static int
abt_init(int argc, char *argv[])
{
int nrequested = abt_max_num_xstreams();
int nrequired = 1 /* primary xstream */ + DSS_XS_NR_TOTAL;
int rc;
/*
* Set ABT_MAX_NUM_XSTREAMS to the larger of nrequested and nrequired.
* If we don't do this, Argobots may use a default or requested value
* less than nrequired. We may then hit Argobots assertion failures
* because xstream_data.xd_mutex's internal queue has fewer slots than
* some xstreams' rank numbers need.
*/
rc = set_abt_max_num_xstreams(max(nrequested, nrequired));
if (rc != 0)
return daos_errno2der(errno);
/* Now, initialize Argobots. */
rc = ABT_init(argc, argv);
if (rc != ABT_SUCCESS) {
D_ERROR("failed to init ABT: %d\n", rc);
return dss_abterr2der(rc);
}
return 0;
}
int main(int argc, char *argv[])
{
ABT_xstream xstreams[NUM_XSTREAMS];
ABT_sched scheds[NUM_XSTREAMS];
int num_pools[NUM_XSTREAMS];
ABT_pool *pools[NUM_XSTREAMS];
int i, k;
ABT_init(argc, argv);
/* Create schedulers */
for (i = 0; i < NUM_XSTREAMS; i++) {
ABT_sched_predef predef;
predef = (i % 2) ? ABT_SCHED_BASIC : ABT_SCHED_PRIO;
ABT_sched_create_basic(predef, 0, NULL, ABT_SCHED_CONFIG_NULL,
&scheds[i]);
}
/* Create ESs */
ABT_xstream_self(&xstreams[0]);
ABT_xstream_set_main_sched(xstreams[0], scheds[0]);
for (i = 1; i < NUM_XSTREAMS; i++) {
ABT_xstream_create(scheds[i], &xstreams[i]);
}
/* Get the pools associated with each scheduler */
for (i = 0; i < NUM_XSTREAMS; i++) {
ABT_sched_get_num_pools(scheds[i], &num_pools[i]);
pools[i] = (ABT_pool *)malloc(num_pools[i] * sizeof(ABT_pool));
ABT_sched_get_pools(scheds[i], num_pools[i], 0, pools[i]);
}
/* Create ULTs */
for (i = 0; i < NUM_XSTREAMS; i++) {
for (k = num_pools[i] - 1; k >= 0; k--) {
size_t tid = (i + 1) * 10 + k;
ABT_thread_create(pools[i][k], thread_hello, (void *)tid,
ABT_THREAD_ATTR_NULL, NULL);
}
}
/* Join & Free */
for (i = 1; i < NUM_XSTREAMS; i++) {
ABT_xstream_join(xstreams[i]);
ABT_xstream_free(&xstreams[i]);
}
for (i = 0; i < NUM_XSTREAMS; i++) {
free(pools[i]);
}
ABT_finalize();
return 0;
}
GLT_func_prefix void glt_init(int argc, char * argv[]) {
int num_threads = get_nprocs();
main_team = (glt_team_t *) malloc(sizeof (glt_team_t));
CHECK(ABT_init(argc, argv),ABT_SUCCESS);
if (getenv("GLT_NUM_THREADS") != NULL) {
num_threads = atoi(getenv("GLT_NUM_THREADS"));
}
int num_pools = num_threads;
if (getenv("GLT_NUM_POOLS") != NULL) {
num_pools = atoi(getenv("GLT_NUM_POOLS"));
}
CHECK(ABT_xstream_self(&main_team->master),ABT_SUCCESS);
main_team->num_xstreams = num_threads;
main_team->num_pools = num_pools;
main_team->max_elem = get_nprocs();
main_team->team = (ABT_xstream *) malloc(sizeof (ABT_xstream) * num_threads);//main_team->max_elem);
main_team->pools = (ABT_pool *) malloc(sizeof (ABT_pool) * num_pools);//main_team->max_elem);
for (int i = 0; i < num_pools; i++) {
CHECK(ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE,
&main_team->pools[i]),ABT_SUCCESS);
}
CHECK(ABT_xstream_self(&main_team->team[0]),ABT_SUCCESS);
CHECK(ABT_xstream_set_main_sched_basic(main_team->team[0], ABT_SCHED_DEFAULT,
1, &main_team->pools[0]),ABT_SUCCESS);
for (int i = 1; i < num_threads; i++) {
CHECK(ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1,
&main_team->pools[i % main_team->num_pools],
ABT_SCHED_CONFIG_NULL, &main_team->team[i]),ABT_SUCCESS);
CHECK(ABT_xstream_start(main_team->team[i]),ABT_SUCCESS);
}
}
int main(int argc, char *argv[])
{
if (argc != 5) {
printf("Usage: %s <max # of ESs> <timeout> <server> <port>\n", argv[0]);
return EXIT_SUCCESS;
}
max_xstreams = atoi(argv[1]);
if (max_xstreams < 1) {
max_xstreams = 1;
}
num_xstreams = max_xstreams;
g_timeout = atof(argv[2]);
printf("# of ESs : %d\n", max_xstreams);
printf("Timeout : %.1f sec.\n", g_timeout);
abt_connect(argv[3], argv[4]);
/* Initialize */
ABT_init(argc, argv);
init(num_xstreams);
thread_add_sched((void *)0);
finalize();
/* Finalize */
ABT_finalize();
abt_disconnect();
printf("Done.\n"); fflush(stdout);
return EXIT_SUCCESS;
}
/* Main function */
int main(int argc, char *argv[])
{
int n, i, expected;
int num_xstreams;
ABT_xstream *xstreams;
ABT_thread thread;
thread_args args;
if (argc > 1 && strcmp(argv[1], "-h") == 0) {
printf("Usage: %s [N=10] [num_ES=4]\n", argv[0]);
return EXIT_SUCCESS;
}
n = argc > 1 ? atoi(argv[1]) : N;
num_xstreams = argc > 2 ? atoi(argv[2]) : NUM_XSTREAMS;
printf("# of ESs: %d\n", num_xstreams);
/* initialization */
ABT_init(argc, argv);
/* shared pool creation */
ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE,
&g_pool);
/* ES creation */
xstreams = (ABT_xstream *)malloc(sizeof(ABT_xstream) * num_xstreams);
ABT_xstream_self(&xstreams[0]);
ABT_xstream_set_main_sched_basic(xstreams[0], ABT_SCHED_DEFAULT,
1, &g_pool);
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1, &g_pool,
ABT_SCHED_CONFIG_NULL, &xstreams[i]);
ABT_xstream_start(xstreams[i]);
}
args.n = n;
args.future = ABT_FUTURE_NULL;
ABT_thread_create(g_pool, fibonacci, &args, ABT_THREAD_ATTR_NULL, &thread);
/* join the thread */
ABT_thread_join(thread);
ABT_thread_free(&thread);
/* join ESs */
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_join(xstreams[i]);
ABT_xstream_free(&xstreams[i]);
}
ABT_finalize();
free(xstreams);
printf("Fib(%d): %d\n", n, args.result);
expected = verify(n);
if (args.result != expected) {
fprintf(stderr, "ERROR: expected=%d\n", expected);
exit(EXIT_FAILURE);
}
return EXIT_SUCCESS;
}
int main(int argc, char *argv[])
{
int i, provided;
int ret;
int num_threads = DEFAULT_NUM_THREADS;
if (argc > 1)
num_threads = atoi(argv[1]);
if (num_threads % 2 != 0) {
if (rank == 0)
printf("number of user level threads should be even\n");
exit(0);
}
assert(num_threads >= 0);
num_loop = NUM_LOOP;
if (argc > 2)
num_loop = atoi(argv[2]);
assert(num_loop >= 0);
buf_size = BUF_SIZE;
if (argc > 3)
buf_size = atoi(argv[3]);
assert(buf_size >= 0);
/* Initialize */
ret = ABT_init(argc, argv);
if (ret != ABT_SUCCESS) {
printf("Failed to initialize Argobots\n");
return EXIT_FAILURE;
}
ret = MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
if (provided != MPI_THREAD_MULTIPLE) {
printf("Cannot initialize with MPI_THREAD_MULTIPLE\n");
return EXIT_FAILURE;
}
send_buf = (int *) malloc(buf_size * sizeof(int));
recv_buf = (int *) malloc(buf_size * sizeof(int));
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
if (profiling)
t1 = MPI_Wtime();
ABT_xstream xstream;
ABT_pool pools;
ABT_thread *threads = (ABT_thread *) malloc(num_threads * sizeof(ABT_thread));
ABT_xstream_self(&xstream);
ABT_xstream_get_main_pools(xstream, 1, &pools);
for (i = 0; i < num_threads; i++) {
size_t tid = i + 1;
if (i % 2 == 0) {
ret = ABT_thread_create(pools,
thread_send_func, (void *) tid,
ABT_THREAD_ATTR_NULL, &threads[i]);
} else {
ret = ABT_thread_create(pools,
thread_recv_func, (void *) tid,
ABT_THREAD_ATTR_NULL, &threads[i]);
}
HANDLE_ERROR(ret, "ABT_thread_create");
}
/* Join and free the ULT children */
for (i = 0; i < num_threads; i++)
ABT_thread_free(&threads[i]);
if (profiling) {
t2 = MPI_Wtime();
if (rank == 0) {
fprintf(stdout, "%*s%*f\n", FIELD_WIDTH, "Time", FIELD_WIDTH, t2 - t1);
}
}
/* Finalize */
free(threads);
free(send_buf);
free(recv_buf);
MPI_Finalize();
ABT_finalize();
return EXIT_SUCCESS;
}
void
__kmp_global_initialize(void)
{
int i;
int status;
/* Initialize Argobots before other initializations. */
status = ABT_init(0, NULL);
KMP_CHECK_SYSFAIL( "ABT_init", status );
__kmp_global.g = { 0 };
/* --------------------------------------------------------------------------- */
/* map OMP 3.0 schedule types with our internal schedule types */
static sched_type sch_map[ kmp_sched_upper - kmp_sched_lower_ext + kmp_sched_upper_std - kmp_sched_lower - 2 ] = {
kmp_sch_static_chunked, // ==> kmp_sched_static = 1
kmp_sch_dynamic_chunked, // ==> kmp_sched_dynamic = 2
kmp_sch_guided_chunked, // ==> kmp_sched_guided = 3
kmp_sch_auto, // ==> kmp_sched_auto = 4
kmp_sch_trapezoidal // ==> kmp_sched_trapezoidal = 101
// will likely not used, introduced here just to debug the code
// of public intel extension schedules
};
KMP_MEMCPY(__kmp_global.sch_map, sch_map, sizeof(sch_map));
#if OMP_40_ENABLED
__kmp_global.nested_proc_bind = { NULL, 0, 0 };
__kmp_global.affinity_num_places = 0;
#endif
__kmp_global.place_num_sockets = 0;
__kmp_global.place_socket_offset = 0;
__kmp_global.place_num_cores = 0;
__kmp_global.place_core_offset = 0;
__kmp_global.place_num_threads_per_core = 0;
__kmp_global.tasking_mode = tskm_task_teams;
__kmp_global.task_stealing_constraint = 1; /* Constrain task stealing by default */
#if OMP_41_ENABLED
__kmp_global.max_task_priority = 0;
#endif
__kmp_global.settings = FALSE;
__kmp_global.duplicate_library_ok = 0;
__kmp_global.force_reduction_method = reduction_method_not_defined;
__kmp_global.determ_red = FALSE;
__kmp_global.cpuinfo = { 0 };
/* ------------------------------------------------------------------------ */
__kmp_global.init_serial = FALSE;
__kmp_global.init_gtid = FALSE;
__kmp_global.init_common = FALSE;
__kmp_global.init_middle = FALSE;
__kmp_global.init_parallel = FALSE;
__kmp_global.init_runtime = FALSE;
__kmp_global.init_counter = 0;
__kmp_global.root_counter = 0;
__kmp_global.version = 0;
/* list of address of allocated caches for commons */
__kmp_global.threadpriv_cache_list = NULL;
/* Global Locks */
ABT_mutex_create(&__kmp_global.stdio_lock);
ABT_mutex_create(&__kmp_global.cat_lock);
ABT_mutex_create(&__kmp_global.initz_lock);
ABT_mutex_create(&__kmp_global.task_team_lock);
for (i = 0; i < KMP_NUM_CRIT_LOCKS; i++) {
ABT_mutex_create(&__kmp_global.crit_lock[i]);
}
__kmp_global.library = library_none;
__kmp_global.sched = kmp_sch_default; /* scheduling method for runtime scheduling */
__kmp_global.sched_static = kmp_sch_static_greedy; /* default static scheduling method */
__kmp_global.sched_guided = kmp_sch_guided_iterative_chunked; /* default guided scheduling method */
__kmp_global.sched_auto = kmp_sch_guided_analytical_chunked; /* default auto scheduling method */
__kmp_global.chunk = 0;
__kmp_global.stksize = KMP_DEFAULT_STKSIZE;
__kmp_global.stkoffset = KMP_DEFAULT_STKOFFSET;
__kmp_global.stkpadding = KMP_MIN_STKPADDING;
__kmp_global.malloc_pool_incr = KMP_DEFAULT_MALLOC_POOL_INCR;
__kmp_global.env_chunk = FALSE; /* KMP_CHUNK specified? */
__kmp_global.env_stksize = FALSE; /* KMP_STACKSIZE specified? */
__kmp_global.env_omp_stksize = FALSE; /* OMP_STACKSIZE specified? */
__kmp_global.env_all_threads = FALSE;/* KMP_ALL_THREADS or KMP_MAX_THREADS specified? */
__kmp_global.env_omp_all_threads = FALSE;/* OMP_THREAD_LIMIT specified? */
__kmp_global.env_checks = FALSE; /* KMP_CHECKS specified? */
__kmp_global.env_consistency_check = FALSE; /* KMP_CONSISTENCY_CHECK specified? */
__kmp_global.generate_warnings = kmp_warnings_low;
__kmp_global.reserve_warn = 0;
#ifdef DEBUG_SUSPEND
__kmp_global.suspend_count = 0;
#endif
/* ------------------------------------------------------------------------- */
__kmp_global.allThreadsSpecified = 0;
__kmp_global.align_alloc = CACHE_LINE;
__kmp_global.xproc = 0;
__kmp_global.avail_proc = 0;
__kmp_global.sys_min_stksize = KMP_MIN_STKSIZE;
__kmp_global.sys_max_nth = KMP_MAX_NTH;
__kmp_global.max_nth = 0;
__kmp_global.threads_capacity = 0;
__kmp_global.dflt_team_nth = 0;
__kmp_global.dflt_team_nth_ub = 0;
__kmp_global.tp_capacity = 0;
__kmp_global.tp_cached = 0;
__kmp_global.dflt_nested = TRUE;
__kmp_global.dflt_max_active_levels = KMP_MAX_ACTIVE_LEVELS_LIMIT; /* max_active_levels limit */
#ifdef KMP_DFLT_NTH_CORES
__kmp_global.ncores = 0;
#endif
__kmp_global.abort_delay = 0;
/* Initialize the library data structures when we fork a child process, defaults to TRUE */
__kmp_global.need_register_atfork = TRUE; /* At initialization, call pthread_atfork to install fork handler */
__kmp_global.need_register_atfork_specified = TRUE;
__kmp_global.tls_gtid_min = INT_MAX;
__kmp_global.foreign_tp = TRUE;
#if KMP_ARCH_X86 || KMP_ARCH_X86_64
__kmp_global.inherit_fp_control = TRUE;
__kmp_global.init_x87_fpu_control_word = 0;
__kmp_global.init_mxcsr = 0;
#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
#if KMP_NESTED_HOT_TEAMS
__kmp_global.hot_teams_mode = 0; /* 0 - free extra threads when reduced */
/* 1 - keep extra threads when reduced */
__kmp_global.hot_teams_max_level = 1; /* nesting level of hot teams */
#endif
#if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_WINDOWS)
__kmp_global.mic_type = non_mic;
#endif
#ifdef USE_LOAD_BALANCE
__kmp_global.load_balance_interval = 1.0;
#endif /* USE_LOAD_BALANCE */
__kmp_global.nested_nth = { NULL, 0, 0 };
#if OMP_40_ENABLED
__kmp_global.display_env = FALSE;
__kmp_global.display_env_verbose = FALSE;
__kmp_global.omp_cancellation = FALSE;
#endif
/* ------------------------------------------------------ */
/* STATE mostly syncronized with global lock */
/* data written to rarely by masters, read often by workers */
__kmp_global.threads = NULL;
__kmp_global.team_pool = NULL;
__kmp_global.thread_pool = NULL;
/* data read/written to often by masters */
__kmp_global.nth = 0;
__kmp_global.all_nth = 0;
__kmp_global.thread_pool_nth = 0;
__kmp_global.thread_pool_active_nth = 0;
__kmp_global.root = NULL;
/* ------------------------------------------------------ */
__kmp_init_global = TRUE;
}
int main(int argc, char *argv[])
{
ABT_xstream xstream;
ABT_pool pool;
ABT_timer timer;
int i, iter;
double t_overhead;
ABT_test_read_args(argc, argv);
iter = ABT_test_get_arg_val(ABT_TEST_ARG_N_ITER);
ABT_timer_create(&timer);
ABT_timer_start(timer);
ABT_timer_stop(timer);
ABT_timer_get_overhead(&t_overhead);
for (i = 0; i < T_LAST; i++) t_timers[i] = 0.0;
/* measure init/finalize time (cold) */
ABT_timer_start(timer);
ABT_init(argc, argv);
ABT_finalize();
ABT_timer_stop_and_read(timer, &t_timers[T_INIT_FINALIZE_COLD]);
t_timers[T_INIT_FINALIZE_COLD] -= t_overhead;
/* measure init/finalize time */
for (i = 0; i < iter; i++) {
ABT_timer_start(timer);
ABT_init(argc, argv);
ABT_timer_stop_and_add(timer, &t_timers[T_INIT]);
ABT_timer_start(timer);
ABT_finalize();
ABT_timer_stop_and_add(timer, &t_timers[T_FINALIZE]);
}
t_timers[T_INIT] /= iter;
t_timers[T_INIT] -= t_overhead;
t_timers[T_FINALIZE] /= iter;
t_timers[T_FINALIZE] -= t_overhead;
t_timers[T_INIT_FINALIZE] = t_timers[T_INIT] + t_timers[T_FINALIZE];
/* measure time of init/finalize with work */
for (i = 0; i < iter; i++) {
ABT_timer_start(timer);
ABT_init(argc, argv);
ABT_timer_stop_and_add(timer, &t_timers[T_INIT_FINALIZE_WITH_WORK]);
ABT_xstream_create(ABT_SCHED_NULL, &xstream);
ABT_xstream_get_main_pools(xstream, 1, &pool);
ABT_thread_create(pool, thread_func, NULL, ABT_THREAD_ATTR_NULL,
NULL);
ABT_xstream_join(xstream);
ABT_xstream_free(&xstream);
ABT_timer_start(timer);
ABT_finalize();
ABT_timer_stop_and_add(timer, &t_timers[T_INIT_FINALIZE_WITH_WORK]);
}
t_timers[T_INIT_FINALIZE_WITH_WORK] /= iter;
t_timers[T_INIT_FINALIZE_WITH_WORK] -= t_overhead;
/* output */
int line_size = 45;
ABT_test_print_line(stdout, '-', line_size);
printf("Avg. execution time (in seconds, %d times)\n", iter);
ABT_test_print_line(stdout, '-', line_size);
for (i = 0; i < T_LAST; i++) {
printf("%-23s %.9f\n", t_names[i], t_timers[i]);
}
ABT_test_print_line(stdout, '-', line_size);
ABT_timer_free(&timer);
return EXIT_SUCCESS;
}
/* Main function */
int main(int argc, char *argv[])
{
int n, i, result, expected;
int num_xstreams;
ABT_xstream *xstreams;
ABT_thread thread;
thread_args args_thread;
ABT_task task;
task_args *args_task;
if (argc > 1 && strcmp(argv[1], "-h") == 0) {
printf("Usage: %s [N=10] [num_ES=4]\n", argv[0]);
return EXIT_SUCCESS;
}
n = argc > 1 ? atoi(argv[1]) : N;
num_xstreams = argc > 2 ? atoi(argv[2]) : NUM_XSTREAMS;
printf("# of ESs: %d\n", num_xstreams);
if (n <= 2) {
result = 1;
goto fn_result;
}
/* initialization */
ABT_init(argc, argv);
/* shared pool creation */
ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE,
&g_pool);
/* ES creation */
xstreams = (ABT_xstream *)malloc(sizeof(ABT_xstream) * num_xstreams);
ABT_xstream_self(&xstreams[0]);
ABT_xstream_set_main_sched_basic(xstreams[0], ABT_SCHED_DEFAULT,
1, &g_pool);
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1, &g_pool,
ABT_SCHED_CONFIG_NULL, &xstreams[i]);
ABT_xstream_start(xstreams[i]);
}
/* creating thread */
args_thread.n = n - 1;
args_thread.eventual = ABT_EVENTUAL_NULL;
ABT_thread_create(g_pool, fibonacci_thread, &args_thread, ABT_THREAD_ATTR_NULL, &thread);
/* creating task */
args_task = (task_args *)malloc(sizeof(task_args));
args_task->n = n - 2;
args_task->result = 0;
ABT_mutex_create(&args_task->mutex);
args_task->parent = NULL;
ABT_task_create(g_pool, fibonacci_task, args_task, &task);
/* switch to other user-level threads */
ABT_thread_yield();
/* join other threads */
ABT_thread_join(thread);
ABT_thread_free(&thread);
/* join ESs */
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_join(xstreams[i]);
ABT_xstream_free(&xstreams[i]);
}
result = args_thread.result + args_task->result;
free(args_task);
ABT_finalize();
free(xstreams);
fn_result:
printf("Fib(%d): %d\n", n, result);
expected = verify(n);
if (result != expected) {
fprintf(stderr, "ERROR: expected=%d\n", expected);
exit(EXIT_FAILURE);
}
return EXIT_SUCCESS;
}
int main(int argc, char *argv[])
{
int i, j, r;
int num_xstreams;
char *str, *endptr;
ABT_xstream *xstreams;
ABT_thread *threads;
vector_scal_task_args_t *args;
int inner_xstreams;
double *time, avg_time = 0.0;
num_xstreams = (argc > 1) ? atoi(argv[1]) : NUM_XSTREAMS;
inner_xstreams = (argc > 2) ? atoi(argv[2]) : NUM_XSTREAMS;
int rep = (argc > 3) ? atoi(argv[3]) : NUM_REPS;
time = (double *)malloc(sizeof(double) * rep);
init();
g_pools = (ABT_pool *)malloc(sizeof(ABT_pool) * num_xstreams);
xstreams = (ABT_xstream *)malloc(sizeof(ABT_xstream) * num_xstreams);
threads = (ABT_thread *)malloc(sizeof(ABT_thread) * num_xstreams);
args = (vector_scal_task_args_t *)malloc(sizeof(vector_scal_task_args_t)
* num_xstreams);
/* initialization */
ABT_init(argc, argv);
for (i = 0; i < num_xstreams; i++) {
ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE,
&g_pools[i]);
}
/* ES creation */
ABT_xstream_self(&xstreams[0]);
ABT_xstream_set_main_sched_basic(xstreams[0], ABT_SCHED_DEFAULT,
1, &g_pools[0]);
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1, &g_pools[i],
ABT_SCHED_CONFIG_NULL, &xstreams[i]);
ABT_xstream_start(xstreams[i]);
}
/* Each task is created on the xstream which is going to execute it */
for (r = 0; r < rep; r++) {
time[r] = ABT_get_wtime();
int bloc = NUM / (num_xstreams);
int rest = NUM % (num_xstreams);
int start = 0;
int end = 0;
for (j = 0; j < num_xstreams; j++) {
start = end;
int inc = (j < rest) ? 1 : 0;
end += bloc + inc;
args[j].start = start;
args[j].end = end;
args[j].it = NUM;
args[j].nxstreams = inner_xstreams;
if (j > 0) {
ABT_thread_create(g_pools[j], vector_scal_launch,
(void *)&args[j], ABT_THREAD_ATTR_NULL,
&threads[j]);
}
}
vector_scal_launch((void *)&args[0]);
for (j = 1; j < num_xstreams; j++) {
ABT_thread_free(&threads[j]);
}
time[r] = ABT_get_wtime() - time[r];
avg_time += time[r];
}
avg_time /= rep;
printf("%d %d %f\n", num_xstreams, inner_xstreams, avg_time);
check();
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_join(xstreams[i]);
ABT_xstream_free(&xstreams[i]);
}
ABT_finalize();
free(g_pools);
free(xstreams);
free(threads);
free(args);
free(time);
return EXIT_SUCCESS;
}
void accalt_init(int argc, char * argv[]) {
int num_threads = 1;
main_team = (accalt_team_t *) malloc(sizeof (accalt_team_t));
#ifdef ARGOBOTS
ABT_init(argc, argv);
int num_pools = 1;
if (getenv("ACCALT_NUM_THREADS") != NULL) {
num_threads = atoi(getenv("ACCALT_NUM_THREADS"));
}
if (getenv("ACCALT_NUM_POOLS") != NULL) {
num_pools = atoi(getenv("ACCALT_NUM_POOLS"));
}
main_team->num_xstreams = num_threads;
main_team->num_pools = num_pools;
//printf("Argobots %d ES, %d Pools\n", num_threads, num_pools);
ABT_xstream_self(&main_team->master);
main_team->team = (ABT_xstream *) malloc(sizeof (ABT_xstream) * num_threads);
main_team->pools = (ABT_pool *) malloc(sizeof (ABT_pool) * num_pools);
for (int i = 0; i < num_pools; i++) {
ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE,
&main_team->pools[i]);
}
ABT_xstream_self(&main_team->team[0]);
ABT_xstream_set_main_sched_basic(main_team->team[0], ABT_SCHED_DEFAULT,
1, &main_team->pools[0]);
for (int i = 1; i < num_threads; i++) {
ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1,
&main_team->pools[i % main_team->num_pools],
ABT_SCHED_CONFIG_NULL, &main_team->team[i]);
ABT_xstream_start(main_team->team[i]);
}
#endif
#ifdef MASSIVETHREADS
char buff[10];
if (getenv("ACCALT_NUM_THREADS") != NULL) {
num_threads = atoi(getenv("ACCALT_NUM_THREADS"));
sprintf(buff, "%d", num_threads);
setenv("MYTH_WORKER_NUM", buff, 1);
} else
num_threads = atoi(getenv("MYTH_WORKER_NUM"));
setenv("MYTH_BIND_WORKERS", "1", 1);
//printf("Massive %d Workers\n", num_threads);
main_team->num_workers = num_threads;
myth_init(); //MassiveThreads
#endif
#ifdef QTHREADS
char buff[10];
int num_workers_per_thread;
if (getenv("ACCALT_NUM_THREADS") != NULL) {
num_threads = atoi(getenv("ACCALT_NUM_THREADS"));
sprintf(buff, "%d", num_threads);
setenv("QTHREAD_NUM_SHEPHERDS", buff, 1);
} else
num_threads = atoi(getenv("QTHREAD_NUM_SHEPHERDS"));
if (getenv("ACCALT_NUM_WORKERS_PER_THREAD") != NULL) {
num_workers_per_thread = atoi(getenv("ACCALT_NUM_WORKERS_PER_THREAD"));
sprintf(buff, "%d", num_workers_per_thread);
setenv("QTHREAD_NUM_WORKERS_PER_SHEPHERD", buff, 1);
} else
num_workers_per_thread = atoi(getenv("QTHREAD_NUM_WORKERS_PER_SHEPHERD"));
if (num_threads == 1 && num_workers_per_thread > 1) {
setenv("QTHREAD_SHEPHERDS_BOUNDARY", "node", 1);
setenv("QTHREAD_WORKER_UNIT", "core", 1);
}
if (num_threads > 1) {
setenv("QTHREAD_SHEPHERDS_BOUNDARY", "core", 1);
setenv("QTHREAD_WORKER_UNIT", "core", 1);
}
setenv("QTHREAD_AFFINITY", "yes", 1);
//printf("Qthreads %d Shepherds, %d Workers_per_shepherd\n", num_threads, num_workers_per_thread);
main_team->num_shepherds = num_threads;
main_team->num_workers_per_shepherd = num_workers_per_thread;
qthread_initialize(); //qthreads
#endif
}
int main(int argc, char *argv[])
{
// Argobots definitions
int ret, i;
num_threads = 0;
// Sockets Definitions
int fd, cfd;
struct sockaddr_in svaddr;
struct sockaddr_storage claddr;
socklen_t addrlen;
signal(SIGINT, sighandler);
ABT_init(argc, argv);
int abts = 0;//ABT_snoozer_xstream_self_set();
if (abts != 0){
fprintf(stderr, "%s\n", "ABT snoozer xstream self error");
exit(-1);
}
xstreams = (ABT_xstream *)malloc(sizeof(ABT_xstream) * CORES);
ret = ABT_xstream_self(&xstreams[0]);
if(ret != 0){
fprintf(stderr, "%s\n", "ABT xstream self error");
exit(-1);
}
ret = ABT_xstream_get_main_pools(xstreams[0], 1, &pool);
if(ret != 0){
fprintf(stderr, "%s\n", "ABT xstream pool error");
exit(-1);
}
ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE, &g_pool);
/* ES creation */
ABT_xstream_self(&xstreams[0]);
ABT_xstream_set_main_sched_basic(xstreams[0], ABT_SCHED_DEFAULT,
1, &g_pool);
for (i = 1; i < CORES; i++) {
ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1, &g_pool,
ABT_SCHED_CONFIG_NULL, &xstreams[i]);
ABT_xstream_start(xstreams[i]);
}
//abtio = abt_io_init(CORES);
//assert(abtio != NULL);
fd = socket(AF_INET, SOCK_STREAM, 0);
if(fd < 0){
fprintf(stderr, "%s\n", "Socket creating error");
exit(-1);
}
memset(&svaddr, 0, sizeof(struct sockaddr_in));
svaddr.sin_family = AF_INET;
svaddr.sin_addr.s_addr = INADDR_ANY;
svaddr.sin_port = htons(PORTNUM);
ret = bind(fd, (struct sockaddr *) &svaddr, sizeof(struct sockaddr_in));
if(ret < 0){
fprintf(stderr, "%s\n", "Socket binding error");
exit(-1);
}
if( listen(fd, BACKLOG) == -1)
{
fprintf(stderr, "%s\n", "Socket listening error");
exit(-1);
}
int aio_sock = abt_io_socket_initialize(1000);
if(aio_sock <= 0)
printf("Initialize io_sock error\n");
//printf("USer: main: epoll fd = %d\n", aio_sock);
addrlen = sizeof(struct sockaddr_storage);
while(1){
cfd = accept(fd, (struct sockaddr *)&claddr, &addrlen);
if(cfd == -1){
fprintf(stderr, "%s\n", "Socket accepting error");
exit(-1);
}
//printf("accepted client on file descriptor %d\n", cfd);
struct thread_args* ta = (struct thread_args*) malloc (sizeof(ta));
ta->epfd = aio_sock;
ta->fd = cfd;
ABT_thread_create(g_pool, handle_client, (void *) ta, ABT_THREAD_ATTR_NULL, threads[num_threads++]);
}
return 0;
}
int main(int argc, char *argv[])
{
int i, j;
int ntasks;
int start, end;
int num_xstreams;
ABT_xstream *xstreams;
vector_scal_task_args_t *args;
struct timeval t_start, t_end, t_end2;
char *str, *endptr;
float *a;
num_xstreams = argc > 1 ? atoi(argv[1]) : NUM_XSTREAMS;
if (argc > 2) {
str = argv[2];
}
ntasks = argc > 2 ? strtoll(str, &endptr, 10) : NUM_TASKS;
if (ntasks < num_xstreams) {
ntasks = num_xstreams;
}
printf("# of ESs: %d\n", num_xstreams);
xstreams = (ABT_xstream *)malloc(sizeof(ABT_xstream) * num_xstreams);
args = (vector_scal_task_args_t *)malloc(sizeof(vector_scal_task_args_t)
* num_xstreams);
g_pools = (ABT_pool *)malloc(sizeof(ABT_pool) * num_xstreams);
a = malloc(sizeof(float) * ntasks);
for (i = 0; i < ntasks; i++) {
a[i] = i + 100.0f;
}
/* initialization */
ABT_init(argc, argv);
for (i = 0; i < num_xstreams; i++) {
ABT_pool_create_basic(ABT_POOL_FIFO, ABT_POOL_ACCESS_MPMC, ABT_TRUE,
&g_pools[i]);
}
/* ES creation */
ABT_xstream_self(&xstreams[0]);
ABT_xstream_set_main_sched_basic(xstreams[0], ABT_SCHED_DEFAULT,
1, &g_pools[0]);
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_create_basic(ABT_SCHED_DEFAULT, 1, &g_pools[i],
ABT_SCHED_CONFIG_NULL, &xstreams[i]);
ABT_xstream_start(xstreams[i]);
}
/* Work here */
start = end = 0;
int bloc = ntasks / num_xstreams;
int rest = ntasks % num_xstreams;
gettimeofday(&t_start, NULL);
for (j = 0; j < num_xstreams; j++) {
start = end;
end = start + bloc;
if (j < rest) {
end++;
}
args[j].ptr = a;
args[j].value = 0.9f;
args[j].start = start;
args[j].end = end;
args[j].id = j;
ABT_thread_create_on_xstream(xstreams[j], task_creator,
(void *)&args[j], ABT_THREAD_ATTR_NULL,
NULL);
}
gettimeofday(&t_end2, NULL);
for (i = 0; i < num_xstreams; i++) {
size_t size;
do {
ABT_thread_yield();
ABT_pool_get_size(g_pools[i], &size);
} while (size != 0);
}
gettimeofday(&t_end, NULL);
for (i = 0; i < ntasks; i++) {
if (a[i] != (i + 100.0f) * 0.9f) {
printf("error: a[%d]\n", i);
}
}
double time = (t_end.tv_sec * 1000000 + t_end.tv_usec)
- (t_start.tv_sec * 1000000 + t_start.tv_usec);
double time2 = (t_end2.tv_sec * 1000000 + t_end2.tv_usec)
- (t_start.tv_sec * 1000000 + t_start.tv_usec);
printf("nxstreams: %d\nntasks %d\nTime(s): %f\n",
num_xstreams, ntasks, time / 1000000.0);
/* join ESs */
for (i = 1; i < num_xstreams; i++) {
ABT_xstream_join(xstreams[i]);
ABT_xstream_free(&xstreams[i]);
}
printf("Creation time=%f\n", time2 / 1000000.0);
ABT_finalize();
free(xstreams);
return EXIT_SUCCESS;
}