARPU_DISP("Could not measure energy used by CUDA device %d\n", devid);
@@ -225,8 +228,11 @@ int starpu_energy_stop(struct starpu_perfmodel *model, struct starpu_task *task,
 #ifdef HAVE_NVMLDEVICEGETTOTALENERGYCONSUMPTION
 	case STARPU_CUDA_WORKER:
 	{
+		if (!_starpu_nvmlDeviceGetTotalEnergyConsumption)
+			return -1;
+
 		STARPU_ASSERT_MSG(workerid != -1, "For CUDA GPUs we measure each GPU separately, please specify a worker\n");
-		int ret = nvmlDeviceGetTotalEnergyConsumption(device, &energy_end);
+		int ret = _starpu_nvmlDeviceGetTotalEnergyConsumption(device, &energy_end);
 		if (ret != NVML_SUCCESS)
 			return -1;
 		energy = (energy_end - energy_begin) / 1000.;

src/core/perfmodel/perfmodel_bus.c

@@ -51,6 +51,8 @@
 #ifdef STARPU_HAVE_HWLOC
 #include <hwloc.h>
 #ifdef STARPU_HAVE_NVML_H
+#define nvmlDeviceGetPciInfo _starpu_nvmlDeviceGetPciInfo
+#define nvmlDeviceGetUUID _starpu_nvmlDeviceGetUUID
 #include <hwloc/nvml.h>
 #endif
 #ifndef HWLOC_API_VERSION
@@ -2090,7 +2092,7 @@ static hwloc_obj_t get_hwloc_cuda_obj(hwloc_topology_t topology, unsigned devid)
 #if defined(STARPU_HAVE_NVML_H) && !defined(STARPU_USE_CUDA0) && !defined(STARPU_USE_CUDA1)
 		nvmlDevice_t nvmldev = _starpu_cuda_get_nvmldev(&props);
 
-		if (nvmldev && _starpu_nvmlDeviceGetIndex)
+		if (nvmldev && _starpu_nvmlDeviceGetIndex && _starpu_nvmlDeviceGetPciInfo && _starpu_nvmlDeviceGetUUID)
 		{
 			unsigned int index;
 			if (_starpu_nvmlDeviceGetIndex(nvmldev, &index) == NVML_SUCCESS)