vector<CVector> CellInitHelper::generateInitCellNodes() {
bool isSuccess = false;
vector<CVector> attemptedPoss;
while (!isSuccess) {
attemptedPoss = tryGenInitCellNodes();
if (isPosQualify(attemptedPoss)) {
isSuccess = true;
}
}
// also need to make sure center point is (0,0,0).
CVector tmpSum(0, 0, 0);
for (uint i = 0; i < attemptedPoss.size(); i++) {
tmpSum = tmpSum + attemptedPoss[i];
}
tmpSum = tmpSum / (double) (attemptedPoss.size());
for (uint i = 0; i < attemptedPoss.size(); i++) {
attemptedPoss[i] = attemptedPoss[i] - tmpSum;
}
return attemptedPoss;
}
void Foam::GAMGAgglomeration::createTarget
(
const labelgpuList& list,
const labelgpuList& sort,
labelgpuList& target,
labelgpuList& targetStart
)
{
labelgpuList ones(list.size(),1);
labelgpuList tmpTarget(list.size());
labelgpuList tmpSum(list.size());
labelgpuList listSort(list.size());
thrust::copy
(
thrust::make_permutation_iterator
(
list.begin(),
sort.begin()
),
thrust::make_permutation_iterator
(
list.begin(),
sort.end()
),
listSort.begin()
);
target = listSort;
label targetSize =
thrust::unique
(
target.begin(),
target.end()
)
- target.begin();
target.setSize(targetSize);
targetStart.setSize(list.size());
thrust::reduce_by_key
(
listSort.begin(),
listSort.end(),
ones.begin(),
tmpTarget.begin(),
tmpSum.begin()
);
thrust::exclusive_scan
(
tmpSum.begin(),
tmpSum.end(),
targetStart.begin()
);
targetStart.setSize(targetSize+1);
targetStart.set(targetSize,list.size());
}
void Foam::lduAddressing::calcLosortStart() const
{
if (losortStartPtr_)
{
FatalErrorIn("lduAddressing::calcLosortStart() const")
<< "losort start already calculated"
<< abort(FatalError);
}
const labelgpuList& nbr = upperAddr();
const labelgpuList& lsrt = losortAddr();
losortStartPtr_ = new labelgpuList(size() + 1, nbr.size());
labelgpuList& lsrtStart = *losortStartPtr_;
labelgpuList ones(nbr.size()+size(),1);
labelgpuList tmpCell(size());
labelgpuList tmpSum(size());
labelgpuList nbrSort(nbr.size()+size());
thrust::copy
(
thrust::make_permutation_iterator
(
nbr.begin(),
lsrt.begin()
),
thrust::make_permutation_iterator
(
nbr.begin(),
lsrt.end()
),
nbrSort.begin()
);
thrust::copy
(
thrust::make_counting_iterator(0),
thrust::make_counting_iterator(0)+size(),
nbrSort.begin()+nbr.size()
);
thrust::fill
(
ones.begin()+nbr.size(),
ones.end(),
0
);
thrust::stable_sort_by_key
(
nbrSort.begin(),
nbrSort.end(),
ones.begin()
);
thrust::reduce_by_key
(
nbrSort.begin(),
nbrSort.end(),
ones.begin(),
tmpCell.begin(),
tmpSum.begin()
);
thrust::exclusive_scan
(
tmpSum.begin(),
tmpSum.begin()+size(),
lsrtStart.begin()
);
}
void Foam::lduAddressing::calcOwnerStart() const
{
if (ownerStartPtr_)
{
FatalErrorIn("lduAddressing::calcOwnerStart() const")
<< "owner start already calculated"
<< abort(FatalError);
}
const labelgpuList& own = lowerAddr();
ownerStartPtr_ = new labelgpuList(size() + 1, own.size());
labelgpuList& ownStart = *ownerStartPtr_;
labelgpuList ones(own.size()+size(),1);
labelgpuList tmpCell(size());
labelgpuList tmpSum(size());
labelgpuList ownSort(own.size()+size());
thrust::copy
(
own.begin(),
own.end(),
ownSort.begin()
);
thrust::copy
(
thrust::make_counting_iterator(0),
thrust::make_counting_iterator(0)+size(),
ownSort.begin()+own.size()
);
thrust::fill
(
ones.begin()+own.size(),
ones.end(),
0
);
thrust::stable_sort_by_key
(
ownSort.begin(),
ownSort.end(),
ones.begin()
);
thrust::reduce_by_key
(
ownSort.begin(),
ownSort.end(),
ones.begin(),
tmpCell.begin(),
tmpSum.begin()
);
thrust::exclusive_scan
(
tmpSum.begin(),
tmpSum.end(),
ownStart.begin()
);
}
void Foam::lduAddressing::calcPatchSortStart() const
{
if (patchSortStartAddr_.size() == nPatches())
{
FatalErrorIn("lduAddressing::calcLosortStart() const")
<< "losort start already calculated"
<< abort(FatalError);
}
patchSortStartAddr_.setSize(nPatches());
for(label i = 0; i < nPatches(); i++)
{
if( ! patchAvailable(i))
continue;
const labelgpuList& nbr = patchAddr(i);
const labelgpuList& lsrt = patchSortAddr(i);
labelgpuList* patchSortStartPtr_ = new labelgpuList(nbr.size() + 1, nbr.size());
patchSortStartAddr_.set(i,patchSortStartPtr_);
labelgpuList& lsrtStart = *patchSortStartPtr_;
labelgpuList ones(nbr.size(),1);
labelgpuList tmpCell(nbr.size());
labelgpuList tmpSum(nbr.size());
labelgpuList nbrSort(nbr.size());
thrust::copy
(
thrust::make_permutation_iterator
(
nbr.begin(),
lsrt.begin()
),
thrust::make_permutation_iterator
(
nbr.begin(),
lsrt.end()
),
nbrSort.begin()
);
thrust::reduce_by_key
(
nbrSort.begin(),
nbrSort.end(),
ones.begin(),
tmpCell.begin(),
tmpSum.begin()
);
thrust::exclusive_scan
(
tmpSum.begin(),
tmpSum.end(),
lsrtStart.begin()
);
}
}
Slice * PithExtractor::convolution(const Slice &slice, arma::fcolvec verticalFilter, arma::frowvec horizontalFilter) const {
const uint height = slice.n_rows;
const uint width = slice.n_cols;
Slice *resultat = new Slice(height, width);
arma::fmat tmpSum;
arma::fvec sum = arma::fvec(width);
resultat->zeros();
int kOffset;
int kCenter;
int lag;
int endIndex;
//convolve horizontal direction
//find center position of kernel
kCenter = horizontalFilter.n_cols/2;
endIndex = width-kCenter;
tmpSum = arma::fmat(height, width);
tmpSum.zeros();
for (uint i=0; i<height; i++) { //height pictures (nb rows)
kOffset = 0;
//index=0 to index=kCenter-1
for (int j=0; j<kCenter; j++) {
for (int k=kCenter+kOffset, m=0; k>=0; k--, m++) {
tmpSum(i,j) += slice(i,m) * horizontalFilter[k];
}
kOffset++;
}
//index=kCenter to index=(width pictures)-kCenter-1)
for (int j=kCenter; j<endIndex; j++) {
lag = j-kCenter;
for (int k=horizontalFilter.n_cols-1, m=0; k >= 0; k--, m++) {
tmpSum(i,j) += slice(i,(lag+m)) * horizontalFilter[k];
}
}
//index=(width pictures)-kCenter to index=(width pictures)-1
kOffset = 1;
for (uint j=endIndex; j<width; j++) {
lag = j-kCenter;
for (int k=horizontalFilter.n_cols-1, m=0; k >= kOffset; k--, m++) {
tmpSum(i,j) += slice(i,lag+m) * horizontalFilter[k];
}
kOffset++;
}
}
//convolve vertical direction
//find center position of kernel
kCenter = verticalFilter.n_rows/2;
endIndex = width-kCenter;
sum.zeros();
kOffset = 0;
//index=0 to index=(kCenter-1)
for (int i=0; i <kCenter; i++) {
lag = -i;
for (int k = kCenter + kOffset; k>=0; k--) { //convolve with partial kernel
for (uint j=0; j<width; j++) {
sum[j] += tmpSum(i+lag,j) * verticalFilter[k];
}
lag++;
}
for (uint n=0; n<width; n++) { //convert to output format
if(sum[n]>=0)
(*resultat)(i,n) = (int)(sum[n] + 0.5f);
else
(*resultat)(i,n) = (int)(sum[n] - 0.5f);
sum[n] = 0; //reset before next summing
}
kOffset++;
}
//index=kCenter to index=(height pictures)-kCenter-1
for (int i=kCenter; i<endIndex; i++) {
lag = -kCenter;
for (int k=verticalFilter.n_rows-1; k>=0; k--) { //convolve with full kernel
for(uint j=0; j<width; j++){
sum[j] += tmpSum(i+lag,j) * verticalFilter[k];
}
lag++;
}
for (uint n=0; n<width; n++) { //convert to output format
if(sum[n]>=0)
(*resultat)(i,n) = (int)(sum[n] + 0.5f);
else
(*resultat)(i,n) = (int)(sum[n] - 0.5f);
sum[n] = 0; //reset before next summing
}
}
//index=(height pictures)-kcenter to index=(height pictures)-1
kOffset = 1;
for (uint i=endIndex; i<height; i++) {
lag = -kCenter;
for (int k=verticalFilter.n_rows-1; k>=kOffset; k--) { //convolve with partial kernel
for (uint j=0; j<width; j++) { //height p corrigeictures
sum[j] += tmpSum(i+lag,j) * verticalFilter[k];
}
lag++;
}
for (uint n=0; n<width; n++) { //convert to output format
if(sum[n]>=0)
(*resultat)(i,n) = (int)(sum[n] + 0.5f);
else
(*resultat)(i,n) = (int)(sum[n] - 0.5f);
sum[n] = 0; //reset before next summing
}
kOffset++;
}
return resultat;
}