small fixes

doc/doxygen/chapters/api/workers.doxy

@@ -252,12 +252,12 @@ on which device the memory needs to be allocated.
 
 \fn int starpu_memory_nodes_numa_id_to_devid(int osid)
 \ingroup API_Workers_Properties
-This function returns the identifier of the memory node associated to the NUMA
+Return the identifier of the memory node associated to the NUMA
 node identified by \p osid by the Operating System.
 
 \fn int starpu_memory_nodes_numa_devid_to_id(unsigned id);
 \ingroup API_Workers_Properties
-This function returns the Operating System identifier of the memory node
+Return the Operating System identifier of the memory node
 whose StarPU identifier is \p id.
 
 \fn char *starpu_worker_get_type_as_string(enum starpu_worker_archtype type)
@@ -287,7 +287,7 @@ Must be called after a potentially blocking call is complete, to
 restore the relax state in place before the corresponding relax_on.
 Decreases \c state_relax_refcnt. Calls to \ref starpu_worker_relax_on
 and \c starpu_worker_relax_off must be well parenthesized. This
-function is automatically called by \ref starpu_worker_unlock after the 
+function is automatically called by \ref starpu_worker_unlock after the
 target worker has been unlocked.
 
 \fn int starpu_worker_get_relax_state(void)

src/core/jobs.c

@@ -184,7 +184,7 @@ void _starpu_wait_job(struct _starpu_job *j)
 	{
 		STARPU_PTHREAD_COND_WAIT(&j->sync_cond, &j->sync_mutex);
 	}
-	
+
 	STARPU_PTHREAD_MUTEX_UNLOCK(&j->sync_mutex);
         _STARPU_LOG_OUT();
 }
@@ -368,7 +368,7 @@ void _starpu_handle_job_termination(struct _starpu_job *j)
 	 * to tell them that we will not exist any more before notifying the
 	 * tasks waiting for us
 	 *
-	 * For continuations, implicit dependency handles are only released 
+	 * For continuations, implicit dependency handles are only released
 	 * when the task fully completes */
 	if (j->implicit_dep_handle && !continuation)
 	{

src/core/perfmodel/perfmodel_bus.c

@@ -177,7 +177,7 @@ static void measure_bandwidth_between_host_and_dev_on_numa_with_cuda(int dev, in
 
 	/* Allocate a buffer on the host */
 	unsigned char *h_buffer;
-	
+
 #if defined(STARPU_HAVE_HWLOC)
 	if (nnuma_nodes > 1)
 	{
@@ -261,7 +261,7 @@ static void measure_bandwidth_between_host_and_dev_on_numa_with_cuda(int dev, in
 
 	/* Free buffers */
 	cudaHostUnregister(h_buffer);
-#if defined(STARPU_HAVE_HWLOC) 
+#if defined(STARPU_HAVE_HWLOC)
 	if (nnuma_nodes > 1)
 	{
 		/* NUMA mode activated */
@@ -669,11 +669,11 @@ static void measure_bandwidth_latency_between_numa(int numa_src, int numa_dst)
 		double start, end, timing;
 		unsigned iter;
 
-		unsigned char *h_buffer;	
+		unsigned char *h_buffer;
 		hwloc_obj_t obj_src = hwloc_get_obj_by_type(hwtopology, HWLOC_OBJ_NODE, numa_src);
 		h_buffer = hwloc_alloc_membind_nodeset(hwtopology, SIZE, obj_src->nodeset, HWLOC_MEMBIND_BIND, 0);
 
-		unsigned char *d_buffer;	
+		unsigned char *d_buffer;
 		hwloc_obj_t obj_dst = hwloc_get_obj_by_type(hwtopology, HWLOC_OBJ_NODE, numa_dst);
 		d_buffer = hwloc_alloc_membind_nodeset(hwtopology, SIZE, obj_dst->nodeset, HWLOC_MEMBIND_BIND, 0);
 
@@ -1038,7 +1038,7 @@ static int check_bus_affinity_file(void)
 
 	ret = fscanf(f, "# GPU\t");
 	STARPU_ASSERT(ret == 0);
-	
+
 	ret = fscanf(f, "NUMA%u\t", &dummy);
 
 	if (locked)
@@ -1255,7 +1255,7 @@ static double search_bus_best_latency(int src, char * type, int htod)
 		}
 #endif
 #ifdef STARPU_USE_OPENCL
-		if (strncmp(type, "OpenCL", 6) == 0)		
+		if (strncmp(type, "OpenCL", 6) == 0)
 		{
 			if (htod)
 				actual = opencldev_timing_per_numa[src*STARPU_MAXNUMANODES+numa].latency_htod;
@@ -1338,7 +1338,7 @@ static void write_bus_latency_file_content(void)
 			{
 				b_low = b_up = 0;
 
-				/* ---- Begin NUMA ---- */				
+				/* ---- Begin NUMA ---- */
 				b_up += nnumas;
 
 				if (src >= b_low && src < b_up && dst >= b_low && dst < b_up)
@@ -1347,7 +1347,7 @@ static void write_bus_latency_file_content(void)
 				/* copy interval to check numa index later */
 				unsigned numa_low = b_low;
 				unsigned numa_up = b_up;
-				
+
 				b_low += nnumas;
 				/* ---- End NUMA ---- */
 #ifdef STARPU_USE_CUDA
@@ -1390,7 +1390,7 @@ static void write_bus_latency_file_content(void)
 				b_up += nmic;
 				/* TODO Latency MIC */
 				b_low += nmic;
-#endif                                
+#endif
 #ifdef STARPU_USE_MPI_MASTER_SLAVE
 				b_up += nmpi_ms;
 				/* Modify MPI src and MPI dst if they contain the master node or not
@@ -1577,9 +1577,9 @@ static double search_bus_best_timing(int src, char * type, int htod)
 {
         /* Search the best latency for this node */
         double best = 0.0;
-        double actual = 0.0; 
+        double actual = 0.0;
         unsigned check = 0;
-        unsigned numa; 
+        unsigned numa;
         for (numa = 0; numa < nnumas; numa++)
         {
 #ifdef STARPU_USE_CUDA
@@ -1592,7 +1592,7 @@ static double search_bus_best_timing(int src, char * type, int htod)
 		}
 #endif
 #ifdef STARPU_USE_OPENCL
-		if (strncmp(type, "OpenCL", 6) == 0)		
+		if (strncmp(type, "OpenCL", 6) == 0)
 		{
 			if (htod)
 				actual = opencldev_timing_per_numa[src*STARPU_MAXNUMANODES+numa].timing_htod;
@@ -2852,7 +2852,7 @@ flat_cuda:
 	if (locked)
 		_starpu_fwrunlock(f);
 	fclose(f);
-	
+
 }
 
 static void generate_bus_platform_file(void)

src/core/topology.c

@@ -150,7 +150,7 @@ static int _starpu_get_logical_numa_node_worker(unsigned workerid)
 		struct _starpu_machine_topology *topology = &config->topology ;
 
 		hwloc_obj_t obj;
-		switch(worker->arch) 	
+		switch(worker->arch)
 		{
 			case STARPU_CPU_WORKER:
 				obj = hwloc_get_obj_by_type(topology->hwtopology, HWLOC_OBJ_PU, worker->bindid) ;
@@ -161,8 +161,8 @@ static int _starpu_get_logical_numa_node_worker(unsigned workerid)
 
 		return numa_get_logical_id(obj);
 	}
-	else		
-#endif 
+	else
+#endif
 	{
 		(void) workerid; /* unused */
 		return STARPU_NUMA_MAIN_RAM;
@@ -180,7 +180,7 @@ static int _starpu_get_physical_numa_node_worker(unsigned workerid)
 		struct _starpu_machine_topology *topology = &config->topology ;
 
 		hwloc_obj_t obj;
-		switch(worker->arch) 	
+		switch(worker->arch)
 		{
 			case STARPU_CPU_WORKER:
 				obj = hwloc_get_obj_by_type(topology->hwtopology, HWLOC_OBJ_PU, worker->bindid) ;
@@ -191,8 +191,8 @@ static int _starpu_get_physical_numa_node_worker(unsigned workerid)
 
 		return numa_get_physical_id(obj);
 	}
-	else		
-#endif 
+	else
+#endif
 	{
 		(void) workerid; /* unused */
 		return STARPU_NUMA_MAIN_RAM;
@@ -997,8 +997,8 @@ unsigned _starpu_topology_get_nnumanodes(struct _starpu_machine_config *config S
 		res = nnumanodes > 0 ? nnumanodes : 1 ;
 	}
 	else
-#endif 
-	{	
+#endif
+	{
 		res = 1;
 	}
 
@@ -1701,7 +1701,7 @@ _starpu_init_machine_config(struct _starpu_machine_config *config, int no_mp_con
 					(nworker_per_cuda * topology->ncudagpus) :
 					topology->ncudagpus;
 #endif
-			unsigned already_busy_cpus = mpi_ms_busy_cpus + mic_busy_cpus 
+			unsigned already_busy_cpus = mpi_ms_busy_cpus + mic_busy_cpus
 				+ cuda_busy_cpus
 				+ topology->nopenclgpus + topology->nsccdevices;
 
@@ -2074,7 +2074,7 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 				_starpu_simgrid_memory_node_set_host(memnode, host);
 #endif
 			}
-		}	
+		}
 #endif
 #if defined(STARPU_USE_OPENCL) && defined(STARPU_HAVE_HWLOC)
 		if (config->topology.nopenclgpus > 0)
@@ -2086,7 +2086,7 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 			unsigned nb_opencl_devices = 0, num = 0;
 
 			err = clGetPlatformIDs(_STARPU_OPENCL_PLATFORM_MAX, platform_id, &nb_platforms);
-			if (STARPU_UNLIKELY(err != CL_SUCCESS)) 
+			if (STARPU_UNLIKELY(err != CL_SUCCESS))
 				nb_platforms=0;
 
 			cl_device_type device_type = CL_DEVICE_TYPE_GPU|CL_DEVICE_TYPE_ACCELERATOR;
@@ -2134,7 +2134,7 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 						int memnode = _starpu_memory_node_register(STARPU_CPU_RAM, obj->logical_index);
 						STARPU_ASSERT_MSG(memnode < STARPU_MAXNUMANODES, "Wrong Memory Node : %d (only %d available)", memnode, STARPU_MAXNUMANODES);
 						numa_memory_nodes_to_hwloclogid[memnode] = obj->logical_index;
-						numa_memory_nodes_to_physicalid[memnode] = obj->os_index;	
+						numa_memory_nodes_to_physicalid[memnode] = obj->os_index;
 						nb_numa_nodes++;
 #ifdef STARPU_SIMGRID
 						snprintf(name, sizeof(name), "RAM%d", memnode);
@@ -2143,12 +2143,12 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 						_starpu_simgrid_memory_node_set_host(memnode, host);
 #endif
 					}
-				}	
+				}
 			}
 		}
 #endif
 	}
-	
+
 #if (defined(STARPU_USE_CUDA) || defined(STARPU_USE_OPENCL)) && defined(STARPU_HAVE_HWLOC)
 	//Found NUMA nodes from CUDA nodes
 	if (nb_numa_nodes != 0)
@@ -2164,7 +2164,7 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 	if (nnuma > STARPU_MAXNUMANODES)
 	{
 		_STARPU_MSG("Warning: %u NUMA nodes available. Only %u enabled. Use configure option --enable-maxnumanodes=xxx to update the maximum value of supported NUMA nodes.\n", _starpu_topology_get_nnumanodes(config), STARPU_MAXNUMANODES);
-		nnuma = STARPU_MAXNUMANODES;		
+		nnuma = STARPU_MAXNUMANODES;
 	}
 
 	unsigned numa;
@@ -2182,7 +2182,7 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 
 			numa_memory_nodes_to_hwloclogid[memnode] = numa_logical_id;
 			numa_memory_nodes_to_physicalid[memnode] = numa_physical_id;
-			nb_numa_nodes++;								
+			nb_numa_nodes++;
 
 #ifdef STARPU_SIMGRID
 			snprintf(name, sizeof(name), "RAM%d", memnode);
@@ -2201,7 +2201,7 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 
 			numa_memory_nodes_to_hwloclogid[memnode] = STARPU_NUMA_MAIN_RAM;
 			numa_memory_nodes_to_physicalid[memnode] = STARPU_NUMA_MAIN_RAM;
-			nb_numa_nodes++;								
+			nb_numa_nodes++;
 #ifdef STARPU_SIMGRID
 			char name[16];
 			msg_host_t host = _starpu_simgrid_get_host_by_name("RAM");
@@ -2210,9 +2210,9 @@ static void _starpu_init_numa_node(struct _starpu_machine_config *config)
 #endif
 		}
 
-	}	
-	
-	STARPU_ASSERT_MSG(nb_numa_nodes > 0, "No NUMA node found... We need at least one memory node !\n");	
+	}
+
+	STARPU_ASSERT_MSG(nb_numa_nodes > 0, "No NUMA node found... We need at least one memory node !\n");
 }
 
 static void _starpu_init_numa_bus()
@@ -2524,9 +2524,9 @@ _starpu_init_workers_binding_and_memory (struct _starpu_machine_config *config,
 					mpi_init[devid] = 1;
 					mpi_bindid[devid] = _starpu_get_next_bindid(config, preferred_binding, npreferred);
 					memory_node = mpi_memory_nodes[devid] = _starpu_memory_node_register(STARPU_MPI_MS_RAM, devid);
-		
+
 					for (numa = 0; numa < nb_numa_nodes; numa++)
-					{	
+					{
 						_starpu_register_bus(numa, memory_node);
 						_starpu_register_bus(memory_node, numa);
 					}
@@ -2780,4 +2780,3 @@ starpu_topology_print (FILE *output)
 		fprintf(output, "\n");
 	}
 }
-

src/datawizard/datawizard.h

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009, 2010, 2014, 2017  Université de Bordeaux
+ * Copyright (C) 2009, 2010, 2014  Université de Bordeaux
  * Copyright (C) 2010, 2013  CNRS
  *
  * StarPU is free software; you can redistribute it and/or modify

src/drivers/cpu/driver_cpu.c

@@ -166,12 +166,15 @@ static size_t _starpu_cpu_get_global_mem_size(int nodeid STARPU_ATTRIBUTE_UNUSED
 		int depth_node = hwloc_get_type_depth(topology->hwtopology, HWLOC_OBJ_NODE);
 
 		if (depth_node == HWLOC_TYPE_DEPTH_UNKNOWN)
-		     global_mem = hwloc_get_root_obj(topology->hwtopology)->memory.total_memory;
-		else {
-		     hwloc_obj_t obj = hwloc_get_obj_by_depth(topology->hwtopology, depth_node, nodeid);
-		     global_mem = obj->memory.local_memory;
-		     sprintf(name, "STARPU_LIMIT_CPU_NUMA_%d_MEM", obj->os_index);
-		     limit = starpu_get_env_number(name);
+		{
+			global_mem = hwloc_get_root_obj(topology->hwtopology)->memory.total_memory;
+		}
+		else
+		{
+			hwloc_obj_t obj = hwloc_get_obj_by_depth(topology->hwtopology, depth_node, nodeid);
+			global_mem = obj->memory.local_memory;
+			sprintf(name, "STARPU_LIMIT_CPU_NUMA_%d_MEM", obj->os_index);
+			limit = starpu_get_env_number(name);
 		}
 	}
 	else

src/drivers/cuda/driver_cuda.c

@@ -163,7 +163,7 @@ cudaStream_t starpu_cuda_get_local_in_transfer_stream()
 	int worker = starpu_worker_get_id_check();
 	int devid = starpu_worker_get_devid(worker);
 	cudaStream_t stream;
-	
+
 	stream = in_transfer_streams[devid];
 	STARPU_ASSERT(stream);
 	return stream;
@@ -251,11 +251,13 @@ void starpu_cuda_set_device(unsigned devid STARPU_ATTRIBUTE_UNUSED)
 	}
 #else
 	for (i = 0; i < conf->n_cuda_opengl_interoperability; i++)
+	{
 		if (conf->cuda_opengl_interoperability[i] == devid)
 		{
 			cures = cudaGLSetGLDevice(devid);
 			goto done;
 		}
+	}
 #endif
 
 	cures = cudaSetDevice(devid);