Dune Core Modules (2.4.1)

torus.hh
Go to the documentation of this file.
1// -*- tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*-
2// vi: set et ts=4 sw=2 sts=2:
3#ifndef DUNE_GRID_YASPGRID_TORUS_HH
4#define DUNE_GRID_YASPGRID_TORUS_HH
5
6#include <bitset>
7#include <cmath>
8#include <deque>
9#include <vector>
10
11#if HAVE_MPI
12#include <mpi.h>
13#endif
14
15#include <dune/common/array.hh>
17#include <dune/grid/common/exceptions.hh>
18
19#include "partitioning.hh"
20
25namespace Dune
26{
27
41 template<class CollectiveCommunication, int d>
42 class Torus {
43 public:
45 typedef Dune::array<int, d> iTupel;
46
47
48 private:
49 struct CommPartner {
50 int rank;
51 iTupel delta;
52 int index;
53 };
54
55 struct CommTask {
56 int rank; // process to send to / receive from
57 void *buffer; // buffer to send / receive
58 int size; // size of buffer
59#if HAVE_MPI
60 MPI_Request request; // used by MPI to handle request
61#else
62 int request;
63#endif
64 int flag; // used by MPI
65 };
66
67 public:
70 {}
71
74 : _comm(comm), _tag(tag)
75 {
76 // determine dimensions
77 lb->loadbalance(size, _comm.size(), _dims);
78
79 // compute increments for lexicographic ordering
80 int inc = 1;
81 for (int i=0; i<d; i++)
82 {
83 _increment[i] = inc;
84 inc *= _dims[i];
85 }
86
87 // check whether the load balancing matches the size of the communicator
88 if (inc != _comm.size())
89 DUNE_THROW(Dune::Exception, "Communicator size and result of the given load balancer do not match!");
90
91 // make full schedule
92 proclists();
93 }
94
96 int rank () const
97 {
98 return _comm.rank();
99 }
100
102 iTupel coord () const
103 {
104 return rank_to_coord(_comm.rank());
105 }
106
108 int procs () const
109 {
110 return _comm.size();
111 }
112
114 const iTupel & dims () const
115 {
116 return _dims;
117 }
118
120 int dims (int i) const
121 {
122 return _dims[i];
123 }
124
127 {
128 return _comm;
129 }
130
132 int tag () const
133 {
134 return _tag;
135 }
136
138 bool inside (iTupel c) const
139 {
140 for (int i=d-1; i>=0; i--)
141 if (c[i]<0 || c[i]>=_dims[i]) return false;
142 return true;
143 }
144
147 {
149 rank = rank%_comm.size();
150 for (int i=d-1; i>=0; i--)
151 {
152 coord[i] = rank/_increment[i];
153 rank = rank%_increment[i];
154 }
155 return coord;
156 }
157
160 {
161 for (int i=0; i<d; i++) coord[i] = coord[i]%_dims[i];
162 int rank = 0;
163 for (int i=0; i<d; i++) rank += coord[i]*_increment[i];
164 return rank;
165 }
166
168 int rank_relative (int rank, int dir, int cnt) const
169 {
171 coord[dir] = (coord[dir]+_dims[dir]+cnt)%_dims[dir];
172 return coord_to_rank(coord);
173 }
174
176 int color (const iTupel & coord) const
177 {
178 int c = 0;
179 int power = 1;
180
181 // interior coloring
182 for (int i=0; i<d; i++)
183 {
184 if (coord[i]%2==1) c += power;
185 power *= 2;
186 }
187
188 // extra colors for boundary processes
189 for (int i=0; i<d; i++)
190 {
191 if (_dims[i]>1 && coord[i]==_dims[i]-1) c += power;
192 power *= 2;
193 }
194
195 return c;
196 }
197
199 int color (int rank) const
200 {
201 return color(rank_to_coord(rank));
202 }
203
205 int neighbors () const
206 {
207 int n=1;
208 for (int i=0; i<d; ++i)
209 n *= 3;
210 return n-1;
211 }
212
214 bool is_neighbor (iTupel delta, std::bitset<d> periodic) const
215 {
216 iTupel coord = rank_to_coord(_comm.rank()); // my own coordinate with 0 <= c_i < dims_i
217
218 for (int i=0; i<d; i++)
219 {
220 if (delta[i]<0)
221 {
222 // if I am on the boundary and domain is not periodic => no neighbor
223 if (coord[i]==0 && periodic[i]==false) return false;
224 }
225 if (delta[i]>0)
226 {
227 // if I am on the boundary and domain is not periodic => no neighbor
228 if (coord[i]==_dims[i]-1 && periodic[i]==false) return false;
229 }
230 }
231 return true;
232 }
233
241 double partition (int rank, iTupel origin_in, iTupel size_in, iTupel& origin_out, iTupel& size_out) const
242 {
244 double maxsize = 1;
245 double sz = 1;
246
247 // make a tensor product partition
248 for (int i=0; i<d; i++)
249 {
250 // determine
251 int m = size_in[i]/_dims[i];
252 int r = size_in[i]%_dims[i];
253
254 sz *= size_in[i];
255
256 if (coord[i]<_dims[i]-r)
257 {
258 origin_out[i] = origin_in[i] + coord[i]*m;
259 size_out[i] = m;
260 maxsize *= m;
261 }
262 else
263 {
264 origin_out[i] = origin_in[i] + (_dims[i]-r)*m + (coord[i]-(_dims[i]-r))*(m+1);
265 size_out[i] = m+1;
266 maxsize *= m+1;
267 }
268 }
269 return maxsize/(sz/_comm.size());
270 }
271
279 public:
281 ProcListIterator (typename std::deque<CommPartner>::const_iterator iter)
282 {
283 i = iter;
284 }
285
287 int rank () const
288 {
289 return i->rank;
290 }
291
293 iTupel delta () const
294 {
295 return i->delta;
296 }
297
299 int index () const
300 {
301 return i->index;
302 }
303
305 int distance () const
306 {
307 int dist = 0;
308 iTupel delta=i->delta;
309 for (int j=0; j<d; ++j)
310 dist += std::abs(delta[j]);
311 return dist;
312 }
313
316 {
317 return i == iter.i;
318 }
319
320
323 {
324 return i != iter.i;
325 }
326
329 {
330 ++i;
331 return *this;
332 }
333
334 private:
335 typename std::deque<CommPartner>::const_iterator i;
336 };
337
340 {
341 return ProcListIterator(_sendlist.begin());
342 }
343
346 {
347 return ProcListIterator(_sendlist.end());
348 }
349
352 {
353 return ProcListIterator(_recvlist.begin());
354 }
355
358 {
359 return ProcListIterator(_recvlist.end());
360 }
361
363 void send (int rank, void* buffer, int size) const
364 {
365 CommTask task;
366 task.rank = rank;
367 task.buffer = buffer;
368 task.size = size;
369 if (rank!=_comm.rank())
370 _sendrequests.push_back(task);
371 else
372 _localsendrequests.push_back(task);
373 }
374
376 void recv (int rank, void* buffer, int size) const
377 {
378 CommTask task;
379 task.rank = rank;
380 task.buffer = buffer;
381 task.size = size;
382 if (rank!=_comm.rank())
383 _recvrequests.push_back(task);
384 else
385 _localrecvrequests.push_back(task);
386 }
387
389 void exchange () const
390 {
391 // handle local requests first
392 if (_localsendrequests.size()!=_localrecvrequests.size())
393 {
394 std::cout << "[" << rank() << "]: ERROR: local sends/receives do not match in exchange!" << std::endl;
395 return;
396 }
397 for (unsigned int i=0; i<_localsendrequests.size(); i++)
398 {
399 if (_localsendrequests[i].size!=_localrecvrequests[i].size)
400 {
401 std::cout << "[" << rank() << "]: ERROR: size in local sends/receive does not match in exchange!" << std::endl;
402 return;
403 }
404 memcpy(_localrecvrequests[i].buffer,_localsendrequests[i].buffer,_localsendrequests[i].size);
405 }
406 _localsendrequests.clear();
407 _localrecvrequests.clear();
408
409#if HAVE_MPI
410 // handle foreign requests
411 int sends=0;
412 int recvs=0;
413
414 // issue sends to foreign processes
415 for (unsigned int i=0; i<_sendrequests.size(); i++)
416 if (_sendrequests[i].rank!=rank())
417 {
418 // std::cout << "[" << rank() << "]" << " send " << _sendrequests[i].size << " bytes "
419 // << "to " << _sendrequests[i].rank << " p=" << _sendrequests[i].buffer << std::endl;
420 MPI_Isend(_sendrequests[i].buffer, _sendrequests[i].size, MPI_BYTE,
421 _sendrequests[i].rank, _tag, _comm, &(_sendrequests[i].request));
422 _sendrequests[i].flag = false;
423 sends++;
424 }
425
426 // issue receives from foreign processes
427 for (unsigned int i=0; i<_recvrequests.size(); i++)
428 if (_recvrequests[i].rank!=rank())
429 {
430 // std::cout << "[" << rank() << "]" << " recv " << _recvrequests[i].size << " bytes "
431 // << "fm " << _recvrequests[i].rank << " p=" << _recvrequests[i].buffer << std::endl;
432 MPI_Irecv(_recvrequests[i].buffer, _recvrequests[i].size, MPI_BYTE,
433 _recvrequests[i].rank, _tag, _comm, &(_recvrequests[i].request));
434 _recvrequests[i].flag = false;
435 recvs++;
436 }
437
438 // poll sends
439 while (sends>0)
440 {
441 for (unsigned int i=0; i<_sendrequests.size(); i++)
442 if (!_sendrequests[i].flag)
443 {
444 MPI_Status status;
445 MPI_Test( &(_sendrequests[i].request), &(_sendrequests[i].flag), &status);
446 if (_sendrequests[i].flag)
447 {
448 sends--;
449 // std::cout << "[" << rank() << "]" << " send to " << _sendrequests[i].rank << " OK" << std::endl;
450 }
451 }
452 }
453
454 // poll receives
455 while (recvs>0)
456 {
457 for (unsigned int i=0; i<_recvrequests.size(); i++)
458 if (!_recvrequests[i].flag)
459 {
460 MPI_Status status;
461 MPI_Test( &(_recvrequests[i].request), &(_recvrequests[i].flag), &status);
462 if (_recvrequests[i].flag)
463 {
464 recvs--;
465 // std::cout << "[" << rank() << "]" << " recv fm " << _recvrequests[i].rank << " OK" << std::endl;
466 }
467
468 }
469 }
470
471 // clear request buffers
472 _sendrequests.clear();
473 _recvrequests.clear();
474#endif
475 }
476
478 double global_max (double x) const
479 {
480 double res = 0.0;
481 _comm.template allreduce<Dune::Max<double>,double>(&x, &res, 1);
482 return res;
483 }
484
486 void print (std::ostream& s) const
487 {
488 s << "[" << rank() << "]: Torus " << procs() << " processor(s) arranged as " << dims() << std::endl;
489 for (ProcListIterator i=sendbegin(); i!=sendend(); ++i)
490 {
491 s << "[" << rank() << "]: send to "
492 << "rank=" << i.rank()
493 << " index=" << i.index()
494 << " delta=" << i.delta() << " dist=" << i.distance() << std::endl;
495 }
496 for (ProcListIterator i=recvbegin(); i!=recvend(); ++i)
497 {
498 s << "[" << rank() << "]: recv from "
499 << "rank=" << i.rank()
500 << " index=" << i.index()
501 << " delta=" << i.delta() << " dist=" << i.distance() << std::endl;
502 }
503 }
504
505 private:
506
507 void proclists ()
508 {
509 // compile the full neighbor list
510 CommPartner cp;
511 iTupel delta;
512
513 std::fill(delta.begin(), delta.end(), -1);
514 bool ready = false;
515 iTupel me, nb;
516 me = rank_to_coord(_comm.rank());
517 int index = 0;
518 int last = neighbors()-1;
519 while (!ready)
520 {
521 // find neighbors coordinates
522 for (int i=0; i<d; i++)
523 nb[i] = ( me[i]+_dims[i]+delta[i] ) % _dims[i];
524
525 // find neighbors rank
526 int nbrank = coord_to_rank(nb);
527
528 // check if delta is not zero
529 for (int i=0; i<d; i++)
530 if (delta[i]!=0)
531 {
532 cp.rank = nbrank;
533 cp.delta = delta;
534 cp.index = index;
535 _recvlist.push_back(cp);
536 cp.index = last-index;
537 _sendlist.push_front(cp);
538 index++;
539 break;
540 }
541
542 // next neighbor
543 ready = true;
544 for (int i=0; i<d; i++)
545 if (delta[i]<1)
546 {
547 (delta[i])++;
548 ready=false;
549 break;
550 }
551 else
552 {
553 delta[i] = -1;
554 }
555 }
556
557 }
558
559 CollectiveCommunication _comm;
560
561 iTupel _dims;
562 iTupel _increment;
563 int _tag;
564 std::deque<CommPartner> _sendlist;
565 std::deque<CommPartner> _recvlist;
566
567 mutable std::vector<CommTask> _sendrequests;
568 mutable std::vector<CommTask> _recvrequests;
569 mutable std::vector<CommTask> _localsendrequests;
570 mutable std::vector<CommTask> _localrecvrequests;
571
572 };
573
575 template <class CollectiveCommunication, int d>
576 inline std::ostream& operator<< (std::ostream& s, const Torus<CollectiveCommunication, d> & t)
577 {
578 t.print(s);
579 return s;
580 }
581}
582
583#endif
Fallback implementation of the std::array class (a static array)
Various helper classes derived from from std::binary_function for stl-style functional programming.
Collective communication interface and sequential default implementation.
Definition: collectivecommunication.hh:73
int rank() const
Return rank, is between 0 and size()-1.
Definition: collectivecommunication.hh:82
int size() const
Number of processes in set, is greater than 0.
Definition: collectivecommunication.hh:88
Base class for Dune-Exceptions.
Definition: exceptions.hh:91
Definition: torus.hh:278
ProcListIterator(typename std::deque< CommPartner >::const_iterator iter)
make an iterator
Definition: torus.hh:281
int distance() const
return 1-norm of distance vector
Definition: torus.hh:305
int index() const
return index in proclist
Definition: torus.hh:299
bool operator!=(const ProcListIterator &iter)
Return true when two iterators do not point to same member.
Definition: torus.hh:322
int rank() const
return rank of neighboring process
Definition: torus.hh:287
bool operator==(const ProcListIterator &iter)
Return true when two iterators point to same member.
Definition: torus.hh:315
iTupel delta() const
return distance vector
Definition: torus.hh:293
ProcListIterator & operator++()
Increment iterator to next cell.
Definition: torus.hh:328
Definition: torus.hh:42
Torus()
constructor making uninitialized object
Definition: torus.hh:69
int neighbors() const
return the number of neighbors, which is
Definition: torus.hh:205
iTupel rank_to_coord(int rank) const
map rank to coordinate in torus using lexicographic ordering
Definition: torus.hh:146
ProcListIterator sendbegin() const
first process in send list
Definition: torus.hh:339
int color(const iTupel &coord) const
assign color to given coordinate
Definition: torus.hh:176
int rank() const
return own rank
Definition: torus.hh:96
ProcListIterator recvbegin() const
first process in receive list
Definition: torus.hh:351
int rank_relative(int rank, int dir, int cnt) const
return rank of process where its coordinate in direction dir has offset cnt (handles periodic case)
Definition: torus.hh:168
void recv(int rank, void *buffer, int size) const
store a receive request; buffers are received in order; handles also local requests with memcpy
Definition: torus.hh:376
int coord_to_rank(iTupel coord) const
map coordinate in torus to rank using lexicographic ordering
Definition: torus.hh:159
int dims(int i) const
return dimensions of torus in direction i
Definition: torus.hh:120
int procs() const
return number of processes
Definition: torus.hh:108
Torus(CollectiveCommunication comm, int tag, iTupel size, const YLoadBalance< d > *lb)
make partitioner from communicator and coarse mesh size
Definition: torus.hh:73
ProcListIterator recvend() const
last process in receive list
Definition: torus.hh:357
double global_max(double x) const
global max
Definition: torus.hh:478
Dune::array< int, d > iTupel
type used to pass tupels in and out
Definition: torus.hh:45
bool is_neighbor(iTupel delta, std::bitset< d > periodic) const
return true if neighbor with given delta is a neighbor under the given periodicity
Definition: torus.hh:214
int color(int rank) const
assign color to given rank
Definition: torus.hh:199
CollectiveCommunication comm() const
return communicator
Definition: torus.hh:126
void send(int rank, void *buffer, int size) const
store a send request; buffers are sent in order; handles also local requests with memcpy
Definition: torus.hh:363
void print(std::ostream &s) const
print contents of torus object
Definition: torus.hh:486
double partition(int rank, iTupel origin_in, iTupel size_in, iTupel &origin_out, iTupel &size_out) const
partition the given grid onto the torus and return the piece of the process with given rank; returns ...
Definition: torus.hh:241
iTupel coord() const
return own coordinates
Definition: torus.hh:102
bool inside(iTupel c) const
return true if coordinate is inside torus
Definition: torus.hh:138
int tag() const
return tag used by torus
Definition: torus.hh:132
void exchange() const
exchange messages stored in request buffers; clear request buffers afterwards
Definition: torus.hh:389
ProcListIterator sendend() const
end of send list
Definition: torus.hh:345
const iTupel & dims() const
return dimensions of torus
Definition: torus.hh:114
a base class for the yaspgrid partitioning strategy The name might be irritating. It will probably ch...
Definition: partitioning.hh:24
std::ostream & operator<<(std::ostream &s, const array< T, N > &e)
Output operator for array.
Definition: array.hh:26
#define DUNE_THROW(E, m)
Definition: exceptions.hh:243
Dune namespace.
Definition: alignment.hh:10
This file provides tools to partition YaspGrids. If you want to write your own partitioner,...
Creative Commons License   |  Legal Statements / Impressum  |  Hosted by TU Dresden  |  generated with Hugo v0.111.3 (Nov 21, 23:30, 2024)