| 1 |
/*! \file C_BlasRoutines.hpp |
| 2 |
\brief factorization routines LDL^t, LDU, forward/backward substitution |
| 3 |
\author François-Xavier Roux, ONERA, Laboratoire Jacques-Louis Lions |
| 4 |
\author Atsushi Suzuki, Laboratoire Jacques-Louis Lions |
| 5 |
\date Jul. 12th 2015 |
| 6 |
\date Nov. 30th 2016 |
| 7 |
*/ |
| 8 |
|
| 9 |
// This file is part of Dissection |
| 10 |
// |
| 11 |
// Dissection is free software: you can redistribute it and/or modify |
| 12 |
// it under the terms of the GNU General Public License as published by |
| 13 |
// the Free Software Foundation, either version 3 of the License, or |
| 14 |
// (at your option) any later version. |
| 15 |
// |
| 16 |
// Linking Dissection statically or dynamically with other modules is making |
| 17 |
// a combined work based on Disssection. Thus, the terms and conditions of |
| 18 |
// the GNU General Public License cover the whole combination. |
| 19 |
// |
| 20 |
// As a special exception, the copyright holders of Dissection give you |
| 21 |
// permission to combine Dissection program with free software programs or |
| 22 |
// libraries that are released under the GNU LGPL and with independent modules |
| 23 |
// that communicate with Dissection solely through the Dissection-fortran |
| 24 |
// interface. You may copy and distribute such a system following the terms of |
| 25 |
// the GNU GPL for Dissection and the licenses of the other code concerned, |
| 26 |
// provided that you include the source code of that other code when and as |
| 27 |
// the GNU GPL requires distribution of source code and provided that you do |
| 28 |
// not modify the Dissection-fortran interface. |
| 29 |
// |
| 30 |
// Note that people who make modified versions of Dissection are not obligated |
| 31 |
// to grant this special exception for their modified versions; it is their |
| 32 |
// choice whether to do so. The GNU General Public License gives permission to |
| 33 |
// release a modified version without this exception; this exception also makes |
| 34 |
// it possible to release a modified version which carries forward this |
| 35 |
// exception. If you modify the Dissection-fortran interface, this exception |
| 36 |
// does not apply to your modified version of Dissection, and you must remove |
| 37 |
// this exception when you distribute your modified version. |
| 38 |
// |
| 39 |
// This exception is an additional permission under section 7 of the GNU |
| 40 |
// General Public License, version 3 ("GPLv3") |
| 41 |
// |
| 42 |
// Dissection is distributed in the hope that it will be useful, |
| 43 |
// but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 44 |
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 45 |
// GNU General Public License for more details. |
| 46 |
// |
| 47 |
// You should have received a copy of the GNU General Public License |
| 48 |
// along with Dissection. If not, see <http://www.gnu.org/licenses/>. |
| 49 |
// |
| 50 |
|
| 51 |
# ifndef _DRIVER_C_BLASROUTINES_ |
| 52 |
# define _DRIVER_C_BLASROUTINES_ |
| 53 |
#include <vector> |
| 54 |
#include <list> |
| 55 |
#include "Compiler/blas.hpp" |
| 56 |
#include "Algebra/ColumnMatrix.hpp" |
| 57 |
|
| 58 |
using std::vector; |
| 59 |
using std::list; |
| 60 |
|
| 61 |
template<typename T, typename U, typename Z> |
| 62 |
Z matrix_infty_norm_(const int n, T *a, const int lda); |
| 63 |
|
| 64 |
template<typename T, typename U> |
| 65 |
U matrix_infty_norm(const int n, T *a, const int lda); |
| 66 |
template<> |
| 67 |
double matrix_infty_norm<quadruple, double>(const int n, quadruple *a, |
| 68 |
const int lda); |
| 69 |
|
| 70 |
template<> |
| 71 |
double matrix_infty_norm<complex<quadruple>, |
| 72 |
double>(const int n, complex<quadruple> *a, |
| 73 |
const int lda); |
| 74 |
|
| 75 |
#ifndef NO_OCTRUPLE |
| 76 |
template<> |
| 77 |
quadruple matrix_infty_norm<octruple, quadruple>(const int n, |
| 78 |
octruple *a, const int lda); |
| 79 |
|
| 80 |
template<> |
| 81 |
quadruple matrix_infty_norm<complex<octruple>, |
| 82 |
quadruple>(const int n, |
| 83 |
complex<octruple> *a, const int lda); |
| 84 |
#endif |
| 85 |
template<> |
| 86 |
float matrix_infty_norm<double, float>(const int n, double *a, |
| 87 |
const int lda); |
| 88 |
|
| 89 |
template<> |
| 90 |
float matrix_infty_norm<complex<double>, float>(const int n, complex<double> *a, |
| 91 |
const int lda); |
| 92 |
|
| 93 |
template<typename T> |
| 94 |
void full_ldlt(const int n, T *a, const int lda); |
| 95 |
|
| 96 |
template<typename T> |
| 97 |
void full_ldlh(const int n, T *a, const int lda); |
| 98 |
template<> |
| 99 |
void full_ldlh<double>(const int n, double *a, const int lda); |
| 100 |
template<> |
| 101 |
void full_ldlh<quadruple>(const int n, quadruple *a, const int lda); |
| 102 |
template<> |
| 103 |
void full_ldlh<float>(const int n, float *a, const int lda); |
| 104 |
|
| 105 |
template<typename T> |
| 106 |
void full_ldu(const int n, T *a, const int lda); |
| 107 |
|
| 108 |
template<typename T> |
| 109 |
void FillUpperBlock(const int nrow, T *coef, |
| 110 |
const int *prow, |
| 111 |
const int *indcols, const int *indvals, |
| 112 |
const int *old2new_i, const int *old2new_j, |
| 113 |
ColumnMatrix<T> &b); |
| 114 |
|
| 115 |
template<typename T, typename U> |
| 116 |
bool full_ldlt_permute(int *n0, const int nn0, const int n, |
| 117 |
T *a, const int lda, |
| 118 |
double *pivot, int *permute, |
| 119 |
const double eps, double *fop); |
| 120 |
|
| 121 |
template<typename T, typename U> |
| 122 |
bool full_ldu_permute(int *n0, const int nn0, const int n, |
| 123 |
T *a, const int lda, |
| 124 |
double *pivot, int *permute, |
| 125 |
const double eps, double *fop); |
| 126 |
|
| 127 |
template<typename T, typename U> |
| 128 |
bool ldu_full_permute(int *nn0, const int n0, |
| 129 |
const int n, T *a, const int lda, |
| 130 |
double *pivot, int *permute_right, int *permute_left, |
| 131 |
const double eps, double *fop); |
| 132 |
|
| 133 |
template<typename T> |
| 134 |
void full_fw_multiprofile(bool isTransposed, const int nrow, const int n0, |
| 135 |
const int ncol, |
| 136 |
T *a, |
| 137 |
const int lda, |
| 138 |
T *y, const int ldy, |
| 139 |
vector<int> &i0, double *fop); |
| 140 |
|
| 141 |
template<typename T> |
| 142 |
void full_fw_single(bool isTransposed, const int nrow, const int n0, |
| 143 |
T *a, const int lda, |
| 144 |
T *x, |
| 145 |
double *fop); |
| 146 |
|
| 147 |
template<typename T> |
| 148 |
void full_bw_single(bool isTransposed, const int nrow, const int n0, |
| 149 |
T *a, const int lda, |
| 150 |
T *x, |
| 151 |
double *fop); |
| 152 |
|
| 153 |
template<typename T> |
| 154 |
void full_fw_multi(bool isTransposed, const int nrow, const int n0, |
| 155 |
T *a, const int lda, |
| 156 |
const int ncol, |
| 157 |
T *x, const int ldy, |
| 158 |
double *fop); |
| 159 |
|
| 160 |
template<typename T> |
| 161 |
void full_bw_multi(bool isTransposed, const int nrow, const int n0, |
| 162 |
T *a, const int lda, |
| 163 |
const int ncol, |
| 164 |
T *x, const int ldy, |
| 165 |
double *fop); |
| 166 |
|
| 167 |
template<typename T> |
| 168 |
void SparseSchur(const bool isSym, const int dim2, const int dim1, |
| 169 |
vector<int>& i0, |
| 170 |
ColumnMatrix<T> &upper, |
| 171 |
ColumnMatrix<T> &lower, |
| 172 |
ColumnMatrix<T> &diag, |
| 173 |
double *fop); |
| 174 |
|
| 175 |
template<typename T> |
| 176 |
void full_fwbw_single(const bool isTrans, const int n, const int n0, |
| 177 |
T *a, const int lda, T *z); |
| 178 |
|
| 179 |
template<typename T> |
| 180 |
void full_fwbw_multi(const bool isTrans, const int n, const int n0, |
| 181 |
T *a, const int lda, |
| 182 |
const int m, T *x, const int ldx); |
| 183 |
|
| 184 |
template<typename T> |
| 185 |
void full_fwbw_part(const int n, T *a, const int lda, T*x); |
| 186 |
|
| 187 |
template<typename T> |
| 188 |
void SchurProfileSym(const int nrow, const int ncol, vector<int> &i0, |
| 189 |
ColumnMatrix<T> &b, ColumnMatrix<T> &c, |
| 190 |
T* s, const int size_b1, double *fop); |
| 191 |
|
| 192 |
template<typename T> |
| 193 |
void SchurProfileUnSym(const int nrow, const int ncol, vector<int> &i0, |
| 194 |
ColumnMatrix<T> &b, ColumnMatrix<T> &c, |
| 195 |
T* s, const int size_b1, double *fop); |
| 196 |
|
| 197 |
template<typename T> |
| 198 |
void swap_sym_lower(const int n, T *a, const int lda, |
| 199 |
const int k, const int km, T *col_k, T *col_km); |
| 200 |
|
| 201 |
template<typename T> |
| 202 |
void swap_sym_upper(const int n, T *a, const int lda, |
| 203 |
const int k, const int km, T *col_k, T *col_km); |
| 204 |
|
| 205 |
template<typename T> |
| 206 |
void swap_unsym(const int n, T *a, const int lda, |
| 207 |
const int k, const int km, T *col_k, T *col_km); |
| 208 |
|
| 209 |
template<typename T> |
| 210 |
void swap_full(const int n, T *a, const int lda, |
| 211 |
const int k, const int ki, const int kj, T *coli, T *colj); |
| 212 |
|
| 213 |
// routines for C_KernDetect.cpp |
| 214 |
template<typename T, typename Z> |
| 215 |
void full_fwbw_perturb_single(const int n, |
| 216 |
T *a, const int lda, T *a_fact, T *x, |
| 217 |
const int dim_augkern, const Z &eps, |
| 218 |
bool flag_sym); |
| 219 |
|
| 220 |
template<typename T, typename Z> |
| 221 |
void full_fwbw_perturb_multi(const int n, const int m, |
| 222 |
T *a, const int lda, T *a_fact, T *x, |
| 223 |
const int dim_augkern, const Z &eps, |
| 224 |
bool flag_sym); |
| 225 |
// |
| 226 |
|
| 227 |
template<typename T> |
| 228 |
void full_sym_2x2BK(int n, T *a, T *dd1, |
| 229 |
int *pivot_width, int *permute); |
| 230 |
|
| 231 |
template<typename T> |
| 232 |
void C_gemm_symm(const int ncol, const int nrow, const T &alpha, |
| 233 |
const T *a, const int lda, |
| 234 |
const T *b, const int ldb, |
| 235 |
const T &beta, |
| 236 |
T *c, const int ldc); |
| 237 |
|
| 238 |
#endif |