jiri/LA_library
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

*** empty log message ***

Browse Source
This commit is contained in:
jiri
2021-04-21 13:04:37 +00:00
parent 853008caf1
commit e4937a41f0
18 changed files with 350 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
vec.h
View File

@@ -33,7 +33,7 @@ template <typename T> void lawritemat(FILE *file, const T *a, int r, int c,
/***************************************************************************//**
* static constants used in several cblas-routines
******************************************************************************/
const static complex<double> CONE = 1.0, CMONE = -1.0, CZERO = 0.0;
const static std::complex<double> CONE = 1.0, CMONE = -1.0, CZERO = 0.0;
#ifdef CUDALA
const static cuDoubleComplex CUONE = {1.,0.}, CUMONE = {-1.,0.}, CUZERO = {0.,0.};
#endif
@@ -73,7 +73,7 @@ protected:
public:
friend class NRSMat<T>;
friend class NRMat<T>;
template <typename U> friend NRVec<complex<U> > complexify(const NRVec<U>&);
template <typename U> friend NRVec<std::complex<U> > complexify(const NRVec<U>&);
typedef T ROWTYPE;
@@ -272,6 +272,7 @@ public:
//! get the pointer to the underlying data structure
inline operator T*();
//! get the constant pointer to the underlying data structure
inline operator const T*() const;
@@ -963,14 +964,14 @@ NRVec<T> & NRVec<T>::operator|=(const NRVec<T> &rhs) {
* @see NRVec<T>::copyonwrite()
******************************************************************************/
template<typename T>
NRVec<complex<T> > complexify(const NRVec<T> &rhs) {
NRVec<std::complex<T> > complexify(const NRVec<T> &rhs) {
NOT_GPU(rhs);
NRVec<complex<T> > r(rhs.size(), rhs.getlocation());
NRVec<std::complex<T> > r(rhs.size(), rhs.getlocation());
for(register int i=0; i<rhs.size(); ++i) r[i] = rhs[i];
return r;
}
template<> NRVec<complex<double> > complexify<double>(const NRVec<double> &rhs);
template<> NRVec<std::complex<double> > complexify<double>(const NRVec<double> &rhs);
/***************************************************************************//**
* routine for moving vector data between CPU and GPU memory
@@ -1039,7 +1040,7 @@ inline NRVec<double>& NRVec<double>::operator+=(const double &a) {
* @return reference to the modified vector
******************************************************************************/
template<>
inline NRVec<complex<double> >& NRVec<complex<double> >::operator+=(const complex<double> &a) {
inline NRVec<std::complex<double> >& NRVec<std::complex<double> >::operator+=(const std::complex<double> &a) {
copyonwrite();
#ifdef CUDALA
if(location == cpu){
@@ -1047,7 +1048,7 @@ inline NRVec<complex<double> >& NRVec<complex<double> >::operator+=(const comple
cblas_zaxpy(nn, &CONE, &a, 0, v, 1);
#ifdef CUDALA
}else{
complex<double> *d = gpuputcomplex(a);
std::complex<double> *d = gpuputcomplex(a);
cublasZaxpy(nn, CUONE, (cuDoubleComplex *)d, 0, (cuDoubleComplex *)v, 1);
TEST_CUBLAS("cublasZaxpy");
gpufree(d);
@@ -1087,7 +1088,7 @@ inline NRVec<double>& NRVec<double>::operator-=(const double &a) {
* @return reference to the modified vector
******************************************************************************/
template<>
inline NRVec<complex<double> >& NRVec<complex<double> >::operator-=(const complex<double> &a) {
inline NRVec<std::complex<double> >& NRVec<std::complex<double> >::operator-=(const std::complex<double> &a) {
copyonwrite();
#ifdef CUDALA
if(location == cpu){
@@ -1095,7 +1096,7 @@ inline NRVec<complex<double> >& NRVec<complex<double> >::operator-=(const comple
cblas_zaxpy(nn, &CMONE, &a, 0, v, 1);
#ifdef CUDALA
}else{
complex<double> *d = gpuputcomplex(a);
std::complex<double> *d = gpuputcomplex(a);
cublasZaxpy(nn, CUMONE, (cuDoubleComplex *)d, 0, (cuDoubleComplex *)v, 1);
TEST_CUBLAS("cublasZaxpy");
gpufree(d);
@@ -1137,7 +1138,7 @@ inline NRVec<double>& NRVec<double>::operator+=(const NRVec<double> &rhs) {
* @return reference to the modified vector
******************************************************************************/
template<>
inline NRVec<complex<double> >& NRVec<complex<double> >::operator+=(const NRVec<complex<double> > &rhs) {
inline NRVec<std::complex<double> >& NRVec<std::complex<double> >::operator+=(const NRVec<std::complex<double> > &rhs) {
#ifdef DEBUG
if (nn != rhs.nn) laerror("incompatible dimensions");
#endif
@@ -1189,7 +1190,7 @@ inline NRVec<double> & NRVec<double>::operator-=(const NRVec<double> &rhs) {
* @return reference to the modified vector
******************************************************************************/
template<>
inline NRVec<complex<double> >& NRVec<complex<double> >::operator-=(const NRVec<complex<double> > &rhs) {
inline NRVec<std::complex<double> >& NRVec<std::complex<double> >::operator-=(const NRVec<std::complex<double> > &rhs) {
#ifdef DEBUG
if (nn != rhs.nn) laerror("incompatible dimensions");
#endif
@@ -1237,7 +1238,7 @@ inline NRVec<double>& NRVec<double>::operator*=(const double &a) {
* @return reference to the modified vector
******************************************************************************/
template<>
inline NRVec<complex<double> >& NRVec<complex<double> >::operator*=(const complex<double> &a) {
inline NRVec<std::complex<double> >& NRVec<std::complex<double> >::operator*=(const std::complex<double> &a) {
copyonwrite();
#ifdef CUDALA
if(location == cpu){
@@ -1285,11 +1286,11 @@ inline const double NRVec<double>::operator*(const NRVec<double> &rhs) const {
* @return \f$\sum_{i=1}^N\overbar{\vec{x}_i}\cdot\vec{y}_i\f$
******************************************************************************/
template<>
inline const complex<double> NRVec<complex<double> >::operator*(const NRVec< complex<double> > &rhs) const {
inline const std::complex<double> NRVec<std::complex<double> >::operator*(const NRVec< std::complex<double> > &rhs) const {
#ifdef DEBUG
if(nn != rhs.nn) laerror("incompatible dimensions");
#endif
complex<double> dot;
std::complex<double> dot;
SAME_LOC(*this, rhs);
#ifdef CUDALA
if(location == cpu){
@@ -1299,7 +1300,7 @@ inline const complex<double> NRVec<complex<double> >::operator*(const NRVec< com
}else{
const cuDoubleComplex val = cublasZdotc(nn, (cuDoubleComplex*)v, 1, (cuDoubleComplex*)rhs.v, 1);
TEST_CUBLAS("cublasZdotc");
dot = complex<double>(cuCreal(val), cuCimag(val));
dot = std::complex<double>(cuCreal(val), cuCimag(val));
}
#endif
return dot;
@@ -1324,8 +1325,8 @@ inline const double NRVec<double>::dot(const double *y, const int stride) const
* @return \f$\sum_{i=1}^N\vec{x}_{i}\cdot \overbar{y_{\mathrm{stride}\cdot(i-1) + 1}}\f$
******************************************************************************/
template<>
inline const complex<double> NRVec<complex<double> >::dot(const complex<double> *y, const int stride) const {
complex<double> dot;
inline const std::complex<double> NRVec<std::complex<double> >::dot(const std::complex<double> *y, const int stride) const {
std::complex<double> dot;
NOT_GPU(*this);
cblas_zdotc_sub(nn, y, stride, v, 1, &dot);
return dot;
@@ -1358,7 +1359,7 @@ inline const double NRVec<double>::asum() const {
* @return the value of this sum
******************************************************************************/
template<>
inline const double NRVec<complex<double> >::asum() const {
inline const double NRVec<std::complex<double> >::asum() const {
double ret(0.0);
#ifdef CUDALA
if(location == cpu){
@@ -1398,7 +1399,7 @@ inline const double NRVec<double>::norm() const {
* @return \f$\sum_{i=1}^N\left|\vec{x}_i\right|^2\f$
******************************************************************************/
template<>
inline const double NRVec< complex<double> >::norm() const {
inline const double NRVec< std::complex<double> >::norm() const {
double ret(0.);
#ifdef CUDALA
if(location == cpu){
@@ -1469,8 +1470,8 @@ inline const double NRVec<double>::amin() const {
* @return \f$\vec{v}_{j}\f$ which maximizes \f$\left\{\left|\Re{}\vec{v}_{i}\right|+\left|\Im{}\vec{v}_{i}\right|\right}\f$
******************************************************************************/
template<>
inline const complex<double> NRVec<complex<double> >::amax() const {
complex<double> ret(0., 0.);
inline const std::complex<double> NRVec<std::complex<double> >::amax() const {
std::complex<double> ret(0., 0.);
#ifdef CUDALA
if(location == cpu){
#endif
@@ -1479,7 +1480,7 @@ inline const complex<double> NRVec<complex<double> >::amax() const {
}else{
const int pozice = cublasIzamax(nn, (cuDoubleComplex*)v, 1) - 1;
TEST_CUBLAS("cublasIzamax");
gpuget(1, sizeof(complex<double>), v + pozice, &ret);
gpuget(1, sizeof(std::complex<double>), v + pozice, &ret);
}
#endif
return ret;
@@ -1491,15 +1492,15 @@ inline const complex<double> NRVec<complex<double> >::amax() const {
* @return \f$\vec{v}_{j}\f$ which minimizes \f$\left\{\left|\Re{}\vec{v}_{i}\right|+\left|\Im{}\vec{v}_{i}\right|\right}\f$
******************************************************************************/
template<>
inline const complex<double> NRVec<complex<double> >::amin() const {
complex<double> ret(0., 0.);
inline const std::complex<double> NRVec<std::complex<double> >::amin() const {
std::complex<double> ret(0., 0.);
#ifdef CUDALA
if(location == cpu){
#endif
// izamin seems not to be supported
int index(0);
double val(0.0), min_val(std::numeric_limits<double>::max());
complex<double> z_val(0.0, 0.0);
std::complex<double> z_val(0.0, 0.0);
for(register int i=0; i < nn; i++){
z_val = v[i];
@@ -1511,7 +1512,7 @@ inline const complex<double> NRVec<complex<double> >::amin() const {
}else{
const int pozice = cublasIzamin(nn, (cuDoubleComplex*)v, 1) - 1;
TEST_CUBLAS("cublasIzamin");
gpuget(1, sizeof(complex<double>), v + pozice, &ret);
gpuget(1, sizeof(std::complex<double>), v + pozice, &ret);
}
#endif
return ret;