Also use physical number of cpus, since the users may use specific CPU cores

src/profiling/bound.c

@@ -209,7 +209,7 @@ static double** initialize_arch_duration(int maxdevid, unsigned* maxncore_table)
 static void initialize_duration(struct bound_task *task)
 {
 	struct _starpu_machine_config *conf = _starpu_get_machine_config();
-	task->duration[STARPU_CPU_WORKER] = initialize_arch_duration(1,&conf->topology.ncpus); 
+	task->duration[STARPU_CPU_WORKER] = initialize_arch_duration(1,&conf->topology.nhwcpus); 
 	task->duration[STARPU_CUDA_WORKER] = initialize_arch_duration(conf->topology.nhwcudagpus,NULL); 
 	task->duration[STARPU_OPENCL_WORKER] = initialize_arch_duration(conf->topology.nhwopenclgpus,NULL); 
 	task->duration[STARPU_MIC_WORKER] = initialize_arch_duration(conf->topology.nhwmicdevices,conf->topology.nmiccores);