// allow barrer initization from C
void INTERNAL qt_global_barrier_init(size_t size,
int debug)
{ /*{{{ */
if (MBar == NULL) {
MBar = qt_barrier_create(size, REGION_BARRIER);
assert(MBar);
}
} /*}}} */
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_barrier_create(n*m+1, REGION_BARRIER);
// 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_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_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[])
{
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;
}
