Lines Matching refs:g
36 #define tprintf(x...) do { if (g && g->p.show_details >= 0) printf(x); } while (0)
41 #define dprintf(x...) do { if (g && g->p.show_details >= 1) printf(x); } while (0)
146 static struct global_info *g = NULL; variable
219 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_cpu()
222 BUG_ON(target_cpu < 0 || target_cpu >= g->p.nr_cpus); in bind_to_cpu()
234 int cpus_per_node = g->p.nr_cpus/g->p.nr_nodes; in bind_to_node()
239 BUG_ON(cpus_per_node*g->p.nr_nodes != g->p.nr_cpus); in bind_to_node()
248 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in bind_to_node()
254 BUG_ON(cpu_stop > g->p.nr_cpus); in bind_to_node()
278 ret = set_mempolicy(MPOL_DEFAULT, NULL, g->p.nr_nodes-1); in mempol_restore()
291 BUG_ON(g->p.nr_nodes > (int)sizeof(nodemask)); in bind_to_memnode()
335 if (ret && !g->print_once) { in alloc_data()
336 g->print_once = 1; in alloc_data()
342 if (ret && !g->print_once) { in alloc_data()
343 g->print_once = 1; in alloc_data()
391 return alloc_data(bytes, MAP_SHARED, 1, g->p.init_cpu0, g->p.thp, g->p.init_random); in zalloc_shared_data()
399 return alloc_data(bytes, MAP_SHARED, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_shared_data()
408 return alloc_data(bytes, MAP_PRIVATE, 0, g->p.init_cpu0, g->p.thp, g->p.init_random); in setup_private_data()
438 if (!g->p.cpu_list_str) in parse_setup_cpu_list()
441 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_cpu_list()
443 str0 = str = strdup(g->p.cpu_list_str); in parse_setup_cpu_list()
478 BUG_ON(step <= 0 || step >= g->p.nr_cpus); in parse_setup_cpu_list()
490 BUG_ON(bind_len <= 0 || bind_len > g->p.nr_cpus); in parse_setup_cpu_list()
503 if (bind_cpu_0 >= g->p.nr_cpus || bind_cpu_1 >= g->p.nr_cpus) { in parse_setup_cpu_list()
504 printf("\nTest not applicable, system has only %d CPUs.\n", g->p.nr_cpus); in parse_setup_cpu_list()
517 if (t >= g->p.nr_tasks) { in parse_setup_cpu_list()
521 td = g->threads + t; in parse_setup_cpu_list()
533 BUG_ON(cpu < 0 || cpu >= g->p.nr_cpus); in parse_setup_cpu_list()
544 if (t < g->p.nr_tasks) in parse_setup_cpu_list()
545 printf("# NOTE: %d tasks bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_cpu_list()
575 if (!g->p.node_list_str) in parse_setup_node_list()
578 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks); in parse_setup_node_list()
580 str0 = str = strdup(g->p.node_list_str); in parse_setup_node_list()
614 BUG_ON(step <= 0 || step >= g->p.nr_nodes); in parse_setup_node_list()
627 if (bind_node_0 >= g->p.nr_nodes || bind_node_1 >= g->p.nr_nodes) { in parse_setup_node_list()
628 printf("\nTest not applicable, system has only %d nodes.\n", g->p.nr_nodes); in parse_setup_node_list()
639 if (t >= g->p.nr_tasks) { in parse_setup_node_list()
643 td = g->threads + t; in parse_setup_node_list()
659 if (t < g->p.nr_tasks) in parse_setup_node_list()
660 printf("# NOTE: %d tasks mem-bound, %d tasks unbound\n", t, g->p.nr_tasks - t); in parse_setup_node_list()
693 if (g->p.data_reads) in access_data()
695 if (g->p.data_writes) in access_data()
724 if (g->p.data_zero_memset && !g->p.data_rand_walk) { in do_work()
731 chunk_1 = words/g->p.nr_loops; in do_work()
737 if (g->p.data_rand_walk) { in do_work()
749 if (g->p.data_zero_memset) { in do_work()
756 } else if (!g->p.data_backwards || (nr + loop) & 1) { in do_work()
802 g->threads[task_nr].curr_cpu = cpu; in update_curr_cpu()
822 for (t = 0; t < g->p.nr_threads; t++) { in count_process_nodes()
827 task_nr = process_nr*g->p.nr_threads + t; in count_process_nodes()
828 td = g->threads + task_nr; in count_process_nodes()
857 for (p = 0; p < g->p.nr_proc; p++) { in count_node_processes()
858 for (t = 0; t < g->p.nr_threads; t++) { in count_node_processes()
863 task_nr = p*g->p.nr_threads + t; in count_node_processes()
864 td = g->threads + task_nr; in count_node_processes()
885 for (p = 0; p < g->p.nr_proc; p++) { in calc_convergence_compression()
921 if (!g->p.show_convergence && !g->p.measure_convergence) in calc_convergence()
924 for (node = 0; node < g->p.nr_nodes; node++) in calc_convergence()
930 for (t = 0; t < g->p.nr_tasks; t++) { in calc_convergence()
931 struct thread_data *td = g->threads + t; in calc_convergence()
950 nr_min = g->p.nr_tasks; in calc_convergence()
953 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
961 BUG_ON(sum > g->p.nr_tasks); in calc_convergence()
963 if (0 && (sum < g->p.nr_tasks)) in calc_convergence()
973 for (node = 0; node < g->p.nr_nodes; node++) { in calc_convergence()
997 if (strong && process_groups == g->p.nr_proc) { in calc_convergence()
1001 if (g->p.measure_convergence) { in calc_convergence()
1002 g->all_converged = true; in calc_convergence()
1003 g->stop_work = true; in calc_convergence()
1018 (double)(l+1)/g->p.nr_loops*100.0, runtime_ns_max/1e9 / 60.0); in show_summary()
1022 if (g->p.show_details >= 0) in show_summary()
1034 int details = g->p.show_details; in worker_thread()
1051 global_data = g->data; in worker_thread()
1053 thread_data = setup_private_data(g->p.bytes_thread); in worker_thread()
1058 if (process_nr == g->p.nr_proc-1 && thread_nr == g->p.nr_threads-1) in worker_thread()
1070 if (g->p.serialize_startup) { in worker_thread()
1071 pthread_mutex_lock(&g->startup_mutex); in worker_thread()
1072 g->nr_tasks_started++; in worker_thread()
1073 pthread_mutex_unlock(&g->startup_mutex); in worker_thread()
1076 pthread_mutex_lock(&g->start_work_mutex); in worker_thread()
1077 g->nr_tasks_working++; in worker_thread()
1080 if (g->nr_tasks_working == g->p.nr_tasks) in worker_thread()
1081 pthread_mutex_unlock(&g->startup_done_mutex); in worker_thread()
1083 pthread_mutex_unlock(&g->start_work_mutex); in worker_thread()
1091 for (l = 0; l < g->p.nr_loops; l++) { in worker_thread()
1094 if (g->stop_work) in worker_thread()
1097 val += do_work(global_data, g->p.bytes_global, process_nr, g->p.nr_proc, l, val); in worker_thread()
1098 val += do_work(process_data, g->p.bytes_process, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1099 val += do_work(thread_data, g->p.bytes_thread, 0, 1, l, val); in worker_thread()
1101 if (g->p.sleep_usecs) { in worker_thread()
1103 usleep(g->p.sleep_usecs); in worker_thread()
1109 if (g->p.bytes_process_locked) { in worker_thread()
1111 val += do_work(process_data, g->p.bytes_process_locked, thread_nr, g->p.nr_threads, l, val); in worker_thread()
1115 work_done = g->p.bytes_global + g->p.bytes_process + in worker_thread()
1116 g->p.bytes_process_locked + g->p.bytes_thread; in worker_thread()
1121 if (details < 0 && !g->p.perturb_secs && !g->p.measure_convergence && !g->p.nr_secs) in worker_thread()
1129 if (g->p.nr_secs) { in worker_thread()
1131 if ((u32)diff.tv_sec >= g->p.nr_secs) { in worker_thread()
1132 g->stop_work = true; in worker_thread()
1145 …if (first_task && g->p.perturb_secs && (int)(stop.tv_sec - last_perturbance) >= g->p.perturb_secs)… in worker_thread()
1157 this_cpu = g->threads[task_nr].curr_cpu; in worker_thread()
1158 if (this_cpu < g->p.nr_cpus/2) in worker_thread()
1159 target_cpu = g->p.nr_cpus-1; in worker_thread()
1198 free_data(thread_data, g->p.bytes_thread); in worker_thread()
1200 pthread_mutex_lock(&g->stop_work_mutex); in worker_thread()
1201 g->bytes_done += bytes_done; in worker_thread()
1202 pthread_mutex_unlock(&g->stop_work_mutex); in worker_thread()
1227 task_nr = process_nr*g->p.nr_threads; in worker_process()
1228 td = g->threads + task_nr; in worker_process()
1233 pthreads = zalloc(g->p.nr_threads * sizeof(pthread_t)); in worker_process()
1234 process_data = setup_private_data(g->p.bytes_process); in worker_process()
1236 if (g->p.show_details >= 3) { in worker_process()
1238 process_nr, g->data, process_data); in worker_process()
1241 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1242 task_nr = process_nr*g->p.nr_threads + t; in worker_process()
1243 td = g->threads + task_nr; in worker_process()
1257 for (t = 0; t < g->p.nr_threads; t++) { in worker_process()
1262 free_data(process_data, g->p.bytes_process); in worker_process()
1268 if (g->p.show_details < 0) in print_summary()
1273 g->p.nr_tasks, g->p.nr_tasks == 1 ? "task" : "tasks", g->p.nr_nodes, g->p.nr_cpus); in print_summary()
1275 g->p.nr_loops, g->p.bytes_global/1024/1024); in print_summary()
1277 g->p.nr_loops, g->p.bytes_process/1024/1024); in print_summary()
1279 g->p.nr_loops, g->p.bytes_thread/1024/1024); in print_summary()
1288 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in init_thread_data()
1291 g->threads = zalloc_shared_data(size); in init_thread_data()
1293 for (t = 0; t < g->p.nr_tasks; t++) { in init_thread_data()
1294 struct thread_data *td = g->threads + t; in init_thread_data()
1302 for (cpu = 0; cpu < g->p.nr_cpus; cpu++) in init_thread_data()
1309 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks; in deinit_thread_data()
1311 free_data(g->threads, size); in deinit_thread_data()
1316 g = (void *)alloc_data(sizeof(*g), MAP_SHARED, 1, 0, 0 /* THP */, 0); in init()
1319 g->p = p0; in init()
1321 g->p.nr_cpus = numa_num_configured_cpus(); in init()
1323 g->p.nr_nodes = numa_max_node() + 1; in init()
1326 BUG_ON(g->p.nr_nodes > MAX_NR_NODES || g->p.nr_nodes < 0); in init()
1328 if (g->p.show_quiet && !g->p.show_details) in init()
1329 g->p.show_details = -1; in init()
1332 if (!g->p.mb_global_str && !g->p.mb_proc_str && !g->p.mb_thread_str) in init()
1335 if (g->p.mb_global_str) { in init()
1336 g->p.mb_global = atof(g->p.mb_global_str); in init()
1337 BUG_ON(g->p.mb_global < 0); in init()
1340 if (g->p.mb_proc_str) { in init()
1341 g->p.mb_proc = atof(g->p.mb_proc_str); in init()
1342 BUG_ON(g->p.mb_proc < 0); in init()
1345 if (g->p.mb_proc_locked_str) { in init()
1346 g->p.mb_proc_locked = atof(g->p.mb_proc_locked_str); in init()
1347 BUG_ON(g->p.mb_proc_locked < 0); in init()
1348 BUG_ON(g->p.mb_proc_locked > g->p.mb_proc); in init()
1351 if (g->p.mb_thread_str) { in init()
1352 g->p.mb_thread = atof(g->p.mb_thread_str); in init()
1353 BUG_ON(g->p.mb_thread < 0); in init()
1356 BUG_ON(g->p.nr_threads <= 0); in init()
1357 BUG_ON(g->p.nr_proc <= 0); in init()
1359 g->p.nr_tasks = g->p.nr_proc*g->p.nr_threads; in init()
1361 g->p.bytes_global = g->p.mb_global *1024L*1024L; in init()
1362 g->p.bytes_process = g->p.mb_proc *1024L*1024L; in init()
1363 g->p.bytes_process_locked = g->p.mb_proc_locked *1024L*1024L; in init()
1364 g->p.bytes_thread = g->p.mb_thread *1024L*1024L; in init()
1366 g->data = setup_shared_data(g->p.bytes_global); in init()
1369 init_global_mutex(&g->start_work_mutex); in init()
1370 init_global_mutex(&g->startup_mutex); in init()
1371 init_global_mutex(&g->startup_done_mutex); in init()
1372 init_global_mutex(&g->stop_work_mutex); in init()
1388 free_data(g->data, g->p.bytes_global); in deinit()
1389 g->data = NULL; in deinit()
1393 free_data(g, sizeof(*g)); in deinit()
1394 g = NULL; in deinit()
1406 if (!g->p.show_quiet) in print_res()
1428 pids = zalloc(g->p.nr_proc * sizeof(*pids)); in __bench_numa()
1432 pthread_mutex_lock(&g->start_work_mutex); in __bench_numa()
1434 if (g->p.serialize_startup) { in __bench_numa()
1441 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1456 while (g->nr_tasks_started != g->p.nr_tasks) in __bench_numa()
1459 BUG_ON(g->nr_tasks_started != g->p.nr_tasks); in __bench_numa()
1461 if (g->p.serialize_startup) { in __bench_numa()
1464 pthread_mutex_lock(&g->startup_done_mutex); in __bench_numa()
1467 pthread_mutex_unlock(&g->start_work_mutex); in __bench_numa()
1470 pthread_mutex_lock(&g->startup_done_mutex); in __bench_numa()
1484 pthread_mutex_unlock(&g->startup_done_mutex); in __bench_numa()
1492 for (i = 0; i < g->p.nr_proc; i++) { in __bench_numa()
1502 for (t = 0; t < g->p.nr_tasks; t++) { in __bench_numa()
1503 u64 thread_runtime_ns = g->threads[t].runtime_ns; in __bench_numa()
1523 bytes = g->bytes_done; in __bench_numa()
1524 runtime_avg = (double)runtime_ns_sum / g->p.nr_tasks / 1e9; in __bench_numa()
1526 if (g->p.measure_convergence) { in __bench_numa()
1544 print_res(name, bytes / g->p.nr_tasks / 1e9, in __bench_numa()
1550 print_res(name, runtime_sec_max * 1e9 / (bytes / g->p.nr_tasks), in __bench_numa()
1553 print_res(name, bytes / g->p.nr_tasks / 1e9 / runtime_sec_max, in __bench_numa()