small modifs

examples/cholesky_2ctxs/cholesky/cholesky_models.c

@@ -1,7 +1,8 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009, 2010  Université de Bordeaux 1
- * Copyright (C) 2010  Centre National de la Recherche Scientifique
+ * Copyright (C) 2009, 2010-2011  Université de Bordeaux 1
+ * Copyright (C) 2010, 2011  Centre National de la Recherche Scientifique
+ * Copyright (C) 2011  Télécom-SudParis
  *
  * 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
@@ -26,7 +27,7 @@
 
 #include <starpu.h>
 
-//#define USE_PERTURBATION	1
+/* #define USE_PERTURBATION	1 */
 
 #ifdef USE_PERTURBATION
 #define PERTURBATE(a)	((starpu_drand48()*2.0f*(AMPL) + 1.0f - (AMPL))*(a))
@@ -43,7 +44,7 @@ static double cpu_chol_task_11_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/1000.0f*0.894/0.79176);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cpu_chol_task_11_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cpu_chol_task_11_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -58,7 +59,7 @@ static double cuda_chol_task_11_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/5.088633/0.9883);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cuda_chol_task_11_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cuda_chol_task_11_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -73,7 +74,7 @@ static double cpu_chol_task_21_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/7706.674/0.95/0.9965);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cpu_chol_task_21_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cpu_chol_task_21_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -88,7 +89,7 @@ static double cuda_chol_task_21_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/87.29520);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cuda_chol_task_21_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cuda_chol_task_21_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -103,7 +104,7 @@ static double cpu_chol_task_22_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/8.0760);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cpu_chol_task_22_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cpu_chol_task_22_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -118,34 +119,34 @@ static double cuda_chol_task_22_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/76.30666);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cuda_chol_task_22_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cuda_chol_task_22_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
 }
 
 struct starpu_perfmodel_t chol_model_11 = {
-	.per_arch = { 
-		[STARPU_CPU_DEFAULT] = { .cost_model = cpu_chol_task_11_cost },
-		[STARPU_CUDA_DEFAULT] = { .cost_model = cuda_chol_task_11_cost }
+	.per_arch = {
+		[STARPU_CPU_DEFAULT][0] = { .cost_model = cpu_chol_task_11_cost },
+		[STARPU_CUDA_DEFAULT][0] = { .cost_model = cuda_chol_task_11_cost }
 	},
 	.type = STARPU_HISTORY_BASED,
 	.symbol = "chol_model_11"
 };
 
 struct starpu_perfmodel_t chol_model_21 = {
-	.per_arch = { 
-		[STARPU_CPU_DEFAULT] = { .cost_model = cpu_chol_task_21_cost },
-		[STARPU_CUDA_DEFAULT] = { .cost_model = cuda_chol_task_21_cost }
+	.per_arch = {
+		[STARPU_CPU_DEFAULT][0] = { .cost_model = cpu_chol_task_21_cost },
+		[STARPU_CUDA_DEFAULT][0] = { .cost_model = cuda_chol_task_21_cost }
 	},
 	.type = STARPU_HISTORY_BASED,
 	.symbol = "chol_model_21"
 };
 
 struct starpu_perfmodel_t chol_model_22 = {
-	.per_arch = { 
-		[STARPU_CPU_DEFAULT] = { .cost_model = cpu_chol_task_22_cost },
-		[STARPU_CUDA_DEFAULT] = { .cost_model = cuda_chol_task_22_cost }
+	.per_arch = {
+		[STARPU_CPU_DEFAULT][0] = { .cost_model = cpu_chol_task_22_cost },
+		[STARPU_CUDA_DEFAULT][0] = { .cost_model = cuda_chol_task_22_cost }
 	},
 	.type = STARPU_HISTORY_BASED,
 	.symbol = "chol_model_22"

examples/cholesky_no_ctxs/cholesky/cholesky_models.c

@@ -1,7 +1,8 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2009, 2010  Université de Bordeaux 1
- * Copyright (C) 2010  Centre National de la Recherche Scientifique
+ * Copyright (C) 2009, 2010-2011  Université de Bordeaux 1
+ * Copyright (C) 2010, 2011  Centre National de la Recherche Scientifique
+ * Copyright (C) 2011  Télécom-SudParis
  *
  * 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
@@ -26,7 +27,7 @@
 
 #include <starpu.h>
 
-//#define USE_PERTURBATION	1
+/* #define USE_PERTURBATION	1 */
 
 #ifdef USE_PERTURBATION
 #define PERTURBATE(a)	((starpu_drand48()*2.0f*(AMPL) + 1.0f - (AMPL))*(a))
@@ -43,7 +44,7 @@ static double cpu_chol_task_11_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/1000.0f*0.894/0.79176);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cpu_chol_task_11_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cpu_chol_task_11_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -58,7 +59,7 @@ static double cuda_chol_task_11_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/5.088633/0.9883);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cuda_chol_task_11_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cuda_chol_task_11_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -73,7 +74,7 @@ static double cpu_chol_task_21_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/7706.674/0.95/0.9965);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cpu_chol_task_21_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cpu_chol_task_21_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -88,7 +89,7 @@ static double cuda_chol_task_21_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/87.29520);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cuda_chol_task_21_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cuda_chol_task_21_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -103,7 +104,7 @@ static double cpu_chol_task_22_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/8.0760);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cpu_chol_task_22_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cpu_chol_task_22_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
@@ -118,34 +119,34 @@ static double cuda_chol_task_22_cost(starpu_buffer_descr *descr)
 	double cost = (((double)(n)*n*n)/50.0f/10.75/76.30666);
 
 #ifdef STARPU_MODEL_DEBUG
-	printf("cuda_chol_task_22_cost n %d cost %e\n", n, cost);
+	FPRINTF(stdout, "cuda_chol_task_22_cost n %d cost %e\n", n, cost);
 #endif
 
 	return PERTURBATE(cost);
 }
 
 struct starpu_perfmodel_t chol_model_11 = {
-	.per_arch = { 
-		[STARPU_CPU_DEFAULT] = { .cost_model = cpu_chol_task_11_cost },
-		[STARPU_CUDA_DEFAULT] = { .cost_model = cuda_chol_task_11_cost }
+	.per_arch = {
+		[STARPU_CPU_DEFAULT][0] = { .cost_model = cpu_chol_task_11_cost },
+		[STARPU_CUDA_DEFAULT][0] = { .cost_model = cuda_chol_task_11_cost }
 	},
 	.type = STARPU_HISTORY_BASED,
 	.symbol = "chol_model_11"
 };
 
 struct starpu_perfmodel_t chol_model_21 = {
-	.per_arch = { 
-		[STARPU_CPU_DEFAULT] = { .cost_model = cpu_chol_task_21_cost },
-		[STARPU_CUDA_DEFAULT] = { .cost_model = cuda_chol_task_21_cost }
+	.per_arch = {
+		[STARPU_CPU_DEFAULT][0] = { .cost_model = cpu_chol_task_21_cost },
+		[STARPU_CUDA_DEFAULT][0] = { .cost_model = cuda_chol_task_21_cost }
 	},
 	.type = STARPU_HISTORY_BASED,
 	.symbol = "chol_model_21"
 };
 
 struct starpu_perfmodel_t chol_model_22 = {
-	.per_arch = { 
-		[STARPU_CPU_DEFAULT] = { .cost_model = cpu_chol_task_22_cost },
-		[STARPU_CUDA_DEFAULT] = { .cost_model = cuda_chol_task_22_cost }
+	.per_arch = {
+		[STARPU_CPU_DEFAULT][0] = { .cost_model = cpu_chol_task_22_cost },
+		[STARPU_CUDA_DEFAULT][0] = { .cost_model = cuda_chol_task_22_cost }
 	},
 	.type = STARPU_HISTORY_BASED,
 	.symbol = "chol_model_22"

src/sched_policies/deque_modeling_policy_data_aware.c

@@ -436,7 +436,7 @@ static int _dmda_push_task(struct starpu_task *task, unsigned prio, struct starp
 	int ntasks_best = -1;
 	double ntasks_best_end = 0.0;
 	int calibrating = 0;
-
+	
 	/* A priori, we know all estimations */
 	int unknown = 0;
 
@@ -444,7 +444,7 @@ static int _dmda_push_task(struct starpu_task *task, unsigned prio, struct starp
 	unsigned nimpl=0;
 	for (worker_in_ctx = 0; worker_in_ctx < nworkers_in_ctx; worker_in_ctx++)
 	{
-        worker = sched_ctx->workerid[worker_in_ctx];
+		worker = sched_ctx->workerid[worker_in_ctx];
 		for(nimpl  = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
 	 	{
 			fifo = dt->queue_array[worker_in_ctx];
@@ -462,7 +462,7 @@ static int _dmda_push_task(struct starpu_task *task, unsigned prio, struct starp
 			}
 
 			enum starpu_perf_archtype perf_arch = starpu_worker_get_perf_archtype(worker);
-			local_task_length[worker_in_ctx] = starpu_task_expected_length(task, perf_arch);
+			local_task_length[worker_in_ctx] = starpu_task_expected_length(task, perf_arch, nimpl);
 
 			//_STARPU_DEBUG("Scheduler dmda: task length (%lf) worker (%u) kernel (%u) \n", local_task_length[worker],worker,nimpl);
 

src/sched_policies/heft.c

@@ -242,6 +242,8 @@ static void compute_all_performance_predictions(struct starpu_task *task,
 
   unsigned nworkers = sched_ctx->nworkers_in_ctx;
 
+  unsigned nimpl;
+  unsigned best_impl = 0;
   unsigned worker, worker_in_ctx;
   for (worker_in_ctx = 0; worker_in_ctx < nworkers; worker_in_ctx++)
     {

src/util/starpu_insert_task_utils.c

@@ -144,6 +144,7 @@ int _starpu_pack_cl_args(size_t arg_buffer_size, char **arg_buffer, va_list varg
 int _starpu_insert_task_create_and_submit(char *arg_buffer, starpu_codelet *cl, struct starpu_task **task, va_list varg_list) {
         int arg_type;
 	unsigned current_buffer = 0;
+	unsigned ctx = 0;
 
 	struct insert_task_cb_wrapper *cl_arg_wrapper = (struct insert_task_cb_wrapper *) malloc(sizeof(struct insert_task_cb_wrapper));
 	STARPU_ASSERT(cl_arg_wrapper);