void productor(){
while(1){
omp_set_lock(&CS);
if(pet==0){
rsp=rsp%10;
rsp++;
} else if(pet==1){
rsp=(rsp%10)+10;
rsp++;
} else if(pet==2){
rsp=(rsp%10)+20;
rsp++;
}
int id=omp_get_thread_num();
printf("----Productor %d con petición: %d y respuesta: %d\n",id,pet,rsp);
// omp_unset_lock(&CC); // Para el caso sin distribuidor
omp_unset_lock(&CD);
sleep(1);
}
}
/* Implements using omp lock functions
* */
void foo_locks(long long int n) {
long long int a=0;
long long int i;
omp_lock_t my_lock;
// init lock
omp_init_lock(&my_lock);
double time = omp_get_wtime();
#pragma omp parallel for schedule(static) shared(a)
for(i = 0; i < n; i++)
{
omp_set_lock(&my_lock);
a+=1;
omp_unset_lock(&my_lock);
}
omp_destroy_lock(&my_lock);
time = omp_get_wtime() - time;
printf("Final value = %d \n ", a);
printf("Locks: Total time = %f seconds \n ", time);
} // end foo_locks
/* Gets the next chunk of iterations to perform for the given thread_id
* If USE_LOCKS is set to True, function will implement locks
* Else it will initiate a critical region for the shared variable read-write
*/
void get_chunks(int thread_id, double K, int* start_iter, int* chunk)
{
int remaining_iters_num, chunk_size;
if(USE_LOCKS == FALSE) {
#pragma omp critical (chunk)
{
remaining_iters_num = remaining_iters[thread_id];
chunk_size = (int) ceil((double)remaining_iters_num*K);
if (chunk_size > remaining_iters_num) chunk_size = remaining_iters_num;
remaining_iters[thread_id] = (remaining_iters_num - chunk_size);
}
} else {
omp_set_lock(&(remaining_iters_lock[thread_id]));
remaining_iters_num = remaining_iters[thread_id];
chunk_size = (int) ceil((double)remaining_iters_num*K);
if (chunk_size > remaining_iters_num) chunk_size = remaining_iters_num;
remaining_iters[thread_id] = (remaining_iters_num - chunk_size);
omp_unset_lock(&(remaining_iters_lock[thread_id]));
}
*start_iter = hi[thread_id]-remaining_iters_num;
*chunk = chunk_size;
}
void IntList_Insert(pIntList pList, int x, pArrNode an)
{
pIntListNode prev, p , newNode;
// assert(newNode!=NULL);
omp_set_lock(&listLock);
newNode = ArrNode_getNode(an);
if (pList->head == NULL) { /* list is empty, insert the first element */
pList->head = newNode;
}
else { /* list is not empty, find the right place to insert element */
p = pList->head;
prev = NULL;
while (p != NULL && p->data < newNode->data) {
prev = p;
p = p->next;
}
if (p == NULL) { /* insert as the last element */
prev->next = newNode;
newNode->prev = prev;
}
else if (prev == NULL) { /* insert as the first element */
pList->head = newNode;
newNode->next = p;
p->prev = newNode;
}
else { /* insert right between prev and p */
prev->next = newNode;
newNode->prev = prev;
newNode->next = p;
p->prev = newNode;
}
}
omp_unset_lock(&listLock);
}
double computeGraph(graph* G, graphSDG* SDGdata) {
VERT_T* endV;
LONG_T *degree, *numEdges, *pos, *pSums;
WEIGHT_T* w;
double elapsed_time;
#ifdef _OPENMP
omp_lock_t *vLock;
LONG_T chunkSize;
#endif
elapsed_time = get_seconds();
#ifdef _OPENMP
omp_set_num_threads(NUM_THREADS);
#endif
#ifdef _OPENMP
#pragma omp parallel
#endif
{
LONG_T i, j, u, n, m, tid, nthreads;
#ifdef DIAGNOSTIC
double elapsed_time_part;
#endif
#ifdef _OPENMP
nthreads = omp_get_num_threads();
tid = omp_get_thread_num();
#else
tid = 0;
nthreads = 1;
#endif
n = N;
m = M;
if (tid == 0) {
#ifdef _OPENMP
vLock = (omp_lock_t *) malloc(n*sizeof(omp_lock_t));
assert(vLock != NULL);
chunkSize = n/nthreads;
#endif
pos = (LONG_T *) malloc(m*sizeof(LONG_T));
assert(pos != NULL);
degree = (LONG_T *) calloc(n, sizeof(LONG_T));
assert(degree != NULL);
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds();
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for schedule(static, chunkSize)
for (i=0; i<n; i++) {
omp_init_lock(&vLock[i]);
}
#pragma omp barrier
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "Lock initialization time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
#pragma omp for
#endif
for (i=0; i<m; i++) {
u = SDGdata->startVertex[i];
#ifdef _OPENMP
omp_set_lock(&vLock[u]);
#endif
pos[i] = degree[u]++;
#ifdef _OPENMP
omp_unset_lock(&vLock[u]);
#endif
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "Degree computation time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for schedule(static, chunkSize)
for (i=0; i<n; i++) {
omp_destroy_lock(&vLock[i]);
}
if (tid == 0)
free(vLock);
#endif
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "Lock destruction time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
if (tid == 0) {
numEdges = (LONG_T *) malloc((n+1)*sizeof(LONG_T));
pSums = (LONG_T *) malloc(nthreads*sizeof(LONG_T));
}
#ifdef _OPENMP
#pragma omp barrier
#endif
prefix_sums(degree, numEdges, pSums, n);
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "Prefix sums time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#endif
if (tid == 0) {
free(degree);
free(pSums);
w = (WEIGHT_T *) malloc(m*sizeof(WEIGHT_T));
endV = (VERT_T *) malloc(m* sizeof(VERT_T));
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
#endif
for (i=0; i<m; i++) {
u = SDGdata->startVertex[i];
j = numEdges[u] + pos[i];
endV[j] = SDGdata->endVertex[i];
//TODO:
//w[j] = SDGdata->weight[i];
fprintf(stderr, "%d\n", SDGdata->weight[i]);
w[j] = 1;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "Edge data structure construction time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
if (tid == 0) {
free(pos);
G->n = n;
G->m = m;
G->numEdges = numEdges;
G->endV = endV;
G->weight = w;
}
#ifdef _OPENMP
#endif
}
/* Verification */
#if 0
fprintf(stderr, "SDG data:\n");
for (int i=0; i<SDGdata->m; i++) {
fprintf(stderr, "[%ld %ld %ld] ", SDGdata->startVertex[i],
SDGdata->endVertex[i], SDGdata->weight[i]);
}
fprintf(stderr, "\n");
for (int i=0; i<G->n + 1; i++) {
fprintf(stderr, "[%ld] ", G->numEdges[i]);
}
fprintf(stderr, "\nGraph:\n");
for (int i=0; i<G->n; i++) {
for (int j=G->numEdges[i]; j<G->numEdges[i+1]; j++) {
fprintf(stderr, "[%ld %ld %ld] ", i, G->endV[j], G->weight[j]);
}
}
#endif
free(SDGdata->startVertex);
free(SDGdata->endVertex);
free(SDGdata->weight);
elapsed_time = get_seconds() - elapsed_time;
return elapsed_time;
}
int kdFoF(KD kd,float fEps)
{
PARTICLE *p;
KDN *c;
int pi,pj,pn,cp;
int iGroup;
int *Fifo,iHead,iTail,nFifo;
float fEps2;
float dx,dy,dz,x,y,z,lx,ly,lz,sx,sy,sz,fDist2;
#ifdef _OPENMP
int idSelf;
omp_lock_t *locks;
for (pn=0;pn<kd->nActive;++pn) kd->p[pn].iTouched = -1;
/* We really want to make an independent lock for each particle. However, each lock
* seems to use a buttload of memory (something like 312 bytes per lock). Therefore,
* to ensure that we don't use too much memory, only use 1 lock per 100 particles.
* This should still provide very low lock contention while not using oodles of
* memory at the same time, since it is extremely rare that two threads will be looking
* two particles that map to the same lock at the same time.*/
kd->nHash = (int)(kd->nActive/100);
locks = (omp_lock_t *)malloc(kd->nHash*sizeof(omp_lock_t));
assert(locks != NULL);
for (pn=0;pn<kd->nHash;++pn) omp_init_lock(&locks[pn]);
#endif
p = kd->p;
c = kd->kdNodes;
lx = kd->fPeriod[0];
ly = kd->fPeriod[1];
lz = kd->fPeriod[2];
fEps2 = fEps*fEps;
for (pn=0;pn<kd->nActive;++pn) p[pn].iGroup = 0;
#pragma omp parallel default(none) shared(kd,locks,p,c,lx,ly,lz,fEps2) \
private(pi,pj,pn,cp,iGroup,Fifo,iHead,iTail,dx,dy,dz,x,y,z,sx,sy,sz,fDist2,idSelf,nFifo)
{
#ifdef _OPENMP
nFifo = kd->nActive/omp_get_num_threads();
idSelf = omp_get_thread_num();
#else
nFifo = kd->nActive;
#endif
Fifo = (int *)malloc(nFifo*sizeof(int));
assert(Fifo != NULL);
iHead = 0;
iTail = 0;
iGroup = 0;
#pragma omp for schedule(runtime)
for (pn=0;pn<kd->nActive;++pn) {
if (p[pn].iGroup) continue;
/*
** Mark it and add to the do-fifo.
*/
#ifdef _OPENMP
omp_set_lock(&locks[_hashLock(kd,pn)]);
if (p[pn].iTouched >= 0 && p[pn].iTouched < idSelf ) {
assert(p[pn].iGroup > 0);
omp_unset_lock(&locks[_hashLock(kd,pn)]);
continue;
}
p[pn].iTouched = idSelf;
iGroup = pn+1;
p[pn].iGroup = iGroup;
omp_unset_lock(&locks[_hashLock(kd,pn)]);
#else
++iGroup;
p[pn].iGroup = iGroup;
#endif
Fifo[iTail++] = pn;
if (iTail == nFifo) iTail = 0;
while (iHead != iTail) {
pi = Fifo[iHead++];
if (iHead == nFifo) iHead = 0;
/*
** Now do an fEps-Ball Gather!
*/
x = p[pi].r[0];
y = p[pi].r[1];
z = p[pi].r[2];
cp = ROOT;
while (1) {
INTERSECT(c,cp,fEps2,lx,ly,lz,x,y,z,sx,sy,sz);
/*
** We have an intersection to test.
*/
if (c[cp].iDim >= 0) {
cp = LOWER(cp);
continue;
}
else {
for (pj=c[cp].pLower;pj<=c[cp].pUpper;++pj) {
#ifdef _OPENMP
if (p[pj].iGroup == iGroup) {
/* We have already looked at this particle */
//assert(p[pj].iTouched == idSelf); particle is not locked.
continue;
}
if (p[pj].iTouched >= 0 && p[pj].iTouched < idSelf) {
/* Somebody more important than us is already looking at this
* particle. However, we do not yet know if this particle belongs
* in our group, so just skip it to save time but don't restart the
* entire group. */
// assert(p[pj].iGroup > 0); particle is not locked
continue;
}
#else
if (p[pj].iGroup) continue;
#endif
dx = sx - p[pj].r[0];
dy = sy - p[pj].r[1];
dz = sz - p[pj].r[2];
fDist2 = dx*dx + dy*dy + dz*dz;
if (fDist2 < fEps2) {
/*
** Mark it and add to the do-fifo.
*/
#ifdef _OPENMP
omp_set_lock(&locks[_hashLock(kd,pj)]);
if (p[pj].iTouched >= 0 && p[pj].iTouched < idSelf) {
/* Now we know this particle should be in our group. If somebody more
* important than us touched it, about the entire group. */
assert(p[pj].iGroup > 0);
omp_unset_lock(&locks[_hashLock(kd,pj)]);
iHead = iTail;
/*printf("Thread %d: Aborting group %d. p[%d].iOrder p.iGroup=%d p.iTouched=%d (Per-Particle2)\n",
idSelf, iGroup, pj, p[pj].iOrder, p[pj].iGroup, p[pj].iTouched);*/
goto RestartSnake;
}
p[pj].iTouched = idSelf;
p[pj].iGroup = iGroup;
omp_unset_lock(&locks[_hashLock(kd,pj)]);
#else
p[pj].iGroup = iGroup;
#endif
Fifo[iTail++] = pj;
if (iTail == nFifo) iTail = 0;
}
}
SETNEXT(cp);
if (cp == ROOT) break;
continue;
}
ContainedCell:
for (pj=c[cp].pLower;pj<=c[cp].pUpper;++pj) {
#ifdef _OPENMP
if (p[pj].iGroup == iGroup) continue;
if (p[pj].iTouched >= 0 && p[pj].iTouched < idSelf) {
/* Somebody more important that us is already looking at this
* group. Abort this entire group! */
//assert(p[pj].iGroup > 0); particle is not locked
iHead = iTail;
/*printf("Thread %d: Aborting group %d. p[%d].iOrder=%d p.iGroup=%d p.iTouched=%d (Per-Cell1)\n",
idSelf, iGroup, pj, p[pj].iOrder, p[pj].iGroup, p[pj].iTouched);*/
goto RestartSnake;
}
#else
if (p[pj].iGroup) continue;
#endif
/*
** Mark it and add to the do-fifo.
*/
#ifdef _OPENMP
omp_set_lock(&locks[_hashLock(kd,pj)]);
if (p[pj].iTouched >= 0 && p[pj].iTouched < idSelf) {
/* Check again in case somebody touched it before the lock. */
assert(p[pj].iGroup > 0);
omp_unset_lock(&locks[_hashLock(kd,pj)]);
iHead = iTail;
/*printf("Thread %d: Aborting group %d. p[%d].iGroup=%d p[%d].iTouched=%d (Per-Cell2)\n",
idSelf, iGroup, pj, p[pj].iGroup, pj, p[pj].iTouched);*/
goto RestartSnake;
}
p[pj].iTouched = idSelf;
p[pj].iGroup = iGroup;
omp_unset_lock(&locks[_hashLock(kd,pj)]);
#else
p[pj].iGroup = iGroup;
#endif
Fifo[iTail++] = pj;
if (iTail == nFifo) iTail = 0;
}
GetNextCell:
SETNEXT(cp);
if (cp == ROOT) break;
}
} /* End while(iHead != iTail) */
#ifdef _OPENMP
RestartSnake:
#endif
assert(iHead == iTail);
}
free(Fifo);
} /* End of the OpenMP PARALLEL section */
#ifdef _OPENMP
/* Now we have count how many groups there are. This is straightforward,
* since the number of groups is the number of particles whose groupID equals
* their particleID+1. */
pj = 0;
for (pn=0;pn<kd->nActive;++pn)
if (p[pn].iGroup == pn+1) ++pj;
kd->nGroup = (kd->nActive)+1;
free(locks);
#else
kd->nGroup = iGroup+1;
#endif
return(kd->nGroup-1);
}
void vertex_betweenness_centrality_parBFS(graph_t* G, double* BC, long numSrcs) {
attr_id_t *S; /* stack of vertices in the order of non-decreasing
distance from s. Also used to implicitly
represent the BFS queue */
plist_t* P; /* predecessors of a vertex v on shortest paths from s */
double* sig; /* No. of shortest paths */
attr_id_t* d; /* Length of the shortest path between every pair */
double* del; /* dependency of vertices */
attr_id_t *in_degree, *numEdges, *pSums;
attr_id_t* pListMem;
#if RANDSRCS
attr_id_t* Srcs;
#endif
attr_id_t *start, *end;
long MAX_NUM_PHASES;
attr_id_t *psCount;
#ifdef _OPENMP
omp_lock_t* vLock;
long chunkSize;
#endif
#ifdef DIAGNOSTIC
double elapsed_time;
#endif
int seed = 2387;
#ifdef _OPENMP
#pragma omp parallel firstprivate(G)
{
#endif
attr_id_t *myS, *myS_t;
attr_id_t myS_size;
long i, j, k, p, count, myCount;
long v, w, vert;
long k0, k1;
long numV, num_traversals, n, m, phase_num;
long start_iter, end_iter;
long tid, nthreads;
int* stream;
#ifdef DIAGNOSTIC
double elapsed_time_part;
#endif
#ifdef _OPENMP
int myLock;
tid = omp_get_thread_num();
nthreads = omp_get_num_threads();
#else
tid = 0;
nthreads = 1;
#endif
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time = get_seconds();
elapsed_time_part = get_seconds();
}
#endif
/* numV: no. of vertices to run BFS from = numSrcs */
numV = numSrcs;
n = G->n;
m = G->m;
/* Permute vertices */
if (tid == 0) {
#if RANDSRCS
Srcs = (attr_id_t *) malloc(n*sizeof(attr_id_t));
#endif
#ifdef _OPENMP
vLock = (omp_lock_t *) malloc(n*sizeof(omp_lock_t));
#endif
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
for (i=0; i<n; i++) {
omp_init_lock(&vLock[i]);
}
#endif
/* Initialize RNG stream */
stream = init_sprng(0, tid, nthreads, seed, SPRNG_DEFAULT);
#if RANDSRCS
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
Srcs[i] = i;
}
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
j = n * sprng(stream);
if (i != j) {
#ifdef _OPENMP
int l1 = omp_test_lock(&vLock[i]);
if (l1) {
int l2 = omp_test_lock(&vLock[j]);
if (l2) {
#endif
k = Srcs[i];
Srcs[i] = Srcs[j];
Srcs[j] = k;
#ifdef _OPENMP
omp_unset_lock(&vLock[j]);
}
omp_unset_lock(&vLock[i]);
}
#endif
}
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#endif
if (tid == 0) {
MAX_NUM_PHASES = 500;
}
#ifdef _OPENMP
#pragma omp barrier
#endif
/* Initialize predecessor lists */
/* The size of the predecessor list of each vertex is bounded by
its in-degree. So we first compute the in-degree of every
vertex */
if (tid == 0) {
P = (plist_t *) calloc(n, sizeof(plist_t));
in_degree = (attr_id_t *) calloc(n+1, sizeof(attr_id_t));
numEdges = (attr_id_t *) malloc((n+1)*sizeof(attr_id_t));
pSums = (attr_id_t *) malloc(nthreads*sizeof(attr_id_t));
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
#endif
for (i=0; i<m; i++) {
v = G->endV[i];
#ifdef _OPENMP
omp_set_lock(&vLock[v]);
#endif
in_degree[v]++;
#ifdef _OPENMP
omp_unset_lock(&vLock[v]);
#endif
}
prefix_sums(in_degree, numEdges, pSums, n);
if (tid == 0) {
pListMem = (attr_id_t *) malloc(m*sizeof(attr_id_t));
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
#endif
for (i=0; i<n; i++) {
P[i].list = pListMem + numEdges[i];
P[i].degree = in_degree[i];
P[i].count = 0;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() -elapsed_time_part;
fprintf(stderr, "In-degree computation time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
/* Allocate shared memory */
if (tid == 0) {
free(in_degree);
free(numEdges);
free(pSums);
S = (attr_id_t *) malloc(n*sizeof(attr_id_t));
sig = (double *) malloc(n*sizeof(double));
d = (attr_id_t *) malloc(n*sizeof(attr_id_t));
del = (double *) calloc(n, sizeof(double));
start = (attr_id_t *) malloc(MAX_NUM_PHASES*sizeof(attr_id_t));
end = (attr_id_t *) malloc(MAX_NUM_PHASES*sizeof(attr_id_t));
psCount = (attr_id_t *) malloc((nthreads+1)*sizeof(attr_id_t));
}
/* local memory for each thread */
myS_size = (2*n)/nthreads;
myS = (attr_id_t *) malloc(myS_size*sizeof(attr_id_t));
num_traversals = 0;
myCount = 0;
#ifdef _OPENMP
#pragma omp barrier
#endif
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
d[i] = -1;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "BC initialization time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
for (p=0; p<n; p++) {
#if RANDSRCS
i = Srcs[p];
#else
i = p;
#endif
if (G->numEdges[i+1] - G->numEdges[i] == 0) {
continue;
} else {
num_traversals++;
}
if (num_traversals == numV + 1) {
break;
}
if (tid == 0) {
sig[i] = 1;
d[i] = 0;
S[0] = i;
start[0] = 0;
end[0] = 1;
}
count = 1;
phase_num = 0;
#ifdef _OPENMP
#pragma omp barrier
#endif
while (end[phase_num] - start[phase_num] > 0) {
myCount = 0;
start_iter = start[phase_num];
end_iter = end[phase_num];
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for schedule(dynamic) nowait
#endif
for (vert = start_iter; vert < end_iter; vert++) {
v = S[vert];
for (j=G->numEdges[v]; j<G->numEdges[v+1]; j++) {
w = G->endV[j];
if (v != w) {
#ifdef _OPENMP
myLock = omp_test_lock(&vLock[w]);
if (myLock) {
#endif
/* w found for the first time? */
if (d[w] == -1) {
if (myS_size == myCount) {
/* Resize myS */
myS_t = (attr_id_t *)
malloc(2*myS_size*sizeof(attr_id_t));
memcpy(myS_t, myS,
myS_size*sizeof(attr_id_t));
free(myS);
myS = myS_t;
myS_size = 2*myS_size;
}
myS[myCount++] = w;
d[w] = d[v] + 1;
sig[w] = sig[v];
P[w].list[P[w].count++] = v;
} else if (d[w] == d[v] + 1) {
sig[w] += sig[v];
P[w].list[P[w].count++] = v;
}
#ifdef _OPENMP
omp_unset_lock(&vLock[w]);
} else {
if ((d[w] == -1) || (d[w] == d[v]+ 1)) {
omp_set_lock(&vLock[w]);
sig[w] += sig[v];
P[w].list[P[w].count++] = v;
omp_unset_lock(&vLock[w]);
}
}
#endif
}
}
}
/* Merge all local stacks for next iteration */
phase_num++;
if (tid == 0) {
if (phase_num >= MAX_NUM_PHASES) {
fprintf(stderr, "Error: Max num phases set to %ld\n",
MAX_NUM_PHASES);
fprintf(stderr, "Diameter of input network greater than"
" this value. Increase MAX_NUM_PHASES"
" in vertex_betweenness_centrality_parBFS()\n");
exit(-1);
}
}
psCount[tid+1] = myCount;
#ifdef _OPENMP
#pragma omp barrier
#endif
if (tid == 0) {
start[phase_num] = end[phase_num-1];
psCount[0] = start[phase_num];
for(k=1; k<=nthreads; k++) {
psCount[k] = psCount[k-1] + psCount[k];
}
end[phase_num] = psCount[nthreads];
}
#ifdef _OPENMP
#pragma omp barrier
#endif
k0 = psCount[tid];
k1 = psCount[tid+1];
for (k = k0; k < k1; k++) {
S[k] = myS[k-k0];
}
count = end[phase_num];
}
phase_num--;
while (phase_num > 0) {
start_iter = start[phase_num];
end_iter = end[phase_num];
#ifdef _OPENMP
#pragma omp for schedule(static) nowait
#endif
for (j=start_iter; j<end_iter; j++) {
w = S[j];
for (k = 0; k<P[w].count; k++) {
v = P[w].list[k];
#ifdef _OPENMP
omp_set_lock(&vLock[v]);
#endif
del[v] = del[v] + sig[v]*(1+del[w])/sig[w];
#ifdef _OPENMP
omp_unset_lock(&vLock[v]);
#endif
}
BC[w] += del[w];
}
phase_num--;
#ifdef _OPENMP
#pragma omp barrier
#endif
}
#ifdef _OPENMP
chunkSize = n/nthreads;
#pragma omp for schedule(static, chunkSize) nowait
#endif
for (j=0; j<count; j++) {
w = S[j];
d[w] = -1;
del[w] = 0;
P[w].count = 0;
}
#ifdef _OPENMP
#pragma omp barrier
#endif
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "BC computation time: %lf seconds\n",
elapsed_time_part);
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#endif
#ifdef _OPENMP
#pragma omp for
for (i=0; i<n; i++) {
omp_destroy_lock(&vLock[i]);
}
#endif
free(myS);
if (tid == 0) {
free(S);
free(pListMem);
free(P);
free(sig);
free(d);
free(del);
#ifdef _OPENMP
free(vLock);
#endif
free(start);
free(end);
free(psCount);
#ifdef DIAGNOSTIC
elapsed_time = get_seconds() - elapsed_time;
fprintf(stderr, "Time taken: %lf\n seconds", elapsed_time);
#endif
#if RANDSRCS
free(Srcs);
#endif
}
free_sprng(stream);
#ifdef _OPENMP
}
#endif
}
int
main (void)
{
double d, e;
int l;
omp_lock_t lck;
omp_nest_lock_t nlck;
d = omp_get_wtime ();
omp_init_lock (&lck);
omp_set_lock (&lck);
if (omp_test_lock (&lck))
abort ();
omp_unset_lock (&lck);
if (! omp_test_lock (&lck))
abort ();
if (omp_test_lock (&lck))
abort ();
omp_unset_lock (&lck);
omp_destroy_lock (&lck);
omp_init_nest_lock (&nlck);
if (omp_test_nest_lock (&nlck) != 1)
abort ();
omp_set_nest_lock (&nlck);
if (omp_test_nest_lock (&nlck) != 3)
abort ();
omp_unset_nest_lock (&nlck);
omp_unset_nest_lock (&nlck);
if (omp_test_nest_lock (&nlck) != 2)
abort ();
omp_unset_nest_lock (&nlck);
omp_unset_nest_lock (&nlck);
omp_destroy_nest_lock (&nlck);
omp_set_dynamic (1);
if (! omp_get_dynamic ())
abort ();
omp_set_dynamic (0);
if (omp_get_dynamic ())
abort ();
omp_set_nested (1);
if (! omp_get_nested ())
abort ();
omp_set_nested (0);
if (omp_get_nested ())
abort ();
omp_set_num_threads (5);
if (omp_get_num_threads () != 1)
abort ();
if (omp_get_max_threads () != 5)
abort ();
if (omp_get_thread_num () != 0)
abort ();
omp_set_num_threads (3);
if (omp_get_num_threads () != 1)
abort ();
if (omp_get_max_threads () != 3)
abort ();
if (omp_get_thread_num () != 0)
abort ();
l = 0;
#pragma omp parallel reduction (|:l)
{
l = omp_get_num_threads () != 3;
l |= omp_get_thread_num () < 0;
l |= omp_get_thread_num () >= 3;
#pragma omp master
l |= omp_get_thread_num () != 0;
}
if (l)
abort ();
if (omp_get_num_procs () <= 0)
abort ();
if (omp_in_parallel ())
abort ();
#pragma omp parallel reduction (|:l)
l = ! omp_in_parallel ();
#pragma omp parallel reduction (|:l) if (1)
l = ! omp_in_parallel ();
if (l)
abort ();
e = omp_get_wtime ();
if (d > e)
abort ();
d = omp_get_wtick ();
/* Negative precision is definitely wrong,
bigger than 1s clock resolution is also strange. */
if (d <= 0 || d > 1)
abort ();
return 0;
}
static void
sort1 (int *array, int count)
{
omp_lock_t lock;
struct int_pair_stack global_stack;
int busy = 1;
int num_threads;
omp_init_lock (&lock);
init_int_pair_stack (&global_stack);
#pragma omp parallel firstprivate (array, count)
{
int lo = 0, hi = 0, mid, next_lo, next_hi;
bool idle = true;
struct int_pair_stack local_stack;
init_int_pair_stack (&local_stack);
if (omp_get_thread_num () == 0)
{
num_threads = omp_get_num_threads ();
hi = count - 1;
idle = false;
}
for (;;)
{
if (hi - lo < THRESHOLD)
{
insertsort (array, lo, hi);
lo = hi;
}
if (lo >= hi)
{
if (size_int_pair_stack (&local_stack) == 0)
{
again:
omp_set_lock (&lock);
if (size_int_pair_stack (&global_stack) == 0)
{
if (!idle)
busy--;
if (busy == 0)
{
omp_unset_lock (&lock);
break;
}
omp_unset_lock (&lock);
idle = true;
while (size_int_pair_stack (&global_stack) == 0
&& busy)
busy_wait ();
goto again;
}
if (idle)
busy++;
pop_int_pair_stack (&global_stack, &lo, &hi);
omp_unset_lock (&lock);
idle = false;
}
else
pop_int_pair_stack (&local_stack, &lo, &hi);
}
mid = partition (array, lo, hi);
if (mid - lo < hi - mid)
{
next_lo = mid;
next_hi = hi;
hi = mid - 1;
}
else
{
next_lo = lo;
next_hi = mid - 1;
lo = mid;
}
if (next_hi - next_lo < THRESHOLD)
insertsort (array, next_lo, next_hi);
else
{
if (size_int_pair_stack (&global_stack) < num_threads - 1)
{
int size;
omp_set_lock (&lock);
size = size_int_pair_stack (&global_stack);
if (size < num_threads - 1 && size < STACK_SIZE)
push_int_pair_stack (&global_stack, next_lo, next_hi);
else
push_int_pair_stack (&local_stack, next_lo, next_hi);
omp_unset_lock (&lock);
}
else
push_int_pair_stack (&local_stack, next_lo, next_hi);
}
}
}
omp_destroy_lock (&lock);
}
int main(int argc, char ** argv)
{
long iterations; /* total number of reference pair counter updates */
long stream_size; /* length of stream triad creating private work */
int page_fit; /* indicates that counters fit on different pages */
size_t store_size; /* amount of space reserved for counters */
double *pcounter1,
*pcounter2; /* pointers to counters */
double cosa, sina; /* cosine and sine of rotation angle */
double *counter_space; /* pointer to space reserved for counters */
double refcounter1,
refcounter2; /* reference values for counters */
double epsilon=1.e-7; /* required accuracy */
omp_lock_t counter_lock; /* lock that guards access to counters */
double refcount_time; /* timing parameter */
int nthread_input; /* thread parameters */
int nthread;
/*********************************************************************
** process and test input parameters
*********************************************************************/
printf("Parallel Research Kernels version %s\n", PRKVERSION);
printf("OpenMP exclusive access test RefCount, shared counters\n");
if (argc != 4){
printf("Usage: %s <# threads> <# counter pair updates> <# private stream size>\n", *argv);
return(1);
}
nthread_input = atoi(*++argv);
if ((nthread_input < 1) || (nthread_input > MAX_THREADS)) {
printf("ERROR: Invalid number of threads: %d\n", nthread_input);
exit(EXIT_FAILURE);
}
iterations = atol(*++argv);
if (iterations < 1){
printf("ERROR: iterations must be >= 1 : %d \n",iterations);
exit(EXIT_FAILURE);
}
stream_size = atol(*++argv);
if (stream_size < 0) {
printf("ERROR: private stream size %ld must be non-negative\n", stream_size);
exit(EXIT_FAILURE);
}
omp_set_num_threads(nthread_input);
/* initialize shared counters; we put them on different pages, if possible.
If the page size equals the whole memory, this will fail, and we reduce
the space required */
page_fit = 1;
store_size = (size_t) getpagesize();
#ifdef VERBOSE
printf("Page size = %d\n", getpagesize());
#endif
counter_space = (double *) malloc(store_size+sizeof(double));
while (!counter_space && store_size>2*sizeof(double)) {
page_fit=0;
store_size/=2;
counter_space = (double *) malloc(store_size+sizeof(double));
}
if (!counter_space) {
printf("ERROR: could not allocate space for counters\n");
exit(EXIT_FAILURE);
}
#ifdef VERBOSE
if (!page_fit) printf("Counters do not fit on different pages\n");
else printf("Counters fit on different pages\n");
#endif
pcounter1 = counter_space;
pcounter2 = counter_space + store_size/sizeof(double);
COUNTER1 = 1.0;
COUNTER2 = 0.0;
cosa = cos(1.0);
sina = sin(1.0);
/* initialize the lock on which we will be pounding */
omp_init_lock(&counter_lock);
#pragma omp parallel
{
long iter, j; /* dummies */
double tmp1; /* local copy of previous value of COUNTER1 */
double *a, *b, *c;/* private vectors */
int num_error=0;/* errors in private stream execution */
double aj, bj, cj;
long space;
space = 3*sizeof(double)*stream_size;
a = (double *) malloc(space);
if (!a) {
printf("ERROR: Could not allocate %ld words for private streams\n",
space);
exit(EXIT_FAILURE);
}
b = a + stream_size;
c = b + stream_size;
for (j=0; j<stream_size; j++) {
a[j] = A0;
b[j] = B0;
c[j] = C0;
}
#pragma omp master
{
nthread = omp_get_num_threads();
if (nthread != nthread_input) {
num_error = 1;
printf("ERROR: number of requested threads %d does not equal ",
nthread_input);
printf("number of spawned threads %d\n", nthread);
}
else {
printf("Number of threads = %d\n",nthread_input);
printf("Number of counter pair updates = %ld\n", iterations);
printf("Length of private stream = %ld\n", stream_size);
#ifdef DEPENDENT
printf("Dependent counter pair update\n");
#else
printf("Independent counter pair updates using");
#ifdef ATOMIC
printf(" atomic operations\n");
#else
printf(" using locks\n");
#endif
#endif
}
}
bail_out(num_error);
/* do one warmup iteration outside main loop to avoid overhead */
#ifdef DEPENDENT
omp_set_lock(&counter_lock);
tmp1 = COUNTER1;
COUNTER1 = cosa*tmp1 - sina*COUNTER2;
COUNTER2 = sina*tmp1 + cosa*COUNTER2;
omp_unset_lock(&counter_lock);
#else
#ifndef ATOMIC
omp_set_lock(&counter_lock);
#else
#pragma omp atomic
#endif
COUNTER1++;
#ifdef ATOMIC
#pragma omp atomic
#endif
COUNTER2++;
#ifndef ATOMIC
omp_unset_lock(&counter_lock);
#endif
#endif
/* give each thread some (overlappable) work to do */
private_stream(a, b, c, stream_size);
#pragma omp master
{
refcount_time = wtime();
}
#pragma omp for
/* start with iteration nthread to take into account pre-loop iter */
for (iter=nthread; iter<=iterations; iter++) {
#ifdef DEPENDENT
omp_set_lock(&counter_lock);
tmp1 = COUNTER1;
COUNTER1 = cosa*tmp1 - sina*COUNTER2;
COUNTER2 = sina*tmp1 + cosa*COUNTER2;
omp_unset_lock(&counter_lock);
#else
#ifndef ATOMIC
omp_set_lock(&counter_lock);
#else
#pragma omp atomic
#endif
COUNTER1++;
#ifdef ATOMIC
#pragma omp atomic
#endif
COUNTER2++;
#ifndef ATOMIC
omp_unset_lock(&counter_lock);
#endif
#endif
/* give each thread some (overlappable) work to do */
private_stream(a, b, c, stream_size);
}
#pragma omp master
{
refcount_time = wtime() - refcount_time;
}
/* check whether the private work has been done correctly */
aj = A0; bj = B0; cj = C0;
#pragma omp for
for (iter=0; iter<=iterations; iter++) {
aj += bj + SCALAR*cj;
}
for (j=0; j<stream_size; j++) {
num_error += MAX(ABS(a[j]-aj)>epsilon,num_error);
}
if (num_error>0) {
printf("ERROR: Thread %d encountered errors in private work\n",
omp_get_thread_num());
}
bail_out(num_error);
} /* end of OpenMP parallel region */
#ifdef DEPENDENT
refcounter1 = cos(iterations+1);
refcounter2 = sin(iterations+1);
#else
refcounter1 = (double)(iterations+2);
refcounter2 = (double)(iterations+1);
#endif
if ((ABS(COUNTER1-refcounter1)>epsilon) ||
(ABS(COUNTER2-refcounter2)>epsilon)) {
printf("ERROR: Incorrect or inconsistent counter values %13.10lf %13.10lf; ",
COUNTER1, COUNTER2);
printf("should be %13.10lf, %13.10lf\n", refcounter1, refcounter2);
}
else {
#ifdef VERBOSE
printf("Solution validates; Correct counter values %13.10lf %13.10lf\n",
COUNTER1, COUNTER2);
#else
printf("Solution validates\n");
#endif
printf("Rate (MCPUPs/s): %lf time (s): %lf\n",
iterations/refcount_time*1.e-6, refcount_time);
}
exit(EXIT_SUCCESS);
}
void unlock() { omp_unset_lock( &m_lock ); }
FLA_Error FLA_Gemm_nn_omp_var15( FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj C, fla_gemm_t* cntl )
{
FLA_Obj AT, A0,
AB, A1,
A2;
FLA_Obj CT, C0,
CB, C1,
C2;
FLA_Obj AL, AR, A10, A11, A12;
FLA_Obj BT, B0,
BB, B1,
B2;
FLA_Obj C1_local;
int i, j, lock_ldim, lock_i;
int b_m, b_k;
FLA_Part_2x1( A, &AT,
&AB, 0, FLA_TOP );
FLA_Part_2x1( C, &CT,
&CB, 0, FLA_TOP );
#pragma intel omp parallel taskq
{
while ( FLA_Obj_length( AT ) < FLA_Obj_length( A ) )
{
b_m = FLA_Determine_blocksize( A, AT, FLA_TOP, FLA_Cntl_blocksize( cntl ) );
FLA_Repart_2x1_to_3x1( AT, &A0,
/* ** */ /* ** */
&A1,
AB, &A2, b_m, FLA_BOTTOM );
FLA_Repart_2x1_to_3x1( CT, &C0,
/* ** */ /* ** */
&C1,
CB, &C2, b_m, FLA_BOTTOM );
/*------------------------------------------------------------*/
/* C1 = alpha * A1 * B + C1; */
FLA_Part_1x2( A1, &AL, &AR, 0, FLA_LEFT );
FLA_Part_2x1( B, &BT,
&BB, 0, FLA_TOP );
while ( FLA_Obj_width( AL ) < FLA_Obj_width( A ) )
{
b_k = FLA_Determine_blocksize( A, AL, FLA_LEFT, FLA_Cntl_blocksize( cntl ) );
// Get the index of the current partition.
// FIX THIS: need + b_m - 1 or something like this
//j = FLA_Obj_length( CT ) / b_m;
//i = FLA_Obj_width( AL ) / b_k;
//lock_ldim = FLA_get_num_threads_in_m_dim(omp_get_num_threads());
lock_i = FLA_Obj_length( CT ) / b_m;
FLA_Repart_1x2_to_1x3( AL, /**/ AR, &A10, /**/ &A11, &A12,
b_k, FLA_RIGHT );
FLA_Repart_2x1_to_3x1( BT, &B0,
/* ** */ /* ** */
&B1,
BB, &B2, b_k, FLA_BOTTOM );
/*------------------------------------------------------------*/
/* C1 = alpha * A11 * B1 + C1; */
//// FLA_Gemm( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE,
//// alpha, A11, B1, FLA_ONE, C1 );
#pragma intel omp task captureprivate( lock_i, A11, B1, C1 ), private( C1_local )
{
FLA_Obj_create_conf_to( FLA_NO_TRANSPOSE, C1, &C1_local );
FLA_Obj_set_to_zero( C1_local );
/* C1_local = alpha * A1 * B11 + C1_local; */
FLA_Gemm_external( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE,
alpha, A11, B1, FLA_ONE, C1_local );
// Acquire lock[i] (the lock for C1).
omp_set_lock( &fla_omp_lock[lock_i] );
/* C1 += C1_local */
FLA_Axpy_external( FLA_ONE, C1_local, C1 );
//FLA_Axpy_sync_pipeline2( j*lock_ldim, FLA_ONE, C1_local, C1 );
//FLA_Axpy_sync_circular2( j*lock_ldim, i, FLA_ONE, C1_local, C1 );
//REF_Axpy_sync_circular2( j*lock_ldim, i, FLA_ONE, C1_local, C1 );
// Release lock[i] (the lock for C1).
omp_unset_lock( &fla_omp_lock[lock_i] );
FLA_Obj_free( &C1_local );
}
/*------------------------------------------------------------*/
FLA_Cont_with_1x3_to_1x2( &AL, /**/ &AR, A10, A11, /**/ A12,
FLA_LEFT );
FLA_Cont_with_3x1_to_2x1( &BT, B0,
B1,
/* ** */ /* ** */
&BB, B2, FLA_TOP );
}
/*------------------------------------------------------------*/
FLA_Cont_with_3x1_to_2x1( &AT, A0,
A1,
/* ** */ /* ** */
&AB, A2, FLA_TOP );
FLA_Cont_with_3x1_to_2x1( &CT, C0,
C1,
/* ** */ /* ** */
&CB, C2, FLA_TOP );
}
}
return FLA_SUCCESS;
}
void Roundworld::process()
{
if ( !pause )
{
killHalf();
expireFood();
autoinsertFood();
expireCritters();
autoexchangeCritters();
autosaveCritters();
autoinsertCritters();
// adjust gravity vectors of all entities' rigid bodies
unsigned int j, b;
Food* f;
CritterB* bod;
btRigidBody* bo;
for( j=0; j < entities.size(); j++)
{
if ( entities[j]->type == FOOD )
{
// f = food[j];
Food* f = static_cast<Food*>( entities[j] );
for( b=0; b < f->body.bodyparts.size(); b++)
{
bo = f->body.bodyparts[b]->body;
bo->setGravity( -(bo->getCenterOfMassPosition().normalized()*10) );
}
}
}
for( j=0; j < critters.size(); j++)
{
bod = critters[j];
for( b=0; b < bod->body.bodyparts.size(); b++)
{
bo = bod->body.bodyparts[b]->body;
bo->setGravity( -(bo->getCenterOfMassPosition().normalized()*10) );
}
}
if ( *critter_raycastvision == 0 )
{
renderVision();
grabVision();
}
// do a bullet step
m_dynamicsWorld->stepSimulation(0.016667f, 0, 0.016667f);
// m_dynamicsWorld->stepSimulation(Timer::Instance()->bullet_ms / 1000.f);
int lmax = (int)critters.size();
CritterB *c;
float freeEnergyc = 0.0f;
// FIXME USE FROM WORLDB
omp_set_num_threads( *threads );
#pragma omp parallel for ordered shared(freeEnergyc, lmax) private(c) // ordered
for( int i=0; i < lmax; i++)
{
c = critters[i];
omp_set_lock(&my_lock1);
checkCollisions( c );
omp_unset_lock(&my_lock1);
// process
c->process();
// record critter used energy
freeEnergyc += c->energyUsed;
// process Output Neurons
eat(c);
// procreation if procreation energy trigger is hit
omp_set_lock(&my_lock1);
procreate(c);
omp_unset_lock(&my_lock1);
}
freeEnergy += freeEnergyc;
getGeneralStats();
}
}
void Slic::_GenerateSuperpixels()
{
const int MAX_ITER = 10;
for(int I=0;I<MAX_ITER;++I)
{
clock_t t1 = clock();
std::cout<<"This is the "<<I<<"th circulation:"<<std::endl;
omp_init_lock(&lock);
#pragma omp parallel for
for(int n=0;n<_N;++n)
{
for(int m=0;m<_M;++m)
{
// Init
int nXOff = m*_regionSize;
int nYOff = n*_regionSize;
int nXSize = m==(_M-1)? _width - m*_regionSize : _regionSize;
int nYSize = n==(_N-1)? _height - n*_regionSize : _regionSize;
uchar *bufferSrc = new uchar[nXSize*nYSize*_dataSize*_bandCount];
int *bufferDst = new int[nXSize*nYSize];
omp_set_lock(&lock);
_poSrcDS->RasterIO(GF_Read,nXOff,nYOff,nXSize,nYSize,bufferSrc,nXSize,nYSize,_dataType,_bandCount,0,0,0,0);
_poDstDS->GetRasterBand(1)->RasterIO(GF_Read,nXOff,nYOff,nXSize,nYSize,bufferDst,nXSize,nYSize,GDT_Int32,0,0);
omp_unset_lock(&lock);
std::vector< int > candidateCenterID;
for(int i=-1;i<2;++i)
{
for(int j=-1;j<2;++j)
{
if ((n+i)>=0 && (n+i)<_N && (m+j)>=0 && (m+j)<_M)
candidateCenterID.push_back( (n+i) * _M + (m+j) ) ;
}
}
// GetFeatureInfo
FeatureVector featureVec(_bandCount + 2);
for(int i=0,index=0;i<nYSize;++i)
{
for(int j=0;j<nXSize;++j,++index)
{
uchar* p = bufferSrc;
for(int k=0;k<_bandCount;++k,p += nXSize*nYSize*_dataSize)
{
featureVec[k]= SRCVAL(p,_dataType,index)/_regularizer;
}
featureVec[_bandCount] = static_cast<double>(nXOff+j)/_regionSize; //x
featureVec[_bandCount+1] = static_cast<double>(nYOff+i)/_regionSize; //y
bufferDst[i*nXSize+j] = _GetNearestCenter(candidateCenterID,featureVec);
}
}
omp_set_lock(&lock);
_poDstDS->GetRasterBand(1)->RasterIO(GF_Write,nXOff,nYOff,nXSize,nYSize,bufferDst,nXSize,nYSize,GDT_Int32,0,0);
omp_unset_lock(&lock);
delete []bufferSrc;
delete []bufferDst;
}
}
omp_destroy_lock(&lock);
_ComputeNewCenterVector();
std::cout<<"This circle cost: "<<static_cast<double>(clock()-t1)/CLOCKS_PER_SEC<<"s"<<std::endl;
}
}
void Mutex::unlock() {
omp_unset_lock(&lock_);
}
void PhotonMap::throughputByDensityEstimation(vec3f &color, Path &eyeMergePath,
std::vector<LightPoint> &surfaceVertices, std::vector<LightPoint> &volumeVertices)
{
class Query{
PhotonMap *photonMap;
vec3f contrib;
vec3f position;
vec3f hitNormal;
float radius;
int photonsNum;
Ray outRay;
float GaussianKernel(float mahalanobisDist) const{
double exponent = exp((double)-mahalanobisDist/2);
//photonMap->fout << " Gaussian exp = " << exponent << std::endl;
return exponent / (2*M_PI);
}
float Kernel(float distSqr, float radiusSqr) const{
float s = MAX(0, 1 - distSqr / radiusSqr);
return 3 * s * s / M_PI;
}
public:
Query(PhotonMap *map, float r, int n) : photonMap(map), radius(r), photonsNum(n)
{}
bool volumeMedia;
void SetContrib(const vec3f &color) { contrib = color; }
void SetPosition(const vec3f &pos) { position = pos; }
void SetOutRay(const Ray &ray) { outRay = ray; }
void SetNormal(const vec3f &n) { hitNormal = n; }
vec3f GetContrib() const { return contrib; }
vec3f GetPosition() const { return position; }
void Process(const LightPoint &lightPoint){
if(volumeMedia && lightPoint.photonType != Ray::INVOL) return ;
if(!volumeMedia && lightPoint.photonType != Ray::OUTVOL) return ;
if(!lightPoint.pathThePointIn || lightPoint.indexInThePath < 0)
return;
Path &lightPath = *lightPoint.pathThePointIn;
int index = lightPoint.indexInThePath;
/*
if (volumeMedia && lightPath[index].insideObject != outRay.insideObject)
{
printf("aye\n");
return;
}
if (!volumeMedia && lightPath[index].contactObject != outRay.contactObject)
{
printf("aye\n");
return;
}
*/
vec3f photonThroughput(1,1,1);
for(int i = 0; i < index; i++){
photonThroughput *= lightPath[i].color / lightPath[i].directionProb / lightPath[i].originProb;
photonThroughput *= lightPath[i].getCosineTerm();
float dist = (lightPath[i].origin-lightPath[i+1].origin).length();
photonThroughput *= lightPath[i].getRadianceDecay(dist);
}
photonThroughput /= lightPath[index].originProb;
// runs here, photon's f/p is done.
Ray photonRay = lightPath[index];
photonRay.direction = lightPath[index-1].direction;
vec3f color = photonThroughput * photonRay.getBSDF(outRay);
float distSqr = powf((outRay.origin-lightPath[index].origin).length(), 2);
if(intensity(color) < 1e-6f) return ;
float kernel = Kernel(distSqr, radius*radius);
float normalization = volumeMedia ? kernel/(photonsNum*radius*radius*radius) : kernel/(photonsNum*radius*radius);
//float normalization = volumeMedia==false ? 1.0 / (photonsNum*PI*radius*radius) : 1.0 / (photonsNum*PI*4.0/3*radius*radius*radius);
contrib += color * normalization;
}
double sumWeight;
int photonsCount;
void weightScale(){
contrib /= ( sumWeight / photonsCount );
}
};
Query query(this, mRadius, mPhotonsNum);
vec3f Tr(1,1,1), SurfaceColor(0,0,0), VolumeColor(0,0,0);
int mergeIndex = 1;
for(int i = 1; i < eyeMergePath.size(); i++){
float dist = MAX((eyeMergePath[i-1].origin-eyeMergePath[i].origin).length(), EPSILON);
if(eyeMergePath[i-1].insideObject && eyeMergePath[i-1].insideObject->isVolumetric()){
if(eyeMergePath[i-1].insideObject->isHomogeneous())
{
// ray marching volume radiance
Ray volThroughRay = eyeMergePath[i-1];
SceneVPMObject *volume = static_cast<SceneVPMObject*>(volThroughRay.insideObject);
float stepSize = volume->stepSize;
int N = dist / stepSize;
if(N == 0) N++;
float step = dist / N;
float offset = step * RandGenerator::genFloat();
float t = offset;
Tr *= volume->getRadianceDecay(volThroughRay, offset);
for(int j = 0; j < N; j++, t+=step){
query.SetContrib(vec3f(0,0,0));
query.SetPosition(volThroughRay.origin + volThroughRay.direction*t);
Ray outRay = volThroughRay;
outRay.direction = -volThroughRay.direction;
outRay.origin = volThroughRay.origin + volThroughRay.direction*t;
outRay.contactObject = NULL;
query.SetOutRay(outRay);
query.volumeMedia = true;
volumeHashGrid.Process(volumeVertices, query);
Tr *= volume->getRadianceDecay(outRay, step);
vec3f volColor = query.GetContrib();
VolumeColor += volColor * Tr * step;
}
}
else{
// ray marching volume radiance
Ray volThroughRay = eyeMergePath[i-1];
HeterogeneousVolume *volume = static_cast<HeterogeneousVolume*>(volThroughRay.insideObject);
float stepSize = volume->getStepSize();
int N = dist / stepSize;
if(N == 0) N++;
float step = dist / N;
float offset = step * RandGenerator::genFloat();
float t = offset;
Tr *= volume->getRadianceDecay(volThroughRay, offset);
for(int j = 0; j < N; j++, t+=step){
query.SetContrib(vec3f(0,0,0));
query.SetPosition(volThroughRay.origin + volThroughRay.direction*t);
Ray outRay = volThroughRay;
outRay.direction = -volThroughRay.direction;
outRay.origin = volThroughRay.origin + volThroughRay.direction*t;
outRay.contactObject = NULL;
query.SetOutRay(outRay);
query.volumeMedia = true;
volumeHashGrid.Process(volumeVertices, query);
Tr *= volume->getRadianceDecay(outRay, step);
vec3f volColor = query.GetContrib();
VolumeColor += volColor * Tr * step;
}
}
}
else
{
if (eyeMergePath[i - 1].insideObject)
Tr *= eyeMergePath[i - 1].getRadianceDecay(dist);
}
if(eyeMergePath[i].contactObject && eyeMergePath[i].contactObject->emissive()){
// eye path hit light, surface color equals to light radiance
SurfaceColor = eyeMergePath[i].color;
mergeIndex = i;
break;
}
if(eyeMergePath[i].contactObject && eyeMergePath[i].directionSampleType == Ray::RANDOM){
// non-specular photon density estimation
if(eyeMergePath[i].contactObject->isVolumetric())
continue;
query.SetContrib(vec3f(0,0,0));
query.SetPosition(eyeMergePath[i].origin);
Ray outRay = eyeMergePath[i];
outRay.direction = -eyeMergePath[i-1].direction;
query.SetOutRay(outRay);
query.volumeMedia = false;
Ray fromRay = eyeMergePath[i-1];
omp_set_lock(&surfaceHashGridLock);
surfaceHashGrid.Process(surfaceVertices, query);
omp_unset_lock(&surfaceHashGridLock);
SurfaceColor = query.GetContrib();
mergeIndex = i;
break;
}
}
color = Tr * SurfaceColor + VolumeColor;
if (rayMarching)
{
for(int i = 0; i < 1/*eyeMergePath.size()-1*/; i++){
color *= eyeMergePath[i].getCosineTerm() * eyeMergePath[i].color
/ eyeMergePath[i].directionProb / eyeMergePath[i].originProb;
}
}
else
{
for(int i = 0; i < mergeIndex; i++){
color *= eyeMergePath[i].getCosineTerm() * eyeMergePath[i].color
/ eyeMergePath[i].directionProb / eyeMergePath[i].originProb;
if (i + 1 < mergeIndex)
{
float dist = (eyeMergePath[i].origin - eyeMergePath[i+1].origin).length();
color *= eyeMergePath[i].getRadianceDecay(dist);
}
}
}
}
triangulation triangulate_cube_random(data_list * data) {
int dim = data_list_dim(data);
cube_points cube = gen_cube_points(dim);
triangulation result;
omp_lock_t result_lock;
omp_init_lock(&result_lock); //If we found a triangulation, use this lock!
facet_acute_data parameters; //Parameters for conform_
triangulation tmp_triang; //Triangulation we are expanding in current thread
ptetra tet_list; //List of tetrahedrons, used in the parallel section
unsigned short tet_list_len; //Holds the length of this list
int triangulation_found = 0; //Stop if one of the threads has found a triangulation!
int rand_bound, i;
unsigned short tet_max, tet_min, tet_rand, tet_add;
size_t max_volume;
//Start the parallel loop!
#pragma omp parallel default(none) \
private(parameters, tmp_triang, tet_list, tet_list_len, rand_bound, i,max_volume,tet_max, tet_min, tet_rand, tet_add) \
shared(result, result_lock, cube,data,dim, triangulation_found)
{
//Initalization for each thread
parameters.cube = &cube;
parameters.boundary_func = &triangle_boundary_cube;
parameters.data = data;
parameters.store_acute_ind = 1;
parameters.acute_ind = malloc(sizeof(vert_index) * cube.len);
tet_list = malloc(sizeof(tetra) * cube.len);
max_volume = 0;
while (!triangulation_found) { //Not found a triangulation
//Initalize the triangulation variables
tmp_triang = triangulation_init(dim);
tet_list_len = 0;
//Start triangle (0,0,0), (rand,0,0), (rand,rand,0)
tmp_triang.bound_len = 1;
tmp_triang.bound_tri = triangulation_start_facet(data);
//printf("Thread %d with iteration %zu starts with:\n", omp_get_thread_num(), ++iterations);
//print_triangle(tmp_triang.bound_tri);
//While we have triangles on the boundary
while (tmp_triang.bound_len > 0) {
/*
* We are going to add a tetrahedron on the boundary triangle.
* To do so, we select a random triangle on the boundary. Then we generate all the
* acute tetrahedra (above and below) with facets in our possible list.
* From this list we remove all the tetrahedrons that intersect with our current triangulation.
* Then we add a random tetrahedron to our triangulation and repeat.
*/
rand_bound = rand() % tmp_triang.bound_len;
//
//Calculate the conform tetrahedrons above and below
if (!facet_conform(tmp_triang.bound_tri + rand_bound, ¶meters))
break; //Triangle on the boundary that does not have a conform facet
tet_list_len = parameters.acute_ind_len;
//Form explicit list of the tetrahedrons
for (i = 0; i < tet_list_len; i++)
{
copyArr3(tet_list[i].vertices[0], tmp_triang.bound_tri[rand_bound].vertices[0]);
copyArr3(tet_list[i].vertices[1], tmp_triang.bound_tri[rand_bound].vertices[1]);
copyArr3(tet_list[i].vertices[2], tmp_triang.bound_tri[rand_bound].vertices[2]);
copyArr3(tet_list[i].vertices[3], cube.points[parameters.acute_ind[i]]);
}
//Remove all the tetrahedrons that intersect with current triangulation.
filter_tet_list_disjoint_triangulation(tet_list, &tet_list_len, &tmp_triang);
if (tet_list_len == 0)
break; //We can not find a conform tetrahedron for this boundary.. Restart
//Select a ttetrahedron from the tet_list to add to the triangulation.. Different approaches.
//Combinations between: random tetra, smallest volume, maximum volume. Indices stored in tet_max, tet_min and tet_rand
tet_list_min_max_volume(tet_list, tet_list_len, &tet_max, &tet_min);
tet_rand = rand() % tet_list_len;
switch (omp_get_thread_num() % 6) {
case 0: //Choose tet with max volume
tet_add = tet_max; break;
case 1: //Choose tet with min volume
tet_add = tet_min; break;
case 2: //Choose random tet
tet_add = tet_rand; break;
case 3: //Choose either max or min (random)
tet_add = (rand() % 2)? tet_min : tet_max; break;
case 4: //Choose either max or rand
tet_add = (rand() % 5)? tet_max : tet_rand; break;
case 5: //Either min or rand
tet_add = (rand() % 5)? tet_min : tet_rand; break;
default:
tet_add = 0;
}
/*
* Add the above tetra to the triangulation.
* This removes all the boundary triangles that are covered by this tetrahedron
*/
add_tet_triangulation(tet_list + tet_add,&tmp_triang);
}
if (triangulation_volume(&tmp_triang) > max_volume) {
max_volume = triangulation_volume(&tmp_triang);
printf("Record for thread %d using method %d amount: %zu\n", omp_get_thread_num(), omp_get_thread_num() % 6, max_volume);
triangulation_print(&tmp_triang);
}
if (tmp_triang.bound_len == 0)
{
printf("FOUND A TRIANGULATION!!!\n");
triangulation_print(&tmp_triang);
omp_set_lock(&result_lock);
result = tmp_triang;
triangulation_found = 1;
omp_unset_lock(&result_lock);
} else
triangulation_free(&tmp_triang);
}
free(parameters.acute_ind);
free(tet_list);
}
free(cube.points);
omp_destroy_lock(&result_lock);
return result;
}
int main(int argc, char **argv){
FILE *file = fopen("file1","r");
FILE *out = NULL;
char str_buf[1024][50];
unsigned str_buf_in = 0;
unsigned str_buf_out = 0;
char str[50];
int read_finish = 0;
int num_read = 0, num_write = 0;
char **input_filenames = NULL;
int input_len; //num of input files
FILE **input_files = NULL;
int i,j; double elapsed_time;
int mapping_done = 0;//done when all mapper thread done
struct timeval tvalBefore, tvalAfter;
////locks///
int rank, size, len;
char name[MPI_MAX_PROCESSOR_NAME];
omp_set_num_threads(4);
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Get_processor_name(name, &len);
MPI_Status status;
omp_init_lock(&worklock);
omp_init_lock(&inclock);
omp_init_lock(&readlock);
omp_init_lock(&readerlock);
omp_init_lock(&mapperlock);
if(argc < 5){
printf("Usage ./mapreduce -in [input files].... -out [output file]\n");
return 0;
}else{
if(strcmp("-in",argv[1])){
printf("Usage ./mapreduce -in [input files].... -out [output file]\n");
return 0;
}
for(i=2;i<argc;i++){ //start from first input file
if(!strcmp("-out",argv[i])){
break;
}
}
input_len = i - 2;
input_filenames = (char**)malloc(sizeof(char*)*input_len);
for(j=0;j<input_len;j++)
input_filenames[j] = (char*)malloc(sizeof(char)*50);
for(i=2,j=0;j<input_len;i++,j++){
strcpy(input_filenames[j],argv[i]);
}
input_files = read_in(input_filenames,input_len,0);
if(strcmp("-out",argv[2+input_len])){
printf("output file missing, using default name 'out'\n");
out = fopen("out","w");
}else{
out = fopen(argv[3+input_len],"w");
}
}
omp_set_num_threads(8);
fifoQ *queue_to_map = initQ(1000000, "queue_to_map");
fifoQ *queue_to_reduce = initQ(1000000, "queue_to_map");
fifoQ **queues_to_map = (fifoQ**)malloc(sizeof(fifoQ*)*5);
queues_to_map[0] = initQ(1000000, "queue_to_map0");
queues_to_map[1] = initQ(1000000, "queue_to_map1");
queues_to_map[2] = initQ(1000000, "queue_to_map2");
queues_to_map[3] = initQ(1000000, "queue_to_map3");
queues_to_map[4] = initQ(1000000, "queue_to_map4");
fifoQ **queues_to_reduce = (fifoQ**)malloc(sizeof(fifoQ*)*5);
queues_to_reduce[0] = initQ(1000000, "queue_to_reduce0");
queues_to_reduce[1] = initQ(1000000, "queue_to_reduce1");
queues_to_reduce[2] = initQ(1000000, "queue_to_reduce2");
queues_to_reduce[3] = initQ(1000000, "queue_to_reduce3");
queues_to_reduce[4] = initQ(1000000, "queue_to_reduce4");
fifoQ **queues_reduced = (fifoQ**)malloc(sizeof(fifoQ*)*5);
fifoQ *final_queue = initQ(1000000, "final Q");
int sendsize = input_len/size + (input_len % size - rank > 0 ? 1 : 0); //num of files send to a node
if(rank==0){ //distribute files
int i,j;
char ***files_tosend = (char***)malloc(sizeof(char**)*input_len);
int lsendsize;
FILE **node_files;
for(i=0;i<size;i++){
lsendsize = input_len/size + (input_len % size - i > 0 ? 1 : 0); //num of files send to a node
printf("send size of core %d is %d\n",i,lsendsize);
files_tosend[i] = (char**)malloc(sizeof(char*)*lsendsize);
for(j=0;j<lsendsize;j++){
files_tosend[i][j] = (char*)malloc(sizeof(char)*50);
}
}
for(i=0;i<input_len;i++){
int belongs_to = i % size;
int pos = i/size;
strcpy(files_tosend[belongs_to][pos],input_filenames[i]);
printf("distributing file %s to files_tosend %d,%d, value %s\n",input_filenames[i],belongs_to,pos,files_tosend[belongs_to][pos]);
}
if(size>1){
for(i=1;i<size;i++){
lsendsize = input_len/size + (input_len % size - i > 0 ? 1 : 0);
for(j=0;j<lsendsize;j++){
printf("sending %s to cpu %d\n",files_tosend[i][j],i);
MPI_Send(files_tosend[i][j],50,MPI_BYTE,i,1,MPI_COMM_WORLD);
printf("send done\n");
}
}
}
node_files = (FILE**)malloc(sizeof(FILE*)*sendsize);
for(i=0;i<sendsize;i++){
node_files[i] = fopen(files_tosend[rank][i],"r");
}
gettimeofday (&tvalBefore, NULL);
#pragma omp parallel sections
{
#pragma omp section //reader thread0
{
int i; int odd_even = 0;
//printf("reader 0 is core #%d\n",rank);
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[0], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread0 done\n");
}
#pragma omp section //reader thread1
{
int i; int odd_even = 0;
//printf("reader 1 is core #%d\n",rank);
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[1], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread1 done\n");
}
#pragma omp section //reader thread2
{
int i; int odd_even = 0;
//printf("reader 2 is core #%d\n",rank);
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[2], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread2 done\n");
}
#pragma omp section //reader thread3
{
// printf("reader 3 is core #%d\n",rank);
int i; int odd_even = 0;
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[3], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread3 done %d\n",rank);
}
#pragma omp section //mapper thread 0
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 0");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[0])){
printf("");
if(!is_empty(queues_to_map[0])){
work work = getWork(queues_to_map[0]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread0 done %d\n",rank);
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d in map 0\n",elapsed_time,rank);
}
#pragma omp section //mapper thread 1
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 1");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[1])){
printf("");
if(!is_empty(queues_to_map[1])){
work work = getWork(queues_to_map[1]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread1 done %d\n",rank);
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d in map 1\n",elapsed_time,rank);
}
#pragma omp section //mapper thread 2
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 2");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[2])){
printf("");
if(!is_empty(queues_to_map[2])){
work work = getWork(queues_to_map[2]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread2 done %d\n",rank);
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d in map 2\n",elapsed_time,rank);
}
#pragma omp section //mapper thread 3
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 2");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[3])){
printf("");
if(!is_empty(queues_to_map[3])){
work work = getWork(queues_to_map[3]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread3 done %d\n",rank);
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d in map 3\n",elapsed_time,rank);
}
#pragma omp section //reducer thread 0
{
int i;
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d\n",elapsed_time,rank);
while(mapping_done<NUM_READ_THREADS){
printf("");
}
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d\n",elapsed_time,rank);
queues_reduced[0] = reducer(queues_to_reduce[0]);
//printf("reducer thread 0 done\n");
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d\n",elapsed_time,rank);
}
#pragma omp section //reducer thread 1
{
int i;
while(mapping_done<NUM_READ_THREADS){printf("");}
queues_reduced[1] = reducer(queues_to_reduce[1]);
//printf("reducer thread 1 done\n");
}
#pragma omp section //reducer thread 2
{
int i;
while(mapping_done<NUM_READ_THREADS){printf("");}
queues_reduced[2] = reducer(queues_to_reduce[2]);
//printf("reducer thread 2 done\n");
}
#pragma omp section //reducer thread 3
{
int i;
while(mapping_done<NUM_READ_THREADS){printf("");}
queues_reduced[3] = reducer(queues_to_reduce[3]);
//printf("reducer thread 3 done\n");
}
}
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d\n",elapsed_time,rank);
}
else{
int i;
FILE** node_files = (FILE**)malloc(sizeof(FILE*)*sendsize);
for(i=0;i<sendsize;i++){
char *bufstr = (char*)malloc(sizeof(char)*50);
MPI_Recv(bufstr,50,MPI_BYTE, 0,1, MPI_COMM_WORLD, &status);
//printf("%s received\n",bufstr);
node_files[i] = fopen(bufstr,"r");
}
#pragma omp parallel sections shared(input_files) private(str)
{
//printf("using %d threads in core %d\n",omp_get_num_threads(),rank);
#pragma omp section //reader thread0
{
int i; int odd_even = 0;
// printf("reader 0 is core #%d\n",rank);
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[0], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread0 done\n");
}
#pragma omp section //reader thread1
{
int i; int odd_even = 0;
// printf("reader 1 is core #%d\n",rank);
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[1], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread1 done\n");
}
#pragma omp section //reader thread2
{
int i; int odd_even = 0;
//printf("reader 2 is core #%d\n",rank);
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[2], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread2 done\n");
}
#pragma omp section //reader thread3
{
//printf("reader 3 is core #%d\n",rank);
int i; int odd_even = 0;
for(i=0;i<sendsize;i++){
while(!feof(node_files[i])){
/////////check if full///////////
omp_set_lock(&readerlock);
if(!feof(node_files[i])){
strcpy(str,"");
fscanf(node_files[i],"%s",str);
}
else{
omp_unset_lock(&readerlock);
break;
}
omp_unset_lock(&readerlock);
if(strcmp(str,""))
putWork(queues_to_map[3], constr_work(str));
}
}
omp_set_lock(&inclock);
read_finish++;
omp_unset_lock(&inclock);
//printf("reader thread3 done %d\n",rank);
}
#pragma omp section //mapper thread 0
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 0");
//printf("map1\n");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[0])){
printf("");
if(!is_empty(queues_to_map[0])){
work work = getWork(queues_to_map[0]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread0 done %d\n",rank);
}
#pragma omp section //mapper thread 1
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 1");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[1])){
printf("");
if(!is_empty(queues_to_map[1])){
work work = getWork(queues_to_map[1]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread1 done %d\n",rank);
}
#pragma omp section //mapper thread 2
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 2");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[2])){
printf("");
if(!is_empty(queues_to_map[2])){
work work = getWork(queues_to_map[2]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread2 done %d\n",rank);
}
#pragma omp section //mapper thread 3
{
int i;
fifoQ *innerQ = initQ(50000,"innerQ 2");
while(read_finish<NUM_READ_THREADS || !is_empty(queues_to_map[3])){
printf("");
if(!is_empty(queues_to_map[3])){
work work = getWork(queues_to_map[3]);
//mapper(queues_to_reduce[hash(work.str)], work);
mapper(innerQ, work);
}
}
for(i=0;i<=innerQ->in;i++){
work work = getWork(innerQ);
putWork(queues_to_reduce[hash(work.str)],work);
}
omp_set_lock(&inclock);
mapping_done++;
omp_unset_lock(&inclock);
//printf("mapper thread3 done %d\n",rank);
}
#pragma omp section //reducer thread 0
{
int i;
while(mapping_done<NUM_READ_THREADS){
printf("");
}
queues_reduced[0] = reducer(queues_to_reduce[0]);
//printf("reducer thread 0 done\n");
}
#pragma omp section //reducer thread 1
{
int i;
while(mapping_done<NUM_READ_THREADS){printf("");}
queues_reduced[1] = reducer(queues_to_reduce[1]);
//printf("reducer thread 1 done\n");
}
#pragma omp section //reducer thread 2
{
int i;
while(mapping_done<NUM_READ_THREADS){printf("");}
queues_reduced[2] = reducer(queues_to_reduce[2]);
//printf("reducer thread 2 done\n");
}
#pragma omp section //reducer thread 3
{
int i;
while(mapping_done<NUM_READ_THREADS){printf("");}
queues_reduced[3] = reducer(queues_to_reduce[3]);
//printf("reducer thread 3 done\n");
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
gettimeofday (&tvalAfter, NULL);
elapsed_time = (float)(tvalAfter.tv_sec - tvalBefore.tv_sec)+((float)(tvalAfter.tv_usec - tvalBefore.tv_usec)/1000000);
if(rank==0)
printf("elapsed time = %.2f sec,rank %d\n",elapsed_time,rank);
if(rank==0){ //final reducuction
int i,j,revbuf;int mainct;
for(i=0;i<NUM_READ_THREADS;i++){
combine_queue(final_queue,queues_reduced[i]);
}
//printf("main node has %d to final reduce\n",calcnum(queues_reduced,NUM_READ_THREADS));
for(i=1;i<size;i++){
MPI_Recv(&revbuf,1,MPI_INT,i,1,MPI_COMM_WORLD,&status);
//printf("need to receive %d strings from node %d\n",revbuf,i);
char *strbuf = (char*)malloc(sizeof(char)*50);
char ctbuf = 0;
for(j=0;j<revbuf;j++){
MPI_Recv(strbuf,50,MPI_BYTE,i,1,MPI_COMM_WORLD,&status);
MPI_Recv(&ctbuf,50,MPI_INT,i,1,MPI_COMM_WORLD,&status);
work work;
strcpy(work.str,strbuf);
work.count = ctbuf;
//printf("received <%s,%d> from node %d\n",work.str,work.count,i);
putWork(final_queue,work);
}
}
fifoQ *output = reducer(final_queue);
printQ_to_file(&output,1,out);
}else{
int i,total_num;
total_num = calcnum(queues_reduced,NUM_READ_THREADS);
MPI_Send(&total_num,1,MPI_INT,0,1,MPI_COMM_WORLD);
for(i=0;i<NUM_READ_THREADS;i++){
combine_queue(final_queue,queues_reduced[i]);
}
for(i=0;i<total_num;i++){
MPI_Send(&final_queue->works[i].str,50,MPI_BYTE,0,1,MPI_COMM_WORLD);
MPI_Send(&final_queue->works[i].count,1,MPI_INT,0,1,MPI_COMM_WORLD);
}
}
for(i=0;i<input_len;i++){
fclose(input_files[i]);
}
fclose(out);
/*printQ(queues_to_map[0]);
printQ(queues_to_map[1]);
printQ(queues_to_map[2]);
printQ(queues_to_map[3]);*/
/*printQ(queues_reduced[0]);
printQ(queues_reduced[1]);
printQ(queues_reduced[2]);
printQ(queues_reduced[3]);*/
omp_destroy_lock(&inclock);
omp_destroy_lock(&worklock);
omp_destroy_lock(&readlock);
omp_destroy_lock(&readerlock);
omp_destroy_lock(&mapperlock);
MPI_Finalize();
return 0;
}
double betweennessCentrality(graph* G, DOUBLE_T* BC, int filter) {
VERT_T *S; /* stack of vertices in the order of non-decreasing
distance from s. Also used to implicitly
represent the BFS queue */
plist* P; /* predecessors of a vertex v on shortest paths from s */
DOUBLE_T* sig; /* No. of shortest paths */
LONG_T* d; /* Length of the shortest path between every pair */
DOUBLE_T* del; /* dependency of vertices */
LONG_T *in_degree, *numEdges, *pSums;
LONG_T *pListMem;
LONG_T* Srcs;
LONG_T *start, *end;
LONG_T MAX_NUM_PHASES;
LONG_T *psCount;
#ifdef _OPENMP
omp_lock_t* vLock;
LONG_T chunkSize;
#endif
int seed = 2387;
double elapsed_time;
#ifdef _OPENMP
#pragma omp parallel
{
#endif
VERT_T *myS, *myS_t;
LONG_T myS_size;
LONG_T i, j, k, p, count, myCount;
LONG_T v, w, vert;
LONG_T numV, num_traversals, n, m, phase_num;
LONG_T tid, nthreads;
int* stream;
#ifdef DIAGNOSTIC
double elapsed_time_part;
#endif
#ifdef _OPENMP
int myLock;
tid = omp_get_thread_num();
nthreads = omp_get_num_threads();
#else
tid = 0;
nthreads = 1;
#endif
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds();
}
#endif
/* numV: no. of vertices to run BFS from = 2^K4approx */
numV = 1<<K4approx;
n = G->n;
m = G->m;
/* Permute vertices */
if (tid == 0) {
Srcs = (LONG_T *) malloc(n*sizeof(LONG_T));
#ifdef _OPENMP
vLock = (omp_lock_t *) malloc(n*sizeof(omp_lock_t));
#endif
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
for (i=0; i<n; i++) {
omp_init_lock(&vLock[i]);
}
#endif
/* Initialize RNG stream */
stream = init_sprng(0, tid, nthreads, seed, SPRNG_DEFAULT);
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
Srcs[i] = i;
}
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
j = n*sprng(stream);
if (i != j) {
#ifdef _OPENMP
int l1 = omp_test_lock(&vLock[i]);
if (l1) {
int l2 = omp_test_lock(&vLock[j]);
if (l2) {
#endif
k = Srcs[i];
Srcs[i] = Srcs[j];
Srcs[j] = k;
#ifdef _OPENMP
omp_unset_lock(&vLock[j]);
}
omp_unset_lock(&vLock[i]);
}
#endif
}
}
#ifdef _OPENMP
#pragma omp barrier
#endif
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() -elapsed_time_part;
fprintf(stderr, "Vertex ID permutation time: %lf seconds\n", elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
/* Start timing code from here */
if (tid == 0) {
elapsed_time = get_seconds();
#ifdef VERIFYK4
MAX_NUM_PHASES = 2*sqrt(n);
#else
MAX_NUM_PHASES = 50;
#endif
}
#ifdef _OPENMP
#pragma omp barrier
#endif
/* Initialize predecessor lists */
/* The size of the predecessor list of each vertex is bounded by
its in-degree. So we first compute the in-degree of every
vertex */
if (tid == 0) {
P = (plist *) calloc(n, sizeof(plist));
in_degree = (LONG_T *) calloc(n+1, sizeof(LONG_T));
numEdges = (LONG_T *) malloc((n+1)*sizeof(LONG_T));
pSums = (LONG_T *) malloc(nthreads*sizeof(LONG_T));
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
#endif
for (i=0; i<m; i++) {
v = G->endV[i];
#ifdef _OPENMP
omp_set_lock(&vLock[v]);
#endif
in_degree[v]++;
#ifdef _OPENMP
omp_unset_lock(&vLock[v]);
#endif
}
prefix_sums(in_degree, numEdges, pSums, n);
if (tid == 0) {
pListMem = (LONG_T *) malloc(m*sizeof(LONG_T));
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
#endif
for (i=0; i<n; i++) {
P[i].list = pListMem + numEdges[i];
P[i].degree = in_degree[i];
P[i].count = 0;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "In-degree computation time: %lf seconds\n", elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
/* Allocate shared memory */
if (tid == 0) {
free(in_degree);
free(numEdges);
free(pSums);
S = (VERT_T *) malloc(n*sizeof(VERT_T));
sig = (DOUBLE_T *) malloc(n*sizeof(DOUBLE_T));
d = (LONG_T *) malloc(n*sizeof(LONG_T));
del = (DOUBLE_T *) calloc(n, sizeof(DOUBLE_T));
start = (LONG_T *) malloc(MAX_NUM_PHASES*sizeof(LONG_T));
end = (LONG_T *) malloc(MAX_NUM_PHASES*sizeof(LONG_T));
psCount = (LONG_T *) malloc((nthreads+1)*sizeof(LONG_T));
}
/* local memory for each thread */
myS_size = (2*n)/nthreads;
myS = (LONG_T *) malloc(myS_size*sizeof(LONG_T));
num_traversals = 0;
myCount = 0;
#ifdef _OPENMP
#pragma omp barrier
#endif
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
d[i] = -1;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() -elapsed_time_part;
fprintf(stderr, "BC initialization time: %lf seconds\n", elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
for (p=0; p<n; p++) {
i = Srcs[p];
//printf ("%d \n", i);
// i = p;
if (G->numEdges[i+1] - G->numEdges[i] == 0) {
continue;
} else {
num_traversals++;
}
if (num_traversals == numV + 1) {
break;
}
if (tid == 0) {
sig[i] = 1;
d[i] = 0;
S[0] = i;
start[0] = 0;
end[0] = 1;
}
count = 1;
phase_num = 0;
#ifdef _OPENMP
#pragma omp barrier
#endif
while (end[phase_num] - start[phase_num] > 0) {
myCount = 0;
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for schedule(dynamic)
#endif
for (vert = start[phase_num]; vert < end[phase_num]; vert++) {
v = S[vert];
for (j=G->numEdges[v]; j<G->numEdges[v+1]; j++) {
if ((G->weight[j] & 7) == 0 && filter==1) continue;
w = G->endV[j];
if (v != w) {
#ifdef _OPENMP
myLock = omp_test_lock(&vLock[w]);
if (myLock) {
#endif
/* w found for the first time? */
if (d[w] == -1) {
if (myS_size == myCount) {
/* Resize myS */
myS_t = (LONG_T *)
malloc(2*myS_size*sizeof(VERT_T));
memcpy(myS_t, myS, myS_size*sizeof(VERT_T));
free(myS);
myS = myS_t;
myS_size = 2*myS_size;
}
myS[myCount++] = w;
d[w] = d[v] + 1;
sig[w] = sig[v];
P[w].list[P[w].count++] = v;
} else if (d[w] == d[v] + 1) {
sig[w] += sig[v];
P[w].list[P[w].count++] = v;
}
#ifdef _OPENMP
omp_unset_lock(&vLock[w]);
} else {
if ((d[w] == -1) || (d[w] == d[v]+ 1)) {
omp_set_lock(&vLock[w]);
sig[w] += sig[v];
P[w].list[P[w].count++] = v;
omp_unset_lock(&vLock[w]);
}
}
#endif
}
}
}
/* Merge all local stacks for next iteration */
phase_num++;
psCount[tid+1] = myCount;
#ifdef _OPENMP
#pragma omp barrier
#endif
if (tid == 0) {
start[phase_num] = end[phase_num-1];
psCount[0] = start[phase_num];
for(k=1; k<=nthreads; k++) {
psCount[k] = psCount[k-1] + psCount[k];
}
end[phase_num] = psCount[nthreads];
}
#ifdef _OPENMP
#pragma omp barrier
#endif
for (k = psCount[tid]; k < psCount[tid+1]; k++) {
S[k] = myS[k-psCount[tid]];
}
#ifdef _OPENMP
#pragma omp barrier
#endif
count = end[phase_num];
}
phase_num--;
#ifdef _OPENMP
#pragma omp barrier
#endif
//printf ("%d\n", phase_num);
while (phase_num > 0) {
#ifdef _OPENMP
#pragma omp for
#endif
for (j=start[phase_num]; j<end[phase_num]; j++) {
w = S[j];
for (k = 0; k<P[w].count; k++) {
v = P[w].list[k];
#ifdef _OPENMP
omp_set_lock(&vLock[v]);
#endif
del[v] = del[v] + sig[v]*(1+del[w])/sig[w];
#ifdef _OPENMP
omp_unset_lock(&vLock[v]);
#endif
}
BC[w] += del[w];
}
phase_num--;
#ifdef _OPENMP
#pragma omp barrier
#endif
}
#ifdef _OPENMP
chunkSize = n/nthreads;
#pragma omp for schedule(static, chunkSize)
#endif
for (j=0; j<count; j++) {
w = S[j];
//fprintf (stderr, "w: %d\n", w);
d[w] = -1;
del[w] = 0;
P[w].count = 0;
}
#ifdef _OPENMP
#pragma omp barrier
#endif
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() -elapsed_time_part;
fprintf(stderr, "BC computation time: %lf seconds\n", elapsed_time_part);
}
#endif
#ifdef _OPENMP
#pragma omp for
for (i=0; i<n; i++) {
omp_destroy_lock(&vLock[i]);
}
#endif
free(myS);
if (tid == 0) {
free(S);
free(pListMem);
free(P);
free(sig);
free(d);
free(del);
#ifdef _OPENMP
free(vLock);
#endif
free(start);
free(end);
free(psCount);
elapsed_time = get_seconds() - elapsed_time;
free(Srcs);
}
free_sprng(stream);
#ifdef _OPENMP
}
#endif
/* Verification */
#ifdef VERIFYK4
double BCval;
if (SCALE % 2 == 0) {
BCval = 0.5*pow(2, 3*SCALE/2)-pow(2, SCALE)+1.0;
} else {
BCval = 0.75*pow(2, (3*SCALE-1)/2)-pow(2, SCALE)+1.0;
}
int failed = 0;
for (int i=0; i<G->n; i++) {
if (round(BC[i] - BCval) != 0) {
failed = 1;
break;
}
}
if (failed) {
fprintf(stderr, "Kernel 4 failed validation!\n");
} else {
fprintf(stderr, "Kernel 4 validation successful!\n");
}
#endif
for (int i = 0; i < G->n; i++) printf ("BC: %d %f\n",i, BC[i]);
return elapsed_time;
}
static void unLockNode(IDnum preNodeID)
{
omp_unset_lock(nodeLocks + preNodeID);
}
void vertex_betweenness_centrality_simple(graph_t* G, double* BC, long numSrcs) {
attr_id_t *in_degree, *numEdges, *pSums;
#if RANDSRCS
attr_id_t* Srcs;
#endif
long num_traversals = 0;
#ifdef _OPENMP
omp_lock_t* vLock;
long chunkSize;
#endif
#ifdef DIAGNOSTIC
double elapsed_time;
#endif
int seed = 2387;
/* The outer loop is parallelized in this case. Each thread does a BFS
and the vertex BC values are incremented atomically */
#ifdef _OPENMP
#pragma omp parallel firstprivate(G)
{
#endif
attr_id_t *S; /* stack of vertices in the order of non-decreasing
distance from s. Also used to implicitly
represent the BFS queue */
plist_t* P; /* predecessors of a vertex v on shortest paths
from s */
attr_id_t* pListMem;
double* sig; /* No. of shortest paths */
attr_id_t* d; /* Length of the shortest path between every pair */
double* del; /* dependency of vertices */
attr_id_t *start, *end;
long MAX_NUM_PHASES;
long i, j, k, p, count;
long v, w, vert;
long numV, n, m, phase_num;
long tid, nthreads;
int* stream;
#ifdef DIAGNOSTIC
double elapsed_time_part;
#endif
#ifdef _OPENMP
int myLock;
tid = omp_get_thread_num();
nthreads = omp_get_num_threads();
#else
tid = 0;
nthreads = 1;
#endif
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time = get_seconds();
elapsed_time_part = get_seconds();
}
#endif
/* numV: no. of vertices to run BFS from = numSrcs */
numV = numSrcs;
n = G->n;
m = G->m;
/* Permute vertices */
if (tid == 0) {
#if RANDSRCS
Srcs = (attr_id_t *) malloc(n*sizeof(attr_id_t));
#endif
#ifdef _OPENMP
vLock = (omp_lock_t *) malloc(n*sizeof(omp_lock_t));
#endif
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
for (i=0; i<n; i++) {
omp_init_lock(&vLock[i]);
}
#endif
/* Initialize RNG stream */
stream = init_sprng(0, tid, nthreads, seed, SPRNG_DEFAULT);
#if RANDSRCS
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
Srcs[i] = i;
}
#ifdef _OPENMP
#pragma omp for
#endif
for (i=0; i<n; i++) {
j = n * sprng(stream);
if (i != j) {
#ifdef _OPENMP
int l1 = omp_test_lock(&vLock[i]);
if (l1) {
int l2 = omp_test_lock(&vLock[j]);
if (l2) {
#endif
k = Srcs[i];
Srcs[i] = Srcs[j];
Srcs[j] = k;
#ifdef _OPENMP
omp_unset_lock(&vLock[j]);
}
omp_unset_lock(&vLock[i]);
}
#endif
}
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#endif
MAX_NUM_PHASES = 50;
/* Initialize predecessor lists */
/* The size of the predecessor list of each vertex is bounded by
its in-degree. So we first compute the in-degree of every
vertex */
if (tid == 0) {
in_degree = (attr_id_t *) calloc(n+1, sizeof(attr_id_t));
numEdges = (attr_id_t *) malloc((n+1)*sizeof(attr_id_t));
pSums = (attr_id_t *) malloc(nthreads*sizeof(attr_id_t));
}
#ifdef _OPENMP
#pragma omp barrier
#pragma omp for
#endif
for (i=0; i<m; i++) {
v = G->endV[i];
#ifdef _OPENMP
omp_set_lock(&vLock[v]);
#endif
in_degree[v]++;
#ifdef _OPENMP
omp_unset_lock(&vLock[v]);
#endif
}
prefix_sums(in_degree, numEdges, pSums, n);
P = (plist_t *) calloc(n, sizeof(plist_t));
pListMem = (attr_id_t *) malloc(m*sizeof(attr_id_t));
for (i=0; i<n; i++) {
P[i].list = pListMem + numEdges[i];
P[i].degree = in_degree[i];
P[i].count = 0;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() -elapsed_time_part;
fprintf(stderr, "In-degree computation time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#endif
/* Allocate shared memory */
if (tid == 0) {
free(in_degree);
free(numEdges);
free(pSums);
}
S = (attr_id_t *) malloc(n*sizeof(attr_id_t));
sig = (double *) malloc(n*sizeof(double));
d = (attr_id_t *) malloc(n*sizeof(attr_id_t));
del = (double *) calloc(n, sizeof(double));
start = (attr_id_t *) malloc(MAX_NUM_PHASES*sizeof(attr_id_t));
end = (attr_id_t *) malloc(MAX_NUM_PHASES*sizeof(attr_id_t));
#ifdef _OPENMP
#pragma omp barrier
#endif
for (i=0; i<n; i++) {
d[i] = -1;
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "BC initialization time: %lf seconds\n",
elapsed_time_part);
elapsed_time_part = get_seconds();
}
#endif
#ifdef _OPENMP
#pragma omp for reduction(+:num_traversals)
#endif
for (p=0; p<numV; p++) {
#if RANDSRCS
i = Srcs[p];
#else
i = p;
#endif
if (G->numEdges[i+1] - G->numEdges[i] == 0) {
continue;
} else {
num_traversals++;
}
sig[i] = 1;
d[i] = 0;
S[0] = i;
start[0] = 0;
end[0] = 1;
count = 1;
phase_num = 0;
while (end[phase_num] - start[phase_num] > 0) {
for (vert = start[phase_num]; vert < end[phase_num]; vert++) {
v = S[vert];
for (j=G->numEdges[v]; j<G->numEdges[v+1]; j++) {
w = G->endV[j];
if (v != w) {
/* w found for the first time? */
if (d[w] == -1) {
S[count++] = w;
d[w] = d[v] + 1;
sig[w] = sig[v];
P[w].list[P[w].count++] = v;
} else if (d[w] == d[v] + 1) {
sig[w] += sig[v];
P[w].list[P[w].count++] = v;
}
}
}
}
phase_num++;
start[phase_num] = end[phase_num-1];
end[phase_num] = count;
}
phase_num--;
while (phase_num > 0) {
for (j=start[phase_num]; j<end[phase_num]; j++) {
w = S[j];
for (k = 0; k<P[w].count; k++) {
v = P[w].list[k];
del[v] = del[v] + sig[v]*(1+del[w])/sig[w];
}
#ifdef _OPENMP
omp_set_lock(&vLock[w]);
BC[w] += del[w];
omp_unset_lock(&vLock[w]);
#else
BC[w] += del[w];
#endif
}
phase_num--;
}
for (j=0; j<count; j++) {
w = S[j];
d[w] = -1;
del[w] = 0;
P[w].count = 0;
}
}
#ifdef DIAGNOSTIC
if (tid == 0) {
elapsed_time_part = get_seconds() - elapsed_time_part;
fprintf(stderr, "BC computation time: %lf seconds\n",
elapsed_time_part);
}
#endif
#ifdef _OPENMP
#pragma omp barrier
#endif
#ifdef _OPENMP
#pragma omp for
for (i=0; i<n; i++) {
omp_destroy_lock(&vLock[i]);
}
#endif
free(S);
free(pListMem);
free(P);
free(sig);
free(d);
free(del);
free(start);
free(end);
if (tid == 0) {
#ifdef _OPENMP
free(vLock);
#endif
#if RANDSRCS
free(Srcs);
#endif
#ifdef DIAGNOSTIC
elapsed_time = get_seconds() - elapsed_time;
fprintf(stderr, "Total time taken: %lf seconds\n", elapsed_time);
#endif
}
free_sprng(stream);
#ifdef _OPENMP
#pragma omp barrier
}
#endif
}
/*
* Computes clusters' centroids.
*/
static void compute_centroids(void)
{
int i, j; /* Loop indexes. */
int population; /* Centroid population. */
start = timer_get();
memcpy(lcentroids, CENTROID(rank*(ncentroids/nprocs)), lncentroids[rank]*dimension*sizeof(float));
memset(&has_changed[rank*NUM_THREADS], 0, NUM_THREADS*sizeof(int));
memset(centroids, 0, (ncentroids + DELTA*nprocs)*dimension*sizeof(float));
memset(ppopulation, 0, (ncentroids + nprocs*DELTA)*sizeof(int));
/* Compute partial centroids. */
#pragma omp parallel for schedule(static) default(shared) private(i, j)
for (i = 0; i < lnpoints; i++)
{
j = map[i]%NUM_THREADS;
omp_set_lock(&lock[j]);
vector_add(CENTROID(map[i]), POINT(i));
ppopulation[map[i]]++;
omp_unset_lock(&lock[j]);
}
end = timer_get();
total += timer_diff(start, end);
sync_pcentroids();
sync_ppopulation();
start = timer_get();
/* Compute centroids. */
#pragma omp parallel for schedule(static) default(shared) private(i, j, population)
for (j = 0; j < lncentroids[rank]; j++)
{
population = 0;
for (i = 0; i < nprocs; i++)
{
if (*POPULATION(i, j) == 0)
continue;
population += *POPULATION(i, j);
if (i == rank)
continue;
vector_add(PCENTROID(rank, j), PCENTROID(i, j));
}
if (population > 1)
vector_mult(PCENTROID(rank, j), 1.0/population);
/* Cluster mean has changed. */
if (!vector_equal(PCENTROID(rank, j), LCENTROID(j)))
{
has_changed[rank*NUM_THREADS + omp_get_thread_num()] = 1;
vector_assign(LCENTROID(j), PCENTROID(rank, j));
}
}
end = timer_get();
total += timer_diff(start, end);
sync_centroids();
sync_status();
}
void OpenMPCounter::reset()
{
omp_set_lock(&_lock);
_counter = 0;
omp_unset_lock(&_lock);
}
void YsOpenMPMutex::Unlock(void)
{
omp_unset_lock(&lock);
}
void Unlock() { omp_unset_lock(&lock); }
void IptTracer::genLightPaths(omp_lock_t& cmdLock , vector<Path*>& lightPathList , bool isFirstIter)
{
#pragma omp parallel for
for(int p=0; p<lightPathNum; p++)
{
if (!renderer->scene.usingGPU())
{
Ray lightRay = genEmissiveSurfaceSample(true , false);
lightPathList[p] = new Path;
samplePath(*lightPathList[p] , lightRay);
}
Path& lightPath = *lightPathList[p];
if (lightPath.size() <= 1)
continue;
IptPathState lightState;
lightState.originRay = &lightPath[0];
Real cosAtLight = lightPath[0].getCosineTerm();
lightState.throughput = lightPath[0].color * cosAtLight /
(lightPath[0].originProb * lightPath[0].directionProb *
lightPath[1].originProb);
lightState.indirContrib = vec3f(0.0);
lightState.mergedPath = 1;
/*
fprintf(fp , "====================\n");
vec3f decay = lightPath[0].getRadianceDecay((lightPath[0].origin - lightPath[1].origin).length());
fprintf(fp , "l = 0 , thr = (%.8f,%.8f,%.8f) , color = (%.8f,%.8f,%.8f)\ncosine = %.8f , dirPdf = %.8f , oPdf = %.8f\ndecay=(%.8f,%.8f,%.8f)\n" ,
lightState.throughput.x , lightState.throughput.y , lightState.throughput.z ,
lightPath[0].color[0] , lightPath[0].color[1] , lightPath[0].color[2] ,
lightPath[0].getCosineTerm() , lightPath[0].directionProb , lightPath[1].originProb ,
decay.x , decay.y , decay.z);
*/
int nonSpecPathLength = 0;
for(unsigned i = 1; i < lightPath.size(); i++)
//for (unsigned i = 1; i < 2; i++)
{
Real dist = std::max((lightPath[i].origin - lightPath[i - 1].origin).length() , 1e-5f);
vec3f decayFactor = lightPath[i - 1].getRadianceDecay(dist);
lightState.throughput *= decayFactor;
if(lightPath[i].contactObject && lightPath[i].contactObject->emissive())
break;
lightState.pos = lightPath[i].origin;
lightState.lastRay = &lightPath[i - 1];
lightState.ray = &lightPath[i];
lightState.pathLen = i;
if(lightPath[i].directionSampleType == Ray::RANDOM &&
(lightPath[i].insideObject != NULL || lightPath[i].contactObject != NULL) &&
(lightPath[i].origin != lightPath[i - 1].origin))
{
//if (lightPath[i].insideObject && !lightPath[i].contactObject)
// fprintf(fp , "path length = %d, dirContrib = (%.8f,%.8f,%.8f)\n" ,
// i , lightState.dirContrib[0] , lightState.dirContrib[1] , lightState.dirContrib[2]);
omp_set_lock(&cmdLock);
partialSubPathList.push_back(lightState);
omp_unset_lock(&cmdLock);
}
if (i == lightPath.size() - 1)
break;
if (lightPath[i].direction.length() < 0.5f)
break;
vec3f scatterFactor = (lightPath[i].color * lightPath[i].getCosineTerm() /
(lightPath[i + 1].originProb * lightPath[i].directionProb));
lightState.throughput *= scatterFactor;
/*
vec3f decay = lightPath[i].getRadianceDecay((lightPath[i].origin - lightPath[i + 1].origin).length());
fprintf(fp , "l = %d , thr = (%.8f,%.8f,%.8f) , color = (%.8f,%.8f,%.8f)\ncosine = %.8f , dirPdf = %.8f , oPdf = %.8f\ndecay=(%.8f,%.8f,%.8f)\n" ,
i , lightState.throughput.x , lightState.throughput.y , lightState.throughput.z ,
lightPath[i].color[0] , lightPath[i].color[1] , lightPath[i].color[2] ,
lightPath[i].getCosineTerm() , lightPath[i].directionProb , lightPath[i + 1].originProb ,
decay.x , decay.y , decay.z);
*/
if (lightPath[i].directionSampleType == Ray::RANDOM && useWeight)
{
Real pdf = lightPath[i].directionProb;
if (pdf < 1e-7f)
break;
Real weightFactor;
Real volMergeScale = 1;
Real originProb;
Real dirProb;
if (lightPath[i].contactObject)
{
if (isFirstIter || useUniformSur)
originProb = 1.f / totArea;
else
originProb = lightPath[i].contactObject->getOriginProb(lightPath[i].contactObjectTriangleID);
if (useUniformDir)
dirProb = INV_2_PI;
else
dirProb = lightPath[i].getCosineTerm() / M_PI;
}
//if (lightPath[i].insideObject && lightPath[i].contactObject)
// printf("!!!\n");
if (lightPath[i].insideObject && !lightPath[i].contactObject && lightPath[i].insideObject->isVolumetric())
{
volMergeScale = 4.f / 3.f * mergeRadius;
if (isFirstIter || useUniformVol)
originProb = 1.f / totVol;
else
originProb = lightPath[i].insideObject->getOriginProb(lightPath[i].origin);
dirProb = 0.25f / M_PI;
}
weightFactor = connectFactor(pdf) /
(connectFactor(pdf) + mergeFactor(&volMergeScale , &originProb , &dirProb , &lightPathNum));
if (_isnan(weightFactor) || abs(pdf) < 1e-6f)
{
fprintf(err , "sample light path error, %.8f , %.8f\n" , connectFactor(pdf) ,
mergeFactor(&volMergeScale , &originProb , &dirProb , &lightPathNum));
}
/*
if (abs(volMergeScale - 1.f) < 1e-6)
printf("surface %.8f\n" , weightFactor);
else
printf("volume %.8f %.8f\n" , weightFactor);
*/
//if (lightPath[i].contactObject && lightPath[i].contactObject->objectIndex == 7)
lightState.throughput *= weightFactor;
}
}
}
lightPhotonNum = partialPhotonNum = partialSubPathList.size();
}
inline void unsetLock() { omp_unset_lock(&lock_); }
void IptTracer::genIntermediatePaths(omp_lock_t& cmdLock , vector<Path*>& interPathList)
{
#pragma omp parallel for
for(int p=0; p<interPathNum; p++)
{
if (!renderer->scene.usingGPU())
{
Ray interRay = genIntermediateSamples(renderer->scene);
interPathList[p] = new Path;
samplePath(*interPathList[p] , interRay);
}
Path& interPath = *interPathList[p];
//fprintf(fp , "=================\n");
partPathMergeIndex[p].clear();
if (interPath.size() <= 1)
continue;
IptPathState interState;
interState.originRay = &interPath[0];
interState.throughput = interPath[0].color * interPath[0].getCosineTerm() /
(interPath[0].originProb * interPath[0].directionProb * interPath[1].originProb);
interState.indirContrib = vec3f(0.f);
interState.mergedPath = 0;
//if (intensity(interState.throughput) > 30.f)
// continue;
/*
fprintf(fp , "====================\n");
vec3f decay = interPath[0].getRadianceDecay((interPath[0].origin - interPath[1].origin).length());
fprintf(fp , "l = 0 , thr = (%.8f,%.8f,%.8f) , color = (%.8f,%.8f,%.8f)\ncosine = %.8f , dirPdf = %.8f , oPdf = %.8f\ndecay=(%.8f,%.8f,%.8f)\n" ,
interState.throughput.x , interState.throughput.y , interState.throughput.z ,
interPath[0].color[0] , interPath[0].color[1] , interPath[0].color[2] ,
interPath[0].getCosineTerm() , interPath[0].directionProb , interPath[1].originProb ,
decay.x , decay.y , decay.z);
*/
for(unsigned i = 1; i < interPath.size(); i++)
//for (unsigned i = 1; i < 2; i++)
{
Real dist = std::max((interPath[i].origin - interPath[i - 1].origin).length() , 1e-5f);
interState.throughput *= interPath[i - 1].getRadianceDecay(dist);
if(interPath[i].contactObject && interPath[i].contactObject->emissive())
break;
interState.pos = interPath[i].origin;
interState.lastRay = &interPath[i - 1];
interState.ray = &interPath[i];
interState.pathLen = i;
if(interPath[i].directionSampleType != Ray::DEFINITE &&
(interPath[i].insideObject != NULL || interPath[i].contactObject != NULL) &&
(interPath[i].origin != interPath[i - 1].origin))
//(interPath[i].insideObject && !interPath[i].contactObject)) // only volume
{
//fprintf(fp , "path length = %d, dirContrib = (%.8f,%.8f,%.8f)\n" ,
// i , interState.dirContrib[0] , interState.dirContrib[1] , interState.dirContrib[2]);
omp_set_lock(&cmdLock);
partialSubPathList.push_back(interState);
partPathMergeIndex[p].push_back(partialSubPathList.size() - 1);
omp_unset_lock(&cmdLock);
}
if (i == interPath.size() - 1)
break;
if (interPath[i].direction.length() < 0.5f)
break;
vec3f scatterFactor = (interPath[i].color * interPath[i].getCosineTerm() /
(interPath[i + 1].originProb * interPath[i].directionProb));
interState.throughput *= scatterFactor;
/*
vec3f decay = interPath[i].getRadianceDecay((interPath[i].origin - interPath[i + 1].origin).length());
fprintf(fp , "l = %d , thr = (%.8f,%.8f,%.8f) , color = (%.8f,%.8f,%.8f)\ncosine = %.8f , dirPdf = %.8f , oPdf = %.8f\ndecay=(%.8f,%.8f,%.8f)\n" ,
i , interState.throughput.x , interState.throughput.y , interState.throughput.z ,
interPath[i].color[0] , interPath[i].color[1] , interPath[i].color[2] ,
interPath[i].getCosineTerm() , interPath[i].directionProb , interPath[i + 1].originProb ,
decay.x , decay.y , decay.z);
*/
if (interPath[i].directionSampleType != Ray::DEFINITE && useWeight)
{
Real pdf = interPath[i].directionProb;
if (pdf < 1e-7f)
break;
Real weightFactor;
Real volMergeScale = 1.f;
Real originProb;
Real dirProb;
if (interPath[i].contactObject)
{
if (useUniformSur)
originProb = 1.f / totArea;
else
originProb = interPath[i].contactObject->getOriginProb(interPath[i].contactObjectTriangleID);
if (useUniformDir)
dirProb = INV_2_PI;
else
dirProb = interPath[i].getCosineTerm() / M_PI;
}
//if (interPath[i].insideObject && interPath[i].contactObject)
// printf("!!!\n");
if (interPath[i].insideObject && !interPath[i].contactObject && interPath[i].insideObject->isVolumetric())
{
volMergeScale = 4.f / 3.f * mergeRadius;
if (useUniformVol)
originProb = 1.f / totVol;
else
originProb = interPath[i].insideObject->getOriginProb(interPath[i].origin);
dirProb = 0.25f / M_PI;
}
weightFactor = connectFactor(pdf) /
(connectFactor(pdf) + mergeFactor(&volMergeScale , &originProb , &dirProb , &partialPathNum));
if (_isnan(weightFactor) || abs(pdf) < 1e-6f)
{
fprintf(err , "sample inter path error, %.8f , %.8f\n" , connectFactor(pdf) ,
mergeFactor(&volMergeScale , &originProb , &dirProb , &partialPathNum));
}
//if (interPath[i].contactObject && interPath[i].contactObject->objectIndex == 7)
interState.throughput *= weightFactor;
}
}
}
partialPhotonNum = partialSubPathList.size();
}
void Shape::splitshapes(vector<Shape*> &shapes, ViewProgress *progress)
{
int n_tr = (int)triangles.size();
if (progress) progress->start(_("Split Shapes"), n_tr);
int progress_steps = max(1,(int)(n_tr/100));
vector<bool> done(n_tr);
bool cont = true;
// make list of adjacent triangles for each triangle
vector< vector<uint> > adj(n_tr);
if (progress) progress->set_label(_("Split: Sorting Triangles ..."));
#ifdef _OPENMP
omp_lock_t progress_lock;
omp_init_lock(&progress_lock);
#pragma omp parallel for schedule(dynamic)
#endif
for (int i = 0; i < n_tr; i++) {
if (progress && i%progress_steps==0) {
#ifdef _OPENMP
omp_set_lock(&progress_lock);
#endif
cont = progress->update(i);
#ifdef _OPENMP
omp_unset_lock(&progress_lock);
#endif
}
vector<uint> trv;
for (int j = 0; j < n_tr; j++) {
if (i!=j) {
bool add = false;
if (j<i) // maybe(!) we have it already
for (uint k = 0; k<adj[j].size(); k++) {
if ((int)adj[j][k] == i) {
add = true; break;
}
}
add |= (triangles[i].isConnectedTo(triangles[j], 0.01));
if (add) trv.push_back(j);
}
}
adj[i] = trv;
if (!cont) i=n_tr;
}
if (progress) progress->set_label(_("Split: Building shapes ..."));
// triangle indices of shapes
vector< vector<uint> > shape_tri;
for (int i = 0; i < n_tr; i++) done[i] = false;
for (int i = 0; i < n_tr; i++) {
if (progress && i%progress_steps==0)
cont = progress->update(i);
if (!done[i]){
cerr << _("Shape ") << shapes.size()+1 << endl;
vector<uint> current;
addtoshape(i, adj, current, done);
Shape *shape = new Shape();
shapes.push_back(shape);
shapes.back()->triangles.resize(current.size());
for (uint i = 0; i < current.size(); i++)
shapes.back()->triangles[i] = triangles[current[i]];
shapes.back()->CalcBBox();
}
if (!cont) i=n_tr;
}
if (progress) progress->stop("_(Done)");
}
vector<vec3f> PhotonMap::renderPixels(const Camera& camera){
uint width = camera.width, height = camera.height;
std::vector<vec3f> pixelColors(width * height, vec3f(0,0,0));
omp_init_lock(&surfaceHashGridLock);
omp_init_lock(&volumeHashGridLock);
omp_init_lock(&debugPrintLock);
//std::vector<int> pixelMaps(pixelColors.size(), 0);
preprocessEmissionSampler();
mRadius = mBaseRadius;
clock_t startTime = clock();
for(uint s = 0; s < spp; s++){
std::cout << "iteration : " << s << std::endl;
std::vector<vec3f> oneIterColors(pixelColors.size(), vec3f(0,0,0));
#ifdef PPM
//if (renderer->scene.getTotalVolume() > 1e-6f)
if (true)
{
rayMarching = true;
mRadius = MAX(mBaseRadius * powf(powf(s+1 , mAlpha-1) , 1.f / 3.f) , EPSILON);
}
else
{
rayMarching = false;
mRadius = MAX(mBaseRadius * sqrt(powf(s+1, mAlpha-1)), EPSILON);
}
#endif
std::vector<Path*> pixelLightPaths(mPhotonsNum, NULL);
std::vector<LightPoint> surfaceLightVertices(0);
std::vector<LightPoint> volumeLightVertices(0);
surfaceHashGrid.Reserve(pixelColors.size());
volumeHashGrid.Reserve(pixelColors.size());
#pragma omp parallel for
// step1: sample light paths and build range search struct independently for surface and volume
for(int p = 0; p < mPhotonsNum; p++){
Ray lightRay = genEmissiveSurfaceSample(true , false);
pixelLightPaths[p] = new Path;
Path &lightPath = *pixelLightPaths[p];
samplePath(lightPath, lightRay);
for(int i = 1; i < lightPath.size(); i++){
// light is not reflective
if(lightPath[i].contactObject && lightPath[i].contactObject->emissive())
break;
// only store particles non-specular
if(lightPath[i].directionSampleType == Ray::DEFINITE)
continue;
LightPoint lightPoint;
lightPoint.position = lightPath[i].origin;
lightPoint.indexInThePath = i;
lightPoint.pathThePointIn = &lightPath;
lightPoint.photonType = lightPath[i].photonType;
if(lightPoint.photonType == Ray::OUTVOL){
omp_set_lock(&surfaceHashGridLock);
surfaceLightVertices.push_back(lightPoint);
omp_unset_lock(&surfaceHashGridLock);
}
if(lightPoint.photonType == Ray::INVOL){
omp_set_lock(&volumeHashGridLock);
volumeLightVertices.push_back(lightPoint);
omp_unset_lock(&volumeHashGridLock);
}
}
}
std::cout<< "vol vertices= " << volumeLightVertices.size() << " sur vertices= " << surfaceLightVertices.size() << std::endl;
surfaceHashGrid.Build(surfaceLightVertices, mRadius);
volumeHashGrid.Build(volumeLightVertices, mRadius);
std::cout<< "finish building hashgrid" << std::endl;
// step2: calculate pixel colors by progressive photon mapping
#pragma omp parallel for
for(int p = 0; p < pixelColors.size(); p++){
Path eyePath;
if (rayMarching)
sampleMergePath(eyePath, camera.generateRay(p), 0);
else
samplePath(eyePath, camera.generateRay(p));
//fprintf(fp , "===================\n");
//for (int i = 0; i < eyePath.size(); i++)
//{
// fprintf(fp , "l=%d, bsdf=(%.8f,%.8f,%.8f), originPdf=%.8f, dirPdf=%.8f\n" , i , eyePath[i].color.x ,
// eyePath[i].color.y , eyePath[i].color.z , eyePath[i].originProb , eyePath[i].directionProb);
//}
/*if(eyePath[1].contactObj && eyePath[1].contactObj->anisotropic()){
pixelMaps[p] = 1;
}*/
throughputByDensityEstimation(oneIterColors[p], eyePath, surfaceLightVertices, volumeLightVertices);
}
/*std::ofstream fout(engine->renderer->name + engine->scene.name+"pixelMap.txt");
for(int p = 0; p < pixelMaps.size(); p++)
fout << pixelMaps[p] << ' ' ;
fout << std::endl;
fout.close();*/
std::cout << "calculation done" << std::endl;
for(uint i = 0; i < pixelColors.size(); i++){
pixelColors[i] *= s / float(s+1);
pixelColors[i] += camera.eliminateVignetting(oneIterColors[i], i) / (s + 1);
delete pixelLightPaths[i];
}
unsigned nowTime = (float)(clock() - startTime) / 1000;
//if (nowTime > recordTime)
if (s % outputIter == 0)
{
showCurrentResult(pixelColors , &nowTime , &s);
//showCurrentResult(pixelColors , &lastTime , &s);
//recordTime += timeInterval;
}
else
showCurrentResult(pixelColors);
}
return pixelColors;
}