int test_kmp_set_defaults_lock_bug()
{
/* checks that omp_get_num_threads is equal to the number of
threads */
int nthreads_lib;
int nthreads = 0;
nthreads_lib = -1;
#pragma omp parallel
{
omp_set_lock(&lock);
nthreads++;
omp_unset_lock(&lock);
#pragma omp single
{
nthreads_lib = omp_get_num_threads ();
} /* end of single */
} /* end of parallel */
kmp_set_defaults("OMP_NUM_THREADS");
#pragma omp parallel
{
omp_set_lock(&lock);
nthreads++;
omp_unset_lock(&lock);
} /* end of parallel */
return (nthreads == 2*nthreads_lib);
}
/* Assumes node is already locked */
static inline void lockTwoNodes(Node *node, Node *node2)
{
IDnum nodeID = getNodeID(node);
IDnum node2ID = getNodeID(node2);
if (nodeID < 0)
nodeID = -nodeID;
if (node2ID < 0)
node2ID = -node2ID;
if (nodeID == node2ID)
return;
/* Lock lowest ID first to avoid deadlocks */
if (nodeID < node2ID)
{
omp_set_lock (nodeLocks + node2ID);
}
else if (!omp_test_lock (nodeLocks + node2ID))
{
omp_unset_lock (nodeLocks + nodeID);
omp_set_lock (nodeLocks + node2ID);
omp_set_lock (nodeLocks + nodeID);
}
}
void multWith_LockedArrays(){
#pragma omp for schedule(static, N / NUM_THREADS)
for(int i = 0; i < N; i++){
forces[i][X_Pos] = 0;
forces[i][Y_Pos] = 0;
}
#pragma omp for schedule(static, N / NUM_THREADS)
for(int i = 0; i < N; i++){
double x_dif, y_dif, dist, dist_cubed;
double tempForcX, tempForcY;
for(int j = i + 1; j < N; j++){
x_dif = pos[i][X_Pos] - pos[j][X_Pos];
y_dif = pos[i][Y_Pos] - pos[j][Y_Pos];
dist = sqrt(x_dif * x_dif + y_dif * y_dif);
dist_cubed = dist * dist * dist;
tempForcX = constG * masses[i] * masses[j] / dist_cubed * x_dif;
tempForcY = constG * masses[i] * masses[j] / dist_cubed * y_dif;
omp_set_lock(&locks[i]);
forces[i][X_Pos] += tempForcX;
forces[i][Y_Pos] += tempForcY;
omp_unset_lock(&locks[i]);
omp_set_lock(&locks[j]);
forces[j][X_Pos] -= tempForcX;
forces[j][Y_Pos] -= tempForcY;
omp_unset_lock(&locks[j]);
}
}
}
int kernel(int npoints, int nclusters) {
int i, j, delta = 0, index;
#pragma omp parallel for default(shared) \
private(i,j,index) \
schedule(static)
for(i = 0; i < npoints; i++) {
index = find_nearest_point(feature[i], 2, clusters, nclusters);
if (membership[i] != index) {
omp_set_lock(&delta_lock);
delta += 1;
omp_unset_lock(&delta_lock);
}
membership[i] = index;
omp_set_lock(&new_centers_len_locks[index]);
new_centers_len[index]++;
omp_unset_lock(&new_centers_len_locks[index]);
for(j = 0; j < 2; j++) {
omp_set_lock(&new_centers_locks[index][j]);
new_centers[index][j] += feature[i][j];
omp_unset_lock(&new_centers_locks[index][j]);
}
}
return delta;
}
int main(){
/*объявляем mutex*/
omp_lock_t lock_1;
/*!!!обязательно инициализируем его!!!*/
omp_init_lock(&lock_1);
#pragma omp parallel num_threads(3)
{
#pragma omp sections
{
#pragma omp section
{
omp_set_lock(&lock_1); //устанавливать и снимать надо в одном и том же потоке
std::cout<<"После установки блокировки 1\n";
omp_unset_lock(&lock_1);
}
#pragma omp section
{
omp_set_lock(&lock_1);
std::cout<<"После установки блокировки 2\n";
omp_unset_lock(&lock_1);
}
#pragma omp section
{
omp_set_lock(&lock_1);
std::cout<<"После установки блокировки 3\n";
omp_unset_lock(&lock_1);
}
}
}
return 0;
}
int dequeue(){
while(tail == NULL);
if(head == tail){
int data = head->data;
omp_set_lock(&head_lock);
head = NULL;
#pragma omp flush(head)
omp_unset_lock(&head_lock);
omp_set_lock(&tail_lock);
tail = NULL;
#pragma omp flush(tail)
omp_unset_lock(&tail_lock);
return data;
}
else{
int data = tail->data;
NODE *ptr = head;
while(ptr->next != tail)
ptr = ptr->next;
omp_set_lock(&tail_lock);
tail = ptr;
tail->next = NULL;
#pragma omp flush(tail)
omp_unset_lock(&tail_lock);
return data;
}
}
int main (int argc, char *argv[])
{
int nthreads, tid, i;
float a[N], b[N];
omp_lock_t locka, lockb;
/* Initialize the locks */
omp_init_lock(&locka);
omp_init_lock(&lockb);
/* Fork a team of threads giving them their own copies of variables */
#pragma omp parallel shared(a, b, nthreads, locka, lockb) private(tid)
{
/* Obtain thread number and number of threads */
tid = omp_get_thread_num();
#pragma omp master
{
nthreads = omp_get_num_threads();
printf("Number of threads = %d\n", nthreads);
}
printf("Thread %d starting...\n", tid);
#pragma omp barrier
#pragma omp sections nowait
{
#pragma omp section
{
printf("Thread %d initializing a[]\n",tid);
omp_set_lock(&locka);
for (i=0; i<N; i++)
a[i] = i * DELTA;
omp_unset_lock(&locka);// if unset, the other thread will wait
omp_set_lock(&lockb);
printf("Thread %d adding a[] to b[]\n",tid);
for (i=0; i<N; i++)
b[i] += a[i];
omp_unset_lock(&lockb);
//omp_unset_lock(&locka);
}
#pragma omp section
{
printf("Thread %d initializing b[]\n",tid);
omp_set_lock(&lockb);
for (i=0; i<N; i++)
b[i] = i * PI;
omp_unset_lock(&lockb);// if unset, the other thread will wait
omp_set_lock(&locka);
printf("Thread %d adding b[] to a[]\n",tid);
for (i=0; i<N; i++)
a[i] += b[i];
omp_unset_lock(&locka);
//omp_unset_lock(&lockb);
}
} /* end of sections */
} /* end of parallel region */
}
void POMP2_Set_lock(omp_lock_t *s) {
if ( pomp2_tracing ) {
esd_enter(epk_omp_regid[EPK__OMP_SET_LOCK]);
omp_set_lock(s);
if ( pomp2_tracing > 1 ) esd_omp_alock(epk_lock_id(s));
esd_exit(epk_omp_regid[EPK__OMP_SET_LOCK]);
} else {
omp_set_lock(s);
}
}
void incrementer(omp_lock_t* lock_b) {
for(int i=0; i< LOOP_ITS; ++i) {
omp_set_lock(&g_lock_a);
g_a++;
omp_unset_lock(&g_lock_a);
omp_set_lock(lock_b);
g_b++;
omp_unset_lock(lock_b);
}
}
void POMP_Set_lock(omp_lock_t *s) {
if ( IS_POMP_TRACE_ON ) {
uint64_t time = vt_pform_wtime();
vt_enter(&time, vt_omp_regid[VT__OMP_SET_LOCK]);
omp_set_lock(s);
time = vt_pform_wtime();
vt_omp_alock(&time, vt_lock_id(s));
vt_exit(&time);
} else {
omp_set_lock(s);
}
}
DEF_FPOMP_FUNC(void POMP_Set_lock_f(omp_lock_t *s)) {
if ( IS_POMP_TRACE_ON ) {
uint64_t time = vt_pform_wtime();
vt_enter(&time, vt_omp_regid[VT__OMP_SET_LOCK]);
omp_set_lock(s);
time = vt_pform_wtime();
vt_omp_alock(&time, vt_lock_id(s));
vt_exit(&time);
} else {
omp_set_lock(s);
}
} VT_GENERATE_F77_BINDINGS(pomp_set_lock, POMP_SET_LOCK,
VT_DECLDEF(void POMP_Set_lock_f(omp_lock_t *s)) {
if ( IS_POMP_TRACE_ON ) {
uint64_t time;
time = vt_pform_wtime();
vt_enter(VT_CURRENT_THREAD, &time, vt_omp_regid[VT__OMP_SET_LOCK]);
omp_set_lock(s);
time = vt_pform_wtime();
vt_exit(VT_CURRENT_THREAD, &time);
} else {
omp_set_lock(s);
}
} VT_GENERATE_F77_BINDINGS(pomp_set_lock, POMP_SET_LOCK,
void POMP_Set_lock(omp_lock_t *s) {
if ( IS_POMP_TRACE_ON ) {
uint64_t time;
time = vt_pform_wtime();
vt_enter(VT_CURRENT_THREAD, &time, vt_omp_regid[VT__OMP_SET_LOCK]);
omp_set_lock(s);
time = vt_pform_wtime();
vt_exit(VT_CURRENT_THREAD, &time);
} else {
omp_set_lock(s);
}
}
/*
Looks up a previously stored state, or generates a new one. No assumptions are
made about whether the board state is in reduced form already. Never fails. If
the memory is full it allocates a new state regardless. If the state is found
and returned it's OpenMP lock is first set. Thread-safe.
RETURNS the state information
*/
tt_stats * tt_lookup_create(
const board * b,
bool is_black,
u64 hash
){
u32 key = (u32)(hash % ((u64)number_of_buckets));
omp_lock_t * bucket_lock = is_black ? &b_table_lock : &w_table_lock;
omp_set_lock(bucket_lock);
tt_stats * ret = find_state(hash, b, is_black);
if(ret == NULL) /* doesnt exist */
{
if(states_in_use >= max_allocated_states)
{
/*
It is possible in theory for a complex ko to produce a situation
where freeing the game tree that is not reachable doesn't free any
states.
*/
tt_log_status();
char * s = alloc();
board_to_string(s, b->p, b->last_played, b->last_eaten);
flog_warn("tt", s);
release(s);
flog_warn("tt", "memory exceeded on root lookup");
}
ret = create_state(hash);
memcpy(ret->p, b->p, TOTAL_BOARD_SIZ);
ret->last_eaten_passed =
(b->last_played == PASS) ? PASS : b->last_eaten;
omp_set_lock(&ret->lock);
if(is_black)
{
ret->next = b_stats_table[key];
b_stats_table[key] = ret;
}
else
{
ret->next = w_stats_table[key];
w_stats_table[key] = ret;
}
omp_unset_lock(bucket_lock);
}
else /* update */
{
omp_set_lock(&ret->lock);
omp_unset_lock(bucket_lock);
}
return ret;
}
/**
* PRIVATE. Insert a region in a context.
*/
static inline int
bfwork_region_insert(bam_fwork_t *fwork, bam_region_t *region)
{
linked_list_t *list;
size_t list_l;
assert(fwork);
assert(region);
omp_set_lock(&fwork->regions_lock);
//List handle
list = fwork->regions_list;
list_l = linked_list_size(list);
if(list_l >= FWORK_REGIONS_MAX)
{
omp_unset_lock(&fwork->regions_lock);
//Wait for free slots
if(!omp_test_lock(&fwork->free_slots))
{
if(omp_get_num_threads() == 2)
{
#pragma omp taskwait //Force processing
}
omp_set_lock(&fwork->free_slots);
}
//LOG_FATAL_F("Not enough region slots, current: %d\n", list_l);
}
//This lock must be always locked until regions buffer have regions free
omp_test_lock(&fwork->free_slots);
//Add region to list
linked_list_insert_last(region, list);
omp_set_lock(®ion->lock);
LOG_INFO_F("Inserting region %d:%lu-%lu with %d reads\n",
region->chrom + 1, region->init_pos + 1,
region->end_pos + 1, region->size);
LOG_INFO_F("Regions to process %lu\n", linked_list_size(list));
omp_unset_lock(®ion->lock);
omp_unset_lock(&fwork->regions_lock);
return NO_ERROR;
}
void *popBusyWait(struct Stack *s) {
struct Node *temp;
void *d;
omp_set_lock(&s->lock);
while (s->top == NULL) {
omp_unset_lock(&s->lock);
omp_set_lock(&s->lock);
}
d = s->top->data;
temp = s->top;
s->top = s->top->next;
omp_unset_lock(&s->lock);
free(temp);
return d;
}
void
openmp_print_tid (void)
{
omp_set_lock (&writelock);
SC_PRODUCTIONF ("Hello from thread %i.\n", omp_get_thread_num ());
omp_unset_lock (&writelock);
}
OpenMPCounter& OpenMPCounter::operator ++()
{
omp_set_lock(&_lock);
++_counter;
omp_unset_lock(&_lock);
return *this;
}
int ATL_DecAtomicCount(void *vp)
{
char *cp=vp;
#ifdef ATL_OMP_THREADS
omp_lock_t *mp;
#else
pthread_mutex_t *mp;
#endif
int *cntp;
int iret;
cntp = (int*)(cp+128);
#ifdef ATL_OMP_THREADS
mp = (omp_lock_t*)(cntp+2);
omp_set_lock(mp);
#else
mp = (pthread_mutex_t*)(cntp+2);
pthread_mutex_lock(mp);
#endif
iret = *cntp;
if (iret) (*cntp)--;
#ifdef ATL_OMP_THREADS
omp_unset_lock(mp);
#else
pthread_mutex_unlock(mp);
#endif
return(iret);
}
int main()
{
omp_lock_t lck;
int id;
omp_init_lock(&lck);
#pragma omp parallel shared(lck) private(id)
{
id = omp_get_thread_num();
omp_set_lock(&lck);
/* only one thread at a time can execute this printf */
printf("My thread id is %d.\n", id);
omp_unset_lock(&lck);
while (! omp_test_lock(&lck)) {
skip(id); /* we do not yet have the lock,
so we must do something else */
}
work(id); /* we now have the lock
and can do the work */
omp_unset_lock(&lck);
}
omp_destroy_lock(&lck);
return 0;
}
FLA_Error FLA_Axpy_sync_circular( FLA_Obj alpha, FLA_Obj X, FLA_Obj B )
{
FLA_Obj XL, XR, X0, X1, X2;
FLA_Obj BL, BR, B0, B1, B2;
int n_stages = FLA_omp_get_num_stages();
int stage_width = FLA_omp_compute_stage_width( X );
int thread_num = omp_get_thread_num();
int n_done = 0;
int b, i;
// Start thread i on the ith panel partition of B.
FLA_Part_1x2( X, &XL, &XR, stage_width*thread_num, FLA_LEFT );
FLA_Part_1x2( B, &BL, &BR, stage_width*thread_num, FLA_LEFT );
while ( n_done++ < n_stages ){
// The last lockable partition may be smaller than the others.
b = min( FLA_Obj_width( XR ), stage_width );
FLA_Repart_1x2_to_1x3( XL, /**/ XR, &X0, /**/ &X1, &X2,
b, FLA_RIGHT );
FLA_Repart_1x2_to_1x3( BL, /**/ BR, &B0, /**/ &B1, &B2,
b, FLA_RIGHT );
/*------------------------------------------------------------*/
// Get the index of the current partition.
i = FLA_Obj_width(XL)/stage_width;
// Acquire lock[i] (the lock for X1 and B1).
omp_set_lock( &fla_omp_lock[i] );
// B1 := alpha * X1 + B1
FLA_Axpy_external( alpha, X1, B1 );
// Release lock[i] (the lock for X1 and B1).
omp_unset_lock( &fla_omp_lock[i] );
/*------------------------------------------------------------*/
FLA_Cont_with_1x3_to_1x2( &XL, /**/ &XR, X0, X1, /**/ X2,
FLA_LEFT );
FLA_Cont_with_1x3_to_1x2( &BL, /**/ &BR, B0, B1, /**/ B2,
FLA_LEFT );
// If this thread reaches the last partition, wrap back around to
// the first partition for the next iteration.
if( FLA_Obj_width( XL ) == FLA_Obj_width( X ) )
{
FLA_Part_1x2( X, &XL, &XR, 0, FLA_LEFT );
FLA_Part_1x2( B, &BL, &BR, 0, FLA_LEFT );
}
}
return FLA_SUCCESS;
}
/* Locks a single cell
* i, j the coordinates of the cell; board, the current board
*/
void lockCell(int i, int j, Board board){
int width = board->width + 1;
int offset = i * width + j;
omp_set_lock(&(board->locks[offset]));
}
void
computeHistogram (const Eigen::MatrixXf &data, Eigen::MatrixXf &histogram, size_t bins, float min, float max)
{
float bin_size = (max-min) / bins;
int num_dim = data.cols();
histogram = Eigen::MatrixXf::Zero (bins, num_dim);
for (int dim = 0; dim < num_dim; dim++)
{
omp_lock_t bin_lock[bins];
for(size_t pos=0; pos<bins; pos++)
omp_init_lock(&bin_lock[pos]);
#pragma omp parallel for firstprivate(min, max, bins) schedule(dynamic)
for (int j = 0; j<data.rows(); j++)
{
int pos = std::floor( (data(j,dim) - min) / bin_size);
if(pos < 0)
pos = 0;
if(pos > (int)bins)
pos = bins - 1;
omp_set_lock(&bin_lock[pos]);
histogram(pos,dim)++;
omp_unset_lock(&bin_lock[pos]);
}
for(size_t pos=0; pos<bins; pos++)
omp_destroy_lock(&bin_lock[pos]);
}
}
inline void CReferenceCounter::Release(void* pObject)
{
PNL_CHECK_IS_NULL_POINTER(pObject);
omp_set_lock(&m_release_lock);
std::list<void*>::iterator location = std::find( m_refList.begin(),
m_refList.end(), pObject );
assert( location != m_refList.end() );
while( location != m_refList.end() )
{
location = std::find( m_refList.erase(location), m_refList.end(),
pObject );
}
if( m_refList.empty() )
{
omp_unset_lock(&m_release_lock);
delete this;
}
else
{
omp_unset_lock(&m_release_lock);
};
}
int colvarproxy_smp::smp_lock()
{
#if defined(_OPENMP)
omp_set_lock(reinterpret_cast<omp_lock_t *>(omp_lock_state));
#endif
return COLVARS_OK;
}
static tt_stats * create_state(
u64 hash
){
tt_stats * ret = NULL;
omp_set_lock(&freed_nodes_lock);
if(freed_nodes != NULL)
{
ret = freed_nodes;
freed_nodes = freed_nodes->next;
}
else
++allocated_states;
++states_in_use;
omp_unset_lock(&freed_nodes_lock);
if(ret == NULL)
{
ret = (tt_stats *)malloc(sizeof(tt_stats));
if(ret == NULL)
flog_crit("tt", "create_state: system out of memory");
omp_init_lock(&ret->lock);
}
/* careful that some fields are not initialized here */
ret->zobrist_hash = hash;
ret->maintenance_mark = maintenance_mark;
ret->plays_count = 0;
ret->expansion_delay = expansion_delay;
return ret;
}
/*Each thread will call this function independantly. They calculate dataset
*and then write it out to hdf, for NUM_MDSET times */
void CalcWriteData(hid_t fid, hid_t dataspace, hid_t datatype)
{
double data[NX][NY];
hid_t dataset;
char dname[16];
int tid;
int i, j, k;
tid = omp_get_thread_num();
for(i=0; i<NUM_MDSET; i++) {
/*Weird calculation to extend computing time*/
for(j=0; j<NX; j++) {
for(k=0; k<NY; k++)
data[j][k] = ( pow(sin(tid), 2.0) + pow(cos(tid), 2.0) ) *
( tid + i ) +
( pow(123456789.0, 8.0) - pow(123456789.0, 8.0) );
}
sprintf(dname, "tid%d-dataset%d", tid, i);
/*Serialize HDF dataset writing using OpenMP lock*/
omp_set_lock(&lock);
dataset = H5Dcreate(fid, dname, datatype, dataspace, H5P_DEFAULT);
H5Dwrite(dataset, H5T_NATIVE_DOUBLE, H5S_ALL, H5S_ALL, H5P_DEFAULT,
data);
H5Dclose(dataset);
/*Release lock*/
omp_unset_lock(&lock);
}
}
int test_omp_lock()
{
int nr_threads_in_single = 0;
int result = 0;
int nr_iterations = 0;
int i;
omp_init_lock(&lck);
#pragma omp parallel shared(lck)
{
#pragma omp for
for(i = 0; i < LOOPCOUNT; i++) {
omp_set_lock(&lck);
#pragma omp flush
nr_threads_in_single++;
#pragma omp flush
nr_iterations++;
nr_threads_in_single--;
result = result + nr_threads_in_single;
omp_unset_lock(&lck);
}
}
omp_destroy_lock(&lck);
return ((result == 0) && (nr_iterations == LOOPCOUNT));
}
void
ColorTransformOMP::initializeLUT()
{
omp_set_lock(&initialization_lock_);
sRGB_LUT.resize(256);
sXYZ_LUT.resize(4000);
#pragma omp parallel for schedule (dynamic)
for (int i = 0; i < 256; i++)
{
float f = static_cast<float> (i) / 255.0f;
if (f > 0.04045)
sRGB_LUT[i] = powf ((f + 0.055f) / 1.055f, 2.4f);
else
sRGB_LUT[i] = f / 12.92f;
}
#pragma omp parallel for schedule (dynamic)
for (int i = 0; i < 4000; i++)
{
float f = static_cast<float> (i) / 4000.0f;
if (f > 0.008856)
sXYZ_LUT[i] = static_cast<float> (powf (f, 0.3333f));
else
sXYZ_LUT[i] = static_cast<float>((7.787 * f) + (16.0 / 116.0));
}
omp_unset_lock(&initialization_lock_);
}
int CoutLock::set()
{
#ifdef _OPENMP
omp_set_lock(&CoutLock::coutLock);
#endif
return 0;
}
