void indri::collection::Repository::_merge(index_state& state) {
// this is only legal if we're not readOnly
if (_readOnly) return;
size_t memoryBound = (size_t) (0.75 * _memory);
std::vector<indri::index::Index*>* result = new std::vector<indri::index::Index*>;
if (state->size() <= 2 ||
(_mergeMemory(*state) < memoryBound &&
_mergeFiles(*state) < MERGE_FILE_LIMIT)) {
indri::index::Index* index = _mergeStage(state);
result->push_back(index);
} else {
// divide and conquer
index_state first = new std::vector<indri::index::Index*>;
index_state second = new std::vector<indri::index::Index*>;
first->assign(state->begin(), state->begin() + state->size() / 2);
second->assign(state->begin() + state->size() / 2, state->end());
// release the previous state object
state = 0;
_merge(second);
_merge(first);
std::copy(first->begin(), first->end(), std::back_inserter(*result));
std::copy(second->begin(), second->end(), std::back_inserter(*result));
}
state = result;
}
// a step
void epoch() {
// this->_pop.clear();
// _pareto_front.clear();
_selection (_parent_pop, _child_pop);
parallel::p_for(parallel::range_t(0, _child_pop.size()),
mutate<crowd::Indiv<Phen> >(_child_pop));
#ifndef EA_EVAL_ALL
_eval_subpop(_child_pop);
_merge(_parent_pop, _child_pop, _mixed_pop);
#else
_merge(_parent_pop, _child_pop, _mixed_pop);
_eval_subpop(_mixed_pop);
#endif
_apply_modifier(_mixed_pop);
#ifndef NDEBUG
BOOST_FOREACH(indiv_t& ind, _mixed_pop)
for (size_t i = 0; i < ind->fit().objs().size(); ++i) {
assert(!std::isnan(ind->fit().objs()[i]));
}
#endif
_fill_nondominated_sort(_mixed_pop, _parent_pop);
_mixed_pop.clear();
_child_pop.clear();
_convert_pop(_parent_pop, this->_pop);
assert(_parent_pop.size() == Params::pop::size);
assert(_pareto_front.size() <= Params::pop::size * 2);
assert(_mixed_pop.size() == 0);
// assert(_child_pop.size() == 0);
assert(this->_pop.size() == Params::pop::size);
}
int main (int argc, char *argv[]) {
graph_t gin;
graph_t gout;
args_t args;
startup("cmerge", argc, argv, NULL, NULL);
memset(&args, 0, sizeof(args_t));
_parse_opt(argc, argv, &args);
if (ngdb_read(args.input, &gin)) {
printf("Could not read in %s\n", args.input);
goto fail;
}
printf("merging...\n");
if (_merge(&gin, &gout, args.mergesets, args.nmergesets)) {
printf("Could not perform merge\n");
goto fail;
}
if (ngdb_write(&gout, args.output)) {
printf("Could not write to %s\n", args.output);
goto fail;
}
return 0;
fail:
return 1;
}
void _split(T A[], T temp[], int begin, int end)
{
if ((end - begin) < 2)
return;
// recursively split into halves until size is 1,
// then merge and go back up.
int mid = (end + begin) / 2;
// I chose 16 because it's a power of 2 and precedence.
if ((end - mid) <= 16)
{
insertion_sort(A, end, mid); // mid included.
insertion_sort(A, mid, begin); // mid excluded.
}
// Split if large enough.
else
{
_split(A, temp, begin, mid); // mid will be excluded.
_split(A, temp, mid, end); // mid will be included.
}
_merge(A, temp, begin, mid, end); // merge them back again.
_copy (A, temp, begin, end); // Copy sorted array to original array.
}
/* [begin, end)区间
* 归并排序
*/
void _sort(ListNode *begin, ListNode *end)
{
//处理0~1个节点
if (begin == NULL || begin->next == end)
{
return ;
}
//处理两个节点，交换数据
else if (begin->next->next == end)
{
if ( begin->val > begin->next->val)
{
int t = begin->val;
begin->val = begin->next->val;
begin->next->val = t;
}
return ;
}
// 寻找中间节点
ListNode * p1 = begin, *p2 = begin->next;
while(p2 != end)
{
p1 = p1->next;
p2 = p2->next;
if(p2 != end)
{
p2 = p2->next;
}
}
//排序
_sort(begin, p1);
_sort(p1, end);
_merge(begin, p1, end);
}
static SdbListIter * _merge_sort(SdbListIter *head, SdbListComparator cmp) {
SdbListIter *second;
if (!head || !head->n) {
return head;
}
second = _sdb_list_split (head);
head = _merge_sort (head, cmp);
second = _merge_sort (second, cmp);
return _merge (head, second, cmp);
}
void _mergesort(std::vector<T>& arr, int64_t begin, int64_t end,
std::vector<T>& scratch) {
// trivially sorted since it is 1
if (end - begin <= 1)
return;
int64_t mid = (begin + end) / 2;
_mergesort(arr, begin, mid, scratch);
_mergesort(arr, mid, end, scratch);
_merge(arr, begin, mid, end, scratch);
_copy(scratch, begin, end, arr);
}
node<V>* _cut(node<V>* heap,node<V>* n) {
if(n->next==n) {
n->parent->child=NULL;
} else {
n->next->prev=n->prev;
n->prev->next=n->next;
n->parent->child=n->next;
}
n->next=n->prev=n;
n->marked=false;
return _merge(heap,n);
}
int _MergeSort(int li, int ls) {
if(li==ls) return vec[ li ];
else {
int middle = (li+ls)>>1;
_MergeSort(li, middle);
_MergeSort(middle + 1, ls);
_merge(li, middle, ls);
}
};
/* Substring sort */
void
substringsort(const sauchar_t *T, const saidx_t *PA,
saidx_t *first, saidx_t *last,
saidx_t *buf, saidx_t bufsize,
saidx_t depth, saint_t lastsuffix) {
saidx_t *a, *b;
saidx_t *curbuf;
saidx_t i, j, k;
saidx_t curbufsize;
if(lastsuffix != 0) { ++first; }
for(a = first, i = 0; (a + SS_BLOCKSIZE) < last; a += SS_BLOCKSIZE, ++i) {
_multikey_introsort(T, PA, a, a + SS_BLOCKSIZE, depth);
curbuf = a + SS_BLOCKSIZE;
curbufsize = last - (a + SS_BLOCKSIZE);
if(curbufsize <= bufsize) { curbufsize = bufsize, curbuf = buf; }
for(b = a, k = SS_BLOCKSIZE, j = i; j & 1; b -= k, k <<= 1, j >>= 1) {
_merge(T, PA, b - k, b, b + k, curbuf, curbufsize, depth);
}
}
_multikey_introsort(T, PA, a, last, depth);
for(k = SS_BLOCKSIZE; i != 0; k <<= 1, i >>= 1) {
if(i & 1) {
_merge(T, PA, a - k, a, last, buf, bufsize, depth);
a -= k;
}
}
if(lastsuffix != 0) {
/* Insert last type B* suffix. */
for(a = first, i = *(first - 1);
(a < last) && ((*a < 0) || (0 < _compare(T, PA + i, PA + *a, depth)));
++a) {
*(a - 1) = *a;
}
*(a - 1) = i;
}
}
node<V>* _removeMinimum(node<V>* n) {
_unMarkAndUnParentAll(n->child);
if(n->next==n) {
n=n->child;
} else {
n->next->prev=n->prev;
n->prev->next=n->next;
n=_merge(n->next,n->child);
}
if(n==NULL)return n;
node<V>* trees[64]={NULL};
while(true) {
if(trees[n->degree]!=NULL) {
node<V>* t=trees[n->degree];
if(t==n)break;
trees[n->degree]=NULL;
if(n->value<t->value) {
t->prev->next=t->next;
t->next->prev=t->prev;
_addChild(n,t);
} else {
t->prev->next=t->next;
t->next->prev=t->prev;
if(n->next==n) {
t->next=t->prev=t;
_addChild(t,n);
n=t;
} else {
n->prev->next=t;
n->next->prev=t;
t->next=n->next;
t->prev=n->prev;
_addChild(t,n);
n=t;
}
}
continue;
} else {
trees[n->degree]=n;
}
n=n->next;
}
node<V>* min=n;
do {
if(n->value<min->value)min=n;
n=n->next;
} while(n!=n);
return min;
}
list_t mergesort_l(list_t src, comparator_t cmp)
{
if (!src || !src->cdr)
return src;
_queue_t qs[2] = { {NULL, NULL}
, {NULL, NULL}
};
size_t i;
for (i = 0; src; i++) {
enqueue(&qs[i & 1], shift(&src), NULL);
}
return _merge(mergesort_l(qs[0].head, cmp),
mergesort_l(qs[1].head, cmp), cmp);
}
void _mergesort(int left, int right) {
if(right == left) return;
else {
uint middle = (left + right) >> 1;
_mergesort(left, middle);
_mergesort(middle + 1, right);
_merge(left, middle, right);
}
};
void mergeSort(int *input, int n) {
int *bak = malloc(sizeof(int)*n);
int *handle = input;
int *store = bak;
int width,i;
for(width=1; width<n; width=2*width) {
for(i=0; i<n; i=i+2*width) {
_merge(handle, i, MIN(i+width,n), MIN(i+2*width,n),store);
}
if (handle==input) handle = bak;
else if (handle==bak) handle = input;
if (store==input) store = bak;
else if (store==bak) store = input;
}
free(bak);
}
void merge_sort(int* const a,const int high)
{
const int LOW=0;
const int TMPSIZE=high-LOW;
const int STACKSIZE=log2(TMPSIZE)*2+3;
int sort_count=0;
strct_prms2 stack_prms[STACKSIZE];
std::unique_ptr<int[]> buff(new int[TMPSIZE]);
int* _buff = buff.get();
stack_prms[0].low=LOW;
stack_prms[0].high=high-1;
strct_prms2* _ptr=nullptr;
int _low=-1;
int _high=-1;
int _mid=-1;
while(sort_count>=0)
{
strct_prms2* _ptr=stack_prms+sort_count;
_low=_ptr->low;
_high=_ptr->high;
if((_ptr->check_))
{
_merge(a,_buff,_low,_high);
_ptr->check_=0;
--sort_count;
}
else
{
if(_low<_high)
{
_ptr->check_=1;
_mid=(_low+_high)>>1;
++sort_count;++_ptr;
_ptr->low=_mid+1;
_ptr->high=_high;
++sort_count;++_ptr;
_ptr->low=_low;
_ptr->high=_mid;
}
else
--sort_count;
}
/**
* \brief Merge pairs of nodes and reverse the order.
*
* Given a list of \p src nodes, merge adjacent pairs and reverse the ordering
* of the result. The resulting merged nodes are stored in \p dst. This is the
* first merge pass when deleting the minimum node from the heap.
*
* \param [in] ph Pairing heap containing the nodes to merge.
* \param [out] dst List into which the nodes are merged.
* \param [in] src List from which to merge nodes.
*
* \post \c list_isempty(src)
*
* \note This operation has a time complexity of O(n) with respect to the length
* of \p src.
*/
static void _merge_pairs_reverse(const struct pheap *ph, struct list *dst,
struct list *src)
{
struct pheap_elem *left, *right;
/* Move left to right, merging pairs */
while (!list_isempty(src))
{
left = containerof(list_popfront(src),
struct pheap_elem, child_le);
if (!list_isempty(src))
{
right = containerof(list_popfront(src),
struct pheap_elem, child_le);
left = _merge(ph, left, right);
}
/* Push the merged element onto the front of the destination, so that
* the result is in reverse order.
*/
list_pushfront(dst, &left->child_le);
}
void indri::collection::Repository::_merge() {
// this is only legal if we're not readOnly
if (_readOnly)
return;
// grab a copy of the current state
index_state state = indexes();
index_state mergers = state;
if (state->size() && state->back()->documentCount() == 0) {
// if the current index is empty, don't need to add a new one; write the others
mergers = new index_vector;
mergers->assign(state->begin(), state->end() - 1);
} else {
// current index isn't empty, so add a new one and write the old ones
_addMemoryIndex();
}
// no need to merge when there's only one index (or none)
bool needsWrite = (mergers->size() > 1) ||
(mergers->size() == 1 && dynamic_cast<indri::index::MemoryIndex*>((*mergers)[0]));
if (!needsWrite)
return;
state = 0;
// merge all the indexes together
while (needsWrite) {
_merge(mergers);
needsWrite = (mergers->size() > 1) ||
(mergers->size() == 1 && dynamic_cast<indri::index::MemoryIndex*>((*mergers)[0]));
}
_checkpoint();
}
void indri::collection::Repository::_write() {
// this is only legal if we're not readOnly
if (_readOnly) return;
// grab a copy of the current state
index_state state = indexes();
// if the current index is empty, don't need to write it
if (state->size() && state->back()->documentCount() == 0) return;
// make a new MemoryIndex, cutting off the old one from updates
_addMemoryIndex();
// if we just added the first, no need to write the "old" one
if (state->size() == 0) return;
// write out the last index
index_state lastState = new std::vector<indri::index::Index*>;
lastState->push_back(state->back());
state = 0;
_merge(lastState);
_checkpoint();
}
void _addChild(node<V>* parent,node<V>* child) {
child->prev=child->next=child;
child->parent=parent;
parent->degree++;
parent->child=_merge(parent->child,child);
}
void indri::collection::Repository::_trim() {
// this is only legal if we're not readOnly
if (_readOnly)
return;
// grab a copy of the current state
index_state state = indexes();
if (state->size() <= 3)
return;
size_t count = state->size();
int position;
// here's how this works:
// we're trying to just 'trim' the indexes so that we merge
// together the small indexes, leaving the large ones as they are.
// We merge together a minimum of 3 indexes every time. We may merge
// more, however. We start at the most recent indexes, and search
// backward in time. When we get to an index that is significantly
// larger than the previous index, we stop.
//
// have to merge at least the last three indexes
int firstDocumentCount = (int)(*state)[count-1]->documentCount();
int lastDocumentCount = (int)(*state)[count-3]->documentCount();
int documentCount = 0;
// move back until we find a really big index--don't merge with that one
for(position = count-4; position>=0; position--) {
// compute the average number of documents in the indexes we've seen so far
documentCount = (int)(*state)[position]->documentCount();
// break if we find an index more than 8 times as large
// as the preceding one.
if (documentCount > lastDocumentCount*8.0)
{
position++;
break;
}
lastDocumentCount = documentCount;
}
// make sure position is greater than or equal to 0
position = lemur_compat::max<int>(position, 0);
// make a new MemoryIndex, cutting off the old one from updates
_addMemoryIndex();
// write out the last index
index_state substate = new std::vector<indri::index::Index*>;
substate->assign(state->begin() + position, state->end());
state = 0;
// substate may be larger than 1 if we didn't have enough
// memory to merge everything together. That's okay,
// because we were just trimming.
_merge(substate);
_checkpoint();
}
shared_ptr<Way> WayAverager::average()
{
_sumMovement1 = 0.0;
_sumMovement2 = 0.0;
_moveCount1 = 0;
_moveCount2 = 0;
_maxMovement1 = 0.0;
_maxMovement2 = 0.0;
if (DirectionFinder::isSimilarDirection(_map.shared_from_this(), _w1, _w2) == false)
{
if (_w1->isOneWay() == true)
{
_w2->reverseOrder();
}
else
{
_w1->reverseOrder();
}
}
shared_ptr<const LineString> ls1 = ElementConverter(_map.shared_from_this()).
convertToLineString(_w1);
shared_ptr<const LineString> ls2 = ElementConverter(_map.shared_from_this()).
convertToLineString(_w2);
// All of the fancy stats here are compliments of Mike Porter.
// calculate standard deviation in meters
Meters sd1 = _w1->getCircularError() / 2.0;
Meters sd2 = _w2->getCircularError() / 2.0;
// calculate variance;
double v1 = sd1 * sd1;
double v2 = sd2 * sd2;
// calculate weights for averaging.
double weight1 = 1 - v1 / (v1 + v2);
double weight2 = 1 - v2 / (v1 + v2);
Meters newAcc = 2.0 * sqrt(weight1 * weight1 * v1 + weight2 * weight2 * v2);
shared_ptr<Way> result(new Way(_w1->getStatus(), _map.createNextWayId(),
newAcc));
_map.addWay(result);
result->addNode(_merge(_w1->getNodeIds()[0], weight1, _w2->getNodeIds()[0], weight2));
// we're getting the vectors affter the above merge because the merge will change node ids.
const std::vector<long>& ns1 = _w1->getNodeIds();
const std::vector<long>& ns2 = _w2->getNodeIds();
size_t i1 = 1;
size_t i2 = 1;
// while there is more than one point available in each line
while (i1 < ns1.size() - 1 || i2 < ns2.size() - 1)
{
// if we're all out of cs1 points
if (i1 == ns1.size() - 1)
{
result->addNode(_moveToLine(ns2[i2++], weight2, ls1.get(), weight1, 2));
}
// if we're all out of cs2 points
else if (i2 == ns2.size() - 1)
{
result->addNode(_moveToLine(ns1[i1++], weight1, ls2.get(), weight2, 1));
}
else
{
const Coordinate& last = _map.getNode(result->getLastNodeId())->toCoordinate();
const Coordinate& nc1 = _moveToLineAsCoordinate(ns1[i1], weight1, ls2.get(), weight2);
const Coordinate& nc2 = _moveToLineAsCoordinate(ns2[i2], weight2, ls1.get(), weight1);
if (nc1.distance(last) < nc2.distance(last))
{
result->addNode(_moveToLine(ns1[i1++], weight1, ls2.get(), weight2, 1));
}
else
{
result->addNode(_moveToLine(ns2[i2++], weight2, ls1.get(), weight1, 2));
}
}
}
// merge the last two nodes and move to the average location
result->addNode(_merge(ns1[i1], weight1, ns2[i2], weight2));
// use the default tag merging mechanism
Tags tags = TagMergerFactory::mergeTags(_w1->getTags(), _w2->getTags(), ElementType::Way);
result->setTags(tags);
_map.removeWay(_w1->getId());
_map.removeWay(_w2->getId());
_meanMovement1 = _sumMovement1 / (double)_moveCount1;
_meanMovement2 = _sumMovement2 / (double)_moveCount2;
return result;
}
void merge(FibonacciHeap& other) {
heap=_merge(heap,other.heap);
other.heap=_empty();
}
node<V>* insert(V value) {
node<V>* ret=_singleton(value);
heap=_merge(heap,ret);
return ret;
}
