Reference documentation for deal.II version 9.5.0
\(\newcommand{\dealvcentcolon}{\mathrel{\mathop{:}}}\) \(\newcommand{\dealcoloneq}{\dealvcentcolon\mathrel{\mkern-1.2mu}=}\) \(\newcommand{\jump}[1]{\left[\!\left[ #1 \right]\!\right]}\) \(\newcommand{\average}[1]{\left\{\!\left\{ #1 \right\}\!\right\}}\)
Loading...
Searching...
No Matches
trilinos_sparse_matrix.cc
Go to the documentation of this file.
1// ---------------------------------------------------------------------
2//
3// Copyright (C) 2008 - 2023 by the deal.II authors
4//
5// This file is part of the deal.II library.
6//
7// The deal.II library is free software; you can use it, redistribute
8// it, and/or modify it under the terms of the GNU Lesser General
9// Public License as published by the Free Software Foundation; either
10// version 2.1 of the License, or (at your option) any later version.
11// The full text of the license can be found in the file LICENSE.md at
12// the top level directory of deal.II.
13//
14// ---------------------------------------------------------------------
15
18
19#ifdef DEAL_II_WITH_TRILINOS
20
23
30
31# include <boost/container/small_vector.hpp>
32
33# ifdef DEAL_II_TRILINOS_WITH_EPETRAEXT
34# include <EpetraExt_MatrixMatrix.h>
35# endif
36# include <Epetra_Export.h>
37# include <Teuchos_RCP.hpp>
38# include <ml_epetra_utils.h>
39# include <ml_struct.h>
40
41# include <memory>
42
44
45namespace TrilinosWrappers
46{
47 namespace internal
48 {
49 template <typename VectorType>
50 typename VectorType::value_type *
51 begin(VectorType &V)
52 {
53 return V.begin();
54 }
55
56 template <typename VectorType>
57 const typename VectorType::value_type *
58 begin(const VectorType &V)
59 {
60 return V.begin();
61 }
62
63 template <typename VectorType>
64 typename VectorType::value_type *
65 end(VectorType &V)
66 {
67 return V.end();
68 }
69
70 template <typename VectorType>
71 const typename VectorType::value_type *
72 end(const VectorType &V)
73 {
74 return V.end();
75 }
76
77 template <>
78 double *
80 {
81 return V.trilinos_vector()[0];
82 }
83
84 template <>
85 const double *
87 {
88 return V.trilinos_vector()[0];
89 }
90
91 template <>
92 double *
94 {
95 return V.trilinos_vector()[0] + V.trilinos_vector().MyLength();
96 }
97
98 template <>
99 const double *
101 {
102 return V.trilinos_vector()[0] + V.trilinos_vector().MyLength();
103 }
104
105# ifdef DEAL_II_TRILINOS_WITH_TPETRA
106 template <typename Number>
107 Number *
109 {
110 return V.trilinos_vector().getDataNonConst().get();
111 }
112
113 template <typename Number>
114 const Number *
116 {
117 return V.trilinos_vector().getData().get();
118 }
119
120 template <typename Number>
121 Number *
123 {
124 return V.trilinos_vector().getDataNonConst().get() +
125 V.trilinos_vector().getLocalLength();
126 }
127
128 template <typename Number>
129 const Number *
131 {
132 return V.trilinos_vector().getData().get() +
133 V.trilinos_vector().getLocalLength();
134 }
135# endif
136 } // namespace internal
137
138
139 namespace SparseMatrixIterators
140 {
141 void
143 {
144 // if we are asked to visit the past-the-end line, then simply
145 // release all our caches and go on with life.
146 //
147 // do the same if the row we're supposed to visit is not locally
148 // owned. this is simply going to make non-locally owned rows
149 // look like they're empty
150 if ((this->a_row == matrix->m()) ||
151 (matrix->in_local_range(this->a_row) == false))
152 {
153 colnum_cache.reset();
154 value_cache.reset();
155
156 return;
157 }
158
159 // get a representation of the present row
160 int ncols;
162 if (value_cache.get() == nullptr)
163 {
165 std::make_shared<std::vector<TrilinosScalar>>(matrix->n());
166 colnum_cache = std::make_shared<std::vector<size_type>>(matrix->n());
167 }
168 else
169 {
170 value_cache->resize(matrix->n());
171 colnum_cache->resize(matrix->n());
172 }
173
174 int ierr = matrix->trilinos_matrix().ExtractGlobalRowCopy(
175 this->a_row,
176 colnums,
177 ncols,
178 value_cache->data(),
179 reinterpret_cast<TrilinosWrappers::types::int_type *>(
180 colnum_cache->data()));
181 value_cache->resize(ncols);
182 colnum_cache->resize(ncols);
183 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
184
185 // copy it into our caches if the
186 // line isn't empty. if it is, then
187 // we've done something wrong, since
188 // we shouldn't have initialized an
189 // iterator for an empty line (what
190 // would it point to?)
191 }
192 } // namespace SparseMatrixIterators
193
194
195 // The constructor is actually the
196 // only point where we have to check
197 // whether we build a serial or a
198 // parallel Trilinos matrix.
199 // Actually, it does not even matter
200 // how many threads there are, but
201 // only if we use an MPI compiler or
202 // a standard compiler. So, even one
203 // thread on a configuration with
204 // MPI will still get a parallel
205 // interface.
207 : column_space_map(new Epetra_Map(0, 0, Utilities::Trilinos::comm_self()))
208 , matrix(
209 new Epetra_FECrsMatrix(View, *column_space_map, *column_space_map, 0))
210 , last_action(Zero)
211 , compressed(true)
212 {
213 matrix->FillComplete();
214 }
215
216
217
219 const size_type n,
220 const unsigned int n_max_entries_per_row)
221 : column_space_map(
222 new Epetra_Map(static_cast<TrilinosWrappers::types::int_type>(n),
223 0,
224 Utilities::Trilinos::comm_self()))
225 ,
226
227 // on one processor only, we know how the
228 // columns of the matrix will be
229 // distributed (everything on one
230 // processor), so we can hand in this
231 // information to the constructor. we
232 // can't do so in parallel, where the
233 // information from columns is only
234 // available when entries have been added
235 matrix(new Epetra_FECrsMatrix(
236 Copy,
237 Epetra_Map(static_cast<TrilinosWrappers::types::int_type>(m),
238 0,
239 Utilities::Trilinos::comm_self()),
240 *column_space_map,
241 n_max_entries_per_row,
242 false))
243 , last_action(Zero)
244 , compressed(false)
245 {}
246
247
248
250 const size_type n,
251 const std::vector<unsigned int> &n_entries_per_row)
252 : column_space_map(
253 new Epetra_Map(static_cast<TrilinosWrappers::types::int_type>(n),
254 0,
255 Utilities::Trilinos::comm_self()))
256 , matrix(new Epetra_FECrsMatrix(
257 Copy,
258 Epetra_Map(static_cast<TrilinosWrappers::types::int_type>(m),
259 0,
260 Utilities::Trilinos::comm_self()),
261 *column_space_map,
262 reinterpret_cast<int *>(
263 const_cast<unsigned int *>(n_entries_per_row.data())),
264 false))
265 , last_action(Zero)
266 , compressed(false)
267 {}
268
269
270
271 SparseMatrix::SparseMatrix(const IndexSet & parallel_partitioning,
272 const MPI_Comm communicator,
273 const unsigned int n_max_entries_per_row)
274 : column_space_map(new Epetra_Map(
275 parallel_partitioning.make_trilinos_map(communicator, false)))
276 , matrix(new Epetra_FECrsMatrix(Copy,
277 *column_space_map,
278 n_max_entries_per_row,
279 false))
280 , last_action(Zero)
281 , compressed(false)
282 {}
283
284
285
286 SparseMatrix::SparseMatrix(const IndexSet &parallel_partitioning,
287 const MPI_Comm communicator,
288 const std::vector<unsigned int> &n_entries_per_row)
289 : column_space_map(new Epetra_Map(
290 parallel_partitioning.make_trilinos_map(communicator, false)))
291 , matrix(new Epetra_FECrsMatrix(Copy,
292 *column_space_map,
293 reinterpret_cast<int *>(
294 const_cast<unsigned int *>(
295 n_entries_per_row.data())),
296 false))
297 , last_action(Zero)
298 , compressed(false)
299 {}
300
301
302
303 SparseMatrix::SparseMatrix(const IndexSet &row_parallel_partitioning,
304 const IndexSet &col_parallel_partitioning,
305 const MPI_Comm communicator,
306 const size_type n_max_entries_per_row)
307 : column_space_map(new Epetra_Map(
308 col_parallel_partitioning.make_trilinos_map(communicator, false)))
309 , matrix(new Epetra_FECrsMatrix(
310 Copy,
311 row_parallel_partitioning.make_trilinos_map(communicator, false),
312 n_max_entries_per_row,
313 false))
314 , last_action(Zero)
315 , compressed(false)
316 {}
317
318
319
320 SparseMatrix::SparseMatrix(const IndexSet &row_parallel_partitioning,
321 const IndexSet &col_parallel_partitioning,
322 const MPI_Comm communicator,
323 const std::vector<unsigned int> &n_entries_per_row)
324 : column_space_map(new Epetra_Map(
325 col_parallel_partitioning.make_trilinos_map(communicator, false)))
326 , matrix(new Epetra_FECrsMatrix(
327 Copy,
328 row_parallel_partitioning.make_trilinos_map(communicator, false),
329 reinterpret_cast<int *>(
330 const_cast<unsigned int *>(n_entries_per_row.data())),
331 false))
332 , last_action(Zero)
333 , compressed(false)
334 {}
335
336
337
339 : column_space_map(new Epetra_Map(sparsity_pattern.domain_partitioner()))
340 , matrix(
341 new Epetra_FECrsMatrix(Copy,
342 sparsity_pattern.trilinos_sparsity_pattern(),
343 false))
344 , last_action(Zero)
345 , compressed(true)
346 {
347 Assert(sparsity_pattern.trilinos_sparsity_pattern().Filled() == true,
349 "The Trilinos sparsity pattern has not been compressed."));
351 }
352
353
354
356 : column_space_map(std::move(other.column_space_map))
357 , matrix(std::move(other.matrix))
358 , nonlocal_matrix(std::move(other.nonlocal_matrix))
359 , nonlocal_matrix_exporter(std::move(other.nonlocal_matrix_exporter))
360 , last_action(other.last_action)
361 , compressed(other.compressed)
362 {
363 other.last_action = Zero;
364 other.compressed = false;
365 }
366
367
368
369 void
371 {
372 if (this == &rhs)
373 return;
374
375 nonlocal_matrix.reset();
377
378 // check whether we need to update the whole matrix layout (we have
379 // different maps or if we detect a row where the columns of the two
380 // matrices do not match)
381 bool needs_deep_copy =
382 !matrix->RowMap().SameAs(rhs.matrix->RowMap()) ||
383 !matrix->ColMap().SameAs(rhs.matrix->ColMap()) ||
384 !matrix->DomainMap().SameAs(rhs.matrix->DomainMap()) ||
386 if (!needs_deep_copy)
387 {
388 // Try to copy all the rows of the matrix one by one. In case of error
389 // (i.e., the column indices are different), we need to abort and blow
390 // away the matrix.
391 for (const auto row : locally_owned_range_indices())
392 {
393 const int row_local = matrix->RowMap().LID(
394 static_cast<TrilinosWrappers::types::int_type>(row));
395 Assert((row_local >= 0), ExcAccessToNonlocalRow(row));
396
397 int n_entries, rhs_n_entries;
398 TrilinosScalar *value_ptr, *rhs_value_ptr;
399 int * index_ptr, *rhs_index_ptr;
400 int ierr = rhs.matrix->ExtractMyRowView(row_local,
401 rhs_n_entries,
402 rhs_value_ptr,
403 rhs_index_ptr);
404 (void)ierr;
405 Assert(ierr == 0, ExcTrilinosError(ierr));
406
407 ierr = matrix->ExtractMyRowView(row_local,
408 n_entries,
409 value_ptr,
410 index_ptr);
411 Assert(ierr == 0, ExcTrilinosError(ierr));
412
413 if (n_entries != rhs_n_entries ||
414 std::memcmp(static_cast<void *>(index_ptr),
415 static_cast<void *>(rhs_index_ptr),
416 sizeof(int) * n_entries) != 0)
417 {
418 needs_deep_copy = true;
419 break;
420 }
421
422 for (int i = 0; i < n_entries; ++i)
423 value_ptr[i] = rhs_value_ptr[i];
424 }
425 }
426
427 if (needs_deep_copy)
428 {
430 std::make_unique<Epetra_Map>(rhs.trilinos_matrix().DomainMap());
431
432 // release memory before reallocation
433 matrix = std::make_unique<Epetra_FECrsMatrix>(*rhs.matrix);
434
435 matrix->FillComplete(*column_space_map, matrix->RowMap());
436 }
437
438 if (rhs.nonlocal_matrix.get() != nullptr)
440 std::make_unique<Epetra_CrsMatrix>(Copy, rhs.nonlocal_matrix->Graph());
441 }
442
443
444
445 namespace
446 {
447 using size_type = SparseMatrix::size_type;
448
449 template <typename SparsityPatternType>
450 void
451 reinit_matrix(const IndexSet & row_parallel_partitioning,
452 const IndexSet & column_parallel_partitioning,
453 const SparsityPatternType & sparsity_pattern,
454 const bool exchange_data,
455 const MPI_Comm communicator,
456 std::unique_ptr<Epetra_Map> &column_space_map,
457 std::unique_ptr<Epetra_FECrsMatrix> &matrix,
458 std::unique_ptr<Epetra_CrsMatrix> & nonlocal_matrix,
459 std::unique_ptr<Epetra_Export> & nonlocal_matrix_exporter)
460 {
461 // release memory before reallocation
462 matrix.reset();
463 nonlocal_matrix.reset();
464 nonlocal_matrix_exporter.reset();
465
466 column_space_map = std::make_unique<Epetra_Map>(
467 column_parallel_partitioning.make_trilinos_map(communicator, false));
468
469 if (column_space_map->Comm().MyPID() == 0)
470 {
471 AssertDimension(sparsity_pattern.n_rows(),
472 row_parallel_partitioning.size());
473 AssertDimension(sparsity_pattern.n_cols(),
474 column_parallel_partitioning.size());
475 }
476
477 Epetra_Map row_space_map =
478 row_parallel_partitioning.make_trilinos_map(communicator, false);
479
480 // if we want to exchange data, build a usual Trilinos sparsity pattern
481 // and let that handle the exchange. otherwise, manually create a
482 // CrsGraph, which consumes considerably less memory because it can set
483 // correct number of indices right from the start
484 if (exchange_data)
485 {
486 SparsityPattern trilinos_sparsity;
487 trilinos_sparsity.reinit(row_parallel_partitioning,
488 column_parallel_partitioning,
489 sparsity_pattern,
490 communicator,
491 exchange_data);
492 matrix = std::make_unique<Epetra_FECrsMatrix>(
493 Copy, trilinos_sparsity.trilinos_sparsity_pattern(), false);
494
495 return;
496 }
497
498 const size_type first_row = TrilinosWrappers::min_my_gid(row_space_map),
499 last_row =
500 TrilinosWrappers::max_my_gid(row_space_map) + 1;
501 std::vector<int> n_entries_per_row(last_row - first_row);
502
503 for (size_type row = first_row; row < last_row; ++row)
504 n_entries_per_row[row - first_row] = sparsity_pattern.row_length(row);
505
506 // The deal.II notation of a Sparsity pattern corresponds to the Epetra
507 // concept of a Graph. Hence, we generate a graph by copying the
508 // sparsity pattern into it, and then build up the matrix from the
509 // graph. This is considerable faster than directly filling elements
510 // into the matrix. Moreover, it consumes less memory, since the
511 // internal reordering is done on ints only, and we can leave the
512 // doubles aside.
513
514 // for more than one processor, need to specify only row map first and
515 // let the matrix entries decide about the column map (which says which
516 // columns are present in the matrix, not to be confused with the
517 // col_map that tells how the domain dofs of the matrix will be
518 // distributed). for only one processor, we can directly assign the
519 // columns as well. Compare this with bug # 4123 in the Sandia Bugzilla.
520 std::unique_ptr<Epetra_CrsGraph> graph;
521 if (row_space_map.Comm().NumProc() > 1)
522 graph = std::make_unique<Epetra_CrsGraph>(Copy,
523 row_space_map,
524 n_entries_per_row.data(),
525 true);
526 else
527 graph = std::make_unique<Epetra_CrsGraph>(Copy,
528 row_space_map,
529 *column_space_map,
530 n_entries_per_row.data(),
531 true);
532
533 // This functions assumes that the sparsity pattern sits on all
534 // processors (completely). The parallel version uses an Epetra graph
535 // that is already distributed.
536
537 // now insert the indices
538 std::vector<TrilinosWrappers::types::int_type> row_indices;
539
540 for (size_type row = first_row; row < last_row; ++row)
541 {
542 const int row_length = sparsity_pattern.row_length(row);
543 if (row_length == 0)
544 continue;
545
546 row_indices.resize(row_length, -1);
547 {
548 typename SparsityPatternType::iterator p =
549 sparsity_pattern.begin(row);
550 for (size_type col = 0; p != sparsity_pattern.end(row); ++p, ++col)
551 row_indices[col] = p->column();
552 }
553 graph->Epetra_CrsGraph::InsertGlobalIndices(row,
554 row_length,
555 row_indices.data());
556 }
557
558 // Eventually, optimize the graph structure (sort indices, make memory
559 // contiguous, etc). note that the documentation of the function indeed
560 // states that we first need to provide the column (domain) map and then
561 // the row (range) map
562 graph->FillComplete(*column_space_map, row_space_map);
563 graph->OptimizeStorage();
564
565 // check whether we got the number of columns right.
566 AssertDimension(sparsity_pattern.n_cols(),
568 (void)n_global_cols;
569
570 // And now finally generate the matrix.
571 matrix = std::make_unique<Epetra_FECrsMatrix>(Copy, *graph, false);
572 }
573
574
575
576 // for the non-local graph, we need to circumvent the problem that some
577 // processors will not add into the non-local graph at all: We do not want
578 // to insert dummy elements on >5000 processors because that gets very
579 // slow. Thus, we set a flag in Epetra_CrsGraph that sets the correct
580 // flag. Since it is protected, we need to expose this information by
581 // deriving a class from Epetra_CrsGraph for the purpose of creating the
582 // data structure
583 class Epetra_CrsGraphMod : public Epetra_CrsGraph
584 {
585 public:
586 Epetra_CrsGraphMod(const Epetra_Map &row_map,
587 const int * n_entries_per_row)
588 : Epetra_CrsGraph(Copy, row_map, n_entries_per_row, true)
589 {}
590
591 void
592 SetIndicesAreGlobal()
593 {
594 this->Epetra_CrsGraph::SetIndicesAreGlobal(true);
595 }
596 };
597
598
599
600 // specialization for DynamicSparsityPattern which can provide us with
601 // more information about the non-locally owned rows
602 template <>
603 void
604 reinit_matrix(const IndexSet & row_parallel_partitioning,
605 const IndexSet & column_parallel_partitioning,
606 const DynamicSparsityPattern &sparsity_pattern,
607 const bool exchange_data,
608 const MPI_Comm communicator,
609 std::unique_ptr<Epetra_Map> & column_space_map,
610 std::unique_ptr<Epetra_FECrsMatrix> &matrix,
611 std::unique_ptr<Epetra_CrsMatrix> & nonlocal_matrix,
612 std::unique_ptr<Epetra_Export> & nonlocal_matrix_exporter)
613 {
614 matrix.reset();
615 nonlocal_matrix.reset();
616 nonlocal_matrix_exporter.reset();
617
618 column_space_map = std::make_unique<Epetra_Map>(
619 column_parallel_partitioning.make_trilinos_map(communicator, false));
620
621 AssertDimension(sparsity_pattern.n_rows(),
622 row_parallel_partitioning.size());
623 AssertDimension(sparsity_pattern.n_cols(),
624 column_parallel_partitioning.size());
625
626 Epetra_Map row_space_map =
627 row_parallel_partitioning.make_trilinos_map(communicator, false);
628
629 IndexSet relevant_rows(sparsity_pattern.row_index_set());
630 // serial case
631 if (relevant_rows.size() == 0)
632 {
633 relevant_rows.set_size(
635 relevant_rows.add_range(
636 0, TrilinosWrappers::n_global_elements(row_space_map));
637 }
638 relevant_rows.compress();
639 Assert(relevant_rows.n_elements() >=
640 static_cast<unsigned int>(row_space_map.NumMyElements()),
642 "Locally relevant rows of sparsity pattern must contain "
643 "all locally owned rows"));
644
645 // check whether the relevant rows correspond to exactly the same map as
646 // the owned rows. In that case, do not create the nonlocal graph and
647 // fill the columns by demand
648 const bool have_ghost_rows = [&]() {
649 const std::vector<::types::global_dof_index> indices =
650 relevant_rows.get_index_vector();
651 Epetra_Map relevant_map(
653 TrilinosWrappers::types::int_type(relevant_rows.n_elements()),
654 (indices.empty() ?
655 nullptr :
656 reinterpret_cast<const TrilinosWrappers::types::int_type *>(
657 indices.data())),
658 0,
659 row_space_map.Comm());
660 return !relevant_map.SameAs(row_space_map);
661 }();
662
663 const unsigned int n_rows = relevant_rows.n_elements();
664 std::vector<TrilinosWrappers::types::int_type> ghost_rows;
665 std::vector<int> n_entries_per_row(row_space_map.NumMyElements());
666 std::vector<int> n_entries_per_ghost_row;
667 for (unsigned int i = 0, own = 0; i < n_rows; ++i)
668 {
669 const TrilinosWrappers::types::int_type global_row =
670 relevant_rows.nth_index_in_set(i);
671 if (row_space_map.MyGID(global_row))
672 n_entries_per_row[own++] = sparsity_pattern.row_length(global_row);
673 else if (sparsity_pattern.row_length(global_row) > 0)
674 {
675 ghost_rows.push_back(global_row);
676 n_entries_per_ghost_row.push_back(
677 sparsity_pattern.row_length(global_row));
678 }
679 }
680
681 Epetra_Map off_processor_map(-1,
682 ghost_rows.size(),
683 (ghost_rows.size() > 0) ?
684 (ghost_rows.data()) :
685 nullptr,
686 0,
687 row_space_map.Comm());
688
689 std::unique_ptr<Epetra_CrsGraph> graph;
690 std::unique_ptr<Epetra_CrsGraphMod> nonlocal_graph;
691 if (row_space_map.Comm().NumProc() > 1)
692 {
693 graph =
694 std::make_unique<Epetra_CrsGraph>(Copy,
695 row_space_map,
696 (n_entries_per_row.size() > 0) ?
697 (n_entries_per_row.data()) :
698 nullptr,
699 exchange_data ? false : true);
700 if (have_ghost_rows == true)
701 nonlocal_graph = std::make_unique<Epetra_CrsGraphMod>(
702 off_processor_map, n_entries_per_ghost_row.data());
703 }
704 else
705 graph =
706 std::make_unique<Epetra_CrsGraph>(Copy,
707 row_space_map,
708 *column_space_map,
709 (n_entries_per_row.size() > 0) ?
710 (n_entries_per_row.data()) :
711 nullptr,
712 true);
713
714 // now insert the indices, select between the right matrix
715 std::vector<TrilinosWrappers::types::int_type> row_indices;
716
717 for (unsigned int i = 0; i < n_rows; ++i)
718 {
719 const TrilinosWrappers::types::int_type global_row =
720 relevant_rows.nth_index_in_set(i);
721 const int row_length = sparsity_pattern.row_length(global_row);
722 if (row_length == 0)
723 continue;
724
725 row_indices.resize(row_length, -1);
726 for (int col = 0; col < row_length; ++col)
727 row_indices[col] = sparsity_pattern.column_number(global_row, col);
728
729 if (row_space_map.MyGID(global_row))
730 graph->InsertGlobalIndices(global_row,
731 row_length,
732 row_indices.data());
733 else
734 {
735 Assert(nonlocal_graph.get() != nullptr, ExcInternalError());
736 nonlocal_graph->InsertGlobalIndices(global_row,
737 row_length,
738 row_indices.data());
739 }
740 }
741
742 // finalize nonlocal graph and create nonlocal matrix
743 if (nonlocal_graph.get() != nullptr)
744 {
745 // must make sure the IndicesAreGlobal flag is set on all processors
746 // because some processors might not call InsertGlobalIndices (and
747 // we do not want to insert dummy indices on all processors for
748 // large-scale simulations due to the bad impact on performance)
749 nonlocal_graph->SetIndicesAreGlobal();
750 Assert(nonlocal_graph->IndicesAreGlobal() == true,
752 nonlocal_graph->FillComplete(*column_space_map, row_space_map);
753 nonlocal_graph->OptimizeStorage();
754
755 // insert data from nonlocal graph into the final sparsity pattern
756 if (exchange_data)
757 {
758 Epetra_Export exporter(nonlocal_graph->RowMap(), row_space_map);
759 int ierr = graph->Export(*nonlocal_graph, exporter, Add);
760 (void)ierr;
761 Assert(ierr == 0, ExcTrilinosError(ierr));
762 }
763
764 nonlocal_matrix =
765 std::make_unique<Epetra_CrsMatrix>(Copy, *nonlocal_graph);
766 }
767
768 graph->FillComplete(*column_space_map, row_space_map);
769 graph->OptimizeStorage();
770
771 AssertDimension(sparsity_pattern.n_cols(),
773
774 matrix = std::make_unique<Epetra_FECrsMatrix>(Copy, *graph, false);
775 }
776 } // namespace
777
778
779
780 template <typename SparsityPatternType>
781 void
782 SparseMatrix::reinit(const SparsityPatternType &sparsity_pattern)
783 {
784 reinit_matrix(complete_index_set(sparsity_pattern.n_rows()),
785 complete_index_set(sparsity_pattern.n_cols()),
786 sparsity_pattern,
787 false,
788 MPI_COMM_SELF,
790 matrix,
793 }
794
795
796
797 template <typename SparsityPatternType>
798 inline std::enable_if_t<
799 !std::is_same<SparsityPatternType, ::SparseMatrix<double>>::value>
800 SparseMatrix::reinit(const IndexSet & row_parallel_partitioning,
801 const IndexSet & col_parallel_partitioning,
802 const SparsityPatternType &sparsity_pattern,
803 const MPI_Comm communicator,
804 const bool exchange_data)
805 {
806 reinit_matrix(row_parallel_partitioning,
807 col_parallel_partitioning,
808 sparsity_pattern,
809 exchange_data,
810 communicator,
812 matrix,
815
816 // In the end, the matrix needs to be compressed in order to be really
817 // ready.
818 last_action = Zero;
820 }
821
822
823
824 void
825 SparseMatrix::reinit(const SparsityPattern &sparsity_pattern)
826 {
827 matrix.reset();
829
830 // reinit with a (parallel) Trilinos sparsity pattern.
832 std::make_unique<Epetra_Map>(sparsity_pattern.domain_partitioner());
833 matrix = std::make_unique<Epetra_FECrsMatrix>(
834 Copy, sparsity_pattern.trilinos_sparsity_pattern(), false);
835
836 if (sparsity_pattern.nonlocal_graph.get() != nullptr)
838 std::make_unique<Epetra_CrsMatrix>(Copy,
839 *sparsity_pattern.nonlocal_graph);
840 else
841 nonlocal_matrix.reset();
842
843 last_action = Zero;
845 }
846
847
848
849 void
850 SparseMatrix::reinit(const SparseMatrix &sparse_matrix)
851 {
852 if (this == &sparse_matrix)
853 return;
854
856 std::make_unique<Epetra_Map>(sparse_matrix.trilinos_matrix().DomainMap());
857 matrix.reset();
859 matrix = std::make_unique<Epetra_FECrsMatrix>(
860 Copy, sparse_matrix.trilinos_sparsity_pattern(), false);
861
862 if (sparse_matrix.nonlocal_matrix != nullptr)
863 nonlocal_matrix = std::make_unique<Epetra_CrsMatrix>(
864 Copy, sparse_matrix.nonlocal_matrix->Graph());
865 else
866 nonlocal_matrix.reset();
867
868 last_action = Zero;
870 }
871
872
873
874 template <typename number>
875 inline void
877 const IndexSet & row_parallel_partitioning,
878 const IndexSet & col_parallel_partitioning,
879 const ::SparseMatrix<number> &dealii_sparse_matrix,
880 const MPI_Comm communicator,
881 const double drop_tolerance,
882 const bool copy_values,
883 const ::SparsityPattern * use_this_sparsity)
884 {
885 if (copy_values == false)
886 {
887 // in case we do not copy values, just
888 // call the other function.
889 if (use_this_sparsity == nullptr)
890 reinit(row_parallel_partitioning,
891 col_parallel_partitioning,
892 dealii_sparse_matrix.get_sparsity_pattern(),
893 communicator,
894 false);
895 else
896 reinit(row_parallel_partitioning,
897 col_parallel_partitioning,
898 *use_this_sparsity,
899 communicator,
900 false);
901 return;
902 }
903
904 const size_type n_rows = dealii_sparse_matrix.m();
905
906 AssertDimension(row_parallel_partitioning.size(), n_rows);
907 AssertDimension(col_parallel_partitioning.size(), dealii_sparse_matrix.n());
908
909 const ::SparsityPattern &sparsity_pattern =
910 (use_this_sparsity != nullptr) ?
911 *use_this_sparsity :
912 dealii_sparse_matrix.get_sparsity_pattern();
913
914 if (matrix.get() == nullptr || m() != n_rows ||
915 n_nonzero_elements() != sparsity_pattern.n_nonzero_elements())
916 {
917 reinit(row_parallel_partitioning,
918 col_parallel_partitioning,
919 sparsity_pattern,
920 communicator,
921 false);
922 }
923
924 // fill the values. the same as above: go through all rows of the
925 // matrix, and then all columns. since the sparsity patterns of the
926 // input matrix and the specified sparsity pattern might be different,
927 // need to go through the row for both these sparsity structures
928 // simultaneously in order to really set the correct values.
929 size_type maximum_row_length = matrix->MaxNumEntries();
930 std::vector<size_type> row_indices(maximum_row_length);
931 std::vector<TrilinosScalar> values(maximum_row_length);
932
933 for (size_type row = 0; row < n_rows; ++row)
934 // see if the row is locally stored on this processor
935 if (row_parallel_partitioning.is_element(row) == true)
936 {
937 ::SparsityPattern::iterator select_index =
938 sparsity_pattern.begin(row);
939 typename ::SparseMatrix<number>::const_iterator it =
940 dealii_sparse_matrix.begin(row);
941 size_type col = 0;
942 if (sparsity_pattern.n_rows() == sparsity_pattern.n_cols())
943 {
944 // optimized diagonal
945 AssertDimension(it->column(), row);
946 if (std::fabs(it->value()) > drop_tolerance)
947 {
948 values[col] = it->value();
949 row_indices[col++] = it->column();
950 }
951 ++select_index;
952 ++it;
953 }
954
955 while (it != dealii_sparse_matrix.end(row) &&
956 select_index != sparsity_pattern.end(row))
957 {
958 while (select_index->column() < it->column() &&
959 select_index != sparsity_pattern.end(row))
960 ++select_index;
961 while (it->column() < select_index->column() &&
962 it != dealii_sparse_matrix.end(row))
963 ++it;
964
965 if (it == dealii_sparse_matrix.end(row))
966 break;
967 if (std::fabs(it->value()) > drop_tolerance)
968 {
969 values[col] = it->value();
970 row_indices[col++] = it->column();
971 }
972 ++select_index;
973 ++it;
974 }
975 set(row,
976 col,
977 reinterpret_cast<size_type *>(row_indices.data()),
978 values.data(),
979 false);
980 }
982 }
983
984
985
986 template <typename number>
987 void
989 const ::SparseMatrix<number> &dealii_sparse_matrix,
990 const double drop_tolerance,
991 const bool copy_values,
992 const ::SparsityPattern * use_this_sparsity)
993 {
994 reinit(complete_index_set(dealii_sparse_matrix.m()),
995 complete_index_set(dealii_sparse_matrix.n()),
996 dealii_sparse_matrix,
997 MPI_COMM_SELF,
998 drop_tolerance,
999 copy_values,
1000 use_this_sparsity);
1001 }
1002
1003
1004
1005 void
1006 SparseMatrix::reinit(const Epetra_CrsMatrix &input_matrix,
1007 const bool copy_values)
1008 {
1009 Assert(input_matrix.Filled() == true,
1010 ExcMessage("Input CrsMatrix has not called FillComplete()!"));
1011
1012 column_space_map = std::make_unique<Epetra_Map>(input_matrix.DomainMap());
1013
1014 const Epetra_CrsGraph *graph = &input_matrix.Graph();
1015
1016 nonlocal_matrix.reset();
1018 matrix.reset();
1019 matrix = std::make_unique<Epetra_FECrsMatrix>(Copy, *graph, false);
1020
1021 matrix->FillComplete(*column_space_map, input_matrix.RangeMap(), true);
1022
1023 if (copy_values == true)
1024 {
1025 // point to the first data entry in the two
1026 // matrices and copy the content
1027 const TrilinosScalar *in_values = input_matrix[0];
1028 TrilinosScalar * values = (*matrix)[0];
1029 const size_type my_nonzeros = input_matrix.NumMyNonzeros();
1030 std::memcpy(values, in_values, my_nonzeros * sizeof(TrilinosScalar));
1031 }
1032
1033 last_action = Zero;
1035 }
1036
1037
1038
1039 void
1041 {
1042 Epetra_CombineMode mode = last_action;
1043 if (last_action == Zero)
1044 {
1045 if ((operation == VectorOperation::add) ||
1046 (operation == VectorOperation::unknown))
1047 mode = Add;
1048 else if (operation == VectorOperation::insert)
1049 mode = Insert;
1050 else
1051 Assert(
1052 false,
1053 ExcMessage(
1054 "compress() can only be called with VectorOperation add, insert, or unknown"));
1055 }
1056 else
1057 {
1058 Assert(
1059 ((last_action == Add) && (operation != VectorOperation::insert)) ||
1060 ((last_action == Insert) && (operation != VectorOperation::add)),
1061 ExcMessage("Operation and argument to compress() do not match"));
1062 }
1063
1064 // flush buffers
1065 int ierr;
1066 if (nonlocal_matrix.get() != nullptr && mode == Add)
1067 {
1068 // do only export in case of an add() operation, otherwise the owning
1069 // processor must have set the correct entry
1070 nonlocal_matrix->FillComplete(*column_space_map, matrix->RowMap());
1071 if (nonlocal_matrix_exporter.get() == nullptr)
1073 std::make_unique<Epetra_Export>(nonlocal_matrix->RowMap(),
1074 matrix->RowMap());
1075 ierr =
1077 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1078 ierr = matrix->FillComplete(*column_space_map, matrix->RowMap());
1079 nonlocal_matrix->PutScalar(0);
1080 }
1081 else
1082 ierr =
1083 matrix->GlobalAssemble(*column_space_map, matrix->RowMap(), true, mode);
1084
1085 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1086
1087 ierr = matrix->OptimizeStorage();
1088 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1089
1090 last_action = Zero;
1091
1092 compressed = true;
1093 }
1094
1095
1096
1097 void
1099 {
1100 // When we clear the matrix, reset
1101 // the pointer and generate an
1102 // empty matrix.
1104 std::make_unique<Epetra_Map>(0, 0, Utilities::Trilinos::comm_self());
1105 matrix = std::make_unique<Epetra_FECrsMatrix>(View, *column_space_map, 0);
1106 nonlocal_matrix.reset();
1108
1109 matrix->FillComplete();
1110
1111 compressed = true;
1112 }
1113
1114
1115
1116 void
1118 const TrilinosScalar new_diag_value)
1119 {
1120 Assert(matrix->Filled() == true, ExcMatrixNotCompressed());
1121
1122 // Only do this on the rows owned
1123 // locally on this processor.
1124 int local_row =
1125 matrix->LRID(static_cast<TrilinosWrappers::types::int_type>(row));
1126 if (local_row >= 0)
1127 {
1128 TrilinosScalar *values;
1129 int * col_indices;
1130 int num_entries;
1131 const int ierr =
1132 matrix->ExtractMyRowView(local_row, num_entries, values, col_indices);
1133 (void)ierr;
1134
1135 Assert(ierr == 0, ExcTrilinosError(ierr));
1136
1137 const std::ptrdiff_t diag_index =
1138 std::find(col_indices, col_indices + num_entries, local_row) -
1139 col_indices;
1140
1141 for (TrilinosWrappers::types::int_type j = 0; j < num_entries; ++j)
1142 if (diag_index != j || new_diag_value == 0)
1143 values[j] = 0.;
1144
1145 if (diag_index != num_entries)
1146 values[diag_index] = new_diag_value;
1147 }
1148 }
1149
1150
1151
1152 void
1153 SparseMatrix::clear_rows(const std::vector<size_type> &rows,
1154 const TrilinosScalar new_diag_value)
1155 {
1156 for (const auto row : rows)
1157 clear_row(row, new_diag_value);
1158 }
1159
1160
1161
1164 {
1165 // Extract local indices in
1166 // the matrix.
1167 int trilinos_i =
1168 matrix->LRID(static_cast<TrilinosWrappers::types::int_type>(i)),
1169 trilinos_j =
1170 matrix->LCID(static_cast<TrilinosWrappers::types::int_type>(j));
1171 TrilinosScalar value = 0.;
1172
1173 // If the data is not on the
1174 // present processor, we throw
1175 // an exception. This is one of
1176 // the two tiny differences to
1177 // the el(i,j) call, which does
1178 // not throw any assertions.
1179 if (trilinos_i == -1)
1180 {
1181 Assert(false,
1183 i, j, local_range().first, local_range().second - 1));
1184 }
1185 else
1186 {
1187 // Check whether the matrix has
1188 // already been transformed to local
1189 // indices.
1190 Assert(matrix->Filled(), ExcMatrixNotCompressed());
1191
1192 // Prepare pointers for extraction
1193 // of a view of the row.
1194 int nnz_present = matrix->NumMyEntries(trilinos_i);
1195 int nnz_extracted;
1196 int * col_indices;
1197 TrilinosScalar *values;
1198
1199 // Generate the view and make
1200 // sure that we have not generated
1201 // an error.
1202 // TODO Check that col_indices are int and not long long
1203 int ierr = matrix->ExtractMyRowView(trilinos_i,
1204 nnz_extracted,
1205 values,
1206 col_indices);
1207 (void)ierr;
1208 Assert(ierr == 0, ExcTrilinosError(ierr));
1209
1210 Assert(nnz_present == nnz_extracted,
1211 ExcDimensionMismatch(nnz_present, nnz_extracted));
1212
1213 // Search the index where we
1214 // look for the value, and then
1215 // finally get it.
1216 const std::ptrdiff_t local_col_index =
1217 std::find(col_indices, col_indices + nnz_present, trilinos_j) -
1218 col_indices;
1219
1220 // This is actually the only
1221 // difference to the el(i,j)
1222 // function, which means that
1223 // we throw an exception in
1224 // this case instead of just
1225 // returning zero for an
1226 // element that is not present
1227 // in the sparsity pattern.
1228 if (local_col_index == nnz_present)
1229 {
1230 Assert(false, ExcInvalidIndex(i, j));
1231 }
1232 else
1233 value = values[local_col_index];
1234 }
1235
1236 return value;
1237 }
1238
1239
1240
1242 SparseMatrix::el(const size_type i, const size_type j) const
1243 {
1244 // Extract local indices in
1245 // the matrix.
1246 int trilinos_i =
1247 matrix->LRID(static_cast<TrilinosWrappers::types::int_type>(i)),
1248 trilinos_j =
1249 matrix->LCID(static_cast<TrilinosWrappers::types::int_type>(j));
1250 TrilinosScalar value = 0.;
1251
1252 // If the data is not on the
1253 // present processor, we can't
1254 // continue. Just print out zero
1255 // as discussed in the
1256 // documentation of this
1257 // function. if you want error
1258 // checking, use operator().
1259 if ((trilinos_i == -1) || (trilinos_j == -1))
1260 return 0.;
1261 else
1262 {
1263 // Check whether the matrix
1264 // already is transformed to
1265 // local indices.
1266 Assert(matrix->Filled(), ExcMatrixNotCompressed());
1267
1268 // Prepare pointers for extraction
1269 // of a view of the row.
1270 int nnz_present = matrix->NumMyEntries(trilinos_i);
1271 int nnz_extracted;
1272 int * col_indices;
1273 TrilinosScalar *values;
1274
1275 // Generate the view and make
1276 // sure that we have not generated
1277 // an error.
1278 int ierr = matrix->ExtractMyRowView(trilinos_i,
1279 nnz_extracted,
1280 values,
1281 col_indices);
1282 (void)ierr;
1283 Assert(ierr == 0, ExcTrilinosError(ierr));
1284
1285 Assert(nnz_present == nnz_extracted,
1286 ExcDimensionMismatch(nnz_present, nnz_extracted));
1287
1288 // Search the index where we
1289 // look for the value, and then
1290 // finally get it.
1291 const std::ptrdiff_t local_col_index =
1292 std::find(col_indices, col_indices + nnz_present, trilinos_j) -
1293 col_indices;
1294
1295 // This is actually the only
1296 // difference to the () function
1297 // querying (i,j), where we throw an
1298 // exception instead of just
1299 // returning zero for an element
1300 // that is not present in the
1301 // sparsity pattern.
1302 if (local_col_index == nnz_present)
1303 value = 0;
1304 else
1305 value = values[local_col_index];
1306 }
1307
1308 return value;
1309 }
1310
1311
1312
1315 {
1316 Assert(m() == n(), ExcNotQuadratic());
1317
1318# ifdef DEBUG
1319 // use operator() in debug mode because
1320 // it checks if this is a valid element
1321 // (in parallel)
1322 return operator()(i, i);
1323# else
1324 // Trilinos doesn't seem to have a
1325 // more efficient way to access the
1326 // diagonal than by just using the
1327 // standard el(i,j) function.
1328 return el(i, i);
1329# endif
1330 }
1331
1332
1333
1334 unsigned int
1336 {
1337 Assert(row < m(), ExcInternalError());
1338
1339 // get a representation of the
1340 // present row
1341 int ncols = -1;
1342 int local_row =
1343 matrix->LRID(static_cast<TrilinosWrappers::types::int_type>(row));
1344 Assert((local_row >= 0), ExcAccessToNonlocalRow(row));
1345
1346 // on the processor who owns this
1347 // row, we'll have a non-negative
1348 // value.
1349 if (local_row >= 0)
1350 {
1351 int ierr = matrix->NumMyRowEntries(local_row, ncols);
1352 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1353 }
1354
1355 return static_cast<unsigned int>(ncols);
1356 }
1357
1358
1359
1360 void
1361 SparseMatrix::set(const std::vector<size_type> & row_indices,
1362 const std::vector<size_type> & col_indices,
1363 const FullMatrix<TrilinosScalar> &values,
1364 const bool elide_zero_values)
1365 {
1366 Assert(row_indices.size() == values.m(),
1367 ExcDimensionMismatch(row_indices.size(), values.m()));
1368 Assert(col_indices.size() == values.n(),
1369 ExcDimensionMismatch(col_indices.size(), values.n()));
1370
1371 for (size_type i = 0; i < row_indices.size(); ++i)
1372 set(row_indices[i],
1373 col_indices.size(),
1374 col_indices.data(),
1375 &values(i, 0),
1376 elide_zero_values);
1377 }
1378
1379
1380
1381 void
1383 const std::vector<size_type> & col_indices,
1384 const std::vector<TrilinosScalar> &values,
1385 const bool elide_zero_values)
1386 {
1387 Assert(col_indices.size() == values.size(),
1388 ExcDimensionMismatch(col_indices.size(), values.size()));
1389
1390 set(row,
1391 col_indices.size(),
1392 col_indices.data(),
1393 values.data(),
1394 elide_zero_values);
1395 }
1396
1397
1398
1399 template <>
1400 void
1401 SparseMatrix::set<TrilinosScalar>(const size_type row,
1402 const size_type n_cols,
1403 const size_type * col_indices,
1404 const TrilinosScalar *values,
1405 const bool elide_zero_values)
1406 {
1407 AssertIndexRange(row, this->m());
1408
1409 int ierr;
1410 if (last_action == Add)
1411 {
1412 ierr =
1413 matrix->GlobalAssemble(*column_space_map, matrix->RowMap(), true);
1414
1415 Assert(ierr == 0, ExcTrilinosError(ierr));
1416 }
1417
1418 last_action = Insert;
1419
1420 const TrilinosWrappers::types::int_type *col_index_ptr;
1421 const TrilinosScalar * col_value_ptr;
1422 const TrilinosWrappers::types::int_type trilinos_row = row;
1424
1425 boost::container::small_vector<TrilinosScalar, 200> local_value_array(
1426 n_cols);
1427 boost::container::small_vector<TrilinosWrappers::types::int_type, 200>
1428 local_index_array(n_cols);
1429
1430 // If we don't elide zeros, the pointers are already available... need to
1431 // cast to non-const pointers as that is the format taken by Trilinos (but
1432 // we will not modify const data)
1433 if (elide_zero_values == false)
1434 {
1435 col_index_ptr =
1436 reinterpret_cast<const TrilinosWrappers::types::int_type *>(
1437 col_indices);
1438 col_value_ptr = values;
1439 n_columns = n_cols;
1440 }
1441 else
1442 {
1443 // Otherwise, extract nonzero values in each row and get the
1444 // respective indices.
1445 col_index_ptr = local_index_array.data();
1446 col_value_ptr = local_value_array.data();
1447
1448 n_columns = 0;
1449 for (size_type j = 0; j < n_cols; ++j)
1450 {
1451 const double value = values[j];
1452 AssertIsFinite(value);
1453 if (value != 0)
1454 {
1455 local_index_array[n_columns] = col_indices[j];
1456 local_value_array[n_columns] = value;
1457 n_columns++;
1458 }
1459 }
1460
1461 AssertIndexRange(n_columns, n_cols + 1);
1462 }
1463
1464
1465 // If the calling matrix owns the row to which we want to insert values,
1466 // we can directly call the Epetra_CrsMatrix input function, which is much
1467 // faster than the Epetra_FECrsMatrix function. We distinguish between two
1468 // cases: the first one is when the matrix is not filled (i.e., it is
1469 // possible to add new elements to the sparsity pattern), and the second
1470 // one is when the pattern is already fixed. In the former case, we add
1471 // the possibility to insert new values, and in the second we just replace
1472 // data.
1473 if (matrix->RowMap().MyGID(
1474 static_cast<TrilinosWrappers::types::int_type>(row)) == true)
1475 {
1476 if (matrix->Filled() == false)
1477 {
1478 ierr = matrix->Epetra_CrsMatrix::InsertGlobalValues(
1479 row, static_cast<int>(n_columns), col_value_ptr, col_index_ptr);
1480
1481 // When inserting elements, we do not want to create exceptions in
1482 // the case when inserting non-local data (since that's what we
1483 // want to do right now).
1484 if (ierr > 0)
1485 ierr = 0;
1486 }
1487 else
1488 ierr = matrix->Epetra_CrsMatrix::ReplaceGlobalValues(row,
1489 n_columns,
1490 col_value_ptr,
1491 col_index_ptr);
1492 }
1493 else
1494 {
1495 // When we're at off-processor data, we have to stick with the
1496 // standard Insert/ReplaceGlobalValues function. Nevertheless, the way
1497 // we call it is the fastest one (any other will lead to repeated
1498 // allocation and deallocation of memory in order to call the function
1499 // we already use, which is very inefficient if writing one element at
1500 // a time).
1501 compressed = false;
1502
1503 if (matrix->Filled() == false)
1504 {
1505 ierr = matrix->InsertGlobalValues(1,
1506 &trilinos_row,
1507 n_columns,
1508 col_index_ptr,
1509 &col_value_ptr,
1510 Epetra_FECrsMatrix::ROW_MAJOR);
1511 if (ierr > 0)
1512 ierr = 0;
1513 }
1514 else
1515 ierr = matrix->ReplaceGlobalValues(1,
1516 &trilinos_row,
1517 n_columns,
1518 col_index_ptr,
1519 &col_value_ptr,
1520 Epetra_FECrsMatrix::ROW_MAJOR);
1521 // use the FECrsMatrix facilities for set even in the case when we
1522 // have explicitly set the off-processor rows because that only works
1523 // properly when adding elements, not when setting them (since we want
1524 // to only touch elements that have been set explicitly, and there is
1525 // no way on the receiving processor to identify them otherwise)
1526 }
1527
1528 Assert(ierr <= 0, ExcAccessToNonPresentElement(row, col_index_ptr[0]));
1529 AssertThrow(ierr >= 0, ExcTrilinosError(ierr));
1530 }
1531
1532
1533
1534 void
1535 SparseMatrix::add(const std::vector<size_type> & indices,
1536 const FullMatrix<TrilinosScalar> &values,
1537 const bool elide_zero_values)
1538 {
1539 Assert(indices.size() == values.m(),
1540 ExcDimensionMismatch(indices.size(), values.m()));
1541 Assert(values.m() == values.n(), ExcNotQuadratic());
1542
1543 for (size_type i = 0; i < indices.size(); ++i)
1544 add(indices[i],
1545 indices.size(),
1546 indices.data(),
1547 &values(i, 0),
1548 elide_zero_values);
1549 }
1550
1551
1552
1553 void
1554 SparseMatrix::add(const std::vector<size_type> & row_indices,
1555 const std::vector<size_type> & col_indices,
1556 const FullMatrix<TrilinosScalar> &values,
1557 const bool elide_zero_values)
1558 {
1559 Assert(row_indices.size() == values.m(),
1560 ExcDimensionMismatch(row_indices.size(), values.m()));
1561 Assert(col_indices.size() == values.n(),
1562 ExcDimensionMismatch(col_indices.size(), values.n()));
1563
1564 for (size_type i = 0; i < row_indices.size(); ++i)
1565 add(row_indices[i],
1566 col_indices.size(),
1567 col_indices.data(),
1568 &values(i, 0),
1569 elide_zero_values);
1570 }
1571
1572
1573
1574 void
1576 const std::vector<size_type> & col_indices,
1577 const std::vector<TrilinosScalar> &values,
1578 const bool elide_zero_values)
1579 {
1580 Assert(col_indices.size() == values.size(),
1581 ExcDimensionMismatch(col_indices.size(), values.size()));
1582
1583 add(row,
1584 col_indices.size(),
1585 col_indices.data(),
1586 values.data(),
1587 elide_zero_values);
1588 }
1589
1590
1591
1592 void
1594 const size_type n_cols,
1595 const size_type * col_indices,
1596 const TrilinosScalar *values,
1597 const bool elide_zero_values,
1598 const bool /*col_indices_are_sorted*/)
1599 {
1600 AssertIndexRange(row, this->m());
1601 int ierr;
1602 if (last_action == Insert)
1603 {
1604 // TODO: this could lead to a dead lock when only one processor
1605 // calls GlobalAssemble.
1606 ierr =
1607 matrix->GlobalAssemble(*column_space_map, matrix->RowMap(), false);
1608
1609 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1610 }
1611
1612 last_action = Add;
1613
1614 const TrilinosWrappers::types::int_type *col_index_ptr;
1615 const TrilinosScalar * col_value_ptr;
1616 const TrilinosWrappers::types::int_type trilinos_row = row;
1618
1619 boost::container::small_vector<TrilinosScalar, 100> local_value_array(
1620 n_cols);
1621 boost::container::small_vector<TrilinosWrappers::types::int_type, 100>
1622 local_index_array(n_cols);
1623
1624 // If we don't elide zeros, the pointers are already available... need to
1625 // cast to non-const pointers as that is the format taken by Trilinos (but
1626 // we will not modify const data)
1627 if (elide_zero_values == false)
1628 {
1629 col_index_ptr =
1630 reinterpret_cast<const TrilinosWrappers::types::int_type *>(
1631 col_indices);
1632 col_value_ptr = values;
1633 n_columns = n_cols;
1634# ifdef DEBUG
1635 for (size_type j = 0; j < n_cols; ++j)
1636 AssertIsFinite(values[j]);
1637# endif
1638 }
1639 else
1640 {
1641 // Otherwise, extract nonzero values in each row and the corresponding
1642 // index.
1643 col_index_ptr = local_index_array.data();
1644 col_value_ptr = local_value_array.data();
1645
1646 n_columns = 0;
1647 for (size_type j = 0; j < n_cols; ++j)
1648 {
1649 const double value = values[j];
1650
1651 AssertIsFinite(value);
1652 if (value != 0)
1653 {
1654 local_index_array[n_columns] = col_indices[j];
1655 local_value_array[n_columns] = value;
1656 ++n_columns;
1657 }
1658 }
1659
1660 AssertIndexRange(n_columns, n_cols + 1);
1661 }
1662 // Exit early if there is nothing to do
1663 if (n_columns == 0)
1664 {
1665 return;
1666 }
1667
1668 // If the calling processor owns the row to which we want to add values, we
1669 // can directly call the Epetra_CrsMatrix input function, which is much
1670 // faster than the Epetra_FECrsMatrix function.
1671 if (matrix->RowMap().MyGID(
1672 static_cast<TrilinosWrappers::types::int_type>(row)) == true)
1673 {
1674 ierr = matrix->Epetra_CrsMatrix::SumIntoGlobalValues(row,
1675 n_columns,
1676 col_value_ptr,
1677 col_index_ptr);
1678 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1679 }
1680 else if (nonlocal_matrix.get() != nullptr)
1681 {
1682 compressed = false;
1683 // this is the case when we have explicitly set the off-processor rows
1684 // and want to create a separate matrix object for them (to retain
1685 // thread-safety)
1686 Assert(nonlocal_matrix->RowMap().LID(
1687 static_cast<TrilinosWrappers::types::int_type>(row)) != -1,
1688 ExcMessage("Attempted to write into off-processor matrix row "
1689 "that has not be specified as being writable upon "
1690 "initialization"));
1691 ierr = nonlocal_matrix->SumIntoGlobalValues(row,
1692 n_columns,
1693 col_value_ptr,
1694 col_index_ptr);
1695 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1696 }
1697 else
1698 {
1699 // When we're at off-processor data, we have to stick with the
1700 // standard SumIntoGlobalValues function. Nevertheless, the way we
1701 // call it is the fastest one (any other will lead to repeated
1702 // allocation and deallocation of memory in order to call the function
1703 // we already use, which is very inefficient if writing one element at
1704 // a time).
1705 compressed = false;
1706
1707 ierr = matrix->SumIntoGlobalValues(1,
1708 &trilinos_row,
1709 n_columns,
1710 col_index_ptr,
1711 &col_value_ptr,
1712 Epetra_FECrsMatrix::ROW_MAJOR);
1713 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1714 }
1715
1716# ifdef DEBUG
1717 if (ierr > 0)
1718 {
1719 std::cout << "------------------------------------------" << std::endl;
1720 std::cout << "Got error " << ierr << " in row " << row << " of proc "
1721 << matrix->RowMap().Comm().MyPID()
1722 << " when trying to add the columns:" << std::endl;
1723 for (TrilinosWrappers::types::int_type i = 0; i < n_columns; ++i)
1724 std::cout << col_index_ptr[i] << " ";
1725 std::cout << std::endl << std::endl;
1726 std::cout << "Matrix row "
1727 << (matrix->RowMap().MyGID(
1728 static_cast<TrilinosWrappers::types::int_type>(row)) ==
1729 false ?
1730 "(nonlocal part)" :
1731 "")
1732 << " has the following indices:" << std::endl;
1733 std::vector<TrilinosWrappers::types::int_type> indices;
1734 const Epetra_CrsGraph * graph =
1735 (nonlocal_matrix.get() != nullptr &&
1736 matrix->RowMap().MyGID(
1737 static_cast<TrilinosWrappers::types::int_type>(row)) == false) ?
1738 &nonlocal_matrix->Graph() :
1739 &matrix->Graph();
1740
1741 indices.resize(graph->NumGlobalIndices(row));
1742 int n_indices = 0;
1743 graph->ExtractGlobalRowCopy(row,
1744 indices.size(),
1745 n_indices,
1746 indices.data());
1747 AssertDimension(n_indices, indices.size());
1748
1749 for (TrilinosWrappers::types::int_type i = 0; i < n_indices; ++i)
1750 std::cout << indices[i] << " ";
1751 std::cout << std::endl << std::endl;
1752 Assert(ierr <= 0, ExcAccessToNonPresentElement(row, col_index_ptr[0]));
1753 }
1754# endif
1755 Assert(ierr >= 0, ExcTrilinosError(ierr));
1756 }
1757
1758
1759
1760 SparseMatrix &
1762 {
1764 compress(VectorOperation::unknown); // TODO: why do we do this? Should we
1765 // not check for is_compressed?
1766
1767 const int ierr = matrix->PutScalar(d);
1768 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1769 if (nonlocal_matrix.get() != nullptr)
1770 nonlocal_matrix->PutScalar(d);
1771
1772 return *this;
1773 }
1774
1775
1776
1777 void
1779 {
1780 AssertDimension(rhs.m(), m());
1781 AssertDimension(rhs.n(), n());
1784 Assert(matrix->RowMap().SameAs(rhs.matrix->RowMap()),
1785 ExcMessage("Can only add matrices with same distribution of rows"));
1786 Assert(matrix->Filled() && rhs.matrix->Filled(),
1787 ExcMessage("Addition of matrices only allowed if matrices are "
1788 "filled, i.e., compress() has been called"));
1789
1790 const bool same_col_map = matrix->ColMap().SameAs(rhs.matrix->ColMap());
1791
1792 for (const auto row : locally_owned_range_indices())
1793 {
1794 const int row_local = matrix->RowMap().LID(
1795 static_cast<TrilinosWrappers::types::int_type>(row));
1796 Assert((row_local >= 0), ExcAccessToNonlocalRow(row));
1797
1798 // First get a view to the matrix columns of both matrices. Note that
1799 // the data is in local index spaces so we need to be careful not only
1800 // to compare column indices in case they are derived from the same
1801 // map.
1802 int n_entries, rhs_n_entries;
1803 TrilinosScalar *value_ptr, *rhs_value_ptr;
1804 int * index_ptr, *rhs_index_ptr;
1805 int ierr = rhs.matrix->ExtractMyRowView(row_local,
1806 rhs_n_entries,
1807 rhs_value_ptr,
1808 rhs_index_ptr);
1809 (void)ierr;
1810 Assert(ierr == 0, ExcTrilinosError(ierr));
1811
1812 ierr =
1813 matrix->ExtractMyRowView(row_local, n_entries, value_ptr, index_ptr);
1814 Assert(ierr == 0, ExcTrilinosError(ierr));
1815 bool expensive_checks = (n_entries != rhs_n_entries || !same_col_map);
1816 if (!expensive_checks)
1817 {
1818 // check if the column indices are the same. If yes, can simply
1819 // copy over the data.
1820 expensive_checks = std::memcmp(static_cast<void *>(index_ptr),
1821 static_cast<void *>(rhs_index_ptr),
1822 sizeof(int) * n_entries) != 0;
1823 if (!expensive_checks)
1824 for (int i = 0; i < n_entries; ++i)
1825 value_ptr[i] += rhs_value_ptr[i] * factor;
1826 }
1827 // Now to the expensive case where we need to check all column indices
1828 // against each other (transformed into global index space) and where
1829 // we need to make sure that all entries we are about to add into the
1830 // lhs matrix actually exist
1831 if (expensive_checks)
1832 {
1833 for (int i = 0; i < rhs_n_entries; ++i)
1834 {
1835 if (rhs_value_ptr[i] == 0.)
1836 continue;
1837 const TrilinosWrappers::types::int_type rhs_global_col =
1838 global_column_index(*rhs.matrix, rhs_index_ptr[i]);
1839 int local_col = matrix->ColMap().LID(rhs_global_col);
1840 int *local_index = Utilities::lower_bound(index_ptr,
1841 index_ptr + n_entries,
1842 local_col);
1843 Assert(local_index != index_ptr + n_entries &&
1844 *local_index == local_col,
1845 ExcMessage(
1846 "Adding the entries from the other matrix "
1847 "failed, because the sparsity pattern "
1848 "of that matrix includes more elements than the "
1849 "calling matrix, which is not allowed."));
1850 value_ptr[local_index - index_ptr] += factor * rhs_value_ptr[i];
1851 }
1852 }
1853 }
1854 }
1855
1856
1857
1858 void
1860 {
1861 // This only flips a flag that tells
1862 // Trilinos that any vmult operation
1863 // should be done with the
1864 // transpose. However, the matrix
1865 // structure is not reset.
1866 int ierr;
1867
1868 if (!matrix->UseTranspose())
1869 {
1870 ierr = matrix->SetUseTranspose(true);
1871 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1872 }
1873 else
1874 {
1875 ierr = matrix->SetUseTranspose(false);
1876 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
1877 }
1878 }
1879
1880
1881
1882 SparseMatrix &
1884 {
1885 const int ierr = matrix->Scale(a);
1886 Assert(ierr == 0, ExcTrilinosError(ierr));
1887 (void)ierr; // removes -Wunused-variable in optimized mode
1888
1889 return *this;
1890 }
1891
1892
1893
1894 SparseMatrix &
1896 {
1897 Assert(a != 0, ExcDivideByZero());
1898
1899 const TrilinosScalar factor = 1. / a;
1900
1901 const int ierr = matrix->Scale(factor);
1902 Assert(ierr == 0, ExcTrilinosError(ierr));
1903 (void)ierr; // removes -Wunused-variable in optimized mode
1904
1905 return *this;
1906 }
1907
1908
1909
1912 {
1913 Assert(matrix->Filled(), ExcMatrixNotCompressed());
1914 return matrix->NormOne();
1915 }
1916
1917
1918
1921 {
1922 Assert(matrix->Filled(), ExcMatrixNotCompressed());
1923 return matrix->NormInf();
1924 }
1925
1926
1927
1930 {
1931 Assert(matrix->Filled(), ExcMatrixNotCompressed());
1932 return matrix->NormFrobenius();
1933 }
1934
1935
1936
1937 namespace internal
1938 {
1939 namespace SparseMatrixImplementation
1940 {
1941 template <typename VectorType>
1942 inline void
1943 check_vector_map_equality(const Epetra_CrsMatrix &,
1944 const VectorType &,
1945 const VectorType &)
1946 {}
1947
1948 inline void
1949 check_vector_map_equality(const Epetra_CrsMatrix & m,
1952 {
1953 Assert(in.trilinos_partitioner().SameAs(m.DomainMap()) == true,
1954 ExcMessage("The column partitioning of a matrix does not match "
1955 "the partitioning of a vector you are trying to "
1956 "multiply it with. Are you multiplying the "
1957 "matrix with a vector that has ghost elements?"));
1958 Assert(out.trilinos_partitioner().SameAs(m.RangeMap()) == true,
1959 ExcMessage("The row partitioning of a matrix does not match "
1960 "the partitioning of a vector you are trying to "
1961 "put the result of a matrix-vector product in. "
1962 "Are you trying to put the product of the "
1963 "matrix with a vector into a vector that has "
1964 "ghost elements?"));
1965 (void)m;
1966 (void)in;
1967 (void)out;
1968 }
1969 } // namespace SparseMatrixImplementation
1970 } // namespace internal
1971
1972
1973 template <typename VectorType>
1974 std::enable_if_t<
1975 std::is_same<typename VectorType::value_type, TrilinosScalar>::value>
1976 SparseMatrix::vmult(VectorType &dst, const VectorType &src) const
1977 {
1978 Assert(&src != &dst, ExcSourceEqualsDestination());
1979 Assert(matrix->Filled(), ExcMatrixNotCompressed());
1980 (void)src;
1981 (void)dst;
1982
1984 src,
1985 dst);
1986 const size_type dst_local_size = internal::end(dst) - internal::begin(dst);
1987 AssertDimension(dst_local_size, matrix->RangeMap().NumMyPoints());
1988 const size_type src_local_size = internal::end(src) - internal::begin(src);
1989 AssertDimension(src_local_size, matrix->DomainMap().NumMyPoints());
1990
1991 Epetra_MultiVector tril_dst(
1992 View, matrix->RangeMap(), internal::begin(dst), dst_local_size, 1);
1993 Epetra_MultiVector tril_src(View,
1994 matrix->DomainMap(),
1995 const_cast<TrilinosScalar *>(
1996 internal::begin(src)),
1997 src_local_size,
1998 1);
1999
2000 const int ierr = matrix->Multiply(false, tril_src, tril_dst);
2001 Assert(ierr == 0, ExcTrilinosError(ierr));
2002 (void)ierr; // removes -Wunused-variable in optimized mode
2003 }
2004
2005
2006
2007 template <typename VectorType>
2008 std::enable_if_t<
2009 !std::is_same<typename VectorType::value_type, TrilinosScalar>::value>
2010 SparseMatrix::vmult(VectorType & /*dst*/, const VectorType & /*src*/) const
2011 {
2013 }
2014
2015
2016
2017 template <typename VectorType>
2018 std::enable_if_t<
2019 std::is_same<typename VectorType::value_type, TrilinosScalar>::value>
2020 SparseMatrix::Tvmult(VectorType &dst, const VectorType &src) const
2021 {
2022 Assert(&src != &dst, ExcSourceEqualsDestination());
2023 Assert(matrix->Filled(), ExcMatrixNotCompressed());
2024
2026 dst,
2027 src);
2028 const size_type dst_local_size = internal::end(dst) - internal::begin(dst);
2029 AssertDimension(dst_local_size, matrix->DomainMap().NumMyPoints());
2030 const size_type src_local_size = internal::end(src) - internal::begin(src);
2031 AssertDimension(src_local_size, matrix->RangeMap().NumMyPoints());
2032
2033 Epetra_MultiVector tril_dst(
2034 View, matrix->DomainMap(), internal::begin(dst), dst_local_size, 1);
2035 Epetra_MultiVector tril_src(View,
2036 matrix->RangeMap(),
2037 const_cast<double *>(internal::begin(src)),
2038 src_local_size,
2039 1);
2040
2041 const int ierr = matrix->Multiply(true, tril_src, tril_dst);
2042 Assert(ierr == 0, ExcTrilinosError(ierr));
2043 (void)ierr; // removes -Wunused-variable in optimized mode
2044 }
2045
2046
2047
2048 template <typename VectorType>
2049 std::enable_if_t<
2050 !std::is_same<typename VectorType::value_type, TrilinosScalar>::value>
2051 SparseMatrix::Tvmult(VectorType & /*dst*/, const VectorType & /*src*/) const
2052 {
2054 }
2055
2056
2057
2058 template <typename VectorType>
2059 void
2060 SparseMatrix::vmult_add(VectorType &dst, const VectorType &src) const
2061 {
2062 Assert(&src != &dst, ExcSourceEqualsDestination());
2063
2064 // Reinit a temporary vector with fast argument set, which does not
2065 // overwrite the content (to save time).
2066 VectorType tmp_vector;
2067 tmp_vector.reinit(dst, true);
2068 vmult(tmp_vector, src);
2069 dst += tmp_vector;
2070 }
2071
2072
2073
2074 template <typename VectorType>
2075 void
2076 SparseMatrix::Tvmult_add(VectorType &dst, const VectorType &src) const
2077 {
2078 Assert(&src != &dst, ExcSourceEqualsDestination());
2079
2080 // Reinit a temporary vector with fast argument set, which does not
2081 // overwrite the content (to save time).
2082 VectorType tmp_vector;
2083 tmp_vector.reinit(dst, true);
2084 Tvmult(tmp_vector, src);
2085 dst += tmp_vector;
2086 }
2087
2088
2089
2092 {
2093 Assert(matrix->RowMap().SameAs(matrix->DomainMap()), ExcNotQuadratic());
2094
2095 MPI::Vector temp_vector;
2096 temp_vector.reinit(v, true);
2097
2098 vmult(temp_vector, v);
2099 return temp_vector * v;
2100 }
2101
2102
2103
2106 const MPI::Vector &v) const
2107 {
2108 Assert(matrix->RowMap().SameAs(matrix->DomainMap()), ExcNotQuadratic());
2109
2110 MPI::Vector temp_vector;
2111 temp_vector.reinit(v, true);
2112
2113 vmult(temp_vector, v);
2114 return u * temp_vector;
2115 }
2116
2117
2118
2121 const MPI::Vector &x,
2122 const MPI::Vector &b) const
2123 {
2124 vmult(dst, x);
2125 dst -= b;
2126 dst *= -1.;
2127
2128 return dst.l2_norm();
2129 }
2130
2131
2132
2133 namespace internals
2134 {
2136
2137 void
2138 perform_mmult(const SparseMatrix &inputleft,
2139 const SparseMatrix &inputright,
2140 SparseMatrix & result,
2141 const MPI::Vector & V,
2142 const bool transpose_left)
2143 {
2144 const bool use_vector = (V.size() == inputright.m() ? true : false);
2145 if (transpose_left == false)
2146 {
2147 Assert(inputleft.n() == inputright.m(),
2148 ExcDimensionMismatch(inputleft.n(), inputright.m()));
2149 Assert(inputleft.trilinos_matrix().DomainMap().SameAs(
2150 inputright.trilinos_matrix().RangeMap()),
2151 ExcMessage("Parallel partitioning of A and B does not fit."));
2152 }
2153 else
2154 {
2155 Assert(inputleft.m() == inputright.m(),
2156 ExcDimensionMismatch(inputleft.m(), inputright.m()));
2157 Assert(inputleft.trilinos_matrix().RangeMap().SameAs(
2158 inputright.trilinos_matrix().RangeMap()),
2159 ExcMessage("Parallel partitioning of A and B does not fit."));
2160 }
2161
2162 result.clear();
2163
2164 // create a suitable operator B: in case
2165 // we do not use a vector, all we need to
2166 // do is to set the pointer. Otherwise,
2167 // we insert the data from B, but
2168 // multiply each row with the respective
2169 // vector element.
2170 Teuchos::RCP<Epetra_CrsMatrix> mod_B;
2171 if (use_vector == false)
2172 {
2173 mod_B = Teuchos::rcp(const_cast<Epetra_CrsMatrix *>(
2174 &inputright.trilinos_matrix()),
2175 false);
2176 }
2177 else
2178 {
2179 mod_B = Teuchos::rcp(
2180 new Epetra_CrsMatrix(Copy, inputright.trilinos_sparsity_pattern()),
2181 true);
2182 mod_B->FillComplete(inputright.trilinos_matrix().DomainMap(),
2183 inputright.trilinos_matrix().RangeMap());
2184 Assert(inputright.local_range() == V.local_range(),
2185 ExcMessage("Parallel distribution of matrix B and vector V "
2186 "does not match."));
2187
2188 const int local_N = inputright.local_size();
2189 for (int i = 0; i < local_N; ++i)
2190 {
2191 int N_entries = -1;
2192 double *new_data, *B_data;
2193 mod_B->ExtractMyRowView(i, N_entries, new_data);
2194 inputright.trilinos_matrix().ExtractMyRowView(i,
2195 N_entries,
2196 B_data);
2197 double value = V.trilinos_vector()[0][i];
2198 for (TrilinosWrappers::types::int_type j = 0; j < N_entries; ++j)
2199 new_data[j] = value * B_data[j];
2200 }
2201 }
2202
2203
2204 SparseMatrix tmp_result(transpose_left ?
2205 inputleft.locally_owned_domain_indices() :
2206 inputleft.locally_owned_range_indices(),
2207 inputright.locally_owned_domain_indices(),
2208 inputleft.get_mpi_communicator());
2209
2210# ifdef DEAL_II_TRILINOS_WITH_EPETRAEXT
2211 EpetraExt::MatrixMatrix::Multiply(inputleft.trilinos_matrix(),
2212 transpose_left,
2213 *mod_B,
2214 false,
2215 const_cast<Epetra_CrsMatrix &>(
2216 tmp_result.trilinos_matrix()));
2217# else
2218 Assert(false,
2219 ExcMessage("This function requires that the Trilinos "
2220 "installation found while running the deal.II "
2221 "CMake scripts contains the optional Trilinos "
2222 "package 'EpetraExt'. However, this optional "
2223 "part of Trilinos was not found."));
2224# endif
2225 result.reinit(tmp_result.trilinos_matrix());
2226 }
2227 } // namespace internals
2228
2229
2230 void
2232 const SparseMatrix &B,
2233 const MPI::Vector & V) const
2234 {
2235 internals::perform_mmult(*this, B, C, V, false);
2236 }
2237
2238
2239
2240 void
2242 const SparseMatrix &B,
2243 const MPI::Vector & V) const
2244 {
2245 internals::perform_mmult(*this, B, C, V, true);
2246 }
2247
2248
2249
2250 void
2252 {
2253 Assert(false, ExcNotImplemented());
2254 }
2255
2256
2257
2258 // As of now, no particularly neat
2259 // output is generated in case of
2260 // multiple processors.
2261 void
2262 SparseMatrix::print(std::ostream &out,
2263 const bool print_detailed_trilinos_information) const
2264 {
2265 if (print_detailed_trilinos_information == true)
2266 out << *matrix;
2267 else
2268 {
2269 double *values;
2270 int * indices;
2271 int num_entries;
2272
2273 for (int i = 0; i < matrix->NumMyRows(); ++i)
2274 {
2275 const int ierr =
2276 matrix->ExtractMyRowView(i, num_entries, values, indices);
2277 (void)ierr;
2278 Assert(ierr == 0, ExcTrilinosError(ierr));
2279
2280 for (TrilinosWrappers::types::int_type j = 0; j < num_entries; ++j)
2282 << ","
2284 << ") " << values[j] << std::endl;
2285 }
2286 }
2287
2288 AssertThrow(out.fail() == false, ExcIO());
2289 }
2290
2291
2292
2295 {
2296 size_type static_memory =
2297 sizeof(*this) + sizeof(*matrix) + sizeof(*matrix->Graph().DataPtr());
2298 return (
2300 matrix->NumMyNonzeros() +
2301 sizeof(int) * local_size() + static_memory);
2302 }
2303
2304
2305
2306 MPI_Comm
2308 {
2309 const Epetra_MpiComm *mpi_comm =
2310 dynamic_cast<const Epetra_MpiComm *>(&matrix->RangeMap().Comm());
2311 Assert(mpi_comm != nullptr, ExcInternalError());
2312 return mpi_comm->Comm();
2313 }
2314} // namespace TrilinosWrappers
2315
2316
2317namespace TrilinosWrappers
2318{
2319 namespace internal
2320 {
2321 namespace LinearOperatorImplementation
2322 {
2324 : use_transpose(false)
2325 , communicator(MPI_COMM_SELF)
2326 , domain_map(IndexSet().make_trilinos_map(communicator.Comm()))
2327 , range_map(IndexSet().make_trilinos_map(communicator.Comm()))
2328 {
2329 vmult = [](Range &, const Domain &) {
2330 Assert(false,
2331 ExcMessage("Uninitialized TrilinosPayload::vmult called "
2332 "(Default constructor)"));
2333 };
2334
2335 Tvmult = [](Domain &, const Range &) {
2336 Assert(false,
2337 ExcMessage("Uninitialized TrilinosPayload::Tvmult called "
2338 "(Default constructor)"));
2339 };
2340
2341 inv_vmult = [](Domain &, const Range &) {
2342 Assert(false,
2343 ExcMessage("Uninitialized TrilinosPayload::inv_vmult called "
2344 "(Default constructor)"));
2345 };
2346
2347 inv_Tvmult = [](Range &, const Domain &) {
2348 Assert(false,
2349 ExcMessage("Uninitialized TrilinosPayload::inv_Tvmult called "
2350 "(Default constructor)"));
2351 };
2352 }
2353
2354
2355
2357 const TrilinosWrappers::SparseMatrix &matrix_exemplar,
2358 const TrilinosWrappers::SparseMatrix &matrix)
2359 : TrilinosPayload(const_cast<Epetra_CrsMatrix &>(
2360 matrix.trilinos_matrix()),
2361 /*op_supports_inverse_operations = */ false,
2362 matrix_exemplar.trilinos_matrix().UseTranspose(),
2363 matrix_exemplar.get_mpi_communicator(),
2364 matrix_exemplar.locally_owned_domain_indices(),
2365 matrix_exemplar.locally_owned_range_indices())
2366 {}
2367
2368
2369
2371 const TrilinosPayload & payload_exemplar,
2372 const TrilinosWrappers::SparseMatrix &matrix)
2373
2374 : TrilinosPayload(const_cast<Epetra_CrsMatrix &>(
2375 matrix.trilinos_matrix()),
2376 /*op_supports_inverse_operations = */ false,
2377 payload_exemplar.UseTranspose(),
2378 payload_exemplar.get_mpi_communicator(),
2379 payload_exemplar.locally_owned_domain_indices(),
2380 payload_exemplar.locally_owned_range_indices())
2381 {}
2382
2383
2384
2386 const TrilinosWrappers::SparseMatrix & matrix_exemplar,
2387 const TrilinosWrappers::PreconditionBase &preconditioner)
2388 : TrilinosPayload(preconditioner.trilinos_operator(),
2389 /*op_supports_inverse_operations = */ true,
2390 matrix_exemplar.trilinos_matrix().UseTranspose(),
2391 matrix_exemplar.get_mpi_communicator(),
2392 matrix_exemplar.locally_owned_domain_indices(),
2393 matrix_exemplar.locally_owned_range_indices())
2394 {}
2395
2396
2397
2399 const TrilinosWrappers::PreconditionBase &preconditioner_exemplar,
2400 const TrilinosWrappers::PreconditionBase &preconditioner)
2402 preconditioner.trilinos_operator(),
2403 /*op_supports_inverse_operations = */ true,
2404 preconditioner_exemplar.trilinos_operator().UseTranspose(),
2405 preconditioner_exemplar.get_mpi_communicator(),
2406 preconditioner_exemplar.locally_owned_domain_indices(),
2407 preconditioner_exemplar.locally_owned_range_indices())
2408 {}
2409
2410
2411
2413 const TrilinosPayload & payload_exemplar,
2414 const TrilinosWrappers::PreconditionBase &preconditioner)
2415 : TrilinosPayload(preconditioner.trilinos_operator(),
2416 /*op_supports_inverse_operations = */ true,
2417 payload_exemplar.UseTranspose(),
2418 payload_exemplar.get_mpi_communicator(),
2419 payload_exemplar.locally_owned_domain_indices(),
2420 payload_exemplar.locally_owned_range_indices())
2421 {}
2422
2423
2424
2426 : vmult(payload.vmult)
2427 , Tvmult(payload.Tvmult)
2428 , inv_vmult(payload.inv_vmult)
2429 , inv_Tvmult(payload.inv_Tvmult)
2430 , use_transpose(payload.use_transpose)
2431 , communicator(payload.communicator)
2432 , domain_map(payload.domain_map)
2433 , range_map(payload.range_map)
2434 {}
2435
2436
2437
2438 // Composite copy constructor
2439 // This is required for PackagedOperations
2441 const TrilinosPayload &second_op)
2442 : use_transpose(false)
2443 , // The combination of operators provides the exact
2444 // definition of the operation
2445 communicator(first_op.communicator)
2446 , domain_map(second_op.domain_map)
2447 , range_map(first_op.range_map)
2448 {}
2449
2450
2451
2454 {
2455 TrilinosPayload return_op(*this);
2456
2457 return_op.vmult = [](Range &tril_dst, const Range &tril_src) {
2458 tril_dst = tril_src;
2459 };
2460
2461 return_op.Tvmult = [](Range &tril_dst, const Range &tril_src) {
2462 tril_dst = tril_src;
2463 };
2464
2465 return_op.inv_vmult = [](Range &tril_dst, const Range &tril_src) {
2466 tril_dst = tril_src;
2467 };
2468
2469 return_op.inv_Tvmult = [](Range &tril_dst, const Range &tril_src) {
2470 tril_dst = tril_src;
2471 };
2472
2473 return return_op;
2474 }
2475
2476
2477
2480 {
2481 TrilinosPayload return_op(*this);
2482
2483 return_op.vmult = [](Range &tril_dst, const Domain &) {
2484 const int ierr = tril_dst.PutScalar(0.0);
2485
2486 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2487 };
2488
2489 return_op.Tvmult = [](Domain &tril_dst, const Range &) {
2490 const int ierr = tril_dst.PutScalar(0.0);
2491
2492 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2493 };
2494
2495 return_op.inv_vmult = [](Domain &tril_dst, const Range &) {
2496 AssertThrow(false,
2497 ExcMessage("Cannot compute inverse of null operator"));
2498
2499 const int ierr = tril_dst.PutScalar(0.0);
2500
2501 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2502 };
2503
2504 return_op.inv_Tvmult = [](Range &tril_dst, const Domain &) {
2505 AssertThrow(false,
2506 ExcMessage("Cannot compute inverse of null operator"));
2507
2508 const int ierr = tril_dst.PutScalar(0.0);
2509
2510 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2511 };
2512
2513 return return_op;
2514 }
2515
2516
2517
2520 {
2521 TrilinosPayload return_op(*this);
2522 return_op.transpose();
2523 return return_op;
2524 }
2525
2526
2527
2528 IndexSet
2530 {
2531 return IndexSet(domain_map);
2532 }
2533
2534
2535
2536 IndexSet
2538 {
2539 return IndexSet(range_map);
2540 }
2541
2542
2543
2544 MPI_Comm
2546 {
2547 return communicator.Comm();
2548 }
2549
2550
2551
2552 void
2554 {
2556 }
2557
2558
2559
2560 bool
2562 {
2563 return use_transpose;
2564 }
2565
2566
2567
2568 int
2570 {
2572 {
2574 std::swap(domain_map, range_map);
2575 std::swap(vmult, Tvmult);
2576 std::swap(inv_vmult, inv_Tvmult);
2577 }
2578 return 0;
2579 }
2580
2581
2582
2583 int
2585 {
2586 // The transposedness of the operations is taken care of
2587 // when we hit the transpose flag.
2588 vmult(Y, X);
2589 return 0;
2590 }
2591
2592
2593
2594 int
2596 {
2597 // The transposedness of the operations is taken care of
2598 // when we hit the transpose flag.
2599 inv_vmult(X, Y);
2600 return 0;
2601 }
2602
2603
2604
2605 const char *
2607 {
2608 return "TrilinosPayload";
2609 }
2610
2611
2612
2613 const Epetra_Comm &
2615 {
2616 return communicator;
2617 }
2618
2619
2620
2621 const Epetra_Map &
2623 {
2624 return domain_map;
2625 }
2626
2627
2628
2629 const Epetra_Map &
2631 {
2632 return range_map;
2633 }
2634
2635
2636
2637 bool
2639 {
2640 return false;
2641 }
2642
2643
2644
2645 double
2647 {
2649 return 0.0;
2650 }
2651
2652
2653
2656 const TrilinosPayload &second_op)
2657 {
2658 using Domain = typename TrilinosPayload::Domain;
2659 using Range = typename TrilinosPayload::Range;
2660 using Intermediate = typename TrilinosPayload::VectorType;
2661 using GVMVectorType = TrilinosWrappers::MPI::Vector;
2662
2664 second_op.locally_owned_domain_indices(),
2665 ExcMessage(
2666 "Operators are set to work on incompatible IndexSets."));
2668 second_op.locally_owned_range_indices(),
2669 ExcMessage(
2670 "Operators are set to work on incompatible IndexSets."));
2671
2672 TrilinosPayload return_op(first_op, second_op);
2673
2674 // Capture by copy so the payloads are always valid
2675 return_op.vmult = [first_op, second_op](Range & tril_dst,
2676 const Domain &tril_src) {
2677 // Duplicated from LinearOperator::operator*
2678 // TODO: Template the constructor on GrowingVectorMemory vector type?
2680 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2681
2682 // Initialize intermediate vector
2683 const Epetra_Map &first_op_init_map = first_op.OperatorDomainMap();
2684 i->reinit(IndexSet(first_op_init_map),
2685 first_op.get_mpi_communicator(),
2686 /*bool omit_zeroing_entries =*/true);
2687
2688 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2689 const size_type i_local_size = i->end() - i->begin();
2690 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2691 const Epetra_Map &second_op_init_map = second_op.OperatorDomainMap();
2692 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2693 (void)second_op_init_map;
2694 Intermediate tril_int(View,
2695 first_op_init_map,
2696 const_cast<TrilinosScalar *>(i->begin()),
2697 i_local_size,
2698 1);
2699
2700 // These operators may themselves be transposed or not, so we let them
2701 // decide what the intended outcome is
2702 second_op.Apply(tril_src, tril_int);
2703 first_op.Apply(tril_src, tril_dst);
2704 const int ierr = tril_dst.Update(1.0, tril_int, 1.0);
2705 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2706 };
2707
2708 return_op.Tvmult = [first_op, second_op](Domain & tril_dst,
2709 const Range &tril_src) {
2710 // Duplicated from LinearOperator::operator*
2711 // TODO: Template the constructor on GrowingVectorMemory vector type?
2713 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2714
2715 // These operators may themselves be transposed or not, so we let them
2716 // decide what the intended outcome is
2717 // We must first transpose the operators to get the right IndexSets
2718 // for the input, intermediate and result vectors
2719 const_cast<TrilinosPayload &>(first_op).transpose();
2720 const_cast<TrilinosPayload &>(second_op).transpose();
2721
2722 // Initialize intermediate vector
2723 const Epetra_Map &first_op_init_map = first_op.OperatorRangeMap();
2724 i->reinit(IndexSet(first_op_init_map),
2725 first_op.get_mpi_communicator(),
2726 /*bool omit_zeroing_entries =*/true);
2727
2728 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2729 const size_type i_local_size = i->end() - i->begin();
2730 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2731 const Epetra_Map &second_op_init_map = second_op.OperatorRangeMap();
2732 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2733 (void)second_op_init_map;
2734 Intermediate tril_int(View,
2735 first_op_init_map,
2736 const_cast<TrilinosScalar *>(i->begin()),
2737 i_local_size,
2738 1);
2739
2740 // These operators may themselves be transposed or not, so we let them
2741 // decide what the intended outcome is
2742 second_op.Apply(tril_src, tril_int);
2743 first_op.Apply(tril_src, tril_dst);
2744 const int ierr = tril_dst.Update(1.0, tril_int, 1.0);
2745 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2746
2747 // Reset transpose flag
2748 const_cast<TrilinosPayload &>(first_op).transpose();
2749 const_cast<TrilinosPayload &>(second_op).transpose();
2750 };
2751
2752 return_op.inv_vmult = [first_op, second_op](Domain & tril_dst,
2753 const Range &tril_src) {
2754 // Duplicated from LinearOperator::operator*
2755 // TODO: Template the constructor on GrowingVectorMemory vector type?
2757 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2758
2759 // Initialize intermediate vector
2760 const Epetra_Map &first_op_init_map = first_op.OperatorRangeMap();
2761 i->reinit(IndexSet(first_op_init_map),
2762 first_op.get_mpi_communicator(),
2763 /*bool omit_zeroing_entries =*/true);
2764
2765 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2766 const size_type i_local_size = i->end() - i->begin();
2767 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2768 const Epetra_Map &second_op_init_map = second_op.OperatorRangeMap();
2769 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2770 (void)second_op_init_map;
2771 Intermediate tril_int(View,
2772 first_op_init_map,
2773 const_cast<TrilinosScalar *>(i->begin()),
2774 i_local_size,
2775 1);
2776
2777 // These operators may themselves be transposed or not, so we let them
2778 // decide what the intended outcome is
2779 second_op.ApplyInverse(tril_src, tril_int);
2780 first_op.ApplyInverse(tril_src, tril_dst);
2781 const int ierr = tril_dst.Update(1.0, tril_int, 1.0);
2782 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2783 };
2784
2785 return_op.inv_Tvmult = [first_op, second_op](Range & tril_dst,
2786 const Domain &tril_src) {
2787 // Duplicated from LinearOperator::operator*
2788 // TODO: Template the constructor on GrowingVectorMemory vector type?
2790 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2791
2792 // These operators may themselves be transposed or not, so we let them
2793 // decide what the intended outcome is
2794 // We must first transpose the operators to get the right IndexSets
2795 // for the input, intermediate and result vectors
2796 const_cast<TrilinosPayload &>(first_op).transpose();
2797 const_cast<TrilinosPayload &>(second_op).transpose();
2798
2799 // Initialize intermediate vector
2800 const Epetra_Map &first_op_init_map = first_op.OperatorDomainMap();
2801 i->reinit(IndexSet(first_op_init_map),
2802 first_op.get_mpi_communicator(),
2803 /*bool omit_zeroing_entries =*/true);
2804
2805 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2806 const size_type i_local_size = i->end() - i->begin();
2807 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2808 const Epetra_Map &second_op_init_map = second_op.OperatorDomainMap();
2809 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2810 (void)second_op_init_map;
2811 Intermediate tril_int(View,
2812 first_op_init_map,
2813 const_cast<TrilinosScalar *>(i->begin()),
2814 i_local_size,
2815 1);
2816
2817 // These operators may themselves be transposed or not, so we let them
2818 // decide what the intended outcome is
2819 second_op.ApplyInverse(tril_src, tril_int);
2820 first_op.ApplyInverse(tril_src, tril_dst);
2821 const int ierr = tril_dst.Update(1.0, tril_int, 1.0);
2822 AssertThrow(ierr == 0, ExcTrilinosError(ierr));
2823
2824 // Reset transpose flag
2825 const_cast<TrilinosPayload &>(first_op).transpose();
2826 const_cast<TrilinosPayload &>(second_op).transpose();
2827 };
2828
2829 return return_op;
2830 }
2831
2832
2833
2834 TrilinosPayload
2836 const TrilinosPayload &second_op)
2837 {
2838 using Domain = typename TrilinosPayload::Domain;
2839 using Range = typename TrilinosPayload::Range;
2840 using Intermediate = typename TrilinosPayload::VectorType;
2841 using GVMVectorType = TrilinosWrappers::MPI::Vector;
2842
2844 second_op.locally_owned_range_indices(),
2845 ExcMessage(
2846 "Operators are set to work on incompatible IndexSets."));
2847
2848 TrilinosPayload return_op(first_op, second_op);
2849
2850 // Capture by copy so the payloads are always valid
2851 return_op.vmult = [first_op, second_op](Range & tril_dst,
2852 const Domain &tril_src) {
2853 // Duplicated from LinearOperator::operator*
2854 // TODO: Template the constructor on GrowingVectorMemory vector type?
2856 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2857
2858 // Initialize intermediate vector
2859 const Epetra_Map &first_op_init_map = first_op.OperatorDomainMap();
2860 i->reinit(IndexSet(first_op_init_map),
2861 first_op.get_mpi_communicator(),
2862 /*bool omit_zeroing_entries =*/true);
2863
2864 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2865 const size_type i_local_size = i->end() - i->begin();
2866 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2867 const Epetra_Map &second_op_init_map = second_op.OperatorRangeMap();
2868 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2869 (void)second_op_init_map;
2870 Intermediate tril_int(View,
2871 first_op_init_map,
2872 const_cast<TrilinosScalar *>(i->begin()),
2873 i_local_size,
2874 1);
2875
2876 // These operators may themselves be transposed or not, so we let them
2877 // decide what the intended outcome is
2878 second_op.Apply(tril_src, tril_int);
2879 first_op.Apply(tril_int, tril_dst);
2880 };
2881
2882 return_op.Tvmult = [first_op, second_op](Domain & tril_dst,
2883 const Range &tril_src) {
2884 // Duplicated from LinearOperator::operator*
2885 // TODO: Template the constructor on GrowingVectorMemory vector type?
2887 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2888
2889 // These operators may themselves be transposed or not, so we let them
2890 // decide what the intended outcome is
2891 // We must first transpose the operators to get the right IndexSets
2892 // for the input, intermediate and result vectors
2893 const_cast<TrilinosPayload &>(first_op).transpose();
2894 const_cast<TrilinosPayload &>(second_op).transpose();
2895
2896 // Initialize intermediate vector
2897 const Epetra_Map &first_op_init_map = first_op.OperatorRangeMap();
2898 i->reinit(IndexSet(first_op_init_map),
2899 first_op.get_mpi_communicator(),
2900 /*bool omit_zeroing_entries =*/true);
2901
2902 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2903 const size_type i_local_size = i->end() - i->begin();
2904 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2905 const Epetra_Map &second_op_init_map = second_op.OperatorDomainMap();
2906 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2907 (void)second_op_init_map;
2908 Intermediate tril_int(View,
2909 first_op_init_map,
2910 const_cast<TrilinosScalar *>(i->begin()),
2911 i_local_size,
2912 1);
2913
2914 // Apply the operators in the reverse order to vmult
2915 first_op.Apply(tril_src, tril_int);
2916 second_op.Apply(tril_int, tril_dst);
2917
2918 // Reset transpose flag
2919 const_cast<TrilinosPayload &>(first_op).transpose();
2920 const_cast<TrilinosPayload &>(second_op).transpose();
2921 };
2922
2923 return_op.inv_vmult = [first_op, second_op](Domain & tril_dst,
2924 const Range &tril_src) {
2925 // Duplicated from LinearOperator::operator*
2926 // TODO: Template the constructor on GrowingVectorMemory vector type?
2928 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2929
2930 // Initialize intermediate vector
2931 const Epetra_Map &first_op_init_map = first_op.OperatorRangeMap();
2932 i->reinit(IndexSet(first_op_init_map),
2933 first_op.get_mpi_communicator(),
2934 /*bool omit_zeroing_entries =*/true);
2935
2936 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2937 const size_type i_local_size = i->end() - i->begin();
2938 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2939 const Epetra_Map &second_op_init_map = second_op.OperatorDomainMap();
2940 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2941 (void)second_op_init_map;
2942 Intermediate tril_int(View,
2943 first_op_init_map,
2944 const_cast<TrilinosScalar *>(i->begin()),
2945 i_local_size,
2946 1);
2947
2948 // Apply the operators in the reverse order to vmult
2949 // and the same order as Tvmult
2950 first_op.ApplyInverse(tril_src, tril_int);
2951 second_op.ApplyInverse(tril_int, tril_dst);
2952 };
2953
2954 return_op.inv_Tvmult = [first_op, second_op](Range & tril_dst,
2955 const Domain &tril_src) {
2956 // Duplicated from LinearOperator::operator*
2957 // TODO: Template the constructor on GrowingVectorMemory vector type?
2959 VectorMemory<GVMVectorType>::Pointer i(vector_memory);
2960
2961 // These operators may themselves be transposed or not, so we let them
2962 // decide what the intended outcome is
2963 // We must first transpose the operators to get the right IndexSets
2964 // for the input, intermediate and result vectors
2965 const_cast<TrilinosPayload &>(first_op).transpose();
2966 const_cast<TrilinosPayload &>(second_op).transpose();
2967
2968 // Initialize intermediate vector
2969 const Epetra_Map &first_op_init_map = first_op.OperatorDomainMap();
2970 i->reinit(IndexSet(first_op_init_map),
2971 first_op.get_mpi_communicator(),
2972 /*bool omit_zeroing_entries =*/true);
2973
2974 // Duplicated from TrilinosWrappers::SparseMatrix::vmult
2975 const size_type i_local_size = i->end() - i->begin();
2976 AssertDimension(i_local_size, first_op_init_map.NumMyPoints());
2977 const Epetra_Map &second_op_init_map = second_op.OperatorRangeMap();
2978 AssertDimension(i_local_size, second_op_init_map.NumMyPoints());
2979 (void)second_op_init_map;
2980 Intermediate tril_int(View,
2981 first_op_init_map,
2982 const_cast<TrilinosScalar *>(i->begin()),
2983 i_local_size,
2984 1);
2985
2986 // These operators may themselves be transposed or not, so we let them
2987 // decide what the intended outcome is
2988 // Apply the operators in the reverse order to Tvmult
2989 // and the same order as vmult
2990 second_op.ApplyInverse(tril_src, tril_int);
2991 first_op.ApplyInverse(tril_int, tril_dst);
2992
2993 // Reset transpose flag
2994 const_cast<TrilinosPayload &>(first_op).transpose();
2995 const_cast<TrilinosPayload &>(second_op).transpose();
2996 };
2997
2998 return return_op;
2999 }
3000
3001 } // namespace LinearOperatorImplementation
3002 } /* namespace internal */
3003} /* namespace TrilinosWrappers */
3004
3005
3006
3007// explicit instantiations
3008# include "trilinos_sparse_matrix.inst"
3009
3010# ifndef DOXYGEN
3011// TODO: put these instantiations into generic file
3012namespace TrilinosWrappers
3013{
3014 template void
3015 SparseMatrix::reinit(const ::SparsityPattern &);
3016
3017 template void
3019
3020 template void
3022 const IndexSet &,
3023 const ::SparsityPattern &,
3024 const MPI_Comm,
3025 const bool);
3026
3027 template void
3029 const IndexSet &,
3030 const DynamicSparsityPattern &,
3031 const MPI_Comm,
3032 const bool);
3033
3034 template void
3035 SparseMatrix::vmult(MPI::Vector &, const MPI::Vector &) const;
3036
3037 template void
3039 const ::Vector<double> &) const;
3040
3041 template void
3044 const ::LinearAlgebra::distributed::Vector<double> &) const;
3045
3046# ifdef DEAL_II_TRILINOS_WITH_TPETRA
3047# if defined(HAVE_TPETRA_INST_DOUBLE)
3048 template void
3051 const ::LinearAlgebra::TpetraWrappers::Vector<double> &) const;
3052# endif
3053
3054# if defined(HAVE_TPETRA_INST_FLOAT)
3055 template void
3058 const ::LinearAlgebra::TpetraWrappers::Vector<float> &) const;
3059# endif
3060# endif
3061
3062 template void
3065 const ::LinearAlgebra::EpetraWrappers::Vector &) const;
3066
3067 template void
3068 SparseMatrix::Tvmult(MPI::Vector &, const MPI::Vector &) const;
3069
3070 template void
3072 const ::Vector<double> &) const;
3073
3074 template void
3077 const ::LinearAlgebra::distributed::Vector<double> &) const;
3078
3079# ifdef DEAL_II_TRILINOS_WITH_TPETRA
3080# if defined(HAVE_TPETRA_INST_DOUBLE)
3081 template void
3084 const ::LinearAlgebra::TpetraWrappers::Vector<double> &) const;
3085# endif
3086
3087# if defined(HAVE_TPETRA_INST_FLOAT)
3088 template void
3091 const ::LinearAlgebra::TpetraWrappers::Vector<float> &) const;
3092# endif
3093# endif
3094
3095 template void
3098 const ::LinearAlgebra::EpetraWrappers::Vector &) const;
3099
3100 template void
3101 SparseMatrix::vmult_add(MPI::Vector &, const MPI::Vector &) const;
3102
3103 template void
3105 const ::Vector<double> &) const;
3106
3107 template void
3110 const ::LinearAlgebra::distributed::Vector<double> &) const;
3111
3112# ifdef DEAL_II_TRILINOS_WITH_TPETRA
3113# if defined(HAVE_TPETRA_INST_DOUBLE)
3114 template void
3117 const ::LinearAlgebra::TpetraWrappers::Vector<double> &) const;
3118# endif
3119
3120# if defined(HAVE_TPETRA_INST_FLOAT)
3121 template void
3124 const ::LinearAlgebra::TpetraWrappers::Vector<float> &) const;
3125# endif
3126# endif
3127
3128 template void
3131 const ::LinearAlgebra::EpetraWrappers::Vector &) const;
3132
3133 template void
3134 SparseMatrix::Tvmult_add(MPI::Vector &, const MPI::Vector &) const;
3135
3136 template void
3138 const ::Vector<double> &) const;
3139
3140 template void
3143 const ::LinearAlgebra::distributed::Vector<double> &) const;
3144
3145# ifdef DEAL_II_TRILINOS_WITH_TPETRA
3146# if defined(HAVE_TPETRA_INST_DOUBLE)
3147 template void
3150 const ::LinearAlgebra::TpetraWrappers::Vector<double> &) const;
3151# endif
3152
3153# if defined(HAVE_TPETRA_INST_FLOAT)
3154 template void
3157 const ::LinearAlgebra::TpetraWrappers::Vector<float> &) const;
3158# endif
3159# endif
3160
3161 template void
3164 const ::LinearAlgebra::EpetraWrappers::Vector &) const;
3165} // namespace TrilinosWrappers
3166# endif // DOXYGEN
3167
3169
3170#endif // DEAL_II_WITH_TRILINOS
const IndexSet & row_index_set() const
size_type row_length(const size_type row) const
size_type column_number(const size_type row, const size_type index) const
size_type size() const
Definition index_set.h:1661
bool is_element(const size_type index) const
Definition index_set.h:1776
Epetra_Map make_trilinos_map(const MPI_Comm communicator=MPI_COMM_WORLD, const bool overlapping=false) const
Definition index_set.cc:789
const Epetra_FEVector & trilinos_vector() const
const Tpetra::Vector< Number, int, types::signed_global_dof_index > & trilinos_vector() const
size_type n_rows() const
size_type n_cols() const
void reinit(const size_type m, const size_type n, const ArrayView< const unsigned int > &row_lengths)
const Epetra_BlockMap & trilinos_partitioner() const
const Epetra_MultiVector & trilinos_vector() const
void reinit(const Vector &v, const bool omit_zeroing_entries=false, const bool allow_different_maps=false)
std::pair< size_type, size_type > local_range() const
std::shared_ptr< std::vector< TrilinosScalar > > value_cache
std::shared_ptr< std::vector< size_type > > colnum_cache
void set(const size_type i, const size_type j, const TrilinosScalar value)
std::unique_ptr< Epetra_Map > column_space_map
SparseMatrix & operator*=(const TrilinosScalar factor)
void mmult(SparseMatrix &C, const SparseMatrix &B, const MPI::Vector &V=MPI::Vector()) const
std::unique_ptr< Epetra_Export > nonlocal_matrix_exporter
void compress(VectorOperation::values operation)
std::unique_ptr< Epetra_FECrsMatrix > matrix
::types::global_dof_index size_type
const Epetra_CrsMatrix & trilinos_matrix() const
TrilinosScalar matrix_norm_square(const MPI::Vector &v) const
IndexSet locally_owned_range_indices() const
void print(std::ostream &out, const bool write_extended_trilinos_info=false) const
void Tmmult(SparseMatrix &C, const SparseMatrix &B, const MPI::Vector &V=MPI::Vector()) const
void clear_row(const size_type row, const TrilinosScalar new_diag_value=0)
void reinit(const SparsityPatternType &sparsity_pattern)
std::enable_if_t< std::is_same< typename VectorType::value_type, TrilinosScalar >::value > Tvmult(VectorType &dst, const VectorType &src) const
void vmult_add(VectorType &dst, const VectorType &src) const
SparseMatrix & operator=(const SparseMatrix &)=delete
SparseMatrix & operator/=(const TrilinosScalar factor)
void Tvmult_add(VectorType &dst, const VectorType &src) const
void clear_rows(const std::vector< size_type > &rows, const TrilinosScalar new_diag_value=0)
TrilinosScalar el(const size_type i, const size_type j) const
IndexSet locally_owned_domain_indices() const
TrilinosScalar residual(MPI::Vector &dst, const MPI::Vector &x, const MPI::Vector &b) const
bool in_local_range(const size_type index) const
void copy_from(const SparseMatrix &source)
unsigned int row_length(const size_type row) const
std::uint64_t n_nonzero_elements() const
const Epetra_CrsGraph & trilinos_sparsity_pattern() const
TrilinosScalar diag_element(const size_type i) const
void add(const size_type i, const size_type j, const TrilinosScalar value)
unsigned int local_size() const
TrilinosScalar operator()(const size_type i, const size_type j) const
TrilinosScalar matrix_scalar_product(const MPI::Vector &u, const MPI::Vector &v) const
std::pair< size_type, size_type > local_range() const
std::enable_if_t< std::is_same< typename VectorType::value_type, TrilinosScalar >::value > vmult(VectorType &dst, const VectorType &src) const
std::unique_ptr< Epetra_CrsMatrix > nonlocal_matrix
std::unique_ptr< Epetra_CrsGraph > nonlocal_graph
const Epetra_FECrsGraph & trilinos_sparsity_pattern() const
std::function< void(VectorType &, const VectorType &)> inv_Tvmult
virtual int ApplyInverse(const VectorType &Y, VectorType &X) const override
virtual int Apply(const VectorType &X, VectorType &Y) const override
#define DEAL_II_NAMESPACE_OPEN
Definition config.h:472
#define DEAL_II_NAMESPACE_CLOSE
Definition config.h:473
DerivativeForm< 1, spacedim, dim, Number > transpose(const DerivativeForm< 1, dim, spacedim, Number > &DF)
Point< 2 > second
Definition grid_out.cc:4616
Point< 2 > first
Definition grid_out.cc:4615
static ::ExceptionBase & ExcIO()
static ::ExceptionBase & ExcNotImplemented()
static ::ExceptionBase & ExcScalarAssignmentOnlyForZeroValue()
static ::ExceptionBase & ExcInvalidIndex(size_type arg1, size_type arg2)
#define Assert(cond, exc)
static ::ExceptionBase & ExcAccessToNonPresentElement(size_type arg1, size_type arg2)
#define AssertIsFinite(number)
static ::ExceptionBase & ExcAccessToNonlocalRow(std::size_t arg1)
#define AssertDimension(dim1, dim2)
#define AssertIndexRange(index, range)
static ::ExceptionBase & ExcInternalError()
static ::ExceptionBase & ExcDivideByZero()
static ::ExceptionBase & ExcSourceEqualsDestination()
static ::ExceptionBase & ExcDimensionMismatch(std::size_t arg1, std::size_t arg2)
static ::ExceptionBase & ExcNotQuadratic()
static ::ExceptionBase & ExcMatrixNotCompressed()
static ::ExceptionBase & ExcTrilinosError(int arg1)
static ::ExceptionBase & ExcAccessToNonLocalElement(size_type arg1, size_type arg2, size_type arg3, size_type arg4)
static ::ExceptionBase & ExcMessage(std::string arg1)
static ::ExceptionBase & ExcTrilinosError(int arg1)
#define AssertThrow(cond, exc)
IndexSet complete_index_set(const IndexSet::size_type N)
Definition index_set.h:1089
@ matrix
Contents is actually a matrix.
types::global_dof_index size_type
TrilinosPayload operator+(const TrilinosPayload &first_op, const TrilinosPayload &second_op)
TrilinosPayload operator*(const TrilinosPayload &first_op, const TrilinosPayload &second_op)
void check_vector_map_equality(const Epetra_CrsMatrix &, const VectorType &, const VectorType &)
VectorType::value_type * end(VectorType &V)
VectorType::value_type * begin(VectorType &V)
void perform_mmult(const SparseMatrix &inputleft, const SparseMatrix &inputright, SparseMatrix &result, const MPI::Vector &V, const bool transpose_left)
TrilinosWrappers::types::int_type min_my_gid(const Epetra_BlockMap &map)
TrilinosWrappers::types::int64_type n_global_elements(const Epetra_BlockMap &map)
TrilinosWrappers::types::int_type global_column_index(const Epetra_CrsMatrix &matrix, const ::types::global_dof_index i)
TrilinosWrappers::types::int_type max_my_gid(const Epetra_BlockMap &map)
TrilinosWrappers::types::int_type global_row_index(const Epetra_CrsMatrix &matrix, const ::types::global_dof_index i)
TrilinosWrappers::types::int_type n_global_cols(const Epetra_CrsGraph &graph)
const Epetra_Comm & comm_self()
Iterator lower_bound(Iterator first, Iterator last, const T &val)
Definition utilities.h:1016
Definition types.h:33
unsigned int global_dof_index
Definition types.h:82
double TrilinosScalar
Definition types.h:175