unsigned long long parTreeSearch(int depth, Node *parent, int numChildren)
{
Node *n = (Node *)malloc(numChildren * sizeof(Node));
Node *nodePtr;
int i, j;
unsigned long long subtreesize = 1;
unsigned long long *partialCount = (unsigned long long *)malloc(numChildren * sizeof(unsigned long long));
// Recurse on the children
for (i = 0; i < numChildren; i++) {
nodePtr = &n[i];
nodePtr->height = parent->height + 1;
// The following line is the work (one or more SHA-1 ops)
for (j = 0; j < computeGranularity; j++) {
rng_spawn(parent->state.state, nodePtr->state.state, i);
}
nodePtr->numChildren = uts_numChildren(nodePtr);
hclib_pragma_marker("omp", "task untied firstprivate(i, nodePtr) shared(partialCount)", "pragma221_omp_task");
partialCount[i] = parTreeSearch(depth+1, nodePtr, nodePtr->numChildren);
}
hclib_pragma_marker("omp", "taskwait", "pragma225_omp_taskwait");
for (i = 0; i < numChildren; i++) {
subtreesize += partialCount[i];
}
free(n);
free(partialCount);
return subtreesize;
}
unsigned long long parTreeSearch(int depth, Node *parent, int numChildren)
{
Node n[numChildren], *nodePtr;
int i, j;
unsigned long long subtreesize = 1, partialCount[numChildren];
// Recurse on the children
for (i = 0; i < numChildren; i++) {
nodePtr = &n[i];
nodePtr->height = parent->height + 1;
// The following line is the work (one or more SHA-1 ops)
for (j = 0; j < computeGranularity; j++) {
rng_spawn(parent->state.state, nodePtr->state.state, i);
}
nodePtr->numChildren = uts_numChildren(nodePtr);
#pragma omp task untied firstprivate(i, nodePtr) shared(partialCount)
partialCount[i] = parTreeSearch(depth+1, nodePtr, nodePtr->numChildren);
}
#pragma omp taskwait
for (i = 0; i < numChildren; i++) {
subtreesize += partialCount[i];
}
return subtreesize;
}
int main(int argc, char *argv[]) {
double t1, t2;
Node root;
StealStack *ss;
/* initialize stealstacks and comm. layer */
ss = ss_init(&argc, &argv);
/* determine benchmark parameters */
uts_parseParams(argc, argv);
/* Initialize trace collection structures */
ss_initStats(ss);
/* show parameter settings */
if (ss_get_thread_num() == 0) {
uts_printParams();
}
fflush(NULL);
// Workers will return 1 from ss_start(), all others (managers)
// will return 0 here once the computation ends
if (ss_start(sizeof(Node), chunkSize)) {
/* initialize root node and push on thread 0 stack */
if (ss_get_thread_num() == 0) {
uts_initRoot(&root, type);
#ifdef TRACE
ss_markSteal(ss, 0); // first session is own "parent session"
#endif
ss_put_work(ss, &root);
}
/* time parallel search */
t1 = uts_wctime();
parTreeSearch(ss);
t2 = uts_wctime();
ss->walltime = t2 - t1;
#ifdef TRACE
ss->startTime = t1;
ss->sessionRecords[SS_IDLE][ss->entries[SS_IDLE] - 1].endTime = t2;
#endif
}
ss_stop();
/* display results */
showStats();
ss_finalize();
return 0;
}
counter_t parallel_uts ( Node *root )
{
counter_t num_nodes;
bots_message("Computing Unbalance Tree Search algorithm ");
#pragma omp parallel
#pragma omp single nowait
#pragma omp task untied
num_nodes = parTreeSearch( 0, root, getNumRootChildren(root) );
bots_message(" completed!");
return num_nodes;
}
unsigned long long parallel_uts ( Node *root )
{
unsigned long long num_nodes = 0 ;
root->numChildren = uts_numChildren(root);
bots_message("Computing Unbalance Tree Search algorithm ");
#pragma omp parallel
#pragma omp single nowait
#pragma omp task untied
num_nodes = parTreeSearch( 0, root, root->numChildren );
bots_message(" completed!");
return num_nodes;
}
void parallel_uts( int children, int factor ) {
struct thread_data input;
counter_t num_nodes = 0;
input.depth = 1;
input.numChildren = children;
input.factor = factor;
input.subtree = true;
//printf("main: %d %d %d\n", input.depth, input.numChildren, input.factor);
printf("numChildren: %d, factor: %d, ", input.numChildren, input.factor);
num_nodes = parTreeSearch(&input);
printf("Tree size = %llu\n", (unsigned long long)num_nodes );
}
counter_t parTreeSearch(int depth, Node *parent, int numChildren)
{
//JK
//Node n[numChildren], *nodePtr;
Node *n, *nodePtr;
int i, j;
counter_t subtreesize = 1;
counter_t *partialCount;
//counter_t partialCount[numChildren];
n = (Node*)malloc(numChildren * sizeof(Node));
partialCount = (counter_t*)malloc(numChildren * sizeof(counter_t));
// Recurse on the children
for (i = 0; i < numChildren; i++) {
nodePtr = &n[i];
nodePtr->height = parent->height + 1;
// The following line is the work (one or more SHA-1 ops)
for (j = 0; j < computeGranularity; j++) {
rng_spawn(parent->state.state, nodePtr->state.state, i);
}
nodePtr->numChildren = uts_numChildren(nodePtr);
#pragma omp task firstprivate(i, nodePtr) shared(partialCount) untied
partialCount[i] = parTreeSearch(depth+1, nodePtr, nodePtr->numChildren);
}
#pragma omp taskwait
for (i = 0; i < numChildren; i++) {
subtreesize += partialCount[i];
}
free(n);
free(partialCount);
return subtreesize;
}
unsigned long long parallel_uts ( Node *root )
{
unsigned long long num_nodes = 0 ;
root->numChildren = uts_numChildren(root);
bots_message("Computing Unbalance Tree Search algorithm ");
hclib_pragma_marker("omp_to_hclib", "", "pragma183_omp_to_hclib");
{
hclib_pragma_marker("omp", "parallel", "pragma185_omp_parallel");
{
hclib_pragma_marker("omp", "single nowait", "pragma187_omp_single");
{
hclib_pragma_marker("omp", "task untied", "pragma189_omp_task");
num_nodes = parTreeSearch( 0, root, root->numChildren );
}
}
}
bots_message(" completed!");
return num_nodes;
}
counter_t parTreeSearch( struct thread_data *args )
{
counter_t subtreesize = 1, partialCount[args->numChildren];
struct thread_data temp, input;
int i;
temp.depth = args->depth;
temp.numChildren = args->numChildren;
temp.factor = args->factor;
//printf("par: %d %d %d\n", temp.depth, temp.numChildren, temp.factor);
if(!args->subtree) return subtreesize;
if(args->depth == args->factor) return subtreesize;
int pivot = 0;
time_t t;
srand((unsigned) time(&t));
pivot = rand() % temp.numChildren;
for (i = 0; i < temp.numChildren; i++) {
input.depth = temp.depth+1;
input.numChildren = temp.numChildren;
input.factor = temp.factor;
if(i == pivot) input.subtree = false;
else input.subtree = true;
partialCount[i] = parTreeSearch(&input);
//printf("[%d] %llu\n", i, partialCount[i]);
}
for (i = 0; i < temp.numChildren; i++) {
subtreesize += partialCount[i];
}
return subtreesize;
}
counter_t parTreeSearch(int depth, Node *parent, int numChildren)
{
Node n[numChildren], *nodePtr;
int i, j;
counter_t subtreesize = 1, partialCount[numChildren];
//printf("[p] *** depth = %d ***\n", depth);
//printf("[p] *** height = %d ***\n", parent->height);
//printf("[p] *** numChildren = %d ***\n", parent->numChildren);
// Recurse on the children
for (i = 0; i < numChildren; i++) {
nodePtr = &n[i];
nodePtr->height = parent->height + 1;
// The following line is the work (one or more SHA-1 ops)
for (j = 0; j < computeGranularity; j++) {
rng_spawn(parent->state.state, nodePtr->state.state, i);
}
nodePtr->numChildren = uts_numChildren(nodePtr);
//#pragma omp task firstprivate(i, nodePtr) shared(partialCount) untied
partialCount[i] = parTreeSearch(depth+1, nodePtr, nodePtr->numChildren);
}
//#pragma omp taskwait
for (i = 0; i < numChildren; i++) {
subtreesize += partialCount[i];
}
return subtreesize;
}