too expensive the lock in copmpute_all_performance_prediction, exp_start is update in post_exec/pre_exec/etc also

sc_hypervisor/src/hypervisor_policies/feft_lp_policy.c

@@ -143,8 +143,8 @@ static feft_lp_handle_idle_cycle(unsigned sched_ctx, int worker)
 			
 			if(sc_hypervisor_check_idle(sched_ctx, worker))
 			{
-//				_try_resizing();
-				sc_hypervisor_move_workers(sched_ctx, 3 - sched_ctx, &worker, 1, 1);
+				_try_resizing();
+//				sc_hypervisor_move_workers(sched_ctx, 3 - sched_ctx, &worker, 1, 1);
 			}
 		}
 		starpu_pthread_mutex_unlock(&act_hypervisor_mutex);

sc_hypervisor/src/hypervisor_policies/teft_lp_policy.c

@@ -274,8 +274,8 @@ static int teft_lp_handle_idle_cycle(unsigned sched_ctx, int worker)
 			
 			if(sc_hypervisor_check_idle(sched_ctx, worker))
 			{
-//				_try_resizing();
-				sc_hypervisor_move_workers(sched_ctx, 3 - sched_ctx, &worker, 1, 1);
+				_try_resizing();
+//				sc_hypervisor_move_workers(sched_ctx, 3 - sched_ctx, &worker, 1, 1);
 			}
 		}
 		starpu_pthread_mutex_unlock(&act_hypervisor_mutex);

sc_hypervisor/src/sc_hypervisor.c

@@ -137,7 +137,7 @@ struct starpu_sched_ctx_performance_counters* sc_hypervisor_init(struct sc_hyper
 	char* vel_gap = getenv("SC_HYPERVISOR_MAX_VELOCITY_GAP");
 	hypervisor.max_velocity_gap = vel_gap ? atof(vel_gap) : SC_VELOCITY_MAX_GAP_DEFAULT;
 	char* crit =  getenv("SC_HYPERVISOR_TRIGGER_RESIZE");
-	hypervisor.resize_criteria = !crit ? SC_NOTHING : strcmp(crit,"idle") == 0 ? SC_IDLE : (strcmp(crit,"speed") == 0 ? SC_VELOCITY : SC_NOTHING);
+	hypervisor.resize_criteria = !crit ? SC_IDLE : strcmp(crit,"idle") == 0 ? SC_IDLE : (strcmp(crit,"speed") == 0 ? SC_VELOCITY : SC_NOTHING);
 
 	starpu_pthread_mutex_init(&act_hypervisor_mutex, NULL);
 	hypervisor.start_executing_time = starpu_timing_now();

src/sched_policies/deque_modeling_policy_data_aware.c

@@ -547,13 +547,7 @@ static void compute_all_performance_predictions(struct starpu_task *task,
 		unsigned memory_node = starpu_worker_get_memory_node(worker);
 
 		/* Sometimes workers didn't take the tasks as early as we expected */
-		starpu_pthread_mutex_t *sched_mutex;
-		starpu_pthread_cond_t *sched_cond;
-		starpu_worker_get_sched_condition(worker, &sched_mutex, &sched_cond);
-
-		STARPU_PTHREAD_MUTEX_LOCK(sched_mutex);
-		fifo->exp_start = STARPU_MAX(fifo->exp_start, starpu_timing_now());
-		STARPU_PTHREAD_MUTEX_UNLOCK(sched_mutex);
+		double exp_start = STARPU_MAX(fifo->exp_start, starpu_timing_now());
 
 		for(nimpl  = 0; nimpl < STARPU_MAXIMPLEMENTATIONS; nimpl++)
 	 	{
@@ -639,7 +633,7 @@ static void compute_all_performance_predictions(struct starpu_task *task,
 			if (unknown)
 				continue;
 
-			exp_end[worker_ctx][nimpl] = fifo->exp_start + fifo->exp_len + local_task_length[worker_ctx][nimpl];
+			exp_end[worker_ctx][nimpl] = exp_start + fifo->exp_len + local_task_length[worker_ctx][nimpl];
 
 			if (exp_end[worker_ctx][nimpl] < best_exp_end)
 			{