/* =============================================================================
* prefix_sums
* =============================================================================
*/
static void
prefix_sums (ULONGINT_T* result, LONGINT_T* input, ULONGINT_T arraySize)
{
long myId = thread_getId();
long numThread = thread_getNumThread();
ULONGINT_T* p = NULL;
if (myId == 0) {
p = (ULONGINT_T*)P_MALLOC(NOSHARE(numThread) * sizeof(ULONGINT_T));
assert(p);
global_p = p;
}
thread_barrier_wait();
p = global_p;
long start;
long end;
long r = arraySize / numThread;
start = myId * r + 1;
end = (myId + 1) * r;
if (myId == (numThread - 1)) {
end = arraySize;
}
ULONGINT_T j;
for (j = start; j < end; j++) {
result[j] = input[j-1] + result[j-1];
}
p[NOSHARE(myId)] = result[end-1];
thread_barrier_wait();
if (myId == 0) {
for (j = 1; j < numThread; j++) {
p[NOSHARE(j)] += p[NOSHARE(j-1)];
}
}
thread_barrier_wait();
if (myId > 0) {
ULONGINT_T add_value = p[NOSHARE(myId-1)];
for (j = start-1; j < end; j++) {
result[j] += add_value;
}
}
thread_barrier_wait();
if (myId == 0) {
P_FREE(p);
}
}
/* =============================================================================
* Pregion_free
* =============================================================================
*/
void
Pregion_free (region_t* regionPtr)
{
PVECTOR_FREE(regionPtr->badVectorPtr);
PLIST_FREE(regionPtr->borderListPtr);
PLIST_FREE(regionPtr->beforeListPtr);
PQUEUE_FREE(regionPtr->expandQueuePtr);
P_FREE(regionPtr);
}
void controller_free()
{
for (long i = 1 ; i < global_numThreads ; i++)
{
SEM_DESTROY(global_metadata[i].semaphore);
}
P_FREE(global_metadata);
global_metadata = NULL;
pthread_join(controllerThread, NULL);
}
/* =============================================================================
* Pqueue_push
* =============================================================================
*/
bool_t
Pqueue_push (queue_t* queuePtr, void* dataPtr)
{
long pop = queuePtr->pop;
long push = queuePtr->push;
long capacity = queuePtr->capacity;
assert(pop != push);
/* Need to resize */
long newPush = (push + 1) % capacity;
if (newPush == pop) {
long newCapacity = capacity * QUEUE_GROWTH_FACTOR;
void** newElements = (void**)P_MALLOC(newCapacity * sizeof(void*));
if (newElements == NULL) {
return FALSE;
}
long dst = 0;
void** elements = queuePtr->elements;
if (pop < push) {
long src;
for (src = (pop + 1); src < push; src++, dst++) {
newElements[dst] = elements[src];
}
} else {
long src;
for (src = (pop + 1); src < capacity; src++, dst++) {
newElements[dst] = elements[src];
}
for (src = 0; src < push; src++, dst++) {
newElements[dst] = elements[src];
}
}
P_FREE(elements);
queuePtr->elements = newElements;
queuePtr->pop = newCapacity - 1;
queuePtr->capacity = newCapacity;
push = dst;
newPush = push + 1; /* no need modulo */
}
queuePtr->elements[push] = dataPtr;
queuePtr->push = newPush;
return TRUE;
}
/* =============================================================================
* Pvector_pushBack
* -- Returns FALSE if fail, else TRUE
* =============================================================================
*/
bool_t
Pvector_pushBack (vector_t* vectorPtr, void* dataPtr)
{
if (vectorPtr->size == vectorPtr->capacity) {
long i;
long newCapacity = vectorPtr->capacity * 2;
void** newElements = (void**)P_MALLOC(newCapacity * sizeof(void*));
if (newElements == NULL) {
return FALSE;
}
vectorPtr->capacity = newCapacity;
for (i = 0; i < vectorPtr->size; i++) {
newElements[i] = vectorPtr->elements[i];
}
P_FREE(vectorPtr->elements);
vectorPtr->elements = newElements;
}
vectorPtr->elements[vectorPtr->size++] = dataPtr;
return TRUE;
}
/* =============================================================================
* Pvector_copy
* =============================================================================
*/
bool_t
Pvector_copy (vector_t* dstVectorPtr, vector_t* srcVectorPtr)
{
long dstCapacity = dstVectorPtr->capacity;
long srcSize = srcVectorPtr->size;
if (dstCapacity < srcSize) {
long srcCapacity = srcVectorPtr->capacity;
void** elements = (void**)P_MALLOC(srcCapacity * sizeof(void*));
if (elements == NULL) {
return FALSE;
}
P_FREE(dstVectorPtr->elements);
dstVectorPtr->elements = elements;
dstVectorPtr->capacity = srcCapacity;
}
memcpy(dstVectorPtr->elements,
srcVectorPtr->elements,
(srcSize * sizeof(void*)));
dstVectorPtr->size = srcSize;
return TRUE;
}
/* =============================================================================
* findSubGraphs2
* =============================================================================
*/
void
findSubGraphs2 (void* argPtr)
{
graph* GPtr = ((findSubGraphs2_arg_t*)argPtr)->GPtr;
Vd* intWtVDList = ((findSubGraphs2_arg_t*)argPtr)->intWtVDList;
Vd* strWtVDList = ((findSubGraphs2_arg_t*)argPtr)->strWtVDList;
edge* maxIntWtList = ((findSubGraphs2_arg_t*)argPtr)->maxIntWtList;
long maxIntWtListSize = ((findSubGraphs2_arg_t*)argPtr)->maxIntWtListSize;
edge* soughtStrWtList = ((findSubGraphs2_arg_t*)argPtr)->soughtStrWtList;
long soughtStrWtListSize = ((findSubGraphs2_arg_t*)argPtr)->soughtStrWtListSize;
long myId = thread_getId();
long numThread = thread_getNumThread();
long numSubArray = 30;
long arraySize = 5*MAX_CLUSTER_SIZE;
long i;
long i_start;
long i_stop;
createPartition(0,
(maxIntWtListSize + soughtStrWtListSize),
myId,
numThread,
&i_start,
&i_stop);
for (i = i_start; i < i_stop; i++) {
if (i < maxIntWtListSize) {
/* Initialize the DS and create one sub-array */
intWtVDList[i].numArrays = 1;
intWtVDList[i].arraySize =
(ULONGINT_T*)P_MALLOC(numSubArray * sizeof(ULONGINT_T));
assert(intWtVDList[i].arraySize);
intWtVDList[i].arraySize[0] = 0;
intWtVDList[i].vList = (V**)P_MALLOC(numSubArray*sizeof(V *));
assert(intWtVDList[i].vList);
intWtVDList[i].vList[0] = (V*)P_MALLOC(arraySize*sizeof(V));
assert(intWtVDList[i].vList[0]);
long j;
for (j = 0; j < arraySize; j++) {
intWtVDList[i].vList[0][j].num = 0;
intWtVDList[i].vList[0][j].depth = 0;
}
} else {
long t = i - maxIntWtListSize;
strWtVDList[t].numArrays = 1;
strWtVDList[t].arraySize =
(ULONGINT_T*)P_MALLOC(numSubArray * sizeof(ULONGINT_T));
assert(strWtVDList[t].arraySize);
strWtVDList[t].arraySize[0] = 0;
strWtVDList[t].vList = (V**)P_MALLOC(numSubArray * sizeof(V*));
assert(strWtVDList[t].vList);
strWtVDList[t].vList[0] = (V*)P_MALLOC(arraySize * sizeof(V));
assert(strWtVDList[t].vList[0]);
long j;
for (j = 0; j < arraySize; j++) {
strWtVDList[t].vList[0][j].num = 0;
strWtVDList[t].vList[0][j].depth = 0;
}
}
}
thread_barrier_wait();
char* visited = (char*)P_MALLOC(GPtr->numVertices * sizeof(char));
assert(visited);
/*
* Each thread runs a BFS from endvertex of maxIntWtList edgeList
*/
for (i = i_start; i < i_stop; i++) {
unsigned long k;
for (k = 0; k < GPtr->numVertices; k++) {
visited[k] = 'u';
}
if (i < maxIntWtListSize) {
intWtVDList[i].vList[0][0].num = maxIntWtList[i].startVertex;
intWtVDList[i].vList[0][0].depth = -1;
intWtVDList[i].vList[0][1].num = maxIntWtList[i].endVertex;
intWtVDList[i].vList[0][1].depth = 1;
intWtVDList[i].arraySize[0] = 2;
visited[intWtVDList[i].vList[0][0].num] = 'v';
visited[intWtVDList[i].vList[0][1].num] = 'v';
long depth = 1;
long verticesVisited = 2;
long currIndex = 1;
while ((depth < SUBGR_EDGE_LENGTH) ||
(verticesVisited == (long)GPtr->numVertices))
{
long currVListX = currIndex / arraySize;
long currVListY = currIndex % arraySize;
V* currV = &intWtVDList[i].vList[currVListX][currVListY];
long vNum = currV->num;
depth = currV->depth + 1;
long j;
long j_start = GPtr->outVertexIndex[vNum];
long j_stop = j_start + GPtr->outDegree[vNum];
for (j = j_start; j < j_stop; j++) {
if (visited[GPtr->outVertexList[j]] == 'u') {
visited[GPtr->outVertexList[j]] = 'v';
long vListX = verticesVisited/arraySize;
long vListY = verticesVisited % arraySize;
V* v = &intWtVDList[i].vList[vListX][vListY];
v->num = GPtr->outVertexList[j];
v->depth = depth;
intWtVDList[i].arraySize[vListX]++;
verticesVisited++;
}
}
/* Check if we need to create a new array */
if (((float) verticesVisited / (float) arraySize) > 0.5) {
/* create a new sub-array */
if (intWtVDList[i].numArrays !=
(unsigned long)(verticesVisited/arraySize + 2))
{
intWtVDList[i].numArrays++;
intWtVDList[i].vList[intWtVDList[i].numArrays-1] =
(V*)P_MALLOC(arraySize * sizeof(V));
assert(intWtVDList[i].vList[intWtVDList[i].numArrays-1]);
intWtVDList[i].arraySize[intWtVDList[i].numArrays-1] = 0;
}
}
if ((currIndex < verticesVisited - 1) &&
(verticesVisited < (long)GPtr->numVertices))
{
currIndex++;
long vListX = currIndex / arraySize;
long vListY = currIndex % arraySize;
depth = intWtVDList[i].vList[vListX][vListY].depth;
} else {
break;
}
}
} else {
long t = i - maxIntWtListSize;
strWtVDList[t].vList[0][0].num = soughtStrWtList[t].startVertex;
strWtVDList[t].vList[0][0].depth = -1;
strWtVDList[t].vList[0][1].num = soughtStrWtList[t].endVertex;
strWtVDList[t].vList[0][1].depth = 1;
strWtVDList[t].arraySize[0] = 2;
visited[strWtVDList[t].vList[0][0].num] = 'v';
visited[strWtVDList[t].vList[0][1].num] = 'v';
long depth = 1;
long verticesVisited = 2;
long currIndex = 1;
while ((depth < SUBGR_EDGE_LENGTH) ||
(verticesVisited == (long)GPtr->numVertices))
{
long currVListX = currIndex / arraySize;
long currVListY = currIndex % arraySize;
V* currV = &strWtVDList[t].vList[currVListX][currVListY];
long vNum = currV->num;
depth = currV->depth + 1;
long j;
long j_start = GPtr->outVertexIndex[vNum];
long j_stop = j_start + GPtr->outDegree[vNum];
for (j = j_start; j < j_stop; j++) {
if (visited[GPtr->outVertexList[j]] == 'u') {
visited[GPtr->outVertexList[j]] = 'v';
long vListX = verticesVisited / arraySize;
long vListY = verticesVisited % arraySize;
V* v = &strWtVDList[t].vList[vListX][vListY];
v->num = GPtr->outVertexList[j];
v->depth = depth;
strWtVDList[t].arraySize[vListX]++;
verticesVisited++;
}
}
/* Check if we need to create a new array */
if (((float)verticesVisited /(float) arraySize) > 0.5) {
/* create a new sub-array */
if (strWtVDList[t].numArrays !=
(unsigned long)(verticesVisited/arraySize + 2))
{
strWtVDList[t].numArrays++;
strWtVDList[t].vList[strWtVDList[t].numArrays-1] =
(V*)P_MALLOC(arraySize * sizeof(V));
assert(strWtVDList[t].vList[strWtVDList[t].numArrays-1]);
strWtVDList[t].arraySize[strWtVDList[t].numArrays-1] = 0;
}
}
if ((currIndex < verticesVisited - 1) &&
((unsigned long)verticesVisited < GPtr->numVertices))
{
currIndex++;
long currVListX = currIndex / arraySize;
long currVListY = currIndex % arraySize;
depth = strWtVDList[t].vList[currVListX][currVListY].depth;
} else {
break;
}
}
}
} /* for i */
P_FREE(visited);
}
/* =============================================================================
* genScalData
* =============================================================================
*/
void
genScalData (void* argPtr)
{
TM_THREAD_ENTER();
graphSDG* SDGdataPtr = (graphSDG*)argPtr;
long myId = thread_getId();
long numThread = thread_getNumThread();
/*
* STEP 0: Create the permutations required to randomize the vertices
*/
random_t* stream = PRANDOM_ALLOC();
assert(stream);
PRANDOM_SEED(stream, myId);
ULONGINT_T* permV; /* the vars associated with the graph tuple */
if (myId == 0) {
permV = (ULONGINT_T*)P_MALLOC(TOT_VERTICES * sizeof(ULONGINT_T));
assert(permV);
global_permV = permV;
}
thread_barrier_wait();
permV = global_permV;
long i;
long i_start;
long i_stop;
createPartition(0, TOT_VERTICES, myId, numThread, &i_start, &i_stop);
/* Initialize the array */
for (i = i_start; i < i_stop; i++) {
permV[i] = i;
}
thread_barrier_wait();
for (i = i_start; i < i_stop; i++) {
long t1 = PRANDOM_GENERATE(stream);
long t = i + t1 % (TOT_VERTICES - i);
if (t != i) {
AL_LOCK(0);
TM_BEGIN();
long t2 = (long)TM_SHARED_READ(permV[t]);
TM_SHARED_WRITE(permV[t], TM_SHARED_READ(permV[i]));
TM_SHARED_WRITE(permV[i], t2);
TM_END();
}
}
/*
* STEP 1: Create Cliques
*/
long* cliqueSizes;
long estTotCliques = ceil(1.5 * TOT_VERTICES / ((1+MAX_CLIQUE_SIZE)/2));
/*
* Allocate mem for Clique array
* Estimate number of clique required and pad by 50%
*/
if (myId == 0) {
cliqueSizes = (long*)P_MALLOC(estTotCliques * sizeof(long));
assert(cliqueSizes);
global_cliqueSizes = cliqueSizes;
}
thread_barrier_wait();
cliqueSizes = global_cliqueSizes;
createPartition(0, estTotCliques, myId, numThread, &i_start, &i_stop);
/* Generate random clique sizes. */
for (i = i_start; i < i_stop; i++) {
cliqueSizes[i] = 1 + (PRANDOM_GENERATE(stream) % MAX_CLIQUE_SIZE);
}
thread_barrier_wait();
long totCliques = 0;
/*
* Allocate memory for cliqueList
*/
ULONGINT_T* lastVsInCliques;
ULONGINT_T* firstVsInCliques;
if (myId == 0) {
lastVsInCliques = (ULONGINT_T*)P_MALLOC(estTotCliques * sizeof(ULONGINT_T));
assert(lastVsInCliques);
global_lastVsInCliques = lastVsInCliques;
firstVsInCliques = (ULONGINT_T*)P_MALLOC(estTotCliques * sizeof(ULONGINT_T));
assert(firstVsInCliques);
global_firstVsInCliques = firstVsInCliques;
/*
* Sum up vertices in each clique to determine the lastVsInCliques array
*/
lastVsInCliques[0] = cliqueSizes[0] - 1;
for (i = 1; i < estTotCliques; i++) {
lastVsInCliques[i] = cliqueSizes[i] + lastVsInCliques[i-1];
if (lastVsInCliques[i] >= TOT_VERTICES-1) {
break;
}
}
totCliques = i + 1;
global_totCliques = totCliques;
/*
* Fix the size of the last clique
*/
cliqueSizes[totCliques-1] =
TOT_VERTICES - lastVsInCliques[totCliques-2] - 1;
lastVsInCliques[totCliques-1] = TOT_VERTICES - 1;
firstVsInCliques[0] = 0;
}
thread_barrier_wait();
lastVsInCliques = global_lastVsInCliques;
firstVsInCliques = global_firstVsInCliques;
totCliques = global_totCliques;
/* Compute start Vertices in cliques. */
createPartition(1, totCliques, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
firstVsInCliques[i] = lastVsInCliques[i-1] + 1;
}
#ifdef WRITE_RESULT_FILES
thread_barrier_wait();
/* Write the generated cliques to file for comparison with Kernel 4 */
if (myId == 0) {
FILE* outfp = fopen("cliques.txt", "w");
fprintf(outfp, "No. of cliques - %lu\n", totCliques);
for (i = 0; i < totCliques; i++) {
fprintf(outfp, "Clq %lu - ", i);
long j;
for (j = firstVsInCliques[i]; j <= lastVsInCliques[i]; j++) {
fprintf(outfp, "%lu ", permV[j]);
}
fprintf(outfp, "\n");
}
fclose(outfp);
}
thread_barrier_wait();
#endif
/*
* STEP 2: Create the edges within the cliques
*/
/*
* Estimate number of edges - using an empirical measure
*/
long estTotEdges;
if (SCALE >= 12) {
estTotEdges = ceil(((MAX_CLIQUE_SIZE-1) * TOT_VERTICES));
} else {
estTotEdges = ceil(1.2 * (((MAX_CLIQUE_SIZE-1)*TOT_VERTICES)
* ((1 + MAX_PARAL_EDGES)/2) + TOT_VERTICES*2));
}
/*
* Initialize edge counter
*/
long i_edgePtr = 0;
float p = PROB_UNIDIRECTIONAL;
/*
* Partial edgeLists
*/
ULONGINT_T* startV;
ULONGINT_T* endV;
if (numThread > 3) {
long numByte = 1.5 * (estTotEdges/numThread) * sizeof(ULONGINT_T);
startV = (ULONGINT_T*)P_MALLOC(numByte);
endV = (ULONGINT_T*)P_MALLOC(numByte);
} else {
long numByte = (estTotEdges/numThread) * sizeof(ULONGINT_T);
startV = (ULONGINT_T*)P_MALLOC(numByte);
endV = (ULONGINT_T*)P_MALLOC(numByte);
}
assert(startV);
assert(endV);
/*
* Tmp array to keep track of the no. of parallel edges in each direction
*/
ULONGINT_T** tmpEdgeCounter =
(ULONGINT_T**)P_MALLOC(MAX_CLIQUE_SIZE * sizeof(ULONGINT_T *));
assert(tmpEdgeCounter);
for (i = 0; i < MAX_CLIQUE_SIZE; i++) {
tmpEdgeCounter[i] =
(ULONGINT_T*)P_MALLOC(MAX_CLIQUE_SIZE * sizeof(ULONGINT_T));
assert(tmpEdgeCounter[i]);
}
/*
* Create edges in parallel
*/
long i_clique;
createPartition(0, totCliques, myId, numThread, &i_start, &i_stop);
for (i_clique = i_start; i_clique < i_stop; i_clique++) {
/*
* Get current clique parameters
*/
long i_cliqueSize = cliqueSizes[i_clique];
long i_firstVsInClique = firstVsInCliques[i_clique];
/*
* First create at least one edge between two vetices in a clique
*/
for (i = 0; i < i_cliqueSize; i++) {
long j;
for (j = 0; j < i; j++) {
float r = (float)(PRANDOM_GENERATE(stream) % 1000) / (float)1000;
if (r >= p) {
startV[i_edgePtr] = i + i_firstVsInClique;
endV[i_edgePtr] = j + i_firstVsInClique;
i_edgePtr++;
tmpEdgeCounter[i][j] = 1;
startV[i_edgePtr] = j + i_firstVsInClique;
endV[i_edgePtr] = i + i_firstVsInClique;
i_edgePtr++;
tmpEdgeCounter[j][i] = 1;
} else if (r >= 0.5) {
startV[i_edgePtr] = i + i_firstVsInClique;
endV[i_edgePtr] = j + i_firstVsInClique;
i_edgePtr++;
tmpEdgeCounter[i][j] = 1;
tmpEdgeCounter[j][i] = 0;
} else {
startV[i_edgePtr] = j + i_firstVsInClique;
endV[i_edgePtr] = i + i_firstVsInClique;
i_edgePtr++;
tmpEdgeCounter[j][i] = 1;
tmpEdgeCounter[i][j] = 0;
}
} /* for j */
} /* for i */
if (i_cliqueSize != 1) {
long randNumEdges = (long)(PRANDOM_GENERATE(stream)
% (2*i_cliqueSize*MAX_PARAL_EDGES));
long i_paralEdge;
for (i_paralEdge = 0; i_paralEdge < randNumEdges; i_paralEdge++) {
i = (PRANDOM_GENERATE(stream) % i_cliqueSize);
long j = (PRANDOM_GENERATE(stream) % i_cliqueSize);
if ((i != j) && (tmpEdgeCounter[i][j] < MAX_PARAL_EDGES)) {
float r = (float)(PRANDOM_GENERATE(stream) % 1000) / (float)1000;
if (r >= p) {
/* Copy to edge structure. */
startV[i_edgePtr] = i + i_firstVsInClique;
endV[i_edgePtr] = j + i_firstVsInClique;
i_edgePtr++;
tmpEdgeCounter[i][j]++;
}
}
}
}
} /* for i_clique */
for (i = 0; i < MAX_CLIQUE_SIZE; i++) {
P_FREE(tmpEdgeCounter[i]);
}
P_FREE(tmpEdgeCounter);
/*
* Merge partial edge lists
*/
ULONGINT_T* i_edgeStartCounter;
ULONGINT_T* i_edgeEndCounter;
if (myId == 0) {
i_edgeStartCounter = (ULONGINT_T*)P_MALLOC(numThread * sizeof(ULONGINT_T));
assert(i_edgeStartCounter);
global_i_edgeStartCounter = i_edgeStartCounter;
i_edgeEndCounter = (ULONGINT_T*)P_MALLOC(numThread * sizeof(ULONGINT_T));
assert(i_edgeEndCounter);
global_i_edgeEndCounter = i_edgeEndCounter;
}
thread_barrier_wait();
i_edgeStartCounter = global_i_edgeStartCounter;
i_edgeEndCounter = global_i_edgeEndCounter;
i_edgeEndCounter[myId] = i_edgePtr;
i_edgeStartCounter[myId] = 0;
thread_barrier_wait();
if (myId == 0) {
for (i = 1; i < numThread; i++) {
i_edgeEndCounter[i] = i_edgeEndCounter[i-1] + i_edgeEndCounter[i];
i_edgeStartCounter[i] = i_edgeEndCounter[i-1];
}
}
AL_LOCK(0);
TM_BEGIN();
TM_SHARED_WRITE(global_edgeNum,
((long)TM_SHARED_READ(global_edgeNum) + i_edgePtr));
TM_END();
thread_barrier_wait();
long edgeNum = global_edgeNum;
/*
* Initialize edge list arrays
*/
ULONGINT_T* startVertex;
ULONGINT_T* endVertex;
if (myId == 0) {
if (SCALE < 10) {
long numByte = 2 * edgeNum * sizeof(ULONGINT_T);
startVertex = (ULONGINT_T*)P_MALLOC(numByte);
endVertex = (ULONGINT_T*)P_MALLOC(numByte);
} else {
long numByte = (edgeNum + MAX_PARAL_EDGES * TOT_VERTICES)
* sizeof(ULONGINT_T);
startVertex = (ULONGINT_T*)P_MALLOC(numByte);
endVertex = (ULONGINT_T*)P_MALLOC(numByte);
}
assert(startVertex);
assert(endVertex);
global_startVertex = startVertex;
global_endVertex = endVertex;
}
thread_barrier_wait();
startVertex = global_startVertex;
endVertex = global_endVertex;
for (i = i_edgeStartCounter[myId]; i < i_edgeEndCounter[myId]; i++) {
startVertex[i] = startV[i-i_edgeStartCounter[myId]];
endVertex[i] = endV[i-i_edgeStartCounter[myId]];
}
ULONGINT_T numEdgesPlacedInCliques = edgeNum;
thread_barrier_wait();
/*
* STEP 3: Connect the cliques
*/
i_edgePtr = 0;
p = PROB_INTERCL_EDGES;
/*
* Generating inter-clique edges as given in the specs
*/
createPartition(0, TOT_VERTICES, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
ULONGINT_T tempVertex1 = i;
long h = totCliques;
long l = 0;
long t = -1;
while (h - l > 1) {
long m = (h + l) / 2;
if (tempVertex1 >= firstVsInCliques[m]) {
l = m;
} else {
if ((tempVertex1 < firstVsInCliques[m]) && (m > 0)) {
if (tempVertex1 >= firstVsInCliques[m-1]) {
t = m - 1;
break;
} else {
h = m;
}
}
}
}
if (t == -1) {
long m;
for (m = (l + 1); m < h; m++) {
if (tempVertex1<firstVsInCliques[m]) {
break;
}
}
t = m-1;
}
long t1 = firstVsInCliques[t];
ULONGINT_T d;
for (d = 1, p = PROB_INTERCL_EDGES; d < TOT_VERTICES; d *= 2, p /= 2) {
float r = (float)(PRANDOM_GENERATE(stream) % 1000) / (float)1000;
if (r <= p) {
ULONGINT_T tempVertex2 = (i+d) % TOT_VERTICES;
h = totCliques;
l = 0;
t = -1;
while (h - l > 1) {
long m = (h + l) / 2;
if (tempVertex2 >= firstVsInCliques[m]) {
l = m;
} else {
if ((tempVertex2 < firstVsInCliques[m]) && (m > 0)) {
if (firstVsInCliques[m-1] <= tempVertex2) {
t = m - 1;
break;
} else {
h = m;
}
}
}
}
if (t == -1) {
long m;
for (m = (l + 1); m < h; m++) {
if (tempVertex2 < firstVsInCliques[m]) {
break;
}
}
t = m - 1;
}
long t2 = firstVsInCliques[t];
if (t1 != t2) {
long randNumEdges =
PRANDOM_GENERATE(stream) % MAX_PARAL_EDGES + 1;
long j;
for (j = 0; j < randNumEdges; j++) {
startV[i_edgePtr] = tempVertex1;
endV[i_edgePtr] = tempVertex2;
i_edgePtr++;
}
}
} /* r <= p */
float r0 = (float)(PRANDOM_GENERATE(stream) % 1000) / (float)1000;
if ((r0 <= p) && (i-d>=0)) {
ULONGINT_T tempVertex2 = (i-d) % TOT_VERTICES;
h = totCliques;
l = 0;
t = -1;
while (h - l > 1) {
long m = (h + l) / 2;
if (tempVertex2 >= firstVsInCliques[m]) {
l = m;
} else {
if ((tempVertex2 < firstVsInCliques[m]) && (m > 0)) {
if (firstVsInCliques[m-1] <= tempVertex2) {
t = m - 1;
break;
} else {
h = m;
}
}
}
}
if (t == -1) {
long m;
for (m = (l + 1); m < h; m++) {
if (tempVertex2 < firstVsInCliques[m]) {
break;
}
}
t = m - 1;
}
long t2 = firstVsInCliques[t];
if (t1 != t2) {
long randNumEdges =
PRANDOM_GENERATE(stream) % MAX_PARAL_EDGES + 1;
long j;
for (j = 0; j < randNumEdges; j++) {
startV[i_edgePtr] = tempVertex1;
endV[i_edgePtr] = tempVertex2;
i_edgePtr++;
}
}
} /* r0 <= p && (i-d) > 0 */
} /* for d, p */
} /* for i */
i_edgeEndCounter[myId] = i_edgePtr;
i_edgeStartCounter[myId] = 0;
if (myId == 0) {
global_edgeNum = 0;
}
thread_barrier_wait();
if (myId == 0) {
for (i = 1; i < numThread; i++) {
i_edgeEndCounter[i] = i_edgeEndCounter[i-1] + i_edgeEndCounter[i];
i_edgeStartCounter[i] = i_edgeEndCounter[i-1];
}
}
AL_LOCK(0);
TM_BEGIN();
TM_SHARED_WRITE(global_edgeNum,
((long)TM_SHARED_READ(global_edgeNum) + i_edgePtr));
TM_END();
thread_barrier_wait();
edgeNum = global_edgeNum;
ULONGINT_T numEdgesPlacedOutside = global_edgeNum;
for (i = i_edgeStartCounter[myId]; i < i_edgeEndCounter[myId]; i++) {
startVertex[i+numEdgesPlacedInCliques] = startV[i-i_edgeStartCounter[myId]];
endVertex[i+numEdgesPlacedInCliques] = endV[i-i_edgeStartCounter[myId]];
}
thread_barrier_wait();
ULONGINT_T numEdgesPlaced = numEdgesPlacedInCliques + numEdgesPlacedOutside;
if (myId == 0) {
SDGdataPtr->numEdgesPlaced = numEdgesPlaced;
printf("Finished generating edges\n");
printf("No. of intra-clique edges - %lu\n", numEdgesPlacedInCliques);
printf("No. of inter-clique edges - %lu\n", numEdgesPlacedOutside);
printf("Total no. of edges - %lu\n", numEdgesPlaced);
P_FREE(i_edgeStartCounter);
P_FREE(i_edgeEndCounter);
P_FREE(cliqueSizes);
P_FREE(firstVsInCliques);
P_FREE(lastVsInCliques);
}
thread_barrier_wait();
P_FREE(startV);
P_FREE(endV);
/*
* STEP 4: Generate edge weights
*/
if (myId == 0) {
SDGdataPtr->intWeight =
(LONGINT_T*)P_MALLOC(numEdgesPlaced * sizeof(LONGINT_T));
assert(SDGdataPtr->intWeight);
}
thread_barrier_wait();
p = PERC_INT_WEIGHTS;
ULONGINT_T numStrWtEdges = 0;
createPartition(0, numEdgesPlaced, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
float r = (float)(PRANDOM_GENERATE(stream) % 1000) / (float)1000;
if (r <= p) {
SDGdataPtr->intWeight[i] =
1 + (PRANDOM_GENERATE(stream) % (MAX_INT_WEIGHT-1));
} else {
SDGdataPtr->intWeight[i] = -1;
numStrWtEdges++;
}
}
thread_barrier_wait();
if (myId == 0) {
long t = 0;
for (i = 0; i < numEdgesPlaced; i++) {
if (SDGdataPtr->intWeight[i] < 0) {
SDGdataPtr->intWeight[i] = -t;
t++;
}
}
}
AL_LOCK(0);
TM_BEGIN();
TM_SHARED_WRITE(global_numStrWtEdges,
((long)TM_SHARED_READ(global_numStrWtEdges) + numStrWtEdges));
TM_END();
thread_barrier_wait();
numStrWtEdges = global_numStrWtEdges;
if (myId == 0) {
SDGdataPtr->strWeight =
(char*)P_MALLOC(numStrWtEdges * MAX_STRLEN * sizeof(char));
assert(SDGdataPtr->strWeight);
}
thread_barrier_wait();
createPartition(0, numEdgesPlaced, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
if (SDGdataPtr->intWeight[i] <= 0) {
long j;
for (j = 0; j < MAX_STRLEN; j++) {
SDGdataPtr->strWeight[(-SDGdataPtr->intWeight[i])*MAX_STRLEN+j] =
(char) (1 + PRANDOM_GENERATE(stream) % 127);
}
}
}
/*
* Choose SOUGHT STRING randomly if not assigned
*/
if (myId == 0) {
if (strlen(SOUGHT_STRING) != MAX_STRLEN) {
SOUGHT_STRING = (char*)P_MALLOC(MAX_STRLEN * sizeof(char));
assert(SOUGHT_STRING);
}
long t = PRANDOM_GENERATE(stream) % numStrWtEdges;
long j;
for (j = 0; j < MAX_STRLEN; j++) {
SOUGHT_STRING[j] =
(char) ((long) SDGdataPtr->strWeight[t*MAX_STRLEN+j]);
}
}
thread_barrier_wait();
/*
* STEP 5: Permute Vertices
*/
for (i = i_start; i < i_stop; i++) {
startVertex[i] = permV[(startVertex[i])];
endVertex[i] = permV[(endVertex[i])];
}
thread_barrier_wait();
/*
* STEP 6: Sort Vertices
*/
/*
* Radix sort with StartVertex as primary key
*/
if (myId == 0) {
long numByte = numEdgesPlaced * sizeof(ULONGINT_T);
SDGdataPtr->startVertex = (ULONGINT_T*)P_MALLOC(numByte);
assert(SDGdataPtr->startVertex);
SDGdataPtr->endVertex = (ULONGINT_T*)P_MALLOC(numByte);
assert(SDGdataPtr->endVertex);
}
thread_barrier_wait();
all_radixsort_node_aux_s3(numEdgesPlaced,
startVertex,
SDGdataPtr->startVertex,
endVertex,
SDGdataPtr->endVertex);
thread_barrier_wait();
if (myId == 0) {
P_FREE(startVertex);
P_FREE(endVertex);
}
thread_barrier_wait();
if (SCALE < 12) {
/*
* Sort with endVertex as secondary key
*/
if (myId == 0) {
long i0 = 0;
long i1 = 0;
i = 0;
while (i < numEdgesPlaced) {
for (i = i0; i < numEdgesPlaced; i++) {
if (SDGdataPtr->startVertex[i] !=
SDGdataPtr->startVertex[i1])
{
i1 = i;
break;
}
}
long j;
for (j = i0; j < i1; j++) {
long k;
for (k = j+1; k < i1; k++) {
if (SDGdataPtr->endVertex[k] <
SDGdataPtr->endVertex[j])
{
long t = SDGdataPtr->endVertex[j];
SDGdataPtr->endVertex[j] = SDGdataPtr->endVertex[k];
SDGdataPtr->endVertex[k] = t;
}
}
}
if (SDGdataPtr->startVertex[i0] != TOT_VERTICES-1) {
i0 = i1;
} else {
long j;
for (j=i0; j<numEdgesPlaced; j++) {
long k;
for (k=j+1; k<numEdgesPlaced; k++) {
if (SDGdataPtr->endVertex[k] <
SDGdataPtr->endVertex[j])
{
long t = SDGdataPtr->endVertex[j];
SDGdataPtr->endVertex[j] = SDGdataPtr->endVertex[k];
SDGdataPtr->endVertex[k] = t;
}
}
}
}
} /* while i < numEdgesPlaced */
}
} else {
ULONGINT_T* tempIndex;
if (myId == 0) {
tempIndex =
(ULONGINT_T*)P_MALLOC((TOT_VERTICES + 1) * sizeof(ULONGINT_T));
assert(tempIndex);
global_tempIndex = tempIndex;
/*
* Update degree of each vertex
*/
tempIndex[0] = 0;
tempIndex[TOT_VERTICES] = numEdgesPlaced;
long i0 = 0;
for (i=0; i < TOT_VERTICES; i++) {
tempIndex[i+1] = tempIndex[i];
long j;
for (j = i0; j < numEdgesPlaced; j++) {
if (SDGdataPtr->startVertex[j] !=
SDGdataPtr->startVertex[i0])
{
if (SDGdataPtr->startVertex[i0] == i) {
tempIndex[i+1] = j;
i0 = j;
break;
}
}
}
}
}
thread_barrier_wait();
tempIndex = global_tempIndex;
/*
* Insertion sort for now, replace with something better later on
*/
#if 0
createPartition(0, TOT_VERTICES, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
long j;
for (j = tempIndex[i]; j < tempIndex[i+1]; j++) {
long k;
for (k = (j + 1); k < tempIndex[i+1]; k++) {
if (SDGdataPtr->endVertex[k] <
SDGdataPtr->endVertex[j])
{
long t = SDGdataPtr->endVertex[j];
SDGdataPtr->endVertex[j] = SDGdataPtr->endVertex[k];
SDGdataPtr->endVertex[k] = t;
}
}
}
}
#else
if (myId == 0) {
for (i = 0; i < TOT_VERTICES; i++) {
long j;
for (j = tempIndex[i]; j < tempIndex[i+1]; j++) {
long k;
for (k = (j + 1); k < tempIndex[i+1]; k++) {
if (SDGdataPtr->endVertex[k] <
SDGdataPtr->endVertex[j])
{
long t = SDGdataPtr->endVertex[j];
SDGdataPtr->endVertex[j] = SDGdataPtr->endVertex[k];
SDGdataPtr->endVertex[k] = t;
}
}
}
}
}
#endif
if (myId == 0) {
P_FREE(tempIndex);
}
} /* SCALE >= 12 */
PRANDOM_FREE(stream);
if (myId == 0) {
P_FREE(permV);
}
TM_THREAD_EXIT();
}
/* =============================================================================
* findSubGraphs1
* =============================================================================
*/
void
findSubGraphs1 (void* argPtr)
{
graph* GPtr = ((findSubGraphs1_arg_t*)argPtr)->GPtr;
Vl** intWtVLList = ((findSubGraphs1_arg_t*)argPtr)->intWtVLList;
Vl** strWtVLList = ((findSubGraphs1_arg_t*)argPtr)->strWtVLList;
edge* maxIntWtList = ((findSubGraphs1_arg_t*)argPtr)->maxIntWtList;
long maxIntWtListSize = ((findSubGraphs1_arg_t*)argPtr)->maxIntWtListSize;
edge* soughtStrWtList = ((findSubGraphs1_arg_t*)argPtr)->soughtStrWtList;
long soughtStrWtListSize = ((findSubGraphs1_arg_t*)argPtr)->soughtStrWtListSize;
long myId = thread_getId();
long numThread = thread_getNumThread();
char* visited = (char*)P_MALLOC(GPtr->numVertices * sizeof(char));
assert(visited);
long i;
long i_start;
long i_stop;
createPartition(0,
(maxIntWtListSize + soughtStrWtListSize),
myId,
numThread,
&i_start,
&i_stop);
for (i = i_start; i < i_stop; i++) {
unsigned long k;
for (k = 0; k < GPtr->numVertices; k++) {
visited[k] = 'u';
}
if (i < maxIntWtListSize) {
intWtVLList[i] = (Vl*)P_MALLOC(sizeof(Vl));
assert(intWtVLList[i]);
(intWtVLList[i])->num = maxIntWtList[i].startVertex;
(intWtVLList[i])->depth = 0;
visited[(intWtVLList[i])->num] = 'v';
(intWtVLList[i])->next = (Vl*)P_MALLOC(sizeof(Vl));
P_FREE((intWtVLList[i])->next);
((intWtVLList[i])->next)->num = maxIntWtList[i].endVertex;
((intWtVLList[i])->next)->depth = 1;
visited[((intWtVLList[i])->next)->num] = 'v';
Vl* currV = (intWtVLList[i])->next;
Vl* startV = (intWtVLList[i])->next;
long depth = 1;
long verticesVisited = 2;
long currIndex = 1;
while ((startV->depth < SUBGR_EDGE_LENGTH) ||
(verticesVisited == (long)GPtr->numVertices))
{
depth = startV->depth + 1;
long j;
long j_start = GPtr->outVertexIndex[startV->num];
long j_stop = j_start + GPtr->outDegree[startV->num];
for (j = j_start; j < j_stop; j++) {
if (visited[GPtr->outVertexList[j]] == 'u') {
visited[GPtr->outVertexList[j]] = 'v';
currV->next = (Vl*)P_MALLOC(sizeof(Vl));
assert(currV->next);
(currV->next)->num = GPtr->outVertexList[j];
(currV->next)->depth = depth;
verticesVisited++;
currV = currV->next;
}
}
if ((currIndex < verticesVisited - 1) &&
(verticesVisited < (long)GPtr->numVertices))
{
currIndex++;
startV = startV->next;
} else {
break;
}
}
currV->next = NULL;
} else {
long t = i - maxIntWtListSize;
strWtVLList[t] = (Vl*)P_MALLOC(sizeof(Vl));
assert(strWtVLList[t]);
(strWtVLList[t])->num = soughtStrWtList[t].startVertex;
(strWtVLList[t])->depth = 0;
visited[(strWtVLList[t])->num] = 'v';
(strWtVLList[t])->next = (Vl*)P_MALLOC(sizeof(Vl));
assert((strWtVLList[t])->next);
((strWtVLList[t])->next)->num = soughtStrWtList[t].endVertex;
((strWtVLList[t])->next)->depth = 1;
visited[((strWtVLList[t])->next)->num] = 'v';
Vl* currV = (strWtVLList[t])->next;
Vl* startV = (strWtVLList[t])->next;
long depth = 1;
long verticesVisited = 2;
long currIndex = 1;
while ((startV->depth < SUBGR_EDGE_LENGTH) ||
(verticesVisited == (long)GPtr->numVertices))
{
depth = startV->depth + 1;
long j;
long j_start = GPtr->outVertexIndex[startV->num];
long j_stop = j_start + GPtr->outDegree[startV->num];
for (j = j_start; j < j_stop; j++) {
if (visited[GPtr->outVertexList[j]] == 'u') {
visited[GPtr->outVertexList[j]] = 'v';
currV->next = (Vl*)P_MALLOC(sizeof(Vl));
assert(currV->next);
(currV->next)->num = GPtr->outVertexList[j];
(currV->next)->depth = depth;
verticesVisited++;
currV = currV->next;
}
}
if ((currIndex < verticesVisited - 1) &&
(verticesVisited < (long)GPtr->numVertices))
{
currIndex++;
startV = startV->next;
} else {
break;
}
}
currV->next = NULL;
}
} /* for i */
P_FREE(visited);
}
/* =============================================================================
* computeGraph
* =============================================================================
*/
void
computeGraph (void* argPtr)
{
TM_THREAD_ENTER();
graph* GPtr = ((computeGraph_arg_t*)argPtr)->GPtr;
graphSDG* SDGdataPtr = ((computeGraph_arg_t*)argPtr)->SDGdataPtr;
long myId = thread_getId();
long numThread = thread_getNumThread();
ULONGINT_T j;
ULONGINT_T maxNumVertices = 0;
ULONGINT_T numEdgesPlaced = SDGdataPtr->numEdgesPlaced;
/*
* First determine the number of vertices by scanning the tuple
* startVertex list
*/
long i;
long i_start;
long i_stop;
createPartition(0, numEdgesPlaced, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
if (SDGdataPtr->startVertex[i] > maxNumVertices) {
maxNumVertices = SDGdataPtr->startVertex[i];
}
}
TM_BEGIN();
long tmp_maxNumVertices = (long)TM_SHARED_READ_L(global_maxNumVertices);
long new_maxNumVertices = MAX(tmp_maxNumVertices, maxNumVertices) + 1;
TM_SHARED_WRITE_L(global_maxNumVertices, new_maxNumVertices);
TM_END();
thread_barrier_wait();
maxNumVertices = global_maxNumVertices;
if (myId == 0) {
GPtr->numVertices = maxNumVertices;
GPtr->numEdges = numEdgesPlaced;
GPtr->intWeight = SDGdataPtr->intWeight;
GPtr->strWeight = SDGdataPtr->strWeight;
for (i = 0; i < numEdgesPlaced; i++) {
if (GPtr->intWeight[numEdgesPlaced-i-1] < 0) {
GPtr->numStrEdges = -(GPtr->intWeight[numEdgesPlaced-i-1]) + 1;
GPtr->numIntEdges = numEdgesPlaced - GPtr->numStrEdges;
break;
}
}
GPtr->outDegree =
(LONGINT_T*)P_MALLOC((GPtr->numVertices) * sizeof(LONGINT_T));
assert(GPtr->outDegree);
GPtr->outVertexIndex =
(ULONGINT_T*)P_MALLOC((GPtr->numVertices) * sizeof(ULONGINT_T));
assert(GPtr->outVertexIndex);
}
thread_barrier_wait();
createPartition(0, GPtr->numVertices, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
GPtr->outDegree[i] = 0;
GPtr->outVertexIndex[i] = 0;
}
ULONGINT_T outVertexListSize = 0;
thread_barrier_wait();
ULONGINT_T i0 = -1UL;
for (i = i_start; i < i_stop; i++) {
ULONGINT_T k = i;
if ((outVertexListSize == 0) && (k != 0)) {
while (i0 == -1UL) {
for (j = 0; j < numEdgesPlaced; j++) {
if (k == SDGdataPtr->startVertex[j]) {
i0 = j;
break;
}
}
k--;
}
}
if ((outVertexListSize == 0) && (k == 0)) {
i0 = 0;
}
for (j = i0; j < numEdgesPlaced; j++) {
if (i == GPtr->numVertices-1) {
break;
}
if ((i != SDGdataPtr->startVertex[j])) {
if ((j > 0) && (i == SDGdataPtr->startVertex[j-1])) {
if (j-i0 >= 1) {
outVertexListSize++;
GPtr->outDegree[i]++;
ULONGINT_T t;
for (t = i0+1; t < j; t++) {
if (SDGdataPtr->endVertex[t] !=
SDGdataPtr->endVertex[t-1])
{
outVertexListSize++;
GPtr->outDegree[i] = GPtr->outDegree[i]+1;
}
}
}
}
i0 = j;
break;
}
}
if (i == GPtr->numVertices-1) {
if (numEdgesPlaced-i0 >= 0) {
outVertexListSize++;
GPtr->outDegree[i]++;
ULONGINT_T t;
for (t = i0+1; t < numEdgesPlaced; t++) {
if (SDGdataPtr->endVertex[t] != SDGdataPtr->endVertex[t-1]) {
outVertexListSize++;
GPtr->outDegree[i]++;
}
}
}
}
} /* for i */
thread_barrier_wait();
prefix_sums(GPtr->outVertexIndex, GPtr->outDegree, GPtr->numVertices);
thread_barrier_wait();
TM_BEGIN();
TM_SHARED_WRITE_L(
global_outVertexListSize,
((long)TM_SHARED_READ_L(global_outVertexListSize) + outVertexListSize)
);
TM_END();
thread_barrier_wait();
outVertexListSize = global_outVertexListSize;
if (myId == 0) {
GPtr->numDirectedEdges = outVertexListSize;
GPtr->outVertexList =
(ULONGINT_T*)P_MALLOC(outVertexListSize * sizeof(ULONGINT_T));
assert(GPtr->outVertexList);
GPtr->paralEdgeIndex =
(ULONGINT_T*)P_MALLOC(outVertexListSize * sizeof(ULONGINT_T));
assert(GPtr->paralEdgeIndex);
GPtr->outVertexList[0] = SDGdataPtr->endVertex[0];
}
thread_barrier_wait();
/*
* Evaluate outVertexList
*/
i0 = -1UL;
for (i = i_start; i < i_stop; i++) {
ULONGINT_T k = i;
while ((i0 == -1UL) && (k != 0)) {
for (j = 0; j < numEdgesPlaced; j++) {
if (k == SDGdataPtr->startVertex[j]) {
i0 = j;
break;
}
}
k--;
}
if ((i0 == -1) && (k == 0)) {
i0 = 0;
}
for (j = i0; j < numEdgesPlaced; j++) {
if (i == GPtr->numVertices-1) {
break;
}
if (i != SDGdataPtr->startVertex[j]) {
if ((j > 0) && (i == SDGdataPtr->startVertex[j-1])) {
if (j-i0 >= 1) {
long ii = GPtr->outVertexIndex[i];
ULONGINT_T r = 0;
GPtr->paralEdgeIndex[ii] = i0;
GPtr->outVertexList[ii] = SDGdataPtr->endVertex[i0];
r++;
ULONGINT_T t;
for (t = i0+1; t < j; t++) {
if (SDGdataPtr->endVertex[t] !=
SDGdataPtr->endVertex[t-1])
{
GPtr->paralEdgeIndex[ii+r] = t;
GPtr->outVertexList[ii+r] = SDGdataPtr->endVertex[t];
r++;
}
}
}
}
i0 = j;
break;
}
} /* for j */
if (i == GPtr->numVertices-1) {
ULONGINT_T r = 0;
if (numEdgesPlaced-i0 >= 0) {
long ii = GPtr->outVertexIndex[i];
GPtr->paralEdgeIndex[ii+r] = i0;
GPtr->outVertexList[ii+r] = SDGdataPtr->endVertex[i0];
r++;
ULONGINT_T t;
for (t = i0+1; t < numEdgesPlaced; t++) {
if (SDGdataPtr->endVertex[t] != SDGdataPtr->endVertex[t-1]) {
GPtr->paralEdgeIndex[ii+r] = t;
GPtr->outVertexList[ii+r] = SDGdataPtr->endVertex[t];
r++;
}
}
}
}
} /* for i */
thread_barrier_wait();
if (myId == 0) {
P_FREE(SDGdataPtr->startVertex);
P_FREE(SDGdataPtr->endVertex);
GPtr->inDegree =
(LONGINT_T*)P_MALLOC(GPtr->numVertices * sizeof(LONGINT_T));
assert(GPtr->inDegree);
GPtr->inVertexIndex =
(ULONGINT_T*)P_MALLOC(GPtr->numVertices * sizeof(ULONGINT_T));
assert(GPtr->inVertexIndex);
}
thread_barrier_wait();
for (i = i_start; i < i_stop; i++) {
GPtr->inDegree[i] = 0;
GPtr->inVertexIndex[i] = 0;
}
/* A temp. array to store the inplied edges */
ULONGINT_T* impliedEdgeList;
if (myId == 0) {
impliedEdgeList = (ULONGINT_T*)P_MALLOC(GPtr->numVertices
* MAX_CLUSTER_SIZE
* sizeof(ULONGINT_T));
global_impliedEdgeList = impliedEdgeList;
}
thread_barrier_wait();
impliedEdgeList = global_impliedEdgeList;
createPartition(0,
(GPtr->numVertices * MAX_CLUSTER_SIZE),
myId,
numThread,
&i_start,
&i_stop);
for (i = i_start; i < i_stop; i++) {
impliedEdgeList[i] = 0;
}
/*
* An auxiliary array to store implied edges, in case we overshoot
* MAX_CLUSTER_SIZE
*/
ULONGINT_T** auxArr;
if (myId == 0) {
auxArr = (ULONGINT_T**)P_MALLOC(GPtr->numVertices * sizeof(ULONGINT_T*));
assert(auxArr);
global_auxArr = auxArr;
}
thread_barrier_wait();
auxArr = global_auxArr;
createPartition(0, GPtr->numVertices, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
/* Inspect adjacency list of vertex i */
for (j = GPtr->outVertexIndex[i];
j < (GPtr->outVertexIndex[i] + GPtr->outDegree[i]);
j++)
{
ULONGINT_T v = GPtr->outVertexList[j];
ULONGINT_T k;
for (k = GPtr->outVertexIndex[v];
k < (GPtr->outVertexIndex[v] + GPtr->outDegree[v]);
k++)
{
if (GPtr->outVertexList[k] == i) {
break;
}
}
if (k == GPtr->outVertexIndex[v]+GPtr->outDegree[v]) {
TM_BEGIN();
/* Add i to the impliedEdgeList of v */
long inDegree = (long)TM_SHARED_READ_L(GPtr->inDegree[v]);
TM_SHARED_WRITE_L(GPtr->inDegree[v], (inDegree + 1));
if (inDegree < MAX_CLUSTER_SIZE) {
TM_SHARED_WRITE_L(impliedEdgeList[v*MAX_CLUSTER_SIZE+inDegree],
i);
} else {
/* Use auxiliary array to store the implied edge */
/* Create an array if it's not present already */
ULONGINT_T* a = NULL;
if ((inDegree % MAX_CLUSTER_SIZE) == 0) {
a = (ULONGINT_T*)TM_MALLOC(MAX_CLUSTER_SIZE
* sizeof(ULONGINT_T));
assert(a);
TM_SHARED_WRITE_P(auxArr[v], a);
} else {
a = auxArr[v];
}
TM_SHARED_WRITE_L(a[inDegree % MAX_CLUSTER_SIZE], i);
}
TM_END();
}
}
} /* for i */
thread_barrier_wait();
prefix_sums(GPtr->inVertexIndex, GPtr->inDegree, GPtr->numVertices);
if (myId == 0) {
GPtr->numUndirectedEdges = GPtr->inVertexIndex[GPtr->numVertices-1]
+ GPtr->inDegree[GPtr->numVertices-1];
GPtr->inVertexList =
(ULONGINT_T *)P_MALLOC(GPtr->numUndirectedEdges * sizeof(ULONGINT_T));
}
thread_barrier_wait();
/*
* Create the inVertex List
*/
for (i = i_start; i < i_stop; i++) {
for (j = GPtr->inVertexIndex[i];
j < (GPtr->inVertexIndex[i] + GPtr->inDegree[i]);
j++)
{
if ((j - GPtr->inVertexIndex[i]) < MAX_CLUSTER_SIZE) {
GPtr->inVertexList[j] =
impliedEdgeList[i*MAX_CLUSTER_SIZE+j-GPtr->inVertexIndex[i]];
} else {
GPtr->inVertexList[j] =
auxArr[i][(j-GPtr->inVertexIndex[i]) % MAX_CLUSTER_SIZE];
}
}
}
thread_barrier_wait();
if (myId == 0) {
P_FREE(impliedEdgeList);
}
for (i = i_start; i < i_stop; i++) {
if (GPtr->inDegree[i] > MAX_CLUSTER_SIZE) {
P_FREE(auxArr[i]);
}
}
thread_barrier_wait();
if (myId == 0) {
P_FREE(auxArr);
}
TM_THREAD_EXIT();
}
/* =============================================================================
* Pelement_free
* =============================================================================
*/
void
Pelement_free (element_t* elementPtr)
{
PLIST_FREE(elementPtr->neighborListPtr);
P_FREE(elementPtr);
}
/* =============================================================================
* Prandom_free
* =============================================================================
*/
void
Prandom_free (random_t* randomPtr)
{
P_FREE(randomPtr);
}
/* =============================================================================
* Plist_free
* =============================================================================
*/
void
list_stm_v2::Plist_free (list_t* listPtr)
{
PfreeList(listPtr->head.nextPtr);
P_FREE(listPtr);
}
/* =============================================================================
* processPackets
* =============================================================================
*/
void
processPackets (void* argPtr)
{
TM_THREAD_ENTER();
long threadId = thread_getId();
stream_t* streamPtr = ((arg_t*)argPtr)->streamPtr;
decoder_t* decoderPtr = ((arg_t*)argPtr)->decoderPtr;
vector_t** errorVectors = ((arg_t*)argPtr)->errorVectors;
detector_t* detectorPtr = PDETECTOR_ALLOC();
assert(detectorPtr);
PDETECTOR_ADDPREPROCESSOR(detectorPtr, &preprocessor_toLower);
vector_t* errorVectorPtr = errorVectors[threadId];
while (1) {
char* bytes;
unsigned int locks[1];
TM_BEGIN();
SINGLE_LOCK(streamPtr);
bytes = TMSTREAM_GETPACKET(streamPtr);
SINGLE_UNLOCK(streamPtr);
TM_END();
if (!bytes) {
break;
}
packet_t* packetPtr = (packet_t*)bytes;
long flowId = packetPtr->flowId;
error_t error;
TM_BEGIN();
error = TMDECODER_PROCESS(decoderPtr,
bytes,
(PACKET_HEADER_LENGTH + packetPtr->length));
TM_END();
if (error) {
/*
* Currently, stream_generate() does not create these errors.
*/
assert(0);
bool_t status = PVECTOR_PUSHBACK(errorVectorPtr, (void*)flowId);
assert(status);
}
char* data;
long decodedFlowId;
TM_BEGIN();
SINGLE_LOCK(decoderPtr);
data = TMDECODER_GETCOMPLETE(decoderPtr, &decodedFlowId);
SINGLE_UNLOCK(decoderPtr);
TM_END();
if (data) {
error_t error = PDETECTOR_PROCESS(detectorPtr, data);
P_FREE(data);
if (error) {
bool_t status = PVECTOR_PUSHBACK(errorVectorPtr,
(void*)decodedFlowId);
assert(status);
}
}
}
PDETECTOR_FREE(detectorPtr);
TM_THREAD_EXIT();
}
/* =============================================================================
* getStartLists
* =============================================================================
*/
void
getStartLists (void* argPtr)
{
TM_THREAD_ENTER();
graph* GPtr = ((getStartLists_arg_t*)argPtr)->GPtr;
edge** maxIntWtListPtr = ((getStartLists_arg_t*)argPtr)->maxIntWtListPtr;
long* maxIntWtListSize = ((getStartLists_arg_t*)argPtr)->maxIntWtListSize;
edge** soughtStrWtListPtr = ((getStartLists_arg_t*)argPtr)->soughtStrWtListPtr;
long* soughtStrWtListSize = ((getStartLists_arg_t*)argPtr)->soughtStrWtListSize;
long myId = thread_getId();
long numThread = thread_getNumThread();
/*
* Find Max Wt on each thread
*/
LONGINT_T maxWeight = 0;
long i;
long i_start;
long i_stop;
createPartition(0, GPtr->numEdges, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
if (GPtr->intWeight[i] > maxWeight) {
maxWeight = GPtr->intWeight[i];
}
}
AL_LOCK(0);
TM_BEGIN(9);
long tmp_maxWeight = (long)TM_SHARED_READ(global_maxWeight);
if (maxWeight > tmp_maxWeight) {
TM_SHARED_WRITE(global_maxWeight, maxWeight);
}
TM_END();
thread_barrier_wait();
maxWeight = global_maxWeight;
/*
* Create partial lists
*/
/*
* Allocate mem. for temp edge list for each thread
*/
long numTmpEdge = (5+ceil(1.5*(GPtr->numIntEdges)/MAX_INT_WEIGHT));
edge* tmpEdgeList = (edge*)P_MALLOC(numTmpEdge * sizeof(edge));
long i_edgeCounter = 0;
for (i = i_start; i < i_stop; i++) {
if (GPtr->intWeight[i] == maxWeight) {
/* Find the corresponding endVertex */
long j;
for (j = 0; j < GPtr->numDirectedEdges; j++) {
if (GPtr->paralEdgeIndex[j] > i) {
break;
}
}
tmpEdgeList[i_edgeCounter].endVertex = GPtr->outVertexList[j-1];
tmpEdgeList[i_edgeCounter].edgeNum = j-1;
long t;
for (t = 0; t < GPtr->numVertices; t++) {
if (GPtr->outVertexIndex[t] > j-1) {
break;
}
}
tmpEdgeList[i_edgeCounter].startVertex = t-1;
i_edgeCounter++;
}
}
/*
* Merge partial edge lists
*/
long* i_edgeStartCounter;
long* i_edgeEndCounter;
if (myId == 0) {
i_edgeStartCounter = (long*)P_MALLOC(numThread * sizeof(long));
assert(i_edgeStartCounter);
global_i_edgeStartCounter = i_edgeStartCounter;
i_edgeEndCounter = (long*)P_MALLOC(numThread * sizeof(long));
assert(i_edgeEndCounter);
global_i_edgeEndCounter = i_edgeEndCounter;
*maxIntWtListSize = 0;
}
thread_barrier_wait();
i_edgeStartCounter = global_i_edgeStartCounter;
i_edgeEndCounter = global_i_edgeEndCounter;
i_edgeEndCounter[myId] = i_edgeCounter;
i_edgeStartCounter[myId] = 0;
thread_barrier_wait();
if (myId == 0) {
for (i = 1; i < numThread; i++) {
i_edgeEndCounter[i] = i_edgeEndCounter[i-1] + i_edgeEndCounter[i];
i_edgeStartCounter[i] = i_edgeEndCounter[i-1];
}
}
*maxIntWtListSize += i_edgeCounter;
thread_barrier_wait();
edge* maxIntWtList;
if (myId == 0) {
P_FREE(*maxIntWtListPtr);
maxIntWtList = (edge*)P_MALLOC((*maxIntWtListSize) * sizeof(edge));
assert(maxIntWtList);
global_maxIntWtList = maxIntWtList;
}
thread_barrier_wait();
maxIntWtList = global_maxIntWtList;
for (i = i_edgeStartCounter[myId]; i<i_edgeEndCounter[myId]; i++) {
(maxIntWtList[i]).startVertex = tmpEdgeList[i-i_edgeStartCounter[myId]].startVertex;
(maxIntWtList[i]).endVertex = tmpEdgeList[i-i_edgeStartCounter[myId]].endVertex;
(maxIntWtList[i]).edgeNum = tmpEdgeList[i-i_edgeStartCounter[myId]].edgeNum;
}
if (myId == 0) {
*maxIntWtListPtr = maxIntWtList;
}
i_edgeCounter = 0;
createPartition(0, GPtr->numStrEdges, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
if (strncmp(GPtr->strWeight+i*MAX_STRLEN,
SOUGHT_STRING,
MAX_STRLEN) == 0)
{
/*
* Find the corresponding endVertex
*/
long t;
for (t = 0; t < GPtr->numEdges; t++) {
if (GPtr->intWeight[t] == -i) {
break;
}
}
long j;
for (j = 0; j < GPtr->numDirectedEdges; j++) {
if (GPtr->paralEdgeIndex[j] > t) {
break;
}
}
tmpEdgeList[i_edgeCounter].endVertex = GPtr->outVertexList[j-1];
tmpEdgeList[i_edgeCounter].edgeNum = j-1;
for (t = 0; t < GPtr->numVertices; t++) {
if (GPtr->outVertexIndex[t] > j-1) {
break;
}
}
tmpEdgeList[i_edgeCounter].startVertex = t-1;
i_edgeCounter++;
}
}
thread_barrier_wait();
i_edgeEndCounter[myId] = i_edgeCounter;
i_edgeStartCounter[myId] = 0;
if (myId == 0) {
*soughtStrWtListSize = 0;
}
thread_barrier_wait();
if (myId == 0) {
for (i = 1; i < numThread; i++) {
i_edgeEndCounter[i] = i_edgeEndCounter[i-1] + i_edgeEndCounter[i];
i_edgeStartCounter[i] = i_edgeEndCounter[i-1];
}
}
*soughtStrWtListSize += i_edgeCounter;
thread_barrier_wait();
edge* soughtStrWtList;
if (myId == 0) {
P_FREE(*soughtStrWtListPtr);
soughtStrWtList = (edge*)P_MALLOC((*soughtStrWtListSize) * sizeof(edge));
assert(soughtStrWtList);
global_soughtStrWtList = soughtStrWtList;
}
thread_barrier_wait();
soughtStrWtList = global_soughtStrWtList;
for (i = i_edgeStartCounter[myId]; i < i_edgeEndCounter[myId]; i++) {
(soughtStrWtList[i]).startVertex =
tmpEdgeList[i-i_edgeStartCounter[myId]].startVertex;
(soughtStrWtList[i]).endVertex =
tmpEdgeList[i-i_edgeStartCounter[myId]].endVertex;
(soughtStrWtList[i]).edgeNum =
tmpEdgeList[i-i_edgeStartCounter[myId]].edgeNum;
}
thread_barrier_wait();
if (myId == 0) {
*soughtStrWtListPtr = soughtStrWtList;
P_FREE(i_edgeStartCounter);
P_FREE(i_edgeEndCounter);
}
P_FREE(tmpEdgeList);
TM_THREAD_EXIT();
}
/* =============================================================================
* TMgrid_free
* =============================================================================
*/
void
Pgrid_free (grid_t* gridPtr)
{
P_FREE(gridPtr->points_unaligned);
P_FREE(gridPtr);
}
/* =============================================================================
* cutClusters
* =============================================================================
*/
void
cutClusters (void* argPtr)
{
TM_THREAD_ENTER();
graph* GPtr = (graph*)argPtr;
long myId = thread_getId();
long numThread = thread_getNumThread();
/*
* Sort the vertex list by their degree
*/
ULONGINT_T* Index;
ULONGINT_T* neighbourArray;
ULONGINT_T* IndexSorted;
ULONGINT_T* neighbourArraySorted;
if (myId == 0) {
long numByte = GPtr->numVertices * sizeof(ULONGINT_T);
Index = (ULONGINT_T*)P_MALLOC(numByte);
assert(Index);
global_Index = Index;
neighbourArray = (ULONGINT_T*)P_MALLOC(numByte);
assert(neighbourArray);
global_neighbourArray = neighbourArray;
IndexSorted = (ULONGINT_T*)P_MALLOC(numByte);
assert(IndexSorted);
global_IndexSorted = IndexSorted;
neighbourArraySorted = (ULONGINT_T*)P_MALLOC(numByte);
assert(neighbourArraySorted);
global_neighbourArraySorted = neighbourArraySorted;
}
thread_barrier_wait();
Index = global_Index;
neighbourArray = global_neighbourArray;
IndexSorted = global_IndexSorted;
neighbourArraySorted = global_neighbourArraySorted;
long i;
long i_start;
long i_stop;
createPartition(0, GPtr->numVertices, myId, numThread, &i_start, &i_stop);
for (i = i_start; i < i_stop; i++) {
neighbourArray[i] = GPtr->inDegree[i] + GPtr->outDegree[i];
Index[i] = i;
}
thread_barrier_wait();
all_radixsort_node_aux_s3(GPtr->numVertices,
neighbourArray,
neighbourArraySorted,
Index,
IndexSorted);
thread_barrier_wait();
/*
* Global array to keep track of vertex status:
* -1 if a vertex hasn't been assigned to a cluster yet
* t if it belongs to a cluster; t = iteration*numThread + myId
*/
long* vStatus;
edge* pCutSet;
ULONGINT_T* startV;
ULONGINT_T* clusterSize;
if (myId == 0) {
P_FREE(Index);
P_FREE(neighbourArray);
vStatus = (long*)P_MALLOC(GPtr->numVertices * sizeof(long));
assert(vStatus);
global_vStatus = vStatus;
/*
* Allocate mem. for the cut set list
* Maintain local arrays initially and merge them in the end
*/
if (SCALE < 12) {
pCutSet =(edge*)P_MALLOC((1*(GPtr->numDirectedEdges)/numThread)
* sizeof(edge));
} else {
pCutSet = (edge*)P_MALLOC((0.2*(GPtr->numDirectedEdges)/numThread)
* sizeof(edge));
}
assert(pCutSet);
global_pCutSet = pCutSet;
/*
* Vertex to start from, on each thread
*/
startV = (ULONGINT_T*)P_MALLOC(numThread * sizeof(ULONGINT_T));
assert(startV);
global_startV = startV;
clusterSize = (ULONGINT_T*)P_MALLOC(numThread * sizeof(ULONGINT_T));
assert(clusterSize);
global_clusterSize = clusterSize;
}
thread_barrier_wait();
vStatus = global_vStatus;
pCutSet = global_pCutSet;
startV = global_startV;
clusterSize = global_clusterSize;
for (i = i_start; i < i_stop; i++) {
vStatus[i] = -1;
}
thread_barrier_wait();
ULONGINT_T verticesVisited = 0;
#ifdef WRITE_RESULT_FILES
FILE* outfp1 = NULL;
if (myId == 0) {
outfp1 = fopen("clusters.txt", "w");
fprintf(outfp1, "\nKernel 4 - Extracted Clusters\n");
}
#endif
long iter = 0;
ULONGINT_T currIndex = 0;
ULONGINT_T cutSetIndex = 0;
while (verticesVisited < GPtr->numVertices) {
/* Clear start vertex array */
startV[myId] = -1;
clusterSize[myId] = 0;
if (currIndex == GPtr->numVertices) {
currIndex = 0;
}
thread_barrier_wait();
/*
* Choose vertices to start from
* Done sequentially right now, can be parallelized
*/
if (myId == 0) {
long t;
for (t = 0; t < numThread; t++) {
long r;
for (r = currIndex; r < GPtr->numVertices; r++) {
if (vStatus[IndexSorted[GPtr->numVertices - r - 1]] == -1) {
startV[t] = IndexSorted[GPtr->numVertices - r - 1];
vStatus[startV[t]] = iter * numThread + t;
long j;
for (j = 0; j < GPtr->outDegree[startV[t]]; j++) {
long outVertexListIndex =
j+GPtr->outVertexIndex[startV[t]];
long vStatusIndex =
GPtr->outVertexList[outVertexListIndex];
if (vStatus[vStatusIndex] == -1) {
vStatus[vStatusIndex] = iter * numThread + t;
clusterSize[t]++;
}
}
for (j = 0; j < GPtr->inDegree[startV[t]]; j++) {
long inVertexIndex = j+GPtr->inVertexIndex[startV[t]];
long vStatusIndex = GPtr->inVertexList[inVertexIndex];
if (vStatus[vStatusIndex] == -1) {
vStatus[vStatusIndex] = iter * numThread + t;
clusterSize[t]++;
}
}
currIndex = r+1;
break;
}
}
}
}
thread_barrier_wait();
/*
* Determine clusters and cut sets in parallel
*/
i = startV[myId];
ULONGINT_T cliqueSize = 0;
/* If the thread has some vertex to start from */
if (i != -1) {
cliqueSize = 1;
/* clusterSize[myId] gives the no. of 'unassigned' vertices adjacent to the current vertex */
if ((clusterSize[myId] >= 0.6*(GPtr->inDegree[i]+GPtr->outDegree[i])) ||
((iter > (GPtr->numVertices)/(numThread*MAX_CLUSTER_SIZE)) &&
(clusterSize[myId] > 0)))
{
/*
* Most of the adjacent vertices are unassigned,
* should be able to extract a cluster easily
*/
/* Inspect adjacency list */
long j;
for (j = 0; j < GPtr->outDegree[i]; j++) {
ULONGINT_T clusterCounter = 0;
ULONGINT_T cutSetIndexPrev = cutSetIndex;
ULONGINT_T cutSetCounter = 0;
if (vStatus[GPtr->outVertexList[j+GPtr->outVertexIndex[i]]] ==
iter * numThread + myId)
{
long v = GPtr->outVertexList[j+GPtr->outVertexIndex[i]];
/*
* Inspect vertices adjacent to v and determine if it belongs
* to a cluster or not
*/
long k;
for (k = 0; k < GPtr->outDegree[v]; k++) {
long outVertexListIndex = k+GPtr->outVertexIndex[v];
long vStatusIndex = GPtr->outVertexList[outVertexListIndex];
if (vStatus[vStatusIndex] == (iter * numThread + myId)) {
clusterCounter++;
} else {
cutSetCounter++;
if (vStatus[vStatusIndex] == -1) {
/* Ensure that an edge is not added twice to the list */
pCutSet[cutSetIndex].startVertex = v;
pCutSet[cutSetIndex].endVertex = vStatusIndex;
cutSetIndex++;
}
}
}
if ((cutSetCounter >= clusterCounter) ||
((SCALE < 9) &&
(clusterCounter <= 2) &&
(GPtr->inDegree[v]+GPtr->outDegree[v] >
clusterCounter + cutSetCounter) &&
(clusterSize[myId] > clusterCounter + 2)) ||
((SCALE > 9) &&
(clusterCounter < 0.5*clusterSize[myId])))
{
/* v doesn't belong to this clique, free it */
vStatus[v] = -1;
/* Also add this edge to cutset list, removing previously added edges */
cutSetIndex = cutSetIndexPrev;
pCutSet[cutSetIndex].startVertex = i;
pCutSet[cutSetIndex].endVertex = v;
cutSetIndex++;
} else {
cliqueSize++;
/* Add edges in inVertexList also to cut Set */
for (k = 0; k < GPtr->inDegree[v]; k++) {
long inVertexListIndex = k+GPtr->inVertexIndex[v];
long vStatusIndex = GPtr->inVertexList[inVertexListIndex];
if (vStatus[vStatusIndex] == -1) {
pCutSet[cutSetIndex].startVertex = v;
pCutSet[cutSetIndex].endVertex = vStatusIndex;
cutSetIndex++;
}
}
}
}
}
/* Do the same for the implied edges too */
for (j = 0; j < GPtr->inDegree[i]; j++) {
ULONGINT_T clusterCounter = 0;
ULONGINT_T cutSetIndexPrev = cutSetIndex;
ULONGINT_T cutSetCounter = 0;
if (vStatus[GPtr->inVertexList[j+GPtr->inVertexIndex[i]]] ==
iter*numThread+myId)
{
long v = GPtr->inVertexList[j+GPtr->inVertexIndex[i]];
/* Inspect vertices adjacent to v and determine if it belongs to a cluster or not */
long k;
for (k = 0; k < GPtr->outDegree[v]; k++) {
long outVertexListIndex = k+GPtr->outVertexIndex[v];
long vStatusIndex = GPtr->outVertexList[outVertexListIndex];
if (vStatus[vStatusIndex] == iter*numThread+myId) {
clusterCounter++;
} else {
cutSetCounter++;
if (vStatus[vStatusIndex] == -1) {
/* To ensure that an edge is not added twice to the list */
pCutSet[cutSetIndex].startVertex = v;
pCutSet[cutSetIndex].endVertex = vStatusIndex;
cutSetIndex++;
}
}
}
if ((cutSetCounter >= clusterCounter) ||
((SCALE < 9) &&
(clusterCounter <= 2) &&
(GPtr->inDegree[v]+GPtr->outDegree[v] >
clusterCounter + cutSetCounter) &&
(clusterSize[myId] > clusterCounter + 2)) ||
((SCALE > 9) &&
(clusterCounter < 0.5*clusterSize[myId])))
{
/* v doesn't belong to this clique, free it */
vStatus[v] = -1;
cutSetIndex = cutSetIndexPrev;
pCutSet[cutSetIndex].startVertex = i;
pCutSet[cutSetIndex].endVertex = v;
cutSetIndex++;
} else {
cliqueSize++;
/* Add edges in inVertexList also to cut Set */
for (k = 0; k < GPtr->inDegree[v]; k++) {
long inVertexListIndex = k+GPtr->inVertexIndex[v];
long vStatusIndex = GPtr->inVertexList[inVertexListIndex];
if (vStatus[vStatusIndex] == -1) {
pCutSet[cutSetIndex].startVertex = v;
pCutSet[cutSetIndex].endVertex = vStatusIndex;
cutSetIndex++;
}
}
}
}
}
} /* i != -1 */
if (clusterSize[myId] == 0) {
/* Only one vertex in cluster */
cliqueSize = 1;
} else {
if ((clusterSize[myId] < 0.6*(GPtr->inDegree[i]+GPtr->outDegree[i])) &&
(iter <= GPtr->numVertices/(numThread*MAX_CLUSTER_SIZE)))
{
/* High perc. of intra-clique edges, do not commit clique */
cliqueSize = 0;
vStatus[i] = -1;
long j;
for (j=0; j<GPtr->outDegree[i]; j++) {
long outVertexListIndex = j+GPtr->outVertexIndex[i];
long vStatusIndex = GPtr->outVertexList[outVertexListIndex];
if (vStatus[vStatusIndex] == iter*numThread+myId) {
vStatus[vStatusIndex] = -1;
}
}
for (j=0; j<GPtr->inDegree[i]; j++) {
long inVertexListIndex = j+GPtr->inVertexIndex[i];
long vStatusIndex = GPtr->inVertexList[inVertexListIndex];
if (vStatus[vStatusIndex] == iter*numThread+myId) {
vStatus[vStatusIndex] = -1;
}
}
}
}
} /* if i != -1 */
if (myId == 0) {
global_cliqueSize = 0;
}
thread_barrier_wait();
#ifdef WRITE_RESULT_FILES
/* Print to results.clq file */
if (myId == 0) {
long t;
for (t = 0; t < numThread; t++) {
if (startV[t] != -1) {
if (vStatus[startV[t]] == iter*numThread+t) {
fprintf(outfp1, "%lu ", startV[t]);
long j;
for (j = 0; j < GPtr->outDegree[startV[t]]; j++) {
long outVertexListIndex = j+GPtr->outVertexIndex[startV[t]];
long vStatusIndex = GPtr->outVertexList[outVertexListIndex];
if (vStatus[vStatusIndex] == iter*numThread+t) {
fprintf(outfp1, "%lu ", vStatusIndex);
}
}
for (j = 0; j < GPtr->inDegree[startV[t]]; j++) {
long inVertexListIndex = j+GPtr->inVertexIndex[startV[t]];
long vStatusIndex = GPtr->inVertexList[inVertexListIndex];
if (vStatus[vStatusIndex] == iter*numThread+t) {
fprintf(outfp1, "%lu ", vStatusIndex);
}
}
fprintf(outfp1, "\n");
}
}
}
}
thread_barrier_wait();
#endif /* WRITE_RESULTS_FILE */
if (myId == 0) {
iter++;
global_iter = iter;
}
TM_BEGIN();
long tmp_cliqueSize = (long)TM_SHARED_READ(global_cliqueSize);
TM_SHARED_WRITE(global_cliqueSize, (tmp_cliqueSize + cliqueSize));
TM_END();
thread_barrier_wait();
iter = global_iter;
verticesVisited += global_cliqueSize;
if ((verticesVisited >= 0.95*GPtr->numVertices) ||
(iter > GPtr->numVertices/2))
{
break;
}
} /* while (verticesVisited < GPtr->numVertices) */
thread_barrier_wait();
#ifdef WRITE_RESULT_FILES
/* Take care of unmarked vertices */
if (myId == 0) {
if (verticesVisited < GPtr->numVertices) {
for(i = 0; i < GPtr->numVertices; i++) {
if (vStatus[i] == -1) {
vStatus[i] = iter*numThread+myId;
fprintf(outfp1, "%lu\n", i);
iter++;
}
}
}
}
thread_barrier_wait();
#endif
/*
* Merge partial Cutset Lists
*/
/* Temp vars for merging edge lists */
ULONGINT_T* edgeStartCounter;
ULONGINT_T* edgeEndCounter;
if (myId == 0) {
edgeStartCounter = (ULONGINT_T*)P_MALLOC(numThread * sizeof(ULONGINT_T));
assert(edgeStartCounter);
global_edgeStartCounter = edgeStartCounter;
edgeEndCounter = (ULONGINT_T*)P_MALLOC(numThread * sizeof(ULONGINT_T));
assert(edgeEndCounter);
global_edgeEndCounter = edgeEndCounter;
}
thread_barrier_wait();
edgeStartCounter = global_edgeStartCounter;
edgeEndCounter = global_edgeEndCounter;
edgeEndCounter[myId] = cutSetIndex;
edgeStartCounter[myId] = 0;
thread_barrier_wait();
if (myId == 0) {
long t;
for (t = 1; t < numThread; t++) {
edgeEndCounter[t] = edgeEndCounter[t-1] + edgeEndCounter[t];
edgeStartCounter[t] = edgeEndCounter[t-1];
}
}
TM_BEGIN();
long tmp_cutSetIndex = (long)TM_SHARED_READ(global_cutSetIndex);
TM_SHARED_WRITE(global_cutSetIndex, (tmp_cutSetIndex + cutSetIndex));
TM_END();
thread_barrier_wait();
cutSetIndex = global_cutSetIndex;
ULONGINT_T cutSetCounter = cutSetIndex;
/* Data struct. for storing edgeCut */
edge* cutSet;
if (myId == 0) {
cutSet = (edge*)P_MALLOC(cutSetCounter * sizeof(edge));
assert(cutSet);
global_cutSet = cutSet;
}
thread_barrier_wait();
cutSet = global_cutSet;
long j;
for (j = edgeStartCounter[myId]; j < edgeEndCounter[myId]; j++) {
cutSet[j].startVertex = pCutSet[j-edgeStartCounter[myId]].startVertex;
cutSet[j].endVertex = pCutSet[j-edgeStartCounter[myId]].endVertex;
}
thread_barrier_wait();
#ifdef WRITE_RESULT_FILES
FILE* outfp2 = NULL;
if (myId == 0) {
outfp2 = fopen("edgeCut.txt", "w");
fprintf(outfp2, "\nEdges in Cut Set - \n");
for (i = 0; i < cutSetCounter; i++) {
fprintf(outfp2, "[%lu %lu] ",
cutSet[i].startVertex, cutSet[i].endVertex);
}
fclose(outfp2);
fclose(outfp1);
}
#endif
if (myId == 0) {
P_FREE(edgeStartCounter);
P_FREE(edgeEndCounter);
P_FREE(pCutSet);
P_FREE(IndexSorted);
P_FREE(neighbourArraySorted);
P_FREE(startV);
P_FREE(clusterSize);
P_FREE(cutSet);
P_FREE(vStatus);
}
TM_THREAD_EXIT();
}
MAIN(argc, argv)
{
GOTO_REAL();
SETUP_NUMBER_TASKS(10);
/*
* Tuple for Scalable Data Generation
* stores startVertex, endVertex, long weight and other info
*/
graphSDG* SDGdata;
/*
* The graph data structure for this benchmark - see defs.h
*/
graph* G;
#ifdef ENABLE_KERNEL2
/*
* Kernel 2
*/
edge* maxIntWtList;
edge* soughtStrWtList;
long maxIntWtListSize;
long soughtStrWtListSize;
#endif /* ENABLE_KERNEL2 */
#ifdef ENABLE_KERNEL3
# ifndef ENABLE_KERNEL2
# error KERNEL3 requires KERNEL2
# endif
/*
* Kernel 3
*/
V* intWtVList = NULL;
V* strWtVList = NULL;
Vl** intWtVLList = NULL;
Vl** strWtVLList = NULL;
Vd* intWtVDList = NULL;
Vd* strWtVDList = NULL;
#endif /* ENABLE_KERNEL3 */
double totalTime = 0.0;
/* -------------------------------------------------------------------------
* Preamble
* -------------------------------------------------------------------------
*/
/*
* User Interface: Configurable parameters, and global program control
*/
getUserParameters(argc, (char** const) argv);
SIM_GET_NUM_CPU(THREADS);
TM_STARTUP(THREADS, 0);
P_MEMORY_STARTUP(THREADS);
SETUP_NUMBER_THREADS(THREADS);
thread_startup(THREADS);
double time_total = 0.0;
int repeat = REPEATS;
for (; repeat > 0; --repeat) {
SDGdata = (graphSDG*)malloc(sizeof(graphSDG));
assert(SDGdata);
genScalData_seq(SDGdata);
G = (graph*)malloc(sizeof(graph));
assert(G);
computeGraph_arg_t computeGraphArgs;
computeGraphArgs.GPtr = G;
computeGraphArgs.SDGdataPtr = SDGdata;
TIMER_T start;
TIMER_READ(start);
GOTO_SIM();
thread_start(computeGraph, (void*)&computeGraphArgs);
GOTO_REAL();
TIMER_T stop;
TIMER_READ(stop);
double time_tmp = TIMER_DIFF_SECONDS(start, stop);
PRINT_STATS();
time_total += time_tmp;
}
totalTime += time_total;
#ifdef ENABLE_KERNEL2
/* -------------------------------------------------------------------------
* Kernel 2 - Find Max weight and sought string
* -------------------------------------------------------------------------
*/
printf("\nKernel 2 - getStartLists() beginning execution...\n");
maxIntWtListSize = 0;
soughtStrWtListSize = 0;
maxIntWtList = (edge*)malloc(sizeof(edge));
assert(maxIntWtList);
soughtStrWtList = (edge*)malloc(sizeof(edge));
assert(soughtStrWtList);
getStartLists_arg_t getStartListsArg;
getStartListsArg.GPtr = G;
getStartListsArg.maxIntWtListPtr = &maxIntWtList;
getStartListsArg.maxIntWtListSize = &maxIntWtListSize;
getStartListsArg.soughtStrWtListPtr = &soughtStrWtList;
getStartListsArg.soughtStrWtListSize = &soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
getStartLists((void*)&getStartListsArg);
}
#else
thread_start(getStartLists, (void*)&getStartListsArg);
#endif
GOTO_REAL();
TIMER_T stop;
TIMER_READ(stop);
time = TIMER_DIFF_SECONDS(start, stop);
totalTime += time;
printf("\n\tgetStartLists() completed execution.\n");
printf("\nTime taken for kernel 2 is %9.6f sec.\n\n", time);
#endif /* ENABLE_KERNEL2 */
#ifdef ENABLE_KERNEL3
/* -------------------------------------------------------------------------
* Kernel 3 - Graph Extraction
* -------------------------------------------------------------------------
*/
printf("\nKernel 3 - findSubGraphs() beginning execution...\n");
if (K3_DS == 0) {
intWtVList = (V*)malloc(G->numVertices * maxIntWtListSize * sizeof(V));
assert(intWtVList);
strWtVList = (V*)malloc(G->numVertices * soughtStrWtListSize * sizeof(V));
assert(strWtVList);
findSubGraphs0_arg_t findSubGraphs0Arg;
findSubGraphs0Arg.GPtr = G;
findSubGraphs0Arg.intWtVList = intWtVList;
findSubGraphs0Arg.strWtVList = strWtVList;
findSubGraphs0Arg.maxIntWtList = maxIntWtList;
findSubGraphs0Arg.maxIntWtListSize = maxIntWtListSize;
findSubGraphs0Arg.soughtStrWtList = soughtStrWtList;
findSubGraphs0Arg.soughtStrWtListSize = soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
findSubGraphs0((void*)&findSubGraphs0Arg);
}
#else
thread_start(findSubGraphs0, (void*)&findSubGraphs0Arg);
#endif
GOTO_REAL();
TIMER_READ(stop);
} else if (K3_DS == 1) {
intWtVLList = (Vl**)malloc(maxIntWtListSize * sizeof(Vl*));
assert(intWtVLList);
strWtVLList = (Vl**)malloc(soughtStrWtListSize * sizeof(Vl*));
assert(strWtVLList);
findSubGraphs1_arg_t findSubGraphs1Arg;
findSubGraphs1Arg.GPtr = G;
findSubGraphs1Arg.intWtVLList = intWtVLList;
findSubGraphs1Arg.strWtVLList = strWtVLList;
findSubGraphs1Arg.maxIntWtList = maxIntWtList;
findSubGraphs1Arg.maxIntWtListSize = maxIntWtListSize;
findSubGraphs1Arg.soughtStrWtList = soughtStrWtList;
findSubGraphs1Arg.soughtStrWtListSize = soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
findSubGraphs1((void*)&findSubGraphs1Arg);
}
#else
thread_start(findSubGraphs1, (void*)&findSubGraphs1Arg);
#endif
GOTO_REAL();
TIMER_READ(stop);
/* Verification
on_one_thread {
for (i=0; i<maxIntWtListSize; i++) {
printf("%ld -- ", i);
currV = intWtVLList[i];
while (currV != NULL) {
printf("[%ld %ld] ", currV->num, currV->depth);
currV = currV->next;
}
printf("\n");
}
for (i=0; i<soughtStrWtListSize; i++) {
printf("%ld -- ", i);
currV = strWtVLList[i];
while (currV != NULL) {
printf("[%ld %ld] ", currV->num, currV->depth);
currV = currV->next;
}
printf("\n");
}
}
*/
} else if (K3_DS == 2) {
intWtVDList = (Vd *) malloc(maxIntWtListSize * sizeof(Vd));
assert(intWtVDList);
strWtVDList = (Vd *) malloc(soughtStrWtListSize * sizeof(Vd));
assert(strWtVDList);
findSubGraphs2_arg_t findSubGraphs2Arg;
findSubGraphs2Arg.GPtr = G;
findSubGraphs2Arg.intWtVDList = intWtVDList;
findSubGraphs2Arg.strWtVDList = strWtVDList;
findSubGraphs2Arg.maxIntWtList = maxIntWtList;
findSubGraphs2Arg.maxIntWtListSize = maxIntWtListSize;
findSubGraphs2Arg.soughtStrWtList = soughtStrWtList;
findSubGraphs2Arg.soughtStrWtListSize = soughtStrWtListSize;
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
findSubGraphs2((void*)&findSubGraphs2Arg);
}
#else
thread_start(findSubGraphs2, (void*)&findSubGraphs2Arg);
#endif
GOTO_REAL();
TIMER_READ(stop);
/* Verification */
/*
on_one_thread {
printf("\nInt weight sub-graphs \n");
for (i=0; i<maxIntWtListSize; i++) {
printf("%ld -- ", i);
for (j=0; j<intWtVDList[i].numArrays; j++) {
printf("\n [Array %ld] - \n", j);
for (k=0; k<intWtVDList[i].arraySize[j]; k++) {
printf("[%ld %ld] ", intWtVDList[i].vList[j][k].num, intWtVDList[i].vList[j][k].depth);
}
}
printf("\n");
}
printf("\nStr weight sub-graphs \n");
for (i=0; i<soughtStrWtListSize; i++) {
printf("%ld -- ", i);
for (j=0; j<strWtVDList[i].numArrays; j++) {
printf("\n [Array %ld] - \n", j);
for (k=0; k<strWtVDList[i].arraySize[j]; k++) {
printf("[%ld %ld] ", strWtVDList[i].vList[j][k].num, strWtVDList[i].vList[j][k].depth);
}
}
printf("\n");
}
}
*/
} else {
assert(0);
}
time = TIMER_DIFF_SECONDS(start, stop);
totalTime += time;
printf("\n\tfindSubGraphs() completed execution.\n");
printf("\nTime taken for kernel 3 is %9.6f sec.\n\n", time);
#endif /* ENABLE_KERNEL3 */
#ifdef ENABLE_KERNEL4
/* -------------------------------------------------------------------------
* Kernel 4 - Graph Clustering
* -------------------------------------------------------------------------
*/
printf("\nKernel 4 - cutClusters() beginning execution...\n");
TIMER_READ(start);
GOTO_SIM();
#ifdef OTM
#pragma omp parallel
{
cutClusters((void*)G);
}
#else
thread_start(cutClusters, (void*)G);
#endif
GOTO_REAL();
TIMER_READ(stop);
time = TIMER_DIFF_SECONDS(start, stop);
totalTime += time;
printf("\n\tcutClusters() completed execution.\n");
printf("\nTime taken for Kernel 4 is %9.6f sec.\n\n", time);
#endif /* ENABLE_KERNEL4 */
printf("Time = %9.6f \n", totalTime);
/* -------------------------------------------------------------------------
* Cleanup
* -------------------------------------------------------------------------
*/
P_FREE(G->outDegree);
P_FREE(G->outVertexIndex);
P_FREE(G->outVertexList);
P_FREE(G->paralEdgeIndex);
P_FREE(G->inDegree);
P_FREE(G->inVertexIndex);
P_FREE(G->inVertexList);
P_FREE(G->intWeight);
P_FREE(G->strWeight);
#ifdef ENABLE_KERNEL3
LONGINT_T i;
LONGINT_T j;
Vl* currV;
Vl* tempV;
if (K3_DS == 0) {
P_FREE(strWtVList);
P_FREE(intWtVList);
}
if (K3_DS == 1) {
for (i = 0; i < maxIntWtListSize; i++) {
currV = intWtVLList[i];
while (currV != NULL) {
tempV = currV->next;
P_FREE(currV);
currV = tempV;
}
}
for (i = 0; i < soughtStrWtListSize; i++) {
currV = strWtVLList[i];
while (currV != NULL) {
tempV = currV->next;
P_FREE(currV);
currV = tempV;
}
}
P_FREE(strWtVLList);
P_FREE(intWtVLList);
}
if (K3_DS == 2) {
for (i = 0; i < maxIntWtListSize; i++) {
for (j = 0; j < intWtVDList[i].numArrays; j++) {
P_FREE(intWtVDList[i].vList[j]);
}
P_FREE(intWtVDList[i].vList);
P_FREE(intWtVDList[i].arraySize);
}
for (i = 0; i < soughtStrWtListSize; i++) {
for (j = 0; j < strWtVDList[i].numArrays; j++) {
P_FREE(strWtVDList[i].vList[j]);
}
P_FREE(strWtVDList[i].vList);
P_FREE(strWtVDList[i].arraySize);
}
P_FREE(strWtVDList);
P_FREE(intWtVDList);
}
P_FREE(soughtStrWtList);
P_FREE(maxIntWtList);
#endif /* ENABLE_KERNEL2 */
P_FREE(SOUGHT_STRING);
P_FREE(G);
P_FREE(SDGdata);
TM_SHUTDOWN();
P_MEMORY_SHUTDOWN();
GOTO_SIM();
thread_shutdown();
MAIN_RETURN(0);
}
/* =============================================================================
* Pbitmap_free
* =============================================================================
*/
void
Pbitmap_free (bitmap_t* bitmapPtr)
{
P_FREE(bitmapPtr->bits);
P_FREE(bitmapPtr);
}
/*
* n_vars is the number of variables to be considered,
* d is the data array of variables d[0],...,d[n_vars-1],
* pred determines which estimate is required: BLUE, BLUP, or BLP
*/
void gls(DATA **d /* pointer to DATA array */,
int n_vars, /* length of DATA array (to consider) */
enum GLS_WHAT pred, /* what type of prediction is requested */
DPOINT *where, /* prediction location */
double *est /* output: array that holds the predicted values and variances */)
{
GLM *glm = NULL; /* to be copied to/from d */
static MAT *X0 = MNULL, *C0 = MNULL, *MSPE = MNULL, *CinvC0 = MNULL,
*Tmp1 = MNULL, *Tmp2 = MNULL, *Tmp3 = MNULL, *R = MNULL;
static VEC *blup = VNULL, *tmpa = VNULL, *tmpb = VNULL;
PERM *piv = PNULL;
volatile unsigned int i, rows_C;
unsigned int j, k, l = 0, row, col, start_i, start_j, start_X, global,
one_nbh_empty;
VARIOGRAM *v = NULL;
static enum GLS_WHAT last_pred = GLS_INIT; /* the initial value */
double c_value, *X_ori;
int info;
if (d == NULL) { /* clean up */
if (X0 != MNULL) M_FREE(X0);
if (C0 != MNULL) M_FREE(C0);
if (MSPE != MNULL) M_FREE(MSPE);
if (CinvC0 != MNULL) M_FREE(CinvC0);
if (Tmp1 != MNULL) M_FREE(Tmp1);
if (Tmp2 != MNULL) M_FREE(Tmp2);
if (Tmp3 != MNULL) M_FREE(Tmp3);
if (R != MNULL) M_FREE(R);
if (blup != VNULL) V_FREE(blup);
if (tmpa != VNULL) V_FREE(tmpa);
if (tmpb != VNULL) V_FREE(tmpb);
last_pred = GLS_INIT;
return;
}
if (DEBUG_COV) {
printlog("we're at %s X: %g Y: %g Z: %g\n",
IS_BLOCK(where) ? "block" : "point",
where->x, where->y, where->z);
}
if (pred != UPDATE) /* it right away: */
last_pred = pred;
assert(last_pred != GLS_INIT);
if (d[0]->glm == NULL) { /* allocate and initialize: */
glm = new_glm();
d[0]->glm = (void *) glm;
} else
glm = (GLM *) d[0]->glm;
glm->mu0 = v_resize(glm->mu0, n_vars);
MSPE = m_resize(MSPE, n_vars, n_vars);
if (pred == GLS_BLP || UPDATE_BLP) {
X_ori = where->X;
for (i = 0; i < n_vars; i++) { /* mu(0) */
glm->mu0->ve[i] = calc_mu(d[i], where);
blup = v_copy(glm->mu0, v_resize(blup, glm->mu0->dim));
where->X += d[i]->n_X; /* shift to next x0 entry */
}
where->X = X_ori; /* ... and set back */
for (i = 0; i < n_vars; i++) { /* Cij(0,0): */
for (j = 0; j <= i; j++) {
v = get_vgm(LTI(d[i]->id,d[j]->id));
ME(MSPE, i, j) = ME(MSPE, j, i) = COVARIANCE0(v, where, where, d[j]->pp_norm2);
}
}
fill_est(NULL, blup, MSPE, n_vars, est); /* in case of empty neighbourhood */
}
/* xxx */
/*
logprint_variogram(v, 1);
*/
/*
* selection dependent problem dimensions:
*/
for (i = rows_C = 0, one_nbh_empty = 0; i < n_vars; i++) {
rows_C += d[i]->n_sel;
if (d[i]->n_sel == 0)
one_nbh_empty = 1;
}
if (rows_C == 0 /* all selection lists empty */
|| one_nbh_empty == 1) { /* one selection list empty */
if (pred == GLS_BLP || UPDATE_BLP)
debug_result(blup, MSPE, pred);
return;
}
for (i = 0, global = 1; i < n_vars && global; i++)
global = (d[i]->sel == d[i]->list
&& d[i]->n_list == d[i]->n_original
&& d[i]->n_list == d[i]->n_sel);
/*
* global things: enter whenever (a) first time, (b) local selections or
* (c) the size of the problem grew since the last call (e.g. simulation)
*/
if (glm->C == NULL || !global || rows_C > glm->C->m) {
/*
* fill y:
*/
glm->y = get_y(d, glm->y, n_vars);
if (pred != UPDATE) {
glm->C = m_resize(glm->C, rows_C, rows_C);
if (gl_choleski == 0) /* use LDL' decomposition, allocate piv: */
piv = px_resize(piv, rows_C);
m_zero(glm->C);
glm->X = get_X(d, glm->X, n_vars);
M_DEBUG(glm->X, "X");
glm->CinvX = m_resize(glm->CinvX, rows_C, glm->X->n);
glm->XCinvX = m_resize(glm->XCinvX, glm->X->n, glm->X->n);
glm->beta = v_resize(glm->beta, glm->X->n);
for (i = start_X = start_i = 0; i < n_vars; i++) { /* row var */
/* fill C, mu: */
for (j = start_j = 0; j <= i; j++) { /* col var */
v = get_vgm(LTI(d[i]->id,d[j]->id));
for (k = 0; k < d[i]->n_sel; k++) { /* rows */
row = start_i + k;
for (l = 0, col = start_j; col <= row && l < d[j]->n_sel; l++, col++) {
if (pred == GLS_BLUP)
c_value = GCV(v, d[i]->sel[k], d[j]->sel[l]);
else
c_value = COVARIANCE(v, d[i]->sel[k], d[j]->sel[l]);
/* on the diagonal, if necessary, add measurement error variance */
if (d[i]->colnvariance && i == j && k == l)
c_value += d[i]->sel[k]->variance;
ME(glm->C, col, row) = c_value; /* fill upper */
if (col != row)
ME(glm->C, row, col) = c_value; /* fill all */
} /* for l */
} /* for k */
start_j += d[j]->n_sel;
} /* for j */
start_i += d[i]->n_sel;
if (d[i]->n_sel > 0)
start_X += d[i]->n_X - d[i]->n_merge;
} /* for i */
/*
if (d[0]->colnvmu)
glm->C = convert_vmuC(glm->C, d[0]);
*/
if (d[0]->variance_fn) {
glm->mu = get_mu(glm->mu, glm->y, d, n_vars);
convert_C(glm->C, glm->mu, d[0]->variance_fn);
}
if (DEBUG_COV && pred == GLS_BLUP)
printlog("[using generalized covariances: max_val - semivariance()]");
M_DEBUG(glm->C, "Covariances (x_i, x_j) matrix C (upper triangle)");
/*
* factorize C:
*/
CHfactor(glm->C, piv, &info);
if (info != 0) { /* singular: */
pr_warning("Covariance matrix singular at location [%g,%g,%g]: skipping...",
where->x, where->y, where->z);
m_free(glm->C);
glm->C = MNULL; /* assure re-entrance if global */
P_FREE(piv);
return;
}
if (piv == NULL)
M_DEBUG(glm->C, "glm->C, Choleski decomposed:")
else
M_DEBUG(glm->C, "glm->C, LDL' decomposed:")
} /* if (pred != UPDATE) */
/* =============================================================================
* findSubGraphs0
* =============================================================================
*/
void
findSubGraphs0 (void* argPtr)
{
graph* GPtr = ((findSubGraphs0_arg_t*)argPtr)->GPtr;
V* intWtVList = ((findSubGraphs0_arg_t*)argPtr)->intWtVList;
V* strWtVList = ((findSubGraphs0_arg_t*)argPtr)->strWtVList;
edge* maxIntWtList = ((findSubGraphs0_arg_t*)argPtr)->maxIntWtList;
long maxIntWtListSize = ((findSubGraphs0_arg_t*)argPtr)->maxIntWtListSize;
edge* soughtStrWtList = ((findSubGraphs0_arg_t*)argPtr)->soughtStrWtList;
long soughtStrWtListSize = ((findSubGraphs0_arg_t*)argPtr)->soughtStrWtListSize;
long myId = thread_getId();
long numThread = thread_getNumThread();
long i;
long i_start;
long i_stop;
createPartition(0,
(maxIntWtListSize + soughtStrWtListSize),
myId,
numThread,
&i_start,
&i_stop);
for (i = i_start; i < i_stop; i++) {
if (i < maxIntWtListSize) {
unsigned long j;
for (j = 0; j < GPtr->numVertices; j++) {
intWtVList[i*(GPtr->numVertices)+j].num = 0;
intWtVList[i*(GPtr->numVertices)+j].depth = 0;
}
} else {
long t = i - maxIntWtListSize;
unsigned long j;
for (j = 0; j < GPtr->numVertices; j++) {
strWtVList[t*(GPtr->numVertices)+j].num = 0;
strWtVList[t*(GPtr->numVertices)+j].depth = 0;
}
}
}
thread_barrier_wait();
LONGINT_T* visited = (LONGINT_T*)P_MALLOC(GPtr->numVertices * sizeof(LONGINT_T));
assert(visited);
/*
* Each thread runs a BFS from endvertex of maxIntWtList edgeList
*/
for (i = i_start; i < i_stop; i++) {
unsigned long k;
for (k = 0; k < GPtr->numVertices; k++) {
visited[k] = 0;
}
if (i < maxIntWtListSize) {
intWtVList[i*(GPtr->numVertices)+0].num = maxIntWtList[i].startVertex;
intWtVList[i*(GPtr->numVertices)+0].depth = -1;
intWtVList[i*(GPtr->numVertices)+1].num = maxIntWtList[i].endVertex;
intWtVList[i*(GPtr->numVertices)+1].depth = 1;
visited[(intWtVList[i*(GPtr->numVertices)+0]).num] = 1;
visited[(intWtVList[i*(GPtr->numVertices)+1]).num] = 1;
long depth = 1;
long verticesVisited = 2;
long currIndex = 1;
while ((depth < SUBGR_EDGE_LENGTH) ||
(verticesVisited == (long)GPtr->numVertices))
{
long intWtListIndex = i*(GPtr->numVertices)+currIndex;
depth = intWtVList[intWtListIndex].depth + 1;
long j;
long j_start = GPtr->outVertexIndex[intWtVList[intWtListIndex].num];
long j_stop = j_start + GPtr->outDegree[intWtVList[intWtListIndex].num];
for (j = j_start; j < j_stop; j++) {
if (visited[GPtr->outVertexList[j]] == 0) {
visited[GPtr->outVertexList[j]] = 1;
intWtVList[i*(GPtr->numVertices)+verticesVisited].num =
GPtr->outVertexList[j];
intWtVList[i*(GPtr->numVertices)+verticesVisited].depth =
depth;
verticesVisited++;
}
}
if ((currIndex < verticesVisited - 1) &&
(verticesVisited < (long)GPtr->numVertices))
{
currIndex++;
depth = intWtVList[i*(GPtr->numVertices)+currIndex].depth;
} else {
break;
}
}
} else {
long t = i - maxIntWtListSize;
strWtVList[t*(GPtr->numVertices)+0].num =
(soughtStrWtList[t]).startVertex;
strWtVList[t*(GPtr->numVertices)+0].depth = -1;
strWtVList[t*(GPtr->numVertices)+1].num =
(soughtStrWtList[t]).endVertex;
strWtVList[t*(GPtr->numVertices)+1].depth = 1;
visited[(strWtVList[t*(GPtr->numVertices)+0]).num] = 1;
visited[(strWtVList[t*(GPtr->numVertices)+1]).num] = 1;
long depth = 1;
long verticesVisited = 2;
long currIndex = 1;
while ((depth < SUBGR_EDGE_LENGTH) ||
(verticesVisited == (long)GPtr->numVertices))
{
long strWtVListIndex = t*(GPtr->numVertices)+currIndex;
depth = strWtVList[strWtVListIndex].depth + 1;
long j;
long j_start = GPtr->outVertexIndex[strWtVList[strWtVListIndex].num];
long j_stop = j_start + GPtr->outDegree[strWtVList[strWtVListIndex].num];
for (j = j_start; j < j_stop; j++) {
if (visited[GPtr->outVertexList[j]] == 0) {
visited[GPtr->outVertexList[j]] = 1;
strWtVList[t*(GPtr->numVertices)+verticesVisited].num =
GPtr->outVertexList[j];
strWtVList[t*(GPtr->numVertices)+verticesVisited].depth =
depth;
verticesVisited++;
}
}
if (currIndex < verticesVisited - 1) {
currIndex++;
depth = strWtVList[t*(GPtr->numVertices)+currIndex].depth;
} else {
break;
}
}
}
}
P_FREE(visited);
}
/* =============================================================================
* PfreeNode
* =============================================================================
*/
static void
PfreeNode (list_node_t* nodePtr)
{
P_FREE(nodePtr);
}
/* =============================================================================
* Pvector_free
* =============================================================================
*/
void
Pvector_free (vector_t* vectorPtr)
{
P_FREE(vectorPtr->elements);
P_FREE(vectorPtr);
}
/* =============================================================================
* Plist_free
* =============================================================================
*/
void
Plist_free (list_t* listPtr)
{
PfreeList(listPtr->head.nextPtr);
P_FREE(listPtr);
}
/* =============================================================================
* Pqueue_free
* =============================================================================
*/
void
queue_stm_v2::Pqueue_free (queue_t* queuePtr)
{
P_FREE(queuePtr->elements);
P_FREE(queuePtr);
}
/* =============================================================================
* Pqueue_free
* =============================================================================
*/
void
Pqueue_free (queue_t* queuePtr)
{
P_FREE(queuePtr->elements);
P_FREE(queuePtr);
}
