void
func_count (void* argPtr)
{
TM_THREAD_ENTER();
while (1) {
int stop_counting = 0;
//pair_t* coordinatePairPtr;
TM_BEGIN();
long local_counter = (long)TM_SHARED_READ(my_counter);
local_counter++;
if(local_counter > max_count)
stop_counting = 1;
TM_SHARED_WRITE(my_counter, local_counter);
TM_END();
//pthread_yield();
if (stop_counting == 1) {
break;
}
}//end of while
TM_THREAD_EXIT();
}
/* =============================================================================
* 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) {
__transaction_atomic {
long t2 = permV[t];
permV[t] = permV[i];
permV[i] = t2;
}
}
}
void *test_maintenance(void *data) {
#ifdef TINY10B
int i;
free_list_item **t_list_items;
#endif
maintenance_thread_data_t *d = (maintenance_thread_data_t *)data;
#ifdef TINY10B
t_list_items = (free_list_item **)malloc(d->nb_threads * sizeof(free_list_item *));
for(i = 0; i < d->nb_threads; i++) {
t_list_items[i] = d->set->t_free_list[i];
}
#endif
/* Create transaction */
TM_THREAD_ENTER();
/* Wait on barrier */
barrier_cross(d->barrier);
/* Is the first op an update? */
//unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
#ifdef ICC
while (stop == 0) {
#else
while (AO_load_full(&stop) == 0) {
#endif /* ICC */
#ifdef TINY10B
do_maintenance_thread(d->set, d->id, d->nb_maint);
#endif
#ifdef ICC
}
#else
}
#endif /* ICC */
/* Free transaction */
TM_THREAD_EXIT();
return NULL;
}
void *test(void *data) {
int unext, last = -1;
val_t val = 0;
int result;
int id;
ulong *tloc;
#ifdef BIAS_RANGE
val_t increase;
#endif
thread_data_t *d = (thread_data_t *)data;
id = d->id;
tloc = d->set->nb_committed;
#ifdef BIAS_RANGE
increase = d->range;
#endif
/* Create transaction */
TM_THREAD_ENTER();
/* Wait on barrier */
barrier_cross(d->barrier);
/* Is the first op an update? */
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
#ifdef ICC
while (stop == 0) {
#else
while (AO_load_full(&stop) == 0) {
#endif /* ICC */
if (unext) { // update
if (last < 0) { // add
val = rand_range_re(&d->seed, d->range);
#ifdef BIAS_RANGE
if(rand_range_re(&d->seed, 1000) < 50) {
increase += rand_range_re(&d->seed, 10);
if(increase > d->range * 20) {
increase = d->range;
}
val = increase;
}
#endif
if ((result = avl_add(d->set, val, TRANSACTIONAL, id)) > 0) {
d->nb_added++;
if(result > 1) {
d->nb_modifications++;
}
last = val;
}
d->nb_trans++;
tloc[id]++;
d->nb_add++;
} else { // remove
if (d->alternate) { // alternate mode (default)
#ifdef TINY10B
if ((result = avl_remove(d->set, last, TRANSACTIONAL, id)) > 0) {
#else
if ((result = avl_remove(d->set, last, TRANSACTIONAL, 0)) > 0) {
#endif
d->nb_removed++;
#ifdef REMOVE_LATER
finish_removal(d->set, id);
#endif
if(result > 1) {
d->nb_modifications++;
}
}
last = -1;
} else {
/* Random computation only in non-alternated cases */
val = rand_range_re(&d->seed, d->range);
/* Remove one random value */
#ifdef BIAS_RANGE
if(rand_range_re(&d->seed, 1000) < 300) {
//val = d->range + rand_range_re(&d->seed, increase - d->range);
val = increase - rand_range_re(&d->seed, 10);
}
#endif
#ifdef TINY10B
if ((result = avl_remove(d->set, val, TRANSACTIONAL, id)) > 0) {
#else
if ((result = avl_remove(d->set, val, TRANSACTIONAL, 0)) > 0) {
#endif
d->nb_removed++;
#ifdef REMOVE_LATER
finish_removal(d->set, id);
#endif
if(result > 1) {
d->nb_modifications++;
}
/* Repeat until successful, to avoid size variations */
last = -1;
}
}
d->nb_trans++;
tloc[id]++;
d->nb_remove++;
}
} else { // read
if (d->alternate) {
if (d->update == 0) {
if (last < 0) {
val = d->first;
last = val;
} else { // last >= 0
val = rand_range_re(&d->seed, d->range);
last = -1;
}
} else { // update != 0
if (last < 0) {
val = rand_range_re(&d->seed, d->range);
//last = val;
} else {
val = last;
}
}
} else val = rand_range_re(&d->seed, d->range);
#ifdef BIAS_RANGE
if(rand_range_re(&d->seed, 1000) < 100) {
val = increase;
}
#endif
if (avl_contains(d->set, val, TRANSACTIONAL, id))
d->nb_found++;
d->nb_trans++;
tloc[id]++;
d->nb_contains++;
}
/* Is the next op an update? */
if (d->effective) { // a failed remove/add is a read-only tx
unext = ((100 * (d->nb_added + d->nb_removed))
< (d->update * (d->nb_add + d->nb_remove + d->nb_contains)));
} else { // remove/add (even failed) is considered as an update
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
}
#ifdef ICC
}
#else
}
#endif /* ICC */
/* Free transaction */
TM_THREAD_EXIT();
return NULL;
}
void *test_maintenance(void *data) {
#ifdef TINY10B
int i;
free_list_item **t_list_items;
#endif
maintenance_thread_data_t *d = (maintenance_thread_data_t *)data;
#ifdef TINY10B
t_list_items = (free_list_item **)malloc(d->nb_threads * sizeof(free_list_item *));
for(i = 0; i < d->nb_threads; i++) {
t_list_items[i] = d->set->t_free_list[i];
}
#endif
/* Create transaction */
TM_THREAD_ENTER();
/* Wait on barrier */
barrier_cross(d->barrier);
/* Is the first op an update? */
//unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
#ifdef ICC
while (stop == 0) {
#else
while (AO_load_full(&stop) == 0) {
#endif /* ICC */
#ifdef TINY10B
do_maintenance_thread(d->set, d->id, d->nb_maint);
#endif
#ifdef ICC
}
#else
}
#endif /* ICC */
/* Free transaction */
TM_THREAD_EXIT();
return NULL;
}
void catcher(int sig)
{
printf("CAUGHT SIGNAL %d\n", sig);
}
/* =============================================================================
* router_solve
* =============================================================================
*/
void
router_solve (void* argPtr)
{
TM_THREAD_ENTER();
router_solve_arg_t* routerArgPtr = (router_solve_arg_t*)argPtr;
router_t* routerPtr = routerArgPtr->routerPtr;
maze_t* mazePtr = routerArgPtr->mazePtr;
vector_t* myPathVectorPtr = PVECTOR_ALLOC(1);
assert(myPathVectorPtr);
queue_t* workQueuePtr = mazePtr->workQueuePtr;
grid_t* gridPtr = mazePtr->gridPtr;
grid_t* myGridPtr =
PGRID_ALLOC(gridPtr->width, gridPtr->height, gridPtr->depth);
assert(myGridPtr);
long bendCost = routerPtr->bendCost;
queue_t* myExpansionQueuePtr = PQUEUE_ALLOC(-1);
/*
* Iterate over work list to route each path. This involves an
* 'expansion' and 'traceback' phase for each source/destination pair.
*/
while (1) {
pair_t* coordinatePairPtr;
int mode = 0;
TM_BEGIN(0,mode);
if (mode == 0) {
if (queue_htm::TMqueue_isEmpty(TM_ARG workQueuePtr)) {
coordinatePairPtr = NULL;
} else {
coordinatePairPtr = (pair_t*)queue_htm::TMqueue_pop(TM_ARG workQueuePtr);
}
} else {
if (queue_stm::TMqueue_isEmpty(TM_ARG workQueuePtr)) {
coordinatePairPtr = NULL;
} else {
coordinatePairPtr = (pair_t*)queue_stm::TMqueue_pop(TM_ARG workQueuePtr);
}
}
TM_END();
if (coordinatePairPtr == NULL) {
break;
}
coordinate_t* srcPtr = coordinatePairPtr->firstPtr;
coordinate_t* dstPtr = coordinatePairPtr->secondPtr;
bool_t success = FALSE;
vector_t* pointVectorPtr = NULL;
mode = 0;
TM_BEGIN(1,mode);
if (mode == 0) {
grid_copy(myGridPtr, gridPtr); /* ok if not most up-to-date */
if (PdoExpansion(routerPtr, myGridPtr, myExpansionQueuePtr,
srcPtr, dstPtr)) {
pointVectorPtr = PdoTraceback(gridPtr, myGridPtr, dstPtr, bendCost);
/*
* TODO: fix memory leak
*
* pointVectorPtr will be a memory leak if we abort this transaction
*/
if (pointVectorPtr) {
TMGRID_ADDPATH_HTM(gridPtr, pointVectorPtr);
TM_LOCAL_WRITE(success, TRUE);
}
}
} else {
grid_copy(myGridPtr, gridPtr); /* ok if not most up-to-date */
if (PdoExpansion(routerPtr, myGridPtr, myExpansionQueuePtr,
srcPtr, dstPtr)) {
pointVectorPtr = PdoTraceback(gridPtr, myGridPtr, dstPtr, bendCost);
/*
* TODO: fix memory leak
*
* pointVectorPtr will be a memory leak if we abort this transaction
*/
if (pointVectorPtr) {
TMGRID_ADDPATH_STM(gridPtr, pointVectorPtr);
TM_LOCAL_WRITE(success, TRUE);
}
}
}
TM_END();
if (success) {
bool_t status = PVECTOR_PUSHBACK(myPathVectorPtr,
(void*)pointVectorPtr);
assert(status);
}
}
/*
* Add my paths to global list
*/
list_t* pathVectorListPtr = routerArgPtr->pathVectorListPtr;
int mode = 0;
TM_BEGIN(2,mode);
if (mode == 0) {
list_htm::TMlist_insert(TM_ARG pathVectorListPtr, (void*)myPathVectorPtr);
} else {
list_stm::TMlist_insert(TM_ARG pathVectorListPtr, (void*)myPathVectorPtr);
}
TM_END();
PGRID_FREE(myGridPtr);
PQUEUE_FREE(myExpansionQueuePtr);
#if DEBUG
puts("\nFinal Grid:");
grid_print(gridPtr);
#endif /* DEBUG */
TM_THREAD_EXIT();
}
void client_run (void* argPtr) {
TM_THREAD_ENTER();
random_t* randomPtr = random_alloc();
random_seed(randomPtr, time(0));
// unsigned long myId = thread_getId();
// long numThread = *((long*)argPtr);
long operations = (long)global_params[PARAM_OPERATIONS] / (long)global_params[PARAM_THREADS];
long interval = (long)global_params[PARAM_INTERVAL];
printf("operations: %ld \tinterval: %ld\n", operations, interval);
long total = 0;
long total2 = 0;
long i = 0;
unsigned int cont_size = (unsigned int) global_params[PARAM_CONTENTION];
unsigned int* sorted_locks = (unsigned int*) malloc((2 + cont_size) * sizeof(int));
unsigned int* read_idxs = (unsigned int*) malloc(cont_size * sizeof(int));
for (; i < operations; i++) {
long random_number = ((long) random_generate(randomPtr)) % ((long)global_params[PARAM_SIZE]);
long random_number2 = ((long) random_generate(randomPtr)) % ((long)global_params[PARAM_SIZE]);
if (random_number == random_number2) {
random_number2 = (random_number2 + 1) % ((long)global_params[PARAM_SIZE]);
}
int repeat = 0;
for (; repeat < cont_size; repeat++) {
read_idxs[repeat] = ((unsigned int) random_generate(randomPtr)) % ((unsigned int)global_params[PARAM_SIZE]);
LI_HASH(&global_array[read_idxs[repeat]], &sorted_locks[repeat + 2]);
}
// TM_BEGIN();
LI_HASH(&global_array[random_number], &sorted_locks[0]);
LI_HASH(&global_array[random_number2], &sorted_locks[1]);
TM_BEGIN_ARGS(sorted_locks, cont_size + 2);
long r1 = (long)TM_SHARED_READ(global_array[random_number].value);
long r2 = (long)TM_SHARED_READ(global_array[random_number2].value);
for (repeat--; repeat >= 0; repeat--) {
total2 += (long) TM_SHARED_READ(global_array[read_idxs[repeat]].value);
}
r1 = r1 + 1;
r2 = r2 - 1;
int f = 1;
int ii;
for(ii = 1; ii <= ((unsigned int) global_params[PARAM_WORK]); ii++)
{
f *= ii;
}
total += f / 1000000;
TM_SHARED_WRITE(global_array[random_number].value, r1);
TM_SHARED_WRITE(global_array[random_number2].value, r2);
TM_END_ARGS(sorted_locks, cont_size + 2);
long k = 0;
for (;k < (long)global_params[PARAM_INTERVAL]; k++) {
long ru = ((long) random_generate(randomPtr)) % 2;
total += ru;
}
}
TM_THREAD_EXIT();
printf("ru ignore %ld - %ld\n", total, total2);
}
/* =============================================================================
* 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();
}
/* =============================================================================
* 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();
}
void* test(void *data) {
int unext, last = -1;
val_t val = 0;
pval_t pval = 0;
thread_data_t *d = (thread_data_t *)data;
/* Create transaction */
TM_THREAD_ENTER(d->id);
set_cpu(the_cores[d->id]);
/* Wait on barrier */
ssalloc_init();
PF_CORRECTION;
seeds = seed_rand();
#ifdef PIN
int id = d->id;
int cpu = 40*(id/40) + 4*(id%10) + (id%40)/10;
// printf("Pinning %d to %d\n",id,cpu);
pin(pthread_self(), cpu);
// pin(pthread_self(), id);
#endif
#ifdef PAPI
if (PAPI_OK != PAPI_start_counters(g_events, G_EVENT_COUNT))
{
printf("Problem starting counters 1.");
}
#endif
barrier_cross(d->barrier);
/* Is the first op an update? */
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
#ifdef DISTRIBUTION_EXPERIMENT
while (1)
#else
while (*running)
#endif
{
if (d->es) { // event simulator experiment
if (d->lin) {
if (!empty(d->linden_set)) {
d->nb_remove++;
pval_t pval = deletemin(d->linden_set, d);
d->nb_removed++;
// printf("%d %d\n", pval, deps[pval][0]);
int i = 0;
val_t dep;
while ((dep = deps[pval][i]) != -1 && i < MAX_DEPS) {
d->nb_add++;
if (insert(d->linden_set, dep, dep)) {
d->nb_added++;
}
i++;
}
}
} else {
if (d->set->head->next[0]->next[0] != NULL) {// set not empty
d->nb_remove++;
if (d->sl) { // spray list
if (spray_delete_min(d->set, &val, d)) {
d->nb_removed++;
} else {
continue;
}
} else if (d->pq) { // lotan_shavit pq
if (lotan_shavit_delete_min(d->set, &val, d)) {
d->nb_removed++;
// continue; // TODO: maybe try remove this to simulate task handling (dependency checks still occur)
} else {
continue;
}
}
// struct timespec ten_usec;
// ten_usec.tv_sec = 0;
// ten_usec.tv_nsec = 10000;
// nanosleep(&ten_usec, NULL);
// dependency handling
int i = 0;
val_t dep;
while ((dep = deps[val][i]) != -1 && i < MAX_DEPS) {
if (!sl_contains(d->set, dep, TRANSACTIONAL)) { // dependent has been removed, need to add it again
if (sl_add(d->set, dep, TRANSACTIONAL)) { // check if insert actually succeeded (otherwise someone else did it first)
d->nb_added++;
}
d->nb_add++;
}
i++;
}
}
}
} else { // not event simulator
if (unext) { // update
if (last < 0) { // add
val = rand_range_re(&d->seed, d->range);
if (d->lin) {
pval = val;
insert(d->linden_set, pval, pval);
d->nb_added++;
last = pval;
} else { // not linden
if (sl_add(d->set, val, TRANSACTIONAL)) {
d->nb_added++;
last = val;
}
}
d->nb_add++;
} else { // remove
if (d->pq) {
if (lotan_shavit_delete_min(d->set, &val, d)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
}
last = -1;
}
else if (d->sl) {
if (spray_delete_min(d->set, &val, d)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
last = -1;
}
}
else if (d->lin) {
if ((pval = deletemin(d->linden_set, d))) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = pval;
}
last = -1;
}
}
else if (d->alternate) { // alternate mode (default)
if (sl_remove(d->set, last, TRANSACTIONAL)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
}
last = -1;
} else {
/* Random computation only in non-alternated cases */
val = rand_range_re(&d->seed, d->range);
/* Remove one random value */
if (sl_remove_succ(d->set, val, TRANSACTIONAL)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
/* Repeat until successful, to avoid size variations */
last = -1;
}
}
d->nb_remove++;
}
} else { // read
if (d->alternate) {
if (d->update == 0) {
if (last < 0) {
val = d->first;
last = val;
} else { // last >= 0
val = rand_range_re(&d->seed, d->range);
last = -1;
}
} else { // update != 0
if (last < 0) {
val = rand_range_re(&d->seed, d->range);
//last = val;
} else {
val = last;
}
}
} else val = rand_range_re(&d->seed, d->range);
PF_START(2);
if (sl_contains(d->set, val, TRANSACTIONAL))
d->nb_found++;
PF_STOP(2);
d->nb_contains++;
}
/* Is the next op an update? */
if (d->effective) { // a failed remove/add is a read-only tx
unext = ((100 * (d->nb_added + d->nb_removed))
< (d->update * (d->nb_add + d->nb_remove + d->nb_contains)));
} else { // remove/add (even failed) is considered as an update
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
}
}
#ifdef DISTRIBUTION_EXPERIMENT
if (d->first_remove != -1) {
break; //only one run
}
#endif
}
#ifdef PAPI
if (PAPI_OK != PAPI_read_counters(g_values[d->id], G_EVENT_COUNT))
{
printf("Problem reading counters 2.");
}
#endif
/* Free transaction */
TM_THREAD_EXIT();
PF_PRINT;
return NULL;
}
/* =============================================================================
* process
* =============================================================================
*/
void
process ()
{
TM_THREAD_ENTER();
heap_t* workHeapPtr = global_workHeapPtr;
mesh_t* meshPtr = global_meshPtr;
region_t* regionPtr;
long totalNumAdded = 0;
long numProcess = 0;
regionPtr = PREGION_ALLOC();
assert(regionPtr);
while (1) {
element_t* elementPtr;
AL_LOCK(0);
TM_BEGIN(0);
elementPtr = TMHEAP_REMOVE(workHeapPtr);
TM_END();
if (elementPtr == NULL) {
break;
}
bool_t isGarbage;
AL_LOCK(0);
TM_BEGIN(1);
isGarbage = TMELEMENT_ISGARBAGE(elementPtr);
TM_END();
if (isGarbage) {
/*
* Handle delayed deallocation
*/
PELEMENT_FREE(elementPtr);
continue;
}
long numAdded;
AL_LOCK(0);
TM_BEGIN(2);
PREGION_CLEARBAD(regionPtr);
numAdded = TMREGION_REFINE(regionPtr, elementPtr, meshPtr);
TM_END();
AL_LOCK(0);
TM_BEGIN(3);
TMELEMENT_SETISREFERENCED(elementPtr, FALSE);
isGarbage = TMELEMENT_ISGARBAGE(elementPtr);
TM_END();
if (isGarbage) {
/*
* Handle delayed deallocation
*/
PELEMENT_FREE(elementPtr);
}
totalNumAdded += numAdded;
AL_LOCK(0);
TM_BEGIN(4);
TMREGION_TRANSFERBAD(regionPtr, workHeapPtr);
TM_END();
numProcess++;
}
AL_LOCK(0);
TM_BEGIN(5);
TM_SHARED_WRITE(global_totalNumAdded,
TM_SHARED_READ(global_totalNumAdded) + totalNumAdded);
TM_SHARED_WRITE(global_numProcess,
TM_SHARED_READ(global_numProcess) + numProcess);
TM_END();
PREGION_FREE(regionPtr);
TM_THREAD_EXIT();
}
/* =============================================================================
* main
* =============================================================================
*/
MAIN(argc, argv)
{
int max_nclusters = 13;
int min_nclusters = 4;
char* filename = 0;
float* buf;
float** attributes;
float** cluster_centres = NULL;
int i;
int j;
int best_nclusters;
int* cluster_assign;
int numAttributes;
int numObjects;
int use_zscore_transform = 1;
char* line;
int isBinaryFile = 0;
int nloops;
int len;
int nthreads;
float threshold = 0.001;
int opt;
nthreads = 1;
while ((opt = getopt(argc,(char**)argv,"p:i:m:n:t:bz")) != EOF) {
switch (opt) {
case 'i': filename = optarg;
break;
case 'b': isBinaryFile = 1;
break;
case 't': threshold = atof(optarg);
break;
case 'm': max_nclusters = atoi(optarg);
break;
case 'n': min_nclusters = atoi(optarg);
break;
case 'z': use_zscore_transform = 0;
break;
case 'p': nthreads = atoi(optarg);
break;
case '?': usage((char*)argv[0]);
break;
default: usage((char*)argv[0]);
break;
}
}
// [RSTM] moved this allocation so that we only allocate after
// an MMPolicy has been created
SIM_GET_NUM_CPU(nthreads);
TM_STARTUP(nthreads);
P_MEMORY_STARTUP(numThread);
TM_THREAD_ENTER();
TM_BEGIN();
line = (char*)SEQ_MALLOC(MAX_LINE_LENGTH); /* reserve memory line */
if (filename == 0) {
usage((char*)argv[0]);
}
if (max_nclusters < min_nclusters) {
fprintf(stderr, "Error: max_clusters must be >= min_clusters\n");
usage((char*)argv[0]);
}
numAttributes = 0;
numObjects = 0;
/*
* From the input file, get the numAttributes and numObjects
*/
if (isBinaryFile) {
int infile;
if ((infile = open(filename, O_RDONLY, "0600")) == -1) {
fprintf(stderr, "Error: no such file (%s)\n", filename);
exit(1);
}
read(infile, &numObjects, sizeof(int));
read(infile, &numAttributes, sizeof(int));
/* Allocate space for attributes[] and read attributes of all objects */
buf = (float*)SEQ_MALLOC(numObjects * numAttributes * sizeof(float));
assert(buf);
attributes = (float**)SEQ_MALLOC(numObjects * sizeof(float*));
assert(attributes);
attributes[0] = (float*)SEQ_MALLOC(numObjects * numAttributes * sizeof(float));
assert(attributes[0]);
for (i = 1; i < numObjects; i++) {
attributes[i] = attributes[i-1] + numAttributes;
}
read(infile, buf, (numObjects * numAttributes * sizeof(float)));
close(infile);
} else {
FILE *infile;
if ((infile = fopen(filename, "r")) == NULL) {
fprintf(stderr, "Error: no such file (%s)\n", filename);
exit(1);
}
while (fgets(line, MAX_LINE_LENGTH, infile) != NULL) {
if (strtok(line, " \t\n") != 0) {
numObjects++;
}
}
rewind(infile);
while (fgets(line, MAX_LINE_LENGTH, infile) != NULL) {
if (strtok(line, " \t\n") != 0) {
/* Ignore the id (first attribute): numAttributes = 1; */
while (strtok(NULL, " ,\t\n") != NULL) {
numAttributes++;
}
break;
}
}
/* Allocate space for attributes[] and read attributes of all objects */
buf = (float*)SEQ_MALLOC(numObjects * numAttributes * sizeof(float));
assert(buf);
attributes = (float**)SEQ_MALLOC(numObjects * sizeof(float*));
assert(attributes);
attributes[0] = (float*)SEQ_MALLOC(numObjects * numAttributes * sizeof(float));
assert(attributes[0]);
for (i = 1; i < numObjects; i++) {
attributes[i] = attributes[i-1] + numAttributes;
}
rewind(infile);
i = 0;
while (fgets(line, MAX_LINE_LENGTH, infile) != NULL) {
if (strtok(line, " \t\n") == NULL) {
continue;
}
for (j = 0; j < numAttributes; j++) {
buf[i] = atof(strtok(NULL, " ,\t\n"));
i++;
}
}
fclose(infile);
}
/*
* The core of the clustering
*/
cluster_assign = (int*)SEQ_MALLOC(numObjects * sizeof(int));
assert(cluster_assign);
nloops = 1;
len = max_nclusters - min_nclusters + 1;
TM_END();
thread_startup(nthreads);
for (i = 0; i < nloops; i++) {
/*
* Since zscore transform may perform in cluster() which modifies the
* contents of attributes[][], we need to re-store the originals
*/
memcpy(attributes[0], buf, (numObjects * numAttributes * sizeof(float)));
//thread_barrier_wait();
cluster_centres = NULL;
cluster_exec(nthreads,
numObjects,
numAttributes,
attributes, /* [numObjects][numAttributes] */
use_zscore_transform, /* 0 or 1 */
min_nclusters, /* pre-define range from min to max */
max_nclusters,
threshold,
&best_nclusters, /* return: number between min and max */
&cluster_centres, /* return: [best_nclusters][numAttributes] */
cluster_assign); /* return: [numObjects] cluster id for each object */
}
#ifdef GNUPLOT_OUTPUT
{
FILE** fptr;
char outFileName[1024];
fptr = (FILE**)SEQ_MALLOC(best_nclusters * sizeof(FILE*));
for (i = 0; i < best_nclusters; i++) {
sprintf(outFileName, "group.%d", i);
fptr[i] = fopen(outFileName, "w");
}
for (i = 0; i < numObjects; i++) {
fprintf(fptr[cluster_assign[i]],
"%6.4f %6.4f\n",
attributes[i][0],
attributes[i][1]);
}
for (i = 0; i < best_nclusters; i++) {
fclose(fptr[i]);
}
SEQ_FREE(fptr);
}
#endif /* GNUPLOT_OUTPUT */
#ifdef OUTPUT_TO_FILE
{
/* Output: the coordinates of the cluster centres */
FILE* cluster_centre_file;
FILE* clustering_file;
char outFileName[1024];
sprintf(outFileName, "%s.cluster_centres", filename);
cluster_centre_file = fopen(outFileName, "w");
for (i = 0; i < best_nclusters; i++) {
fprintf(cluster_centre_file, "%d ", i);
for (j = 0; j < numAttributes; j++) {
fprintf(cluster_centre_file, "%f ", cluster_centres[i][j]);
}
fprintf(cluster_centre_file, "\n");
}
fclose(cluster_centre_file);
/* Output: the closest cluster centre to each of the data points */
sprintf(outFileName, "%s.cluster_assign", filename);
clustering_file = fopen(outFileName, "w");
for (i = 0; i < numObjects; i++) {
fprintf(clustering_file, "%d %d\n", i, cluster_assign[i]);
}
fclose(clustering_file);
}
#endif /* OUTPUT TO_FILE */
#ifdef OUTPUT_TO_STDOUT
{
/* Output: the coordinates of the cluster centres */
for (i = 0; i < best_nclusters; i++) {
printf("%d ", i);
for (j = 0; j < numAttributes; j++) {
printf("%f ", cluster_centres[i][j]);
}
printf("\n");
}
}
#endif /* OUTPUT TO_STDOUT */
printf("Time: %lg seconds\n", global_time);
SEQ_FREE(cluster_assign);
SEQ_FREE(attributes);
SEQ_FREE(cluster_centres[0]);
SEQ_FREE(cluster_centres);
SEQ_FREE(buf);
TM_SHUTDOWN();
thread_shutdown();
MAIN_RETURN(0);
}
/* =============================================================================
* 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();
}
void *test2(void *data)
{
int val, newval, last, flag = 1;
int id;
ulong *tloc;
thread_data_t *d = (thread_data_t *)data;
id = d->id;
tloc = d->set->nb_committed;
/* Create transaction */
TM_THREAD_ENTER();
/* Wait on barrier */
barrier_cross(d->barrier);
last = 0; // to avoid warning
while (stop == 0) {
val = rand_range_re(&d->seed, 100) - 1;
/* added for HashTables */
if (val < d->update) {
if (val >= d->move) { /* update without move */
if (flag) {
/* Add random value */
val = (rand_r(&d->seed) % d->range) + 1;
if (avl_add(d->set, val, TRANSACTIONAL, id)) {
d->nb_added++;
last = val;
flag = 0;
}
d->nb_trans++;
tloc[id]++;
d->nb_add++;
} else {
if (d->alternate) {
/* Remove last value */
if (avl_remove(d->set, last, TRANSACTIONAL, id))
d->nb_removed++;
d->nb_trans++;
tloc[id]++;
d->nb_remove++;
flag = 1;
} else {
/* Random computation only in non-alternated cases */
newval = rand_range_re(&d->seed, d->range);
if (avl_remove(d->set, newval, TRANSACTIONAL, id)) {
d->nb_removed++;
/* Repeat until successful, to avoid size variations */
flag = 1;
}
d->nb_trans++;
tloc[id]++;
d->nb_remove++;
}
}
} else { /* move */
val = rand_range_re(&d->seed, d->range);
if (avl_move(d->set, last, val, TRANSACTIONAL, id)) {
d->nb_moved++;
last = val;
}
d->nb_trans++;
tloc[id]++;
d->nb_move++;
}
} else {
if (val >= d->update + d->snapshot) { /* read-only without snapshot */
/* Look for random value */
val = rand_range_re(&d->seed, d->range);
if (avl_contains(d->set, val, TRANSACTIONAL, id))
d->nb_found++;
d->nb_trans++;
tloc[id]++;
d->nb_contains++;
} else { /* snapshot */
if (avl_snapshot(d->set, TRANSACTIONAL, id))
d->nb_snapshoted++;
d->nb_trans++;
tloc[id]++;
d->nb_snapshot++;
}
}
}
/* Free transaction */
TM_THREAD_EXIT();
return NULL;
}
/* =============================================================================
* router_solve
* =============================================================================
*/
void
router_solve (void* argPtr)
{
TM_THREAD_ENTER();
router_solve_arg_t* routerArgPtr = (router_solve_arg_t*)argPtr;
router_t* routerPtr = routerArgPtr->routerPtr;
maze_t* mazePtr = routerArgPtr->mazePtr;
vector_t* myPathVectorPtr = PVECTOR_ALLOC(1);
assert(myPathVectorPtr);
queue_t* workQueuePtr = mazePtr->workQueuePtr;
grid_t* gridPtr = mazePtr->gridPtr;
grid_t* myGridPtr =
PGRID_ALLOC(gridPtr->width, gridPtr->height, gridPtr->depth);
assert(myGridPtr);
long bendCost = routerPtr->bendCost;
queue_t* myExpansionQueuePtr = PQUEUE_ALLOC(-1);
/*
* Iterate over work list to route each path. This involves an
* 'expansion' and 'traceback' phase for each source/destination pair.
*/
while (1) {
pair_t* coordinatePairPtr;
TM_BEGIN();
if (TMQUEUE_ISEMPTY(workQueuePtr)) {
coordinatePairPtr = NULL;
} else {
coordinatePairPtr = (pair_t*)TMQUEUE_POP(workQueuePtr);
}
TM_END();
if (coordinatePairPtr == NULL) {
break;
}
coordinate_t* srcPtr = (coordinate_t*)coordinatePairPtr->firstPtr;
coordinate_t* dstPtr = (coordinate_t*)coordinatePairPtr->secondPtr;
bool success = false;
vector_t* pointVectorPtr = NULL;
TM_BEGIN();
grid_copy(myGridPtr, gridPtr); /* ok if not most up-to-date */
if (PdoExpansion(routerPtr, myGridPtr, myExpansionQueuePtr,
srcPtr, dstPtr)) {
pointVectorPtr = PdoTraceback(gridPtr, myGridPtr, dstPtr, bendCost);
/*
* TODO: fix memory leak
*
* pointVectorPtr will be a memory leak if we abort this transaction
*/
if (pointVectorPtr) {
TMGRID_ADDPATH(gridPtr, pointVectorPtr);
TM_LOCAL_WRITE_L(success, true);
}
}
TM_END();
if (success) {
bool status = PVECTOR_PUSHBACK(myPathVectorPtr, (void*)pointVectorPtr);
assert(status);
}
}
/*
* Add my paths to global list
*/
list_t* pathVectorListPtr = routerArgPtr->pathVectorListPtr;
TM_BEGIN();
TMLIST_INSERT(pathVectorListPtr, (void*)myPathVectorPtr);
TM_END();
PGRID_FREE(myGridPtr);
PQUEUE_FREE(myExpansionQueuePtr);
#if DEBUG
puts("\nFinal Grid:");
grid_print(gridPtr);
#endif /* DEBUG */
TM_THREAD_EXIT();
}
/* =============================================================================
* process
* =============================================================================
*/
void
process ()
{
TM_THREAD_ENTER();
heap_t* workHeapPtr = global_workHeapPtr;
mesh_t* meshPtr = global_meshPtr;
region_t* regionPtr;
long totalNumAdded = 0;
long numProcess = 0;
regionPtr = PREGION_ALLOC();
assert(regionPtr);
while (1) {
element_t* elementPtr;
__transaction_atomic {
elementPtr = TMHEAP_REMOVE(workHeapPtr);
}
if (elementPtr == NULL) {
break;
}
bool_t isGarbage;
__transaction_atomic {
isGarbage = TMELEMENT_ISGARBAGE(elementPtr);
}
if (isGarbage) {
/*
* Handle delayed deallocation
*/
PELEMENT_FREE(elementPtr);
continue;
}
long numAdded;
__transaction_atomic {
PREGION_CLEARBAD(regionPtr);
numAdded = TMREGION_REFINE(regionPtr, elementPtr, meshPtr);
}
__transaction_atomic {
TMELEMENT_SETISREFERENCED(elementPtr, FALSE);
isGarbage = TMELEMENT_ISGARBAGE(elementPtr);
}
if (isGarbage) {
/*
* Handle delayed deallocation
*/
PELEMENT_FREE(elementPtr);
}
totalNumAdded += numAdded;
__transaction_atomic {
TMREGION_TRANSFERBAD(regionPtr, workHeapPtr);
}
numProcess++;
}
__transaction_atomic {
global_totalNumAdded = global_totalNumAdded + totalNumAdded;
global_numProcess = global_numProcess + numProcess;
}
PREGION_FREE(regionPtr);
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();
}
void *test(void *data) {
int val2, numtx, r, last = -1;
val_t val = 0;
int unext, mnext, cnext;
thread_data_t *d = (thread_data_t *)data;
/* Create transaction */
TM_THREAD_ENTER();
/* Wait on barrier */
barrier_cross(d->barrier);
/* Is the first op an update, a move? */
r = rand_range_re(&d->seed, 100) - 1;
unext = (r < d->update);
mnext = (r < d->move);
cnext = (r >= d->update + d->snapshot);
#ifdef ICC
while (stop == 0) {
#else
while (AO_load_full(&stop) == 0) {
#endif /* ICC */
if (unext) { // update
if (mnext) { // move
if (last == -1) val = rand_range_re(&d->seed, d->range);
val2 = rand_range_re(&d->seed, d->range);
if (ht_move(d->set, val, val2, TRANSACTIONAL)) {
d->nb_moved++;
last = val2;
}
d->nb_move++;
} else if (last < 0) { // add
val = rand_range_re(&d->seed, d->range);
if (ht_add(d->set, val, TRANSACTIONAL)) {
d->nb_added++;
last = val;
}
d->nb_add++;
} else { // remove
if (d->alternate) { // alternate mode
if (ht_remove(d->set, last, TRANSACTIONAL)) {
d->nb_removed++;
last = -1;
}
} else {
/* Random computation only in non-alternated cases */
val = rand_range_re(&d->seed, d->range);
/* Remove one random value */
if (ht_remove(d->set, val, TRANSACTIONAL)) {
d->nb_removed++;
/* Repeat until successful, to avoid size variations */
last = -1;
}
}
d->nb_remove++;
}
} else { // reads
if (cnext) { // contains (no snapshot)
if (d->alternate) {
if (d->update == 0) {
if (last < 0) {
val = d->first;
last = val;
} else { // last >= 0
val = rand_range_re(&d->seed, d->range);
last = -1;
}
} else { // update != 0
if (last < 0) {
val = rand_range_re(&d->seed, d->range);
//last = val;
} else {
val = last;
}
}
} else val = rand_range_re(&d->seed, d->range);
if (ht_contains(d->set, val, TRANSACTIONAL))
d->nb_found++;
d->nb_contains++;
} else { // snapshot
if (ht_snapshot(d->set, TRANSACTIONAL))
d->nb_snapshoted++;
d->nb_snapshot++;
}
}
/* Is the next op an update, a move, a contains? */
if (d->effective) { // a failed remove/add is a read-only tx
numtx = d->nb_contains + d->nb_add + d->nb_remove + d->nb_move + d->nb_snapshot;
unext = ((100.0 * (d->nb_added + d->nb_removed + d->nb_moved)) < (d->update * numtx));
mnext = ((100.0 * d->nb_moved) < (d->move * numtx));
cnext = !((100.0 * d->nb_snapshoted) < (d->snapshot * numtx));
} else { // remove/add (even failed) is considered as an update
r = rand_range_re(&d->seed, 100) - 1;
unext = (r < d->update);
mnext = (r < d->move);
cnext = (r >= d->update + d->snapshot);
}
#ifdef ICC
}
#else
}
#endif /* ICC */
/* Free transaction */
TM_THREAD_EXIT();
return NULL;
}
/* =============================================================================
* main
* =============================================================================
*/
MAIN (argc,argv)
{
TIMER_T start;
TIMER_T stop;
/* Initialization */
parseArgs(argc, (char** const)argv);
SIM_GET_NUM_CPU(global_params[PARAM_THREAD]);
printf("Creating gene and segments... ");
fflush(stdout);
long geneLength = global_params[PARAM_GENE];
long segmentLength = global_params[PARAM_SEGMENT];
long minNumSegment = global_params[PARAM_NUMBER];
long numThread = global_params[PARAM_THREAD];
random_t* randomPtr;
gene_t* genePtr;
char* gene;
segments_t* segmentsPtr;
sequencer_t* sequencerPtr;
TM_STARTUP(numThread);
P_MEMORY_STARTUP(numThread);
TM_THREAD_ENTER();
// TM_BEGIN();
randomPtr= random_alloc();
assert(randomPtr != NULL);
random_seed(randomPtr, 0);
genePtr = gene_alloc(geneLength);
assert( genePtr != NULL);
gene_create(genePtr, randomPtr);
gene = genePtr->contents;
segmentsPtr = segments_alloc(segmentLength, minNumSegment);
assert(segmentsPtr != NULL);
segments_create(segmentsPtr, genePtr, randomPtr);
sequencerPtr = sequencer_alloc(geneLength, segmentLength, segmentsPtr);
assert(sequencerPtr != NULL);
//TM_END();
thread_startup(numThread);
puts("done.");
printf("Gene length = %li\n", genePtr->length);
printf("Segment length = %li\n", segmentsPtr->length);
printf("Number segments = %li\n", vector_getSize(segmentsPtr->contentsPtr));
fflush(stdout);
/* Benchmark */
printf("Sequencing gene... ");
fflush(stdout);
// 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_READ(start);
#ifdef OTM
#pragma omp parallel
{
sequencer_run(sequencerPtr);
}
#else
thread_start(sequencer_run, (void*)sequencerPtr);
#endif
TIMER_READ(stop);
// NB: As above, timer reads must be done inside of the simulated region
// for PTLSim/ASF
GOTO_REAL();
puts("done.");
printf("Time = %lf\n", TIMER_DIFF_SECONDS(start, stop));
fflush(stdout);
/* Check result */
{
char* sequence;
int result;
//TM_BEGIN();
sequence= sequencerPtr->sequence;
result = strcmp(gene, sequence);
//TM_END();
printf("Sequence matches gene: %s\n", (result ? "no" : "yes"));
if (result) {
printf("gene = %s\n", gene);
printf("sequence = %s\n", sequence);
}
fflush(stdout);
assert(strlen(sequence) >= strlen(gene));
}
/* Clean up */
printf("Deallocating memory... ");
fflush(stdout);
sequencer_free(sequencerPtr);
segments_free(segmentsPtr);
gene_free(genePtr);
random_free(randomPtr);
puts("done.");
fflush(stdout);
TM_SHUTDOWN();
P_MEMORY_SHUTDOWN();
thread_shutdown();
MAIN_RETURN(0);
}
/* =============================================================================
* 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();
}
/* =============================================================================
* work
* =============================================================================
*/
static void
work (void* argPtr)
{
TM_THREAD_ENTER();
args_t* args = (args_t*)argPtr;
double** feature = args->feature;
int nfeatures = args->nfeatures;
int npoints = args->npoints;
int nclusters = args->nclusters;
int* membership = args->membership;
double** clusters = args->clusters;
int** new_centers_len = args->new_centers_len;
double** new_centers = args->new_centers;
double delta = 0.0;
int index;
int i;
int j;
int start;
int stop;
int myId;
myId = thread_getId();
start = myId * CHUNK;
while (start < npoints) {
stop = (((start + CHUNK) < npoints) ? (start + CHUNK) : npoints);
for (i = start; i < stop; i++) {
index = common_findNearestPoint(feature[i],
nfeatures,
clusters,
nclusters);
/*
* If membership changes, increase delta by 1.
* membership[i] cannot be changed by other threads
*/
if (membership[i] != index) {
delta += 1.0;
}
/* Assign the membership to object i */
/* membership[i] can't be changed by other thread */
membership[i] = index;
/* Update new cluster centers : sum of objects located within */
int mode = 0;
TM_BEGIN(0,mode);
if (mode == 0) {
FAST_PATH_SHARED_WRITE(*new_centers_len[index],
FAST_PATH_SHARED_READ(*new_centers_len[index]) + 1);
for (j = 0; j < nfeatures; j++) {
FAST_PATH_SHARED_WRITE_D(
new_centers[index][j],
(FAST_PATH_SHARED_READ_D(new_centers[index][j]) + feature[i][j])
);
}
} else {
SLOW_PATH_SHARED_WRITE(*new_centers_len[index],
SLOW_PATH_SHARED_READ(*new_centers_len[index]) + 1);
for (j = 0; j < nfeatures; j++) {
SLOW_PATH_SHARED_WRITE_D(
new_centers[index][j],
(SLOW_PATH_SHARED_READ_D(new_centers[index][j]) + feature[i][j])
);
}
}
TM_END();
}
/* Update task queue */
if (start + CHUNK < npoints) {
int mode = 0;
TM_BEGIN(1, mode);
if (mode == 0) {
start = (int)FAST_PATH_SHARED_READ(global_i);
FAST_PATH_SHARED_WRITE(global_i, (start + CHUNK));
} else {
start = (int)SLOW_PATH_SHARED_READ(global_i);
SLOW_PATH_SHARED_WRITE(global_i, (start + CHUNK));
}
TM_END();
} else {
break;
}
}
int mode = 0;
TM_BEGIN(2,mode);
if (mode == 0) {
FAST_PATH_SHARED_WRITE_D(global_delta, FAST_PATH_SHARED_READ_D(global_delta) + delta);
} else {
SLOW_PATH_SHARED_WRITE_D(global_delta, SLOW_PATH_SHARED_READ_D(global_delta) + delta);
}
TM_END();
TM_THREAD_EXIT();
}
/* =============================================================================
* work
* =============================================================================
*/
static void
work (void* argPtr)
{
TM_THREAD_ENTER();
args_t* args = (args_t*)argPtr;
float** feature = args->feature;
int nfeatures = args->nfeatures;
int npoints = args->npoints;
int nclusters = args->nclusters;
int* membership = args->membership;
float** clusters = args->clusters;
int** new_centers_len = args->new_centers_len;
float** new_centers = args->new_centers;
float delta = 0.0;
int index;
long i;
int j;
int start;
int stop;
int myId;
myId = thread_getId();
start = myId * CHUNK;
while (start < npoints) {
stop = (((start + CHUNK) < npoints) ? (start + CHUNK) : npoints);
for (i = start; i < stop; TMHT_LOCAL_WRITE(i, i+1)) {
index = common_findNearestPoint(feature[i],
nfeatures,
clusters,
nclusters);
/*
* If membership changes, increase delta by 1.
* membership[i] cannot be changed by other threads
*/
if (membership[i] != index) {
delta += 1.0;
}
/* Assign the membership to object i */
/* membership[i] can't be changed by other thread */
membership[i] = index;
/* Update new cluster centers : sum of objects located within */
TM_BEGIN();
TM_SHARED_WRITE_I(*new_centers_len[index],
TM_SHARED_READ_I(*new_centers_len[index]) + 1);
for (j = 0; j < nfeatures; j++) {
TM_SHARED_WRITE_F(
new_centers[index][j],
(TM_SHARED_READ_F(new_centers[index][j]) + feature[i][j])
);
}
TM_END();
}
/* Update task queue */
if (start + CHUNK < npoints) {
TM_BEGIN();
start = (int)TM_SHARED_READ_L(global_i);
TM_SHARED_WRITE_L(global_i, (long)(start + CHUNK));
TM_END();
} else {
break;
}
}
TM_BEGIN();
TM_SHARED_WRITE_F(global_delta, TM_SHARED_READ_F(global_delta) + delta);
TM_END();
TM_THREAD_EXIT();
}
/* =============================================================================
* work
* =============================================================================
*/
static void
work (void* argPtr)
{
TM_THREAD_ENTER();
args_t* args = (args_t*)argPtr;
float** feature = args->feature;
int nfeatures = args->nfeatures;
int npoints = args->npoints;
int nclusters = args->nclusters;
int* membership = args->membership;
float** clusters = args->clusters;
long long int** new_centers_len = args->new_centers_len;
float** new_centers = args->new_centers;
float delta = 0.0;
int index;
int i;
int j;
int start;
int stop;
int myId;
bool indexx[1000];
myId = thread_getId();
start = myId * CHUNK;
int cnt=0;
while (start < npoints) {
stop = (((start + CHUNK) < npoints) ? (start + CHUNK) : npoints);
for (i = start; i < stop; i++) {
index = common_findNearestPoint(feature[i],
nfeatures,
clusters,
nclusters);
/*
* If membership changes, increase delta by 1.
* membership[i] cannot be changed by other threads
*/
if (membership[i] != index) {
delta += 1.0;
}
/* Assign the membership to object i */
/* membership[i] can't be changed by other thread */
membership[i] = index;
/* Update new cluster centers : sum of objects located within */
TM_BEGIN();
//printf("shared write to begin: \n");
// int write = *new_centers_len[index];
//int* pt = new_centers_len[i];
//int dat = TM_SHARED_READ_I(*new_centers_len[i]);
//printf("in loop write centers lendata: %i %i\n", dat, *new_centers_len[i]);
TM_SHARED_WRITE_I(*new_centers_len[index],
TM_SHARED_READ_I(*new_centers_len[index]) + 1);
//printf("befor loop len P: %p data: %i\n", new_centers_len[index], *new_centers_len[index]);
//new *new_centers_len[index] = *new_centers_len[index] + 1;
//printf("INDEX %i \n" , index);
indexx[index] = true;
/*if(*new_centers_len[index]==0)*/
//printf("in lloop len P: %p data: %i\n", new_centers_len[index], *new_centers_len[index]);
//*new_centers_len[i] = *new_centers_len[i]+1;
// pt = new_centers_len[i];
//dat = TM_SHARED_READ_I(*new_centers_len[i]);
//printf("in loop write centers len data: %i\n", dat);
for (j = 0; j < nfeatures; j++) {
//printf("featurs\n");
//int feat = feature[i][j];
//printf("write\n");
//float read = TM_SHARED_READ_F(new_centers[index][j]);
//printf("write %p " ,write);
//printf("shared write to:\n");
//printf("feature %f", feature[i][j]);
//float feat = feature[i][j];
//float fl = (TM_SHARED_READ_F(new_centers[index][j])+ feat);//feature[i][j]);
//int len = *new_centers_len[index];
TM_SHARED_WRITE_F(
//write,
new_centers[index][j],
//(read + feat)
//fl
(TM_SHARED_READ_F(new_centers[index][j])+ feature[i][j])
//printf("index %p %p\n", (void*)*new_centers[index][j], (void*)(*new_centers[index][j] +1));
//printf("indexnon p %p %p\n", (void*)new_centers[index][j], (void*)(new_centers[index][j] +1));
);
//new new_centers[index][j] = new_centers[index][j] + feature[i][j];
//if(0==*new_centers_len[index]) printf("ISNAN %i\n", len);
//if(isnanf(new_centers[index][j])) printf("ISNAN2\n\n");
// if(isinf(*new_centers_len[index])) printf("ISINF\n\n");
//if(isinf(*new_centers_len[index])) printf("ISINF2\n\n");
}
TM_END();
}
//printf("update \n");
/* Update task queue */
if (start + CHUNK < npoints) {
TM_BEGIN();
start = (int)TM_SHARED_READ_L(*global_i);
TM_SHARED_WRITE_L(*global_i, (long)(start + CHUNK));
TM_END();
} else {
break;
}
}
TM_BEGIN();
//printf("shared write to: %p", *global_delta);
TM_SHARED_WRITE_F(*global_delta, TM_SHARED_READ_F(*global_delta) + delta);
//new *global_delta = *global_delta + delta;
TM_END();
int u1 =0;
/* for(int i1=0; i1<1000; i1++){
if(indexx[i1]) printf("INDEX %i %i\n", i1, u1++);
}*/
TM_THREAD_EXIT();
}
void client_run (void* argPtr) {
TM_THREAD_ENTER();
/*long id = thread_getId();
volatile long* ptr1 = &(global_array[0].value);
volatile long* ptr2 = &(global_array[100].value);
long tt = 0;
if (id == 0) {
while (1) {
long v1 = 0;
long v2 = 0;
acquire_write(&(local_th_data[phys_id]), &the_lock);
*ptr1 = (*ptr1) + 1;
int f = 1;
int ii;
for(ii = 1; ii <= 100000000; ii++)
{
f *= ii;
}
tt += f;
*ptr2 = (*ptr2) + 1;
v1 = global_array[0].value;
v2 = global_array[100].value;
release_write(cluster_id, &(local_th_data[phys_id]), &the_lock); \
if (v1 != v2) {
printf("different2! %ld %ld\n", v1, v2);
exit(1);
}
}
} else {
while (1) {
int i = 0;
long sum = 0;
for (; i < 100000; i++) {
int status = _xbegin();
if (status == _XBEGIN_STARTED) {
sum += *ptr1;
sum += *ptr2;
_xend();
}
}
while(1) {
long v1 = 0;
long v2 = 0;
int status = _xbegin();
if (status == _XBEGIN_STARTED) {
v1 = *ptr1;
v2 = *ptr2;
_xend();
if (v1 != v2) {
printf("different! %ld %ld\n", v1, v2);
exit(1);
}
}
}
}
}
printf("%ld", tt);*/
random_t* randomPtr = random_alloc();
random_seed(randomPtr, time(0));
// unsigned long myId = thread_getId();
// long numThread = *((long*)argPtr);
long operations = (long)global_params[PARAM_OPERATIONS] / (long)global_params[PARAM_THREADS];
long interval = (long)global_params[PARAM_INTERVAL];
printf("operations: %ld \tinterval: %ld\n", operations, interval);
long total = 0;
long total2 = 0;
long i = 0;
for (; i < operations; i++) {
long random_number = ((long) random_generate(randomPtr)) % ((long)global_params[PARAM_SIZE]);
long random_number2 = ((long) random_generate(randomPtr)) % ((long)global_params[PARAM_SIZE]);
if (random_number == random_number2) {
random_number2 = (random_number2 + 1) % ((long)global_params[PARAM_SIZE]);
}
TM_BEGIN();
long r1 = (long)TM_SHARED_READ_L(global_array[random_number].value);
long r2 = (long)TM_SHARED_READ_L(global_array[random_number2].value);
int repeat = 0;
for (; repeat < (long) global_params[PARAM_CONTENTION]; repeat++) {
total2 += (long) TM_SHARED_READ_L(global_array[((long) random_generate(randomPtr)) % ((long)global_params[PARAM_SIZE])].value);
}
r1 = r1 + 1;
r2 = r2 - 1;
int f = 1;
int ii;
for(ii = 1; ii <= ((unsigned int) global_params[PARAM_WORK]); ii++)
{
f *= ii;
}
total += f / 1000000;
TM_SHARED_WRITE_L(global_array[random_number].value, r1);
TM_SHARED_WRITE_L(global_array[random_number2].value, r2);
TM_END();
long k = 0;
for (;k < (long)global_params[PARAM_INTERVAL]; k++) {
long ru = ((long) random_generate(randomPtr)) % 2;
total += ru;
}
}
TM_THREAD_EXIT();
printf("ru ignore %ld - %ld\n", total, total2);
}
/* =============================================================================
* 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();
}
/* =============================================================================
* client_run
* -- Execute list operations on the database
* =============================================================================
*/
void
client_run (void* argPtr)
{
TM_THREAD_ENTER();
long myId = thread_getId();
client_t* clientPtr = ((client_t**)argPtr)[myId];
manager_t* managerPtr = clientPtr->managerPtr;
random_t* randomPtr = clientPtr->randomPtr;
long numOperation = clientPtr->numOperation;
long numQueryPerTransaction = clientPtr->numQueryPerTransaction;
long queryRange = clientPtr->queryRange;
long percentUser = clientPtr->percentUser;
long* types = (long*)P_MALLOC(numQueryPerTransaction * sizeof(long));
long* ids = (long*)P_MALLOC(numQueryPerTransaction * sizeof(long));
long* ops = (long*)P_MALLOC(numQueryPerTransaction * sizeof(long));
long* prices = (long*)P_MALLOC(numQueryPerTransaction * sizeof(long));
long i;
for (i = 0; i < numOperation; i++) {
long r = random_generate(randomPtr) % 100;
action_t action = selectAction(r, percentUser);
switch (action) {
case ACTION_MAKE_RESERVATION: {
long maxPrices[NUM_RESERVATION_TYPE] = { -1, -1, -1 };
long maxIds[NUM_RESERVATION_TYPE] = { -1, -1, -1 };
long n;
long numQuery = random_generate(randomPtr) % numQueryPerTransaction + 1;
long customerId = random_generate(randomPtr) % queryRange + 1;
for (n = 0; n < numQuery; n++) {
types[n] = random_generate(randomPtr) % NUM_RESERVATION_TYPE;
ids[n] = (random_generate(randomPtr) % queryRange) + 1;
}
bool_t isFound = FALSE;
TM_BEGIN();
for (n = 0; n < numQuery; n++) {
long t = types[n];
long id = ids[n];
long price = -1;
switch (t) {
case RESERVATION_CAR:
if (MANAGER_QUERY_CAR(managerPtr, id) >= 0) {
price = MANAGER_QUERY_CAR_PRICE(managerPtr, id);
}
break;
case RESERVATION_FLIGHT:
if (MANAGER_QUERY_FLIGHT(managerPtr, id) >= 0) {
price = MANAGER_QUERY_FLIGHT_PRICE(managerPtr, id);
}
break;
case RESERVATION_ROOM:
if (MANAGER_QUERY_ROOM(managerPtr, id) >= 0) {
price = MANAGER_QUERY_ROOM_PRICE(managerPtr, id);
}
break;
default:
assert(0);
}
if (price > maxPrices[t]) {
maxPrices[t] = price;
maxIds[t] = id;
isFound = TRUE;
}
} /* for n */
if (isFound) {
MANAGER_ADD_CUSTOMER(managerPtr, customerId);
}
if (maxIds[RESERVATION_CAR] > 0) {
MANAGER_RESERVE_CAR(managerPtr,
customerId, maxIds[RESERVATION_CAR]);
}
if (maxIds[RESERVATION_FLIGHT] > 0) {
MANAGER_RESERVE_FLIGHT(managerPtr,
customerId, maxIds[RESERVATION_FLIGHT]);
}
if (maxIds[RESERVATION_ROOM] > 0) {
MANAGER_RESERVE_ROOM(managerPtr,
customerId, maxIds[RESERVATION_ROOM]);
}
TM_END();
break;
}
case ACTION_DELETE_CUSTOMER: {
long customerId = random_generate(randomPtr) % queryRange + 1;
TM_BEGIN();
long bill = MANAGER_QUERY_CUSTOMER_BILL(managerPtr, customerId);
if (bill >= 0) {
MANAGER_DELETE_CUSTOMER(managerPtr, customerId);
}
TM_END();
break;
}
case ACTION_UPDATE_TABLES: {
long numUpdate = random_generate(randomPtr) % numQueryPerTransaction + 1;
long n;
for (n = 0; n < numUpdate; n++) {
types[n] = random_generate(randomPtr) % NUM_RESERVATION_TYPE;
ids[n] = (random_generate(randomPtr) % queryRange) + 1;
ops[n] = random_generate(randomPtr) % 2;
if (ops[n]) {
prices[n] = ((random_generate(randomPtr) % 5) * 10) + 50;
}
}
TM_BEGIN();
for (n = 0; n < numUpdate; n++) {
long t = types[n];
long id = ids[n];
long doAdd = ops[n];
if (doAdd) {
long newPrice = prices[n];
switch (t) {
case RESERVATION_CAR:
MANAGER_ADD_CAR(managerPtr, id, 100, newPrice);
break;
case RESERVATION_FLIGHT:
MANAGER_ADD_FLIGHT(managerPtr, id, 100, newPrice);
break;
case RESERVATION_ROOM:
MANAGER_ADD_ROOM(managerPtr, id, 100, newPrice);
break;
default:
assert(0);
}
} else { /* do delete */
switch (t) {
case RESERVATION_CAR:
MANAGER_DELETE_CAR(managerPtr, id, 100);
break;
case RESERVATION_FLIGHT:
MANAGER_DELETE_FLIGHT(managerPtr, id);
break;
case RESERVATION_ROOM:
MANAGER_DELETE_ROOM(managerPtr, id, 100);
break;
default:
assert(0);
}
}
}
TM_END();
break;
}
default:
assert(0);
} /* switch (action) */
} /* for i */
TM_THREAD_EXIT();
}
/* =============================================================================
* router_solve
* =============================================================================
*/
void
router_solve (void* argPtr)
{
TM_THREAD_ENTER();
long threadId = thread_getId();
router_solve_arg_t* routerArgPtr = (router_solve_arg_t*)argPtr;
router_t* routerPtr = routerArgPtr->routerPtr;
maze_t* mazePtr = routerArgPtr->mazePtr;
long* numPathArray = routerArgPtr->numPathArray;
vector_t* myPathVectorPtr = PVECTOR_ALLOC(1);
assert(myPathVectorPtr);
queue_t* workQueuePtr = mazePtr->workQueuePtr;
grid_t* gridPtr = mazePtr->gridPtr;
grid_t* myGridPtr =
PGRID_ALLOC(gridPtr->width, gridPtr->height, gridPtr->depth);
assert(myGridPtr);
long bendCost = routerPtr->bendCost;
queue_t* myExpansionQueuePtr = PQUEUE_ALLOC(-1);
long numPath = 0;
/*
* Iterate over work list to route each path. This involves an
* 'expansion' and 'traceback' phase for each source/destination pair.
*/
while ((global_timedExecution && !global_isTerminated) || (!global_timedExecution)) {
//while (1) {
wait_for_turn(threadId);
if (global_timedExecution && global_isTerminated)
break;
ulong_t beginTime;
pair_t* coordinatePairPtr;
TM_BEGIN();
beginTime = get_thread_time();
if (TMQUEUE_ISEMPTY(workQueuePtr)) {
if (TMQUEUE_ISEMPTY(workQueuePtr))
coordinatePairPtr = NULL;
} else {
coordinatePairPtr = (pair_t*)TMQUEUE_POP(workQueuePtr);
}
TM_END();
//add_throughput(threadId , get_thread_time() - beginTime);
if (coordinatePairPtr == NULL) {
break;
}
coordinate_t* srcPtr = (coordinate_t*)coordinatePairPtr->firstPtr;
coordinate_t* dstPtr = (coordinate_t*)coordinatePairPtr->secondPtr;
bool_t success = FALSE;
vector_t* pointVectorPtr = NULL;
TM_BEGIN();
beginTime = get_thread_time();
grid_copy(myGridPtr, gridPtr); /* ok if not most up-to-date */
if (PdoExpansion(routerPtr, myGridPtr, myExpansionQueuePtr,
srcPtr, dstPtr)) {
pointVectorPtr = PdoTraceback(gridPtr, myGridPtr, dstPtr, bendCost);
/*
* TODO: fix memory leak
*
* pointVectorPtr will be a memory leak if we abort this transaction
*/
if (pointVectorPtr) {
TMGRID_ADDPATH(gridPtr, pointVectorPtr);
TM_LOCAL_WRITE_L(success, TRUE);
}
}
TM_END();
add_throughput(threadId , get_thread_time() - beginTime);
numPath++;
if (success) {
bool_t status = PVECTOR_PUSHBACK(myPathVectorPtr,
(void*)pointVectorPtr);
assert(status);
}
}
numPathArray[threadId] = numPath;
/*
* Add my paths to global list
*/
list_t* pathVectorListPtr = routerArgPtr->pathVectorListPtr;
TM_BEGIN();
TMLIST_INSERT(pathVectorListPtr, (void*)myPathVectorPtr);
TM_END();
PGRID_FREE(myGridPtr);
PQUEUE_FREE(myExpansionQueuePtr);
#if DEBUG
puts("\nFinal Grid:");
grid_print(gridPtr);
#endif /* DEBUG */
TM_THREAD_EXIT();
}