Lines Matching full:g
50 #define tprintf(x...) do { if (g && g->p.show_details >= 0) printf(x); } while (0)
56 #define dprintf(x...) do { if (g && g->p.show_details >= 1) printf(x); } while (0)
164 static struct global_info *g = NULL; variable
175 OPT_STRING('G', "mb_global" , &p0.mb_global_str, "MB", "global memory (MBs)"),
233 for (i = 0; i < g->p.nr_nodes; i++) { in nr_numa_nodes()
295 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_cpu()
298 if (target_cpu < 0 || target_cpu >= g->p.nr_cpus) in bind_to_cpu()
341 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_node()
389 ret = set_mempolicy(MPOL_DEFAULT, NULL, g->p.nr_nodes-1); in mempol_restore()
452 if (ret && !g->print_once) { in alloc_data()
453 g->print_once = 1; in alloc_data()
459 if (ret && !g->print_once) { in alloc_data()
460 g->print_once = 1; in alloc_data()
509 return alloc_data(bytes, MAP_SHARED, 1, g->p.init_cpu0, g->p.thp, g->p.init_random); in zalloc_shared_data()
517 return alloc_data(bytes, MAP_SHARED, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_shared_data()
526 return alloc_data(bytes, MAP_PRIVATE, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_private_data()
595 if (!g->p.cpu_list_str) in parse_setup_cpu_list()
598 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_cpu_list()
600 str0 = str = strdup(g->p.cpu_list_str); in parse_setup_cpu_list()
635 BUG_ON(step <= 0 || step >= g->p.nr_cpus); in parse_setup_cpu_list()
647 BUG_ON(bind_len <= 0 || bind_len > g->p.nr_cpus); in parse_setup_cpu_list()
660 if (bind_cpu_0 >= g->p.nr_cpus || bind_cpu_1 >= g->p.nr_cpus) { in parse_setup_cpu_list()
661 printf("\nTest not applicable, system has only %d CPUs.\n", g->p.nr_cpus); in parse_setup_cpu_list()
674 size_t size = CPU_ALLOC_SIZE(g->p.nr_cpus); in parse_setup_cpu_list()
680 if (t >= g->p.nr_tasks) { in parse_setup_cpu_list()
684 td = g->threads + t; in parse_setup_cpu_list()
694 td->bind_cpumask = CPU_ALLOC(g->p.nr_cpus); in parse_setup_cpu_list()
698 if (cpu < 0 || cpu >= g->p.nr_cpus) { in parse_setup_cpu_list()
712 if (t < g->p.nr_tasks) in parse_setup_cpu_list()
713 printf("# NOTE: %d tasks bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_cpu_list()
743 if (!g->p.node_list_str) in parse_setup_node_list()
746 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_node_list()
748 str0 = str = strdup(g->p.node_list_str); in parse_setup_node_list()
782 BUG_ON(step <= 0 || step >= g->p.nr_nodes); in parse_setup_node_list()
795 if (bind_node_0 >= g->p.nr_nodes || bind_node_1 >= g->p.nr_nodes) { in parse_setup_node_list()
796 printf("\nTest not applicable, system has only %d nodes.\n", g->p.nr_nodes); in parse_setup_node_list()
807 if (t >= g->p.nr_tasks || !node_has_cpus(bind_node)) { in parse_setup_node_list()
811 td = g->threads + t; in parse_setup_node_list()
827 if (t < g->p.nr_tasks) in parse_setup_node_list()
828 printf("# NOTE: %d tasks mem-bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_node_list()
857 if (g->p.data_reads) in access_data()
859 if (g->p.data_writes) in access_data()
888 if (g->p.data_zero_memset && !g->p.data_rand_walk) { in do_work()
895 chunk_1 = words/g->p.nr_loops; in do_work()
901 if (g->p.data_rand_walk) { in do_work()
913 if (g->p.data_zero_memset) { in do_work()
920 } else if (!g->p.data_backwards || (nr + loop) & 1) { in do_work()
965 g->threads[task_nr].curr_cpu = cpu; in update_curr_cpu()
974 * to a single node. A count of g->p.nr_nodes means it's
983 node_present = (char *)malloc(g->p.nr_nodes * sizeof(char)); in count_process_nodes()
985 for (nodes = 0; nodes < g->p.nr_nodes; nodes++) in count_process_nodes()
988 for (t = 0; t < g->p.nr_threads; t++) { in count_process_nodes()
993 task_nr = process_nr*g->p.nr_threads + t; in count_process_nodes()
994 td = g->threads + task_nr; in count_process_nodes()
1007 for (n = 0; n < g->p.nr_nodes; n++) in count_process_nodes()
1026 for (p = 0; p < g->p.nr_proc; p++) { in count_node_processes()
1027 for (t = 0; t < g->p.nr_threads; t++) { in count_node_processes()
1032 task_nr = p*g->p.nr_threads + t; in count_node_processes()
1033 td = g->threads + task_nr; in count_node_processes()
1054 for (p = 0; p < g->p.nr_proc; p++) { in calc_convergence_compression()
1090 if (!g->p.show_convergence && !g->p.measure_convergence) in calc_convergence()
1093 nodes = (int *)malloc(g->p.nr_nodes * sizeof(int)); in calc_convergence()
1095 for (node = 0; node < g->p.nr_nodes; node++) in calc_convergence()
1101 for (t = 0; t < g->p.nr_tasks; t++) { in calc_convergence()
1102 struct thread_data *td = g->threads + t; in calc_convergence()
1121 nr_min = g->p.nr_tasks; in calc_convergence()
1124 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
1134 BUG_ON(sum > g->p.nr_tasks); in calc_convergence()
1136 if (0 && (sum < g->p.nr_tasks)) { in calc_convergence()
1144 * to g->p.nr_proc: in calc_convergence()
1148 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
1175 if (strong && process_groups == g->p.nr_proc) { in calc_convergence()
1179 if (g->p.measure_convergence) { in calc_convergence()
1180 g->all_converged = true; in calc_convergence()
1181 g->stop_work = true; in calc_convergence()
1198 (double)(l+1)/g->p.nr_loops*100.0, runtime_ns_max / NSEC_PER_SEC / 60.0); in show_summary()
1202 if (g->p.show_details >= 0) in show_summary()
1214 int details = g->p.show_details; in worker_thread()
1232 global_data = g->data; in worker_thread()
1234 thread_data = setup_private_data(g->p.bytes_thread); in worker_thread()
1239 if (process_nr == g->p.nr_proc-1 && thread_nr == g->p.nr_threads-1) in worker_thread()
1251 if (g->p.serialize_startup) { in worker_thread()
1252 mutex_lock(&g->startup_mutex); in worker_thread()
1253 g->nr_tasks_started++; in worker_thread()
1255 if (g->nr_tasks_started == g->p.nr_tasks) in worker_thread()
1256 cond_signal(&g->startup_cond); in worker_thread()
1258 mutex_unlock(&g->startup_mutex); in worker_thread()
1261 mutex_lock(&g->start_work_mutex); in worker_thread()
1262 g->start_work = false; in worker_thread()
1263 g->nr_tasks_working++; in worker_thread()
1264 while (!g->start_work) in worker_thread()
1265 cond_wait(&g->start_work_cond, &g->start_work_mutex); in worker_thread()
1267 mutex_unlock(&g->start_work_mutex); in worker_thread()
1275 for (l = 0; l < g->p.nr_loops; l++) { in worker_thread()
1278 if (g->stop_work) in worker_thread()
1281 val += do_work(global_data, g->p.bytes_global, process_nr, g->p.nr_proc, l, val); in worker_thread()
1282 val += do_work(process_data, g->p.bytes_process, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1283 val += do_work(thread_data, g->p.bytes_thread, 0, 1, l, val); in worker_thread()
1285 if (g->p.sleep_usecs) { in worker_thread()
1287 usleep(g->p.sleep_usecs); in worker_thread()
1293 if (g->p.bytes_process_locked) { in worker_thread()
1295 val += do_work(process_data, g->p.bytes_process_locked, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1299 work_done = g->p.bytes_global + g->p.bytes_process + in worker_thread()
1300 g->p.bytes_process_locked + g->p.bytes_thread; in worker_thread()
1305 if (details < 0 && !g->p.perturb_secs && !g->p.measure_convergence && !g->p.nr_secs) in worker_thread()
1313 if (g->p.nr_secs) { in worker_thread()
1315 if ((u32)diff.tv_sec >= g->p.nr_secs) { in worker_thread()
1316 g->stop_work = true; in worker_thread()
1326 * Perturb the first task's equilibrium every g->p.perturb_secs seconds, in worker_thread()
1329 …if (first_task && g->p.perturb_secs && (int)(stop.tv_sec - last_perturbance) >= g->p.perturb_secs)… in worker_thread()
1341 this_cpu = g->threads[task_nr].curr_cpu; in worker_thread()
1342 if (this_cpu < g->p.nr_cpus/2) in worker_thread()
1343 target_cpu = g->p.nr_cpus-1; in worker_thread()
1391 free_data(thread_data, g->p.bytes_thread); in worker_thread()
1393 mutex_lock(&g->stop_work_mutex); in worker_thread()
1394 g->bytes_done += bytes_done; in worker_thread()
1395 mutex_unlock(&g->stop_work_mutex); in worker_thread()
1420 task_nr = process_nr*g->p.nr_threads; in worker_process()
1421 td = g->threads + task_nr; in worker_process()
1426 pthreads = zalloc(g->p.nr_threads * sizeof(pthread_t)); in worker_process()
1427 process_data = setup_private_data(g->p.bytes_process); in worker_process()
1429 if (g->p.show_details >= 3) { in worker_process()
1431 process_nr, g->data, process_data); in worker_process()
1434 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1435 task_nr = process_nr*g->p.nr_threads + t; in worker_process()
1436 td = g->threads + task_nr; in worker_process()
1450 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1455 free_data(process_data, g->p.bytes_process); in worker_process()
1461 if (g->p.show_details < 0) in print_summary()
1466 g->p.nr_tasks, g->p.nr_tasks == 1 ? "task" : "tasks", nr_numa_nodes(), g->p.nr_cpus); in print_summary()
1468 g->p.nr_loops, g->p.bytes_global/1024/1024); in print_summary()
1470 g->p.nr_loops, g->p.bytes_process/1024/1024); in print_summary()
1472 g->p.nr_loops, g->p.bytes_thread/1024/1024); in print_summary()
1481 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in init_thread_data()
1484 g->threads = zalloc_shared_data(size); in init_thread_data()
1486 for (t = 0; t < g->p.nr_tasks; t++) { in init_thread_data()
1487 struct thread_data *td = g->threads + t; in init_thread_data()
1488 size_t cpuset_size = CPU_ALLOC_SIZE(g->p.nr_cpus); in init_thread_data()
1495 td->bind_cpumask = CPU_ALLOC(g->p.nr_cpus); in init_thread_data()
1498 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in init_thread_data()
1505 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in deinit_thread_data()
1509 for (t = 0; t < g->p.nr_tasks; t++) { in deinit_thread_data()
1510 struct thread_data *td = g->threads + t; in deinit_thread_data()
1514 free_data(g->threads, size); in deinit_thread_data()
1519 g = (void *)alloc_data(sizeof(*g), MAP_SHARED, 1, 0, 0 /* THP */, 0); in init()
1522 g->p = p0; in init()
1524 g->p.nr_cpus = numa_num_configured_cpus(); in init()
1526 g->p.nr_nodes = numa_max_node() + 1; in init()
1529 BUG_ON(g->p.nr_nodes < 0); in init()
1531 if (quiet && !g->p.show_details) in init()
1532 g->p.show_details = -1; in init()
1535 if (!g->p.mb_global_str && !g->p.mb_proc_str && !g->p.mb_thread_str) in init()
1538 if (g->p.mb_global_str) { in init()
1539 g->p.mb_global = atof(g->p.mb_global_str); in init()
1540 BUG_ON(g->p.mb_global < 0); in init()
1543 if (g->p.mb_proc_str) { in init()
1544 g->p.mb_proc = atof(g->p.mb_proc_str); in init()
1545 BUG_ON(g->p.mb_proc < 0); in init()
1548 if (g->p.mb_proc_locked_str) { in init()
1549 g->p.mb_proc_locked = atof(g->p.mb_proc_locked_str); in init()
1550 BUG_ON(g->p.mb_proc_locked < 0); in init()
1551 BUG_ON(g->p.mb_proc_locked > g->p.mb_proc); in init()
1554 if (g->p.mb_thread_str) { in init()
1555 g->p.mb_thread = atof(g->p.mb_thread_str); in init()
1556 BUG_ON(g->p.mb_thread < 0); in init()
1559 BUG_ON(g->p.nr_threads <= 0); in init()
1560 BUG_ON(g->p.nr_proc <= 0); in init()
1562 g->p.nr_tasks = g->p.nr_proc*g->p.nr_threads; in init()
1564 g->p.bytes_global = g->p.mb_global *1024L*1024L; in init()
1565 g->p.bytes_process = g->p.mb_proc *1024L*1024L; in init()
1566 g->p.bytes_process_locked = g->p.mb_proc_locked *1024L*1024L; in init()
1567 g->p.bytes_thread = g->p.mb_thread *1024L*1024L; in init()
1569 g->data = setup_shared_data(g->p.bytes_global); in init()
1572 mutex_init_pshared(&g->start_work_mutex); in init()
1573 cond_init_pshared(&g->start_work_cond); in init()
1574 mutex_init_pshared(&g->startup_mutex); in init()
1575 cond_init_pshared(&g->startup_cond); in init()
1576 mutex_init_pshared(&g->stop_work_mutex); in init()
1592 free_data(g->data, g->p.bytes_global); in deinit()
1593 g->data = NULL; in deinit()
1597 free_data(g, sizeof(*g)); in deinit()
1598 g = NULL; in deinit()
1632 pids = zalloc(g->p.nr_proc * sizeof(*pids)); in __bench_numa()
1635 if (g->p.serialize_startup) { in __bench_numa()
1642 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1657 if (g->p.serialize_startup) { in __bench_numa()
1665 mutex_lock(&g->startup_mutex); in __bench_numa()
1666 while (g->nr_tasks_started != g->p.nr_tasks) in __bench_numa()
1667 cond_wait(&g->startup_cond, &g->startup_mutex); in __bench_numa()
1669 mutex_unlock(&g->startup_mutex); in __bench_numa()
1673 mutex_lock(&g->start_work_mutex); in __bench_numa()
1674 threads_ready = (g->nr_tasks_working == g->p.nr_tasks); in __bench_numa()
1675 mutex_unlock(&g->start_work_mutex); in __bench_numa()
1693 mutex_lock(&g->start_work_mutex); in __bench_numa()
1694 g->start_work = true; in __bench_numa()
1695 mutex_unlock(&g->start_work_mutex); in __bench_numa()
1696 cond_broadcast(&g->start_work_cond); in __bench_numa()
1704 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1714 for (t = 0; t < g->p.nr_tasks; t++) { in __bench_numa()
1715 u64 thread_runtime_ns = g->threads[t].runtime_ns; in __bench_numa()
1735 bytes = g->bytes_done; in __bench_numa()
1736 runtime_avg = (double)runtime_ns_sum / g->p.nr_tasks / NSEC_PER_SEC; in __bench_numa()
1738 if (g->p.measure_convergence) { in __bench_numa()
1756 print_res(name, bytes / g->p.nr_tasks / 1e9, in __bench_numa()
1762 print_res(name, runtime_sec_max * NSEC_PER_SEC / (bytes / g->p.nr_tasks), in __bench_numa()
1765 print_res(name, bytes / g->p.nr_tasks / 1e9 / runtime_sec_max, in __bench_numa()
1771 if (g->p.show_details >= 2) { in __bench_numa()
1774 for (p = 0; p < g->p.nr_proc; p++) { in __bench_numa()
1775 for (t = 0; t < g->p.nr_threads; t++) { in __bench_numa()
1777 td = g->threads + p*g->p.nr_threads + t; in __bench_numa()