Lines Matching full:g
47 #define tprintf(x...) do { if (g && g->p.show_details >= 0) printf(x); } while (0)
53 #define dprintf(x...) do { if (g && g->p.show_details >= 1) printf(x); } while (0)
162 static struct global_info *g = NULL; variable
173 OPT_STRING('G', "mb_global" , &p0.mb_global_str, "MB", "global memory (MBs)"),
230 for (i = 0; i < g->p.nr_nodes; i++) { in nr_numa_nodes()
292 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_cpu()
295 if (target_cpu < 0 || target_cpu >= g->p.nr_cpus) in bind_to_cpu()
338 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_node()
386 ret = set_mempolicy(MPOL_DEFAULT, NULL, g->p.nr_nodes-1); in mempol_restore()
449 if (ret && !g->print_once) { in alloc_data()
450 g->print_once = 1; in alloc_data()
456 if (ret && !g->print_once) { in alloc_data()
457 g->print_once = 1; in alloc_data()
506 return alloc_data(bytes, MAP_SHARED, 1, g->p.init_cpu0, g->p.thp, g->p.init_random); in zalloc_shared_data()
514 return alloc_data(bytes, MAP_SHARED, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_shared_data()
523 return alloc_data(bytes, MAP_PRIVATE, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_private_data()
541 if (!g->p.cpu_list_str) in parse_setup_cpu_list()
544 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_cpu_list()
546 str0 = str = strdup(g->p.cpu_list_str); in parse_setup_cpu_list()
581 BUG_ON(step <= 0 || step >= g->p.nr_cpus); in parse_setup_cpu_list()
593 BUG_ON(bind_len <= 0 || bind_len > g->p.nr_cpus); in parse_setup_cpu_list()
606 if (bind_cpu_0 >= g->p.nr_cpus || bind_cpu_1 >= g->p.nr_cpus) { in parse_setup_cpu_list()
607 printf("\nTest not applicable, system has only %d CPUs.\n", g->p.nr_cpus); in parse_setup_cpu_list()
620 size_t size = CPU_ALLOC_SIZE(g->p.nr_cpus); in parse_setup_cpu_list()
626 if (t >= g->p.nr_tasks) { in parse_setup_cpu_list()
630 td = g->threads + t; in parse_setup_cpu_list()
640 td->bind_cpumask = CPU_ALLOC(g->p.nr_cpus); in parse_setup_cpu_list()
644 if (cpu < 0 || cpu >= g->p.nr_cpus) { in parse_setup_cpu_list()
658 if (t < g->p.nr_tasks) in parse_setup_cpu_list()
659 printf("# NOTE: %d tasks bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_cpu_list()
689 if (!g->p.node_list_str) in parse_setup_node_list()
692 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_node_list()
694 str0 = str = strdup(g->p.node_list_str); in parse_setup_node_list()
728 BUG_ON(step <= 0 || step >= g->p.nr_nodes); in parse_setup_node_list()
741 if (bind_node_0 >= g->p.nr_nodes || bind_node_1 >= g->p.nr_nodes) { in parse_setup_node_list()
742 printf("\nTest not applicable, system has only %d nodes.\n", g->p.nr_nodes); in parse_setup_node_list()
753 if (t >= g->p.nr_tasks || !node_has_cpus(bind_node)) { in parse_setup_node_list()
757 td = g->threads + t; in parse_setup_node_list()
773 if (t < g->p.nr_tasks) in parse_setup_node_list()
774 printf("# NOTE: %d tasks mem-bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_node_list()
803 if (g->p.data_reads) in access_data()
805 if (g->p.data_writes) in access_data()
834 if (g->p.data_zero_memset && !g->p.data_rand_walk) { in do_work()
841 chunk_1 = words/g->p.nr_loops; in do_work()
847 if (g->p.data_rand_walk) { in do_work()
859 if (g->p.data_zero_memset) { in do_work()
866 } else if (!g->p.data_backwards || (nr + loop) & 1) { in do_work()
911 g->threads[task_nr].curr_cpu = cpu; in update_curr_cpu()
920 * to a single node. A count of g->p.nr_nodes means it's
929 node_present = (char *)malloc(g->p.nr_nodes * sizeof(char)); in count_process_nodes()
931 for (nodes = 0; nodes < g->p.nr_nodes; nodes++) in count_process_nodes()
934 for (t = 0; t < g->p.nr_threads; t++) { in count_process_nodes()
939 task_nr = process_nr*g->p.nr_threads + t; in count_process_nodes()
940 td = g->threads + task_nr; in count_process_nodes()
953 for (n = 0; n < g->p.nr_nodes; n++) in count_process_nodes()
972 for (p = 0; p < g->p.nr_proc; p++) { in count_node_processes()
973 for (t = 0; t < g->p.nr_threads; t++) { in count_node_processes()
978 task_nr = p*g->p.nr_threads + t; in count_node_processes()
979 td = g->threads + task_nr; in count_node_processes()
1000 for (p = 0; p < g->p.nr_proc; p++) { in calc_convergence_compression()
1036 if (!g->p.show_convergence && !g->p.measure_convergence) in calc_convergence()
1039 nodes = (int *)malloc(g->p.nr_nodes * sizeof(int)); in calc_convergence()
1041 for (node = 0; node < g->p.nr_nodes; node++) in calc_convergence()
1047 for (t = 0; t < g->p.nr_tasks; t++) { in calc_convergence()
1048 struct thread_data *td = g->threads + t; in calc_convergence()
1067 nr_min = g->p.nr_tasks; in calc_convergence()
1070 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
1080 BUG_ON(sum > g->p.nr_tasks); in calc_convergence()
1082 if (0 && (sum < g->p.nr_tasks)) { in calc_convergence()
1090 * to g->p.nr_proc: in calc_convergence()
1094 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
1121 if (strong && process_groups == g->p.nr_proc) { in calc_convergence()
1125 if (g->p.measure_convergence) { in calc_convergence()
1126 g->all_converged = true; in calc_convergence()
1127 g->stop_work = true; in calc_convergence()
1144 (double)(l+1)/g->p.nr_loops*100.0, runtime_ns_max / NSEC_PER_SEC / 60.0); in show_summary()
1148 if (g->p.show_details >= 0) in show_summary()
1160 int details = g->p.show_details; in worker_thread()
1178 global_data = g->data; in worker_thread()
1180 thread_data = setup_private_data(g->p.bytes_thread); in worker_thread()
1185 if (process_nr == g->p.nr_proc-1 && thread_nr == g->p.nr_threads-1) in worker_thread()
1197 if (g->p.serialize_startup) { in worker_thread()
1198 mutex_lock(&g->startup_mutex); in worker_thread()
1199 g->nr_tasks_started++; in worker_thread()
1201 if (g->nr_tasks_started == g->p.nr_tasks) in worker_thread()
1202 cond_signal(&g->startup_cond); in worker_thread()
1204 mutex_unlock(&g->startup_mutex); in worker_thread()
1207 mutex_lock(&g->start_work_mutex); in worker_thread()
1208 g->start_work = false; in worker_thread()
1209 g->nr_tasks_working++; in worker_thread()
1210 while (!g->start_work) in worker_thread()
1211 cond_wait(&g->start_work_cond, &g->start_work_mutex); in worker_thread()
1213 mutex_unlock(&g->start_work_mutex); in worker_thread()
1221 for (l = 0; l < g->p.nr_loops; l++) { in worker_thread()
1224 if (g->stop_work) in worker_thread()
1227 val += do_work(global_data, g->p.bytes_global, process_nr, g->p.nr_proc, l, val); in worker_thread()
1228 val += do_work(process_data, g->p.bytes_process, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1229 val += do_work(thread_data, g->p.bytes_thread, 0, 1, l, val); in worker_thread()
1231 if (g->p.sleep_usecs) { in worker_thread()
1233 usleep(g->p.sleep_usecs); in worker_thread()
1239 if (g->p.bytes_process_locked) { in worker_thread()
1241 val += do_work(process_data, g->p.bytes_process_locked, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1245 work_done = g->p.bytes_global + g->p.bytes_process + in worker_thread()
1246 g->p.bytes_process_locked + g->p.bytes_thread; in worker_thread()
1251 if (details < 0 && !g->p.perturb_secs && !g->p.measure_convergence && !g->p.nr_secs) in worker_thread()
1259 if (g->p.nr_secs) { in worker_thread()
1261 if ((u32)diff.tv_sec >= g->p.nr_secs) { in worker_thread()
1262 g->stop_work = true; in worker_thread()
1272 * Perturb the first task's equilibrium every g->p.perturb_secs seconds, in worker_thread()
1275 …if (first_task && g->p.perturb_secs && (int)(stop.tv_sec - last_perturbance) >= g->p.perturb_secs)… in worker_thread()
1287 this_cpu = g->threads[task_nr].curr_cpu; in worker_thread()
1288 if (this_cpu < g->p.nr_cpus/2) in worker_thread()
1289 target_cpu = g->p.nr_cpus-1; in worker_thread()
1337 free_data(thread_data, g->p.bytes_thread); in worker_thread()
1339 mutex_lock(&g->stop_work_mutex); in worker_thread()
1340 g->bytes_done += bytes_done; in worker_thread()
1341 mutex_unlock(&g->stop_work_mutex); in worker_thread()
1366 task_nr = process_nr*g->p.nr_threads; in worker_process()
1367 td = g->threads + task_nr; in worker_process()
1372 pthreads = zalloc(g->p.nr_threads * sizeof(pthread_t)); in worker_process()
1373 process_data = setup_private_data(g->p.bytes_process); in worker_process()
1375 if (g->p.show_details >= 3) { in worker_process()
1377 process_nr, g->data, process_data); in worker_process()
1380 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1381 task_nr = process_nr*g->p.nr_threads + t; in worker_process()
1382 td = g->threads + task_nr; in worker_process()
1396 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1401 free_data(process_data, g->p.bytes_process); in worker_process()
1407 if (g->p.show_details < 0) in print_summary()
1412 g->p.nr_tasks, g->p.nr_tasks == 1 ? "task" : "tasks", nr_numa_nodes(), g->p.nr_cpus); in print_summary()
1414 g->p.nr_loops, g->p.bytes_global/1024/1024); in print_summary()
1416 g->p.nr_loops, g->p.bytes_process/1024/1024); in print_summary()
1418 g->p.nr_loops, g->p.bytes_thread/1024/1024); in print_summary()
1427 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in init_thread_data()
1430 g->threads = zalloc_shared_data(size); in init_thread_data()
1432 for (t = 0; t < g->p.nr_tasks; t++) { in init_thread_data()
1433 struct thread_data *td = g->threads + t; in init_thread_data()
1434 size_t cpuset_size = CPU_ALLOC_SIZE(g->p.nr_cpus); in init_thread_data()
1441 td->bind_cpumask = CPU_ALLOC(g->p.nr_cpus); in init_thread_data()
1444 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in init_thread_data()
1451 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in deinit_thread_data()
1455 for (t = 0; t < g->p.nr_tasks; t++) { in deinit_thread_data()
1456 struct thread_data *td = g->threads + t; in deinit_thread_data()
1460 free_data(g->threads, size); in deinit_thread_data()
1465 g = (void *)alloc_data(sizeof(*g), MAP_SHARED, 1, 0, 0 /* THP */, 0); in init()
1468 g->p = p0; in init()
1470 g->p.nr_cpus = numa_num_configured_cpus(); in init()
1472 g->p.nr_nodes = numa_max_node() + 1; in init()
1475 BUG_ON(g->p.nr_nodes < 0); in init()
1477 if (g->p.show_quiet && !g->p.show_details) in init()
1478 g->p.show_details = -1; in init()
1481 if (!g->p.mb_global_str && !g->p.mb_proc_str && !g->p.mb_thread_str) in init()
1484 if (g->p.mb_global_str) { in init()
1485 g->p.mb_global = atof(g->p.mb_global_str); in init()
1486 BUG_ON(g->p.mb_global < 0); in init()
1489 if (g->p.mb_proc_str) { in init()
1490 g->p.mb_proc = atof(g->p.mb_proc_str); in init()
1491 BUG_ON(g->p.mb_proc < 0); in init()
1494 if (g->p.mb_proc_locked_str) { in init()
1495 g->p.mb_proc_locked = atof(g->p.mb_proc_locked_str); in init()
1496 BUG_ON(g->p.mb_proc_locked < 0); in init()
1497 BUG_ON(g->p.mb_proc_locked > g->p.mb_proc); in init()
1500 if (g->p.mb_thread_str) { in init()
1501 g->p.mb_thread = atof(g->p.mb_thread_str); in init()
1502 BUG_ON(g->p.mb_thread < 0); in init()
1505 BUG_ON(g->p.nr_threads <= 0); in init()
1506 BUG_ON(g->p.nr_proc <= 0); in init()
1508 g->p.nr_tasks = g->p.nr_proc*g->p.nr_threads; in init()
1510 g->p.bytes_global = g->p.mb_global *1024L*1024L; in init()
1511 g->p.bytes_process = g->p.mb_proc *1024L*1024L; in init()
1512 g->p.bytes_process_locked = g->p.mb_proc_locked *1024L*1024L; in init()
1513 g->p.bytes_thread = g->p.mb_thread *1024L*1024L; in init()
1515 g->data = setup_shared_data(g->p.bytes_global); in init()
1518 mutex_init_pshared(&g->start_work_mutex); in init()
1519 cond_init_pshared(&g->start_work_cond); in init()
1520 mutex_init_pshared(&g->startup_mutex); in init()
1521 cond_init_pshared(&g->startup_cond); in init()
1522 mutex_init_pshared(&g->stop_work_mutex); in init()
1538 free_data(g->data, g->p.bytes_global); in deinit()
1539 g->data = NULL; in deinit()
1543 free_data(g, sizeof(*g)); in deinit()
1544 g = NULL; in deinit()
1556 if (!g->p.show_quiet) in print_res()
1578 pids = zalloc(g->p.nr_proc * sizeof(*pids)); in __bench_numa()
1581 if (g->p.serialize_startup) { in __bench_numa()
1588 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1603 if (g->p.serialize_startup) { in __bench_numa()
1611 mutex_lock(&g->startup_mutex); in __bench_numa()
1612 while (g->nr_tasks_started != g->p.nr_tasks) in __bench_numa()
1613 cond_wait(&g->startup_cond, &g->startup_mutex); in __bench_numa()
1615 mutex_unlock(&g->startup_mutex); in __bench_numa()
1619 mutex_lock(&g->start_work_mutex); in __bench_numa()
1620 threads_ready = (g->nr_tasks_working == g->p.nr_tasks); in __bench_numa()
1621 mutex_unlock(&g->start_work_mutex); in __bench_numa()
1639 mutex_lock(&g->start_work_mutex); in __bench_numa()
1640 g->start_work = true; in __bench_numa()
1641 mutex_unlock(&g->start_work_mutex); in __bench_numa()
1642 cond_broadcast(&g->start_work_cond); in __bench_numa()
1650 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1660 for (t = 0; t < g->p.nr_tasks; t++) { in __bench_numa()
1661 u64 thread_runtime_ns = g->threads[t].runtime_ns; in __bench_numa()
1681 bytes = g->bytes_done; in __bench_numa()
1682 runtime_avg = (double)runtime_ns_sum / g->p.nr_tasks / NSEC_PER_SEC; in __bench_numa()
1684 if (g->p.measure_convergence) { in __bench_numa()
1702 print_res(name, bytes / g->p.nr_tasks / 1e9, in __bench_numa()
1708 print_res(name, runtime_sec_max * NSEC_PER_SEC / (bytes / g->p.nr_tasks), in __bench_numa()
1711 print_res(name, bytes / g->p.nr_tasks / 1e9 / runtime_sec_max, in __bench_numa()
1717 if (g->p.show_details >= 2) { in __bench_numa()
1720 for (p = 0; p < g->p.nr_proc; p++) { in __bench_numa()
1721 for (t = 0; t < g->p.nr_threads; t++) { in __bench_numa()
1723 td = g->threads + p*g->p.nr_threads + t; in __bench_numa()