409 lines
9.5 KiB
C++
409 lines
9.5 KiB
C++
#include <iostream>
|
|
#include "vec.h"
|
|
#include <stdlib.h>
|
|
#include <sys/types.h>
|
|
#include <sys/stat.h>
|
|
#include <fcntl.h>
|
|
extern "C" {
|
|
extern ssize_t read(int, void *, size_t);
|
|
extern ssize_t write(int, const void *, size_t);
|
|
}
|
|
|
|
|
|
//////////////////////////////////////////////////////////////////////////////
|
|
//// forced instantization in the corespoding object file
|
|
#define INSTANTIZE(T) \
|
|
template ostream & operator<<(ostream &s, const NRVec< T > &x); \
|
|
template istream & operator>>(istream &s, NRVec< T > &x); \
|
|
template void NRVec<T>::put(int fd, bool dim) const; \
|
|
template void NRVec<T>::get(int fd, bool dim); \
|
|
|
|
|
|
|
|
INSTANTIZE(double)
|
|
INSTANTIZE(complex<double>)
|
|
INSTANTIZE(int)
|
|
INSTANTIZE(unsigned int)
|
|
INSTANTIZE(short)
|
|
INSTANTIZE(unsigned short)
|
|
INSTANTIZE(char)
|
|
INSTANTIZE(unsigned char)
|
|
template NRVec<double>;
|
|
template NRVec<complex<double> >;
|
|
template NRVec<char>;
|
|
template NRVec<short>;
|
|
template NRVec<int>;
|
|
|
|
|
|
|
|
|
|
/*
|
|
* Templates first, specializations for BLAS next
|
|
*/
|
|
|
|
// conversion ctor
|
|
#ifndef MATPTR
|
|
template <typename T>
|
|
NRVec<T>::NRVec(const NRMat<T> &rhs)
|
|
{
|
|
nn = rhs.nn*rhs.mm;
|
|
v = rhs.v;
|
|
count = rhs.count;
|
|
(*count)++;
|
|
}
|
|
#endif
|
|
|
|
|
|
|
|
|
|
// formatted I/O
|
|
template <typename T>
|
|
ostream & operator<<(ostream &s, const NRVec<T> &x)
|
|
{
|
|
int i, n;
|
|
|
|
n = x.size();
|
|
s << n << endl;
|
|
for(i=0; i<n; i++) s << x[i] << (i == n-1 ? '\n' : ' ');
|
|
return s;
|
|
}
|
|
|
|
template <typename T>
|
|
istream & operator>>(istream &s, NRVec<T> &x)
|
|
{
|
|
int i,n;
|
|
|
|
s >> n;
|
|
x.resize(n);
|
|
for(i=0; i<n; i++) s >> x[i];
|
|
return s;
|
|
}
|
|
|
|
|
|
//raw I/O
|
|
template <typename T>
|
|
void NRVec<T>::put(int fd, bool dim) const
|
|
{
|
|
errno=0;
|
|
int pad=1; //align at least 8-byte
|
|
if(dim)
|
|
{
|
|
if(sizeof(int) != write(fd,&nn,sizeof(int))) laerror("cannot write");
|
|
if(sizeof(int) != write(fd,&pad,sizeof(int))) laerror("cannot write");
|
|
}
|
|
LA_traits<T>::multiput(nn,fd,v,dim);
|
|
}
|
|
|
|
template <typename T>
|
|
void NRVec<T>::get(int fd, bool dim)
|
|
{
|
|
int nn0[2]; //align at least 8-byte
|
|
errno=0;
|
|
if(dim)
|
|
{
|
|
if(2*sizeof(int) != read(fd,&nn0,2*sizeof(int))) laerror("cannot read");
|
|
resize(nn0[0]);
|
|
}
|
|
else
|
|
copyonwrite();
|
|
LA_traits<T>::multiget(nn,fd,v,dim);
|
|
}
|
|
|
|
|
|
|
|
|
|
// formatted print for NRVec
|
|
template<typename T>
|
|
void NRVec<T>::fprintf(FILE *file, const char *format, const int modulo) const
|
|
{
|
|
lawritemat(file, v, 1, nn, format, 1, modulo, 0);
|
|
}
|
|
|
|
// formatted scan for NRVec
|
|
template <typename T>
|
|
void NRVec<T>::fscanf(FILE *f, const char *format)
|
|
{
|
|
int n;
|
|
|
|
if(std::fscanf(f, "%d", &n) != 1) laerror("cannot read vector dimension");
|
|
resize(n);
|
|
for (int i=0; i<n; i++)
|
|
if (std::fscanf(f, format, v+i) != 1)
|
|
laerror("cannot read the vector eleemnt");
|
|
}
|
|
|
|
// unary minus
|
|
template <typename T>
|
|
const NRVec<T> NRVec<T>::operator-() const
|
|
{
|
|
NRVec<T> result(nn);
|
|
for (int i=0; i<nn; i++) result.v[i]= -v[i];
|
|
return result;
|
|
}
|
|
|
|
|
|
//comparison operators (for lexical order)
|
|
|
|
template <typename T>
|
|
const bool NRVec<T>::operator>(const NRVec &rhs) const
|
|
{
|
|
int n=nn; if(rhs.nn<n) n=rhs.nn;
|
|
for(int i=0; i<n;++i)
|
|
{
|
|
if(LA_traits<T>::bigger(v[i],rhs.v[i])) return true;
|
|
if(LA_traits<T>::smaller(v[i],rhs.v[i])) return false;
|
|
}
|
|
return nn>rhs.nn;
|
|
}
|
|
|
|
template <typename T>
|
|
const bool NRVec<T>::operator<(const NRVec &rhs) const
|
|
{
|
|
int n=nn; if(rhs.nn<n) n=rhs.nn;
|
|
for(int i=0; i<n;++i)
|
|
{
|
|
if(LA_traits<T>::smaller(v[i],rhs.v[i])) return true;
|
|
if(LA_traits<T>::bigger(v[i],rhs.v[i])) return false;
|
|
}
|
|
return nn<rhs.nn;
|
|
}
|
|
|
|
|
|
|
|
// axpy call for T = double (not strided)
|
|
void NRVec<double>::axpy(const double alpha, const NRVec<double> &x)
|
|
{
|
|
#ifdef DEBUG
|
|
if (nn != x.nn) laerror("axpy of incompatible vectors");
|
|
#endif
|
|
copyonwrite();
|
|
cblas_daxpy(nn, alpha, x.v, 1, v, 1);
|
|
}
|
|
|
|
// axpy call for T = complex<double> (not strided)
|
|
void NRVec< complex<double> >::axpy(const complex<double> alpha,
|
|
const NRVec< complex<double> > &x)
|
|
{
|
|
#ifdef DEBUG
|
|
if (nn != x.nn) laerror("axpy of incompatible vectors");
|
|
#endif
|
|
copyonwrite();
|
|
cblas_zaxpy(nn, (void *)(&alpha), (void *)(x.v), 1, (void *)v, 1);
|
|
}
|
|
|
|
// axpy call for T = double (strided)
|
|
void NRVec<double>::axpy(const double alpha, const double *x, const int stride)
|
|
{
|
|
copyonwrite();
|
|
cblas_daxpy(nn, alpha, x, stride, v, 1);
|
|
}
|
|
|
|
// axpy call for T = complex<double> (strided)
|
|
void NRVec< complex<double> >::axpy(const complex<double> alpha,
|
|
const complex<double> *x, const int stride)
|
|
{
|
|
copyonwrite();
|
|
cblas_zaxpy(nn, (void *)(&alpha), (void *)x, stride, v, 1);
|
|
}
|
|
|
|
// unary minus
|
|
const NRVec<double> NRVec<double>::operator-() const
|
|
{
|
|
NRVec<double> result(*this);
|
|
result.copyonwrite();
|
|
cblas_dscal(nn, -1.0, result.v, 1);
|
|
return result;
|
|
}
|
|
const NRVec< complex<double> >
|
|
NRVec< complex<double> >::operator-() const
|
|
{
|
|
NRVec< complex<double> > result(*this);
|
|
result.copyonwrite();
|
|
cblas_zdscal(nn, -1.0, (void *)(result.v), 1);
|
|
return result;
|
|
}
|
|
|
|
// assignment of scalar to every element
|
|
template <typename T>
|
|
NRVec<T> & NRVec<T>::operator=(const T &a)
|
|
{
|
|
copyonwrite();
|
|
if(a != (T)0)
|
|
for (int i=0; i<nn; i++) v[i] = a;
|
|
else
|
|
memset(v, 0, nn*sizeof(T));
|
|
return *this;
|
|
}
|
|
|
|
// Normalization of NRVec<double>
|
|
NRVec<double> & NRVec<double>::normalize()
|
|
{
|
|
double tmp;
|
|
|
|
tmp = cblas_dnrm2(nn, v, 1);
|
|
#ifdef DEBUG
|
|
if(!tmp) laerror("normalization of zero vector");
|
|
#endif
|
|
copyonwrite();
|
|
tmp = 1.0/tmp;
|
|
cblas_dscal(nn, tmp, v, 1);
|
|
return *this;
|
|
}
|
|
|
|
// Normalization of NRVec< complex<double> >
|
|
NRVec< complex<double> > & NRVec< complex<double> >::normalize()
|
|
{
|
|
complex<double> tmp;
|
|
tmp = cblas_dznrm2(nn, (void *)v, 1);
|
|
#ifdef DEBUG
|
|
if(!(tmp.real()) && !(tmp.imag())) laerror("normalization of zero vector");
|
|
#endif
|
|
copyonwrite();
|
|
tmp = 1.0/tmp;
|
|
cblas_zscal(nn, (void *)(&tmp), (void *)v, 1);
|
|
return *this;
|
|
}
|
|
|
|
//stubs for linkage
|
|
NRVec<int> & NRVec<int>::normalize() {laerror("normalize() impossible for integer types"); return *this;}
|
|
NRVec<short> & NRVec<short>::normalize() {laerror("normalize() impossible for integer types"); return *this;}
|
|
NRVec<char> & NRVec<char>::normalize() {laerror("normalize() impossible for integer types"); return *this;}
|
|
void NRVec<int>::gemv(const int beta,
|
|
const NRSMat<int> &A, const char trans,
|
|
const int alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
void NRVec<short>::gemv(const short beta,
|
|
const NRSMat<short> &A, const char trans,
|
|
const short alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
|
|
void NRVec<char>::gemv(const char beta,
|
|
const NRSMat<char> &A, const char trans,
|
|
const char alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
void NRVec<int>::gemv(const int beta,
|
|
const NRMat<int> &A, const char trans,
|
|
const int alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
void NRVec<short>::gemv(const short beta,
|
|
const NRMat<short> &A, const char trans,
|
|
const short alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
|
|
void NRVec<char>::gemv(const char beta,
|
|
const NRMat<char> &A, const char trans,
|
|
const char alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
void NRVec<int>::gemv(const int beta,
|
|
const SparseMat<int> &A, const char trans,
|
|
const int alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
void NRVec<short>::gemv(const short beta,
|
|
const SparseMat<short> &A, const char trans,
|
|
const short alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
|
|
void NRVec<char>::gemv(const char beta,
|
|
const SparseMat<char> &A, const char trans,
|
|
const char alpha, const NRVec &x)
|
|
{
|
|
laerror("not yet implemented");
|
|
}
|
|
|
|
|
|
|
|
|
|
// gemv calls
|
|
void NRVec<double>::gemv(const double beta, const NRMat<double> &A,
|
|
const char trans, const double alpha, const NRVec &x)
|
|
{
|
|
#ifdef DEBUG
|
|
if ((trans == 'n'?A.ncols():A.nrows()) != x.size())
|
|
laerror("incompatible sizes in gemv A*x");
|
|
#endif
|
|
cblas_dgemv(CblasRowMajor, (trans=='n' ? CblasNoTrans:CblasTrans),
|
|
A.nrows(), A.ncols(), alpha, A, A.ncols(), x.v, 1, beta, v, 1);
|
|
}
|
|
|
|
void NRVec< complex<double> >::gemv(const complex<double> beta,
|
|
const NRMat< complex<double> > &A, const char trans,
|
|
const complex<double> alpha, const NRVec &x)
|
|
{
|
|
#ifdef DEBUG
|
|
if ((trans == 'n'?A.ncols():A.nrows()) != x.size())
|
|
laerror("incompatible sizes in gemv A*x");
|
|
#endif
|
|
cblas_zgemv(CblasRowMajor, (trans=='n' ? CblasNoTrans:CblasTrans),
|
|
A.nrows(), A.ncols(), &alpha, A, A.ncols(),
|
|
x.v, 1, &beta, v, 1);
|
|
}
|
|
|
|
|
|
void NRVec<double>::gemv(const double beta, const NRSMat<double> &A,
|
|
const char trans, const double alpha, const NRVec &x)
|
|
{
|
|
#ifdef DEBUG
|
|
if (A.ncols()!=x.size()) laerror("incompatible dimension in gemv A*x");
|
|
#endif
|
|
NRVec<double> result(nn);
|
|
cblas_dspmv(CblasRowMajor, CblasLower, A.ncols(), alpha, A, x.v, 1, beta, v, 1);
|
|
}
|
|
|
|
|
|
void NRVec< complex<double> >::gemv(const complex<double> beta,
|
|
const NRSMat< complex<double> > &A, const char trans,
|
|
const complex<double> alpha, const NRVec &x)
|
|
{
|
|
#ifdef DEBUG
|
|
if (A.ncols()!=x.size()) laerror("incompatible dimension in gemv");
|
|
#endif
|
|
NRVec< complex<double> > result(nn);
|
|
cblas_zhpmv(CblasRowMajor, CblasLower, A.ncols(), &alpha, A,
|
|
x.v, 1, &beta, v, 1);
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
// Direc product Mat = Vec | Vec
|
|
const NRMat<double> NRVec<double>::operator|(const NRVec<double> &b) const
|
|
{
|
|
NRMat<double> result(0.,nn,b.nn);
|
|
cblas_dger(CblasRowMajor, nn, b.nn, 1., v, 1, b.v, 1, result, b.nn);
|
|
return result;
|
|
}
|
|
const NRMat< complex<double> >
|
|
NRVec< complex<double> >::operator|(const NRVec< complex<double> > &b) const
|
|
{
|
|
NRMat< complex<double> > result(0.,nn,b.nn);
|
|
cblas_zgerc(CblasRowMajor, nn, b.nn, &CONE, v, 1, b.v, 1, result, b.nn);
|
|
return result;
|
|
}
|
|
|
|
|