perf_model: rename some functions

	    starpu_perfmodel_get_per_arch --> starpu_perfmodel_get_model_per_devices
	    starpu_perfmodel_set_per_arch_cost_function --> starpu_perfmodel_set_per_devices_cost_function
	    starpu_perfmodel_set_per_arch_size_base --> starpu_perfmodel_set_per_devices_size_base

examples/cholesky/cholesky_models.c

@@ -138,14 +138,14 @@ void initialize_chol_model(struct starpu_perfmodel* model, char * symbol,
 
 	starpu_perfmodel_init(NULL, model);
 
-	per_arch = starpu_perfmodel_get_per_arch(model, 0, STARPU_CPU_WORKER, 0, 1, -1);
+	per_arch = starpu_perfmodel_get_model_per_devices(model, 0, STARPU_CPU_WORKER, 0, 1, -1);
         per_arch->cost_function = cpu_cost_function;
 	// We could also call directly:
-	// starpu_perfmodel_set_per_arch_cost_function(model, 0, cpu_cost_function, STARPU_CPU_WORKER, 0, 1, -1);
+	// starpu_perfmodel_set_per_devices_cost_function(model, 0, cpu_cost_function, STARPU_CPU_WORKER, 0, 1, -1);
 
 	if(starpu_worker_get_count_by_type(STARPU_CUDA_WORKER) != 0)
 	{
-	     	per_arch = starpu_perfmodel_get_per_arch(model, 0, STARPU_CUDA_WORKER, 0, 1, -1);
+	     	per_arch = starpu_perfmodel_get_model_per_devices(model, 0, STARPU_CUDA_WORKER, 0, 1, -1);
 		per_arch->cost_function = cuda_cost_function;
 
 	}

examples/heat/lu_kernels_model.c

@@ -224,10 +224,10 @@ void initialize_lu_kernels_model(struct starpu_perfmodel* model, char * symbol,
 
 	starpu_perfmodel_init(NULL, model);
 
-	starpu_perfmodel_set_per_arch_cost_function(model, 0, cpu_cost_function, STARPU_CPU_WORKER, 0, 1, -1);
+	starpu_perfmodel_set_per_devices_cost_function(model, 0, cpu_cost_function, STARPU_CPU_WORKER, 0, 1, -1);
 
 	if(starpu_worker_get_count_by_type(STARPU_CUDA_WORKER) != 0)
 	{
-		starpu_perfmodel_set_per_arch_cost_function(model, 0, cuda_cost_function, STARPU_CUDA_WORKER, 0, 1, -1);
+		starpu_perfmodel_set_per_devices_cost_function(model, 0, cuda_cost_function, STARPU_CUDA_WORKER, 0, 1, -1);
 	}
 }

include/starpu_perfmodel.h

@@ -157,17 +157,17 @@ void starpu_perfmodel_init(FILE *f, struct starpu_perfmodel *model);
 
 struct starpu_perfmodel_arch *starpu_worker_get_perf_archtype(int workerid, unsigned sched_ctx_id);
 
-struct starpu_perfmodel_per_arch *starpu_perfmodel_get_model_per_arch(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, unsigned impl);
-
 int starpu_perfmodel_load_symbol(const char *symbol, struct starpu_perfmodel *model);
 int starpu_perfmodel_unload_model(struct starpu_perfmodel *model);
 int starpu_get_narch_combs();
 int starpu_perfmodel_arch_comb_add(int ndevices, struct starpu_perfmodel_device* devices);
 int starpu_perfmodel_arch_comb_get(int ndevices, struct starpu_perfmodel_device *devices);
 
-struct starpu_perfmodel_per_arch *starpu_perfmodel_get_per_arch(struct starpu_perfmodel *model, int impl, ...);
-int starpu_perfmodel_set_per_arch_cost_function(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_cost_function func, ...);
-int starpu_perfmodel_set_per_arch_size_base(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_size_base func, ...);
+struct starpu_perfmodel_per_arch *starpu_perfmodel_get_model_per_arch(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, unsigned impl);
+struct starpu_perfmodel_per_arch *starpu_perfmodel_get_model_per_devices(struct starpu_perfmodel *model, int impl, ...);
+
+int starpu_perfmodel_set_per_devices_cost_function(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_cost_function func, ...);
+int starpu_perfmodel_set_per_devices_size_base(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_size_base func, ...);
 
 void starpu_perfmodel_debugfilepath(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, char *path, size_t maxlen, unsigned nimpl);
 char* starpu_perfmodel_get_archtype_name(enum starpu_worker_archtype archtype);

src/core/perfmodel/perfmodel_history.c

@@ -1410,14 +1410,6 @@ void starpu_perfmodel_update_history(struct starpu_perfmodel *model, struct star
 	_starpu_set_calibrate_flag(1);
 }
 
-struct starpu_perfmodel_per_arch *starpu_perfmodel_get_model_per_arch(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, unsigned impl)
-{
-	int comb = starpu_perfmodel_arch_comb_get(arch->ndevices, arch->devices);
-	if(comb == -1) return NULL;
-
-	return &model->per_arch[comb][impl];
-}
-
 int starpu_perfmodel_list_combs(FILE *output, struct starpu_perfmodel *model)
 {
 	int comb;
@@ -1439,7 +1431,15 @@ int starpu_perfmodel_list_combs(FILE *output, struct starpu_perfmodel *model)
 	return 0;
 }
 
-struct starpu_perfmodel_per_arch *_starpu_perfmodel_get_per_arch(struct starpu_perfmodel *model, int impl, va_list varg_list)
+struct starpu_perfmodel_per_arch *starpu_perfmodel_get_model_per_arch(struct starpu_perfmodel *model, struct starpu_perfmodel_arch *arch, unsigned impl)
+{
+	int comb = starpu_perfmodel_arch_comb_get(arch->ndevices, arch->devices);
+	if(comb == -1) return NULL;
+
+	return &model->per_arch[comb][impl];
+}
+
+struct starpu_perfmodel_per_arch *_starpu_perfmodel_get_model_per_devices(struct starpu_perfmodel *model, int impl, va_list varg_list)
 {
 	struct starpu_perfmodel_arch arch;
 	va_list varg_list_copy;
@@ -1500,38 +1500,38 @@ struct starpu_perfmodel_per_arch *_starpu_perfmodel_get_per_arch(struct starpu_p
 	return &model->per_arch[comb][impl];
 }
 
-struct starpu_perfmodel_per_arch *starpu_perfmodel_get_per_arch(struct starpu_perfmodel *model, int impl, ...)
+struct starpu_perfmodel_per_arch *starpu_perfmodel_get_model_per_devices(struct starpu_perfmodel *model, int impl, ...)
 {
 	va_list varg_list;
 	struct starpu_perfmodel_per_arch *per_arch;
 
 	va_start(varg_list, impl);
-	per_arch = _starpu_perfmodel_get_per_arch(model, impl, varg_list);
+	per_arch = _starpu_perfmodel_get_model_per_devices(model, impl, varg_list);
 	va_end(varg_list);
 
 	return per_arch;
 }
 
-int starpu_perfmodel_set_per_arch_cost_function(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_cost_function func, ...)
+int starpu_perfmodel_set_per_devices_cost_function(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_cost_function func, ...)
 {
 	va_list varg_list;
 	struct starpu_perfmodel_per_arch *per_arch;
 
 	va_start(varg_list, func);
-	per_arch = _starpu_perfmodel_get_per_arch(model, impl, varg_list);
+	per_arch = _starpu_perfmodel_get_model_per_devices(model, impl, varg_list);
 	per_arch->cost_function = func;
 	va_end(varg_list);
 
 	return 0;
 }
 
-int starpu_perfmodel_set_per_arch_size_base(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_size_base func, ...)
+int starpu_perfmodel_set_per_devices_size_base(struct starpu_perfmodel *model, int impl, starpu_perfmodel_per_arch_size_base func, ...)
 {
 	va_list varg_list;
 	struct starpu_perfmodel_per_arch *per_arch;
 
 	va_start(varg_list, func);
-	per_arch = _starpu_perfmodel_get_per_arch(model, impl, varg_list);
+	per_arch = _starpu_perfmodel_get_model_per_devices(model, impl, varg_list);
 	per_arch->size_base = func;
 	va_end(varg_list);
 

tests/sched_policies/simple_cpu_gpu_sched.c

@@ -106,13 +106,13 @@ init_perfmodels_gpu(int gpu_type)
 	starpu_worker_get_ids_by_type(gpu_type, worker_gpu_ids, nb_worker_gpu);
 	for(worker_gpu = 0 ; worker_gpu < nb_worker_gpu ; worker_gpu ++)
 	{
-		starpu_perfmodel_set_per_arch_cost_function(&model_cpu_task, 0, cpu_task_gpu,
-							    gpu_type, starpu_worker_get_devid(worker_gpu_ids[worker_gpu]), 1,
-							    -1);
+		starpu_perfmodel_set_per_devices_cost_function(&model_cpu_task, 0, cpu_task_gpu,
+							       gpu_type, starpu_worker_get_devid(worker_gpu_ids[worker_gpu]), 1,
+							       -1);
 
-		starpu_perfmodel_set_per_arch_cost_function(&model_gpu_task, 0, gpu_task_gpu,
-							    gpu_type, starpu_worker_get_devid(worker_gpu_ids[worker_gpu]), 1,
-						    -1);
+		starpu_perfmodel_set_per_devices_cost_function(&model_gpu_task, 0, gpu_task_gpu,
+							       gpu_type, starpu_worker_get_devid(worker_gpu_ids[worker_gpu]), 1,
+							       -1);
 	}
 }
 
@@ -124,8 +124,8 @@ init_perfmodels(void)
 	starpu_perfmodel_init(NULL, &model_cpu_task);
 	starpu_perfmodel_init(NULL, &model_gpu_task);
 
-	starpu_perfmodel_set_per_arch_cost_function(&model_cpu_task, 0, cpu_task_cpu, STARPU_CPU_WORKER, 0, 1, -1);
-	starpu_perfmodel_set_per_arch_cost_function(&model_gpu_task, 0, gpu_task_cpu, STARPU_CPU_WORKER, 0, 1, -1);
+	starpu_perfmodel_set_per_devices_cost_function(&model_cpu_task, 0, cpu_task_cpu, STARPU_CPU_WORKER, 0, 1, -1);
+	starpu_perfmodel_set_per_devices_cost_function(&model_gpu_task, 0, gpu_task_cpu, STARPU_CPU_WORKER, 0, 1, -1);
 
 	// We need to set the cost function for each combination with a CUDA or a OpenCL worker
 	init_perfmodels_gpu(STARPU_CUDA_WORKER);