void accalt_tasklet_join(ACCALT_tasklet *tasklet) {
#ifdef ARGOBOTS
ABT_task_free(tasklet);
#endif
#ifdef MASSIVETHREADS
myth_join(*tasklet, NULL);
#endif
#ifdef QTHREADS
qthread_readFF(NULL, tasklet);
#endif
}
void vector_scal_launch(void *arguments)
{
int i, it, j, num_ults, rank, mystart, myend, p;
ABT_task *tasks;
ABT_xstream xstream;
ABT_xstream_self(&xstream);
vector_scal_task_args_t *arg;
arg = (vector_scal_task_args_t *) arguments;
vector_scal_args_t *args;
it = arg->it;
num_ults = arg->nxstreams;
mystart = arg->start;
myend = arg->end;
args = (vector_scal_args_t *)malloc(sizeof(vector_scal_args_t)
* num_ults);
tasks = (ABT_task *)malloc(sizeof(ABT_task) * num_ults);
int bloc = it / (num_ults);
int rest = it % (num_ults);
ABT_xstream_self_rank(&rank);
for (i = mystart; i < myend; i++) {
int start = 0;
int end = 0;
for (j = 0; j < num_ults; j++) {
start = end;
int inc = (j < rest) ? 1 : 0;
end += bloc + inc;
args[j].start = start;
args[j].end = end;
args[j].x = i;
if (j > 0) {
ABT_task_create(g_pools[rank], vector_scal, (void *)&args[j], &tasks[j]);
}
}
vector_scal((void *)&args[0]);
for (p = 1; p < num_ults; p++) {
ABT_task_free(&tasks[p]);
}
}
free(tasks);
free(args);
}
void eventual_test(void *arg)
{
int i, t, ret;
size_t pid = (size_t)arg;
int eid = (int)pid;
ABT_thread *threads;
arg_t *thread_args;
int *nbytes;
ABT_eventual *evs1;
ABT_thread *waiters;
arg_t *waiter_args;
ABT_task *tasks;
arg_t *task_args;
ABT_eventual *evs2;
ABT_test_printf(1, "[M%u] start\n", (unsigned)(size_t)arg);
assert(num_tasks * 2 == num_threads);
threads = (ABT_thread *)malloc(num_threads * sizeof(ABT_thread));
thread_args = (arg_t *)malloc(num_threads * sizeof(arg_t));
nbytes = (int *)malloc(num_tasks * sizeof(int));
evs1 = (ABT_eventual *)malloc(num_tasks * sizeof(ABT_eventual));
assert(threads && thread_args && nbytes && evs1);
waiters = (ABT_thread *)malloc(num_tasks * sizeof(ABT_thread));
waiter_args = (arg_t *)malloc(num_tasks * sizeof(arg_t));
tasks = (ABT_task *)malloc(num_tasks * sizeof(ABT_task));
task_args = (arg_t *)malloc(num_tasks * sizeof(arg_t));
evs2 = (ABT_eventual *)malloc(num_tasks * sizeof(ABT_eventual));
assert(waiters && waiter_args && tasks && task_args && evs2);
for (t = 0; t < num_tasks; t++) {
nbytes[t] = (t & 1) ? sizeof(int) : 0;
ret = ABT_eventual_create(nbytes[t], &evs1[t]);
ABT_TEST_ERROR(ret, "ABT_eventual_create");
ret = ABT_eventual_create(nbytes[t], &evs2[t]);
ABT_TEST_ERROR(ret, "ABT_eventual_create");
}
for (i = 0; i < num_iter; i++) {
ABT_test_printf(2, "[M%u] iter=%d\n", (unsigned)(size_t)arg, i);
/* Use eventual between ULTs */
for (t = 0; t < num_threads; t += 2) {
thread_args[t].eid = eid;
thread_args[t].tid = t;
thread_args[t].ev = evs1[t/2];
thread_args[t].nbytes = nbytes[t/2];
thread_args[t].op_type = OP_WAIT;
ret = ABT_thread_create(pools[pid], thread_func, &thread_args[t],
ABT_THREAD_ATTR_NULL, &threads[t]);
ABT_TEST_ERROR(ret, "ABT_thread_create");
pid = (pid + 1) % num_xstreams;
thread_args[t+1].eid = eid;
thread_args[t+1].tid = t + 1;
thread_args[t+1].ev = evs1[t/2];
thread_args[t+1].nbytes = nbytes[t/2];
thread_args[t+1].op_type = OP_SET;
ret = ABT_thread_create(pools[pid], thread_func, &thread_args[t+1],
ABT_THREAD_ATTR_NULL, &threads[t+1]);
ABT_TEST_ERROR(ret, "ABT_thread_create");
pid = (pid + 1) % num_xstreams;
}
/* Use eventual between ULT and tasklet */
for (t = 0; t < num_tasks; t++) {
waiter_args[t].ev = evs2[t];
waiter_args[t].nbytes = nbytes[t];
waiter_args[t].op_type = OP_WAIT;
ret = ABT_thread_create(pools[pid], thread_func, &waiter_args[t],
ABT_THREAD_ATTR_NULL, &waiters[t]);
ABT_TEST_ERROR(ret, "ABT_thread_create");
pid = (pid + 1) % num_xstreams;
task_args[t].ev = evs2[t];
task_args[t].nbytes = nbytes[t];
task_args[t].op_type = OP_SET;
ret = ABT_task_create(pools[pid], task_func, &task_args[t], &tasks[t]);
ABT_TEST_ERROR(ret, "ABT_task_create");
pid = (pid + 1) % num_xstreams;
}
for (t = 0; t < num_threads; t++) {
ret = ABT_thread_free(&threads[t]);
ABT_TEST_ERROR(ret, "ABT_thread_free");
}
for (t = 0; t < num_tasks; t++) {
ret = ABT_thread_free(&waiters[t]);
ABT_TEST_ERROR(ret, "ABT_thread_free");
ret = ABT_task_free(&tasks[t]);
ABT_TEST_ERROR(ret, "ABT_task_free");
}
for (t = 0; t < num_tasks; t++) {
ret = ABT_eventual_reset(evs1[t]);
ABT_TEST_ERROR(ret, "ABT_eventual_reset");
ret = ABT_eventual_reset(evs2[t]);
ABT_TEST_ERROR(ret, "ABT_eventual_reset");
}
}
for (t = 0; t < num_tasks; t++) {
ret = ABT_eventual_free(&evs1[t]);
ABT_TEST_ERROR(ret, "ABT_eventual_free");
ret = ABT_eventual_free(&evs2[t]);
ABT_TEST_ERROR(ret, "ABT_eventual_free");
}
free(threads);
free(thread_args);
free(nbytes);
free(evs1);
free(waiters);
free(waiter_args);
free(tasks);
free(task_args);
free(evs2);
ABT_test_printf(1, "[M%u] end\n", (unsigned)(size_t)arg);
}
int main(int argc, char *argv[])
{
int i, j, ret;
int num_xstreams = DEFAULT_NUM_XSTREAMS;
int num_threads = DEFAULT_NUM_THREADS;
int num_tasks = DEFAULT_NUM_TASKS;
if (argc > 1) num_xstreams = atoi(argv[1]);
assert(num_xstreams >= 0);
if (argc > 2) num_threads = atoi(argv[2]);
assert(num_threads >= 0);
if (argc > 3) num_tasks = atoi(argv[3]);
assert(num_tasks >= 0);
ABT_xstream *xstreams;
ABT_thread **threads;
thread_arg_t **thread_args;
ABT_task *tasks;
task_arg_t *task_args;
xstreams = (ABT_xstream *)malloc(sizeof(ABT_xstream) * num_xstreams);
threads = (ABT_thread **)malloc(sizeof(ABT_thread *) * num_xstreams);
thread_args = (thread_arg_t **)malloc(sizeof(thread_arg_t*) * num_xstreams);
for (i = 0; i < num_xstreams; i++) {
threads[i] = (ABT_thread *)malloc(sizeof(ABT_thread) * num_threads);
for (j = 0; j < num_threads; j++) {
threads[i][j] = ABT_THREAD_NULL;
}
thread_args[i] = (thread_arg_t *)malloc(sizeof(thread_arg_t) *
num_threads);
}
tasks = (ABT_task *)malloc(sizeof(ABT_task) * num_tasks);
task_args = (task_arg_t *)malloc(sizeof(task_arg_t) * num_tasks);
/* Initialize */
ABT_test_init(argc, argv);
/* Create Execution Streams */
ret = ABT_xstream_self(&xstreams[0]);
ABT_TEST_ERROR(ret, "ABT_xstream_self");
for (i = 1; i < num_xstreams; i++) {
ret = ABT_xstream_create(ABT_SCHED_NULL, &xstreams[i]);
ABT_TEST_ERROR(ret, "ABT_xstream_create");
}
/* Get the pools attached to an execution stream */
ABT_pool *pools;
pools = (ABT_pool *)malloc(sizeof(ABT_pool) * num_xstreams);
for (i = 0; i < num_xstreams; i++) {
ret = ABT_xstream_get_main_pools(xstreams[i], 1, pools+i);
ABT_TEST_ERROR(ret, "ABT_xstream_get_main_pools");
}
/* Create threads */
for (i = 0; i < num_xstreams; i++) {
for (j = 0; j < num_threads; j++) {
int tid = i * num_threads + j + 1;
thread_args[i][j].id = tid;
thread_args[i][j].num_threads = num_threads;
thread_args[i][j].threads = &threads[i][0];
ret = ABT_thread_create(pools[i],
thread_func, (void *)&thread_args[i][j],
ABT_THREAD_ATTR_NULL,
&threads[i][j]);
ABT_TEST_ERROR(ret, "ABT_thread_create");
}
}
/* Create tasks with task_func1 */
for (i = 0; i < num_tasks; i++) {
size_t num = 100 + i;
ret = ABT_task_create(pools[i % num_xstreams],
task_func1, (void *)num,
NULL);
ABT_TEST_ERROR(ret, "ABT_task_create");
}
/* Create tasks with task_func2 */
for (i = 0; i < num_tasks; i++) {
task_args[i].num = 100 + i;
ret = ABT_task_create(pools[i % num_xstreams],
task_func2, (void *)&task_args[i],
&tasks[i]);
ABT_TEST_ERROR(ret, "ABT_task_create");
}
/* Switch to other work units */
ABT_thread_yield();
/* Results of task_funcs2 */
for (i = 0; i < num_tasks; i++) {
ABT_task_state state;
do {
ABT_task_get_state(tasks[i], &state);
ABT_thread_yield();
} while (state != ABT_TASK_STATE_TERMINATED);
ABT_test_printf(1, "task_func2: num=%lu result=%llu\n",
task_args[i].num, task_args[i].result);
/* Free named tasks */
ret = ABT_task_free(&tasks[i]);
ABT_TEST_ERROR(ret, "ABT_task_free");
}
/* Join Execution Streams */
for (i = 1; i < num_xstreams; i++) {
ret = ABT_xstream_join(xstreams[i]);
ABT_TEST_ERROR(ret, "ABT_xstream_join");
}
/* Free Execution Streams */
for (i = 0; i < num_xstreams; i++) {
for (j = 0; j < num_threads; j++) {
ret = ABT_thread_free(&threads[i][j]);
ABT_TEST_ERROR(ret, "ABT_thread_free");
}
if (i == 0) continue;
ret = ABT_xstream_free(&xstreams[i]);
ABT_TEST_ERROR(ret, "ABT_xstream_free");
}
/* Finalize */
ret = ABT_test_finalize(0);
for (i = 0; i < num_xstreams; i++) {
free(thread_args[i]);
free(threads[i]);
}
free(thread_args);
free(threads);
free(task_args);
free(tasks);
free(pools);
free(xstreams);
return ret;
}
