/* =============================================================================
* net_findAncestors
* -- Contents of bitmapPtr set to 1 if ancestor, else 0
* -- Returns false if id is not root node (i.e., has cycle back id)
* =============================================================================
*/
bool_t
net_findAncestors (net_t* netPtr,
long id,
bitmap_t* ancestorBitmapPtr,
queue_t* workQueuePtr)
{
bool_t status;
vector_t* nodeVectorPtr = LocalLoad(&netPtr->nodeVectorPtr);
assert(LocalLoad(&ancestorBitmapPtr->numBit) == vector_getSize(nodeVectorPtr));
bitmap_clearAll(ancestorBitmapPtr);
queue_clear(workQueuePtr);
{
net_node_t* nodePtr = (net_node_t*)vector_at(nodeVectorPtr, id);
list_t* parentIdListPtr = LocalLoad(&nodePtr->parentIdListPtr);
list_iter_t it;
list_iter_reset(&it, parentIdListPtr);
while (list_iter_hasNext(&it, parentIdListPtr)) {
long parentId = (long)list_iter_next(&it, parentIdListPtr);
status = bitmap_set(ancestorBitmapPtr, parentId);
assert(status);
status = queue_push(workQueuePtr, (void*)parentId);
assert(status);
}
}
while (!queue_isEmpty(workQueuePtr)) {
long parentId = (long)queue_pop(workQueuePtr);
if (parentId == id) {
queue_clear(workQueuePtr);
return FALSE;
}
net_node_t* nodePtr = (net_node_t*)vector_at(nodeVectorPtr, parentId);
list_t* grandParentIdListPtr = LocalLoad(&nodePtr->parentIdListPtr);
list_iter_t it;
list_iter_reset(&it, grandParentIdListPtr);
while (list_iter_hasNext(&it, grandParentIdListPtr)) {
long grandParentId = (long)list_iter_next(&it, grandParentIdListPtr);
if (!bitmap_isSet(ancestorBitmapPtr, grandParentId)) {
status = bitmap_set(ancestorBitmapPtr, grandParentId);
assert(status);
status = queue_push(workQueuePtr, (void*)grandParentId);
assert(status);
}
}
}
return TRUE;
}
/* =============================================================================
* net_findDescendants
* -- Contents of bitmapPtr set to 1 if descendants, else 0
* -- Returns false if id is not root node (i.e., has cycle back id)
* =============================================================================
*/
bool_t
net_findDescendants (net_t* netPtr,
long id,
bitmap_t* descendantBitmapPtr,
queue_t* workQueuePtr)
{
bool_t status;
vector_t* nodeVectorPtr = netPtr->nodeVectorPtr;
assert(descendantBitmapPtr->numBit == vector_getSize(nodeVectorPtr));
bitmap_clearAll(descendantBitmapPtr);
queue_clear(workQueuePtr);
{
net_node_t* nodePtr = (net_node_t*)vector_at(nodeVectorPtr, id);
list_t* childIdListPtr = nodePtr->childIdListPtr;
list_iter_t it;
list_iter_reset(&it, childIdListPtr);
while (list_iter_hasNext(&it, childIdListPtr)) {
long childId = (long)list_iter_next(&it, childIdListPtr);
status = bitmap_set(descendantBitmapPtr, childId);
assert(status);
status = queue_push(workQueuePtr, (void*)childId);
assert(status);
}
}
while (!queue_isEmpty(workQueuePtr)) {
long childId = (long)queue_pop(workQueuePtr);
if (childId == id) {
queue_clear(workQueuePtr);
return FALSE;
}
net_node_t* nodePtr = (net_node_t*)vector_at(nodeVectorPtr, childId);
list_t* grandChildIdListPtr = nodePtr->childIdListPtr;
list_iter_t it;
list_iter_reset(&it, grandChildIdListPtr);
while (list_iter_hasNext(&it, grandChildIdListPtr)) {
long grandChildId = (long)list_iter_next(&it, grandChildIdListPtr);
if (!bitmap_isSet(descendantBitmapPtr, grandChildId)) {
status = bitmap_set(descendantBitmapPtr, grandChildId);
assert(status);
status = queue_push(workQueuePtr, (void*)grandChildId);
assert(status);
}
}
}
return TRUE;
}
/* =============================================================================
* hashtable_iter_next
* =============================================================================
*/
void*
hashtable_iter_next (hashtable_iter_t* itPtr, hashtable_t* hashtablePtr)
{
long bucket;
long numBucket = hashtablePtr->numBucket;
list_t** buckets = hashtablePtr->buckets;
list_iter_t it = itPtr->it;
void* dataPtr = NULL;
for (bucket = itPtr->bucket; bucket < numBucket; /* inside body */) {
list_t* chainPtr = hashtablePtr->buckets[bucket];
if (list_iter_hasNext(&it, chainPtr)) {
pair_t* pairPtr = (pair_t*)list_iter_next(&it, chainPtr);
dataPtr = pairPtr->secondPtr;
break;
}
/* May use dummy bucket; see allocBuckets() */
list_iter_reset(&it, buckets[++bucket]);
}
itPtr->bucket = bucket;
itPtr->it = it;
return dataPtr;
}
static void
printHashtable (hashtable_t* hashtablePtr)
{
long i;
hashtable_iter_t it;
printf("[");
hashtable_iter_reset(&it, hashtablePtr);
while (hashtable_iter_hasNext(&it, hashtablePtr)) {
printf("%li ", *((long*)(hashtable_iter_next(&it, hashtablePtr))));
}
puts("]");
/* Low-level to see structure */
for (i = 0; i < hashtablePtr->numBucket; i++) {
list_iter_t it;
printf("%2li: [", i);
list_iter_reset(&it, hashtablePtr->buckets[i]);
while (list_iter_hasNext(&it, hashtablePtr->buckets[i])) {
void* pairPtr = list_iter_next(&it, hashtablePtr->buckets[i]);
printf("%li ", *(long*)(((pair_t*)pairPtr)->secondPtr));
}
puts("]");
}
}
/* =============================================================================
* isCycle
* =============================================================================
*/
static bool_t
isCycle (vector_t* nodeVectorPtr, net_node_t* nodePtr)
{
switch (nodePtr->mark) {
case NET_NODE_MARK_INIT: {
nodePtr->mark = NET_NODE_MARK_TEST;
list_t* childIdListPtr = nodePtr->childIdListPtr;
list_iter_t it;
list_iter_reset(&it, childIdListPtr);
while (list_iter_hasNext(&it, childIdListPtr)) {
long childId = (long)list_iter_next(&it, childIdListPtr);
net_node_t* childNodePtr =
(net_node_t*)vector_at(nodeVectorPtr, childId);
if (isCycle(nodeVectorPtr, childNodePtr)) {
return TRUE;
}
}
break;
}
case NET_NODE_MARK_TEST:
return TRUE;
case NET_NODE_MARK_DONE:
return FALSE;
break;
default:
assert(0);
}
nodePtr->mark = NET_NODE_MARK_DONE;
return FALSE;
}
static void
printList (list_t* listPtr)
{
list_iter_t it;
printf("[");
list_iter_reset(&it, listPtr);
while (list_iter_hasNext(&it, listPtr)) {
printf("%li ", *((long*)(list_iter_next(&it, listPtr))));
}
puts("]");
}
static void
printTable (table_t* tablePtr)
{
long i;
for (i = 0; i < tablePtr->numBucket; i++) {
list_iter_t it;
printf("%2i: [", i);
list_iter_reset(&it, tablePtr->buckets[i]);
while (list_iter_hasNext(&it, tablePtr->buckets[i])) {
printf("%li ", *(long*)list_iter_next(&it, tablePtr->buckets[i]));
}
puts("]");
}
}
/* =============================================================================
* net_hasEdge
* =============================================================================
*/
bool_t
net_hasEdge (net_t* netPtr, long fromId, long toId)
{
vector_t* nodeVectorPtr = netPtr->nodeVectorPtr;
net_node_t* childNodePtr = (net_node_t*)vector_at(nodeVectorPtr, toId);
list_t* parentIdListPtr = childNodePtr->parentIdListPtr;
list_iter_t it;
list_iter_reset(&it, parentIdListPtr);
while (list_iter_hasNext(&it, parentIdListPtr)) {
long parentId = (long)list_iter_next(&it, parentIdListPtr);
if (parentId == fromId) {
return TRUE;
}
}
return FALSE;
}
/* =============================================================================
* hashtable_iter_hasNext
* =============================================================================
*/
bool_t
hashtable_iter_hasNext (hashtable_iter_t* itPtr, hashtable_t* hashtablePtr)
{
long bucket;
long numBucket = hashtablePtr->numBucket;
list_t** buckets = hashtablePtr->buckets;
list_iter_t it = itPtr->it;
for (bucket = itPtr->bucket; bucket < numBucket; /* inside body */) {
list_t* chainPtr = buckets[bucket];
if (list_iter_hasNext(&it, chainPtr)) {
return TRUE;
}
/* May use dummy bucket; see allocBuckets() */
list_iter_reset(&it, buckets[++bucket]);
}
return FALSE;
}
/* =============================================================================
* net_isPath
* =============================================================================
*/
bool_t
net_isPath (net_t* netPtr,
long fromId,
long toId,
bitmap_t* visitedBitmapPtr,
queue_t* workQueuePtr)
{
bool_t status;
vector_t* nodeVectorPtr = netPtr->nodeVectorPtr;
assert(visitedBitmapPtr->numBit == vector_getSize(nodeVectorPtr));
bitmap_clearAll(visitedBitmapPtr);
queue_clear(workQueuePtr);
status = queue_push(workQueuePtr, (void*)fromId);
assert(status);
while (!queue_isEmpty(workQueuePtr)) {
long id = (long)queue_pop(workQueuePtr);
if (id == toId) {
queue_clear(workQueuePtr);
return TRUE;
}
status = bitmap_set(visitedBitmapPtr, id);
assert(status);
net_node_t* nodePtr = (net_node_t*)vector_at(nodeVectorPtr, id);
list_t* childIdListPtr = nodePtr->childIdListPtr;
list_iter_t it;
list_iter_reset(&it, childIdListPtr);
while (list_iter_hasNext(&it, childIdListPtr)) {
long childId = (long)list_iter_next(&it, childIdListPtr);
if (!bitmap_isSet(visitedBitmapPtr, childId)) {
status = queue_push(workQueuePtr, (void*)childId);
assert(status);
}
}
}
return FALSE;
}
/* =============================================================================
* main
* =============================================================================
*/
MAIN(argc, argv)
{
GOTO_REAL();
/*
* Initialization
*/
parseArgs(argc, (char** const)argv);
long numThread = global_params[PARAM_THREAD];
SIM_GET_NUM_CPU(numThread);
TM_STARTUP(numThread);
P_MEMORY_STARTUP(numThread);
thread_startup(numThread);
maze_t* mazePtr = maze_alloc();
assert(mazePtr);
long numPathToRoute = maze_read(mazePtr, global_inputFile);
router_t* routerPtr = router_alloc(global_params[PARAM_XCOST],
global_params[PARAM_YCOST],
global_params[PARAM_ZCOST],
global_params[PARAM_BENDCOST]);
assert(routerPtr);
list_t* pathVectorListPtr = list_alloc(NULL);
assert(pathVectorListPtr);
/*
* Run transactions
*/
router_solve_arg_t routerArg = {routerPtr, mazePtr, pathVectorListPtr};
TIMER_T startTime;
TIMER_READ(startTime);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
router_solve((void *)&routerArg);
}
#else
thread_start(router_solve, (void*)&routerArg);
#endif
GOTO_REAL();
TIMER_T stopTime;
TIMER_READ(stopTime);
long numPathRouted = 0;
list_iter_t it;
list_iter_reset(&it, pathVectorListPtr);
while (list_iter_hasNext(&it, pathVectorListPtr)) {
vector_t* pathVectorPtr = (vector_t*)list_iter_next(&it, pathVectorListPtr);
numPathRouted += vector_getSize(pathVectorPtr);
}
printf("Paths routed = %li\n", numPathRouted);
printf("Elapsed time = %f seconds\n", TIMER_DIFF_SECONDS(startTime, stopTime));
/*
* Check solution and clean up
*/
assert(numPathRouted <= numPathToRoute);
bool_t status = maze_checkPaths(mazePtr, pathVectorListPtr, global_doPrint);
assert(status == TRUE);
puts("Verification passed.");
maze_free(mazePtr);
router_free(routerPtr);
TM_SHUTDOWN();
P_MEMORY_SHUTDOWN();
GOTO_SIM();
thread_shutdown();
MAIN_RETURN(0);
}
/* =============================================================================
* hashtable_iter_reset
* =============================================================================
*/
void
hashtable_iter_reset (hashtable_iter_t* itPtr, hashtable_t* hashtablePtr)
{
itPtr->bucket = 0;
list_iter_reset(&(itPtr->it), hashtablePtr->buckets[0]);
}
void
prio_iter_reset(prio_queue_t *q)
{
list_iter_reset(q);
}
/* =============================================================================
* TMretriangulate
* -- Returns net amount of elements added to mesh
* =============================================================================
*/
long
TMretriangulate (TM_ARGDECL
element_t* elementPtr,
region_t* regionPtr,
mesh_t* meshPtr,
MAP_T* edgeMapPtr)
{
vector_t* badVectorPtr = regionPtr->badVectorPtr; /* private */
list_t* beforeListPtr = regionPtr->beforeListPtr; /* private */
list_t* borderListPtr = regionPtr->borderListPtr; /* private */
list_iter_t it;
long numDelta = 0L;
assert(edgeMapPtr);
coordinate_t centerCoordinate = element_getNewPoint(elementPtr);
/*
* Remove the old triangles
*/
TMLIST_ITER_RESET(&it, beforeListPtr);
while (TMLIST_ITER_HASNEXT(&it, beforeListPtr)) {
element_t* beforeElementPtr =
(element_t*)TMLIST_ITER_NEXT(&it, beforeListPtr);
TMMESH_REMOVE(meshPtr, beforeElementPtr);
}
numDelta -= PLIST_GETSIZE(beforeListPtr);
/*
* If segment is encroached, split it in half
*/
if (element_getNumEdge(elementPtr) == 1) {
coordinate_t coordinates[2];
edge_t* edgePtr = element_getEdge(elementPtr, 0);
coordinates[0] = centerCoordinate;
coordinates[1] = *(coordinate_t*)(edgePtr->firstPtr);
element_t* aElementPtr = TMELEMENT_ALLOC(coordinates, 2);
assert(aElementPtr);
TMMESH_INSERT(meshPtr, aElementPtr, edgeMapPtr);
coordinates[1] = *(coordinate_t*)(edgePtr->secondPtr);
element_t* bElementPtr = TMELEMENT_ALLOC(coordinates, 2);
assert(bElementPtr);
TMMESH_INSERT(meshPtr, bElementPtr, edgeMapPtr);
bool_t status;
status = TMMESH_REMOVEBOUNDARY(meshPtr, element_getEdge(elementPtr, 0));
assert(status);
status = TMMESH_INSERTBOUNDARY(meshPtr, element_getEdge(aElementPtr, 0));
assert(status);
status = TMMESH_INSERTBOUNDARY(meshPtr, element_getEdge(bElementPtr, 0));
assert(status);
numDelta += 2;
}
/*
* Insert the new triangles. These are contructed using the new
* point and the two points from the border segment.
*/
list_iter_reset(&it, borderListPtr);
while (list_iter_hasNext(&it, borderListPtr)) {
element_t* afterElementPtr;
coordinate_t coordinates[3];
edge_t* borderEdgePtr = (edge_t*)list_iter_next(&it, borderListPtr);
assert(borderEdgePtr);
coordinates[0] = centerCoordinate;
coordinates[1] = *(coordinate_t*)(borderEdgePtr->firstPtr);
coordinates[2] = *(coordinate_t*)(borderEdgePtr->secondPtr);
afterElementPtr = TMELEMENT_ALLOC(coordinates, 3);
assert(afterElementPtr);
TMMESH_INSERT(meshPtr, afterElementPtr, edgeMapPtr);
if (element_isBad(afterElementPtr)) {
TMaddToBadVector(TM_ARG badVectorPtr, afterElementPtr);
}
}
numDelta += PLIST_GETSIZE(borderListPtr);
return numDelta;
}
/* =============================================================================
* main
* =============================================================================
*/
MAIN(argc, argv)
{
/*
* Initialization
*/
parseArgs(argc, (char** const)argv);
long numThread = global_params[PARAM_THREAD];
SIM_GET_NUM_CPU(numThread);
TM_STARTUP(numThread);
P_MEMORY_STARTUP(numThread);
thread_startup(numThread);
maze_t* mazePtr = maze_alloc();
assert(mazePtr);
long numPathToRoute = maze_read(mazePtr, global_inputFile);
router_t* routerPtr = router_alloc(global_params[PARAM_XCOST],
global_params[PARAM_YCOST],
global_params[PARAM_ZCOST],
global_params[PARAM_BENDCOST]);
assert(routerPtr);
list_t* pathVectorListPtr = list_alloc(NULL);
assert(pathVectorListPtr);
/*
* Run transactions
*/
router_solve_arg_t routerArg = {routerPtr, mazePtr, pathVectorListPtr};
// NB: Since ASF/PTLSim "REAL" is native execution, and since we are using
// wallclock time, we want to be sure we read time inside the
// simulator, or else we report native cycles spent on the benchmark
// instead of simulator cycles.
GOTO_SIM();
TIMER_T startTime;
TIMER_READ(startTime);
#ifdef OTM
#pragma omp parallel
{
router_solve((void *)&routerArg);
}
#else
thread_start(router_solve, (void*)&routerArg);
#endif
TIMER_T stopTime;
TIMER_READ(stopTime);
// NB: As above, timer reads must be done inside of the simulated region
// for PTLSim/ASF
GOTO_REAL();
long numPathRouted = 0;
list_iter_t it;
list_iter_reset(&it, pathVectorListPtr);
while (list_iter_hasNext(&it, pathVectorListPtr)) {
vector_t* pathVectorPtr = (vector_t*)list_iter_next(&it, pathVectorListPtr);
numPathRouted += vector_getSize(pathVectorPtr);
}
printf("Paths routed = %li\n", numPathRouted);
printf("Elapsed time = %f seconds\n", TIMER_DIFF_SECONDS(startTime, stopTime));
/*
* Check solution and clean up
*/
assert(numPathRouted <= numPathToRoute);
bool status = maze_checkPaths(mazePtr, pathVectorListPtr, global_doPrint);
assert(status);
puts("Verification passed.");
maze_free(mazePtr);
router_free(routerPtr);
TM_SHUTDOWN();
P_MEMORY_SHUTDOWN();
thread_shutdown();
MAIN_RETURN(0);
}
/* =============================================================================
* maze_checkPaths
* =============================================================================
*/
bool_t
maze_checkPaths (maze_t* mazePtr, list_t* pathVectorListPtr, bool_t doPrintPaths)
{
grid_t* gridPtr = mazePtr->gridPtr;
long width = gridPtr->width;
long height = gridPtr->height;
long depth = gridPtr->depth;
long i;
/* Mark walls */
grid_t* testGridPtr = grid_alloc(width, height, depth);
grid_addPath(testGridPtr, mazePtr->wallVectorPtr);
/* Mark sources */
vector_t* srcVectorPtr = mazePtr->srcVectorPtr;
long numSrc = vector_getSize(srcVectorPtr);
for (i = 0; i < numSrc; i++) {
coordinate_t* srcPtr = (coordinate_t*)vector_at(srcVectorPtr, i);
grid_setPoint(testGridPtr, srcPtr->x, srcPtr->y, srcPtr->z, 0);
}
/* Mark destinations */
vector_t* dstVectorPtr = mazePtr->dstVectorPtr;
long numDst = vector_getSize(dstVectorPtr);
for (i = 0; i < numDst; i++) {
coordinate_t* dstPtr = (coordinate_t*)vector_at(dstVectorPtr, i);
grid_setPoint(testGridPtr, dstPtr->x, dstPtr->y, dstPtr->z, 0);
}
/* Make sure path is contiguous and does not overlap */
long id = 0;
list_iter_t it;
list_iter_reset(&it, pathVectorListPtr);
while (list_iter_hasNext(&it)) {
vector_t* pathVectorPtr = (vector_t*)list_iter_next(&it);
long numPath = vector_getSize(pathVectorPtr);
long i;
for (i = 0; i < numPath; i++) {
id++;
vector_t* pointVectorPtr = (vector_t*)vector_at(pathVectorPtr, i);
/* Check start */
long* prevGridPointPtr = (long*)vector_at(pointVectorPtr, 0);
long x;
long y;
long z;
grid_getPointIndices(gridPtr, prevGridPointPtr, &x, &y, &z);
if (grid_getPoint(testGridPtr, x, y, z) != 0) {
grid_free(testGridPtr);
return FALSE;
}
coordinate_t prevCoordinate;
grid_getPointIndices(gridPtr,
prevGridPointPtr,
&prevCoordinate.x,
&prevCoordinate.y,
&prevCoordinate.z);
long numPoint = vector_getSize(pointVectorPtr);
long j;
for (j = 1; j < (numPoint-1); j++) { /* no need to check endpoints */
long* currGridPointPtr = (long*)vector_at(pointVectorPtr, j);
coordinate_t currCoordinate;
grid_getPointIndices(gridPtr,
currGridPointPtr,
&currCoordinate.x,
&currCoordinate.y,
&currCoordinate.z);
if (!coordinate_areAdjacent(&currCoordinate, &prevCoordinate)) {
grid_free(testGridPtr);
return FALSE;
}
prevCoordinate = currCoordinate;
long x = currCoordinate.x;
long y = currCoordinate.y;
long z = currCoordinate.z;
if (grid_getPoint(testGridPtr, x, y, z) != GRID_POINT_EMPTY) {
grid_free(testGridPtr);
return FALSE;
} else {
grid_setPoint(testGridPtr, x, y, z, id);
}
}
/* Check end */
long* lastGridPointPtr = (long*)vector_at(pointVectorPtr, j);
grid_getPointIndices(gridPtr, lastGridPointPtr, &x, &y, &z);
if (grid_getPoint(testGridPtr, x, y, z) != 0) {
grid_free(testGridPtr);
return FALSE;
}
} /* iteratate over pathVector */
} /* iterate over pathVectorList */
if (doPrintPaths) {
puts("\nRouted Maze:");
grid_print(testGridPtr);
}
grid_free(testGridPtr);
return TRUE;
}
/* =============================================================================
* data_generate
* -- Binary variables of random PDFs
* -- If seed is <0, do not reseed
* -- Returns random network
* =============================================================================
*/
net_t*
data_generate (data_t* dataPtr, long seed, long maxNumParent, long percentParent)
{
random_t* randomPtr = dataPtr->randomPtr;
if (seed >= 0) {
random_seed(randomPtr, seed);
}
/*
* Generate random Bayesian network
*/
long numVar = dataPtr->numVar;
net_t* netPtr = net_alloc(numVar);
assert(netPtr);
net_generateRandomEdges(netPtr, maxNumParent, percentParent, randomPtr);
/*
* Create a threshold for each of the possible permutation of variable
* value instances
*/
long** thresholdsTable = (long**)SEQ_MALLOC(numVar * sizeof(long*));
assert(thresholdsTable);
long v;
for (v = 0; v < numVar; v++) {
list_t* parentIdListPtr = net_getParentIdListPtr(netPtr, v);
long numThreshold = 1 << list_getSize(parentIdListPtr);
long* thresholds = (long*)SEQ_MALLOC(numThreshold * sizeof(long));
assert(thresholds);
long t;
for (t = 0; t < numThreshold; t++) {
long threshold = random_generate(randomPtr) % (DATA_PRECISION + 1);
thresholds[t] = threshold;
}
thresholdsTable[v] = thresholds;
}
/*
* Create variable dependency ordering for record generation
*/
long* order = (long*)SEQ_MALLOC(numVar * sizeof(long));
assert(order);
long numOrder = 0;
queue_t* workQueuePtr = queue_alloc(-1);
assert(workQueuePtr);
vector_t* dependencyVectorPtr = vector_alloc(1);
assert(dependencyVectorPtr);
bitmap_t* orderedBitmapPtr = bitmap_alloc(numVar);
assert(orderedBitmapPtr);
bitmap_clearAll(orderedBitmapPtr);
bitmap_t* doneBitmapPtr = bitmap_alloc(numVar);
assert(doneBitmapPtr);
bitmap_clearAll(doneBitmapPtr);
v = -1;
while ((v = bitmap_findClear(doneBitmapPtr, (v + 1))) >= 0) {
list_t* childIdListPtr = net_getChildIdListPtr(netPtr, v);
long numChild = list_getSize(childIdListPtr);
if (numChild == 0) {
bool status;
/*
* Use breadth-first search to find net connected to this leaf
*/
queue_clear(workQueuePtr);
status = queue_push(workQueuePtr, (void*)v);
assert(status);
while (!queue_isEmpty(workQueuePtr)) {
long id = (long)queue_pop(workQueuePtr);
status = bitmap_set(doneBitmapPtr, id);
assert(status);
status = vector_pushBack(dependencyVectorPtr, (void*)id);
assert(status);
list_t* parentIdListPtr = net_getParentIdListPtr(netPtr, id);
list_iter_t it;
list_iter_reset(&it, parentIdListPtr);
while (list_iter_hasNext(&it, parentIdListPtr)) {
long parentId = (long)list_iter_next(&it, parentIdListPtr);
status = queue_push(workQueuePtr, (void*)parentId);
assert(status);
}
}
/*
* Create ordering
*/
long i;
long n = vector_getSize(dependencyVectorPtr);
for (i = 0; i < n; i++) {
long id = (long)vector_popBack(dependencyVectorPtr);
if (!bitmap_isSet(orderedBitmapPtr, id)) {
bitmap_set(orderedBitmapPtr, id);
order[numOrder++] = id;
}
}
}
}
assert(numOrder == numVar);
/*
* Create records
*/
char* record = dataPtr->records;
long r;
long numRecord = dataPtr->numRecord;
for (r = 0; r < numRecord; r++) {
long o;
for (o = 0; o < numOrder; o++) {
long v = order[o];
list_t* parentIdListPtr = net_getParentIdListPtr(netPtr, v);
long index = 0;
list_iter_t it;
list_iter_reset(&it, parentIdListPtr);
while (list_iter_hasNext(&it, parentIdListPtr)) {
long parentId = (long)list_iter_next(&it, parentIdListPtr);
long value = record[parentId];
assert(value != DATA_INIT);
index = (index << 1) + value;
}
long rnd = random_generate(randomPtr) % DATA_PRECISION;
long threshold = thresholdsTable[v][index];
record[v] = ((rnd < threshold) ? 1 : 0);
}
record += numVar;
assert(record <= (dataPtr->records + numRecord * numVar));
}
/*
* Clean up
*/
bitmap_free(doneBitmapPtr);
bitmap_free(orderedBitmapPtr);
vector_free(dependencyVectorPtr);
queue_free(workQueuePtr);
SEQ_FREE(order);
for (v = 0; v < numVar; v++) {
SEQ_FREE(thresholdsTable[v]);
}
SEQ_FREE(thresholdsTable);
return netPtr;
}
/* =============================================================================
* decoder_process
* =============================================================================
*/
int_error_t
decoder_process (decoder_t* decoderPtr, char* bytes, long numByte)
{
bool_t status;
/*
* Basic error checking
*/
if (numByte < (long)PACKET_HEADER_LENGTH) {
return ERROR_SHORT;
}
packet_t* packetPtr = (packet_t*)bytes;
long flowId = packetPtr->flowId;
long fragmentId = packetPtr->fragmentId;
long numFragment = packetPtr->numFragment;
long length = packetPtr->length;
if (flowId < 0) {
return ERROR_FLOWID;
}
if ((fragmentId < 0) || (fragmentId >= numFragment)) {
return ERROR_FRAGMENTID;
}
if (length < 0) {
return ERROR_LENGTH;
}
#if 0
/*
* With the above checks, this one is redundant
*/
if (numFragment < 1) {
return ERROR_NUMFRAGMENT;
}
#endif
/*
* Add to fragmented map for reassembling
*/
if (numFragment > 1) {
MAP_T* fragmentedMapPtr = decoderPtr->fragmentedMapPtr;
list_t* fragmentListPtr =
(list_t*)MAP_FIND(fragmentedMapPtr, (void*)flowId);
if (fragmentListPtr == NULL) {
fragmentListPtr = list_alloc(&decoder_comparator);
assert(fragmentListPtr);
status = list_insert(fragmentListPtr, (void*)packetPtr);
assert(status);
status = MAP_INSERT(fragmentedMapPtr,
(void*)flowId,
(void*)fragmentListPtr);
assert(status);
} else {
list_iter_t it;
list_iter_reset(&it, fragmentListPtr);
assert(list_iter_hasNext(&it, fragmentListPtr));
packet_t* firstFragmentPtr =
(packet_t*)list_iter_next(&it, fragmentListPtr);
long expectedNumFragment = firstFragmentPtr->numFragment;
if (numFragment != expectedNumFragment) {
status = MAP_REMOVE(fragmentedMapPtr, (void*)flowId);
assert(status);
return ERROR_NUMFRAGMENT;
}
status = list_insert(fragmentListPtr, (void*)packetPtr);
assert(status);
/*
* If we have all the fragments we can reassemble them
*/
if (list_getSize(fragmentListPtr) == numFragment) {
long numByte = 0;
long i = 0;
list_iter_reset(&it, fragmentListPtr);
while (list_iter_hasNext(&it, fragmentListPtr)) {
packet_t* fragmentPtr =
(packet_t*)list_iter_next(&it, fragmentListPtr);
assert(fragmentPtr->flowId == flowId);
if (fragmentPtr->fragmentId != i) {
status = MAP_REMOVE(fragmentedMapPtr, (void*)flowId);
assert(status);
return ERROR_INCOMPLETE; /* should be sequential */
}
numByte += fragmentPtr->length;
i++;
}
char* data = (char*)SEQ_MALLOC(numByte + 1);
assert(data);
data[numByte] = '\0';
char* dst = data;
list_iter_reset(&it, fragmentListPtr);
while (list_iter_hasNext(&it, fragmentListPtr)) {
packet_t* fragmentPtr =
(packet_t*)list_iter_next(&it, fragmentListPtr);
memcpy(dst, (void*)fragmentPtr->data, fragmentPtr->length);
dst += fragmentPtr->length;
}
assert(dst == data + numByte);
decoded_t* decodedPtr = (decoded_t*)SEQ_MALLOC(sizeof(decoded_t));
assert(decodedPtr);
decodedPtr->flowId = flowId;
decodedPtr->data = data;
queue_t* decodedQueuePtr = decoderPtr->decodedQueuePtr;
status = queue_push(decodedQueuePtr, (void*)decodedPtr);
assert(status);
list_free(fragmentListPtr);
status = MAP_REMOVE(fragmentedMapPtr, (void*)flowId);
assert(status);
}
}
} else {
/*
* This is the only fragment, so it is ready
*/
if (fragmentId != 0) {
return ERROR_FRAGMENTID;
}
char* data = (char*)SEQ_MALLOC(length + 1);
assert(data);
data[length] = '\0';
memcpy(data, (void*)packetPtr->data, length);
decoded_t* decodedPtr = (decoded_t*)SEQ_MALLOC(sizeof(decoded_t));
assert(decodedPtr);
decodedPtr->flowId = flowId;
decodedPtr->data = data;
queue_t* decodedQueuePtr = decoderPtr->decodedQueuePtr;
status = queue_push(decodedQueuePtr, (void*)decodedPtr);
assert(status);
}
return ERROR_NONE;
}
/* =============================================================================
* sequencer_run
* =============================================================================
*/
void
sequencer_run (void* argPtr)
{
TM_THREAD_ENTER();
long threadId = thread_getId();
sequencer_t* sequencerPtr = (sequencer_t*)argPtr;
hashtable_t* uniqueSegmentsPtr;
endInfoEntry_t* endInfoEntries;
table_t** startHashToConstructEntryTables;
constructEntry_t* constructEntries;
table_t* hashToConstructEntryTable;
uniqueSegmentsPtr = sequencerPtr->uniqueSegmentsPtr;
endInfoEntries = sequencerPtr->endInfoEntries;
startHashToConstructEntryTables = sequencerPtr->startHashToConstructEntryTables;
constructEntries = sequencerPtr->constructEntries;
hashToConstructEntryTable = sequencerPtr->hashToConstructEntryTable;
segments_t* segmentsPtr = sequencerPtr->segmentsPtr;
assert(segmentsPtr);
vector_t* segmentsContentsPtr = segmentsPtr->contentsPtr;
long numSegment = vector_getSize(segmentsContentsPtr);
long segmentLength = segmentsPtr->length;
long i;
long j;
long i_start;
long i_stop;
long numUniqueSegment;
long substringLength;
long entryIndex;
/*
* Step 1: Remove duplicate segments
*/
// #if defined(HTM) || defined(STM)
long numThread = thread_getNumThread();
{
/* Choose disjoint segments [i_start,i_stop) for each thread */
long partitionSize = (numSegment + numThread/2) / numThread; /* with rounding */
i_start = threadId * partitionSize;
if (threadId == (numThread - 1)) {
i_stop = numSegment;
} else {
i_stop = i_start + partitionSize;
}
}
// #else /* !(HTM || STM) */
// i_start = 0;
// i_stop = numSegment;
// #endif /* !(HTM || STM) */
for (i = i_start; i < i_stop; i+=CHUNK_STEP1) {
TM_BEGIN();
{
long ii;
long ii_stop = MIN(i_stop, (i+CHUNK_STEP1));
for (ii = i; ii < ii_stop; ii++) {
void* segment = vector_at(segmentsContentsPtr, ii);
TMHASHTABLE_INSERT(uniqueSegmentsPtr,
segment,
segment);
} /* ii */
}
TM_END();
}
thread_barrier_wait();
/*
* Step 2a: Iterate over unique segments and compute hashes.
*
* For the gene "atcg", the hashes for the end would be:
*
* "t", "tc", and "tcg"
*
* And for the gene "tcgg", the hashes for the start would be:
*
* "t", "tc", and "tcg"
*
* The names are "end" and "start" because if a matching pair is found,
* they are the substring of the end part of the pair and the start
* part of the pair respectively. In the above example, "tcg" is the
* matching substring so:
*
* (end) (start)
* a[tcg] + [tcg]g = a[tcg]g (overlap = "tcg")
*/
/* uniqueSegmentsPtr is constant now */
numUniqueSegment = hashtable_getSize(uniqueSegmentsPtr);
entryIndex = 0;
// #if defined(HTM) || defined(STM)
{
/* Choose disjoint segments [i_start,i_stop) for each thread */
long num = uniqueSegmentsPtr->numBucket;
long partitionSize = (num + numThread/2) / numThread; /* with rounding */
i_start = threadId * partitionSize;
if (threadId == (numThread - 1)) {
i_stop = num;
} else {
i_stop = i_start + partitionSize;
}
}
{
/* Approximate disjoint segments of element allocation in constructEntries */
long partitionSize = (numUniqueSegment + numThread/2) / numThread; /* with rounding */
entryIndex = threadId * partitionSize;
}
// #else /* !(HTM || STM) */
// i_start = 0;
// i_stop = uniqueSegmentsPtr->numBucket;
// entryIndex = 0;
//#endif /* !(HTM || STM) */
for (i = i_start; i < i_stop; i++) {
list_t* chainPtr = uniqueSegmentsPtr->buckets[i];
list_iter_t it;
list_iter_reset(&it, chainPtr);
while (list_iter_hasNext(&it, chainPtr)) {
char* segment =
(char*)((pair_t*)list_iter_next(&it, chainPtr))->firstPtr;
constructEntry_t* constructEntryPtr;
long j;
ulong_t startHash;
bool_t status;
/* Find an empty constructEntries entry */
TM_BEGIN();
while (((void*)TM_SHARED_READ_P(constructEntries[entryIndex].segment)) != NULL) {
entryIndex = (entryIndex + 1) % numUniqueSegment; /* look for empty */
}
constructEntryPtr = &constructEntries[entryIndex];
TM_SHARED_WRITE_P(constructEntryPtr->segment, segment);
TM_END();
entryIndex = (entryIndex + 1) % numUniqueSegment;
/*
* Save hashes (sdbm algorithm) of segment substrings
*
* endHashes will be computed for shorter substrings after matches
* have been made (in the next phase of the code). This will reduce
* the number of substrings for which hashes need to be computed.
*
* Since we can compute startHashes incrementally, we go ahead
* and compute all of them here.
*/
/* constructEntryPtr is local now */
constructEntryPtr->endHash = (ulong_t)hashString(&segment[1]);
startHash = 0;
for (j = 1; j < segmentLength; j++) {
startHash = (ulong_t)segment[j-1] +
(startHash << 6) + (startHash << 16) - startHash;
TM_BEGIN();
status = TMTABLE_INSERT(startHashToConstructEntryTables[j],
(ulong_t)startHash,
(void*)constructEntryPtr );
TM_END();
assert(status);
}
/*
* For looking up construct entries quickly
*/
startHash = (ulong_t)segment[j-1] +
(startHash << 6) + (startHash << 16) - startHash;
TM_BEGIN();
status = TMTABLE_INSERT(hashToConstructEntryTable,
(ulong_t)startHash,
(void*)constructEntryPtr);
TM_END();
assert(status);
}
}
thread_barrier_wait();
/*
* Step 2b: Match ends to starts by using hash-based string comparison.
*/
for (substringLength = segmentLength-1; substringLength > 0; substringLength--) {
table_t* startHashToConstructEntryTablePtr =
startHashToConstructEntryTables[substringLength];
list_t** buckets = startHashToConstructEntryTablePtr->buckets;
long numBucket = startHashToConstructEntryTablePtr->numBucket;
long index_start;
long index_stop;
// #if defined(HTM) || defined(STM)
{
/* Choose disjoint segments [index_start,index_stop) for each thread */
long partitionSize = (numUniqueSegment + numThread/2) / numThread; /* with rounding */
index_start = threadId * partitionSize;
if (threadId == (numThread - 1)) {
index_stop = numUniqueSegment;
} else {
index_stop = index_start + partitionSize;
}
}
// #else /* !(HTM || STM) */
// index_start = 0;
// index_stop = numUniqueSegment;
//#endif /* !(HTM || STM) */
/* Iterating over disjoint itervals in the range [0, numUniqueSegment) */
for (entryIndex = index_start;
entryIndex < index_stop;
entryIndex += endInfoEntries[entryIndex].jumpToNext)
{
if (!endInfoEntries[entryIndex].isEnd) {
continue;
}
/* ConstructEntries[entryIndex] is local data */
constructEntry_t* endConstructEntryPtr =
&constructEntries[entryIndex];
char* endSegment = endConstructEntryPtr->segment;
ulong_t endHash = endConstructEntryPtr->endHash;
list_t* chainPtr = buckets[endHash % numBucket]; /* buckets: constant data */
list_iter_t it;
list_iter_reset(&it, chainPtr);
/* Linked list at chainPtr is constant */
while (list_iter_hasNext(&it, chainPtr)) {
constructEntry_t* startConstructEntryPtr =
(constructEntry_t*)list_iter_next(&it, chainPtr);
char* startSegment = startConstructEntryPtr->segment;
long newLength = 0;
/* endConstructEntryPtr is local except for properties startPtr/endPtr/length */
TM_BEGIN();
/* Check if matches */
if (TM_SHARED_READ(startConstructEntryPtr->isStart) &&
(TM_SHARED_READ_P(endConstructEntryPtr->startPtr) != startConstructEntryPtr) &&
(strncmp(startSegment,
&endSegment[segmentLength - substringLength],
substringLength) == 0))
{
TM_SHARED_WRITE(startConstructEntryPtr->isStart, FALSE);
constructEntry_t* startConstructEntry_endPtr;
constructEntry_t* endConstructEntry_startPtr;
/* Update endInfo (appended something so no longer end) */
TM_LOCAL_WRITE(endInfoEntries[entryIndex].isEnd, FALSE);
/* Update segment chain construct info */
startConstructEntry_endPtr =
(constructEntry_t*)TM_SHARED_READ_P(startConstructEntryPtr->endPtr);
endConstructEntry_startPtr =
(constructEntry_t*)TM_SHARED_READ_P(endConstructEntryPtr->startPtr);
assert(startConstructEntry_endPtr);
assert(endConstructEntry_startPtr);
TM_SHARED_WRITE_P(startConstructEntry_endPtr->startPtr,
endConstructEntry_startPtr);
TM_LOCAL_WRITE_P(endConstructEntryPtr->nextPtr,
startConstructEntryPtr);
TM_SHARED_WRITE_P(endConstructEntry_startPtr->endPtr,
startConstructEntry_endPtr);
TM_SHARED_WRITE(endConstructEntryPtr->overlap, substringLength);
newLength = (long)TM_SHARED_READ(endConstructEntry_startPtr->length) +
(long)TM_SHARED_READ(startConstructEntryPtr->length) -
substringLength;
TM_SHARED_WRITE(endConstructEntry_startPtr->length, newLength);
} /* if (matched) */
TM_END();
if (!endInfoEntries[entryIndex].isEnd) { /* if there was a match */
break;
}
} /* iterate over chain */
} /* for (endIndex < numUniqueSegment) */
thread_barrier_wait();
/*
* Step 2c: Update jump values and hashes
*
* endHash entries of all remaining ends are updated to the next
* substringLength. Additionally jumpToNext entries are updated such
* that they allow to skip non-end entries. Currently this is sequential
* because parallelization did not perform better.
. */
if (threadId == 0) {
if (substringLength > 1) {
long index = segmentLength - substringLength + 1;
/* initialization if j and i: with i being the next end after j=0 */
for (i = 1; !endInfoEntries[i].isEnd; i+=endInfoEntries[i].jumpToNext) {
/* find first non-null */
}
/* entry 0 is handled seperately from the loop below */
endInfoEntries[0].jumpToNext = i;
if (endInfoEntries[0].isEnd) {
constructEntry_t* constructEntryPtr = &constructEntries[0];
char* segment = constructEntryPtr->segment;
constructEntryPtr->endHash = (ulong_t)hashString(&segment[index]);
}
/* Continue scanning (do not reset i) */
for (j = 0; i < numUniqueSegment; i+=endInfoEntries[i].jumpToNext) {
if (endInfoEntries[i].isEnd) {
constructEntry_t* constructEntryPtr = &constructEntries[i];
char* segment = constructEntryPtr->segment;
constructEntryPtr->endHash = (ulong_t)hashString(&segment[index]);
endInfoEntries[j].jumpToNext = MAX(1, (i - j));
j = i;
}
}
endInfoEntries[j].jumpToNext = i - j;
}
}
thread_barrier_wait();
} /* for (substringLength > 0) */
thread_barrier_wait();
/*
* Step 3: Build sequence string
*/
if (threadId == 0) {
long totalLength = 0;
for (i = 0; i < numUniqueSegment; i++) {
constructEntry_t* constructEntryPtr = &constructEntries[i];
if (constructEntryPtr->isStart) {
totalLength += constructEntryPtr->length;
}
}
sequencerPtr->sequence = (char*)P_MALLOC((totalLength+1) * sizeof(char));
char* sequence = sequencerPtr->sequence;
assert(sequence);
char* copyPtr = sequence;
long sequenceLength = 0;
for (i = 0; i < numUniqueSegment; i++) {
constructEntry_t* constructEntryPtr = &constructEntries[i];
/* If there are several start segments, we append in arbitrary order */
if (constructEntryPtr->isStart) {
long newSequenceLength = sequenceLength + constructEntryPtr->length;
assert( newSequenceLength <= totalLength );
copyPtr = sequence + sequenceLength;
sequenceLength = newSequenceLength;
do {
long numChar = segmentLength - constructEntryPtr->overlap;
if ((copyPtr + numChar) > (sequence + newSequenceLength)) {
TM_PRINT0("ERROR: sequence length != actual length\n");
break;
}
memcpy(copyPtr,
constructEntryPtr->segment,
(numChar * sizeof(char)));
copyPtr += numChar;
} while ((constructEntryPtr = constructEntryPtr->nextPtr) != NULL);
assert(copyPtr <= (sequence + sequenceLength));
}
}
assert(sequence != NULL);
sequence[sequenceLength] = '\0';
}
TM_THREAD_EXIT();
}
/* =============================================================================
* maze_read
* -- Return number of path to route
* =============================================================================
*/
long
maze_read (maze_t* mazePtr, char* inputFileName)
{
FILE* inputFile = fopen(inputFileName, "rt");
if (!inputFile) {
fprintf(stderr, "Error: Could not read %s\n", inputFileName);
exit(1);
}
/*
* Parse input file
*/
long lineNumber = 0;
long height = -1;
long width = -1;
long depth = -1;
char line[256];
list_t* workListPtr = list_alloc(&coordinate_comparePair);
vector_t* wallVectorPtr = mazePtr->wallVectorPtr;
vector_t* srcVectorPtr = mazePtr->srcVectorPtr;
vector_t* dstVectorPtr = mazePtr->dstVectorPtr;
while (fgets(line, sizeof(line), inputFile)) {
char code;
long x1, y1, z1;
long x2, y2, z2;
long numToken = sscanf(line, " %c %li %li %li %li %li %li",
&code, &x1, &y1, &z1, &x2, &y2, &z2);
lineNumber++;
if (numToken < 1) {
continue;
}
switch (code) {
case '#': { /* comment */
/* ignore line */
break;
}
case 'd': { /* dimensions (format: d x y z) */
if (numToken != 4) {
goto PARSE_ERROR;
}
width = x1;
height = y1;
depth = z1;
if (width < 1 || height < 1 || depth < 1) {
goto PARSE_ERROR;
}
break;
}
case 'p': { /* paths (format: p x1 y1 z1 x2 y2 z2) */
if (numToken != 7) {
goto PARSE_ERROR;
}
coordinate_t* srcPtr = coordinate_alloc(x1, y1, z1);
coordinate_t* dstPtr = coordinate_alloc(x2, y2, z2);
assert(srcPtr);
assert(dstPtr);
if (coordinate_isEqual(srcPtr, dstPtr)) {
goto PARSE_ERROR;
}
pair_t* coordinatePairPtr = pair_alloc(srcPtr, dstPtr);
assert(coordinatePairPtr);
bool_t status = list_insert(workListPtr, (void*)coordinatePairPtr);
assert(status == TRUE);
vector_pushBack(srcVectorPtr, (void*)srcPtr);
vector_pushBack(dstVectorPtr, (void*)dstPtr);
break;
}
case 'w': { /* walls (format: w x y z) */
if (numToken != 4) {
goto PARSE_ERROR;
}
coordinate_t* wallPtr = coordinate_alloc(x1, y1, z1);
vector_pushBack(wallVectorPtr, (void*)wallPtr);
break;
}
PARSE_ERROR:
default: { /* error */
fprintf(stderr, "Error: line %li of %s invalid\n",
lineNumber, inputFileName);
exit(1);
}
}
} /* iterate over lines in input file */
fclose(inputFile);
/*
* Initialize grid contents
*/
if (width < 1 || height < 1 || depth < 1) {
fprintf(stderr, "Error: Invalid dimensions (%li, %li, %li)\n",
width, height, depth);
exit(1);
}
grid_t* gridPtr = grid_alloc(width, height, depth);
assert(gridPtr);
mazePtr->gridPtr = gridPtr;
addToGrid(gridPtr, wallVectorPtr, "wall");
addToGrid(gridPtr, srcVectorPtr, "source");
addToGrid(gridPtr, dstVectorPtr, "destination");
printf("Maze dimensions = %li x %li x %li\n", width, height, depth);
printf("Paths to route = %li\n", list_getSize(workListPtr));
/*
* Initialize work queue
*/
queue_t* workQueuePtr = mazePtr->workQueuePtr;
list_iter_t it;
list_iter_reset(&it, workListPtr);
while (list_iter_hasNext(&it)) {
pair_t* coordinatePairPtr = (pair_t*)list_iter_next(&it);
queue_push(workQueuePtr, (void*)coordinatePairPtr);
}
list_free(workListPtr);
return vector_getSize(srcVectorPtr);
}