| 1 |
/*! \file SquareBlockMatrix.cpp |
| 2 |
\brief Block storage for symmetric/unsymmetric Square 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/SquareBlockMatrix.hpp" |
| 52 |
#include "Compiler/blas.hpp" // #include<complex> is inside |
| 53 |
|
| 54 |
template<typename T> |
| 55 |
void SquareBlockMatrix<T>::init(int dim, int block_size, bool isSym, |
| 56 |
int first_block) |
| 57 |
{ |
| 58 |
int itmp; |
| 59 |
_pivrelaxed = false; |
| 60 |
_coefs_alloc_status = false; // for safty |
| 61 |
_dim = dim; |
| 62 |
_block_size = block_size; |
| 63 |
_isSym = isSym; |
| 64 |
_block_size2 = _block_size * _block_size; |
| 65 |
|
| 66 |
if (dim == 0) { |
| 67 |
_isblocked = true; |
| 68 |
_kernelDetected = false; |
| 69 |
_lower_allocated = false; |
| 70 |
_coefs = NULL; |
| 71 |
_coefs_alloc_status = false; |
| 72 |
_allocation_status = NULL; |
| 73 |
return; |
| 74 |
} |
| 75 |
if (first_block > 0) { |
| 76 |
_isdecomposed = true; |
| 77 |
_dim0 = first_block; |
| 78 |
_dim1 = _dim - _dim0; |
| 79 |
|
| 80 |
_num_blocks0 = (_dim0 + _block_size - 1) / _block_size; |
| 81 |
_num_blocks1 = (_dim1 + _block_size - 1) / _block_size; |
| 82 |
_num_blocks = _num_blocks0 + _num_blocks1; |
| 83 |
|
| 84 |
_block_sizes.resize(_num_blocks); |
| 85 |
itmp = _dim0 % _block_size; |
| 86 |
_block_size_last0 = (itmp == 0) ? _block_size : itmp; |
| 87 |
itmp = _dim1 % _block_size; |
| 88 |
_block_size_last1 = (itmp == 0) ? _block_size : itmp; |
| 89 |
for (int i = 0; i < (_num_blocks0 - 1); i++) { |
| 90 |
_block_sizes[i] = _block_size; |
| 91 |
} |
| 92 |
_block_sizes[_num_blocks0 - 1] = _block_size_last0; |
| 93 |
if (_num_blocks1 > 0) { |
| 94 |
for (int i = _num_blocks0; i < (_num_blocks - 1); i++) { |
| 95 |
_block_sizes[i] = _block_size; |
| 96 |
} |
| 97 |
_block_sizes[_num_blocks - 1] = _block_size_last1; |
| 98 |
} |
| 99 |
else { |
| 100 |
_block_size_last1 = 0; |
| 101 |
} |
| 102 |
} |
| 103 |
else { |
| 104 |
_isdecomposed = false; |
| 105 |
_num_blocks = (_dim + _block_size - 1) / _block_size; |
| 106 |
_block_sizes.resize(_num_blocks); |
| 107 |
itmp = _dim % _block_size; |
| 108 |
_block_size_last = (itmp == 0) ? _block_size : itmp; |
| 109 |
for (int i = 0; i < (_num_blocks - 1); i++) { |
| 110 |
_block_sizes[i] = _block_size; |
| 111 |
} |
| 112 |
_block_sizes[_num_blocks - 1] = _block_size_last; |
| 113 |
} |
| 114 |
_nsing_block.resize(_num_blocks); |
| 115 |
if (_num_blocks > 0) { |
| 116 |
for (int i = 0; i < _num_blocks; i++) { |
| 117 |
_nsing_block[i] = 0; |
| 118 |
} |
| 119 |
} |
| 120 |
_isblocked = true; |
| 121 |
_kernelDetected = false; |
| 122 |
_lower_allocated = false; |
| 123 |
|
| 124 |
if (_num_blocks > 0) { |
| 125 |
try { |
| 126 |
_coefs = new T*[_num_blocks * _num_blocks]; // need to be managed |
| 127 |
} catch (const std::bad_alloc& e) { |
| 128 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 129 |
e.what()); |
| 130 |
} |
| 131 |
try { |
| 132 |
_allocation_status = new bool[_num_blocks * _num_blocks]; |
| 133 |
} catch (const std::bad_alloc& e) { |
| 134 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 135 |
e.what()); |
| 136 |
} |
| 137 |
_coefs_alloc_status = true; |
| 138 |
for (int i = 0; i < (_num_blocks * _num_blocks); i++) { |
| 139 |
_allocation_status[i] = false; |
| 140 |
} |
| 141 |
if (_isSym) { |
| 142 |
_size = 0; |
| 143 |
for (int i = 0; i < _num_blocks; i++) { |
| 144 |
for (int j = i; j < _num_blocks; j++) { |
| 145 |
_size += _block_sizes[i] * _block_sizes[j]; |
| 146 |
} |
| 147 |
} |
| 148 |
} |
| 149 |
else { |
| 150 |
_size = 0; |
| 151 |
for (int i = 0; i < _num_blocks; i++) { |
| 152 |
for (int j = 0; j < _num_blocks; j++) { |
| 153 |
_size += _block_sizes[i] * _block_sizes[j]; |
| 154 |
} |
| 155 |
} |
| 156 |
} |
| 157 |
} |
| 158 |
else { |
| 159 |
_coefs = NULL; |
| 160 |
_coefs_alloc_status = false; |
| 161 |
_allocation_status = NULL; |
| 162 |
} |
| 163 |
} |
| 164 |
|
| 165 |
template |
| 166 |
void SquareBlockMatrix<double>::init(int dim, int block_size, bool isSym, |
| 167 |
int first_block); |
| 168 |
template |
| 169 |
void SquareBlockMatrix<quadruple>::init(int dim, int block_size, bool isSym, |
| 170 |
int first_block); |
| 171 |
template |
| 172 |
void SquareBlockMatrix<complex<double> >::init(int dim, int block_size, |
| 173 |
bool isSym, int first_block); |
| 174 |
template |
| 175 |
void SquareBlockMatrix<complex<quadruple> >::init(int dim, int block_size, |
| 176 |
bool isSym, int first_block); |
| 177 |
|
| 178 |
template |
| 179 |
void SquareBlockMatrix<float>::init(int dim, int block_size, bool isSym, |
| 180 |
int first_block); |
| 181 |
template |
| 182 |
void SquareBlockMatrix<complex<float> >::init(int dim, int block_size, |
| 183 |
bool isSym, int first_block); |
| 184 |
|
| 185 |
// |
| 186 |
|
| 187 |
template<typename T> |
| 188 |
void SquareBlockMatrix<T>::allocateBlock(int i, int j) |
| 189 |
{ |
| 190 |
const int nrow = _block_sizes[i]; |
| 191 |
const int ncol = _block_sizes[j]; |
| 192 |
const int itmp = i + j * _num_blocks; |
| 193 |
if (!_allocation_status[itmp]) { |
| 194 |
try { |
| 195 |
_coefs[itmp] = new T[nrow * ncol]; |
| 196 |
} catch (const std::bad_alloc& e) { |
| 197 |
fprintf(stderr, "%s %d : allocation failed : %s", __FILE__, __LINE__, |
| 198 |
e.what()); |
| 199 |
} |
| 200 |
_allocation_status[itmp] = true; |
| 201 |
} |
| 202 |
} |
| 203 |
|
| 204 |
template |
| 205 |
void SquareBlockMatrix<double>::allocateBlock(int i, int j); |
| 206 |
template |
| 207 |
void SquareBlockMatrix<quadruple>::allocateBlock(int i, int j); |
| 208 |
template |
| 209 |
void SquareBlockMatrix<complex<double> >::allocateBlock(int i, int j); |
| 210 |
template |
| 211 |
void SquareBlockMatrix<complex<quadruple> >::allocateBlock(int i, int j); |
| 212 |
template |
| 213 |
void SquareBlockMatrix<float>::allocateBlock(int i, int j); |
| 214 |
template |
| 215 |
void SquareBlockMatrix<complex<float> >::allocateBlock(int i, int j); |
| 216 |
|
| 217 |
// |
| 218 |
|
| 219 |
template<typename T> |
| 220 |
void SquareBlockMatrix<T>::allocate() |
| 221 |
{ |
| 222 |
if (_isSym) { |
| 223 |
// int ioffset = 0; |
| 224 |
for (int i = 0; i < _num_blocks; i++) { |
| 225 |
for (int j = i; j < _num_blocks; j++) { |
| 226 |
const int nrow = _block_sizes[i]; |
| 227 |
const int ncol = _block_sizes[j]; |
| 228 |
const int itmp = i + j * _num_blocks; |
| 229 |
if (!_allocation_status[itmp]) { |
| 230 |
try { |
| 231 |
_coefs[itmp] = new T[nrow * ncol]; |
| 232 |
} catch (const std::bad_alloc& e) { |
| 233 |
fprintf(stderr, "%s %d : allocation failed : %s", |
| 234 |
__FILE__, __LINE__, |
| 235 |
e.what()); |
| 236 |
} |
| 237 |
} |
| 238 |
_allocation_status[itmp] = true; |
| 239 |
} |
| 240 |
} |
| 241 |
} |
| 242 |
else { |
| 243 |
// int ioffset = 0; |
| 244 |
for (int i = 0; i < _num_blocks; i++) { |
| 245 |
for (int j = 0; j < _num_blocks; j++) { |
| 246 |
const int nrow = _block_sizes[i]; |
| 247 |
const int ncol = _block_sizes[j]; |
| 248 |
const int itmp = i + j * _num_blocks; |
| 249 |
if (!_allocation_status[itmp]) { |
| 250 |
try { |
| 251 |
_coefs[itmp] = new T[nrow * ncol]; |
| 252 |
} catch (const std::bad_alloc& e) { |
| 253 |
fprintf(stderr, "%s %d : allocation failed : %s", |
| 254 |
__FILE__, __LINE__, |
| 255 |
e.what()); |
| 256 |
} |
| 257 |
} |
| 258 |
_allocation_status[itmp] = true; |
| 259 |
} |
| 260 |
} |
| 261 |
} |
| 262 |
} |
| 263 |
|
| 264 |
template |
| 265 |
void SquareBlockMatrix<double>::allocate(); |
| 266 |
template |
| 267 |
void SquareBlockMatrix<quadruple>::allocate(); |
| 268 |
template |
| 269 |
void SquareBlockMatrix<complex<double> >::allocate(); |
| 270 |
template |
| 271 |
void SquareBlockMatrix<complex<quadruple> >::allocate(); |
| 272 |
template |
| 273 |
void SquareBlockMatrix<float>::allocate(); |
| 274 |
template |
| 275 |
void SquareBlockMatrix<complex<float> >::allocate(); |
| 276 |
// |
| 277 |
|
| 278 |
template<typename T> |
| 279 |
void SquareBlockMatrix<T>::free(int i, int j) |
| 280 |
{ |
| 281 |
const int itmp = i + j * _num_blocks; |
| 282 |
if (_allocation_status[itmp]) { |
| 283 |
delete [] _coefs[itmp]; // need to be mananged |
| 284 |
} |
| 285 |
_allocation_status[itmp] = false; |
| 286 |
} |
| 287 |
|
| 288 |
template |
| 289 |
void SquareBlockMatrix<double>::free(int i, int j); |
| 290 |
template |
| 291 |
void SquareBlockMatrix<quadruple>::free(int i, int j); |
| 292 |
template |
| 293 |
void SquareBlockMatrix<complex<double> >::free(int i, int j); |
| 294 |
template |
| 295 |
void SquareBlockMatrix<complex<quadruple> >::free(int i, int j); |
| 296 |
template |
| 297 |
void SquareBlockMatrix<float>::free(int i, int j); |
| 298 |
template |
| 299 |
void SquareBlockMatrix<complex<float> >::free(int i, int j); |
| 300 |
|
| 301 |
// |
| 302 |
|
| 303 |
template<typename T> |
| 304 |
void SquareBlockMatrix<T>::free() |
| 305 |
{ |
| 306 |
if (_isSym) { |
| 307 |
for (int i = 0; i < _num_blocks; i++) { |
| 308 |
for (int j = i; j < _num_blocks; j++) { |
| 309 |
const int itmp = i + j * _num_blocks; |
| 310 |
if (_allocation_status[itmp]) { |
| 311 |
delete [] _coefs[itmp]; // need to be mananged |
| 312 |
} |
| 313 |
_allocation_status[itmp]= false; |
| 314 |
} |
| 315 |
} |
| 316 |
} |
| 317 |
else { |
| 318 |
for (int i = 0; i < _num_blocks; i++) { |
| 319 |
for (int j = 0; j < _num_blocks; j++) { |
| 320 |
const int itmp = i + j * _num_blocks; |
| 321 |
if (_allocation_status[itmp]) { |
| 322 |
delete [] _coefs[itmp]; // need to be managed |
| 323 |
} |
| 324 |
_allocation_status[itmp] = false; |
| 325 |
} |
| 326 |
} |
| 327 |
} |
| 328 |
} |
| 329 |
template |
| 330 |
void SquareBlockMatrix<double>::free(); |
| 331 |
template |
| 332 |
void SquareBlockMatrix<quadruple>::free(); |
| 333 |
template |
| 334 |
void SquareBlockMatrix<complex<double> >::free(); |
| 335 |
template |
| 336 |
void SquareBlockMatrix<complex<quadruple> >::free(); |
| 337 |
template |
| 338 |
void SquareBlockMatrix<float>::free(); |
| 339 |
template |
| 340 |
void SquareBlockMatrix<complex<float> >::free(); |
| 341 |
|
| 342 |
// |
| 343 |
|
| 344 |
template<typename T> |
| 345 |
T* SquareBlockMatrix<T>::addrCoefBlock(int i, int j) |
| 346 |
{ |
| 347 |
if (_isdecomposed) { // is necessary? |
| 348 |
if (_isSym && (i > j)) { |
| 349 |
return _coefs[j + i * _num_blocks]; |
| 350 |
} |
| 351 |
} |
| 352 |
return _coefs[i + j * _num_blocks]; |
| 353 |
} |
| 354 |
|
| 355 |
template |
| 356 |
double* SquareBlockMatrix<double>::addrCoefBlock(int i, int j); |
| 357 |
|
| 358 |
template |
| 359 |
quadruple* SquareBlockMatrix<quadruple>::addrCoefBlock(int i, int j); |
| 360 |
|
| 361 |
template |
| 362 |
complex<double>* SquareBlockMatrix<complex<double> >:: |
| 363 |
addrCoefBlock(int i, int j); |
| 364 |
|
| 365 |
template |
| 366 |
complex<quadruple>* SquareBlockMatrix<complex<quadruple> >:: |
| 367 |
addrCoefBlock(int i, int j); |
| 368 |
|
| 369 |
template |
| 370 |
float* SquareBlockMatrix<float>::addrCoefBlock(int i, int j); |
| 371 |
|
| 372 |
template |
| 373 |
complex<float>* SquareBlockMatrix<complex<float> >:: |
| 374 |
addrCoefBlock(int i, int j); |
| 375 |
|
| 376 |
// |
| 377 |
|
| 378 |
template<typename T> |
| 379 |
void SquareBlockMatrix<T>::ZeroClear() |
| 380 |
{ |
| 381 |
const T zero(0.0); |
| 382 |
if (_isSym) { |
| 383 |
for (int i = 0; i < _num_blocks; i++) { |
| 384 |
for (int j = i; j < _num_blocks; j++) { |
| 385 |
const int nrow = _block_sizes[i]; |
| 386 |
const int ncol = _block_sizes[j]; |
| 387 |
const int nsize = nrow * ncol; |
| 388 |
const int itmp = i + j * _num_blocks; |
| 389 |
for (int k = 0; k < nsize; k++) { |
| 390 |
_coefs[itmp][k] = zero; // needs to be replaced by memcpy |
| 391 |
} |
| 392 |
} |
| 393 |
} |
| 394 |
} |
| 395 |
else { |
| 396 |
for (int i = 0; i < _num_blocks; i++) { |
| 397 |
for (int j = 0; j < _num_blocks; j++) { |
| 398 |
const int nrow = _block_sizes[i]; |
| 399 |
const int ncol = _block_sizes[j]; |
| 400 |
const int nsize = nrow * ncol; |
| 401 |
const int itmp = i + j * _num_blocks; |
| 402 |
for (int k = 0; k < nsize; k++) { |
| 403 |
_coefs[itmp][k] = zero; // needs to be replaced by memcpy |
| 404 |
} |
| 405 |
} |
| 406 |
} |
| 407 |
} |
| 408 |
} |
| 409 |
|
| 410 |
template |
| 411 |
void SquareBlockMatrix<double>::ZeroClear(); |
| 412 |
|
| 413 |
template |
| 414 |
void SquareBlockMatrix<quadruple>::ZeroClear(); |
| 415 |
|
| 416 |
template |
| 417 |
void SquareBlockMatrix<complex<double> >::ZeroClear(); |
| 418 |
|
| 419 |
template |
| 420 |
void SquareBlockMatrix<complex<quadruple> >::ZeroClear(); |
| 421 |
|
| 422 |
template |
| 423 |
void SquareBlockMatrix<float>::ZeroClear(); |
| 424 |
|
| 425 |
template |
| 426 |
void SquareBlockMatrix<complex<float> >::ZeroClear(); |
| 427 |
|
| 428 |
// |
| 429 |
|
| 430 |
template<typename T> |
| 431 |
T& SquareBlockMatrix<T>::diag(int i) |
| 432 |
{ |
| 433 |
int i0, i1; |
| 434 |
if(_isdecomposed) { |
| 435 |
fprintf(stderr, "%s %d : operator() is not suppoesd to be used\n", |
| 436 |
__FILE__, __LINE__); |
| 437 |
i0 = BlockIndex(i); |
| 438 |
i1 = BlockOffset(i); |
| 439 |
} |
| 440 |
else { |
| 441 |
i0 = i / _block_size; |
| 442 |
i1 = i % _block_size; |
| 443 |
} |
| 444 |
// const int itmp = offsetBlock(i0, i0) + i1 * (nrowBlock(i0, i0) + 1); |
| 445 |
// return coefs()[itmp]; |
| 446 |
return _coefs[i0 * (_num_blocks + 1)][i1 * (nrowBlock(i0, i0) + 1)]; |
| 447 |
} |
| 448 |
|
| 449 |
template |
| 450 |
double& SquareBlockMatrix<double>::diag(int i); |
| 451 |
|
| 452 |
template |
| 453 |
quadruple& SquareBlockMatrix<quadruple>::diag(int i); |
| 454 |
|
| 455 |
template |
| 456 |
complex<double> & SquareBlockMatrix<complex<double> >::diag(int i); |
| 457 |
|
| 458 |
template |
| 459 |
complex<quadruple> & SquareBlockMatrix<complex<quadruple> >::diag(int i); |
| 460 |
|
| 461 |
template |
| 462 |
float& SquareBlockMatrix<float>::diag(int i); |
| 463 |
|
| 464 |
template |
| 465 |
complex<float> & SquareBlockMatrix<complex<float> >::diag(int i); |
| 466 |
// |
| 467 |
|
| 468 |
template<typename T> |
| 469 |
const T& SquareBlockMatrix<T>::diag(int i) const |
| 470 |
{ |
| 471 |
int i0, i1; |
| 472 |
if(_isdecomposed) { |
| 473 |
fprintf(stderr, "%s %d : operator() is not suppoesd to be used\n", |
| 474 |
__FILE__, __LINE__); |
| 475 |
i0 = BlockIndex(i); |
| 476 |
i1 = BlockOffset(i); |
| 477 |
} |
| 478 |
else { |
| 479 |
i0 = i / _block_size; |
| 480 |
i1 = i % _block_size; |
| 481 |
} |
| 482 |
return _coefs[i0 * (_num_blocks + 1)][i1 * (nrowBlock(i0, i0) + 1)]; |
| 483 |
} |
| 484 |
|
| 485 |
template |
| 486 |
const double & SquareBlockMatrix<double>::diag(int i) const; |
| 487 |
|
| 488 |
template |
| 489 |
const quadruple & SquareBlockMatrix<quadruple>::diag(int i) const; |
| 490 |
|
| 491 |
template |
| 492 |
const complex<double> & SquareBlockMatrix<complex<double> >::diag(int i) const; |
| 493 |
|
| 494 |
template |
| 495 |
const complex<quadruple> & SquareBlockMatrix<complex<quadruple> >:: |
| 496 |
diag(int i) const; |
| 497 |
|
| 498 |
template |
| 499 |
const float & SquareBlockMatrix<float>::diag(int i) const; |
| 500 |
|
| 501 |
template |
| 502 |
const complex<float> & SquareBlockMatrix<complex<float> >::diag(int i) const; |
| 503 |
// |
| 504 |
|
| 505 |
template<typename T> |
| 506 |
T& SquareBlockMatrix<T>::operator () (int i, int j) |
| 507 |
{ |
| 508 |
int i0, i1, j0, j1; |
| 509 |
if(_isdecomposed) { |
| 510 |
fprintf(stderr, "%s %d : operator() is not suppoesd to be used\n", |
| 511 |
__FILE__, __LINE__); |
| 512 |
if (_isSym) { |
| 513 |
if (i < j) { |
| 514 |
i0 = BlockIndex(i); |
| 515 |
j0 = BlockIndex(j); |
| 516 |
i1 = BlockOffset(i); |
| 517 |
j1 = BlockOffset(j); |
| 518 |
} |
| 519 |
else { |
| 520 |
i0 = BlockIndex(j); |
| 521 |
j0 = BlockIndex(i); |
| 522 |
i1 = BlockOffset(j); |
| 523 |
j1 = BlockOffset(i); |
| 524 |
} |
| 525 |
} |
| 526 |
else { |
| 527 |
i0 = BlockIndex(i); |
| 528 |
j0 = BlockIndex(j); |
| 529 |
if(i0 > j0) { |
| 530 |
i1 = BlockOffset(j); |
| 531 |
j1 = BlockOffset(i); |
| 532 |
} |
| 533 |
else { |
| 534 |
i1 = BlockOffset(i); |
| 535 |
j1 = BlockOffset(j); |
| 536 |
} |
| 537 |
} |
| 538 |
} // if(_isdecomposed) { |
| 539 |
else { |
| 540 |
if (_isSym) { |
| 541 |
if (i < j) { |
| 542 |
i0 = i / _block_size; |
| 543 |
i1 = i % _block_size; |
| 544 |
j0 = j / _block_size; |
| 545 |
j1 = j % _block_size; |
| 546 |
} |
| 547 |
else { |
| 548 |
i0 = j / _block_size; |
| 549 |
i1 = j % _block_size; |
| 550 |
j0 = i / _block_size; |
| 551 |
j1 = i % _block_size; |
| 552 |
} |
| 553 |
} |
| 554 |
else { |
| 555 |
i0 = i / _block_size; |
| 556 |
j0 = j / _block_size; |
| 557 |
if (i0 > j0) { |
| 558 |
i1 = j % _block_size; |
| 559 |
j1 = i % _block_size; |
| 560 |
} |
| 561 |
else { |
| 562 |
i1 = i % _block_size; |
| 563 |
j1 = j % _block_size; |
| 564 |
} |
| 565 |
} |
| 566 |
} // if(_isdecomposed) else |
| 567 |
return _coefs[i0 + j0 * _num_blocks][i1 + j1 * nrowBlock(i0, j0)]; |
| 568 |
} |
| 569 |
|
| 570 |
template |
| 571 |
double& SquareBlockMatrix<double>::operator () (int i, int j); |
| 572 |
|
| 573 |
template |
| 574 |
quadruple& SquareBlockMatrix<quadruple>::operator () (int i, int j); |
| 575 |
|
| 576 |
template |
| 577 |
complex<double> & SquareBlockMatrix<complex<double> >:: |
| 578 |
operator () (int i, int j); |
| 579 |
|
| 580 |
template |
| 581 |
complex<quadruple> & SquareBlockMatrix<complex<quadruple> >:: |
| 582 |
operator () (int i, int j); |
| 583 |
|
| 584 |
template |
| 585 |
float& SquareBlockMatrix<float>::operator () (int i, int j); |
| 586 |
|
| 587 |
template |
| 588 |
complex<float> & SquareBlockMatrix<complex<float> >:: |
| 589 |
operator () (int i, int j); |
| 590 |
// |
| 591 |
|
| 592 |
template<typename T> |
| 593 |
const T& SquareBlockMatrix<T>::operator () (int i, int j) const |
| 594 |
{ |
| 595 |
int i0, i1, j0, j1; |
| 596 |
if(_isdecomposed) { |
| 597 |
fprintf(stderr, "%s %d : operator() is not suppoesd to be used\n", |
| 598 |
__FILE__, __LINE__); |
| 599 |
if (_isSym) { |
| 600 |
if (i < j) { |
| 601 |
i0 = BlockIndex(i); |
| 602 |
j0 = BlockIndex(j); |
| 603 |
i1 = BlockOffset(i); |
| 604 |
j1 = BlockOffset(j); |
| 605 |
} |
| 606 |
else { |
| 607 |
i0 = BlockIndex(j); |
| 608 |
j0 = BlockIndex(i); |
| 609 |
i1 = BlockOffset(j); |
| 610 |
j1 = BlockOffset(i); |
| 611 |
} |
| 612 |
} |
| 613 |
else { |
| 614 |
i0 = BlockIndex(i); |
| 615 |
j0 = BlockIndex(j); |
| 616 |
if(i0 > j0) { |
| 617 |
i1 = BlockOffset(j); |
| 618 |
j1 = BlockOffset(i); |
| 619 |
} |
| 620 |
else { |
| 621 |
i1 = BlockOffset(i); |
| 622 |
j1 = BlockOffset(j); |
| 623 |
} |
| 624 |
} |
| 625 |
} // if(_isdecomposed) { |
| 626 |
else { |
| 627 |
if (_isSym) { |
| 628 |
if (i < j) { |
| 629 |
i0 = i / _block_size; |
| 630 |
i1 = i % _block_size; |
| 631 |
j0 = j / _block_size; |
| 632 |
j1 = j % _block_size; |
| 633 |
} |
| 634 |
else { |
| 635 |
i0 = j / _block_size; |
| 636 |
i1 = j % _block_size; |
| 637 |
j0 = i / _block_size; |
| 638 |
j1 = i % _block_size; |
| 639 |
} |
| 640 |
} |
| 641 |
else { |
| 642 |
i0 = i / _block_size; |
| 643 |
j0 = j / _block_size; |
| 644 |
if (i0 > j0) { |
| 645 |
i1 = j % _block_size; |
| 646 |
j1 = i % _block_size; |
| 647 |
} |
| 648 |
else { |
| 649 |
i1 = i % _block_size; |
| 650 |
j1 = j % _block_size; |
| 651 |
} |
| 652 |
} |
| 653 |
} // if(_isdecomposed) else |
| 654 |
return _coefs[i0 + j0 * _num_blocks][i1 + j1 * nrowBlock(i0, j0)]; |
| 655 |
} |
| 656 |
|
| 657 |
template |
| 658 |
const double& SquareBlockMatrix<double>::operator () (int i, int j) const; |
| 659 |
|
| 660 |
template |
| 661 |
const quadruple& SquareBlockMatrix<quadruple>::operator () (int i, int j) const; |
| 662 |
|
| 663 |
template |
| 664 |
const complex<double>& SquareBlockMatrix<complex<double> >:: |
| 665 |
operator () (int i, int j) const; |
| 666 |
|
| 667 |
template |
| 668 |
const complex<quadruple>& SquareBlockMatrix<complex<quadruple> >:: |
| 669 |
operator () (int i, int j) const; |
| 670 |
|
| 671 |
template |
| 672 |
const float& SquareBlockMatrix<float>::operator () (int i, int j) const; |
| 673 |
|
| 674 |
template |
| 675 |
const complex<float>& SquareBlockMatrix<complex<float> >:: |
| 676 |
operator () (int i, int j) const; |
| 677 |
// |
| 678 |
|
| 679 |
template<typename T> |
| 680 |
void SquareBlockMatrix<T>::copyFromArray(const T *a, int dim) |
| 681 |
{ |
| 682 |
if (_isdecomposed) { |
| 683 |
fprintf(stderr, "%s %d : copyFromArray() cannot copy decompesd data\n", |
| 684 |
__FILE__, __LINE__); |
| 685 |
return; |
| 686 |
} |
| 687 |
if (dim != _dim) { |
| 688 |
fprintf(stderr, "%s %d : mismatched dim %d != %d\n", |
| 689 |
__FILE__, __LINE__, dim, _dim); |
| 690 |
return; |
| 691 |
} |
| 692 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 693 |
const int n_row = ((ki == (_num_blocks - 1)) ? |
| 694 |
_block_size_last : _block_size); |
| 695 |
const int dst_row = n_row; |
| 696 |
for (int kj = ki; kj < _num_blocks; kj++) { |
| 697 |
const int n_col = ((kj == (_num_blocks - 1)) ? |
| 698 |
_block_size_last : _block_size); |
| 699 |
T *a_upper = addrCoefBlock(ki, kj); |
| 700 |
int j, j0, j1; |
| 701 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 702 |
j < n_col; j++, j1 += _dim, j0 += dst_row) { |
| 703 |
blas_copy(n_row, (T *)a + j1, 1, a_upper + j0, 1); |
| 704 |
} |
| 705 |
} // loop : kj |
| 706 |
} // loop : ki |
| 707 |
if (!_isSym) { |
| 708 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 709 |
const int n_row = (ki == (_num_blocks - 1)) ? |
| 710 |
_block_size_last : _block_size; |
| 711 |
for (int kj = 0; kj < ki; kj++) { |
| 712 |
const int n_col = _block_size; |
| 713 |
T *a_lower = addrCoefBlock(ki, kj); |
| 714 |
int dst_row = nrowBlock(ki, kj); |
| 715 |
int j, j0, j1; |
| 716 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 717 |
j < n_col; j++, j1 += _dim, j0++) { |
| 718 |
blas_copy(n_row, (T *)a + j1, 1, a_lower + j0, dst_row); |
| 719 |
} |
| 720 |
} // loop : kj |
| 721 |
} // loop : ki |
| 722 |
} // if (!isSym) |
| 723 |
} |
| 724 |
|
| 725 |
template |
| 726 |
void SquareBlockMatrix<double>::copyFromArray(const double *a, int dim); |
| 727 |
|
| 728 |
template |
| 729 |
void SquareBlockMatrix<quadruple>::copyFromArray(const quadruple *a, int dim); |
| 730 |
|
| 731 |
template |
| 732 |
void SquareBlockMatrix<complex<double> >:: |
| 733 |
copyFromArray(const complex<double> *a, int dim); |
| 734 |
|
| 735 |
template |
| 736 |
void SquareBlockMatrix<complex<quadruple> >:: |
| 737 |
copyFromArray(const complex<quadruple> *a, int dim); |
| 738 |
|
| 739 |
template |
| 740 |
void SquareBlockMatrix<float>::copyFromArray(const float *a, int dim); |
| 741 |
|
| 742 |
template |
| 743 |
void SquareBlockMatrix<complex<float> >:: |
| 744 |
copyFromArray(const complex<float> *a, int dim); |
| 745 |
// |
| 746 |
|
| 747 |
template<typename T> |
| 748 |
void SquareBlockMatrix<T>::copyFromArrayPermute(const T *a, int dim, |
| 749 |
int *permute) |
| 750 |
{ |
| 751 |
if (_isdecomposed) { |
| 752 |
fprintf(stderr, "%s %d : copyFromArray() cannot copy decompesd data\n", |
| 753 |
__FILE__, __LINE__); |
| 754 |
return; |
| 755 |
} |
| 756 |
if (dim != _dim) { |
| 757 |
fprintf(stderr, "%s %d : mismatched dim %d != %d\n", |
| 758 |
__FILE__, __LINE__, dim, _dim); |
| 759 |
return; |
| 760 |
} |
| 761 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 762 |
const int n_row = ((ki == (_num_blocks - 1)) ? |
| 763 |
_block_size_last : _block_size); |
| 764 |
const int dst_row = n_row; |
| 765 |
for (int kj = ki; kj < _num_blocks; kj++) { |
| 766 |
const int n_col = ((kj == (_num_blocks - 1)) ? |
| 767 |
_block_size_last : _block_size); |
| 768 |
T *a_upper = addrCoefBlock(ki, kj); |
| 769 |
int j, j0, j1; |
| 770 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 771 |
j < n_col; j++, j1 += _dim, j0 += dst_row) { |
| 772 |
const int jj = permute[kj * _block_size + j]; |
| 773 |
int ii = ki * _block_size; |
| 774 |
int kk = j0; |
| 775 |
if (!_isSym) { |
| 776 |
for (int i = 0; i < n_row; i++, kk++, ii++) { |
| 777 |
a_upper[kk] = a[permute[ii] + jj * _dim]; |
| 778 |
} |
| 779 |
} |
| 780 |
else { |
| 781 |
for (int i = 0; i < n_row; i++, kk++, ii++) { |
| 782 |
const int itmp = permute[ii]; |
| 783 |
a_upper[kk] = a[itmp <= jj ? itmp + jj * _dim : jj + itmp * _dim]; |
| 784 |
} |
| 785 |
} // if (!isSym) { |
| 786 |
// blas_copy(n_row, (T *)a + j1, 1, a_upper + j0, 1); |
| 787 |
} |
| 788 |
} // loop : kj |
| 789 |
} // loop : ki |
| 790 |
if (!_isSym) { |
| 791 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 792 |
const int n_row = ((ki == (_num_blocks - 1)) ? |
| 793 |
_block_size_last : _block_size); |
| 794 |
for (int kj = 0; kj < ki; kj++) { |
| 795 |
const int n_col = _block_size; |
| 796 |
T *a_lower = addrCoefBlock(ki, kj); |
| 797 |
const int dst_row = nrowBlock(ki, kj); |
| 798 |
int j, j0, j1; |
| 799 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 800 |
j < n_col; j++, j1 += _dim, j0++) { |
| 801 |
const int jj = permute[kj * _block_size + j]; |
| 802 |
int ii = ki * _block_size; |
| 803 |
int kk = j0; |
| 804 |
for (int i = 0; i < n_row; i++, ii++, kk += dst_row) { |
| 805 |
a_lower[kk] = a[permute[ii] + jj * _dim]; |
| 806 |
} |
| 807 |
// blas_copy(n_row, (T *)a + j1, 1, a_lower + j0, dst_row); |
| 808 |
} |
| 809 |
} // loop : kj |
| 810 |
} // loop : ki |
| 811 |
} // if (!isSym) |
| 812 |
} |
| 813 |
|
| 814 |
template |
| 815 |
void SquareBlockMatrix<double>::copyFromArrayPermute(const double *a, int dim, |
| 816 |
int *permute); |
| 817 |
|
| 818 |
template |
| 819 |
void SquareBlockMatrix<quadruple>::copyFromArrayPermute(const quadruple *a, |
| 820 |
int dim, |
| 821 |
int *permute); |
| 822 |
|
| 823 |
template |
| 824 |
void SquareBlockMatrix<complex<double> >:: |
| 825 |
copyFromArrayPermute(const complex<double> *a, int dim, |
| 826 |
int *permute); |
| 827 |
|
| 828 |
template |
| 829 |
void SquareBlockMatrix<complex<quadruple> >:: |
| 830 |
copyFromArrayPermute(const complex<quadruple> *a, int dim, |
| 831 |
int *permute); |
| 832 |
template |
| 833 |
void SquareBlockMatrix<float>::copyFromArrayPermute(const float *a, int dim, |
| 834 |
int *permute); |
| 835 |
template |
| 836 |
void SquareBlockMatrix<complex<float> >:: |
| 837 |
copyFromArrayPermute(const complex<float> *a, int dim, |
| 838 |
int *permute); |
| 839 |
|
| 840 |
// |
| 841 |
|
| 842 |
template<typename T> |
| 843 |
void SquareBlockMatrix<T>::copyToArrayFull(const T *a, int dim) |
| 844 |
{ |
| 845 |
if (dim != _dim) { |
| 846 |
fprintf(stderr, "%s %d : mismatched dim %d != %d\n", |
| 847 |
__FILE__, __LINE__, dim, _dim); |
| 848 |
return; |
| 849 |
} |
| 850 |
// copy upper part |
| 851 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 852 |
const int n_row = ((ki == (_num_blocks - 1)) ? |
| 853 |
_block_size_last : _block_size); |
| 854 |
const int dst_row = n_row; |
| 855 |
for (int kj = ki; kj < _num_blocks; kj++) { |
| 856 |
const int n_col = ((kj == (_num_blocks - 1)) ? |
| 857 |
_block_size_last : _block_size); |
| 858 |
T *a_upper = addrCoefBlock(ki, kj); |
| 859 |
int j, j0, j1; |
| 860 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 861 |
j < n_col; j++, j1 += _dim, j0 += dst_row) { |
| 862 |
blas_copy(n_row, a_upper + j0, 1, (T *)a + j1, 1); |
| 863 |
} |
| 864 |
} // loop : kj |
| 865 |
} // loop : ki |
| 866 |
// copy lower part |
| 867 |
if (_isSym) { |
| 868 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 869 |
const int n_row = ((ki == (_num_blocks - 1)) ? |
| 870 |
_block_size_last : _block_size); |
| 871 |
for (int kj = 0; kj < ki; kj++) { |
| 872 |
const int n_col = _block_size; |
| 873 |
const int dst_col = n_col; |
| 874 |
T *a_upper = addrCoefBlock(kj, ki); // transposed access |
| 875 |
int j, j0, j1; |
| 876 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 877 |
j < n_row; j++, j1++, j0 += dst_col) { |
| 878 |
blas_copy(n_col, a_upper + j0, 1, (T *)a + j1, _dim); |
| 879 |
} |
| 880 |
} // loop : kj |
| 881 |
} // loop : ki |
| 882 |
} |
| 883 |
else { |
| 884 |
for (int ki = 0; ki < _num_blocks; ki++) { |
| 885 |
const int n_row = ((ki == (_num_blocks - 1)) ? |
| 886 |
_block_size_last : _block_size); |
| 887 |
for (int kj = 0; kj < ki; kj++) { |
| 888 |
const int n_col = _block_size; |
| 889 |
T *a_upper = addrCoefBlock(ki, kj); |
| 890 |
const int src_row = nrowBlock(ki, kj); |
| 891 |
int j, j0, j1; |
| 892 |
for (j1 = (ki + kj * _dim) * _block_size, j0 = 0, j = 0; |
| 893 |
j < n_col; j++, j1 += _dim, j0++) { |
| 894 |
blas_copy(n_row, a_upper + j0, src_row, (T *)a + j1, 1); |
| 895 |
} |
| 896 |
} // loop : kj |
| 897 |
} // loop : ki |
| 898 |
} |
| 899 |
} |
| 900 |
|
| 901 |
template |
| 902 |
void SquareBlockMatrix<double>::copyToArrayFull(const double *a, int dim); |
| 903 |
|
| 904 |
template |
| 905 |
void SquareBlockMatrix<quadruple>::copyToArrayFull(const quadruple *a, int dim); |
| 906 |
|
| 907 |
template |
| 908 |
void SquareBlockMatrix<complex<double> >:: |
| 909 |
copyToArrayFull(const complex<double> *a, int dim); |
| 910 |
|
| 911 |
template |
| 912 |
void SquareBlockMatrix<complex<quadruple> >:: |
| 913 |
copyToArrayFull(const complex<quadruple> *a, int dim); |
| 914 |
|
| 915 |
template |
| 916 |
void SquareBlockMatrix<float>::copyToArrayFull(const float *a, int dim); |
| 917 |
|
| 918 |
template |
| 919 |
void SquareBlockMatrix<complex<float> >:: |
| 920 |
copyToArrayFull(const complex<float> *a, int dim); |
| 921 |
// |
| 922 |
|
| 923 |
template<typename T> |
| 924 |
void SquareBlockMatrix<T>::copyBlockToArray(int i, int j, const T *a, int dim) |
| 925 |
{ |
| 926 |
if (dim != _dim) { |
| 927 |
fprintf(stderr, "%s %d : mismatched dim %d != %d\n", |
| 928 |
__FILE__, __LINE__, dim, _dim); |
| 929 |
return; |
| 930 |
} |
| 931 |
if (_isSym && i > j) { |
| 932 |
fprintf(stderr, |
| 933 |
"%s %d : symmetric matrix has upper block only : %d > %d\n", |
| 934 |
__FILE__, __LINE__, i, j); |
| 935 |
return; |
| 936 |
} |
| 937 |
T *a_src = addrCoefBlock(i, j); |
| 938 |
int j0, j1, j2; |
| 939 |
if (i <= j) { |
| 940 |
const int n_row = ((i == (_num_blocks - 1)) ? |
| 941 |
_block_size_last : _block_size); |
| 942 |
const int dst_row = n_row; |
| 943 |
const int n_col = ((j == (_num_blocks - 1)) ? |
| 944 |
_block_size_last : _block_size); |
| 945 |
for (j2 = (i + j * _dim) * _block_size, j1 = 0, j0 = 0; |
| 946 |
j0 < n_col; j0++, j1 += dst_row, j2 += _dim) { |
| 947 |
blas_copy(n_row, a_src + j1, 1, (T *)a + j2, 1); |
| 948 |
} |
| 949 |
} |
| 950 |
else { |
| 951 |
const int n_row = ((i == (_num_blocks - 1)) ? |
| 952 |
_block_size_last : _block_size); |
| 953 |
const int n_col = ((j == (_num_blocks - 1)) ? |
| 954 |
_block_size_last : _block_size); |
| 955 |
const int dst_col = n_col; |
| 956 |
for (j2 = (i + j * _dim) * _block_size, j1 = 0, j0 = 0; |
| 957 |
j0 < n_col; j0++, j1++, j2 += _dim) { |
| 958 |
blas_copy(n_row, a_src + j1, dst_col, (T *)a + j2, 1); |
| 959 |
} |
| 960 |
} |
| 961 |
} |
| 962 |
|
| 963 |
template |
| 964 |
void SquareBlockMatrix<double>::copyBlockToArray(int i, int j, |
| 965 |
const double *a, int dim); |
| 966 |
|
| 967 |
template |
| 968 |
void SquareBlockMatrix<quadruple>::copyBlockToArray(int i, int j, |
| 969 |
const quadruple *a, |
| 970 |
int dim); |
| 971 |
template |
| 972 |
void SquareBlockMatrix<complex<double> >:: |
| 973 |
copyBlockToArray(int i, int j, const complex<double> *a, int dim); |
| 974 |
|
| 975 |
template |
| 976 |
void SquareBlockMatrix<complex<quadruple> >:: |
| 977 |
copyBlockToArray(int i, int j, const complex<quadruple> *a, int dim); |
| 978 |
|
| 979 |
template |
| 980 |
void SquareBlockMatrix<float>::copyBlockToArray(int i, int j, |
| 981 |
const float *a, int dim); |
| 982 |
|
| 983 |
template |
| 984 |
void SquareBlockMatrix<complex<float> >:: |
| 985 |
copyBlockToArray(int i, int j, const complex<float> *a, int dim); |