/*
* Generate all children of the parent
*
* details depend on tree type, node type and shape function
*
*/
void genChildren(Node * parent, void * child_buf, Node * child, StealStack * ss) {
int parentHeight = parent->height;
int numChildren, childType;
ss->maxTreeDepth = max(ss->maxTreeDepth, parent->height);
numChildren = uts_numChildren(parent);
childType = uts_childType(parent);
// record number of children in parent
parent->numChildren = numChildren;
// construct children and push onto stack
if (numChildren > 0) {
int i, j;
child->type = childType;
child->height = parentHeight + 1;
for (i = 0; i < numChildren; i++) {
for (j = 0; j < computeGranularity; j++) {
// TBD: add parent height to spawn
// computeGranularity controls number of rng_spawn calls per node
rng_spawn(parent->state.state, child->state.state, i);
}
ss_put_work(ss, child_buf);
}
} else {
ss->nLeaves++;
}
}
TASK_2(Result, parTreeSearch, int, depth, Node *, parent) {
int numChildren, childType;
counter_t parentHeight = parent->height;
Result r = { depth, 1, 0 };
numChildren = uts_numChildren(parent);
childType = uts_childType(parent);
// record number of children in parent
parent->numChildren = numChildren;
// Recurse on the children
if (numChildren > 0) {
int i, j;
for (i = 0; i < numChildren; i++) {
Node *child = (Node*)alloca(sizeof(Node));
child->type = childType;
child->height = parentHeight + 1;
child->numChildren = -1; // not yet determined
for (j = 0; j < computeGranularity; j++) {
rng_spawn(parent->state.state, child->state.state, i);
}
SPAWN(parTreeSearch, depth+1, child);
}
/* Wait a bit */
struct timespec tim = (struct timespec){0, 100L*numChildren};
nanosleep(&tim, NULL);
for (i = 0; i < numChildren; i++) {
Result c = SYNC(parTreeSearch);
if (c.maxdepth>r.maxdepth) r.maxdepth = c.maxdepth;
r.size += c.size;
r.leaves += c.leaves;
}
} else {
/*
* Generate all children of the parent
*
* details depend on tree type, node type and shape function
*
*/
void genChildren(Node * parent, Node * child) {
int parentHeight = parent->height;
int numChildren, childType;
#ifdef THREAD_METADATA
t_metadata[omp_get_thread_num()].ntasks += 1;
#endif
thread_info[hclib::get_current_worker()].n_nodes++;
numChildren = uts_numChildren(parent);
childType = uts_childType(parent);
// record number of children in parent
parent->numChildren = numChildren;
// construct children and push onto stack
if (numChildren > 0) {
int i, j;
child->type = childType;
child->height = parentHeight + 1;
#ifdef UTS_STAT
if (stats) {
child->pp = parent; // pointer to parent
}
#endif
const unsigned char * parent_state = parent->state.state;
unsigned char * child_state = child->state.state;
for (i = 0; i < numChildren; i++) {
for (j = 0; j < computeGranularity; j++) {
// TBD: add parent height to spawn
// computeGranularity controls number of rng_spawn calls per node
rng_spawn(parent_state, child_state, i);
}
Node parent = *child;
int made_available_for_stealing = 0;
if (hclib::get_current_worker() == 0 && n_buffered_steals < N_BUFFERED_STEALS) {
hclib::shmem_set_lock(&steal_buffer_locks[pe]);
if (n_buffered_steals < N_BUFFERED_STEALS) {
steal_buffer[n_buffered_steals++] = parent;
made_available_for_stealing = 1;
}
hclib::shmem_clear_lock(&steal_buffer_locks[pe]);
}
if (!made_available_for_stealing) {
if (parent.height < 9) {
hclib::async([parent] {
Node child;
initNode(&child);
Node tmp = parent;
genChildren(&tmp, &child);
});
} else {
Node child;
initNode(&child);
genChildren(&parent, &child);
}
}
}
} else {
thread_info[hclib::get_current_worker()].n_leaves++;
}
}
