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

bugfixes in mat.h mat.cc and related

Browse Source
This commit is contained in:
jiri committed 2026-10-27 19:14:30 +01:00
1 parent 37f790c20a
commit ad98173078
4 files changed
+44 -45
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

mat.cc
+16 -38
View File
@@ -1025,7 +1025,7 @@ std::complex<double> tmp;
#ifdef CUDALA
if(location == cpu){
#endif
for(register int i=1; i<nn; i++){
for(register int i=0; i<nn; i++){
for(register int j=0; j<=i; j++){
#ifdef MATPTR
tmp = std::complex<double> (v[i][j].real()-v[j][i].real(),v[i][j].imag()+v[j][i].imag());
@@ -1094,9 +1094,9 @@ NRMat<std::complex<double> >::NRMat(const NRMat<double> &rhs, bool imagpart): nn
if(location == cpu){
#endif
#ifdef MATPTR
v = new std::complex<double>*[n];
v = new std::complex<double>*[nn];
v[0] = new std::complex<double>[nn_mm];
for(register int i=1; i<n; i++) v[i] = v[i-1] + m;
for(register int i=1; i<nn; i++) v[i] = v[i-1] + mm;
memset(v[0], 0, nn_mm*sizeof(std::complex<double>));
cblas_dcopy(nn_mm, &rhs[0][0], 1, ((double *)v[0]) + (imagpart?1:0), 2);
@@ -1135,9 +1135,9 @@ NRMat<double>::NRMat(const NRMat<std::complex<double> > &rhs, bool imagpart): nn
if(location == cpu){
#endif
#ifdef MATPTR
v = new double*[n];
v = new double*[nn];
v[0] = new double[nn_mm];
for(register int i=1; i<n; i++) v[i] = v[i-1] + m;
for(register int i=1; i<nn; i++) v[i] = v[i-1] + mm;
cblas_dcopy(nn_mm, ((double *)&rhs[0][0]) + (imagpart?1:0), 2, v[0], 1);
#else
@@ -1357,9 +1357,9 @@ const NRSMat<std::complex<double> > NRMat<std::complex<double> >::timestranspos
for(i=0; i<nn; ++i){
for(j=0; j<=i; ++j){
#ifdef MATPTR
cblas_zdotc_sub(nn, v[i], 1, v[j], 1, &r(i,j));
cblas_zdotc_sub(mm, v[j], 1, v[i], 1, &r(i,j)); //(A A^H)_ij = sum_k A_ik conj(A_jk), lower triangle of the hermitian result
#else
cblas_zdotc_sub(nn, v + i*(size_t)mm, 1, v + j*(size_t)mm, 1, &r(i,j));
cblas_zdotc_sub(mm, v + j*(size_t)mm, 1, v + i*(size_t)mm, 1, &r(i,j)); //(A A^H)_ij = sum_k A_ik conj(A_jk), lower triangle of the hermitian result
#endif
}
}
@@ -2162,7 +2162,8 @@ NRMat< std::complex<double> >::operator*(const NRSMat< std::complex<double> > &r
if(location == cpu){
#endif
for(register int i=0; i<nn; i++){
cblas_zhpmv(CblasRowMajor, CblasLower, mm, &CONE, &rhs[0],
//row i of A H is H^T a_i; the C-order lower packed hermitian H read as fortran-order upper packed is H^T
cblas_zhpmv(CblasColMajor, CblasUpper, mm, &CONE, &rhs[0],
(*this)[i], 1, &CZERO, result[i], 1);
}
#ifdef CUDALA
@@ -3143,24 +3144,9 @@ NRMat<double>& NRMat<double>::swap_rows_cols(){
#ifdef CUDALA
if(location == cpu){
#endif
for(register int i=0; i<n_pul; i++){
std::cout << "swapping row " << i << " and " << nn-i-1 << std::endl;
std::cout << "elements: " << *((*this)[i]) << std::endl;
std::cout << "elements: " << *((*this)[nn - i - 1] + mm - 1) << std::endl;
cblas_dswap(mm, (*this)[i], 1, (*this)[nn - i - 1] + mm - 1, -1);
}
return *this;
if(nn & 1){ // odd number of rows
for(register int i=0; i<=m_pul; i++){
tmp = (*this)(n_pul, i);
(*this)(n_pul, i) = (*this)(n_pul, mm-i-1);
(*this)(n_pul, mm-i-1) = tmp;
}
}
//reversal of the whole C-order array; for a negative increment BLAS expects the pointer to the first element in memory
const size_t dim = (size_t)nn*mm;
cblas_dswap(dim/2, (*this)[0], 1, (*this)[0] + dim - dim/2, -1);
#ifdef CUDALA
}else{
for(register int i=0; i<n_pul; i++){
@@ -3197,17 +3183,9 @@ NRMat<std::complex<double> >& NRMat<std::complex<double> >::swap_rows_cols(){
#ifdef CUDALA
if(location == cpu){
#endif
for(register int i=0;i<n_pul;i++){
cblas_zswap(mm, (*this)[i], 1, (*this)[nn - i - 1] + mm - 1, -1);
}
if(nn & 1){
for(register int i=0; i<=m_pul; i++){ // odd number of rows
tmp = (*this)(n_pul, i);
(*this)(n_pul, i) = (*this)(n_pul, mm-i-1);
(*this)(n_pul, mm-i-1) = tmp;
}
}
//reversal of the whole C-order array; for a negative increment BLAS expects the pointer to the first element in memory
const size_t dim = (size_t)nn*mm;
cblas_zswap(dim/2, (*this)[0], 1, (*this)[0] + dim - dim/2, -1);
#ifdef CUDALA
}else{
for(register int i=0;i<n_pul;i++){
@@ -3247,7 +3225,7 @@ NRMat<T>& NRMat<T>::swap_rows_cols(){
#endif
data_ptr = (*this)[0];
const int dim_pul = dim >> 1;
for(register int i=0; i<=dim_pul; i++){
for(register int i=0; i<dim_pul; i++){
tmp = data_ptr[i];
data_ptr[i] = data_ptr[dim - i - 1];
data_ptr[dim - i - 1] = tmp;