| 1 |
/*! \file RectBlockMatrix.cpp |
| 2 |
\brief Block storage for Rectangular matrix |
| 3 |
\author Atsushi Suzuki, Laboratoire Jacques-Louis Lions |
| 4 |
\date Jun. 11th 2013 |
| 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 |
#include "Algebra/RectBlockMatrix.hpp" |
| 52 |
#include "Compiler/blas.hpp" |
| 53 |
#include <cstdio> |
| 54 |
|
| 55 |
template<typename T> |
| 56 |
void RectBlockMatrix<T>::init(int dim_r, int dim_c, int block_size, int first_block) |
| 57 |
{ |
| 58 |
int itmp; |
| 59 |
|
| 60 |
_dim_r = dim_r; |
| 61 |
_dim_c = dim_c; |
| 62 |
_block_size = block_size; |
| 63 |
|
| 64 |
_block_size2 = _block_size * _block_size; |
| 65 |
|
| 66 |
if ((dim_r * dim_c) == 0) { |
| 67 |
_isdecomposed_c = false; |
| 68 |
_num_blocks_r = 0; |
| 69 |
_num_blocks_c = 0; |
| 70 |
_block_size_last_r = 0; |
| 71 |
_block_size_last_c = 0; |
| 72 |
_coefs = NULL; |
| 73 |
_allocation_status = NULL; |
| 74 |
_deallocation_status = false; |
| 75 |
return; |
| 76 |
} |
| 77 |
if (first_block > 0) { |
| 78 |
_isdecomposed_c = true; |
| 79 |
_dim0_c = first_block; |
| 80 |
_dim1_c = _dim_c - _dim0_c; |
| 81 |
_num_blocks_r = (_dim_r + _block_size - 1) / _block_size; |
| 82 |
_num_blocks0_c = (_dim0_c + _block_size - 1) / _block_size; |
| 83 |
_num_blocks1_c = (_dim1_c + _block_size - 1) / _block_size; |
| 84 |
_num_blocks_c = _num_blocks0_c + _num_blocks1_c; |
| 85 |
_block_sizes_r.resize(_num_blocks_r); |
| 86 |
_block_sizes_c.resize(_num_blocks_c); |
| 87 |
_indexblock_r.resize(_num_blocks_r + 1); |
| 88 |
_indexblock_c.resize(_num_blocks_c + 1); |
| 89 |
|
| 90 |
itmp = _dim_r % _block_size; |
| 91 |
_block_size_last_r = (itmp == 0) ? _block_size : itmp; |
| 92 |
int ktmp; |
| 93 |
ktmp = 0; |
| 94 |
if (_num_blocks_r > 0) { |
| 95 |
for (int i = 0; i < (_num_blocks_r - 1); i++, ktmp += _block_size) { |
| 96 |
_block_sizes_r[i] = _block_size; |
| 97 |
_indexblock_r[i] = ktmp; |
| 98 |
} |
| 99 |
_block_sizes_r[_num_blocks_r - 1] = _block_size_last_r; |
| 100 |
_indexblock_r[_num_blocks_r - 1] = ktmp; |
| 101 |
_indexblock_r[_num_blocks_r] = _dim_r; |
| 102 |
} |
| 103 |
itmp = _dim0_c % _block_size; |
| 104 |
_block_size_last0_c = (itmp == 0) ? _block_size : itmp; |
| 105 |
ktmp = 0; |
| 106 |
if (_num_blocks0_c > 0) { |
| 107 |
for (int i = 0; i < (_num_blocks0_c - 1); i++, ktmp += _block_size) { |
| 108 |
_block_sizes_c[i] = _block_size; |
| 109 |
_indexblock_c[i] = ktmp; |
| 110 |
} |
| 111 |
_block_sizes_c[_num_blocks0_c - 1] = _block_size_last0_c; |
| 112 |
_indexblock_c[_num_blocks0_c - 1] = ktmp; |
| 113 |
} |
| 114 |
itmp = _dim1_c % _block_size; |
| 115 |
_block_size_last1_c = (itmp == 0) ? _block_size : itmp; |
| 116 |
ktmp = _dim0_c; |
| 117 |
if (_num_blocks1_c > 0) { |
| 118 |
for (int i = _num_blocks0_c; i < (_num_blocks_c - 1); |
| 119 |
i++, ktmp += _block_size) { |
| 120 |
_block_sizes_c[i] = _block_size; |
| 121 |
_indexblock_c[i] = ktmp; |
| 122 |
} |
| 123 |
_block_sizes_c[_num_blocks_c - 1] = _block_size_last1_c; |
| 124 |
_indexblock_c[_num_blocks_c - 1] = ktmp; |
| 125 |
} |
| 126 |
_indexblock_c[_num_blocks_c] = _dim_c; |
| 127 |
} |
| 128 |
else { |
| 129 |
_isdecomposed_c = false; |
| 130 |
_num_blocks_r = (_dim_r + _block_size - 1) / _block_size; |
| 131 |
_num_blocks_c = (_dim_c + _block_size - 1) / _block_size; |
| 132 |
_block_sizes_r.resize(_num_blocks_r); |
| 133 |
_block_sizes_c.resize(_num_blocks_c); |
| 134 |
_indexblock_r.resize(_num_blocks_r + 1); |
| 135 |
_indexblock_c.resize(_num_blocks_c + 1); |
| 136 |
|
| 137 |
itmp = _dim_r % _block_size; |
| 138 |
_block_size_last_r = (itmp == 0) ? _block_size : itmp; |
| 139 |
int ktmp; |
| 140 |
ktmp = 0; |
| 141 |
if (_num_blocks_r > 0) { |
| 142 |
for (int i = 0; i < (_num_blocks_r - 1); i++, ktmp += _block_size) { |
| 143 |
_block_sizes_r[i] = _block_size; |
| 144 |
_indexblock_r[i] = ktmp; |
| 145 |
} |
| 146 |
_block_sizes_r[_num_blocks_r - 1] = _block_size_last_r; |
| 147 |
_indexblock_r[_num_blocks_r - 1] = ktmp; |
| 148 |
_indexblock_r[_num_blocks_r] = _dim_r; |
| 149 |
} |
| 150 |
itmp = _dim_c % _block_size; |
| 151 |
ktmp = 0; |
| 152 |
if (_num_blocks_c > 0) { |
| 153 |
_block_size_last_c = (itmp == 0) ? _block_size : itmp; |
| 154 |
for (int i = 0; i < (_num_blocks_c - 1); i++, ktmp += _block_size) { |
| 155 |
_block_sizes_c[i] = _block_size; |
| 156 |
_indexblock_c[i] = ktmp; |
| 157 |
} |
| 158 |
_block_sizes_c[_num_blocks_c - 1] = _block_size_last_c; |
| 159 |
_indexblock_c[_num_blocks_c - 1] = ktmp; |
| 160 |
_indexblock_c[_num_blocks_c] = _dim_c; |
| 161 |
} |
| 162 |
} |
| 163 |
if (_num_blocks_r * _num_blocks_c == 0) { |
| 164 |
_coefs = NULL; // new T*[1]; // dummy |
| 165 |
_allocation_status = NULL; // new bool[1]; // dummy |
| 166 |
_deallocation_status = false; |
| 167 |
// _deallocation_status = true; |
| 168 |
} |
| 169 |
else { |
| 170 |
try { |
| 171 |
_coefs = new T*[_num_blocks_r * _num_blocks_c]; // need to be managed |
| 172 |
} catch (const std::bad_alloc& e) { |
| 173 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 174 |
e.what()); |
| 175 |
} |
| 176 |
try { |
| 177 |
_allocation_status = new bool[_num_blocks_r * _num_blocks_c]; |
| 178 |
} catch (const std::bad_alloc& e) { |
| 179 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 180 |
e.what()); |
| 181 |
} |
| 182 |
for (int i = 0; i < (_num_blocks_r * _num_blocks_c); i++) { |
| 183 |
_allocation_status[i] = false; |
| 184 |
} |
| 185 |
_deallocation_status = false; |
| 186 |
} |
| 187 |
} |
| 188 |
|
| 189 |
template |
| 190 |
void RectBlockMatrix<double>::init(int dim_r, int dim_c, int block_size, |
| 191 |
int first_block); |
| 192 |
template |
| 193 |
void RectBlockMatrix<quadruple>::init(int dim_r, int dim_c, int block_size, |
| 194 |
int first_block); |
| 195 |
template |
| 196 |
void RectBlockMatrix<complex<double> >::init(int dim_r, int dim_c, |
| 197 |
int block_size, int first_block); |
| 198 |
template |
| 199 |
void RectBlockMatrix<complex<quadruple> >::init(int dim_r, int dim_c, |
| 200 |
int block_size, |
| 201 |
int first_block); |
| 202 |
|
| 203 |
template |
| 204 |
void RectBlockMatrix<float>::init(int dim_r, int dim_c, int block_size, |
| 205 |
int first_block); |
| 206 |
template |
| 207 |
void RectBlockMatrix<complex<float> >::init(int dim_r, int dim_c, |
| 208 |
int block_size, int first_block); |
| 209 |
|
| 210 |
// |
| 211 |
|
| 212 |
template<typename T> |
| 213 |
void RectBlockMatrix<T>::allocateBlock(int i, int j) |
| 214 |
{ |
| 215 |
const int nrow = _block_sizes_r[i]; |
| 216 |
const int ncol = _block_sizes_c[j]; |
| 217 |
if (!_allocation_status[i + j * _num_blocks_r]) { // |
| 218 |
try { |
| 219 |
_coefs[i + j * _num_blocks_r] = new T[nrow * ncol]; |
| 220 |
} catch (const std::bad_alloc& e) { |
| 221 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 222 |
e.what()); |
| 223 |
} |
| 224 |
} |
| 225 |
_allocation_status[i + j * _num_blocks_r] = true; |
| 226 |
} |
| 227 |
|
| 228 |
template |
| 229 |
void RectBlockMatrix<double>::allocateBlock(int i, int j); |
| 230 |
template |
| 231 |
void RectBlockMatrix<quadruple>::allocateBlock(int i, int j); |
| 232 |
template |
| 233 |
void RectBlockMatrix<complex<double> >::allocateBlock(int i, int j); |
| 234 |
template |
| 235 |
void RectBlockMatrix<complex<quadruple> >::allocateBlock(int i, int j); |
| 236 |
template |
| 237 |
void RectBlockMatrix<float>::allocateBlock(int i, int j); |
| 238 |
template |
| 239 |
void RectBlockMatrix<complex<float> >::allocateBlock(int i, int j); |
| 240 |
|
| 241 |
// |
| 242 |
|
| 243 |
template<typename T> |
| 244 |
void RectBlockMatrix<T>::allocate() |
| 245 |
{ |
| 246 |
for (int i = 0; i < _num_blocks_r; i++) { |
| 247 |
for (int j = 0; j < _num_blocks_c; j++) { |
| 248 |
const int nrow = _block_sizes_r[i]; |
| 249 |
const int ncol = _block_sizes_c[j]; |
| 250 |
if (!_allocation_status[i + j * _num_blocks_r]) { |
| 251 |
try { |
| 252 |
_coefs[i + j * _num_blocks_r] = new T[nrow * ncol]; |
| 253 |
} catch (const std::bad_alloc& e) { |
| 254 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 255 |
e.what()); |
| 256 |
} |
| 257 |
} |
| 258 |
_allocation_status[i + j * _num_blocks_r] = true; |
| 259 |
} |
| 260 |
} |
| 261 |
_deallocation_status = false; |
| 262 |
} |
| 263 |
|
| 264 |
template |
| 265 |
void RectBlockMatrix<double>::allocate(); |
| 266 |
template |
| 267 |
void RectBlockMatrix<quadruple>::allocate(); |
| 268 |
template |
| 269 |
void RectBlockMatrix<complex<double> >::allocate(); |
| 270 |
template |
| 271 |
void RectBlockMatrix<complex<quadruple> >::allocate(); |
| 272 |
template |
| 273 |
void RectBlockMatrix<float>::allocate(); |
| 274 |
template |
| 275 |
void RectBlockMatrix<complex<float> >::allocate(); |
| 276 |
|
| 277 |
// |
| 278 |
|
| 279 |
template<typename T> |
| 280 |
void RectBlockMatrix<T>::free(int i, int j) |
| 281 |
{ |
| 282 |
if (_allocation_status[i + j * _num_blocks_r]) { |
| 283 |
delete[] _coefs[i + j * _num_blocks_r]; // need to be mananged |
| 284 |
} |
| 285 |
_allocation_status[i + j * _num_blocks_r] = false; |
| 286 |
} |
| 287 |
|
| 288 |
template |
| 289 |
void RectBlockMatrix<double>::free(int i, int j); |
| 290 |
template |
| 291 |
void RectBlockMatrix<quadruple>::free(int i, int j); |
| 292 |
template |
| 293 |
void RectBlockMatrix<complex<double> >::free(int i, int j); |
| 294 |
template |
| 295 |
void RectBlockMatrix<complex<quadruple> >::free(int i, int j); |
| 296 |
template |
| 297 |
void RectBlockMatrix<float>::free(int i, int j); |
| 298 |
template |
| 299 |
void RectBlockMatrix<complex<float> >::free(int i, int j); |
| 300 |
// |
| 301 |
|
| 302 |
template<typename T> |
| 303 |
void RectBlockMatrix<T>::free() |
| 304 |
{ |
| 305 |
for (int i = 0; i < _num_blocks_r; i++) { |
| 306 |
for (int j = 0; j < _num_blocks_c; j++) { |
| 307 |
if (_allocation_status[i + j * _num_blocks_r]) { |
| 308 |
delete[] _coefs[i + j * _num_blocks_r]; // need to be managed |
| 309 |
_allocation_status[i + j * _num_blocks_r] = false; |
| 310 |
} |
| 311 |
} |
| 312 |
} |
| 313 |
} |
| 314 |
|
| 315 |
template |
| 316 |
void RectBlockMatrix<double>::free(); |
| 317 |
template |
| 318 |
void RectBlockMatrix<quadruple>::free(); |
| 319 |
template |
| 320 |
void RectBlockMatrix<complex<double> >::free(); |
| 321 |
template |
| 322 |
void RectBlockMatrix<complex<quadruple> >::free(); |
| 323 |
template |
| 324 |
void RectBlockMatrix<float>::free(); |
| 325 |
template |
| 326 |
void RectBlockMatrix<complex<float> >::free(); |
| 327 |
// |
| 328 |
|
| 329 |
template<typename T> |
| 330 |
T* RectBlockMatrix<T>::addrCoefBlock(int i, int j) |
| 331 |
{ |
| 332 |
return _coefs[i + j * _num_blocks_r]; |
| 333 |
} |
| 334 |
|
| 335 |
template |
| 336 |
double* RectBlockMatrix<double>::addrCoefBlock(int i, int j); |
| 337 |
|
| 338 |
template |
| 339 |
quadruple* RectBlockMatrix<quadruple>::addrCoefBlock(int i, int j); |
| 340 |
|
| 341 |
template |
| 342 |
complex<double>* RectBlockMatrix<complex<double> >:: |
| 343 |
addrCoefBlock(int i, int j); |
| 344 |
|
| 345 |
template |
| 346 |
complex<quadruple>* RectBlockMatrix<complex<quadruple> >:: |
| 347 |
addrCoefBlock(int i, int j); |
| 348 |
|
| 349 |
template |
| 350 |
float* RectBlockMatrix<float>::addrCoefBlock(int i, int j); |
| 351 |
|
| 352 |
template |
| 353 |
complex<float>* RectBlockMatrix<complex<float> >:: |
| 354 |
addrCoefBlock(int i, int j); |
| 355 |
// |
| 356 |
|
| 357 |
template<typename T> |
| 358 |
void RectBlockMatrix<T>::ZeroClear() |
| 359 |
{ |
| 360 |
const T zero(0.0); |
| 361 |
for (int i = 0; i < _num_blocks_r; i++) { |
| 362 |
for (int j = 0; j < _num_blocks_c; j++) { |
| 363 |
const int nrow = _block_sizes_r[i]; |
| 364 |
const int ncol = _block_sizes_c[j]; |
| 365 |
const int nsize = nrow * ncol; |
| 366 |
const int itmp = i + j * _num_blocks_r; |
| 367 |
for (int k = 0; k < nsize; k++) { |
| 368 |
_coefs[itmp][k] = zero; // need to be replaced by memcpy |
| 369 |
} |
| 370 |
} |
| 371 |
} |
| 372 |
} |
| 373 |
|
| 374 |
template |
| 375 |
void RectBlockMatrix<double>::ZeroClear(); |
| 376 |
|
| 377 |
template |
| 378 |
void RectBlockMatrix<quadruple>::ZeroClear(); |
| 379 |
|
| 380 |
template |
| 381 |
void RectBlockMatrix<complex<double> >::ZeroClear(); |
| 382 |
|
| 383 |
template |
| 384 |
void RectBlockMatrix<complex<quadruple> >::ZeroClear(); |
| 385 |
|
| 386 |
template |
| 387 |
void RectBlockMatrix<float>::ZeroClear(); |
| 388 |
|
| 389 |
template |
| 390 |
void RectBlockMatrix<complex<float> >::ZeroClear(); |
| 391 |
// |
| 392 |
|
| 393 |
template<typename T> |
| 394 |
T& RectBlockMatrix<T>::operator () (int i, int j) |
| 395 |
{ |
| 396 |
int i0, i1, j0, j1; |
| 397 |
|
| 398 |
if (_isdecomposed_c) { |
| 399 |
i0 = i / _block_size; |
| 400 |
i1 = i % _block_size; |
| 401 |
if (j < _dim0_c) { |
| 402 |
j0 = j / _block_size; |
| 403 |
j1 = j % _block_size; |
| 404 |
} |
| 405 |
else { |
| 406 |
j0 = (j - _dim0_c) / _block_size + _num_blocks0_c; |
| 407 |
j1 = (j - _dim0_c) % _block_size; |
| 408 |
} |
| 409 |
} |
| 410 |
else { |
| 411 |
i0 = i / _block_size; |
| 412 |
j0 = j / _block_size; |
| 413 |
i1 = i % _block_size; |
| 414 |
j1 = j % _block_size; |
| 415 |
} |
| 416 |
return _coefs[i0 + j0 * _num_blocks_r][i1 + j1 * _block_sizes_r[i0]]; |
| 417 |
} |
| 418 |
|
| 419 |
template |
| 420 |
double& RectBlockMatrix<double>::operator () (int i, int j); |
| 421 |
|
| 422 |
template |
| 423 |
quadruple& RectBlockMatrix<quadruple>::operator () (int i, int j); |
| 424 |
|
| 425 |
template |
| 426 |
complex<double> & RectBlockMatrix<complex<double> >:: |
| 427 |
operator () (int i, int j); |
| 428 |
|
| 429 |
template |
| 430 |
complex<quadruple> & RectBlockMatrix<complex<quadruple> >:: |
| 431 |
operator () (int i, int j); |
| 432 |
|
| 433 |
template |
| 434 |
float& RectBlockMatrix<float>::operator () (int i, int j); |
| 435 |
|
| 436 |
template |
| 437 |
complex<float> & RectBlockMatrix<complex<float> >:: |
| 438 |
operator () (int i, int j); |
| 439 |
|
| 440 |
// |
| 441 |
|
| 442 |
template<typename T> |
| 443 |
const T& RectBlockMatrix<T>::operator () (int i, int j) const |
| 444 |
{ |
| 445 |
int i0, i1, j0, j1; |
| 446 |
|
| 447 |
i0 = i / _block_size; |
| 448 |
j0 = j / _block_size; |
| 449 |
i1 = i % _block_size; |
| 450 |
j1 = j % _block_size; |
| 451 |
return _coefs[i0 + j0 * _num_blocks_r][i1 + j1 * _block_sizes_r[i0]]; |
| 452 |
} |
| 453 |
|
| 454 |
template |
| 455 |
const double& RectBlockMatrix<double>::operator () (int i, int j) const; |
| 456 |
|
| 457 |
template |
| 458 |
const quadruple& RectBlockMatrix<quadruple>::operator () (int i, int j) const; |
| 459 |
|
| 460 |
template |
| 461 |
const complex<double>& RectBlockMatrix<complex<double> >:: |
| 462 |
operator () (int i, int j) const; |
| 463 |
|
| 464 |
template |
| 465 |
const complex<quadruple>& RectBlockMatrix<complex<quadruple> >:: |
| 466 |
operator () (int i, int j) const; |
| 467 |
|
| 468 |
template |
| 469 |
const float& RectBlockMatrix<float>::operator () (int i, int j) const; |
| 470 |
|
| 471 |
template |
| 472 |
const complex<float>& RectBlockMatrix<complex<float> >:: |
| 473 |
operator () (int i, int j) const; |
| 474 |
// |