Reference documentation for deal.II version 9.4.1
\(\newcommand{\dealvcentcolon}{\mathrel{\mathop{:}}}\) \(\newcommand{\dealcoloneq}{\dealvcentcolon\mathrel{\mkern-1.2mu}=}\) \(\newcommand{\jump}[1]{\left[\!\left[ #1 \right]\!\right]}\) \(\newcommand{\average}[1]{\left\{\!\left\{ #1 \right\}\!\right\}}\)
Loading...
Searching...
No Matches
task_info.cc
Go to the documentation of this file.
1// ---------------------------------------------------------------------
2//
3// Copyright (C) 2018 - 2022 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
16
17#include <deal.II/base/conditional_ostream.h>
18#include <deal.II/base/memory_consumption.h>
19#include <deal.II/base/mpi.h>
20#include <deal.II/base/multithread_info.h>
21#include <deal.II/base/parallel.h>
22#include <deal.II/base/utilities.h>
23
24#include <deal.II/lac/dynamic_sparsity_pattern.h>
25
26#include <deal.II/matrix_free/task_info.h>
27
28
29#ifdef DEAL_II_WITH_TBB
30# include <tbb/blocked_range.h>
31# include <tbb/parallel_for.h>
32# include <tbb/task.h>
33# ifndef DEAL_II_TBB_WITH_ONEAPI
34# include <tbb/task_scheduler_init.h>
35# endif
36#endif
37
38#include <iostream>
39#include <set>
40
41//
42// TBB with oneAPI API has deprecated and removed the
43// <code>tbb::tasks</code> backend. With this it is no longer possible to
44// compile the following code that builds a directed acyclic graph (DAG) of
45// (thread parallel) tasks without a major porting effort. It turned out
46// that such a dynamic handling of dependencies and structures is not as
47// competitive as initially assumed. Consequently, this part of the matrix
48// free infrastructure has seen less attention than the rest over the last
49// years and is (presumably) not used that often.
50//
51// In case of detected oneAPI backend we simply disable threading in the
52// matrix free backend for now.
53//
54// Matthias Maier, Martin Kronbichler, 2021
55//
56
57DEAL_II_NAMESPACE_OPEN
58
59
60
61/*-------------------- Implementation of the matrix-free loop --------------*/
62namespace internal
63{
64 namespace MatrixFreeFunctions
65 {
66#if defined(DEAL_II_WITH_TBB) && !defined(DEAL_II_TBB_WITH_ONEAPI)
67
68 // This defines the TBB data structures that are needed to schedule the
69 // partition-partition variant
70
71 namespace partition
72 {
73 class ActualCellWork
74 {
75 public:
76 ActualCellWork(MFWorkerInterface **worker_pointer,
77 const unsigned int partition,
78 const TaskInfo & task_info)
79 : worker(nullptr)
80 , worker_pointer(worker_pointer)
81 , partition(partition)
82 , task_info(task_info)
83 {}
84
85 ActualCellWork(MFWorkerInterface &worker,
86 const unsigned int partition,
87 const TaskInfo & task_info)
88 : worker(&worker)
89 , worker_pointer(nullptr)
90 , partition(partition)
91 , task_info(task_info)
92 {}
93
94 void
95 operator()() const
96 {
97 MFWorkerInterface *used_worker =
98 worker != nullptr ? worker : *worker_pointer;
99 Assert(used_worker != nullptr, ExcInternalError());
100 used_worker->cell(partition);
101
102 if (task_info.face_partition_data.empty() == false)
103 {
104 used_worker->face(partition);
105 used_worker->boundary(partition);
106 }
107 }
108
109 private:
110 MFWorkerInterface * worker;
111 MFWorkerInterface **worker_pointer;
112 const unsigned int partition;
113 const TaskInfo & task_info;
114 };
115
116 class CellWork : public tbb::task
117 {
118 public:
119 CellWork(MFWorkerInterface &worker,
120 const unsigned int partition,
121 const TaskInfo & task_info,
122 const bool is_blocked)
123 : dummy(nullptr)
124 , work(worker, partition, task_info)
125 , is_blocked(is_blocked)
126 {}
127
128 tbb::task *
129 execute() override
130 {
131 work();
132
133 if (is_blocked == true)
134 tbb::empty_task::spawn(*dummy);
135 return nullptr;
136 }
137
138 tbb::empty_task *dummy;
139
140 private:
141 ActualCellWork work;
142 const bool is_blocked;
143 };
144
145
146
147 class PartitionWork : public tbb::task
148 {
149 public:
150 PartitionWork(MFWorkerInterface &function_in,
151 const unsigned int partition_in,
152 const TaskInfo & task_info_in,
153 const bool is_blocked_in = false)
154 : dummy(nullptr)
155 , function(function_in)
156 , partition(partition_in)
157 , task_info(task_info_in)
158 , is_blocked(is_blocked_in)
159 {}
160
161 tbb::task *
162 execute() override
163 {
164 tbb::empty_task *root =
165 new (tbb::task::allocate_root()) tbb::empty_task;
166 const unsigned int evens = task_info.partition_evens[partition];
167 const unsigned int odds = task_info.partition_odds[partition];
168 const unsigned int n_blocked_workers =
169 task_info.partition_n_blocked_workers[partition];
170 const unsigned int n_workers =
171 task_info.partition_n_workers[partition];
172 std::vector<CellWork *> worker(n_workers);
173 std::vector<CellWork *> blocked_worker(n_blocked_workers);
174
175 root->set_ref_count(evens + 1);
176 for (unsigned int j = 0; j < evens; ++j)
177 {
178 worker[j] = new (root->allocate_child())
179 CellWork(function,
180 task_info.partition_row_index[partition] + 2 * j,
181 task_info,
182 false);
183 if (j > 0)
184 {
185 worker[j]->set_ref_count(2);
186 blocked_worker[j - 1]->dummy =
187 new (worker[j]->allocate_child()) tbb::empty_task;
188 tbb::task::spawn(*blocked_worker[j - 1]);
189 }
190 else
191 worker[j]->set_ref_count(1);
192 if (j < evens - 1)
193 {
194 blocked_worker[j] = new (worker[j]->allocate_child())
195 CellWork(function,
196 task_info.partition_row_index[partition] + 2 * j +
197 1,
198 task_info,
199 true);
200 }
201 else
202 {
203 if (odds == evens)
204 {
205 worker[evens] = new (worker[j]->allocate_child())
206 CellWork(function,
207 task_info.partition_row_index[partition] +
208 2 * j + 1,
209 task_info,
210 false);
211 tbb::task::spawn(*worker[evens]);
212 }
213 else
214 {
215 tbb::empty_task *child =
216 new (worker[j]->allocate_child()) tbb::empty_task();
217 tbb::task::spawn(*child);
218 }
219 }
220 }
221
222 root->wait_for_all();
223 root->destroy(*root);
224 if (is_blocked == true)
225 tbb::empty_task::spawn(*dummy);
226 return nullptr;
227 }
228
229 tbb::empty_task *dummy;
230
231 private:
232 MFWorkerInterface &function;
233 const unsigned int partition;
234 const TaskInfo & task_info;
235 const bool is_blocked;
236 };
237
238 } // end of namespace partition
239
240
241
242 namespace color
243 {
244 class CellWork
245 {
246 public:
247 CellWork(MFWorkerInterface &worker_in,
248 const TaskInfo & task_info_in,
249 const unsigned int partition_in)
250 : worker(worker_in)
251 , task_info(task_info_in)
252 , partition(partition_in)
253 {}
254
255 void
256 operator()(const tbb::blocked_range<unsigned int> &r) const
257 {
258 const unsigned int start_index =
259 task_info.cell_partition_data[partition] +
260 task_info.block_size * r.begin();
261 const unsigned int end_index =
262 std::min(start_index + task_info.block_size * (r.end() - r.begin()),
263 task_info.cell_partition_data[partition + 1]);
264 worker.cell(std::make_pair(start_index, end_index));
265
266 if (task_info.face_partition_data.empty() == false)
267 {
268 AssertThrow(false, ExcNotImplemented());
269 }
270 }
271
272 private:
273 MFWorkerInterface &worker;
274 const TaskInfo & task_info;
275 const unsigned int partition;
276 };
277
278
279
280 class PartitionWork : public tbb::task
281 {
282 public:
283 PartitionWork(MFWorkerInterface &worker_in,
284 const unsigned int partition_in,
285 const TaskInfo & task_info_in,
286 const bool is_blocked_in)
287 : dummy(nullptr)
288 , worker(worker_in)
289 , partition(partition_in)
290 , task_info(task_info_in)
291 , is_blocked(is_blocked_in)
292 {}
293
294 tbb::task *
295 execute() override
296 {
297 const unsigned int n_chunks =
298 (task_info.cell_partition_data[partition + 1] -
299 task_info.cell_partition_data[partition] + task_info.block_size -
300 1) /
301 task_info.block_size;
302 parallel_for(tbb::blocked_range<unsigned int>(0, n_chunks, 1),
303 CellWork(worker, task_info, partition));
304 if (is_blocked == true)
305 tbb::empty_task::spawn(*dummy);
306 return nullptr;
307 }
308
309 tbb::empty_task *dummy;
310
311 private:
312 MFWorkerInterface &worker;
313 const unsigned int partition;
314 const TaskInfo & task_info;
315 const bool is_blocked;
316 };
317
318 } // end of namespace color
319
320
321
322 class MPICommunication : public tbb::task
323 {
324 public:
325 MPICommunication(MFWorkerInterface &worker_in, const bool do_compress)
326 : worker(worker_in)
327 , do_compress(do_compress)
328 {}
329
330 tbb::task *
331 execute() override
332 {
333 if (do_compress == false)
334 worker.vector_update_ghosts_finish();
335 else
336 worker.vector_compress_start();
337 return nullptr;
338 }
339
340 private:
341 MFWorkerInterface &worker;
342 const bool do_compress;
343 };
344
345#endif // DEAL_II_WITH_TBB
346
347
348
349 void
350 TaskInfo::loop(MFWorkerInterface &funct) const
351 {
352 // If we use thread parallelism, we do not currently support to schedule
353 // pieces of updates within the loop, so this index will collect all
354 // calls in that case and work like a single complete loop over all
355 // cells
356 if (scheme != none)
357 funct.cell_loop_pre_range(numbers::invalid_unsigned_int);
358 else
359 funct.cell_loop_pre_range(
360 partition_row_index[partition_row_index.size() - 2]);
361
362 funct.vector_update_ghosts_start();
363
364#if defined(DEAL_II_WITH_TBB) && !defined(DEAL_II_TBB_WITH_ONEAPI)
365
366 if (scheme != none)
367 {
368 funct.zero_dst_vector_range(numbers::invalid_unsigned_int);
369 if (scheme == partition_partition && evens > 0)
370 {
371 tbb::empty_task *root =
372 new (tbb::task::allocate_root()) tbb::empty_task;
373 root->set_ref_count(evens + 1);
374 std::vector<partition::PartitionWork *> worker(n_workers);
375 std::vector<partition::PartitionWork *> blocked_worker(
376 n_blocked_workers);
377 MPICommunication *worker_compr =
378 new (root->allocate_child()) MPICommunication(funct, true);
379 worker_compr->set_ref_count(1);
380 for (unsigned int j = 0; j < evens; ++j)
381 {
382 if (j > 0)
383 {
384 worker[j] = new (root->allocate_child())
385 partition::PartitionWork(funct, 2 * j, *this, false);
386 worker[j]->set_ref_count(2);
387 blocked_worker[j - 1]->dummy =
388 new (worker[j]->allocate_child()) tbb::empty_task;
389 tbb::task::spawn(*blocked_worker[j - 1]);
390 }
391 else
392 {
393 worker[j] = new (worker_compr->allocate_child())
394 partition::PartitionWork(funct, 2 * j, *this, false);
395 worker[j]->set_ref_count(2);
396 MPICommunication *worker_dist =
397 new (worker[j]->allocate_child())
398 MPICommunication(funct, false);
399 tbb::task::spawn(*worker_dist);
400 }
401 if (j < evens - 1)
402 {
403 blocked_worker[j] = new (worker[j]->allocate_child())
404 partition::PartitionWork(funct, 2 * j + 1, *this, true);
405 }
406 else
407 {
408 if (odds == evens)
409 {
410 worker[evens] = new (worker[j]->allocate_child())
411 partition::PartitionWork(funct,
412 2 * j + 1,
413 *this,
414 false);
415 tbb::task::spawn(*worker[evens]);
416 }
417 else
418 {
419 tbb::empty_task *child =
420 new (worker[j]->allocate_child()) tbb::empty_task();
421 tbb::task::spawn(*child);
422 }
423 }
424 }
425
426 root->wait_for_all();
427 root->destroy(*root);
428 }
429 else if (scheme == partition_partition)
430 {
431 // catch the case of empty partition list: we still need to call
432 // the vector communication routines to clean up and initiate
433 // things
434 funct.vector_update_ghosts_finish();
435 funct.vector_compress_start();
436 }
437 else // end of partition-partition, start of partition-color
438 {
439 // check whether there is only one partition. if not, build up the
440 // tree of partitions
441 if (odds > 0)
442 {
443 tbb::empty_task *root =
444 new (tbb::task::allocate_root()) tbb::empty_task;
445 root->set_ref_count(evens + 1);
446 const unsigned int n_blocked_workers =
447 odds - (odds + evens + 1) % 2;
448 const unsigned int n_workers =
449 cell_partition_data.size() - 1 - n_blocked_workers;
450 std::vector<color::PartitionWork *> worker(n_workers);
451 std::vector<color::PartitionWork *> blocked_worker(
452 n_blocked_workers);
453 unsigned int worker_index = 0, slice_index = 0;
454 int spawn_index_child = -2;
455 MPICommunication *worker_compr =
456 new (root->allocate_child()) MPICommunication(funct, true);
457 worker_compr->set_ref_count(1);
458 for (unsigned int part = 0;
459 part < partition_row_index.size() - 1;
460 part++)
461 {
462 if (part == 0)
463 worker[worker_index] =
464 new (worker_compr->allocate_child())
465 color::PartitionWork(funct,
466 slice_index,
467 *this,
468 false);
469 else
470 worker[worker_index] = new (root->allocate_child())
471 color::PartitionWork(funct,
472 slice_index,
473 *this,
474 false);
475 slice_index++;
476 for (; slice_index < partition_row_index[part + 1];
477 slice_index++)
478 {
479 worker[worker_index]->set_ref_count(1);
480 worker_index++;
481 worker[worker_index] =
482 new (worker[worker_index - 1]->allocate_child())
483 color::PartitionWork(funct,
484 slice_index,
485 *this,
486 false);
487 }
488 worker[worker_index]->set_ref_count(2);
489 if (part > 0)
490 {
491 blocked_worker[(part - 1) / 2]->dummy =
492 new (worker[worker_index]->allocate_child())
493 tbb::empty_task;
494 worker_index++;
495 if (spawn_index_child == -1)
496 tbb::task::spawn(*blocked_worker[(part - 1) / 2]);
497 else
498 {
499 Assert(spawn_index_child >= 0,
500 ExcInternalError());
501 tbb::task::spawn(*worker[spawn_index_child]);
502 }
503 spawn_index_child = -2;
504 }
505 else
506 {
507 MPICommunication *worker_dist =
508 new (worker[worker_index]->allocate_child())
509 MPICommunication(funct, false);
510 tbb::task::spawn(*worker_dist);
511 worker_index++;
512 }
513 part += 1;
514 if (part < partition_row_index.size() - 1)
515 {
516 if (part < partition_row_index.size() - 2)
517 {
518 blocked_worker[part / 2] =
519 new (worker[worker_index - 1]->allocate_child())
520 color::PartitionWork(funct,
521 slice_index,
522 *this,
523 true);
524 slice_index++;
525 if (slice_index < partition_row_index[part + 1])
526 {
527 blocked_worker[part / 2]->set_ref_count(1);
528 worker[worker_index] = new (
529 blocked_worker[part / 2]->allocate_child())
530 color::PartitionWork(funct,
531 slice_index,
532 *this,
533 false);
534 slice_index++;
535 }
536 else
537 {
538 spawn_index_child = -1;
539 continue;
540 }
541 }
542 for (; slice_index < partition_row_index[part + 1];
543 slice_index++)
544 {
545 if (slice_index > partition_row_index[part])
546 {
547 worker[worker_index]->set_ref_count(1);
548 worker_index++;
549 }
550 worker[worker_index] =
551 new (worker[worker_index - 1]->allocate_child())
552 color::PartitionWork(funct,
553 slice_index,
554 *this,
555 false);
556 }
557 spawn_index_child = worker_index;
558 worker_index++;
559 }
560 else
561 {
562 tbb::empty_task *final =
563 new (worker[worker_index - 1]->allocate_child())
564 tbb::empty_task;
565 tbb::task::spawn(*final);
566 spawn_index_child = worker_index - 1;
567 }
568 }
569 if (evens == odds)
570 {
571 Assert(spawn_index_child >= 0, ExcInternalError());
572 tbb::task::spawn(*worker[spawn_index_child]);
573 }
574 root->wait_for_all();
575 root->destroy(*root);
576 }
577 // case when we only have one partition: this is the usual
578 // coloring scheme, and we just schedule a parallel for loop for
579 // each color
580 else
581 {
582 Assert(evens <= 1, ExcInternalError());
583 funct.vector_update_ghosts_finish();
584
585 for (unsigned int color = 0; color < partition_row_index[1];
586 ++color)
587 {
588 tbb::empty_task *root =
589 new (tbb::task::allocate_root()) tbb::empty_task;
590 root->set_ref_count(2);
591 color::PartitionWork *worker =
592 new (root->allocate_child())
593 color::PartitionWork(funct, color, *this, false);
594 tbb::empty_task::spawn(*worker);
595 root->wait_for_all();
596 root->destroy(*root);
597 }
598
599 funct.vector_compress_start();
600 }
601 }
602 }
603 else
604#endif
605 // serial loop, go through up to three times and do the MPI transfer at
606 // the beginning/end of the second part
607 {
608 for (unsigned int part = 0; part < partition_row_index.size() - 2;
609 ++part)
610 {
611 if (part == 1)
612 funct.vector_update_ghosts_finish();
613
614 for (unsigned int i = partition_row_index[part];
615 i < partition_row_index[part + 1];
616 ++i)
617 {
618 funct.cell_loop_pre_range(i);
619 funct.zero_dst_vector_range(i);
620 AssertIndexRange(i + 1, cell_partition_data.size());
621 if (cell_partition_data[i + 1] > cell_partition_data[i])
622 {
623 funct.cell(i);
624 }
625
626 if (face_partition_data.empty() == false)
627 {
628 if (face_partition_data[i + 1] > face_partition_data[i])
629 funct.face(i);
630 if (boundary_partition_data[i + 1] >
631 boundary_partition_data[i])
632 funct.boundary(i);
633 }
634 funct.cell_loop_post_range(i);
635 }
636
637 if (part == 1)
638 funct.vector_compress_start();
639 }
640 }
641 funct.vector_compress_finish();
642
643 if (scheme != none)
644 funct.cell_loop_post_range(numbers::invalid_unsigned_int);
645 else
646 funct.cell_loop_post_range(
647 partition_row_index[partition_row_index.size() - 2]);
648 }
649
650
651
652 TaskInfo::TaskInfo()
653 {
654 clear();
655 }
656
657
658
659 void
660 TaskInfo::clear()
661 {
662 n_active_cells = 0;
663 n_ghost_cells = 0;
664 vectorization_length = 1;
665 block_size = 0;
666 n_blocks = 0;
667 scheme = none;
668 partition_row_index.clear();
669 partition_row_index.resize(2);
670 cell_partition_data.clear();
671 face_partition_data.clear();
672 boundary_partition_data.clear();
673 evens = 0;
674 odds = 0;
675 n_blocked_workers = 0;
676 n_workers = 0;
677 partition_evens.clear();
678 partition_odds.clear();
679 partition_n_blocked_workers.clear();
680 partition_n_workers.clear();
681 communicator = MPI_COMM_SELF;
682 my_pid = 0;
683 n_procs = 1;
684 }
685
686
687
688 template <typename StreamType>
689 void
690 TaskInfo::print_memory_statistics(StreamType & out,
691 const std::size_t data_length) const
692 {
693 Utilities::MPI::MinMaxAvg memory_c =
694 Utilities::MPI::min_max_avg(1e-6 * data_length, communicator);
695 if (n_procs < 2)
696 out << memory_c.min;
697 else
698 out << memory_c.min << "/" << memory_c.avg << "/" << memory_c.max;
699 out << " MB" << std::endl;
700 }
701
702
703
704 std::size_t
705 TaskInfo::memory_consumption() const
706 {
707 return (
708 sizeof(*this) +
709 MemoryConsumption::memory_consumption(partition_row_index) +
710 MemoryConsumption::memory_consumption(cell_partition_data) +
711 MemoryConsumption::memory_consumption(face_partition_data) +
712 MemoryConsumption::memory_consumption(boundary_partition_data) +
713 MemoryConsumption::memory_consumption(partition_evens) +
714 MemoryConsumption::memory_consumption(partition_odds) +
715 MemoryConsumption::memory_consumption(partition_n_blocked_workers) +
716 MemoryConsumption::memory_consumption(partition_n_workers));
717 }
718
719
720
721 void
722 TaskInfo::make_boundary_cells_divisible(
723 std::vector<unsigned int> &boundary_cells)
724 {
725 // try to make the number of boundary cells divisible by the number of
726 // vectors in vectorization
727 unsigned int fillup_needed =
728 (vectorization_length - boundary_cells.size() % vectorization_length) %
729 vectorization_length;
730 if (fillup_needed > 0 && boundary_cells.size() < n_active_cells)
731 {
732 // fill additional cells into the list of boundary cells to get a
733 // balanced number. Go through the indices successively until we
734 // found enough indices
735 std::vector<unsigned int> new_boundary_cells;
736 new_boundary_cells.reserve(boundary_cells.size());
737
738 unsigned int next_free_slot = 0, bound_index = 0;
739 while (fillup_needed > 0 && bound_index < boundary_cells.size())
740 {
741 if (next_free_slot < boundary_cells[bound_index])
742 {
743 // check if there are enough cells to fill with in the
744 // current slot
745 if (next_free_slot + fillup_needed <=
746 boundary_cells[bound_index])
747 {
748 for (unsigned int j =
749 boundary_cells[bound_index] - fillup_needed;
750 j < boundary_cells[bound_index];
751 ++j)
752 new_boundary_cells.push_back(j);
753 fillup_needed = 0;
754 }
755 // ok, not enough indices, so just take them all up to the
756 // next boundary cell
757 else
758 {
759 for (unsigned int j = next_free_slot;
760 j < boundary_cells[bound_index];
761 ++j)
762 new_boundary_cells.push_back(j);
763 fillup_needed -=
764 boundary_cells[bound_index] - next_free_slot;
765 }
766 }
767 new_boundary_cells.push_back(boundary_cells[bound_index]);
768 next_free_slot = boundary_cells[bound_index] + 1;
769 ++bound_index;
770 }
771 while (fillup_needed > 0 &&
772 (new_boundary_cells.size() == 0 ||
773 new_boundary_cells.back() < n_active_cells - 1))
774 new_boundary_cells.push_back(new_boundary_cells.back() + 1);
775 while (bound_index < boundary_cells.size())
776 new_boundary_cells.push_back(boundary_cells[bound_index++]);
777
778 boundary_cells.swap(new_boundary_cells);
779 }
780
781 // set the number of cells
782 std::sort(boundary_cells.begin(), boundary_cells.end());
783
784 // check that number of boundary cells is divisible by
785 // vectorization_length or that it contains all cells
786 Assert(boundary_cells.size() % vectorization_length == 0 ||
787 boundary_cells.size() == n_active_cells,
788 ExcInternalError());
789 }
790
791
792
793 void
794 TaskInfo::create_blocks_serial(
795 const std::vector<unsigned int> &cells_with_comm,
796 const unsigned int dofs_per_cell,
797 const bool categories_are_hp,
798 const std::vector<unsigned int> &cell_vectorization_categories,
799 const bool cell_vectorization_categories_strict,
800 const std::vector<unsigned int> &parent_relation,
801 std::vector<unsigned int> & renumbering,
802 std::vector<unsigned char> & incompletely_filled_vectorization)
803 {
804 Assert(dofs_per_cell > 0, ExcInternalError());
805 // This function is decomposed into several steps to determine a good
806 // ordering that satisfies the following constraints:
807 // a. Only cells belonging to the same category (or next higher if the
808 // cell_vectorization_categories_strict is false) can be grouped into
809 // the same SIMD batch
810 // b. hp-adaptive computations must form contiguous ranges for the same
811 // degree (category) in cell_partition_data
812 // c. We want to group the cells with the same parent in the same SIMD
813 // lane if possible
814 // d. The cell order should be similar to the initial one
815 // e. Form sets without MPI communication and those with to overlap
816 // communication with computation
817 //
818 // These constraints are satisfied by first grouping by the categories
819 // and, within the groups, to distinguish between cells with a parent
820 // and those without. All of this is set up with batches of cells (with
821 // padding if the size does not match). Then we define a vector of
822 // arrays where we define sorting criteria for the cell batches to
823 // satisfy the items b and d together, split by different parts to
824 // satisfy item e.
825
826 // Give the compiler a chance to detect that vectorization_length is a
827 // power of two, which allows it to replace integer divisions by shifts
828 unsigned int vectorization_length_bits = 0;
829 unsigned int my_length = vectorization_length;
830 while ((my_length >>= 1) != 0u)
831 ++vectorization_length_bits;
832 const unsigned int n_lanes = 1 << vectorization_length_bits;
833
834 // Step 1: find tight map of categories for not taking exceeding amounts
835 // of memory below. Sort the new categories by the numbers in the
836 // old one to ensure we respect the given rules
837 unsigned int n_categories = 1;
838 std::vector<unsigned int> tight_category_map(n_active_cells, 0);
839 if (cell_vectorization_categories.empty() == false)
840 {
841 AssertDimension(cell_vectorization_categories.size(),
842 n_active_cells + n_ghost_cells);
843
844 std::set<unsigned int> used_categories;
845 for (unsigned int i = 0; i < n_active_cells; ++i)
846 used_categories.insert(cell_vectorization_categories[i]);
847 std::vector<unsigned int> used_categories_vector(
848 used_categories.size());
849 n_categories = 0;
850 for (const auto &it : used_categories)
851 used_categories_vector[n_categories++] = it;
852 for (unsigned int i = 0; i < n_active_cells; ++i)
853 {
854 const unsigned int index =
855 std::lower_bound(used_categories_vector.begin(),
856 used_categories_vector.end(),
857 cell_vectorization_categories[i]) -
858 used_categories_vector.begin();
859 AssertIndexRange(index, used_categories_vector.size());
860 tight_category_map[i] = index;
861 }
862 }
863
864 // Step 2: Sort the cells by the category. If we want to fill up the
865 // ranges in vectorization, promote some of the cells to a higher
866 // category
867 std::vector<std::vector<unsigned int>> renumbering_category(n_categories);
868 for (unsigned int i = 0; i < n_active_cells; ++i)
869 renumbering_category[tight_category_map[i]].push_back(i);
870
871 if (cell_vectorization_categories_strict == false && n_categories > 1)
872 for (unsigned int j = n_categories - 1; j > 0; --j)
873 {
874 unsigned int lower_index = j - 1;
875 while ((renumbering_category[j].size() % n_lanes) != 0u)
876 {
877 while (((renumbering_category[j].size() % n_lanes) != 0u) &&
878 !renumbering_category[lower_index].empty())
879 {
880 renumbering_category[j].push_back(
881 renumbering_category[lower_index].back());
882 renumbering_category[lower_index].pop_back();
883 }
884 if (lower_index == 0)
885 break;
886 else
887 --lower_index;
888 }
889 }
890
891 // Step 3: Use the parent relation to find a good grouping of cells. To
892 // do this, we first put cells of each category defined above into two
893 // bins, those which we know can be grouped together by the given parent
894 // relation and those which cannot
895 std::vector<unsigned int> temporary_numbering;
896 temporary_numbering.reserve(n_active_cells +
897 (n_lanes - 1) * n_categories);
898 const unsigned int n_cells_per_parent =
899 std::count(parent_relation.begin(), parent_relation.end(), 0);
900 std::vector<unsigned int> category_size;
901 for (unsigned int j = 0; j < n_categories; ++j)
902 {
903 std::vector<std::pair<unsigned int, unsigned int>> grouped_cells;
904 std::vector<unsigned int> other_cells;
905 for (const unsigned int cell : renumbering_category[j])
906 if (parent_relation.empty() ||
907 parent_relation[cell] == numbers::invalid_unsigned_int)
908 other_cells.push_back(cell);
909 else
910 grouped_cells.emplace_back(parent_relation[cell], cell);
911
912 // Compute the number of cells per group
913 std::sort(grouped_cells.begin(), grouped_cells.end());
914 std::vector<unsigned int> n_cells_per_group;
915 unsigned int length = 0;
916 for (unsigned int i = 0; i < grouped_cells.size(); ++i, ++length)
917 if (i > 0 && grouped_cells[i].first != grouped_cells[i - 1].first)
918 {
919 n_cells_per_group.push_back(length);
920 length = 0;
921 }
922 if (length > 0)
923 n_cells_per_group.push_back(length);
924
925 // Move groups that do not have the complete size (due to
926 // categories) to the 'other_cells'. The cells with correct group
927 // size are immediately appended to the temporary cell numbering
928 auto group_it = grouped_cells.begin();
929 for (unsigned int length : n_cells_per_group)
930 if (length < n_cells_per_parent)
931 for (unsigned int j = 0; j < length; ++j)
932 other_cells.push_back((group_it++)->second);
933 else
934 {
935 // we should not have more cells in a group than in the first
936 // check we did above
937 AssertDimension(length, n_cells_per_parent);
938 for (unsigned int j = 0; j < length; ++j)
939 temporary_numbering.push_back((group_it++)->second);
940 }
941
942 // Sort the remaining cells and append them as well
943 std::sort(other_cells.begin(), other_cells.end());
944 temporary_numbering.insert(temporary_numbering.end(),
945 other_cells.begin(),
946 other_cells.end());
947
948 while (temporary_numbering.size() % n_lanes != 0)
949 temporary_numbering.push_back(numbers::invalid_unsigned_int);
950
951 category_size.push_back(temporary_numbering.size());
952 }
953
954 // Step 4: Identify the batches with cells marked as "comm"
955 std::vector<bool> batch_with_comm(temporary_numbering.size() / n_lanes,
956 false);
957 std::vector<unsigned int> temporary_numbering_inverse(n_active_cells);
958 for (unsigned int i = 0; i < temporary_numbering.size(); ++i)
959 if (temporary_numbering[i] != numbers::invalid_unsigned_int)
960 temporary_numbering_inverse[temporary_numbering[i]] = i;
961 for (const unsigned int cell : cells_with_comm)
962 batch_with_comm[temporary_numbering_inverse[cell] / n_lanes] = true;
963
964 // Step 5: Sort the batches of cells by their last cell index to get
965 // good locality, assuming that the initial cell order is of good
966 // locality. In case we have hp-calculations with categories, we need to
967 // sort also by the category.
968 std::vector<std::array<unsigned int, 3>> batch_order;
969 std::vector<std::array<unsigned int, 3>> batch_order_comm;
970 for (unsigned int i = 0; i < temporary_numbering.size(); i += n_lanes)
971 {
972 unsigned int max_index = 0;
973 for (unsigned int j = 0; j < n_lanes; ++j)
974 if (temporary_numbering[i + j] < numbers::invalid_unsigned_int)
975 max_index = std::max(temporary_numbering[i + j], max_index);
976 const unsigned int category_hp =
977 categories_are_hp ?
978 std::upper_bound(category_size.begin(), category_size.end(), i) -
979 category_size.begin() :
980 0;
981 const std::array<unsigned int, 3> next{{category_hp, max_index, i}};
982 if (batch_with_comm[i / n_lanes])
983 batch_order_comm.emplace_back(next);
984 else
985 batch_order.emplace_back(next);
986 }
987
988 std::sort(batch_order.begin(), batch_order.end());
989 std::sort(batch_order_comm.begin(), batch_order_comm.end());
990
991 // Step 6: Put the cells with communication in the middle of the cell
992 // range. For the MPI case, we need three groups to enable overlap for
993 // communication and computation (part before comm, part with comm, part
994 // after comm), whereas we need one for the other case. And in each
995 // case, we allow for a slot of "ghosted" cells.
996 std::vector<unsigned int> blocks;
997 if (n_procs == 1)
998 {
999 if (batch_order.empty())
1000 std::swap(batch_order_comm, batch_order);
1001 Assert(batch_order_comm.empty(), ExcInternalError());
1002 partition_row_index.resize(3);
1003 blocks = {0, static_cast<unsigned int>(batch_order.size())};
1004 }
1005 else
1006 {
1007 partition_row_index.resize(5);
1008 const unsigned int comm_begin = batch_order.size() / 2;
1009 batch_order.insert(batch_order.begin() + comm_begin,
1010 batch_order_comm.begin(),
1011 batch_order_comm.end());
1012 const unsigned int comm_end = comm_begin + batch_order_comm.size();
1013 const unsigned int end = batch_order.size();
1014 blocks = {0, comm_begin, comm_end, end};
1015 }
1016
1017 // Step 7: Fill in the data by batches for the locally owned cells.
1018 const unsigned int n_cell_batches = batch_order.size();
1019 const unsigned int n_ghost_batches =
1020 (n_ghost_cells + n_lanes - 1) / n_lanes;
1021 incompletely_filled_vectorization.resize(n_cell_batches +
1022 n_ghost_batches);
1023
1024 cell_partition_data.clear();
1025 cell_partition_data.resize(1, 0);
1026
1027 renumbering.clear();
1028 renumbering.resize(n_active_cells + n_ghost_cells,
1029 numbers::invalid_unsigned_int);
1030
1031 unsigned int counter = 0;
1032 for (unsigned int block = 0; block < blocks.size() - 1; ++block)
1033 {
1034 const unsigned int grain_size =
1035 std::max((2048U / dofs_per_cell) / 8 * 4, 2U);
1036 for (unsigned int k = blocks[block]; k < blocks[block + 1];
1037 k += grain_size)
1038 cell_partition_data.push_back(
1039 std::min(k + grain_size, blocks[block + 1]));
1040 partition_row_index[block + 1] = cell_partition_data.size() - 1;
1041
1042 // Set the numbering according to the reordered temporary one
1043 for (unsigned int k = blocks[block]; k < blocks[block + 1]; ++k)
1044 {
1045 const unsigned int pos = batch_order[k][2];
1046 unsigned int j = 0;
1047 for (; j < n_lanes && temporary_numbering[pos + j] !=
1048 numbers::invalid_unsigned_int;
1049 ++j)
1050 renumbering[counter++] = temporary_numbering[pos + j];
1051 if (j < n_lanes)
1052 incompletely_filled_vectorization[k] = j;
1053 }
1054 }
1055 AssertDimension(counter, n_active_cells);
1056
1057 // Step 8: Treat the ghost cells
1058 for (unsigned int cell = n_active_cells;
1059 cell < n_active_cells + n_ghost_cells;
1060 ++cell)
1061 {
1062 if (!cell_vectorization_categories.empty())
1063 AssertDimension(cell_vectorization_categories[cell],
1064 cell_vectorization_categories[n_active_cells]);
1065 renumbering[cell] = cell;
1066 }
1067 if ((n_ghost_cells % n_lanes) != 0u)
1068 incompletely_filled_vectorization.back() = n_ghost_cells % n_lanes;
1069 cell_partition_data.push_back(n_cell_batches + n_ghost_batches);
1070 partition_row_index.back() = cell_partition_data.size() - 1;
1071
1072#ifdef DEBUG
1073 std::vector<unsigned int> renumber_cpy(renumbering);
1074 std::sort(renumber_cpy.begin(), renumber_cpy.end());
1075 for (unsigned int i = 0; i < renumber_cpy.size(); ++i)
1076 AssertDimension(i, renumber_cpy[i]);
1077#endif
1078 }
1079
1080
1081
1082 void
1083 TaskInfo::initial_setup_blocks_tasks(
1084 const std::vector<unsigned int> &boundary_cells,
1085 std::vector<unsigned int> & renumbering,
1086 std::vector<unsigned char> & incompletely_filled_vectorization)
1087 {
1088 const unsigned int n_cell_batches =
1089 (n_active_cells + vectorization_length - 1) / vectorization_length;
1090 const unsigned int n_ghost_slots =
1091 (n_ghost_cells + vectorization_length - 1) / vectorization_length;
1092 incompletely_filled_vectorization.resize(n_cell_batches + n_ghost_slots);
1093 if (n_cell_batches * vectorization_length > n_active_cells)
1094 incompletely_filled_vectorization[n_cell_batches - 1] =
1095 vectorization_length -
1096 (n_cell_batches * vectorization_length - n_active_cells);
1097 if (n_ghost_slots * vectorization_length > n_ghost_cells)
1098 incompletely_filled_vectorization[n_cell_batches + n_ghost_slots - 1] =
1099 vectorization_length -
1100 (n_ghost_slots * vectorization_length - n_ghost_cells);
1101
1102 std::vector<unsigned int> reverse_numbering(
1103 n_active_cells, numbers::invalid_unsigned_int);
1104 for (unsigned int j = 0; j < boundary_cells.size(); ++j)
1105 reverse_numbering[boundary_cells[j]] = j;
1106 unsigned int counter = boundary_cells.size();
1107 for (unsigned int j = 0; j < n_active_cells; ++j)
1108 if (reverse_numbering[j] == numbers::invalid_unsigned_int)
1109 reverse_numbering[j] = counter++;
1110
1111 AssertDimension(counter, n_active_cells);
1112 renumbering = Utilities::invert_permutation(reverse_numbering);
1113
1114 for (unsigned int j = n_active_cells; j < n_active_cells + n_ghost_cells;
1115 ++j)
1116 renumbering.push_back(j);
1117
1118 // TODO: might be able to simplify this code by not relying on the cell
1119 // partition data while computing the thread graph
1120 cell_partition_data.clear();
1121 cell_partition_data.push_back(0);
1122 if (n_procs > 1)
1123 {
1124 const unsigned int n_macro_boundary_cells =
1125 (boundary_cells.size() + vectorization_length - 1) /
1126 vectorization_length;
1127 cell_partition_data.push_back(
1128 (n_cell_batches - n_macro_boundary_cells) / 2);
1129 cell_partition_data.push_back(cell_partition_data[1] +
1130 n_macro_boundary_cells);
1131 }
1132 else
1133 AssertDimension(boundary_cells.size(), 0);
1134 cell_partition_data.push_back(n_cell_batches);
1135 cell_partition_data.push_back(cell_partition_data.back() + n_ghost_slots);
1136 partition_row_index.resize(n_procs > 1 ? 4 : 2);
1137 partition_row_index[0] = 0;
1138 partition_row_index[1] = 1;
1139 if (n_procs > 1)
1140 {
1141 partition_row_index[2] = 2;
1142 partition_row_index[3] = 3;
1143 }
1144 }
1145
1146
1147
1148 void
1149 TaskInfo::guess_block_size(const unsigned int dofs_per_cell)
1150 {
1151 // user did not say a positive number, so we have to guess
1152 if (block_size == 0)
1153 {
1154 // we would like to have enough work to do, so as first guess, try
1155 // to get 16 times as many chunks as we have threads on the system.
1156 block_size = n_active_cells / (MultithreadInfo::n_threads() * 16 *
1157 vectorization_length);
1158
1159 // if there are too few degrees of freedom per cell, need to
1160 // increase the block size
1161 const unsigned int minimum_parallel_grain_size = 200;
1162 if (dofs_per_cell * block_size < minimum_parallel_grain_size)
1163 block_size = (minimum_parallel_grain_size / dofs_per_cell + 1);
1164 if (dofs_per_cell * block_size > 10000)
1165 block_size /= 4;
1166
1167 block_size =
1168 1 << static_cast<unsigned int>(std::log2(block_size + 1));
1169 }
1170 if (block_size > n_active_cells)
1171 block_size = std::max(1U, n_active_cells);
1172 }
1173
1174
1175
1176 void
1177 TaskInfo::make_thread_graph_partition_color(
1178 DynamicSparsityPattern & connectivity_large,
1179 std::vector<unsigned int> & renumbering,
1180 std::vector<unsigned char> &irregular_cells,
1181 const bool)
1182 {
1183 const unsigned int n_cell_batches = *(cell_partition_data.end() - 2);
1184 if (n_cell_batches == 0)
1185 return;
1186
1187 Assert(vectorization_length > 0, ExcInternalError());
1188
1189 unsigned int partition = 0, counter = 0;
1190
1191 // Create connectivity graph for blocks based on connectivity graph for
1192 // cells.
1193 DynamicSparsityPattern connectivity(n_blocks, n_blocks);
1194 make_connectivity_cells_to_blocks(irregular_cells,
1195 connectivity_large,
1196 connectivity);
1197
1198 // Create cell-block partitioning.
1199
1200 // For each block of cells, this variable saves to which partitions the
1201 // block belongs. Initialize all to -1 to mark them as not yet assigned
1202 // a partition.
1203 std::vector<unsigned int> cell_partition(n_blocks,
1204 numbers::invalid_unsigned_int);
1205
1206 // In element j of this variable, one puts the old number of the block
1207 // that should be the jth block in the new numeration.
1208 std::vector<unsigned int> partition_list(n_blocks, 0);
1209 std::vector<unsigned int> partition_color_list(n_blocks, 0);
1210
1211 // This vector points to the start of each partition.
1212 std::vector<unsigned int> partition_size(2, 0);
1213
1214 // blocking_connectivity = true;
1215
1216 // The cluster_size in make_partitioning defines that the no. of cells
1217 // in each partition should be a multiple of cluster_size.
1218 unsigned int cluster_size = 1;
1219
1220 // Make the partitioning of the first layer of the blocks of cells.
1221 make_partitioning(connectivity,
1222 cluster_size,
1223 cell_partition,
1224 partition_list,
1225 partition_size,
1226 partition);
1227
1228 // Color the cells within each partition
1229 make_coloring_within_partitions_pre_blocked(connectivity,
1230 partition,
1231 cell_partition,
1232 partition_list,
1233 partition_size,
1234 partition_color_list);
1235
1236 partition_list = renumbering;
1237
1238#ifdef DEBUG
1239 // in debug mode, check that the partition color list is one-to-one
1240 {
1241 std::vector<unsigned int> sorted_pc_list(partition_color_list);
1242 std::sort(sorted_pc_list.begin(), sorted_pc_list.end());
1243 for (unsigned int i = 0; i < sorted_pc_list.size(); ++i)
1244 Assert(sorted_pc_list[i] == i, ExcInternalError());
1245 }
1246#endif
1247
1248 // set the start list for each block and compute the renumbering of
1249 // cells
1250 std::vector<unsigned int> block_start(n_cell_batches + 1);
1251 std::vector<unsigned char> irregular(n_cell_batches);
1252
1253 unsigned int mcell_start = 0;
1254 block_start[0] = 0;
1255 for (unsigned int block = 0; block < n_blocks; ++block)
1256 {
1257 block_start[block + 1] = block_start[block];
1258 for (unsigned int mcell = mcell_start;
1259 mcell < std::min(mcell_start + block_size, n_cell_batches);
1260 ++mcell)
1261 {
1262 unsigned int n_comp = (irregular_cells[mcell] > 0) ?
1263 irregular_cells[mcell] :
1264 vectorization_length;
1265 block_start[block + 1] += n_comp;
1266 ++counter;
1267 }
1268 mcell_start += block_size;
1269 }
1270 counter = 0;
1271 unsigned int counter_macro = 0;
1272 unsigned int block_size_last =
1273 n_cell_batches - block_size * (n_blocks - 1);
1274 if (block_size_last == 0)
1275 block_size_last = block_size;
1276
1277 unsigned int tick = 0;
1278 for (unsigned int block = 0; block < n_blocks; ++block)
1279 {
1280 unsigned int present_block = partition_color_list[block];
1281 for (unsigned int cell = block_start[present_block];
1282 cell < block_start[present_block + 1];
1283 ++cell)
1284 renumbering[counter++] = partition_list[cell];
1285 unsigned int this_block_size =
1286 (present_block == n_blocks - 1) ? block_size_last : block_size;
1287
1288 // Also re-compute the content of cell_partition_data to
1289 // contain the numbers of cells, not blocks
1290 if (cell_partition_data[tick] == block)
1291 cell_partition_data[tick++] = counter_macro;
1292
1293 for (unsigned int j = 0; j < this_block_size; ++j)
1294 irregular[counter_macro++] =
1295 irregular_cells[present_block * block_size + j];
1296 }
1297 AssertDimension(tick + 1, cell_partition_data.size());
1298 cell_partition_data.back() = counter_macro;
1299
1300 irregular_cells.swap(irregular);
1301 AssertDimension(counter, n_active_cells);
1302 AssertDimension(counter_macro, n_cell_batches);
1303
1304 // check that the renumbering is one-to-one
1305#ifdef DEBUG
1306 {
1307 std::vector<unsigned int> sorted_renumbering(renumbering);
1308 std::sort(sorted_renumbering.begin(), sorted_renumbering.end());
1309 for (unsigned int i = 0; i < sorted_renumbering.size(); ++i)
1310 Assert(sorted_renumbering[i] == i, ExcInternalError());
1311 }
1312#endif
1313
1314
1315 update_task_info(
1316 partition); // Actually sets too much for partition color case
1317
1318 AssertDimension(cell_partition_data.back(), n_cell_batches);
1319 }
1320
1321
1322
1323 void
1324 TaskInfo::make_thread_graph(
1325 const std::vector<unsigned int> &cell_active_fe_index,
1326 DynamicSparsityPattern & connectivity,
1327 std::vector<unsigned int> & renumbering,
1328 std::vector<unsigned char> & irregular_cells,
1329 const bool hp_bool)
1330 {
1331 const unsigned int n_cell_batches = *(cell_partition_data.end() - 2);
1332 if (n_cell_batches == 0)
1333 return;
1334
1335 Assert(vectorization_length > 0, ExcInternalError());
1336
1337 // if we want to block before partitioning, create connectivity graph
1338 // for blocks based on connectivity graph for cells.
1339 DynamicSparsityPattern connectivity_blocks(n_blocks, n_blocks);
1340 make_connectivity_cells_to_blocks(irregular_cells,
1341 connectivity,
1342 connectivity_blocks);
1343
1344 unsigned int n_blocks = 0;
1345 if (scheme == partition_color ||
1346 scheme == color) // blocking_connectivity == true
1347 n_blocks = this->n_blocks;
1348 else
1349 n_blocks = n_active_cells;
1350
1351 // For each block of cells, this variable saves to which partitions the
1352 // block belongs. Initialize all to -1 to mark them as not yet assigned
1353 // a partition.
1354 std::vector<unsigned int> cell_partition(n_blocks,
1355 numbers::invalid_unsigned_int);
1356
1357 // In element j of this variable, one puts the old number (but after
1358 // renumbering according to the input renumbering) of the block that
1359 // should be the jth block in the new numeration.
1360 std::vector<unsigned int> partition_list(n_blocks, 0);
1361 std::vector<unsigned int> partition_2layers_list(n_blocks, 0);
1362
1363 // This vector points to the start of each partition.
1364 std::vector<unsigned int> partition_size(2, 0);
1365
1366 unsigned int partition = 0;
1367
1368 // Within the partitions we want to be able to block for the case that
1369 // we do not block already in the connectivity. The cluster_size in
1370 // make_partitioning defines that the no. of cells in each partition
1371 // should be a multiple of cluster_size.
1372 unsigned int cluster_size = 1;
1373 if (scheme == partition_partition)
1374 cluster_size = block_size * vectorization_length;
1375
1376 // Make the partitioning of the first layer of the blocks of cells.
1377 if (scheme == partition_color || scheme == color)
1378 make_partitioning(connectivity_blocks,
1379 cluster_size,
1380 cell_partition,
1381 partition_list,
1382 partition_size,
1383 partition);
1384 else
1385 make_partitioning(connectivity,
1386 cluster_size,
1387 cell_partition,
1388 partition_list,
1389 partition_size,
1390 partition);
1391
1392 // Partition or color second layer
1393 if (scheme == partition_partition)
1394
1395 {
1396 // Partition within partitions.
1397 make_partitioning_within_partitions_post_blocked(
1398 connectivity,
1399 cell_active_fe_index,
1400 partition,
1401 cluster_size,
1402 hp_bool,
1403 cell_partition,
1404 partition_list,
1405 partition_size,
1406 partition_2layers_list,
1407 irregular_cells);
1408 }
1409 else if (scheme == partition_color || scheme == color)
1410 {
1411 make_coloring_within_partitions_pre_blocked(connectivity_blocks,
1412 partition,
1413 cell_partition,
1414 partition_list,
1415 partition_size,
1416 partition_2layers_list);
1417 }
1418
1419 // in debug mode, check that the partition_2layers_list is one-to-one
1420#ifdef DEBUG
1421 {
1422 std::vector<unsigned int> sorted_pc_list(partition_2layers_list);
1423 std::sort(sorted_pc_list.begin(), sorted_pc_list.end());
1424 for (unsigned int i = 0; i < sorted_pc_list.size(); ++i)
1425 Assert(sorted_pc_list[i] == i, ExcInternalError());
1426 }
1427#endif
1428
1429 // Set the new renumbering
1430 std::vector<unsigned int> renumbering_in(n_active_cells, 0);
1431 renumbering_in.swap(renumbering);
1432 if (scheme == partition_partition) // blocking_connectivity == false
1433 {
1434 // This is the simple case. The renumbering is just a combination of
1435 // the renumbering that we were given as an input and the
1436 // renumbering of partition/coloring given in partition_2layers_list
1437 for (unsigned int j = 0; j < renumbering.size(); ++j)
1438 renumbering[j] = renumbering_in[partition_2layers_list[j]];
1439 // Account for the ghost cells, finally.
1440 for (unsigned int i = 0; i < n_ghost_cells; ++i)
1441 renumbering.push_back(i + n_active_cells);
1442 }
1443 else
1444 {
1445 // set the start list for each block and compute the renumbering of
1446 // cells
1447 std::vector<unsigned int> block_start(n_cell_batches + 1);
1448 std::vector<unsigned char> irregular(n_cell_batches);
1449
1450 unsigned int counter = 0;
1451 unsigned int mcell_start = 0;
1452 block_start[0] = 0;
1453 for (unsigned int block = 0; block < n_blocks; ++block)
1454 {
1455 block_start[block + 1] = block_start[block];
1456 for (unsigned int mcell = mcell_start;
1457 mcell < std::min(mcell_start + block_size, n_cell_batches);
1458 ++mcell)
1459 {
1460 unsigned int n_comp = (irregular_cells[mcell] > 0) ?
1461 irregular_cells[mcell] :
1462 vectorization_length;
1463 block_start[block + 1] += n_comp;
1464 ++counter;
1465 }
1466 mcell_start += block_size;
1467 }
1468 counter = 0;
1469 unsigned int counter_macro = 0;
1470 unsigned int block_size_last =
1471 n_cell_batches - block_size * (n_blocks - 1);
1472 if (block_size_last == 0)
1473 block_size_last = block_size;
1474
1475 unsigned int tick = 0;
1476 for (unsigned int block = 0; block < n_blocks; ++block)
1477 {
1478 unsigned int present_block = partition_2layers_list[block];
1479 for (unsigned int cell = block_start[present_block];
1480 cell < block_start[present_block + 1];
1481 ++cell)
1482 renumbering[counter++] = renumbering_in[cell];
1483 unsigned int this_block_size =
1484 (present_block == n_blocks - 1) ? block_size_last : block_size;
1485
1486 // Also re-compute the content of cell_partition_data to
1487 // contain the numbers of cells, not blocks
1488 if (cell_partition_data[tick] == block)
1489 cell_partition_data[tick++] = counter_macro;
1490
1491 for (unsigned int j = 0; j < this_block_size; ++j)
1492 irregular[counter_macro++] =
1493 irregular_cells[present_block * block_size + j];
1494 }
1495 AssertDimension(tick + 1, cell_partition_data.size());
1496 cell_partition_data.back() = counter_macro;
1497
1498 irregular_cells.swap(irregular);
1499 AssertDimension(counter, n_active_cells);
1500 AssertDimension(counter_macro, n_cell_batches);
1501 // check that the renumbering is one-to-one
1502#ifdef DEBUG
1503 {
1504 std::vector<unsigned int> sorted_renumbering(renumbering);
1505 std::sort(sorted_renumbering.begin(), sorted_renumbering.end());
1506 for (unsigned int i = 0; i < sorted_renumbering.size(); ++i)
1507 Assert(sorted_renumbering[i] == i, ExcInternalError());
1508 }
1509#endif
1510 }
1511
1512 // Update the task_info with the more information for the thread graph.
1513 update_task_info(partition);
1514 }
1515
1516
1517
1518 void
1519 TaskInfo::make_thread_graph_partition_partition(
1520 const std::vector<unsigned int> &cell_active_fe_index,
1521 DynamicSparsityPattern & connectivity,
1522 std::vector<unsigned int> & renumbering,
1523 std::vector<unsigned char> & irregular_cells,
1524 const bool hp_bool)
1525 {
1526 const unsigned int n_cell_batches = *(cell_partition_data.end() - 2);
1527 if (n_cell_batches == 0)
1528 return;
1529
1530 const unsigned int cluster_size = block_size * vectorization_length;
1531
1532 // Create cell-block partitioning.
1533
1534 // For each block of cells, this variable saves to which partitions the
1535 // block belongs. Initialize all to n_cell_batches to mark them as not
1536 // yet assigned a partition.
1537 std::vector<unsigned int> cell_partition(n_active_cells,
1538 numbers::invalid_unsigned_int);
1539
1540
1541 // In element j of this variable, one puts the old number of the block
1542 // that should be the jth block in the new numeration.
1543 std::vector<unsigned int> partition_list(n_active_cells, 0);
1544 std::vector<unsigned int> partition_partition_list(n_active_cells, 0);
1545
1546 // This vector points to the start of each partition.
1547 std::vector<unsigned int> partition_size(2, 0);
1548
1549 unsigned int partition = 0;
1550 // Here, we do not block inside the connectivity graph
1551 // blocking_connectivity = false;
1552
1553 // Make the partitioning of the first layer of the blocks of cells.
1554 make_partitioning(connectivity,
1555 cluster_size,
1556 cell_partition,
1557 partition_list,
1558 partition_size,
1559 partition);
1560
1561 // Partition within partitions.
1562 make_partitioning_within_partitions_post_blocked(connectivity,
1563 cell_active_fe_index,
1564 partition,
1565 cluster_size,
1566 hp_bool,
1567 cell_partition,
1568 partition_list,
1569 partition_size,
1570 partition_partition_list,
1571 irregular_cells);
1572
1573 partition_list.swap(renumbering);
1574
1575 for (unsigned int j = 0; j < renumbering.size(); ++j)
1576 renumbering[j] = partition_list[partition_partition_list[j]];
1577
1578 for (unsigned int i = 0; i < n_ghost_cells; ++i)
1579 renumbering.push_back(i + n_active_cells);
1580
1581 update_task_info(partition);
1582 }
1583
1584
1585
1586 void
1587 TaskInfo::make_connectivity_cells_to_blocks(
1588 const std::vector<unsigned char> &irregular_cells,
1589 const DynamicSparsityPattern & connectivity_cells,
1590 DynamicSparsityPattern & connectivity_blocks) const
1591 {
1592 std::vector<std::vector<unsigned int>> cell_blocks(n_blocks);
1593 std::vector<unsigned int> touched_cells(n_active_cells);
1594 unsigned int cell = 0;
1595 for (unsigned int i = 0, mcell = 0; i < n_blocks; ++i)
1596 {
1597 for (unsigned int c = 0;
1598 c < block_size && mcell < *(cell_partition_data.end() - 2);
1599 ++c, ++mcell)
1600 {
1601 unsigned int ncomp = (irregular_cells[mcell] > 0) ?
1602 irregular_cells[mcell] :
1603 vectorization_length;
1604 for (unsigned int c = 0; c < ncomp; ++c, ++cell)
1605 {
1606 cell_blocks[i].push_back(cell);
1607 touched_cells[cell] = i;
1608 }
1609 }
1610 }
1611 AssertDimension(cell, n_active_cells);
1612 for (unsigned int i = 0; i < cell_blocks.size(); ++i)
1613 for (unsigned int col = 0; col < cell_blocks[i].size(); ++col)
1614 {
1615 for (DynamicSparsityPattern::iterator it =
1616 connectivity_cells.begin(cell_blocks[i][col]);
1617 it != connectivity_cells.end(cell_blocks[i][col]);
1618 ++it)
1619 {
1620 if (touched_cells[it->column()] != i)
1621 connectivity_blocks.add(i, touched_cells[it->column()]);
1622 }
1623 }
1624 }
1625
1626
1627
1628 // Function to create partitioning on the second layer within each
1629 // partition. Version without preblocking.
1630 void
1631 TaskInfo::make_partitioning_within_partitions_post_blocked(
1632 const DynamicSparsityPattern & connectivity,
1633 const std::vector<unsigned int> &cell_active_fe_index,
1634 const unsigned int partition,
1635 const unsigned int cluster_size,
1636 const bool hp_bool,
1637 const std::vector<unsigned int> &cell_partition,
1638 const std::vector<unsigned int> &partition_list,
1639 const std::vector<unsigned int> &partition_size,
1640 std::vector<unsigned int> & partition_partition_list,
1641 std::vector<unsigned char> & irregular_cells)
1642 {
1643 const unsigned int n_cell_batches = *(cell_partition_data.end() - 2);
1644 const unsigned int n_ghost_slots =
1645 *(cell_partition_data.end() - 1) - n_cell_batches;
1646
1647 // List of cells in previous partition
1648 std::vector<unsigned int> neighbor_list;
1649 // List of cells in current partition for use as neighbors in next
1650 // partition
1651 std::vector<unsigned int> neighbor_neighbor_list;
1652
1653 std::vector<unsigned int> renumbering(n_active_cells);
1654
1655 irregular_cells.back() = 0;
1656 irregular_cells.resize(n_active_cells + n_ghost_slots);
1657
1658 unsigned int max_fe_index = 0;
1659 for (const unsigned int fe_index : cell_active_fe_index)
1660 max_fe_index = std::max(fe_index, max_fe_index);
1661
1662 Assert(!hp_bool || cell_active_fe_index.size() == n_active_cells,
1663 ExcInternalError());
1664
1665 {
1666 unsigned int n_cell_batches_before = 0;
1667 // Create partitioning within partitions.
1668
1669 // For each block of cells, this variable saves to which partitions
1670 // the block belongs. Initialize all to n_cell_batches to mark them as
1671 // not yet assigned a partition.
1672 std::vector<unsigned int> cell_partition_l2(
1673 n_active_cells, numbers::invalid_unsigned_int);
1674 partition_row_index.clear();
1675 partition_row_index.resize(partition + 1, 0);
1676 cell_partition_data.resize(1, 0);
1677
1678 unsigned int counter = 0;
1679 unsigned int missing_macros;
1680 for (unsigned int part = 0; part < partition; ++part)
1681 {
1682 neighbor_neighbor_list.resize(0);
1683 neighbor_list.resize(0);
1684 bool work = true;
1685 unsigned int partition_l2 = 0;
1686 unsigned int start_up = partition_size[part];
1687 unsigned int partition_counter = 0;
1688 while (work)
1689 {
1690 if (neighbor_list.size() == 0)
1691 {
1692 work = false;
1693 partition_counter = 0;
1694 for (unsigned int j = start_up;
1695 j < partition_size[part + 1];
1696 ++j)
1697 if (cell_partition[partition_list[j]] == part &&
1698 cell_partition_l2[partition_list[j]] ==
1699 numbers::invalid_unsigned_int)
1700 {
1701 start_up = j;
1702 work = true;
1703 partition_counter = 1;
1704 // To start up, set the start_up cell to partition
1705 // and list all its neighbors.
1706 AssertIndexRange(start_up, partition_size[part + 1]);
1707 cell_partition_l2[partition_list[start_up]] =
1708 partition_l2;
1709 neighbor_neighbor_list.push_back(
1710 partition_list[start_up]);
1711 partition_partition_list[counter++] =
1712 partition_list[start_up];
1713 start_up++;
1714 break;
1715 }
1716 }
1717 else
1718 {
1719 partition_counter = 0;
1720 for (const unsigned int neighbor : neighbor_list)
1721 {
1722 Assert(cell_partition[neighbor] == part,
1723 ExcInternalError());
1724 Assert(cell_partition_l2[neighbor] == partition_l2 - 1,
1725 ExcInternalError());
1726 auto neighbor_it = connectivity.begin(neighbor);
1727 const auto end_it = connectivity.end(neighbor);
1728 for (; neighbor_it != end_it; ++neighbor_it)
1729 {
1730 if (cell_partition[neighbor_it->column()] == part &&
1731 cell_partition_l2[neighbor_it->column()] ==
1732 numbers::invalid_unsigned_int)
1733 {
1734 cell_partition_l2[neighbor_it->column()] =
1735 partition_l2;
1736 neighbor_neighbor_list.push_back(
1737 neighbor_it->column());
1738 partition_partition_list[counter++] =
1739 neighbor_it->column();
1740 partition_counter++;
1741 }
1742 }
1743 }
1744 }
1745 if (partition_counter > 0)
1746 {
1747 int index_before = neighbor_neighbor_list.size(),
1748 index = index_before;
1749 {
1750 // put the cells into separate lists for each FE index
1751 // within one partition-partition
1752 missing_macros = 0;
1753 std::vector<unsigned int> remaining_per_cell_batch(
1754 max_fe_index + 1);
1755 std::vector<std::vector<unsigned int>>
1756 renumbering_fe_index;
1757 unsigned int cell;
1758 bool filled = true;
1759 if (hp_bool == true)
1760 {
1761 renumbering_fe_index.resize(max_fe_index + 1);
1762 for (cell = counter - partition_counter;
1763 cell < counter;
1764 ++cell)
1765 {
1766 renumbering_fe_index
1767 [cell_active_fe_index.empty() ?
1768 0 :
1769 cell_active_fe_index
1770 [partition_partition_list[cell]]]
1771 .push_back(partition_partition_list[cell]);
1772 }
1773 // check how many more cells are needed in the lists
1774 for (unsigned int j = 0; j < max_fe_index + 1; ++j)
1775 {
1776 remaining_per_cell_batch[j] =
1777 renumbering_fe_index[j].size() %
1778 vectorization_length;
1779 if (remaining_per_cell_batch[j] != 0)
1780 filled = false;
1781 missing_macros +=
1782 ((renumbering_fe_index[j].size() +
1783 vectorization_length - 1) /
1784 vectorization_length);
1785 }
1786 }
1787 else
1788 {
1789 remaining_per_cell_batch.resize(1);
1790 remaining_per_cell_batch[0] =
1791 partition_counter % vectorization_length;
1792 missing_macros =
1793 partition_counter / vectorization_length;
1794 if (remaining_per_cell_batch[0] != 0)
1795 {
1796 filled = false;
1797 missing_macros++;
1798 }
1799 }
1800 missing_macros =
1801 cluster_size - (missing_macros % cluster_size);
1802
1803 // now we realized that there are some cells missing.
1804 while (missing_macros > 0 || filled == false)
1805 {
1806 if (index == 0)
1807 {
1808 index = neighbor_neighbor_list.size();
1809 if (index == index_before)
1810 {
1811 if (missing_macros != 0)
1812 {
1813 neighbor_neighbor_list.resize(0);
1814 }
1815 start_up--;
1816 break; // not connected - start again
1817 }
1818 index_before = index;
1819 }
1820 index--;
1821 unsigned int additional =
1822 neighbor_neighbor_list[index];
1823
1824 // go through the neighbors of the last cell in the
1825 // current partition and check if we find some to
1826 // fill up with.
1827 DynamicSparsityPattern::iterator neighbor =
1828 connectivity.begin(
1829 additional),
1830 end =
1831 connectivity.end(
1832 additional);
1833 for (; neighbor != end; ++neighbor)
1834 {
1835 if (cell_partition[neighbor->column()] == part &&
1836 cell_partition_l2[neighbor->column()] ==
1837 numbers::invalid_unsigned_int)
1838 {
1839 unsigned int this_index = 0;
1840 if (hp_bool == true)
1841 this_index =
1842 cell_active_fe_index.empty() ?
1843 0 :
1844 cell_active_fe_index[neighbor
1845 ->column()];
1846
1847 // Only add this cell if we need more macro
1848 // cells in the current block or if there is
1849 // a macro cell with the FE index that is
1850 // not yet fully populated
1851 if (missing_macros > 0 ||
1852 remaining_per_cell_batch[this_index] > 0)
1853 {
1854 cell_partition_l2[neighbor->column()] =
1855 partition_l2;
1856 neighbor_neighbor_list.push_back(
1857 neighbor->column());
1858 if (hp_bool == true)
1859 renumbering_fe_index[this_index]
1860 .push_back(neighbor->column());
1861 partition_partition_list[counter] =
1862 neighbor->column();
1863 counter++;
1864 partition_counter++;
1865 if (remaining_per_cell_batch
1866 [this_index] == 0 &&
1867 missing_macros > 0)
1868 missing_macros--;
1869 remaining_per_cell_batch[this_index]++;
1870 if (remaining_per_cell_batch
1871 [this_index] ==
1872 vectorization_length)
1873 {
1874 remaining_per_cell_batch[this_index] =
1875 0;
1876 }
1877 if (missing_macros == 0)
1878 {
1879 filled = true;
1880 for (unsigned int fe_ind = 0;
1881 fe_ind < max_fe_index + 1;
1882 ++fe_ind)
1883 if (remaining_per_cell_batch
1884 [fe_ind] != 0)
1885 filled = false;
1886 }
1887 if (filled == true)
1888 break;
1889 }
1890 }
1891 }
1892 }
1893 if (hp_bool == true)
1894 {
1895 // set the renumbering according to their active FE
1896 // index within one partition-partition which was
1897 // implicitly assumed above
1898 cell = counter - partition_counter;
1899 for (unsigned int j = 0; j < max_fe_index + 1; ++j)
1900 {
1901 for (const unsigned int jj :
1902 renumbering_fe_index[j])
1903 renumbering[cell++] = jj;
1904 if (renumbering_fe_index[j].size() %
1905 vectorization_length !=
1906 0)
1907 irregular_cells[renumbering_fe_index[j].size() /
1908 vectorization_length +
1909 n_cell_batches_before] =
1910 renumbering_fe_index[j].size() %
1911 vectorization_length;
1912 n_cell_batches_before +=
1913 (renumbering_fe_index[j].size() +
1914 vectorization_length - 1) /
1915 vectorization_length;
1916 renumbering_fe_index[j].resize(0);
1917 }
1918 }
1919 else
1920 {
1921 n_cell_batches_before +=
1922 partition_counter / vectorization_length;
1923 if (partition_counter % vectorization_length != 0)
1924 {
1925 irregular_cells[n_cell_batches_before] =
1926 partition_counter % vectorization_length;
1927 n_cell_batches_before++;
1928 }
1929 }
1930 }
1931 cell_partition_data.push_back(n_cell_batches_before);
1932 partition_l2++;
1933 }
1934 neighbor_list = neighbor_neighbor_list;
1935 neighbor_neighbor_list.resize(0);
1936 }
1937 partition_row_index[part + 1] =
1938 partition_row_index[part] + partition_l2;
1939 }
1940 }
1941 if (hp_bool == true)
1942 {
1943 partition_partition_list.swap(renumbering);
1944 }
1945 }
1946
1947
1948
1949 // Function to create coloring on the second layer within each partition.
1950 // Version assumes preblocking.
1951 void
1952 TaskInfo::make_coloring_within_partitions_pre_blocked(
1953 const DynamicSparsityPattern & connectivity,
1954 const unsigned int partition,
1955 const std::vector<unsigned int> &cell_partition,
1956 const std::vector<unsigned int> &partition_list,
1957 const std::vector<unsigned int> &partition_size,
1958 std::vector<unsigned int> & partition_color_list)
1959 {
1960 const unsigned int n_cell_batches = *(cell_partition_data.end() - 2);
1961 std::vector<unsigned int> cell_color(n_blocks, n_cell_batches);
1962 std::vector<bool> color_finder;
1963
1964 partition_row_index.resize(partition + 1);
1965 cell_partition_data.clear();
1966 unsigned int color_counter = 0, index_counter = 0;
1967 for (unsigned int part = 0; part < partition; ++part)
1968 {
1969 partition_row_index[part] = index_counter;
1970 unsigned int max_color = 0;
1971 for (unsigned int k = partition_size[part];
1972 k < partition_size[part + 1];
1973 k++)
1974 {
1975 unsigned int cell = partition_list[k];
1976 unsigned int n_neighbors = connectivity.row_length(cell);
1977
1978 // In the worst case, each neighbor has a different color. So we
1979 // find at least one available color between 0 and n_neighbors.
1980 color_finder.resize(n_neighbors + 1);
1981 for (unsigned int j = 0; j <= n_neighbors; ++j)
1982 color_finder[j] = true;
1983 DynamicSparsityPattern::iterator neighbor =
1984 connectivity.begin(cell),
1985 end = connectivity.end(cell);
1986 for (; neighbor != end; ++neighbor)
1987 {
1988 // Mark the color that a neighbor within the partition has
1989 // as taken
1990 if (cell_partition[neighbor->column()] == part &&
1991 cell_color[neighbor->column()] <= n_neighbors)
1992 color_finder[cell_color[neighbor->column()]] = false;
1993 }
1994 // Choose the smallest color that is not taken for the block
1995 cell_color[cell] = 0;
1996 while (color_finder[cell_color[cell]] == false)
1997 cell_color[cell]++;
1998 if (cell_color[cell] > max_color)
1999 max_color = cell_color[cell];
2000 }
2001 // Reorder within partition: First, all blocks that belong the 0 and
2002 // then so on until those with color max (Note that the smaller the
2003 // number the larger the partition)
2004 for (unsigned int color = 0; color <= max_color; ++color)
2005 {
2006 cell_partition_data.push_back(color_counter);
2007 index_counter++;
2008 for (unsigned int k = partition_size[part];
2009 k < partition_size[part + 1];
2010 k++)
2011 {
2012 unsigned int cell = partition_list[k];
2013 if (cell_color[cell] == color)
2014 {
2015 partition_color_list[color_counter++] = cell;
2016 }
2017 }
2018 }
2019 }
2020 cell_partition_data.push_back(n_blocks);
2021 partition_row_index[partition] = index_counter;
2022 AssertDimension(color_counter, n_blocks);
2023 }
2024
2025
2026 // Function to create partitioning on the first layer.
2027 void
2028 TaskInfo::make_partitioning(const DynamicSparsityPattern &connectivity,
2029 const unsigned int cluster_size,
2030 std::vector<unsigned int> & cell_partition,
2031 std::vector<unsigned int> & partition_list,
2032 std::vector<unsigned int> & partition_size,
2033 unsigned int & partition) const
2034
2035 {
2036 // For each block of cells, this variable saves to which partitions the
2037 // block belongs. Initialize all to n_cell_batches to mark them as not
2038 // yet assigned a partition.
2039 // std::vector<unsigned int> cell_partition (n_active_cells,
2040 // numbers::invalid_unsigned_int);
2041 // List of cells in previous partition
2042 std::vector<unsigned int> neighbor_list;
2043 // List of cells in current partition for use as neighbors in next
2044 // partition
2045 std::vector<unsigned int> neighbor_neighbor_list;
2046
2047 // In element j of this variable, one puts the old number of the block
2048 // that should be the jth block in the new numeration.
2049 // std::vector<unsigned int> partition_list(n_active_cells,0);
2050
2051 // This vector points to the start of each partition.
2052 // std::vector<unsigned int> partition_size(2,0);
2053
2054 partition = 0;
2055 unsigned int counter = 0;
2056 unsigned int start_nonboundary =
2057 cell_partition_data.size() == 5 ?
2058 vectorization_length *
2059 (cell_partition_data[2] - cell_partition_data[1]) :
2060 0;
2061
2062 const unsigned int n_cell_batches = *(cell_partition_data.end() - 2);
2063 if (n_cell_batches == 0)
2064 return;
2065 if (scheme == color)
2066 start_nonboundary = n_cell_batches;
2067 if (scheme == partition_color ||
2068 scheme == color) // blocking_connectivity == true
2069 start_nonboundary = ((start_nonboundary + block_size - 1) / block_size);
2070 unsigned int n_blocks;
2071 if (scheme == partition_color ||
2072 scheme == color) // blocking_connectivity == true
2073 n_blocks = this->n_blocks;
2074 else
2075 n_blocks = n_active_cells;
2076
2077 if (start_nonboundary > n_blocks)
2078 start_nonboundary = n_blocks;
2079
2080
2081 unsigned int start_up = 0;
2082 bool work = true;
2083 unsigned int remainder = cluster_size;
2084
2085 // this performs a classical breath-first search in the connectivity
2086 // graph of the cells under the restriction that the size of the
2087 // partitions should be a multiple of the given block size
2088 while (work)
2089 {
2090 // put the cells with neighbors on remote MPI processes up front
2091 if (start_nonboundary > 0)
2092 {
2093 for (unsigned int cell = 0; cell < start_nonboundary; ++cell)
2094 {
2095 const unsigned int cell_nn = cell;
2096 cell_partition[cell_nn] = partition;
2097 neighbor_list.push_back(cell_nn);
2098 partition_list[counter++] = cell_nn;
2099 partition_size.back()++;
2100 }
2101 start_nonboundary = 0;
2102 remainder -= (start_nonboundary % cluster_size);
2103 if (remainder == cluster_size)
2104 remainder = 0;
2105 }
2106 else
2107 {
2108 // To start up, set the start_up cell to partition and list all
2109 // its neighbors.
2110 cell_partition[start_up] = partition;
2111 neighbor_list.push_back(start_up);
2112 partition_list[counter++] = start_up;
2113 partition_size.back()++;
2114 start_up++;
2115 remainder--;
2116 if (remainder == cluster_size)
2117 remainder = 0;
2118 }
2119 int index_before = neighbor_list.size(), index = index_before,
2120 index_stop = 0;
2121 while (remainder > 0)
2122 {
2123 if (index == index_stop)
2124 {
2125 index = neighbor_list.size();
2126 if (index == index_before)
2127 {
2128 neighbor_list.resize(0);
2129 goto not_connect;
2130 }
2131 index_stop = index_before;
2132 index_before = index;
2133 }
2134 index--;
2135 unsigned int additional = neighbor_list[index];
2136 DynamicSparsityPattern::iterator neighbor =
2137 connectivity.begin(additional),
2138 end =
2139 connectivity.end(additional);
2140 for (; neighbor != end; ++neighbor)
2141 {
2142 if (cell_partition[neighbor->column()] ==
2143 numbers::invalid_unsigned_int)
2144 {
2145 partition_size.back()++;
2146 cell_partition[neighbor->column()] = partition;
2147 neighbor_list.push_back(neighbor->column());
2148 partition_list[counter++] = neighbor->column();
2149 remainder--;
2150 if (remainder == 0)
2151 break;
2152 }
2153 }
2154 }
2155
2156 while (neighbor_list.size() > 0)
2157 {
2158 partition++;
2159
2160 // counter for number of cells so far in current partition
2161 unsigned int partition_counter = 0;
2162
2163 // Mark the start of the new partition
2164 partition_size.push_back(partition_size.back());
2165
2166 // Loop through the list of cells in previous partition and put
2167 // all their neighbors in current partition
2168 for (const unsigned int cell : neighbor_list)
2169 {
2170 Assert(cell_partition[cell] == partition - 1,
2171 ExcInternalError());
2172 auto neighbor = connectivity.begin(cell);
2173 const auto end = connectivity.end(cell);
2174 for (; neighbor != end; ++neighbor)
2175 {
2176 if (cell_partition[neighbor->column()] ==
2177 numbers::invalid_unsigned_int)
2178 {
2179 partition_size.back()++;
2180 cell_partition[neighbor->column()] = partition;
2181
2182 // collect the cells of the current partition for
2183 // use as neighbors in next partition
2184 neighbor_neighbor_list.push_back(neighbor->column());
2185 partition_list[counter++] = neighbor->column();
2186 partition_counter++;
2187 }
2188 }
2189 }
2190 remainder = cluster_size - (partition_counter % cluster_size);
2191 if (remainder == cluster_size)
2192 remainder = 0;
2193 int index_stop = 0;
2194 int index_before = neighbor_neighbor_list.size(),
2195 index = index_before;
2196 while (remainder > 0)
2197 {
2198 if (index == index_stop)
2199 {
2200 index = neighbor_neighbor_list.size();
2201 if (index == index_before)
2202 {
2203 neighbor_neighbor_list.resize(0);
2204 break;
2205 }
2206 index_stop = index_before;
2207 index_before = index;
2208 }
2209 index--;
2210 unsigned int additional = neighbor_neighbor_list[index];
2211 DynamicSparsityPattern::iterator neighbor =
2212 connectivity.begin(
2213 additional),
2214 end = connectivity.end(
2215 additional);
2216 for (; neighbor != end; ++neighbor)
2217 {
2218 if (cell_partition[neighbor->column()] ==
2219 numbers::invalid_unsigned_int)
2220 {
2221 partition_size.back()++;
2222 cell_partition[neighbor->column()] = partition;
2223 neighbor_neighbor_list.push_back(neighbor->column());
2224 partition_list[counter++] = neighbor->column();
2225 remainder--;
2226 if (remainder == 0)
2227 break;
2228 }
2229 }
2230 }
2231
2232 neighbor_list = neighbor_neighbor_list;
2233 neighbor_neighbor_list.resize(0);
2234 }
2235 not_connect:
2236 // One has to check if the graph is not connected so we have to find
2237 // another partition.
2238 work = false;
2239 for (unsigned int j = start_up; j < n_blocks; ++j)
2240 if (cell_partition[j] == numbers::invalid_unsigned_int)
2241 {
2242 start_up = j;
2243 work = true;
2244 if (remainder == 0)
2245 remainder = cluster_size;
2246 break;
2247 }
2248 }
2249 if (remainder != 0)
2250 partition++;
2251
2252 AssertDimension(partition_size[partition], n_blocks);
2253 }
2254
2255
2256 void
2257 TaskInfo::update_task_info(const unsigned int partition)
2258 {
2259 evens = (partition + 1) / 2;
2260 odds = partition / 2;
2261 n_blocked_workers = odds - (odds + evens + 1) % 2;
2262 n_workers = evens + odds - n_blocked_workers;
2263 // From here only used for partition partition option.
2264 partition_evens.resize(partition);
2265 partition_odds.resize(partition);
2266 partition_n_blocked_workers.resize(partition);
2267 partition_n_workers.resize(partition);
2268 for (unsigned int part = 0; part < partition; ++part)
2269 {
2270 partition_evens[part] =
2271 (partition_row_index[part + 1] - partition_row_index[part] + 1) / 2;
2272 partition_odds[part] =
2273 (partition_row_index[part + 1] - partition_row_index[part]) / 2;
2274 partition_n_blocked_workers[part] =
2275 partition_odds[part] -
2276 (partition_odds[part] + partition_evens[part] + 1) % 2;
2277 partition_n_workers[part] = partition_evens[part] +
2278 partition_odds[part] -
2279 partition_n_blocked_workers[part];
2280 }
2281 }
2282 } // namespace MatrixFreeFunctions
2283} // namespace internal
2284
2285
2286
2287// explicit instantiations of template functions
2288template void
2289internal::MatrixFreeFunctions::TaskInfo::print_memory_statistics<std::ostream>(
2290 std::ostream &,
2291 const std::size_t) const;
2292template void
2293internal::MatrixFreeFunctions::TaskInfo::print_memory_statistics<
2294 ConditionalOStream>(ConditionalOStream &, const std::size_t) const;
2295
2296
2297DEAL_II_NAMESPACE_CLOSE
MultithreadInfo::n_threads
static unsigned int n_threads()
Definition: multithread_info.cc:130
conditional_ostream.h
DEAL_II_NAMESPACE_OPEN
#define DEAL_II_NAMESPACE_OPEN
Definition: config.h:442
DEAL_II_NAMESPACE_CLOSE
#define DEAL_II_NAMESPACE_CLOSE
Definition: config.h:443
dynamic_sparsity_pattern.h
first
Point< 2 > first
Definition: grid_out.cc:4603
StandardExceptions::ExcNotImplemented
static ::ExceptionBase & ExcNotImplemented()
Assert
#define Assert(cond, exc)
Definition: exceptions.h:1473
AssertDimension
#define AssertDimension(dim1, dim2)
Definition: exceptions.h:1667
AssertIndexRange
#define AssertIndexRange(index, range)
Definition: exceptions.h:1732
StandardExceptions::ExcInternalError
static ::ExceptionBase & ExcInternalError()
AssertThrow
#define AssertThrow(cond, exc)
Definition: exceptions.h:1583
DynamicSparsityPatternIterators::Accessor::column
size_type column() const
Definition: dynamic_sparsity_pattern.h:809
DynamicSparsityPattern::begin
iterator begin() const
Definition: dynamic_sparsity_pattern.h:1104
DynamicSparsityPattern::row_length
size_type row_length(const size_type row) const
Definition: dynamic_sparsity_pattern.h:1071
DynamicSparsityPattern::end
iterator end() const
Definition: dynamic_sparsity_pattern.h:1114
DynamicSparsityPattern::add
void add(const size_type i, const size_type j)
Definition: dynamic_sparsity_pattern.h:1031
memory_consumption.h
multithread_info.h
MemoryConsumption::memory_consumption
std::enable_if< std::is_fundamental< T >::value, std::size_t >::type memory_consumption(const T &t)
Definition: memory_consumption.h:266
SparsityTools::partition
void partition(const SparsityPattern &sparsity_pattern, const unsigned int n_partitions, std::vector< unsigned int > &partition_indices, const Partitioner partitioner=Partitioner::metis)
Definition: sparsity_tools.cc:399
Utilities::MPI::min_max_avg
MinMaxAvg min_max_avg(const double my_value, const MPI_Comm &mpi_communicator)
Definition: mpi.cc:114
Utilities::invert_permutation
std::vector< Integer > invert_permutation(const std::vector< Integer > &permutation)
Definition: utilities.h:1813
numbers::invalid_unsigned_int
static const unsigned int invalid_unsigned_int
Definition: types.h:201
parallel::internal::parallel_for
void parallel_for(Iterator x_begin, Iterator x_end, const Functor &functor, const unsigned int grainsize)
Definition: parallel.h:80
std::min
::VectorizedArray< Number, width > min(const ::VectorizedArray< Number, width > &, const ::VectorizedArray< Number, width > &)
Definition: vectorization.h:5605
std::max
::VectorizedArray< Number, width > max(const ::VectorizedArray< Number, width > &, const ::VectorizedArray< Number, width > &)
Definition: vectorization.h:5588
internal::MFWorkerInterface::face
virtual void face(const unsigned int range_index)=0
internal::MFWorkerInterface::zero_dst_vector_range
virtual void zero_dst_vector_range(const unsigned int range_index)=0
internal::MFWorkerInterface::cell
virtual void cell(const std::pair< unsigned int, unsigned int > &cell_range)=0
internal::MFWorkerInterface::boundary
virtual void boundary(const unsigned int range_index)=0
internal::MFWorkerInterface::vector_compress_start
virtual void vector_compress_start()=0
Starts the communication for the vector compress operation.
internal::MFWorkerInterface::cell_loop_post_range
virtual void cell_loop_post_range(const unsigned int range_index)=0
internal::MFWorkerInterface::vector_update_ghosts_start
virtual void vector_update_ghosts_start()=0
Starts the communication for the update ghost values operation.
internal::MFWorkerInterface::cell_loop_pre_range
virtual void cell_loop_pre_range(const unsigned int range_index)=0
internal::MFWorkerInterface::vector_update_ghosts_finish
virtual void vector_update_ghosts_finish()=0
Finishes the communication for the update ghost values operation.
internal::MFWorkerInterface::vector_compress_finish
virtual void vector_compress_finish()=0
Finishes the communication for the vector compress operation.
internal::MatrixFreeFunctions::TaskInfo::memory_consumption
std::size_t memory_consumption() const
Definition: task_info.cc:705
internal::MatrixFreeFunctions::TaskInfo::boundary_partition_data
std::vector< unsigned int > boundary_partition_data
Definition: task_info.h:521
internal::MatrixFreeFunctions::TaskInfo::loop
void loop(MFWorkerInterface &worker) const
Definition: task_info.cc:350
internal::MatrixFreeFunctions::TaskInfo::make_thread_graph_partition_color
void make_thread_graph_partition_color(DynamicSparsityPattern &connectivity, std::vector< unsigned int > &renumbering, std::vector< unsigned char > &irregular_cells, const bool hp_bool)
Definition: task_info.cc:1177
internal::MatrixFreeFunctions::TaskInfo::partition_n_workers
std::vector< unsigned int > partition_n_workers
Definition: task_info.h:579
internal::MatrixFreeFunctions::TaskInfo::create_blocks_serial
void create_blocks_serial(const std::vector< unsigned int > &cells_with_comm, const unsigned int dofs_per_cell, const bool categories_are_hp, const std::vector< unsigned int > &cell_vectorization_categories, const bool cell_vectorization_categories_strict, const std::vector< unsigned int > &parent_relation, std::vector< unsigned int > &renumbering, std::vector< unsigned char > &incompletely_filled_vectorization)
Definition: task_info.cc:794
internal::MatrixFreeFunctions::TaskInfo::make_connectivity_cells_to_blocks
void make_connectivity_cells_to_blocks(const std::vector< unsigned char > &irregular_cells, const DynamicSparsityPattern &connectivity_cells, DynamicSparsityPattern &connectivity_blocks) const
Definition: task_info.cc:1587
internal::MatrixFreeFunctions::TaskInfo::make_partitioning_within_partitions_post_blocked
void make_partitioning_within_partitions_post_blocked(const DynamicSparsityPattern &connectivity, const std::vector< unsigned int > &cell_active_fe_index, const unsigned int partition, const unsigned int cluster_size, const bool hp_bool, const std::vector< unsigned int > &cell_partition, const std::vector< unsigned int > &partition_list, const std::vector< unsigned int > &partition_size, std::vector< unsigned int > &partition_partition_list, std::vector< unsigned char > &irregular_cells)
Definition: task_info.cc:1631
internal::MatrixFreeFunctions::TaskInfo::initial_setup_blocks_tasks
void initial_setup_blocks_tasks(const std::vector< unsigned int > &boundary_cells, std::vector< unsigned int > &renumbering, std::vector< unsigned char > &incompletely_filled_vectorization)
Definition: task_info.cc:1083
internal::MatrixFreeFunctions::TaskInfo::print_memory_statistics
void print_memory_statistics(StreamType &out, std::size_t data_length) const
Definition: task_info.cc:690
internal::MatrixFreeFunctions::TaskInfo::make_coloring_within_partitions_pre_blocked
void make_coloring_within_partitions_pre_blocked(const DynamicSparsityPattern &connectivity, const unsigned int partition, const std::vector< unsigned int > &cell_partition, const std::vector< unsigned int > &partition_list, const std::vector< unsigned int > &partition_size, std::vector< unsigned int > &partition_color_list)
Definition: task_info.cc:1952
internal::MatrixFreeFunctions::TaskInfo::partition_row_index
std::vector< unsigned int > partition_row_index
Definition: task_info.h:469
internal::MatrixFreeFunctions::TaskInfo::make_thread_graph_partition_partition
void make_thread_graph_partition_partition(const std::vector< unsigned int > &cell_active_fe_index, DynamicSparsityPattern &connectivity, std::vector< unsigned int > &renumbering, std::vector< unsigned char > &irregular_cells, const bool hp_bool)
Definition: task_info.cc:1519
internal::MatrixFreeFunctions::TaskInfo::partition_evens
std::vector< unsigned int > partition_evens
Definition: task_info.h:561
internal::MatrixFreeFunctions::TaskInfo::make_thread_graph
void make_thread_graph(const std::vector< unsigned int > &cell_active_fe_index, DynamicSparsityPattern &connectivity, std::vector< unsigned int > &renumbering, std::vector< unsigned char > &irregular_cells, const bool hp_bool)
Definition: task_info.cc:1324
internal::MatrixFreeFunctions::TaskInfo::partition_n_blocked_workers
std::vector< unsigned int > partition_n_blocked_workers
Definition: task_info.h:573
internal::MatrixFreeFunctions::TaskInfo::cell_partition_data
std::vector< unsigned int > cell_partition_data
Definition: task_info.h:477
internal::MatrixFreeFunctions::TaskInfo::make_partitioning
void make_partitioning(const DynamicSparsityPattern &connectivity, const unsigned int cluster_size, std::vector< unsigned int > &cell_partition, std::vector< unsigned int > &partition_list, std::vector< unsigned int > &partition_size, unsigned int &partition) const
Definition: task_info.cc:2028
internal::MatrixFreeFunctions::TaskInfo::update_task_info
void update_task_info(const unsigned int partition)
Definition: task_info.cc:2257
internal::MatrixFreeFunctions::TaskInfo::make_boundary_cells_divisible
void make_boundary_cells_divisible(std::vector< unsigned int > &boundary_cells)
Definition: task_info.cc:722
internal::MatrixFreeFunctions::TaskInfo::guess_block_size
void guess_block_size(const unsigned int dofs_per_cell)
Definition: task_info.cc:1149
internal::MatrixFreeFunctions::TaskInfo::partition_odds
std::vector< unsigned int > partition_odds
Definition: task_info.h:567
internal::MatrixFreeFunctions::TaskInfo::face_partition_data
std::vector< unsigned int > face_partition_data
Definition: task_info.h:499
task_info.h
Generated by doxygen 1.9.5