int main(int argc, char *argv[])
{
aligned_t rets[NUM_THREADS];
qtimer_t timer = qtimer_create();
double cumulative_time = 0.0;
if (qthread_initialize() != QTHREAD_SUCCESS) {
fprintf(stderr, "qthread library could not be initialized!\n");
exit(EXIT_FAILURE);
}
CHECK_VERBOSE();
for (int iteration = 0; iteration < 10; iteration++) {
qtimer_start(timer);
for (int i = 0; i < NUM_THREADS; i++) {
qthread_fork(qincr, NULL, &(rets[i]));
}
for (int i = 0; i < NUM_THREADS; i++) {
qthread_readFF(NULL, &(rets[i]));
}
qtimer_stop(timer);
iprintf("\ttest iteration %i: %f secs\n", iteration,
qtimer_secs(timer));
cumulative_time += qtimer_secs(timer);
}
printf("qthread time: %f\n", cumulative_time / 10.0);
return 0;
}
void Qthread::initialize( int thread_count )
{
// Environment variable: QTHREAD_NUM_SHEPHERDS
// Environment variable: QTHREAD_NUM_WORKERS_PER_SHEP
// Environment variable: QTHREAD_HWPAR
{
char buffer[256];
snprintf(buffer,sizeof(buffer),"QTHREAD_HWPAR=%d",thread_count);
putenv(buffer);
}
const bool ok_init = ( QTHREAD_SUCCESS == qthread_initialize() ) &&
( thread_count == qthread_num_shepherds() * qthread_num_workers_local(NO_SHEPHERD) ) &&
( thread_count == qthread_num_workers() );
bool ok_symmetry = true ;
if ( ok_init ) {
Impl::s_number_shepherds = qthread_num_shepherds();
Impl::s_number_workers_per_shepherd = qthread_num_workers_local(NO_SHEPHERD);
Impl::s_number_workers = Impl::s_number_shepherds * Impl::s_number_workers_per_shepherd ;
for ( int i = 0 ; ok_symmetry && i < Impl::s_number_shepherds ; ++i ) {
ok_symmetry = ( Impl::s_number_workers_per_shepherd == qthread_num_workers_local(i) );
}
}
if ( ! ok_init || ! ok_symmetry ) {
std::ostringstream msg ;
msg << "Kokkos::Qthread::initialize(" << thread_count << ") FAILED" ;
msg << " : qthread_num_shepherds = " << qthread_num_shepherds();
msg << " : qthread_num_workers_per_shepherd = " << qthread_num_workers_local(NO_SHEPHERD);
msg << " : qthread_num_workers = " << qthread_num_workers();
if ( ! ok_symmetry ) {
msg << " : qthread_num_workers_local = {" ;
for ( int i = 0 ; i < Impl::s_number_shepherds ; ++i ) {
msg << " " << qthread_num_workers_local(i) ;
}
msg << " }" ;
}
Impl::s_number_workers = 0 ;
Impl::s_number_shepherds = 0 ;
Impl::s_number_workers_per_shepherd = 0 ;
if ( ok_init ) { qthread_finalize(); }
Kokkos::Impl::throw_runtime_exception( msg.str() );
}
Impl::QthreadExec::resize_worker_scratch( 256 , 256 );
// Init the array for used for arbitrarily sized atomics
Impl::init_lock_array_host_space();
}
// //////////////////////////////////////////////////////////////////////////////
int main(int argc,
char *argv[])
{
int count = 0;
aligned_t max = 0;
aligned_t tmp = 0;
assert(qthread_initialize() == 0);
CHECK_VERBOSE();
NUMARG(count, "COUNT");
iprintf("Main executing in team %lu (w/ parent %lu)\n",
(unsigned long)qt_team_id(), (unsigned long)qt_team_parent_id());
assert(qt_team_id() == default_team_id);
assert(qt_team_parent_id() == non_team_id);
aligned_t hello_in_team_ret;
qthread_fork(hello_in_team, NULL, &hello_in_team_ret);
qthread_readFF(&tmp, &hello_in_team_ret);
max = MAX(max, tmp);
aligned_t hello_new_team_rets[count];
for (int i = 0; i < count; i++) {
qthread_fork_new_team(hello_new_team, NULL, &hello_new_team_rets[i]);
}
for (int i = 0; i < count; i++) {
qthread_readFF(&tmp, &hello_new_team_rets[i]);
max = MAX(max, tmp);
}
aligned_t hello_new_team_in_team_ret;
qthread_fork_new_team(
hello_new_team_in_team, NULL, &hello_new_team_in_team_ret);
qthread_readFF(&tmp, &hello_new_team_in_team_ret);
max = MAX(max, tmp);
aligned_t hello_new_team_new_team_ret;
qthread_fork_new_team(
hello_new_team_new_team, NULL, &hello_new_team_new_team_ret);
qthread_readFF(&tmp, &hello_new_team_new_team_ret);
max = MAX(max, tmp);
iprintf("max is %lu\n", (unsigned long)max);
if (count + 4 == max) {
iprintf("SUCCEEDED with count %lu and max team id %lu\n",
(unsigned long)count,
(unsigned long)max);
return 0;
} else {
iprintf("FAILED with count %lu and max team id %lu\n",
(unsigned long)count,
(unsigned long)max);
return 1;
}
}
int main(){
assert(qthread_initialize() == 0);
qt_sinc_t sinc;
qt_sinc_init(&sinc, 0, NULL, NULL, 1);
int ret;
args_t args = { 30, &sinc, &ret };
qthread_fork_copyargs(fib, &args, sizeof(args_t), NULL);
qt_sinc_wait(&sinc, NULL);
printf("%d\n", ret);
}
static void *initializer(void *junk)
{
qthread_initialize();
MACHINE_FENCE;
chpl_qthread_done_initializing = 1;
qthread_syncvar_readFF(NULL, &canexit);
qthread_finalize();
MACHINE_FENCE;
done_finalizing = 1;
return NULL;
}
void run()
{
setenv("QT_NUM_SHEPHERDS",
boost::lexical_cast<std::string>(this->osthreads_).c_str(), 1);
setenv("QT_NUM_WORKERS_PER_SHEPHERD", "1", 1);
qthread_initialize();
// Cold run
//kernel();
// Hot run
results_type results = kernel();
print_results(results);
}
static void *initializer(void *junk)
{
qthread_initialize();
(void) pthread_mutex_lock(&done_init_final_mux); // implicit memory fence
chpl_qthread_done_initializing = 1;
(void) pthread_mutex_unlock(&done_init_final_mux);
qthread_syncvar_readFF(NULL, &canexit);
qthread_finalize();
(void) pthread_mutex_lock(&done_init_final_mux); // implicit memory fence
done_finalizing = 1;
(void) pthread_mutex_unlock(&done_init_final_mux);
return NULL;
}
int main(int argc,
char *argv[])
{
CHECK_VERBOSE();
aligned_t tmp, ret = 0;
int retval;
long foobar = 1234567890;
setenv("QT_MULTINODE","yes",1);
qthread_initialize();
my_id = qthread_multinode_rank();
world_size = qthread_multinode_size();
iprintf("(%03d) Rank %d of %d is alive\n", my_id, my_id, world_size);
retval = qthread_multinode_register(2, returner);
if (retval != 0){
fprintf(stderr, "(%03d) multinode_register returned %d\n", my_id, retval);
return 1;
}
qthread_multinode_run();
if (my_id != 0) return 2;
int target = (world_size > 1) ? 1 : 0;
retval = qthread_fork_remote(returner, &foobar, &ret, target, sizeof(long));
if (retval != 0) {
fprintf(stderr, "(%03d) fork_remote returned %d\n", my_id, retval);
return 3;
}
retval = qthread_readFE(&tmp, &ret);
iprintf("(%03d) returner returned %ld\n", my_id, (long) tmp);
if (retval != 0) {
fprintf(stderr, "(%03d) readFE returned %d (%d)\n", my_id, retval, (int) tmp);
return 4;
}
qthread_finalize();
return (tmp == foobar) ? 0 : 5;
}
int main(int argc,
char *argv[])
{
assert(qthread_initialize() == QTHREAD_SUCCESS);
CHECK_VERBOSE();
NUMARG(numincrs, "NUM_INCRS");
// future_init(128);
iprintf("%i shepherds\n", qthread_num_shepherds());
iprintf("%i threads\n", qthread_num_workers());
qt_loop_balance_sinc(0, numincrs, sum, NULL);
if (threads != numincrs) {
iprintf("threads == %lu, not %lu\n", (unsigned long)threads, (unsigned long)numincrs);
}
assert(threads == numincrs);
return 0;
}
int main(int argc,
char *argv[])
{
qtimer_t t;
assert(qthread_initialize() == QTHREAD_SUCCESS);
CHECK_VERBOSE();
t = qtimer_create();
assert(t);
qtimer_start(t);
qtimer_stop(t);
if (qtimer_secs(t) == 0) {
fprintf(stderr, "qtimer_secs(t) reported zero length time.\n");
} else if (qtimer_secs(t) < 0) {
fprintf(stderr, "qtimer_secs(t) thinks time went backwards (%g).\n",
qtimer_secs(t));
}
iprintf("time to find self and assert it: %g secs\n", qtimer_secs(t));
qtimer_start(t);
qtimer_stop(t);
assert(qtimer_secs(t) >= 0.0);
if (qtimer_secs(t) == 0.0) {
iprintf("inlining reduces calltime to zero (apparently)\n");
} else {
iprintf("smallest measurable time: %g secs\n", qtimer_secs(t));
}
qtimer_destroy(t);
// Now to test fastrand
ks_test();
runs();
autocorrelation();
qthread_finalize();
return 0;
}
/*
* The main procedure simply creates a producer and a consumer task to run in
* parallel
*/
int main(int argc,
char *argv[])
{
aligned_t t[2];
assert(qthread_initialize() == 0);
CHECK_VERBOSE();
NUMARG(bufferSize, "BUFFERSIZE");
numItems = 8 * bufferSize;
NUMARG(numItems, "NUMITEMS");
iprintf("%i threads...\n", qthread_num_shepherds());
buff = malloc(sizeof(aligned_t) * bufferSize);
for (unsigned int i = 0; i < bufferSize; ++i) {
buff[i] = 0;
}
qthread_fork(consumer, NULL, &t[0]);
qthread_fork(producer, NULL, &t[1]);
qthread_readFF(NULL, &t[0]);
qthread_readFF(NULL, &t[1]);
/* cleanup... unnecessary in general, but for the moment I'm tracking down
* errors in the FEB system, so let's clean up */
for (unsigned int i = 0; i < bufferSize; ++i) {
qthread_fill(buff + i);
}
free(buff);
iprintf("Success!\n");
return 0;
}
int main(int argc, char *argv[])
{
int n = 10;
int m = 10;
num_timesteps = 10;
workload = 0;
workload_per = 0;
workload_var = 0;
int print_final = 0;
int alltime = 0;
CHECK_VERBOSE();
NUMARG(n, "N");
NUMARG(m, "M");
NUMARG(num_timesteps, "TIMESTEPS");
NUMARG(workload, "WORKLOAD");
NUMARG(workload_per, "WORKLOAD_PER");
NUMARG(workload_var, "WORKLOAD_VAR");
NUMARG(print_final, "PRINT_FINAL");
NUMARG(alltime, "ALL_TIME");
assert (n > 0 && m > 0);
// Initialize Qthreads
assert(qthread_initialize() == 0);
qtimer_t alloc_timer = qtimer_create();
qtimer_t init_timer = qtimer_create();
qtimer_t exec_timer = qtimer_create();
// Allocate memory for 3-stage stencil (with boundary padding)
qtimer_start(alloc_timer);
stencil_t points;
points.N = n + 2;
points.M = m + 2;
for (int s = 0; s < NUM_STAGES; s++) {
points.stage[s] = malloc(points.N*sizeof(aligned_t *));
assert(NULL != points.stage[s]);
for (int i = 0; i < points.N; i++) {
points.stage[s][i] = calloc(points.M, sizeof(aligned_t));
assert(NULL != points.stage[s][i]);
}
}
qtimer_stop(alloc_timer);
// Initialize first stage and set boundary conditions
qtimer_start(init_timer);
for (int i = 1; i < points.N-1; i++) {
for (int j = 1; j < points.M-1; j++) {
qthread_writeF_const(&points.stage[0][i][j], 0);
for (int s = 1; s < NUM_STAGES; s++)
qthread_empty(&points.stage[s][i][j]);
}
}
for (int i = 0; i < points.N; i++) {
for (int s = 0; s < NUM_STAGES; s++) {
#ifdef BOUNDARY_SYNC
qthread_writeF_const(&points.stage[s][i][0], BOUNDARY);
qthread_writeF_const(&points.stage[s][i][points.M-1], BOUNDARY);
#else
points.stage[s][i][0] = BOUNDARY;
points.stage[s][i][points.M-1] = BOUNDARY;
#endif
}
}
for (int j = 0; j < points.M; j++) {
for (int s = 0; s < NUM_STAGES; s++) {
#ifdef BOUNDARY_SYNC
qthread_writeF_const(&points.stage[s][0][j], BOUNDARY);
qthread_writeF_const(&points.stage[s][points.N-1][j], BOUNDARY);
#else
points.stage[s][0][j] = BOUNDARY;
points.stage[s][points.N-1][j] = BOUNDARY;
#endif
}
}
qtimer_stop(init_timer);
// Create barrier to synchronize on completion of calculations
qtimer_start(exec_timer);
points.barrier = qt_feb_barrier_create(n*m+1);
// Spawn tasks to start calculating updates at each point
update_args_t args = {&points, -1, -1, 1, 1};
for (int i = 1; i < points.N-1; i++) {
for (int j = 1; j < points.M-1; j++) {
args.i = i;
args.j = j;
qthread_fork_syncvar_copyargs(update, &args, sizeof(update_args_t), NULL);
}
}
// Wait for calculations to finish
qt_feb_barrier_enter(points.barrier);
qtimer_stop(exec_timer);
// Print timing info
if (alltime) {
fprintf(stderr, "Allocation time: %f\n", qtimer_secs(alloc_timer));
fprintf(stderr, "Initialization time: %f\n", qtimer_secs(init_timer));
fprintf(stderr, "Execution time: %f\n", qtimer_secs(exec_timer));
} else {
fprintf(stdout, "%f\n", qtimer_secs(exec_timer));
}
// Print stencils
if (print_final) {
size_t final = (num_timesteps % NUM_STAGES);
iprintf("Stage %lu:\n", prev_stage(prev_stage(final)));
print_stage(&points, prev_stage(prev_stage(final)));
iprintf("\nStage %lu:\n", prev_stage(final));
print_stage(&points, prev_stage(final));
iprintf("\nStage %lu:\n", final);
print_stage(&points, final);
}
qt_feb_barrier_destroy(points.barrier);
qtimer_destroy(alloc_timer);
qtimer_destroy(init_timer);
qtimer_destroy(exec_timer);
// Free allocated memory
for (int i = 0; i < points.N; i++) {
free(points.stage[0][i]);
free(points.stage[1][i]);
free(points.stage[2][i]);
}
free(points.stage[0]);
free(points.stage[1]);
free(points.stage[2]);
return 0;
}
int main(int argc,
char *argv[])
{
aligned_t *t[2];
uint64_t x_value;
uint64_t pairs;
assert(qthread_initialize() == 0);
pairs = qthread_num_shepherds() * 6;
CHECK_VERBOSE();
NUMARG(iterations, "ITERATIONS");
NUMARG(pairs, "PAIRS");
t[0] = calloc(pairs, sizeof(aligned_t));
t[1] = calloc(pairs, sizeof(aligned_t));
iprintf("%i threads...\n", qthread_num_shepherds());
iprintf("Initial value of x: %lu\n", (unsigned long)x.u.w);
qthread_syncvar_empty(&id);
qthread_syncvar_writeF_const(&id, 1);
iprintf("id = 0x%lx\n", (unsigned long)id.u.w);
{
uint64_t tmp = 0;
qthread_syncvar_readFF(&tmp, &id);
assert(tmp == 1);
}
iprintf("x's status is: %s (want full (and nowait))\n",
qthread_syncvar_status(&x) ? "full" : "empty");
assert(qthread_syncvar_status(&x) == 1);
qthread_syncvar_readFE(NULL, &x);
iprintf("x's status became: %s (want empty (and nowait))\n",
qthread_syncvar_status(&x) ? "full" : "empty");
assert(qthread_syncvar_status(&x) == 0);
for (unsigned int i = 0; i < pairs; ++i) {
qthread_fork(consumer, (void *)(uintptr_t)i, &(t[0][i]));
}
for (unsigned int i = 0; i < pairs; ++i) {
qthread_fork(producer, (void *)(uintptr_t)(i + pairs), &(t[1][i]));
}
for (unsigned int i = 0; i < pairs; ++i) {
qthread_readFF(NULL, &(t[0][i]));
qthread_readFF(NULL, &(t[1][i]));
}
iprintf("shouldn't be blocking on x (current status: %s)\n",
qthread_syncvar_status(&x) ? "full" : "empty");
qthread_syncvar_fill(&x);
iprintf("shouldn't be blocking on x (current status: %s)\n",
qthread_syncvar_status(&x) ? "full" : "empty");
qthread_syncvar_readFF(&x_value, &x);
assert(qthread_syncvar_status(&x) == 1);
free(t[0]);
free(t[1]);
if (x_value == iterations - 1) {
iprintf("Success! x==%lu\n", (unsigned long)x_value);
return 0;
} else {
fprintf(stderr, "Final value of x=%lu, expected %lu\n",
(unsigned long)x_value, (unsigned long)(iterations - 1));
return -1;
}
}
int main(int argc,
char *argv[])
{
uint64_t total_num_nodes = 0;
qtimer_t timer;
double total_time = 0.0;
CHECK_VERBOSE();
{
unsigned int tmp = (unsigned int)tree_type;
NUMARG(tmp, "UTS_TREE_TYPE");
if (tmp <= BALANCED) {
tree_type = (tree_t)tmp;
} else {
fprintf(stderr, "invalid tree type\n");
return EXIT_FAILURE;
}
tmp = (unsigned int)shape_fn;
NUMARG(tmp, "UTS_SHAPE_FN");
if (tmp <= FIXED) {
shape_fn = (shape_t)tmp;
} else {
fprintf(stderr, "invalid shape function\n");
return EXIT_FAILURE;
}
}
DBLARG(bf_0, "UTS_BF_0");
NUMARG(root_seed, "UTS_ROOT_SEED");
NUMARG(tree_depth, "UTS_TREE_DEPTH");
DBLARG(non_leaf_prob, "UTS_NON_LEAF_PROB");
NUMARG(non_leaf_bf, "UTS_NON_LEAF_NUM");
NUMARG(shift_depth, "UTS_SHIFT_DEPTH");
NUMARG(num_samples, "UTS_NUM_SAMPLES");
// If the operator did not attempt to set a stack size, force
// a reasonable lower bound
if (!getenv("QT_STACK_SIZE") && !getenv("QTHREAD_STACK_SIZE"))
setenv("QT_STACK_SIZE", "32768", 0);
assert(qthread_initialize() == 0);
#ifdef PRINT_STATS
print_stats();
#else
print_banner();
#endif
timer = qtimer_create();
qtimer_start(timer);
node_t root;
root.height = 0;
rng_init(root.state.state, root_seed);
root.num_children = calc_num_children(&root);
aligned_t donecount = 0;
root.dc = &donecount;
qthread_empty(&donecount);
aligned_t tot = 0;
root.acc = &tot;
root.expect = 1;
qthread_fork_syncvar(visit, &root, NULL);
qthread_readFF(NULL, root.dc);
total_num_nodes = tot;
qtimer_stop(timer);
total_time = qtimer_secs(timer);
qtimer_destroy(timer);
#ifdef PRINT_STATS
printf("tree-size %lu\ntree-depth %d\nnum-leaves %llu\nperc-leaves %.2f\n",
(unsigned long)total_num_nodes,
(int)tree_height,
(unsigned long long)num_leaves,
num_leaves / (float)total_num_nodes * 100.0);
printf("exec-time %.3f\ntotal-perf %.0f\npu-perf %.0f\n\n",
total_time,
total_num_nodes / total_time,
total_num_nodes / total_time / qthread_num_workers());
#else
printf("Tree size = %lu, tree depth = %d, num leaves = %llu (%.2f%%)\n",
(unsigned long)total_num_nodes,
(int)tree_height,
(unsigned long long)num_leaves,
num_leaves / (float)total_num_nodes * 100.0);
printf("Wallclock time = %.3f sec, performance = %.0f "
"nodes/sec (%.0f nodes/sec per PE)\n\n",
total_time,
total_num_nodes / total_time,
total_num_nodes / total_time / qthread_num_workers());
#endif /* ifdef PRINT_STATS */
return 0;
}
int main(int argc,
char *argv[])
{
qarray *a;
distribution_t disttypes[] = {
FIXED_HASH, FIXED_FIELDS,
ALL_LOCAL, ALL_RAND, ALL_LEAST,
DIST_RAND, DIST_STRIPES, DIST_FIELDS, DIST_LEAST
};
const char *distnames[] = {
"FIXED_HASH", "FIXED_FIELDS",
"ALL_LOCAL", "ALL_RAND", "ALL_LEAST",
"DIST_RAND", "DIST_STRIPES", "DIST_FIELDS", "DIST_LEAST"
};
unsigned int dt_index;
unsigned int num_dists = sizeof(disttypes) / sizeof(distribution_t);
unsigned int dists = (1 << num_dists) - 1;
qthread_initialize();
CHECK_VERBOSE();
NUMARG(dists, "TEST_DISTS");
NUMARG(ELEMENT_COUNT, "ELEMENT_COUNT");
/* iterate over all the different distribution types */
for (dt_index = 0; dt_index < num_dists; dt_index++) {
if ((dists & (1 << dt_index)) == 0) {
continue;
}
/* test a basic array of doubles */
count = 0;
a = qarray_create_configured(ELEMENT_COUNT, sizeof(double),
disttypes[dt_index], 0, 0);
assert(a);
iprintf("%s: created basic array of doubles\n", distnames[dt_index]);
qarray_iter(a, 0, ELEMENT_COUNT, assign1);
iprintf("%s: iterated; now checking work...\n", distnames[dt_index]);
if (count != ELEMENT_COUNT) {
printf("count = %lu, dt_index = %u\n", (unsigned long)count,
dt_index);
assert(count == ELEMENT_COUNT);
}
{
size_t i;
for (i = 0; i < ELEMENT_COUNT; i++) {
double elem = *(double *)qarray_elem_nomigrate(a, i);
if (elem != 1.0) {
printf
("element %lu is %f instead of 1.0, disttype = %s\n",
(unsigned long)i, elem, distnames[dt_index]);
assert(elem == 1.0);
}
}
}
iprintf("%s: correct result!\n", distnames[dt_index]);
qarray_destroy(a);
/* now test an array of giant things */
count = 0;
a = qarray_create_configured(ELEMENT_COUNT, sizeof(bigobj),
disttypes[dt_index], 0, 0);
iprintf("%s: created array of big objects\n", distnames[dt_index]);
qarray_iter(a, 0, ELEMENT_COUNT, assignall1);
iprintf("%s: iterated; now checking work...\n", distnames[dt_index]);
if (count != ELEMENT_COUNT) {
printf("count = %lu, dt_index = %u\n", (unsigned long)count,
dt_index);
// assert(count == ELEMENT_COUNT);
}
{
size_t i;
char fail = 0;
for (i = 0; i < ELEMENT_COUNT; i++) {
char *elem = (char *)qarray_elem_nomigrate(a, i);
size_t j;
for (j = 0; j < sizeof(bigobj); j++) {
if (elem[j] != 1) {
printf
(
"byte %lu of element %lu is %i instead of 1, dt_index = %u\n",
(unsigned long)j, (unsigned long)i, elem[j],
dt_index);
fail = 1;
break;
}
}
}
assert(fail == 0);
}
iprintf("%s: correct result!\n", distnames[dt_index]);
qarray_destroy(a);
/* now test an array of weird-sized things */
count = 0;
a = qarray_create_configured(ELEMENT_COUNT, sizeof(offsize),
disttypes[dt_index], 0, 0);
iprintf("%s: created array of odd-sized objects\n",
distnames[dt_index]);
qarray_iter_loop(a, 0, ELEMENT_COUNT, assignoff1, NULL);
iprintf("%s: iterated; now checking work...\n", distnames[dt_index]);
if (count != ELEMENT_COUNT) {
printf("count = %lu, dt_index = %u\n", (unsigned long)count,
dt_index);
assert(count == ELEMENT_COUNT);
}
{
size_t i;
for (i = 0; i < ELEMENT_COUNT; i++) {
char *elem = (char *)qarray_elem_nomigrate(a, i);
size_t j;
for (j = 0; j < sizeof(offsize); j++) {
if (elem[j] != 1) {
printf
(
"byte %lu of element %lu is %i instead of 1, dt_index = %u\n",
(unsigned long)j, (unsigned long)i, elem[j],
dt_index);
assert(elem[j] == 1);
}
}
}
}
iprintf("%s: correct result!\n", distnames[dt_index]);
qarray_destroy(a);
}
return 0;
}
int main(int argc,
char *argv[])
{
size_t threads, i;
aligned_t *rets;
qtimer_t t;
unsigned int iter, iterations = 10;
double tot = 0.0;
assert(qthread_initialize() == 0);
t = qtimer_create();
CHECK_VERBOSE();
NUMARG(iterations, "ITERATIONS");
threads = qthread_num_workers();
iprintf("%i shepherds...\n", qthread_num_shepherds());
iprintf("%i threads...\n", (int)threads);
initme = calloc(threads, sizeof(aligned_t));
assert(initme);
rets = malloc(threads * sizeof(aligned_t));
assert(rets);
iprintf("Creating a barrier to block %i threads\n", threads);
wait_on_me = qt_barrier_create(threads, REGION_BARRIER, 0); // all my spawnees plus me
assert(wait_on_me);
for (iter = 0; iter < iterations; iter++) {
iprintf("%i: forking the threads\n", iter);
for (i = 1; i < threads; i++) {
void *arg[2] = {wait_on_me, (void*)(intptr_t)i};
qthread_spawn(barrier_thread, arg, sizeof(void*)*2, rets + i, 0, NULL, i, 0);
}
iprintf("%i: done forking the threads, entering the barrier\n", iter);
qtimer_start(t);
qt_barrier_enter(wait_on_me, 0);
qtimer_stop(t);
iprintf("%i: main thread exited barrier in %f seconds\n", iter, qtimer_secs(t));
tot += qtimer_secs(t);
// reset
initme_idx = 1;
// check retvals
for (i = 1; i < threads; i++) {
qthread_readFF(NULL, rets + i);
if (initme[i] != iter + 1) {
iprintf("initme[%i] = %i (should be %i)\n", (int)i,
(int)initme[i], iter + 1);
}
assert(initme[i] == iter + 1);
}
}
iprintf("Average barrier time = %f\n", tot / iterations);
iprintf("Destroying the barrier...\n");
qt_barrier_destroy(wait_on_me);
iprintf("Success!\n");
return 0;
}
int main(int argc,
char *argv[])
{
aligned_t return_value = 0;
int status, ret;
CHECK_VERBOSE(); // part of the testing harness; toggles iprintf() output
NUMARG(THREADS_ENQUEUED, "THREADS_ENQUEUED");
status = qthread_initialize();
assert(status == QTHREAD_SUCCESS);
iprintf("%i shepherds...\n", qthread_num_shepherds());
iprintf(" %i threads total\n", qthread_num_workers());
iprintf("Creating the queue...\n");
the_queue = qthread_queue_create(QTHREAD_QUEUE_MULTI_JOIN_LENGTH, 0);
assert(the_queue);
iprintf("---------------------------------------------------------\n");
iprintf("\tSINGLE THREAD TEST\n\n");
iprintf("1/4 Spawning thread to be queued...\n");
status = qthread_fork(tobequeued, NULL, &return_value);
assert(status == QTHREAD_SUCCESS);
iprintf("2/4 Waiting for thread to queue itself...\n");
while(qthread_queue_length(the_queue) != 1) qthread_yield();
assert(qthread_readstate(NODE_BUSYNESS) == 1);
iprintf("3/4 Releasing the queue...\n");
qthread_queue_release_all(the_queue);
ret = qthread_readFF(NULL, &return_value);
assert(ret == QTHREAD_SUCCESS);
assert(threads_in == 1);
assert(awoke == 1);
assert(qthread_queue_length(the_queue) == 0);
assert(qthread_readstate(NODE_BUSYNESS) == 1);
iprintf("4/4 Test passed!\n");
iprintf("---------------------------------------------------------\n");
iprintf("\tMULTI THREAD TEST\n\n");
threads_in = 0;
awoke = 0;
aligned_t *retvals = malloc(sizeof(aligned_t) * THREADS_ENQUEUED);
iprintf("1/6 Spawning %u threads to be queued...\n", THREADS_ENQUEUED);
for (int i=0; i<THREADS_ENQUEUED; i++) {
status = qthread_fork(tobequeued, NULL, retvals + i);
assert(status == QTHREAD_SUCCESS);
}
iprintf("2/6 Waiting for %u threads to queue themselves...\n", THREADS_ENQUEUED);
while(qthread_queue_length(the_queue) != THREADS_ENQUEUED) qthread_yield();
assert(threads_in == THREADS_ENQUEUED);
assert(qthread_readstate(NODE_BUSYNESS) == 1);
iprintf("3/6 Releasing a single thread...\n");
qthread_queue_release_one(the_queue);
iprintf("4/6 Waiting for that thread to exit\n");
while (awoke == 0) qthread_yield();
assert(qthread_queue_length(the_queue) == (THREADS_ENQUEUED - 1));
assert(qthread_readstate(NODE_BUSYNESS) == 1);
iprintf("5/6 Releasing the rest of the threads...\n");
qthread_queue_release_all(the_queue);
for (int i=0; i<THREADS_ENQUEUED; i++) {
ret = qthread_readFF(NULL, retvals + i);
assert(ret == QTHREAD_SUCCESS);
}
assert(qthread_queue_length(the_queue) == 0);
assert(qthread_readstate(NODE_BUSYNESS) == 1);
iprintf("6/6 Test passed!\n");
return EXIT_SUCCESS;
}
// //////////////////////////////////////////////////////////////////////////////
int main(int argc,
char *argv[])
{
size_t depth = 3;
assert(qthread_initialize() == 0);
CHECK_VERBOSE();
NUMARG(depth, "TEST_DEPTH");
// Test creating an empty sinc
{
qt_sinc_t zero_sinc;
qt_sinc_init(&zero_sinc, 0, NULL, NULL, 0);
qt_sinc_wait(&zero_sinc, NULL);
qt_sinc_fini(&zero_sinc);
qt_sinc_t *three_sinc = qt_sinc_create(0, NULL, NULL, 0);
qt_sinc_expect(three_sinc, 3);
qthread_fork(submit_to_sinc, three_sinc, NULL);
qthread_fork(submit_to_sinc, three_sinc, NULL);
qthread_fork(submit_to_sinc, three_sinc, NULL);
qt_sinc_wait(three_sinc, NULL);
qt_sinc_destroy(three_sinc);
}
qt_sinc_t *sinc = qt_sinc_create(0, NULL, NULL, 2);
// Spawn additional waits
aligned_t rets[3];
{
qthread_fork(wait_on_sinc, sinc, &rets[0]);
qthread_fork(wait_on_sinc, sinc, &rets[1]);
qthread_fork(wait_on_sinc, sinc, &rets[2]);
}
{
v_args_t args = { depth, sinc };
// These two spawns covered by qt_sinc_create(...,2)
qthread_fork_syncvar_copyargs(visit, &args, sizeof(v_args_t), NULL);
qthread_fork_syncvar_copyargs(visit, &args, sizeof(v_args_t), NULL);
}
qt_sinc_wait(sinc, NULL);
for (int i = 0; i < 3; i++)
qthread_readFF(NULL, &rets[i]);
// Reset the sinc
qt_sinc_reset(sinc, 2);
// Second use
{
v_args_t args = { depth, sinc };
// These two spawns covered by qt_sinc_reset(...,2)
qthread_fork_syncvar_copyargs(visit, &args, sizeof(v_args_t), NULL);
qthread_fork_syncvar_copyargs(visit, &args, sizeof(v_args_t), NULL);
}
qt_sinc_wait(sinc, NULL);
qt_sinc_destroy(sinc);
return 0;
}
int main(int argc,
char *argv[])
{
size_t threads = 1000, i;
aligned_t *rets;
qtimer_t t;
unsigned int iter, iterations = 10;
assert(qthread_initialize() == 0);
t = qtimer_create();
CHECK_VERBOSE();
NUMARG(threads, "THREADS");
NUMARG(iterations, "ITERATIONS");
initme = (aligned_t *)calloc(threads, sizeof(aligned_t));
assert(initme);
rets = (aligned_t *)malloc(iterations * threads * sizeof(aligned_t));
assert(rets);
iprintf("creating the barrier for %zu threads\n", threads + 1);
wait_on_me = qt_feb_barrier_create(threads + 1); // all my spawnees plus me
assert(wait_on_me);
for (iter = 0; iter < iterations; iter++) {
iprintf("%i: forking the threads\n", iter);
for (i = 0; i < threads; i++) {
qthread_fork(barrier_thread, wait_on_me, rets + (iter * threads) + i);
}
iprintf("%i: done forking the threads, entering the barrier\n", iter);
qtimer_start(t);
qt_feb_barrier_enter(wait_on_me);
qtimer_stop(t);
iprintf("%i: main thread exited barrier in %f seconds\n", iter, qtimer_secs(t));
initme_idx = 0;
for (i = 0; i < threads; i++) {
if (initme[i] != iter + 1) {
iprintf("initme[%i] = %i (should be %i)\n", (int)i,
(int)initme[i], iter + 1);
}
assert(initme[i] == iter + 1);
}
}
iprintf("Destroying barrier...\n");
qt_feb_barrier_destroy(wait_on_me);
iprintf("Success!\n");
/* this loop shouldn't be necessary... but seems to avoid crashes in rare
* cases (in other words there must a race condition in qthread_finalize()
* if there are outstanding threads out there) */
for (i = 0; i < threads * 2; i++) {
aligned_t tmp = 1;
qthread_readFF(&tmp, rets + i);
assert(tmp == 0);
}
return 0;
}
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[])
{
aligned_t *ui_array, *ui_array2;
double *d_array, *d_array2;
size_t len = 1000000;
qtimer_t timer = qtimer_create();
double cumulative_time_qutil = 0.0;
double cumulative_time_libc = 0.0;
int using_doubles = 0;
unsigned long iterations = 10;
qthread_initialize();
CHECK_VERBOSE();
printf("%i threads\n", (int)qthread_num_workers());
NUMARG(len, "TEST_LEN");
NUMARG(iterations, "TEST_ITERATIONS");
NUMARG(using_doubles, "TEST_USING_DOUBLES");
printf("using %s\n", using_doubles ? "doubles" : "aligned_ts");
if (using_doubles) {
d_array = calloc(len, sizeof(double));
printf("array is %s\n", human_readable(len * sizeof(double)));
assert(d_array);
// madvise(d_array,len*sizeof(double), MADV_SEQUENTIAL);
for (unsigned int i = 0; i < len; i++) {
d_array[i] = ((double)random()) / ((double)RAND_MAX) + random();
}
d_array2 = calloc(len, sizeof(double));
assert(d_array2);
// madvise(d_array2,len*sizeof(double), MADV_RANDOM);
iprintf("double array generated...\n");
for (unsigned int i = 0; i < iterations; i++) {
memcpy(d_array2, d_array, len * sizeof(double));
qtimer_start(timer);
qutil_qsort(d_array2, len);
qtimer_stop(timer);
cumulative_time_qutil += qtimer_secs(timer);
iprintf("\t%u: sorting %lu doubles with qutil took: %f seconds\n",
i, (unsigned long)len, qtimer_secs(timer));
}
cumulative_time_qutil /= (double)iterations;
printf("sorting %lu doubles with qutil took: %f seconds (avg)\n",
(unsigned long)len, cumulative_time_qutil);
for (unsigned int i = 0; i < iterations; i++) {
memcpy(d_array2, d_array, len * sizeof(double));
qtimer_start(timer);
qsort(d_array2, len, sizeof(double), dcmp);
qtimer_stop(timer);
cumulative_time_libc += qtimer_secs(timer);
iprintf("\t%u: sorting %lu doubles with libc took: %f seconds\n",
i, (unsigned long)len, qtimer_secs(timer));
}
cumulative_time_libc /= (double)iterations;
printf("sorting %lu doubles with libc took: %f seconds\n",
(unsigned long)len, cumulative_time_libc);
free(d_array);
free(d_array2);
} else {
ui_array = calloc(len, sizeof(aligned_t));
printf("array is %s\n", human_readable(len * sizeof(aligned_t)));
for (unsigned int i = 0; i < len; i++) {
ui_array[i] = random();
}
ui_array2 = calloc(len, sizeof(aligned_t));
iprintf("ui_array generated...\n");
for (int i = 0; i < iterations; i++) {
memcpy(ui_array2, ui_array, len * sizeof(aligned_t));
qtimer_start(timer);
qutil_aligned_qsort(ui_array2, len);
qtimer_stop(timer);
cumulative_time_qutil += qtimer_secs(timer);
}
cumulative_time_qutil /= (double)iterations;
printf("sorting %lu aligned_ts with qutil took: %f seconds\n",
(unsigned long)len, cumulative_time_qutil);
for (int i = 0; i < iterations; i++) {
memcpy(ui_array2, ui_array, len * sizeof(aligned_t));
qtimer_start(timer);
qsort(ui_array2, len, sizeof(double), acmp);
qtimer_stop(timer);
cumulative_time_libc += qtimer_secs(timer);
}
cumulative_time_libc /= (double)iterations;
printf("sorting %lu aligned_ts with libc took: %f seconds (avg)\n",
(unsigned long)len, cumulative_time_libc);
free(ui_array);
free(ui_array2);
}
if (cumulative_time_qutil < cumulative_time_libc) {
printf("qutil with %lu threads provides a %0.2fx speedup.\n", (unsigned long)qthread_num_shepherds(), cumulative_time_libc/cumulative_time_qutil);
} else {
printf("qutil with %lu threads provides a %0.2fx slowdown.\n", (unsigned long)qthread_num_shepherds(), cumulative_time_libc/cumulative_time_qutil);
}
qtimer_destroy(timer);
return 0;
}