void HostMatrixCOO<ValueType>::Apply(const BaseVector<ValueType> &in, BaseVector<ValueType> *out) const {
assert(in. get_size() >= 0);
assert(out->get_size() >= 0);
assert(in. get_size() == this->ncol_);
assert(out->get_size() == this->nrow_);
const HostVector<ValueType> *cast_in = dynamic_cast<const HostVector<ValueType>*> (&in) ;
HostVector<ValueType> *cast_out = dynamic_cast< HostVector<ValueType>*> (out) ;
assert(cast_in != NULL);
assert(cast_out!= NULL);
_set_omp_backend_threads(this->local_backend_, this->nnz_);
#pragma omp parallel for
for (int i=0; i<this->nrow_; ++i) {
cast_out->vec_[i] = ValueType(0.0);
if (i == 0) {
int nthreads = omp_get_num_threads();
std::cout << "variable nthreads IN COO is " << nthreads << std::endl;
}
}
for (int i=0; i<this->nnz_; ++i)
cast_out->vec_[this->mat_.row[i] ] += this->mat_.val[i] * cast_in->vec_[ this->mat_.col[i] ];
}
void HostVector<ValueType>::CopyFrom(const BaseVector<ValueType> &vec) {
if (this != &vec) {
if (const HostVector<ValueType> *cast_vec = dynamic_cast<const HostVector<ValueType>*> (&vec)) {
if (this->size_ == 0)
this->Allocate(cast_vec->size_);
assert(cast_vec->size_ == this->size_);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = cast_vec->vec_[i];
} else {
// non-host type
vec.CopyTo(this);
}
}
}
std::complex<double> HostVector<std::complex<double> >::DotNonConj(const BaseVector<std::complex<double> > &x) const {
assert(&x != NULL);
const HostVector<std::complex<double> > *cast_x = dynamic_cast<const HostVector<std::complex<double> >*> (&x);
assert(cast_x != NULL);
assert(this->size_ == cast_x->size_);
double dot_real = double(0.0);
double dot_imag = double(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:dot_real, dot_imag)
for (int i=0; i<this->size_; ++i) {
dot_real += this->vec_[i].real() * cast_x->vec_[i].real() - this->vec_[i].imag() * cast_x->vec_[i].imag();
dot_imag += this->vec_[i].real() * cast_x->vec_[i].imag() + this->vec_[i].imag() * cast_x->vec_[i].real();
}
return std::complex<double>(dot_real, dot_imag);
}
bool HostMatrixCOO<ValueType>::Permute(const BaseVector<int> &permutation) {
assert(&permutation != NULL);
// symmetric permutation only
assert( (permutation.get_size() == this->nrow_) &&
(permutation.get_size() == this->ncol_) );
const HostVector<int> *cast_perm = dynamic_cast<const HostVector<int>*> (&permutation) ;
assert(cast_perm != NULL);
HostMatrixCOO<ValueType> src(this->local_backend_);
src.AllocateCOO(this->nnz_, this->nrow_, this->ncol_);
src.CopyFrom(*this);
_set_omp_backend_threads(this->local_backend_, this->nnz_);
#pragma omp parallel for
for (int i=0; i<this->nnz_; ++i) {
this->mat_.row[i] = cast_perm->vec_[ src.mat_.row[i] ];
this->mat_.col[i] = cast_perm->vec_[ src.mat_.col[i] ];
}
return true;
}
void HostVector<ValueType>::CopyFrom(const BaseVector<ValueType> &src,
const int src_offset,
const int dst_offset,
const int size) {
assert(&src != NULL);
const HostVector<ValueType> *cast_src = dynamic_cast<const HostVector<ValueType>*> (&src);
assert(cast_src != NULL);
//TOOD check always for == this?
assert(&src != this);
assert(this->size_ > 0);
assert(cast_src->size_ > 0);
assert(size > 0);
assert(src_offset + size <= cast_src->size_);
assert(dst_offset + size <= this->size_);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<size; ++i)
this->vec_[i+dst_offset] = cast_src->vec_[i+src_offset];
}
void HostVector<ValueType>::Scale(const ValueType alpha) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] *= alpha;
}
void HostVector<std::complex<float> >::Power(const double power) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = pow(this->vec_[i], std::complex<float>(float(power)));
}
void HostVector<ValueType>::Power(const double power) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = pow(this->vec_[i], ValueType(power));
}
void HostVector<ValueType>::SetValues(const ValueType val) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = val;
}
void HostVector<ValueType>::Ones(void) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = ValueType(1.0);
}
bool HostMatrixCOO<ValueType>::AddScalar(const ValueType alpha) {
_set_omp_backend_threads(this->local_backend_, this->nnz_);
#pragma omp parallel for
for (int i=0; i<this->nnz_; ++i)
this->mat_.val[i] += alpha;
return true;
}
void HostVector<ValueType>::SetRandom(const ValueType a, const ValueType b, const int seed) {
_set_omp_backend_threads(this->local_backend_, this->size_);
// Fill this with random data from interval [a,b]
srand(seed);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = a + (ValueType)rand() / (ValueType)RAND_MAX * (b - a);
}
void HostVector<std::complex<float> >::SetRandom(const std::complex<float> a, const std::complex<float> b,
const int seed) {
_set_omp_backend_threads(this->local_backend_, this->size_);
// Fill this with random data from interval [a,b]
srand(seed);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = a + (float)rand() / (float)RAND_MAX * (b - a);
}
bool HostMatrixCOO<ValueType>::ScaleOffDiagonal(const ValueType alpha) {
_set_omp_backend_threads(this->local_backend_, this->nnz_);
#pragma omp parallel for
for (int i=0; i<this->nnz_; ++i)
if (this->mat_.row[i] != this->mat_.col[i])
this->mat_.val[i] *= alpha;
return true;
}
void HostVector<ValueType>::CopyToData(ValueType *data) const {
if (this->size_ > 0) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
data[i] = this->vec_[i];
}
}
ValueType HostVector<ValueType>::Reduce(void) const {
ValueType reduce = ValueType(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:reduce)
for (int i=0; i<this->size_; ++i)
reduce += this->vec_[i];
return reduce;
}
ValueType HostVector<ValueType>::Asum(void) const {
ValueType asum = ValueType(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:asum)
for (int i=0; i<this->size_; ++i)
asum += paralution_abs(this->vec_[i]);
return asum;
}
ValueType HostVector<ValueType>::Norm(void) const {
ValueType norm2 = ValueType(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:norm2)
for (int i=0; i<this->size_; ++i)
norm2 += this->vec_[i] * this->vec_[i];
return sqrt(norm2);
}
void HostVector<int>::Power(const double power) {
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i) {
int value = 1;
for (int j=0; j<power; ++j)
value *= this->vec_[i];
this->vec_[i] = value;
}
}
std::complex<double> HostVector<std::complex<double> >::Norm(void) const {
double norm2 = double(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:norm2)
for (int i=0; i<this->size_; ++i)
norm2 += this->vec_[i].real() * this->vec_[i].real() + this->vec_[i].imag() * this->vec_[i].imag();
std::complex<double> res(sqrt(norm2), double(0.0));
return res;
}
void HostVector<ValueType>::ScaleAddScale(const ValueType alpha, const BaseVector<ValueType> &x, const ValueType beta) {
assert(&x != NULL);
const HostVector<ValueType> *cast_x = dynamic_cast<const HostVector<ValueType>*> (&x);
assert(cast_x != NULL);
assert(this->size_ == cast_x->size_);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = alpha*this->vec_[i] + beta*cast_x->vec_[i];
}
void HostVector<ValueType>::PointWiseMult(const BaseVector<ValueType> &x) {
assert(&x != NULL);
const HostVector<ValueType> *cast_x = dynamic_cast<const HostVector<ValueType>*> (&x);
assert(cast_x != NULL);
assert(this->size_ == cast_x->size_);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = this->vec_[i]*cast_x->vec_[i];
}
std::complex<double> HostVector<std::complex<double> >::Reduce(void) const {
double reduce_real = double(0.0);
double reduce_imag = double(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:reduce_real, reduce_imag)
for (int i=0; i<this->size_; ++i) {
reduce_real += this->vec_[i].real();
reduce_imag += this->vec_[i].imag();
}
return std::complex<double>(reduce_real, reduce_imag);
}
std::complex<double> HostVector<std::complex<double> >::Asum(void) const {
double asum_real = double(0.0);
double asum_imag = double(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:asum_real, asum_imag)
for (int i=0; i<this->size_; ++i) {
asum_real += paralution_abs(this->vec_[i].real());
asum_imag += paralution_abs(this->vec_[i].imag());
}
return std::complex<double>(asum_real, asum_imag);
}
void HostVector<ValueType>::CopyFromPermuteBackward(const BaseVector<ValueType> &src,
const BaseVector<int> &permutation) {
assert(this != &src);
const HostVector<ValueType> *cast_vec = dynamic_cast<const HostVector<ValueType>*> (&src);
const HostVector<int> *cast_perm = dynamic_cast<const HostVector<int>*> (&permutation);
assert(cast_perm != NULL);
assert(cast_vec != NULL);
assert(cast_vec ->size_ == this->size_);
assert(cast_perm->size_ == this->size_);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i)
this->vec_[i] = cast_vec->vec_[ cast_perm->vec_[i] ];
}
ValueType HostVector<ValueType>::Dot(const BaseVector<ValueType> &x) const {
assert(&x != NULL);
const HostVector<ValueType> *cast_x = dynamic_cast<const HostVector<ValueType>*> (&x);
assert(cast_x != NULL);
assert(this->size_ == cast_x->size_);
ValueType dot = ValueType(0.0);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for reduction(+:dot)
for (int i=0; i<this->size_; ++i)
dot += this->vec_[i]*cast_x->vec_[i];
return dot;
}
void HostMatrixCOO<ValueType>::CopyFrom(const BaseMatrix<ValueType> &mat) {
// copy only in the same format
assert(this->get_mat_format() == mat.get_mat_format());
if (const HostMatrixCOO<ValueType> *cast_mat = dynamic_cast<const HostMatrixCOO<ValueType>*> (&mat)) {
if (this->nnz_ == 0)
this->AllocateCOO(cast_mat->nnz_, cast_mat->nrow_, cast_mat->ncol_ );
assert((this->nnz_ == cast_mat->nnz_) &&
(this->nrow_ == cast_mat->nrow_) &&
(this->ncol_ == cast_mat->ncol_) );
if (this->nnz_ > 0) {
_set_omp_backend_threads(this->local_backend_, this->nnz_);
#pragma omp parallel for
for (int j=0; j<this->nnz_; ++j)
this->mat_.row[j] = cast_mat->mat_.row[j];
#pragma omp parallel for
for (int j=0; j<this->nnz_; ++j)
this->mat_.col[j] = cast_mat->mat_.col[j];
#pragma omp parallel for
for (int j=0; j<this->nnz_; ++j)
this->mat_.val[j] = cast_mat->mat_.val[j];
}
} else {
// Host matrix knows only host matrices
// -> dispatching
mat.CopyTo(this);
}
}
void HostVector<ValueType>::ScaleAddScale(const ValueType alpha, const BaseVector<ValueType> &x, const ValueType beta,
const int src_offset, const int dst_offset,const int size) {
assert(&x != NULL);
const HostVector<ValueType> *cast_x = dynamic_cast<const HostVector<ValueType>*> (&x);
assert(cast_x != NULL);
assert(this->size_ > 0);
assert(cast_x->size_ > 0);
assert(size > 0);
assert(src_offset + size <= cast_x->size_);
assert(dst_offset + size <= this->size_);
_set_omp_backend_threads(this->local_backend_, size);
#pragma omp parallel for
for (int i=0; i<size; ++i)
this->vec_[i+dst_offset] = alpha*this->vec_[i+dst_offset] + beta*cast_x->vec_[i+src_offset];
}
void HostMatrixELL<ValueType>::Apply(const BaseVector<ValueType> &in, BaseVector<ValueType> *out) const {
if (this->nnz_ > 0) {
assert(in. get_size() >= 0);
assert(out->get_size() >= 0);
assert(in. get_size() == this->ncol_);
assert(out->get_size() == this->nrow_);
const HostVector<ValueType> *cast_in = dynamic_cast<const HostVector<ValueType>*> (&in);
HostVector<ValueType> *cast_out = dynamic_cast< HostVector<ValueType>*> (out);
assert(cast_in != NULL);
assert(cast_out!= NULL);
_set_omp_backend_threads(this->local_backend_, this->nrow_);
#pragma omp parallel for
for (int ai=0; ai<this->nrow_; ++ai) {
ValueType sum = ValueType(0.0);
for (int n=0; n<this->mat_.max_row; ++n) {
int aj = ELL_IND(ai, n, this->nrow_, this->mat_.max_row);
int col_aj = this->mat_.col[aj];
if (col_aj >= 0)
sum += this->mat_.val[aj] * cast_in->vec_[col_aj];
else
break;
}
cast_out->vec_[ai] = sum;
}
}
}
void HostVector<ValueType>::PermuteBackward(const BaseVector<int> &permutation) {
assert(&permutation != NULL);
const HostVector<int> *cast_perm = dynamic_cast<const HostVector<int>*> (&permutation);
assert(cast_perm != NULL);
assert(this->size_ == cast_perm->size_);
HostVector<ValueType> vec_tmp(this->local_backend_);
vec_tmp.Allocate(this->size_);
vec_tmp.CopyFrom(*this);
_set_omp_backend_threads(this->local_backend_, this->size_);
#pragma omp parallel for
for (int i=0; i<this->size_; ++i) {
assert_dbg(cast_perm->vec_[i] >= 0);
assert_dbg(cast_perm->vec_[i] < this->size_);
this->vec_[i] = vec_tmp.vec_[ cast_perm->vec_[i] ];
}
}