Add starpu_worker_get_by_type and starpu_worker_get_by_devid

ChangeLog

@@ -1,6 +1,6 @@
 # StarPU --- Runtime system for heterogeneous multicore architectures.
 #
-# Copyright (C) 2009-2012  Université de Bordeaux 1
+# Copyright (C) 2009-2013  Université de Bordeaux 1
 # Copyright (C) 2010, 2011, 2012, 2013  Centre National de la Recherche Scientifique
 #
 # StarPU is free software; you can redistribute it and/or modify
@@ -114,6 +114,9 @@ Changes:
   * StarPU can now use poti to generate paje traces.
   * Rename scheduling policy "parallel greedy" to "parallel eager"
 
+Small features:
+  * Add starpu_worker_get_by_type and starpu_worker_get_by_devid
+
 Small changes:
   * STARPU_NCPU should now be used instead of STARPU_NCPUS. STARPU_NCPUS is
 	still available for compatibility reasons.

doc/chapters/basic-api.texi

@@ -286,6 +286,17 @@ chosen by the means of the @code{starpu_worker_get_count_by_type} function, or
 by passing a value greater or equal to @code{STARPU_NMAXWORKERS}.
 @end deftypefun
 
+@deftypefun int starpu_worker_get_by_type ({enum starpu_archtype} @var{type}, int @var{n})
+This returns the identifier of the @var{n}-th worker that has the specified type
+@var{type}. If there are no such worker, -1 is returned.
+@end deftypefun
+
+@deftypefun int starpu_worker_get_by_devid ({enum starpu_archtype} @var{type}, int @var{devid})
+This returns the identifier of the worker that has the specified type
+@var{type} and devid @var{devid} (which may not be the n-th, if some devices are
+skipped for instance). If there are no such worker, -1 is returned.
+@end deftypefun
+
 @deftypefun int starpu_worker_get_devid (int @var{id})
 This functions returns the device id of the given worker. The worker
 should be identified with the value returned by the @code{starpu_worker_get_id} function. In the case of a

doc/chapters/perf-optimization.texi

@@ -19,6 +19,7 @@ TODO: improve!
 * Task distribution vs Data transfer::
 * Data prefetch::
 * Power-based scheduling::
+* Forcing scheduling::
 * Profiling::
 * CUDA-specific optimizations::
 * Performance debugging::
@@ -402,6 +403,20 @@ consumption for that series of tasks, divide it by a thousand, repeat for
 varying kinds of tasks and task sizes, and eventually feed StarPU
 with these manual measurements through @code{starpu_perfmodel_update_history}.
 
+@node Static scheduling
+@section Static scheduling
+
+In some cases, one may want to force some scheduling, for instance force a given
+set of tasks to GPU0, another set to GPU1, etc. while letting some other tasks
+be scheduled on any other device. This can indeed be useful to guide StarPU into
+some work distribution, while still letting some degree of dynamism. For
+instance, to force execution of a task on CUDA0:
+
+@example
+task->execute_on_a_specific_worker = 1;
+task->worker = starpu_worker_get_by_type(STARPU_CUDA_WORKER, 0);
+@end example
+
 @node Profiling
 @section Profiling
 

include/starpu_worker.h

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2012  Université de Bordeaux 1
+ * Copyright (C) 2009-2013  Université de Bordeaux 1
  * Copyright (C) 2010, 2011, 2012  Centre National de la Recherche Scientifique
  *
  * StarPU is free software; you can redistribute it and/or modify
@@ -79,6 +79,12 @@ int starpu_worker_get_count_by_type(enum starpu_archtype type);
  * equal to STARPU_NMAXWORKERS. */
 int starpu_worker_get_ids_by_type(enum starpu_archtype type, int *workerids, int maxsize);
 
+/* Return the identifier of the n-th worker of a specific type */
+int starpu_worker_get_by_type(enum starpu_archtype type, int num);
+
+/* Return the identifier of the worker devid of a specific type */
+int starpu_worker_get_by_devid(enum starpu_archtype type, int devid);
+
 /* StarPU associates a unique human readable string to each processing unit.
  * This function copies at most the "maxlen" first bytes of the unique
  * string associated to a worker identified by its identifier "id" into

src/core/perfmodel/perfmodel.c

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2012  Université de Bordeaux 1
+ * Copyright (C) 2009-2013  Université de Bordeaux 1
  * Copyright (C) 2010, 2011, 2012  Centre National de la Recherche Scientifique
  * Copyright (C) 2011  Télécom-SudParis
  *
@@ -224,7 +224,6 @@ double starpu_task_expected_conversion_time(struct starpu_task *task,
 					    unsigned nimpl)
 {
 	unsigned i;
-	int err;
 	double sum = 0.0;
 	int node;
 
@@ -235,25 +234,22 @@ double starpu_task_expected_conversion_time(struct starpu_task *task,
 	 */
 #ifdef STARPU_USE_CPU
 	int cpu_worker, cpu_node;
-	err = starpu_worker_get_ids_by_type(STARPU_CPU_WORKER,
-					    &cpu_worker, 1);
-	if (err != 1 && err != -ERANGE)
+	cpu_worker = starpu_worker_get_by_type(STARPU_CPU_WORKER, 0);
+	if (cpu_worker == -1)
 		return 0.0;
 	cpu_node = starpu_worker_get_memory_node(cpu_worker);
 #endif
 #ifdef STARPU_USE_CUDA
 	int cuda_worker, cuda_node;
-	err = starpu_worker_get_ids_by_type(STARPU_CUDA_WORKER,
-					    &cuda_worker, 1);
-	if (err != 1 && err != -ERANGE)
+	cuda_worker = starpu_worker_get_by_type(STARPU_CUDA_WORKER, 0);
+	if (cuda_worker == -1)
 		return 0.0;
 	cuda_node = starpu_worker_get_memory_node(cuda_worker);
 #endif
 #ifdef STARPU_USE_OPENCL
 	int opencl_worker, opencl_node;
-	err = starpu_worker_get_ids_by_type(STARPU_OPENCL_WORKER,
-					    &opencl_worker, 1);
-	if (err != 1 && err != -ERANGE)
+	opencl_worker = starpu_worker_get_by_type(STARPU_OPENCL_WORKER, 0);
+	if (opencl_worker == -1)
 		return 0.0;
 
 	opencl_node = starpu_worker_get_memory_node(opencl_worker);

src/core/workers.c

@@ -1243,6 +1243,40 @@ int starpu_worker_get_ids_by_type(enum starpu_archtype type, int *workerids, int
 	return cnt;
 }
 
+int starpu_worker_get_by_type(enum starpu_archtype type, int num)
+{
+	unsigned nworkers = starpu_worker_get_count();
+
+	int cnt = 0;
+
+	unsigned id;
+	for (id = 0; id < nworkers; id++)
+	{
+		if (starpu_worker_get_type(id) == type)
+		{
+			if (num == cnt)
+				return id;
+			cnt++;
+		}
+	}
+
+	/* Not found */
+	return -1;
+}
+
+int starpu_worker_get_by_devid(enum starpu_archtype type, int devid)
+{
+	unsigned nworkers = starpu_worker_get_count();
+
+	unsigned id;
+	for (id = 0; id < nworkers; id++)
+		if (starpu_worker_get_type(id) == type && starpu_worker_get_devid(id) == devid)
+			return id;
+
+	/* Not found */
+	return -1;
+}
+
 void starpu_worker_get_name(int id, char *dst, size_t maxlen)
 {
 	char *name = config.workers[id].name;

src/drivers/cpu/driver_cpu.c

@@ -178,13 +178,10 @@ static int execute_job_on_cpu(struct _starpu_job *j, struct starpu_task *worker_
 static struct _starpu_worker*
 _starpu_get_worker_from_driver(struct starpu_driver *d)
 {
-	int workers[d->id.cpu_id + 1];
-	int nworkers;
-	nworkers = starpu_worker_get_ids_by_type(STARPU_CPU_WORKER, workers, d->id.cpu_id+1);
-	if (nworkers >= 0 && (unsigned) nworkers < d->id.cpu_id)
-		return NULL; // No device was found.
-
-	return _starpu_get_worker_struct(workers[d->id.cpu_id]);
+	int n = starpu_worker_get_by_devid(STARPU_CPU_WORKER, d->id.cpu_id);
+	if (n == -1)
+		return NULL;
+	return _starpu_get_worker_struct(n);
 }
 
 int _starpu_cpu_driver_init(struct starpu_driver *d)

src/drivers/cuda/driver_cuda.c

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009-2012  Université de Bordeaux 1
+ * Copyright (C) 2009-2013  Université de Bordeaux 1
  * Copyright (C) 2010  Mehdi Juhoor <mjuhoor@gmail.com>
  * Copyright (C) 2010, 2011, 2012  Centre National de la Recherche Scientifique
  * Copyright (C) 2011  Télécom-SudParis
@@ -629,15 +629,11 @@ int _starpu_run_cuda(struct starpu_driver *d)
 {
 	STARPU_ASSERT(d && d->type == STARPU_CUDA_WORKER);
 
-	int workers[d->id.cuda_id + 1];
-	int nworkers;
-	nworkers = starpu_worker_get_ids_by_type(STARPU_CUDA_WORKER, workers, d->id.cuda_id+1);
-	if (nworkers >= 0 && (unsigned) nworkers < d->id.cuda_id)
-		return -ENODEV;
-	
+	int workerid = starpu_worker_get_by_devid(STARPU_CUDA_WORKER, d->id.cuda_id);
+
 	_STARPU_DEBUG("Running cuda %u from the application\n", d->id.cuda_id);
 
-	struct _starpu_worker *workerarg = _starpu_get_worker_struct(workers[d->id.cuda_id]);
+	struct _starpu_worker *workerarg = _starpu_get_worker_struct(workerid);
 
 	workerarg->set = NULL;
 	workerarg->worker_is_initialized = 0;