/*
* the producer loops over the requested number of items, writing them to the
* buffer starting at location 0 and wrapping around when it hits the end of
* the buffer. It then writes the value -1 as a sentinel to the next position.
*/
static aligned_t producer(void *arg)
{
for (unsigned int i = 0; i < numItems; ++i) {
const unsigned int buffInd = i % bufferSize;
qthread_writeEF_const(&buff[buffInd], i);
iprintf("producer wrote value #%u\n", i);
}
qthread_writeEF_const(&buff[numItems % bufferSize], -1);
iprintf("producer wrote terminus value #%"PRIu64"\n", -1);
return 0;
}
// Test that writeFF waits for empty var to be filled, writes, and leaves full.
// Requires that only one worker is running. Basically does:
// 1: empty var
// 1: fork(writeFF)
// 1: yields
// 2: starts runnning
// 2: hits writeFF, and yields since var is empty
// 1: writeEF
// 1: hits readFF on forked task and yield
// 2: running again, finishes writeFF, task returns
// 1: readFF competes, finishes
static void testWriteFFWaits(void)
{
aligned_t ret;
concurrent_t=45;
qthread_empty(&concurrent_t);
assert(qthread_num_workers() == 1);
iprintf("1: Forking writeFF wrapper\n");
qthread_fork_to(writeFF_wrapper, NULL, &ret, qthread_shep());
iprintf("1: Forked, now yielding to 2\n");
qthread_yield();
iprintf("1: Back from yield\n");
// verify that writeFF has not completed
assert(qthread_feb_status(&concurrent_t) == 0);
assert(concurrent_t != 55);
iprintf("1: Writing EF\n");
qthread_writeEF_const(&concurrent_t, 35);
// wait for writeFF wrapper to complete
qthread_readFF(NULL, &ret);
// veify that writeFF completed and that FEB is full
iprintf("1: concurrent_t=%d\n", concurrent_t);
assert(qthread_feb_status(&concurrent_t) == 1);
assert(concurrent_t == 55);
}
static aligned_t update(void *arg)
{
stencil_t *points = ((update_args_t *)arg)->points;
size_t i = ((update_args_t *)arg)->i;
size_t j = ((update_args_t *)arg)->j;
size_t this_stage = ((update_args_t *)arg)->stage;
size_t step = ((update_args_t *)arg)->step;
size_t next_stage_id = next_stage(this_stage);
// Perform local work
perform_local_work();
aligned_t **prev = points->stage[prev_stage(this_stage)];
aligned_t sum = *(NORTH(prev, i, j))
+ *(WEST(prev, i, j))
+ *(HERE(prev, i, j))
+ *(EAST(prev, i, j))
+ *(SOUTH(prev, i, j));
// Empty the next stage for this index
qthread_empty(&points->stage[next_stage_id][i][j]);
// Update this point
qthread_writeEF_const(&points->stage[this_stage][i][j], sum/NUM_NEIGHBORS);
if (step < num_timesteps) {
// Spawn next stage
update_args_t args = {points, i, j, next_stage_id, step+1};
#ifdef BOUNDARY_SYNC
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, NUM_NEIGHBORS, NEIGHBORS(points->stage[this_stage],i,j));
#else
if (i == 1) { // North edge
if (j == 1) // West edge: EAST & SOUTH
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 2, EAST(points->stage[this_stage],i,j), SOUTH(points->stage[this_stage],i,j));
else if (j == points->M-2) // East edge: WEST & SOUTH
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 2, WEST(points->stage[this_stage],i,j), SOUTH(points->stage[this_stage],i,j));
else // Interior: WEST & EAST & SOUTH
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 3, WEST(points->stage[this_stage],i,j), EAST(points->stage[this_stage],i,j), SOUTH(points->stage[this_stage],i,j));
} else if (i == points->N-2) { // South edge
if (j == 1) // West edge: NORTH & EAST
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 2, NORTH(points->stage[this_stage],i,j), EAST(points->stage[this_stage],i,j));
else if (j == points->M-2) // East edge: NORTH & WEST
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 2, NORTH(points->stage[this_stage],i,j), WEST(points->stage[this_stage],i,j));
else // Interior: NORTH & WEST & EAST
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 3, NORTH(points->stage[this_stage],i,j), WEST(points->stage[this_stage],i,j), EAST(points->stage[this_stage],i,j));
} else { // Interior
if (j == 1) // West edge: NORTH & EAST & SOUTH
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 3 , NORTH(points->stage[this_stage],i,j), EAST(points->stage[this_stage],i,j), SOUTH(points->stage[this_stage],i,j));
else if (j == points->M-2) // East edge: NORTH & WEST & SOUTH
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 3, NORTH(points->stage[this_stage],i,j), WEST(points->stage[this_stage],i,j), SOUTH(points->stage[this_stage],i,j));
else // Interior: ALL
qthread_fork_copyargs_precond(update, &args, sizeof(update_args_t), NULL, 4, NORTH(points->stage[this_stage],i,j), EAST(points->stage[this_stage],i,j), WEST(points->stage[this_stage],i,j), SOUTH(points->stage[this_stage],i,j));
}
#endif
}
else
qt_feb_barrier_enter(points->barrier);
return 0;
}
template < typename T > inline int qthread_writeEF(T *const dest,
const T src)
{
QTHREAD_CHECKSIZE(T);
return qthread_writeEF_const((aligned_t *)dest,
(aligned_t)src);
}
void mt_write(T& target, T2 val)
{
#ifdef __MTA__
writeef(&target, val);
#elif USING_QTHREADS
qthread_writeEF_const((aligned_t*) &target, (aligned_t) val);
#else
target = val;
#endif
}
