merge trunk

doc/doxygen/chapters/api/data_interfaces.doxy

@@ -1000,11 +1000,12 @@ DefiningANewDataInterface.
 \ingroup API_Data_Interfaces
 Allocate \p size bytes on node \p dst_node. This returns 0 if
 allocation failed, the allocation method should then return <c>-ENOMEM</c> as
-allocated size.
+allocated size. Deallocation must be done with starpu_free_on_node.
 
 \fn void starpu_free_on_node(unsigned dst_node, uintptr_t addr, size_t size)
 \ingroup API_Data_Interfaces
-Free \p addr of \p size bytes on node \p dst_node.
+Free \p addr of \p size bytes on node \p dst_node which was previously allocated
+with starpu_malloc_on_node.
 
 \fn int starpu_interface_copy(uintptr_t src, size_t src_offset, unsigned src_node, uintptr_t dst, size_t dst_offset, unsigned dst_node, size_t size, void *async_data)
 \ingroup API_Data_Interfaces

src/common/starpu_spinlock.c

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2010, 2012-2013  Université de Bordeaux 1
+ * Copyright (C) 2010, 2012-2014  Université de Bordeaux 1
  * Copyright (C) 2010, 2011, 2013, 2014  Centre National de la Recherche Scientifique
  *
  * StarPU is free software; you can redistribute it and/or modify
@@ -24,15 +24,17 @@
 int _starpu_spin_init(struct _starpu_spinlock *lock)
 {
 #if defined(STARPU_SPINLOCK_CHECK)
+	starpu_pthread_mutexattr_t errcheck_attr;
 //	memcpy(&lock->errcheck_lock, PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP, sizeof(PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP));
 	int ret;
-	ret = starpu_pthread_mutexattr_init(&lock->errcheck_attr);
+	ret = starpu_pthread_mutexattr_init(&errcheck_attr);
 	STARPU_CHECK_RETURN_VALUE(ret, "pthread_mutexattr_init");
 
-	ret = starpu_pthread_mutexattr_settype(&lock->errcheck_attr, PTHREAD_MUTEX_ERRORCHECK);
+	ret = starpu_pthread_mutexattr_settype(&errcheck_attr, PTHREAD_MUTEX_ERRORCHECK);
 	STARPU_ASSERT(!ret);
 
-	ret = starpu_pthread_mutex_init(&lock->errcheck_lock, &lock->errcheck_attr);
+	ret = starpu_pthread_mutex_init(&lock->errcheck_lock, &errcheck_attr);
+	starpu_pthread_mutexattr_destroy(&errcheck_attr);
 	return ret;
 #else
 	int ret = starpu_pthread_spin_init(&lock->lock, 0);
@@ -44,7 +46,6 @@ int _starpu_spin_init(struct _starpu_spinlock *lock)
 int _starpu_spin_destroy(struct _starpu_spinlock *lock STARPU_ATTRIBUTE_UNUSED)
 {
 #if defined(STARPU_SPINLOCK_CHECK)
-	starpu_pthread_mutexattr_destroy(&lock->errcheck_attr);
 	return starpu_pthread_mutex_destroy(&lock->errcheck_lock);
 #else
 	return starpu_pthread_spin_destroy(&lock->lock);

src/common/starpu_spinlock.h

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2010-2013  Université de Bordeaux 1
+ * Copyright (C) 2010-2014  Université de Bordeaux 1
  * Copyright (C) 2010, 2011, 2013, 2014  Centre National de la Recherche Scientifique
  *
  * StarPU is free software; you can redistribute it and/or modify
@@ -25,7 +25,6 @@
 struct _starpu_spinlock
 {
 #if defined(STARPU_SPINLOCK_CHECK)
-	starpu_pthread_mutexattr_t errcheck_attr;
 	starpu_pthread_mutex_t errcheck_lock;
 	const char *last_taker;
 #else

src/core/perfmodel/perfmodel_history.c

@@ -1177,7 +1177,7 @@ double _starpu_non_linear_regression_based_job_expected_perf(struct starpu_perfm
 			char archname[32];
 
 			starpu_perfmodel_get_arch_name(arch, archname, sizeof(archname), nimpl);
-			_STARPU_DISP("Warning: model %s is not calibrated enough for %s, forcing calibration for this run. Use the STARPU_CALIBRATE environment variable to control this.\n", model->symbol, archname);
+			_STARPU_DISP("Warning: model %s is not calibrated enough for %s (only %u measurements), forcing calibration for this run. Use the STARPU_CALIBRATE environment variable to control this.\n", model->symbol, archname, entry && entry->history_entry ? entry->history_entry->nsample : 0);
 			_starpu_set_calibrate_flag(1);
 			model->benchmarking = 1;
 		}
@@ -1227,7 +1227,7 @@ double _starpu_history_based_job_expected_perf(struct starpu_perfmodel *model, s
 		char archname[32];
 
 		starpu_perfmodel_get_arch_name(arch, archname, sizeof(archname), nimpl);
-		_STARPU_DISP("Warning: model %s is not calibrated enough for %s, forcing calibration for this run. Use the STARPU_CALIBRATE environment variable to control this.\n", model->symbol, archname);
+		_STARPU_DISP("Warning: model %s is not calibrated enough for %s (only %u measurements), forcing calibration for this run. Use the STARPU_CALIBRATE environment variable to control this.\n", model->symbol, archname, entry ? entry->nsample : 0);
 		_starpu_set_calibrate_flag(1);
 		model->benchmarking = 1;
 	}

src/core/workers.c

@@ -1225,6 +1225,7 @@ out:
 			STARPU_ASSERT(worker->local_ordered_tasks[n] == NULL);
 		_starpu_sched_ctx_list_delete(&worker->sched_ctx_list);
 		_starpu_job_list_delete(worker->terminated_jobs);
+		free(worker->local_ordered_tasks);
 	}
 }
 

src/datawizard/data_request.c

@@ -155,17 +155,22 @@ int _starpu_wait_data_request_completion(struct _starpu_data_request *r, unsigne
 {
 	int retval;
 	int do_delete = 0;
+	int completed;
 
 	unsigned local_node = _starpu_memory_node_get_local_key();
 
 	do
 	{
-		_starpu_spin_lock(&r->lock);
-
-		if (r->completed)
-			break;
-
-		_starpu_spin_unlock(&r->lock);
+		STARPU_HG_DISABLE_CHECKING(r->completed);
+		completed = r->completed;
+		STARPU_HG_ENABLE_CHECKING(r->completed);
+		if (completed)
+		{
+			_starpu_spin_lock(&r->lock);
+			if (r->completed)
+				break;
+			_starpu_spin_unlock(&r->lock);
+		}
 
 #ifndef STARPU_NON_BLOCKING_DRIVERS
 		_starpu_wake_all_blocked_workers_on_node(r->handling_node);

src/drivers/cuda/driver_cuda.c

@@ -574,6 +574,11 @@ int _starpu_cuda_driver_init(struct _starpu_worker_set *worker_set)
 		_STARPU_DEBUG("cuda (%s) dev id %u worker %u thread is ready to run on CPU %d !\n", devname, devid, i, worker->bindid);
 
 		worker->pipeline_length = starpu_get_env_number_default("STARPU_CUDA_PIPELINE", 2);
+		if (worker->pipeline_length > STARPU_MAX_PIPELINE)
+		{
+			_STARPU_DISP("Warning: STARPU_CUDA_PIPELINE is %u, but STARPU_MAX_PIPELINE is only %u", worker->pipeline_length, STARPU_MAX_PIPELINE);
+			worker->pipeline_length = STARPU_MAX_PIPELINE;
+		}
 		_STARPU_TRACE_WORKER_INIT_END(worker_set->workers[i].workerid);
 	}
 

src/drivers/opencl/driver_opencl.c

@@ -597,6 +597,11 @@ int _starpu_opencl_driver_init(struct _starpu_worker *worker)
 	snprintf(worker->short_name, sizeof(worker->short_name), "OpenCL %u", devid);
 
 	worker->pipeline_length = starpu_get_env_number_default("STARPU_OPENCL_PIPELINE", 2);
+	if (worker->pipeline_length > STARPU_MAX_PIPELINE)
+	{
+		_STARPU_DISP("Warning: STARPU_OPENCL_PIPELINE is %u, but STARPU_MAX_PIPELINE is only %u", worker->pipeline_length, STARPU_MAX_PIPELINE);
+		worker->pipeline_length = STARPU_MAX_PIPELINE;
+	}
 
 	_STARPU_DEBUG("OpenCL (%s) dev id %d thread is ready to run on CPU %d !\n", devname, devid, worker->bindid);