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)
161 static struct global_info *g = NULL; variable
172 OPT_STRING('G', "mb_global" , &p0.mb_global_str, "MB", "global memory (MBs)"),
229 for (i = 0; i < g->p.nr_nodes; i++) { in nr_numa_nodes()
276 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_cpu()
279 BUG_ON(target_cpu < 0 || target_cpu >= g->p.nr_cpus); in bind_to_cpu()
291 int cpus_per_node = g->p.nr_cpus / nr_numa_nodes(); in bind_to_node()
296 BUG_ON(cpus_per_node * nr_numa_nodes() != g->p.nr_cpus); in bind_to_node()
305 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_node()
311 BUG_ON(cpu_stop > g->p.nr_cpus); in bind_to_node()
335 ret = set_mempolicy(MPOL_DEFAULT, NULL, g->p.nr_nodes-1); in mempol_restore()
348 BUG_ON(g->p.nr_nodes > (int)sizeof(nodemask)*8); in bind_to_memnode()
394 if (ret && !g->print_once) { in alloc_data()
395 g->print_once = 1; in alloc_data()
401 if (ret && !g->print_once) { in alloc_data()
402 g->print_once = 1; in alloc_data()
450 return alloc_data(bytes, MAP_SHARED, 1, g->p.init_cpu0, g->p.thp, g->p.init_random); in zalloc_shared_data()
458 return alloc_data(bytes, MAP_SHARED, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_shared_data()
467 return alloc_data(bytes, MAP_PRIVATE, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_private_data()
497 if (!g->p.cpu_list_str) in parse_setup_cpu_list()
500 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_cpu_list()
502 str0 = str = strdup(g->p.cpu_list_str); in parse_setup_cpu_list()
537 BUG_ON(step <= 0 || step >= g->p.nr_cpus); in parse_setup_cpu_list()
549 BUG_ON(bind_len <= 0 || bind_len > g->p.nr_cpus); in parse_setup_cpu_list()
562 if (bind_cpu_0 >= g->p.nr_cpus || bind_cpu_1 >= g->p.nr_cpus) { in parse_setup_cpu_list()
563 printf("\nTest not applicable, system has only %d CPUs.\n", g->p.nr_cpus); in parse_setup_cpu_list()
576 if (t >= g->p.nr_tasks) { in parse_setup_cpu_list()
580 td = g->threads + t; in parse_setup_cpu_list()
592 BUG_ON(cpu < 0 || cpu >= g->p.nr_cpus); in parse_setup_cpu_list()
603 if (t < g->p.nr_tasks) in parse_setup_cpu_list()
604 printf("# NOTE: %d tasks bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_cpu_list()
634 if (!g->p.node_list_str) in parse_setup_node_list()
637 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_node_list()
639 str0 = str = strdup(g->p.node_list_str); in parse_setup_node_list()
673 BUG_ON(step <= 0 || step >= g->p.nr_nodes); in parse_setup_node_list()
686 if (bind_node_0 >= g->p.nr_nodes || bind_node_1 >= g->p.nr_nodes) { in parse_setup_node_list()
687 printf("\nTest not applicable, system has only %d nodes.\n", g->p.nr_nodes); in parse_setup_node_list()
698 if (t >= g->p.nr_tasks || !node_has_cpus(bind_node)) { in parse_setup_node_list()
702 td = g->threads + t; in parse_setup_node_list()
718 if (t < g->p.nr_tasks) in parse_setup_node_list()
719 printf("# NOTE: %d tasks mem-bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_node_list()
752 if (g->p.data_reads) in access_data()
754 if (g->p.data_writes) in access_data()
783 if (g->p.data_zero_memset && !g->p.data_rand_walk) { in do_work()
790 chunk_1 = words/g->p.nr_loops; in do_work()
796 if (g->p.data_rand_walk) { in do_work()
808 if (g->p.data_zero_memset) { in do_work()
815 } else if (!g->p.data_backwards || (nr + loop) & 1) { in do_work()
861 g->threads[task_nr].curr_cpu = cpu; in update_curr_cpu()
872 * to a single node. A count of g->p.nr_nodes means it's
881 for (t = 0; t < g->p.nr_threads; t++) { in count_process_nodes()
886 task_nr = process_nr*g->p.nr_threads + t; in count_process_nodes()
887 td = g->threads + task_nr; in count_process_nodes()
916 for (p = 0; p < g->p.nr_proc; p++) { in count_node_processes()
917 for (t = 0; t < g->p.nr_threads; t++) { in count_node_processes()
922 task_nr = p*g->p.nr_threads + t; in count_node_processes()
923 td = g->threads + task_nr; in count_node_processes()
944 for (p = 0; p < g->p.nr_proc; p++) { in calc_convergence_compression()
980 if (!g->p.show_convergence && !g->p.measure_convergence) in calc_convergence()
983 for (node = 0; node < g->p.nr_nodes; node++) in calc_convergence()
989 for (t = 0; t < g->p.nr_tasks; t++) { in calc_convergence()
990 struct thread_data *td = g->threads + t; in calc_convergence()
1009 nr_min = g->p.nr_tasks; in calc_convergence()
1012 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
1022 BUG_ON(sum > g->p.nr_tasks); in calc_convergence()
1024 if (0 && (sum < g->p.nr_tasks)) in calc_convergence()
1030 * to g->p.nr_proc: in calc_convergence()
1034 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
1061 if (strong && process_groups == g->p.nr_proc) { in calc_convergence()
1065 if (g->p.measure_convergence) { in calc_convergence()
1066 g->all_converged = true; in calc_convergence()
1067 g->stop_work = true; in calc_convergence()
1082 (double)(l+1)/g->p.nr_loops*100.0, runtime_ns_max / NSEC_PER_SEC / 60.0); in show_summary()
1086 if (g->p.show_details >= 0) in show_summary()
1098 int details = g->p.show_details; in worker_thread()
1116 global_data = g->data; in worker_thread()
1118 thread_data = setup_private_data(g->p.bytes_thread); in worker_thread()
1123 if (process_nr == g->p.nr_proc-1 && thread_nr == g->p.nr_threads-1) in worker_thread()
1135 if (g->p.serialize_startup) { in worker_thread()
1136 pthread_mutex_lock(&g->startup_mutex); in worker_thread()
1137 g->nr_tasks_started++; in worker_thread()
1138 pthread_mutex_unlock(&g->startup_mutex); in worker_thread()
1141 pthread_mutex_lock(&g->start_work_mutex); in worker_thread()
1142 g->nr_tasks_working++; in worker_thread()
1145 if (g->nr_tasks_working == g->p.nr_tasks) in worker_thread()
1146 pthread_mutex_unlock(&g->startup_done_mutex); in worker_thread()
1148 pthread_mutex_unlock(&g->start_work_mutex); in worker_thread()
1156 for (l = 0; l < g->p.nr_loops; l++) { in worker_thread()
1159 if (g->stop_work) in worker_thread()
1162 val += do_work(global_data, g->p.bytes_global, process_nr, g->p.nr_proc, l, val); in worker_thread()
1163 val += do_work(process_data, g->p.bytes_process, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1164 val += do_work(thread_data, g->p.bytes_thread, 0, 1, l, val); in worker_thread()
1166 if (g->p.sleep_usecs) { in worker_thread()
1168 usleep(g->p.sleep_usecs); in worker_thread()
1174 if (g->p.bytes_process_locked) { in worker_thread()
1176 val += do_work(process_data, g->p.bytes_process_locked, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1180 work_done = g->p.bytes_global + g->p.bytes_process + in worker_thread()
1181 g->p.bytes_process_locked + g->p.bytes_thread; in worker_thread()
1186 if (details < 0 && !g->p.perturb_secs && !g->p.measure_convergence && !g->p.nr_secs) in worker_thread()
1194 if (g->p.nr_secs) { in worker_thread()
1196 if ((u32)diff.tv_sec >= g->p.nr_secs) { in worker_thread()
1197 g->stop_work = true; in worker_thread()
1207 * Perturb the first task's equilibrium every g->p.perturb_secs seconds, in worker_thread()
1210 …if (first_task && g->p.perturb_secs && (int)(stop.tv_sec - last_perturbance) >= g->p.perturb_secs)… in worker_thread()
1222 this_cpu = g->threads[task_nr].curr_cpu; in worker_thread()
1223 if (this_cpu < g->p.nr_cpus/2) in worker_thread()
1224 target_cpu = g->p.nr_cpus-1; in worker_thread()
1271 free_data(thread_data, g->p.bytes_thread); in worker_thread()
1273 pthread_mutex_lock(&g->stop_work_mutex); in worker_thread()
1274 g->bytes_done += bytes_done; in worker_thread()
1275 pthread_mutex_unlock(&g->stop_work_mutex); in worker_thread()
1300 task_nr = process_nr*g->p.nr_threads; in worker_process()
1301 td = g->threads + task_nr; in worker_process()
1306 pthreads = zalloc(g->p.nr_threads * sizeof(pthread_t)); in worker_process()
1307 process_data = setup_private_data(g->p.bytes_process); in worker_process()
1309 if (g->p.show_details >= 3) { in worker_process()
1311 process_nr, g->data, process_data); in worker_process()
1314 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1315 task_nr = process_nr*g->p.nr_threads + t; in worker_process()
1316 td = g->threads + task_nr; in worker_process()
1330 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1335 free_data(process_data, g->p.bytes_process); in worker_process()
1341 if (g->p.show_details < 0) in print_summary()
1346 g->p.nr_tasks, g->p.nr_tasks == 1 ? "task" : "tasks", nr_numa_nodes(), g->p.nr_cpus); in print_summary()
1348 g->p.nr_loops, g->p.bytes_global/1024/1024); in print_summary()
1350 g->p.nr_loops, g->p.bytes_process/1024/1024); in print_summary()
1352 g->p.nr_loops, g->p.bytes_thread/1024/1024); in print_summary()
1361 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in init_thread_data()
1364 g->threads = zalloc_shared_data(size); in init_thread_data()
1366 for (t = 0; t < g->p.nr_tasks; t++) { in init_thread_data()
1367 struct thread_data *td = g->threads + t; in init_thread_data()
1375 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in init_thread_data()
1382 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in deinit_thread_data()
1384 free_data(g->threads, size); in deinit_thread_data()
1389 g = (void *)alloc_data(sizeof(*g), MAP_SHARED, 1, 0, 0 /* THP */, 0); in init()
1392 g->p = p0; in init()
1394 g->p.nr_cpus = numa_num_configured_cpus(); in init()
1396 g->p.nr_nodes = numa_max_node() + 1; in init()
1399 BUG_ON(g->p.nr_nodes > MAX_NR_NODES || g->p.nr_nodes < 0); in init()
1401 if (g->p.show_quiet && !g->p.show_details) in init()
1402 g->p.show_details = -1; in init()
1405 if (!g->p.mb_global_str && !g->p.mb_proc_str && !g->p.mb_thread_str) in init()
1408 if (g->p.mb_global_str) { in init()
1409 g->p.mb_global = atof(g->p.mb_global_str); in init()
1410 BUG_ON(g->p.mb_global < 0); in init()
1413 if (g->p.mb_proc_str) { in init()
1414 g->p.mb_proc = atof(g->p.mb_proc_str); in init()
1415 BUG_ON(g->p.mb_proc < 0); in init()
1418 if (g->p.mb_proc_locked_str) { in init()
1419 g->p.mb_proc_locked = atof(g->p.mb_proc_locked_str); in init()
1420 BUG_ON(g->p.mb_proc_locked < 0); in init()
1421 BUG_ON(g->p.mb_proc_locked > g->p.mb_proc); in init()
1424 if (g->p.mb_thread_str) { in init()
1425 g->p.mb_thread = atof(g->p.mb_thread_str); in init()
1426 BUG_ON(g->p.mb_thread < 0); in init()
1429 BUG_ON(g->p.nr_threads <= 0); in init()
1430 BUG_ON(g->p.nr_proc <= 0); in init()
1432 g->p.nr_tasks = g->p.nr_proc*g->p.nr_threads; in init()
1434 g->p.bytes_global = g->p.mb_global *1024L*1024L; in init()
1435 g->p.bytes_process = g->p.mb_proc *1024L*1024L; in init()
1436 g->p.bytes_process_locked = g->p.mb_proc_locked *1024L*1024L; in init()
1437 g->p.bytes_thread = g->p.mb_thread *1024L*1024L; in init()
1439 g->data = setup_shared_data(g->p.bytes_global); in init()
1442 init_global_mutex(&g->start_work_mutex); in init()
1443 init_global_mutex(&g->startup_mutex); in init()
1444 init_global_mutex(&g->startup_done_mutex); in init()
1445 init_global_mutex(&g->stop_work_mutex); in init()
1461 free_data(g->data, g->p.bytes_global); in deinit()
1462 g->data = NULL; in deinit()
1466 free_data(g, sizeof(*g)); in deinit()
1467 g = NULL; in deinit()
1479 if (!g->p.show_quiet) in print_res()
1501 pids = zalloc(g->p.nr_proc * sizeof(*pids)); in __bench_numa()
1505 pthread_mutex_lock(&g->start_work_mutex); in __bench_numa()
1507 if (g->p.serialize_startup) { in __bench_numa()
1514 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1529 while (g->nr_tasks_started != g->p.nr_tasks) in __bench_numa()
1532 BUG_ON(g->nr_tasks_started != g->p.nr_tasks); in __bench_numa()
1534 if (g->p.serialize_startup) { in __bench_numa()
1537 pthread_mutex_lock(&g->startup_done_mutex); in __bench_numa()
1540 pthread_mutex_unlock(&g->start_work_mutex); in __bench_numa()
1543 pthread_mutex_lock(&g->startup_done_mutex); in __bench_numa()
1557 pthread_mutex_unlock(&g->startup_done_mutex); in __bench_numa()
1565 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1575 for (t = 0; t < g->p.nr_tasks; t++) { in __bench_numa()
1576 u64 thread_runtime_ns = g->threads[t].runtime_ns; in __bench_numa()
1596 bytes = g->bytes_done; in __bench_numa()
1597 runtime_avg = (double)runtime_ns_sum / g->p.nr_tasks / NSEC_PER_SEC; in __bench_numa()
1599 if (g->p.measure_convergence) { in __bench_numa()
1617 print_res(name, bytes / g->p.nr_tasks / 1e9, in __bench_numa()
1623 print_res(name, runtime_sec_max * NSEC_PER_SEC / (bytes / g->p.nr_tasks), in __bench_numa()
1626 print_res(name, bytes / g->p.nr_tasks / 1e9 / runtime_sec_max, in __bench_numa()
1632 if (g->p.show_details >= 2) { in __bench_numa()
1635 for (p = 0; p < g->p.nr_proc; p++) { in __bench_numa()
1636 for (t = 0; t < g->p.nr_threads; t++) { in __bench_numa()
1638 td = g->threads + p*g->p.nr_threads + t; in __bench_numa()