| 1 |
// SPDX-FileCopyrightText: Copyright © DUNE Project contributors, see file LICENSE.md in module root |
| 2 |
// SPDX-License-Identifier: LicenseRef-GPL-2.0-only-with-DUNE-exception |
| 3 |
// -*- tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*- |
| 4 |
// vi: set et ts=4 sw=2 sts=2: |
| 5 |
#ifndef DUNE_ISTL_MATRIX_HH |
| 6 |
#define DUNE_ISTL_MATRIX_HH |
| 7 |
|
| 8 |
/** \file |
| 9 |
\brief A dynamic dense block matrix class |
| 10 |
*/ |
| 11 |
|
| 12 |
#include <cmath> |
| 13 |
#include <memory> |
| 14 |
|
| 15 |
#include <dune/common/ftraits.hh> |
| 16 |
#include <dune/common/typetraits.hh> |
| 17 |
#include <dune/common/scalarvectorview.hh> |
| 18 |
#include <dune/common/scalarmatrixview.hh> |
| 19 |
|
| 20 |
#include <dune/istl/bvector.hh> |
| 21 |
#include <dune/istl/istlexception.hh> |
| 22 |
#include <dune/istl/blocklevel.hh> |
| 23 |
|
| 24 |
namespace Dune { |
| 25 |
|
| 26 |
namespace MatrixImp |
| 27 |
{ |
| 28 |
/** |
| 29 |
\brief A Vector of blocks with different blocksizes. |
| 30 |
|
| 31 |
This class started as a copy of VariableBlockVector, which used to be used for the internal memory managerment |
| 32 |
of the 'Matrix' class. However, that mechanism stopped working when I started using the RandomAccessIteratorFacade |
| 33 |
in VariableBlockVector (308dd85483108f8baaa4051251e2c75e2a9aed32, to make VariableBlockVector pass a number of |
| 34 |
tightened interface compliance tests), and I couldn't quite figure out how to fix that. However, using |
| 35 |
VariableBlockVector in Matrix internally was a hack anyway, so I simply took the working version of VariableBlockVector |
| 36 |
and copied it here under the new name of DenseMatrixBase. This is still hacky, but one step closer to an |
| 37 |
elegant solution. |
| 38 |
*/ |
| 39 |
template<class B, class A=std::allocator<B> > |
| 40 |
class DenseMatrixBase : public Imp::block_vector_unmanaged<B,typename A::size_type> |
| 41 |
// this derivation gives us all the blas level 1 and norms |
| 42 |
// on the large array. However, access operators have to be |
| 43 |
// overwritten. |
| 44 |
{ |
| 45 |
public: |
| 46 |
|
| 47 |
//===== type definitions and constants |
| 48 |
|
| 49 |
//! export the type representing the field |
| 50 |
using field_type = typename Imp::BlockTraits<B>::field_type; |
| 51 |
|
| 52 |
//! export the allocator type |
| 53 |
typedef A allocator_type; |
| 54 |
|
| 55 |
//! The size type for the index access |
| 56 |
typedef typename A::size_type size_type; |
| 57 |
|
| 58 |
/** \brief Type of the elements of the outer vector, i.e., dynamic vectors of B |
| 59 |
* |
| 60 |
* Note that this is *not* the type referred to by the iterators and random access operators, |
| 61 |
* which return proxy objects. |
| 62 |
*/ |
| 63 |
typedef BlockVector<B,A> value_type; |
| 64 |
|
| 65 |
/** \brief Same as value_type, here for historical reasons |
| 66 |
*/ |
| 67 |
typedef BlockVector<B,A> block_type; |
| 68 |
|
| 69 |
// just a shorthand |
| 70 |
typedef Imp::BlockVectorWindow<B,A> window_type; |
| 71 |
|
| 72 |
typedef window_type reference; |
| 73 |
|
| 74 |
typedef const window_type const_reference; |
| 75 |
|
| 76 |
|
| 77 |
//===== constructors and such |
| 78 |
|
| 79 |
/** constructor without arguments makes empty vector, |
| 80 |
object cannot be used yet |
| 81 |
*/ |
| 82 |
DenseMatrixBase () : Imp::block_vector_unmanaged<B,size_type>() |
| 83 |
{ |
| 84 |
// nothing is known ... |
| 85 |
rows_ = 0; |
| 86 |
columns_ = 0; |
| 87 |
} |
| 88 |
|
| 89 |
/** make vector with given number of blocks each having a constant size, |
| 90 |
object is fully usable then. |
| 91 |
|
| 92 |
* \param rows Number of rows |
| 93 |
* \param columns Number of columns |
| 94 |
*/ |
| 95 |
DenseMatrixBase (size_type rows, size_type columns) : Imp::block_vector_unmanaged<B,size_type>() |
| 96 |
{ |
| 97 |
// and we can allocate the big array in the base class |
| 98 |
this->n = rows*columns; |
| 99 |
columns_ = columns; |
| 100 |
if (this->n>0) |
| 101 |
{ |
| 102 |
this->p = allocator_.allocate(this->n); |
| 103 |
new (this->p)B[this->n]; |
| 104 |
} |
| 105 |
else |
| 106 |
{ |
| 107 |
this->n = 0; |
| 108 |
this->p = 0; |
| 109 |
} |
| 110 |
|
| 111 |
// we can allocate the windows now |
| 112 |
rows_ = rows; |
| 113 |
} |
| 114 |
|
| 115 |
//! copy constructor, has copy semantics |
| 116 |
DenseMatrixBase (const DenseMatrixBase& a) |
| 117 |
{ |
| 118 |
// allocate the big array in the base class |
| 119 |
this->n = a.n; |
| 120 |
columns_ = a.columns_; |
| 121 |
if (this->n>0) |
| 122 |
{ |
| 123 |
// allocate and construct objects |
| 124 |
this->p = allocator_.allocate(this->n); |
| 125 |
new (this->p)B[this->n]; |
| 126 |
|
| 127 |
// copy data |
| 128 |
for (size_type i=0; i<this->n; i++) |
| 129 |
this->p[i]=a.p[i]; |
| 130 |
} |
| 131 |
else |
| 132 |
{ |
| 133 |
this->n = 0; |
| 134 |
this->p = nullptr; |
| 135 |
} |
| 136 |
|
| 137 |
// we can allocate the windows now |
| 138 |
rows_ = a.rows_; |
| 139 |
} |
| 140 |
|
| 141 |
//! free dynamic memory |
| 142 |
~DenseMatrixBase () |
| 143 |
{ |
| 144 |
if (this->n>0) { |
| 145 |
size_type i=this->n; |
| 146 |
while (i) |
| 147 |
this->p[--i].~B(); |
| 148 |
allocator_.deallocate(this->p,this->n); |
| 149 |
} |
| 150 |
} |
| 151 |
|
| 152 |
//! same effect as constructor with same argument |
| 153 |
void resize (size_type rows, size_type columns) |
| 154 |
{ |
| 155 |
// deconstruct objects and deallocate memory if necessary |
| 156 |
if (this->n>0) { |
| 157 |
size_type i=this->n; |
| 158 |
while (i) |
| 159 |
this->p[--i].~B(); |
| 160 |
allocator_.deallocate(this->p,this->n); |
| 161 |
} |
| 162 |
|
| 163 |
// and we can allocate the big array in the base class |
| 164 |
this->n = rows*columns; |
| 165 |
if (this->n>0) |
| 166 |
{ |
| 167 |
this->p = allocator_.allocate(this->n); |
| 168 |
new (this->p)B[this->n]; |
| 169 |
} |
| 170 |
else |
| 171 |
{ |
| 172 |
this->n = 0; |
| 173 |
this->p = nullptr; |
| 174 |
} |
| 175 |
|
| 176 |
// we can allocate the windows now |
| 177 |
rows_ = rows; |
| 178 |
columns_ = columns; |
| 179 |
} |
| 180 |
|
| 181 |
//! assignment |
| 182 |
DenseMatrixBase& operator= (const DenseMatrixBase& a) |
| 183 |
{ |
| 184 |
if (&a!=this) // check if this and a are different objects |
| 185 |
{ |
| 186 |
columns_ = a.columns_; |
| 187 |
// reallocate arrays if necessary |
| 188 |
// Note: still the block sizes may vary ! |
| 189 |
if (this->n!=a.n || rows_!=a.rows_) |
| 190 |
{ |
| 191 |
// deconstruct objects and deallocate memory if necessary |
| 192 |
if (this->n>0) { |
| 193 |
size_type i=this->n; |
| 194 |
while (i) |
| 195 |
this->p[--i].~B(); |
| 196 |
allocator_.deallocate(this->p,this->n); |
| 197 |
} |
| 198 |
|
| 199 |
// allocate the big array in the base class |
| 200 |
this->n = a.n; |
| 201 |
if (this->n>0) |
| 202 |
{ |
| 203 |
// allocate and construct objects |
| 204 |
this->p = allocator_.allocate(this->n); |
| 205 |
new (this->p)B[this->n]; |
| 206 |
} |
| 207 |
else |
| 208 |
{ |
| 209 |
this->n = 0; |
| 210 |
this->p = nullptr; |
| 211 |
} |
| 212 |
|
| 213 |
// Copy number of rows |
| 214 |
rows_ = a.rows_; |
| 215 |
} |
| 216 |
|
| 217 |
// and copy the data |
| 218 |
for (size_type i=0; i<this->n; i++) |
| 219 |
this->p[i]=a.p[i]; |
| 220 |
} |
| 221 |
|
| 222 |
return *this; |
| 223 |
} |
| 224 |
|
| 225 |
|
| 226 |
//===== assignment from scalar |
| 227 |
|
| 228 |
//! assign from scalar |
| 229 |
DenseMatrixBase& operator= (const field_type& k) |
| 230 |
{ |
| 231 |
(static_cast<Imp::block_vector_unmanaged<B,size_type>&>(*this)) = k; |
| 232 |
return *this; |
| 233 |
} |
| 234 |
|
| 235 |
|
| 236 |
//===== access to components |
| 237 |
// has to be overwritten from base class because it must |
| 238 |
// return access to the windows |
| 239 |
|
| 240 |
//! random access to blocks |
| 241 |
reference operator[] (size_type i) |
| 242 |
{ |
| 243 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 244 |
if (i>=rows_) DUNE_THROW(ISTLError,"index out of range"); |
| 245 |
#endif |
| 246 |
return window_type(this->p + i*columns_, columns_); |
| 247 |
} |
| 248 |
|
| 249 |
//! same for read only access |
| 250 |
const_reference operator[] (size_type i) const |
| 251 |
{ |
| 252 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 253 |
if (i<0 || i>=rows_) DUNE_THROW(ISTLError,"index out of range"); |
| 254 |
#endif |
| 255 |
return window_type(this->p + i*columns_, columns_); |
| 256 |
} |
| 257 |
|
| 258 |
// forward declaration |
| 259 |
class ConstIterator; |
| 260 |
|
| 261 |
//! Iterator class for sequential access |
| 262 |
class Iterator |
| 263 |
{ |
| 264 |
public: |
| 265 |
//! constructor, no arguments |
| 266 |
Iterator () |
| 267 |
: window_(nullptr,0) |
| 268 |
{ |
| 269 |
i = 0; |
| 270 |
} |
| 271 |
|
| 272 |
Iterator (Iterator& other) = default; |
| 273 |
Iterator (Iterator&& other) = default; |
| 274 |
|
| 275 |
//! constructor |
| 276 |
Iterator (B* data, size_type columns, size_type _i) |
| 277 |
: i(_i), |
| 278 |
window_(data + _i*columns, columns) |
| 279 |
{} |
| 280 |
|
| 281 |
/** \brief Move assignment */ |
| 282 |
Iterator& operator=(Iterator&& other) |
| 283 |
{ |
| 284 |
i = other.i; |
| 285 |
// Do NOT use window_.operator=, because that copies the window content, not just the window! |
| 286 |
window_.set(other.window_.getsize(),other.window_.getptr()); |
| 287 |
return *this; |
| 288 |
} |
| 289 |
|
| 290 |
/** \brief Copy assignment */ |
| 291 |
Iterator& operator=(Iterator& other) |
| 292 |
{ |
| 293 |
i = other.i; |
| 294 |
// Do NOT use window_.operator=, because that copies the window content, not just the window! |
| 295 |
window_.set(other.window_.getsize(),other.window_.getptr()); |
| 296 |
return *this; |
| 297 |
} |
| 298 |
|
| 299 |
//! prefix increment |
| 300 |
Iterator& operator++() |
| 301 |
{ |
| 302 |
++i; |
| 303 |
window_.setptr(window_.getptr()+window_.getsize()); |
| 304 |
return *this; |
| 305 |
} |
| 306 |
|
| 307 |
//! prefix decrement |
| 308 |
Iterator& operator--() |
| 309 |
{ |
| 310 |
--i; |
| 311 |
window_.setptr(window_.getptr()-window_.getsize()); |
| 312 |
return *this; |
| 313 |
} |
| 314 |
|
| 315 |
//! equality |
| 316 |
bool operator== (const Iterator& it) const |
| 317 |
{ |
| 318 |
return window_.getptr() == it.window_.getptr(); |
| 319 |
} |
| 320 |
|
| 321 |
//! inequality |
| 322 |
bool operator!= (const Iterator& it) const |
| 323 |
{ |
| 324 |
return window_.getptr() != it.window_.getptr(); |
| 325 |
} |
| 326 |
|
| 327 |
//! equality |
| 328 |
bool operator== (const ConstIterator& it) const |
| 329 |
{ |
| 330 |
return window_.getptr() == it.window_.getptr(); |
| 331 |
} |
| 332 |
|
| 333 |
//! inequality |
| 334 |
bool operator!= (const ConstIterator& it) const |
| 335 |
{ |
| 336 |
return window_.getptr() != it.window_.getptr(); |
| 337 |
} |
| 338 |
|
| 339 |
//! dereferencing |
| 340 |
window_type& operator* () const |
| 341 |
{ |
| 342 |
return window_; |
| 343 |
} |
| 344 |
|
| 345 |
//! arrow |
| 346 |
window_type* operator-> () const |
| 347 |
{ |
| 348 |
return &window_; |
| 349 |
} |
| 350 |
|
| 351 |
// return index corresponding to pointer |
| 352 |
size_type index () const |
| 353 |
{ |
| 354 |
return i; |
| 355 |
} |
| 356 |
|
| 357 |
friend class ConstIterator; |
| 358 |
|
| 359 |
private: |
| 360 |
size_type i; |
| 361 |
mutable window_type window_; |
| 362 |
}; |
| 363 |
|
| 364 |
//! begin Iterator |
| 365 |
Iterator begin () |
| 366 |
{ |
| 367 |
return Iterator(this->p, columns_, 0); |
| 368 |
} |
| 369 |
|
| 370 |
//! end Iterator |
| 371 |
Iterator end () |
| 372 |
{ |
| 373 |
return Iterator(this->p, columns_, rows_); |
| 374 |
} |
| 375 |
|
| 376 |
//! @returns an iterator that is positioned before |
| 377 |
//! the end iterator of the vector, i.e. at the last entry. |
| 378 |
Iterator beforeEnd () |
| 379 |
{ |
| 380 |
return Iterator(this->p, columns_, rows_-1); |
| 381 |
} |
| 382 |
|
| 383 |
//! @returns an iterator that is positioned before |
| 384 |
//! the first entry of the vector. |
| 385 |
Iterator beforeBegin () const |
| 386 |
{ |
| 387 |
return Iterator(this->p, columns_, -1); |
| 388 |
} |
| 389 |
|
| 390 |
//! random access returning iterator (end if not contained) |
| 391 |
Iterator find (size_type i) |
| 392 |
{ |
| 393 |
return Iterator(this->p, columns_, std::min(i,rows_)); |
| 394 |
} |
| 395 |
|
| 396 |
//! random access returning iterator (end if not contained) |
| 397 |
ConstIterator find (size_type i) const |
| 398 |
{ |
| 399 |
return ConstIterator(this->p, columns_, std::min(i,rows_)); |
| 400 |
} |
| 401 |
|
| 402 |
//! ConstIterator class for sequential access |
| 403 |
class ConstIterator |
| 404 |
{ |
| 405 |
public: |
| 406 |
//! constructor |
| 407 |
ConstIterator () |
| 408 |
: window_(nullptr,0) |
| 409 |
{ |
| 410 |
i = 0; |
| 411 |
} |
| 412 |
|
| 413 |
//! constructor from pointer |
| 414 |
ConstIterator (const B* data, size_type columns, size_type _i) |
| 415 |
: i(_i), |
| 416 |
window_(const_cast<B*>(data + _i * columns), columns) |
| 417 |
{} |
| 418 |
|
| 419 |
//! constructor from non_const iterator |
| 420 |
ConstIterator (const Iterator& it) |
| 421 |
: i(it.i), window_(it.window_.getptr(),it.window_.getsize()) |
| 422 |
{} |
| 423 |
|
| 424 |
ConstIterator& operator=(Iterator&& other) |
| 425 |
{ |
| 426 |
i = other.i; |
| 427 |
// Do NOT use window_.operator=, because that copies the window content, not just the window! |
| 428 |
window_.set(other.window_.getsize(),other.window_.getptr()); |
| 429 |
return *this; |
| 430 |
} |
| 431 |
|
| 432 |
ConstIterator& operator=(Iterator& other) |
| 433 |
{ |
| 434 |
i = other.i; |
| 435 |
// Do NOT use window_.operator=, because that copies the window content, not just the window! |
| 436 |
window_.set(other.window_.getsize(),other.window_.getptr()); |
| 437 |
return *this; |
| 438 |
} |
| 439 |
|
| 440 |
//! prefix increment |
| 441 |
ConstIterator& operator++() |
| 442 |
{ |
| 443 |
++i; |
| 444 |
window_.setptr(window_.getptr()+window_.getsize()); |
| 445 |
return *this; |
| 446 |
} |
| 447 |
|
| 448 |
//! prefix decrement |
| 449 |
ConstIterator& operator--() |
| 450 |
{ |
| 451 |
--i; |
| 452 |
window_.setptr(window_.getptr()-window_.getsize()); |
| 453 |
return *this; |
| 454 |
} |
| 455 |
|
| 456 |
//! equality |
| 457 |
bool operator== (const ConstIterator& it) const |
| 458 |
{ |
| 459 |
return window_.getptr() == it.window_.getptr(); |
| 460 |
} |
| 461 |
|
| 462 |
//! inequality |
| 463 |
bool operator!= (const ConstIterator& it) const |
| 464 |
{ |
| 465 |
return window_.getptr() != it.window_.getptr(); |
| 466 |
} |
| 467 |
|
| 468 |
//! equality |
| 469 |
bool operator== (const Iterator& it) const |
| 470 |
{ |
| 471 |
return window_.getptr() == it.window_.getptr(); |
| 472 |
} |
| 473 |
|
| 474 |
//! inequality |
| 475 |
bool operator!= (const Iterator& it) const |
| 476 |
{ |
| 477 |
return window_.getptr() != it.window_.getptr(); |
| 478 |
} |
| 479 |
|
| 480 |
//! dereferencing |
| 481 |
const window_type& operator* () const |
| 482 |
{ |
| 483 |
return window_; |
| 484 |
} |
| 485 |
|
| 486 |
//! arrow |
| 487 |
const window_type* operator-> () const |
| 488 |
{ |
| 489 |
return &window_; |
| 490 |
} |
| 491 |
|
| 492 |
// return index corresponding to pointer |
| 493 |
size_type index () const |
| 494 |
{ |
| 495 |
return i; |
| 496 |
} |
| 497 |
|
| 498 |
friend class Iterator; |
| 499 |
|
| 500 |
private: |
| 501 |
size_type i; |
| 502 |
mutable window_type window_; |
| 503 |
}; |
| 504 |
|
| 505 |
/** \brief Export the iterator type using std naming rules */ |
| 506 |
using iterator = Iterator; |
| 507 |
|
| 508 |
/** \brief Export the const iterator type using std naming rules */ |
| 509 |
using const_iterator = ConstIterator; |
| 510 |
|
| 511 |
//! begin ConstIterator |
| 512 |
ConstIterator begin () const |
| 513 |
{ |
| 514 |
return ConstIterator(this->p, columns_, 0); |
| 515 |
} |
| 516 |
|
| 517 |
//! end ConstIterator |
| 518 |
ConstIterator end () const |
| 519 |
{ |
| 520 |
return ConstIterator(this->p, columns_, rows_); |
| 521 |
} |
| 522 |
|
| 523 |
//! @returns an iterator that is positioned before |
| 524 |
//! the end iterator of the vector. i.e. at the last element. |
| 525 |
ConstIterator beforeEnd() const |
| 526 |
{ |
| 527 |
return ConstIterator(this->p, columns_, rows_-1); |
| 528 |
} |
| 529 |
|
| 530 |
//! end ConstIterator |
| 531 |
ConstIterator rend () const |
| 532 |
{ |
| 533 |
return ConstIterator(this->p, columns_, -1); |
| 534 |
} |
| 535 |
|
| 536 |
//===== sizes |
| 537 |
|
| 538 |
//! number of blocks in the vector (are of variable size here) |
| 539 |
size_type N () const |
| 540 |
{ |
| 541 |
return rows_; |
| 542 |
} |
| 543 |
|
| 544 |
|
| 545 |
private: |
| 546 |
size_type rows_; // number of matrix rows |
| 547 |
size_type columns_; // number of matrix columns |
| 548 |
|
| 549 |
A allocator_; |
| 550 |
}; |
| 551 |
|
| 552 |
} // namespace MatrixImp |
| 553 |
|
| 554 |
/** \addtogroup ISTL_SPMV |
| 555 |
\{ |
| 556 |
*/ |
| 557 |
/** \brief A generic dynamic dense matrix |
| 558 |
*/ |
| 559 |
template<class T, class A=std::allocator<T> > |
| 560 |
class Matrix |
| 561 |
{ |
| 562 |
public: |
| 563 |
|
| 564 |
/** \brief Export the type representing the underlying field */ |
| 565 |
using field_type = typename Imp::BlockTraits<T>::field_type; |
| 566 |
|
| 567 |
/** \brief Export the type representing the components */ |
| 568 |
typedef T block_type; |
| 569 |
|
| 570 |
/** \brief Export the allocator */ |
| 571 |
typedef A allocator_type; |
| 572 |
|
| 573 |
/** \brief The type implementing a matrix row */ |
| 574 |
typedef typename MatrixImp::DenseMatrixBase<T,A>::window_type row_type; |
| 575 |
|
| 576 |
/** \brief Type for indices and sizes */ |
| 577 |
typedef typename A::size_type size_type; |
| 578 |
|
| 579 |
/** \brief Iterator over the matrix rows */ |
| 580 |
typedef typename MatrixImp::DenseMatrixBase<T,A>::Iterator RowIterator; |
| 581 |
|
| 582 |
/** \brief Iterator for the entries of each row */ |
| 583 |
typedef typename row_type::iterator ColIterator; |
| 584 |
|
| 585 |
/** \brief Const iterator over the matrix rows */ |
| 586 |
typedef typename MatrixImp::DenseMatrixBase<T,A>::ConstIterator ConstRowIterator; |
| 587 |
|
| 588 |
/** \brief Const iterator for the entries of each row */ |
| 589 |
typedef typename row_type::const_iterator ConstColIterator; |
| 590 |
|
| 591 |
/** \brief Create empty matrix */ |
| 592 |
Matrix() : data_(0,0), cols_(0) |
| 593 |
{} |
| 594 |
|
| 595 |
/** \brief Create uninitialized matrix of size rows x cols |
| 596 |
*/ |
| 597 |
Matrix(size_type rows, size_type cols) : data_(rows,cols), cols_(cols) |
| 598 |
{} |
| 599 |
|
| 600 |
/** \brief Change the matrix size |
| 601 |
* |
| 602 |
* The way the data is handled is unpredictable. |
| 603 |
*/ |
| 604 |
void setSize(size_type rows, size_type cols) { |
| 605 |
data_.resize(rows,cols); |
| 606 |
cols_ = cols; |
| 607 |
} |
| 608 |
|
| 609 |
/** \brief Get iterator to first row */ |
| 610 |
RowIterator begin() |
| 611 |
{ |
| 612 |
return data_.begin(); |
| 613 |
} |
| 614 |
|
| 615 |
/** \brief Get iterator to one beyond last row */ |
| 616 |
RowIterator end() |
| 617 |
{ |
| 618 |
return data_.end(); |
| 619 |
} |
| 620 |
|
| 621 |
//! @returns an iterator that is positioned before |
| 622 |
//! the end iterator of the rows, i.e. at the last row. |
| 623 |
RowIterator beforeEnd () |
| 624 |
{ |
| 625 |
return data_.beforeEnd(); |
| 626 |
} |
| 627 |
|
| 628 |
//! @returns an iterator that is positioned before |
| 629 |
//! the first row of the matrix. |
| 630 |
RowIterator beforeBegin () |
| 631 |
{ |
| 632 |
return data_.beforeBegin(); |
| 633 |
} |
| 634 |
|
| 635 |
/** \brief Get const iterator to first row */ |
| 636 |
ConstRowIterator begin() const |
| 637 |
{ |
| 638 |
return data_.begin(); |
| 639 |
} |
| 640 |
|
| 641 |
/** \brief Get const iterator to one beyond last row */ |
| 642 |
ConstRowIterator end() const |
| 643 |
{ |
| 644 |
return data_.end(); |
| 645 |
} |
| 646 |
|
| 647 |
//! @returns an iterator that is positioned before |
| 648 |
//! the end iterator of the rows. i.e. at the last row. |
| 649 |
ConstRowIterator beforeEnd() const |
| 650 |
{ |
| 651 |
return data_.beforeEnd(); |
| 652 |
} |
| 653 |
|
| 654 |
//! @returns an iterator that is positioned before |
| 655 |
//! the first row if the matrix. |
| 656 |
ConstRowIterator beforeBegin () const |
| 657 |
{ |
| 658 |
return data_.beforeBegin(); |
| 659 |
} |
| 660 |
|
| 661 |
/** \brief Assignment from scalar */ |
| 662 |
Matrix& operator= (const field_type& t) |
| 663 |
{ |
| 664 |
data_ = t; |
| 665 |
return *this; |
| 666 |
} |
| 667 |
|
| 668 |
/** \brief The index operator */ |
| 669 |
row_type operator[](size_type row) { |
| 670 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 671 |
if (row<0) |
| 672 |
DUNE_THROW(ISTLError, "Can't access negative rows!"); |
| 673 |
if (row>=N()) |
| 674 |
DUNE_THROW(ISTLError, "Row index out of range!"); |
| 675 |
#endif |
| 676 |
return data_[row]; |
| 677 |
} |
| 678 |
|
| 679 |
/** \brief The const index operator */ |
| 680 |
const row_type operator[](size_type row) const { |
| 681 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 682 |
if (row<0) |
| 683 |
DUNE_THROW(ISTLError, "Can't access negative rows!"); |
| 684 |
if (row>=N()) |
| 685 |
DUNE_THROW(ISTLError, "Row index out of range!"); |
| 686 |
#endif |
| 687 |
return data_[row]; |
| 688 |
} |
| 689 |
|
| 690 |
/** \brief Return the number of rows */ |
| 691 |
size_type N() const { |
| 692 |
return data_.N(); |
| 693 |
} |
| 694 |
|
| 695 |
/** \brief Return the number of columns */ |
| 696 |
size_type M() const { |
| 697 |
return cols_; |
| 698 |
} |
| 699 |
|
| 700 |
/** \brief Multiplication with a scalar */ |
| 701 |
Matrix<T>& operator*=(const field_type& scalar) { |
| 702 |
data_ *= scalar; |
| 703 |
return (*this); |
| 704 |
} |
| 705 |
|
| 706 |
/** \brief Division by a scalar */ |
| 707 |
Matrix<T>& operator/=(const field_type& scalar) { |
| 708 |
data_ /= scalar; |
| 709 |
return (*this); |
| 710 |
} |
| 711 |
|
| 712 |
/*! \brief Add the entries of another matrix to this one. |
| 713 |
* |
| 714 |
* \param b The matrix to add to this one. Its size has to |
| 715 |
* be the same as the size of this matrix. |
| 716 |
*/ |
| 717 |
Matrix& operator+= (const Matrix& b) { |
| 718 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 719 |
if(N()!=b.N() || M() != b.M()) |
| 720 |
DUNE_THROW(RangeError, "Matrix sizes do not match!"); |
| 721 |
#endif |
| 722 |
data_ += b.data_; |
| 723 |
return (*this); |
| 724 |
} |
| 725 |
|
| 726 |
/*! \brief Subtract the entries of another matrix from this one. |
| 727 |
* |
| 728 |
* \param b The matrix to subtract from this one. Its size has to |
| 729 |
* be the same as the size of this matrix. |
| 730 |
*/ |
| 731 |
Matrix& operator-= (const Matrix& b) { |
| 732 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 733 |
if(N()!=b.N() || M() != b.M()) |
| 734 |
DUNE_THROW(RangeError, "Matrix sizes do not match!"); |
| 735 |
#endif |
| 736 |
data_ -= b.data_; |
| 737 |
return (*this); |
| 738 |
} |
| 739 |
|
| 740 |
/** \brief Return the transpose of the matrix */ |
| 741 |
Matrix transpose() const { |
| 742 |
Matrix out(M(), N()); |
| 743 |
for (size_type i=0; i<N(); i++) |
| 744 |
for (size_type j=0; j<M(); j++) |
| 745 |
out[j][i] = (*this)[i][j]; |
| 746 |
|
| 747 |
return out; |
| 748 |
} |
| 749 |
|
| 750 |
/// Generic matrix multiplication. |
| 751 |
friend Matrix<T> operator*(const Matrix<T>& m1, const Matrix<T>& m2) { |
| 752 |
Matrix<T> out(m1.N(), m2.M()); |
| 753 |
out = 0; |
| 754 |
|
| 755 |
for (size_type i=0; i<out.N(); i++ ) { |
| 756 |
for ( size_type j=0; j<out.M(); j++ ) |
| 757 |
for (size_type k=0; k<m1.M(); k++) |
| 758 |
out[i][j] += m1[i][k]*m2[k][j]; |
| 759 |
} |
| 760 |
|
| 761 |
return out; |
| 762 |
} |
| 763 |
|
| 764 |
/// Generic matrix-vector multiplication. |
| 765 |
template <class X, class Y> |
| 766 |
friend Y operator*(const Matrix<T>& m, const X& vec) { |
| 767 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 768 |
if (m.M()!=vec.size()) |
| 769 |
DUNE_THROW(ISTLError, "Vector size doesn't match the number of matrix columns!"); |
| 770 |
#endif |
| 771 |
Y out(m.N()); |
| 772 |
out = 0; |
| 773 |
|
| 774 |
for (size_type i=0; i<out.size(); i++ ) { |
| 775 |
for ( size_type j=0; j<vec.size(); j++ ) |
| 776 |
out[i] += m[i][j]*vec[j]; |
| 777 |
} |
| 778 |
|
| 779 |
return out; |
| 780 |
} |
| 781 |
|
| 782 |
//! y = A x |
| 783 |
template <class X, class Y> |
| 784 |
void mv(const X& x, Y& y) const |
| 785 |
{ |
| 786 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 787 |
if (x.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 788 |
if (y.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 789 |
#endif |
| 790 |
for (size_type i=0; i<data_.N(); i++) { |
| 791 |
y[i]=0; |
| 792 |
for (size_type j=0; j<cols_; j++) |
| 793 |
{ |
| 794 |
auto&& xj = Impl::asVector(x[j]); |
| 795 |
auto&& yi = Impl::asVector(y[i]); |
| 796 |
Impl::asMatrix((*this)[i][j]).umv(xj, yi); |
| 797 |
} |
| 798 |
} |
| 799 |
} |
| 800 |
|
| 801 |
//! y = A^T x |
| 802 |
template<class X, class Y> |
| 803 |
void mtv (const X& x, Y& y) const |
| 804 |
{ |
| 805 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 806 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"index out of range"); |
| 807 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"index out of range"); |
| 808 |
#endif |
| 809 |
for(size_type i=0; i<y.N(); ++i) |
| 810 |
y[i]=0; |
| 811 |
umtv(x,y); |
| 812 |
} |
| 813 |
|
| 814 |
//! y += A x |
| 815 |
template <class X, class Y> |
| 816 |
void umv(const X& x, Y& y) const |
| 817 |
{ |
| 818 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 819 |
if (x.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 820 |
if (y.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 821 |
#endif |
| 822 |
for (size_type i=0; i<data_.N(); i++) |
| 823 |
for (size_type j=0; j<cols_; j++) |
| 824 |
{ |
| 825 |
auto&& xj = Impl::asVector(x[j]); |
| 826 |
auto&& yi = Impl::asVector(y[i]); |
| 827 |
Impl::asMatrix((*this)[i][j]).umv(xj, yi); |
| 828 |
} |
| 829 |
} |
| 830 |
|
| 831 |
//! y -= A x |
| 832 |
template<class X, class Y> |
| 833 |
void mmv (const X& x, Y& y) const |
| 834 |
{ |
| 835 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 836 |
if (x.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 837 |
if (y.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 838 |
#endif |
| 839 |
for (size_type i=0; i<data_.N(); i++) |
| 840 |
for (size_type j=0; j<cols_; j++) |
| 841 |
{ |
| 842 |
auto&& xj = Impl::asVector(x[j]); |
| 843 |
auto&& yi = Impl::asVector(y[i]); |
| 844 |
Impl::asMatrix((*this)[i][j]).mmv(xj, yi); |
| 845 |
} |
| 846 |
} |
| 847 |
|
| 848 |
/** \brief \f$ y += \alpha A x \f$ */ |
| 849 |
template <class X, class Y> |
| 850 |
void usmv(const field_type& alpha, const X& x, Y& y) const |
| 851 |
{ |
| 852 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 853 |
if (x.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 854 |
if (y.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 855 |
#endif |
| 856 |
for (size_type i=0; i<data_.N(); i++) |
| 857 |
for (size_type j=0; j<cols_; j++) |
| 858 |
{ |
| 859 |
auto&& xj = Impl::asVector(x[j]); |
| 860 |
auto&& yi = Impl::asVector(y[i]); |
| 861 |
Impl::asMatrix((*this)[i][j]).usmv(alpha, xj, yi); |
| 862 |
} |
| 863 |
} |
| 864 |
|
| 865 |
//! y += A^T x |
| 866 |
template<class X, class Y> |
| 867 |
void umtv (const X& x, Y& y) const |
| 868 |
{ |
| 869 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 870 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 871 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 872 |
#endif |
| 873 |
for (size_type i=0; i<data_.N(); i++) |
| 874 |
for (size_type j=0; j<cols_; j++) |
| 875 |
{ |
| 876 |
auto&& xi = Impl::asVector(x[i]); |
| 877 |
auto&& yj = Impl::asVector(y[j]); |
| 878 |
Impl::asMatrix((*this)[i][j]).umtv(xi, yj); |
| 879 |
} |
| 880 |
} |
| 881 |
|
| 882 |
//! y -= A^T x |
| 883 |
template<class X, class Y> |
| 884 |
void mmtv (const X& x, Y& y) const |
| 885 |
{ |
| 886 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 887 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 888 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 889 |
#endif |
| 890 |
for (size_type i=0; i<data_.N(); i++) |
| 891 |
for (size_type j=0; j<cols_; j++) |
| 892 |
{ |
| 893 |
auto&& xi = Impl::asVector(x[i]); |
| 894 |
auto&& yj = Impl::asVector(y[j]); |
| 895 |
Impl::asMatrix((*this)[i][j]).mmtv(xi, yj); |
| 896 |
} |
| 897 |
} |
| 898 |
|
| 899 |
//! y += alpha A^T x |
| 900 |
template<class X, class Y> |
| 901 |
void usmtv (const field_type& alpha, const X& x, Y& y) const |
| 902 |
{ |
| 903 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 904 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 905 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 906 |
#endif |
| 907 |
for (size_type i=0; i<data_.N(); i++) |
| 908 |
for (size_type j=0; j<cols_; j++) |
| 909 |
{ |
| 910 |
auto&& xi = Impl::asVector(x[i]); |
| 911 |
auto&& yj = Impl::asVector(y[j]); |
| 912 |
Impl::asMatrix((*this)[i][j]).usmtv(alpha, xi, yj); |
| 913 |
} |
| 914 |
} |
| 915 |
|
| 916 |
//! y += A^H x |
| 917 |
template<class X, class Y> |
| 918 |
void umhv (const X& x, Y& y) const |
| 919 |
{ |
| 920 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 921 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 922 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 923 |
#endif |
| 924 |
for (size_type i=0; i<data_.N(); i++) |
| 925 |
for (size_type j=0; j<cols_; j++) |
| 926 |
{ |
| 927 |
auto&& xi = Impl::asVector(x[i]); |
| 928 |
auto&& yj = Impl::asVector(y[j]); |
| 929 |
Impl::asMatrix((*this)[i][j]).umhv(xi,yj); |
| 930 |
} |
| 931 |
} |
| 932 |
|
| 933 |
//! y -= A^H x |
| 934 |
template<class X, class Y> |
| 935 |
void mmhv (const X& x, Y& y) const |
| 936 |
{ |
| 937 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 938 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 939 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 940 |
#endif |
| 941 |
for (size_type i=0; i<data_.N(); i++) |
| 942 |
for (size_type j=0; j<cols_; j++) |
| 943 |
{ |
| 944 |
auto&& xi = Impl::asVector(x[i]); |
| 945 |
auto&& yj = Impl::asVector(y[j]); |
| 946 |
Impl::asMatrix((*this)[i][j]).mmhv(xi,yj); |
| 947 |
} |
| 948 |
} |
| 949 |
|
| 950 |
//! y += alpha A^H x |
| 951 |
template<class X, class Y> |
| 952 |
void usmhv (const field_type& alpha, const X& x, Y& y) const |
| 953 |
{ |
| 954 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 955 |
if (x.N()!=N()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 956 |
if (y.N()!=M()) DUNE_THROW(ISTLError,"vector/matrix size mismatch!"); |
| 957 |
#endif |
| 958 |
for (size_type i=0; i<data_.N(); i++) |
| 959 |
for (size_type j=0; j<cols_; j++) |
| 960 |
{ |
| 961 |
auto&& xi = Impl::asVector(x[i]); |
| 962 |
auto&& yj = Impl::asVector(y[j]); |
| 963 |
Impl::asMatrix((*this)[i][j]).usmhv(alpha,xi,yj); |
| 964 |
} |
| 965 |
} |
| 966 |
|
| 967 |
//===== norms |
| 968 |
|
| 969 |
//! frobenius norm: sqrt(sum over squared values of entries) |
| 970 |
typename FieldTraits<field_type>::real_type frobenius_norm () const |
| 971 |
{ |
| 972 |
return std::sqrt(frobenius_norm2()); |
| 973 |
} |
| 974 |
|
| 975 |
//! square of frobenius norm, need for block recursion |
| 976 |
typename FieldTraits<field_type>::real_type frobenius_norm2 () const |
| 977 |
{ |
| 978 |
typename FieldTraits<field_type>::real_type sum=0; |
| 979 |
for (size_type i=0; i<this->N(); i++) |
| 980 |
for (size_type j=0; j<this->M(); j++) |
| 981 |
sum += Impl::asMatrix(data_[i][j]).frobenius_norm2(); |
| 982 |
return sum; |
| 983 |
} |
| 984 |
|
| 985 |
//! infinity norm (row sum norm, how to generalize for blocks?) |
| 986 |
template <typename ft = field_type, |
| 987 |
typename std::enable_if<!HasNaN<ft>::value, int>::type = 0> |
| 988 |
typename FieldTraits<ft>::real_type infinity_norm() const { |
| 989 |
using real_type = typename FieldTraits<ft>::real_type; |
| 990 |
using std::max; |
| 991 |
|
| 992 |
real_type norm = 0; |
| 993 |
for (auto const &x : *this) { |
| 994 |
real_type sum = 0; |
| 995 |
for (auto const &y : x) |
| 996 |
sum += Impl::asMatrix(y).infinity_norm(); |
| 997 |
norm = max(sum, norm); |
| 998 |
isNaN += sum; |
| 999 |
} |
| 1000 |
|
| 1001 |
return norm; |
| 1002 |
} |
| 1003 |
|
| 1004 |
//! simplified infinity norm (uses Manhattan norm for complex values) |
| 1005 |
template <typename ft = field_type, |
| 1006 |
typename std::enable_if<!HasNaN<ft>::value, int>::type = 0> |
| 1007 |
typename FieldTraits<ft>::real_type infinity_norm_real() const { |
| 1008 |
using real_type = typename FieldTraits<ft>::real_type; |
| 1009 |
using std::max; |
| 1010 |
|
| 1011 |
real_type norm = 0; |
| 1012 |
for (auto const &x : *this) { |
| 1013 |
real_type sum = 0; |
| 1014 |
for (auto const &y : x) |
| 1015 |
sum += Impl::asMatrix(y).infinity_norm_real(); |
| 1016 |
norm = max(sum, norm); |
| 1017 |
} |
| 1018 |
return norm; |
| 1019 |
} |
| 1020 |
|
| 1021 |
//! infinity norm (row sum norm, how to generalize for blocks?) |
| 1022 |
template <typename ft = field_type, |
| 1023 |
typename std::enable_if<HasNaN<ft>::value, int>::type = 0> |
| 1024 |
typename FieldTraits<ft>::real_type infinity_norm() const { |
| 1025 |
using real_type = typename FieldTraits<ft>::real_type; |
| 1026 |
using std::max; |
| 1027 |
|
| 1028 |
real_type norm = 0; |
| 1029 |
real_type is_nan = 1; |
| 1030 |
for (auto const &x : *this) { |
| 1031 |
real_type sum = 0; |
| 1032 |
for (auto const &y : x) |
| 1033 |
sum += Impl::asMatrix(y).infinity_norm(); |
| 1034 |
norm = max(sum, norm); |
| 1035 |
is_nan += sum; |
| 1036 |
} |
| 1037 |
|
| 1038 |
return norm * (is_nan / is_nan); |
| 1039 |
} |
| 1040 |
|
| 1041 |
//! simplified infinity norm (uses Manhattan norm for complex values) |
| 1042 |
template <typename ft = field_type, |
| 1043 |
typename std::enable_if<HasNaN<ft>::value, int>::type = 0> |
| 1044 |
typename FieldTraits<ft>::real_type infinity_norm_real() const { |
| 1045 |
using real_type = typename FieldTraits<ft>::real_type; |
| 1046 |
using std::max; |
| 1047 |
|
| 1048 |
real_type norm = 0; |
| 1049 |
real_type is_nan = 1; |
| 1050 |
for (auto const &x : *this) { |
| 1051 |
real_type sum = 0; |
| 1052 |
for (auto const &y : x) |
| 1053 |
sum += Impl::asMatrix(y).infinity_norm_real(); |
| 1054 |
norm = max(sum, norm); |
| 1055 |
is_nan += sum; |
| 1056 |
} |
| 1057 |
|
| 1058 |
return norm * (is_nan / is_nan); |
| 1059 |
} |
| 1060 |
|
| 1061 |
//===== query |
| 1062 |
|
| 1063 |
//! return true if (i,j) is in pattern |
| 1064 |
bool exists ([[maybe_unused]] size_type i, [[maybe_unused]] size_type j) const |
| 1065 |
{ |
| 1066 |
#ifdef DUNE_ISTL_WITH_CHECKING |
| 1067 |
if (i<0 || i>=N()) DUNE_THROW(ISTLError,"row index out of range"); |
| 1068 |
if (j<0 || i>=M()) DUNE_THROW(ISTLError,"column index out of range"); |
| 1069 |
#endif |
| 1070 |
return true; |
| 1071 |
} |
| 1072 |
|
| 1073 |
protected: |
| 1074 |
|
| 1075 |
/** \brief Abuse DenseMatrixBase as an engine for a 2d array ISTL-style |
| 1076 |
*/ |
| 1077 |
MatrixImp::DenseMatrixBase<T,A> data_; |
| 1078 |
|
| 1079 |
/** \brief Number of columns of the matrix |
| 1080 |
|
| 1081 |
In general you can extract the same information from the data_ member. However if you |
| 1082 |
want to be able to properly handle 0xn matrices then you need a separate member. |
| 1083 |
*/ |
| 1084 |
size_type cols_; |
| 1085 |
}; |
| 1086 |
|
| 1087 |
template<class T, class A> |
| 1088 |
struct FieldTraits< Matrix<T, A> > |
| 1089 |
{ |
| 1090 |
using field_type = typename Matrix<T, A>::field_type; |
| 1091 |
using real_type = typename FieldTraits<field_type>::real_type; |
| 1092 |
}; |
| 1093 |
|
| 1094 |
/** \} */ |
| 1095 |
} // end namespace Dune |
| 1096 |
|
| 1097 |
#endif |