/usr/include/dune/grid/yaspgrid/torus.hh is in libdune-grid-dev 2.4.1-1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 | // -*- tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*-
// vi: set et ts=4 sw=2 sts=2:
#ifndef DUNE_GRID_YASPGRID_TORUS_HH
#define DUNE_GRID_YASPGRID_TORUS_HH
#include <bitset>
#include <cmath>
#include <deque>
#include <vector>
#if HAVE_MPI
#include <mpi.h>
#endif
#include <dune/common/array.hh>
#include <dune/common/binaryfunctions.hh>
#include <dune/grid/common/exceptions.hh>
#include "partitioning.hh"
/** \file
* \brief This file provides the infrastructure for toroidal communication in YaspGrid.
*/
namespace Dune
{
/*! Torus provides all the functionality to handle a toroidal communication structure:
- Map a set of processes (given by an MPI communicator) to a torus of dimension d. The "optimal"
torus dimensions are determined by a coarse mesh. The maximum side length is minimized.
- Provide lists of neighboring processes and a method for nearest neighbor exchange
using asynchronous communication with MPI. The periodic case is handled where one process
might have to exchange several messages with the same process. (Logically, a process has always
\f$3^d-1\f$ neighbors, but several of these logical neighbors might be identical)
- Provide means to partition a grid to the torus.
*/
template<class CollectiveCommunication, int d>
class Torus {
public:
//! type used to pass tupels in and out
typedef Dune::array<int, d> iTupel;
private:
struct CommPartner {
int rank;
iTupel delta;
int index;
};
struct CommTask {
int rank; // process to send to / receive from
void *buffer; // buffer to send / receive
int size; // size of buffer
#if HAVE_MPI
MPI_Request request; // used by MPI to handle request
#else
int request;
#endif
int flag; // used by MPI
};
public:
//! constructor making uninitialized object
Torus ()
{}
//! make partitioner from communicator and coarse mesh size
Torus (CollectiveCommunication comm, int tag, iTupel size, const YLoadBalance<d>* lb)
: _comm(comm), _tag(tag)
{
// determine dimensions
lb->loadbalance(size, _comm.size(), _dims);
// compute increments for lexicographic ordering
int inc = 1;
for (int i=0; i<d; i++)
{
_increment[i] = inc;
inc *= _dims[i];
}
// check whether the load balancing matches the size of the communicator
if (inc != _comm.size())
DUNE_THROW(Dune::Exception, "Communicator size and result of the given load balancer do not match!");
// make full schedule
proclists();
}
//! return own rank
int rank () const
{
return _comm.rank();
}
//! return own coordinates
iTupel coord () const
{
return rank_to_coord(_comm.rank());
}
//! return number of processes
int procs () const
{
return _comm.size();
}
//! return dimensions of torus
const iTupel & dims () const
{
return _dims;
}
//! return dimensions of torus in direction i
int dims (int i) const
{
return _dims[i];
}
//! return communicator
CollectiveCommunication comm () const
{
return _comm;
}
//! return tag used by torus
int tag () const
{
return _tag;
}
//! return true if coordinate is inside torus
bool inside (iTupel c) const
{
for (int i=d-1; i>=0; i--)
if (c[i]<0 || c[i]>=_dims[i]) return false;
return true;
}
//! map rank to coordinate in torus using lexicographic ordering
iTupel rank_to_coord (int rank) const
{
iTupel coord;
rank = rank%_comm.size();
for (int i=d-1; i>=0; i--)
{
coord[i] = rank/_increment[i];
rank = rank%_increment[i];
}
return coord;
}
//! map coordinate in torus to rank using lexicographic ordering
int coord_to_rank (iTupel coord) const
{
for (int i=0; i<d; i++) coord[i] = coord[i]%_dims[i];
int rank = 0;
for (int i=0; i<d; i++) rank += coord[i]*_increment[i];
return rank;
}
//! return rank of process where its coordinate in direction dir has offset cnt (handles periodic case)
int rank_relative (int rank, int dir, int cnt) const
{
iTupel coord = rank_to_coord(rank);
coord[dir] = (coord[dir]+_dims[dir]+cnt)%_dims[dir];
return coord_to_rank(coord);
}
//! assign color to given coordinate
int color (const iTupel & coord) const
{
int c = 0;
int power = 1;
// interior coloring
for (int i=0; i<d; i++)
{
if (coord[i]%2==1) c += power;
power *= 2;
}
// extra colors for boundary processes
for (int i=0; i<d; i++)
{
if (_dims[i]>1 && coord[i]==_dims[i]-1) c += power;
power *= 2;
}
return c;
}
//! assign color to given rank
int color (int rank) const
{
return color(rank_to_coord(rank));
}
//! return the number of neighbors, which is \f$3^d-1\f$
int neighbors () const
{
int n=1;
for (int i=0; i<d; ++i)
n *= 3;
return n-1;
}
//! return true if neighbor with given delta is a neighbor under the given periodicity
bool is_neighbor (iTupel delta, std::bitset<d> periodic) const
{
iTupel coord = rank_to_coord(_comm.rank()); // my own coordinate with 0 <= c_i < dims_i
for (int i=0; i<d; i++)
{
if (delta[i]<0)
{
// if I am on the boundary and domain is not periodic => no neighbor
if (coord[i]==0 && periodic[i]==false) return false;
}
if (delta[i]>0)
{
// if I am on the boundary and domain is not periodic => no neighbor
if (coord[i]==_dims[i]-1 && periodic[i]==false) return false;
}
}
return true;
}
/** \brief partition the given grid onto the torus and return the piece of the process with given rank; returns load imbalance
* @param rank rank of our processor
* @param origin_in global origin
* @param size_in global size
* @param origin_out origin of this processors interior
* @param size_out size of this processors interior
*/
double partition (int rank, iTupel origin_in, iTupel size_in, iTupel& origin_out, iTupel& size_out) const
{
iTupel coord = rank_to_coord(rank);
double maxsize = 1;
double sz = 1;
// make a tensor product partition
for (int i=0; i<d; i++)
{
// determine
int m = size_in[i]/_dims[i];
int r = size_in[i]%_dims[i];
sz *= size_in[i];
if (coord[i]<_dims[i]-r)
{
origin_out[i] = origin_in[i] + coord[i]*m;
size_out[i] = m;
maxsize *= m;
}
else
{
origin_out[i] = origin_in[i] + (_dims[i]-r)*m + (coord[i]-(_dims[i]-r))*(m+1);
size_out[i] = m+1;
maxsize *= m+1;
}
}
return maxsize/(sz/_comm.size());
}
/*!
ProcListIterator provides access to a list of neighboring processes. There are always
\f$ 3^d-1 \f$ entries in such a list. Two lists are maintained, one for sending and one for
receiving. The lists are sorted in such a way that in sequence message delivery ensures that
e.g. a message send to the left neighbor is received as a message from the right neighbor.
*/
class ProcListIterator {
public:
//! make an iterator
ProcListIterator (typename std::deque<CommPartner>::const_iterator iter)
{
i = iter;
}
//! return rank of neighboring process
int rank () const
{
return i->rank;
}
//! return distance vector
iTupel delta () const
{
return i->delta;
}
//! return index in proclist
int index () const
{
return i->index;
}
//! return 1-norm of distance vector
int distance () const
{
int dist = 0;
iTupel delta=i->delta;
for (int j=0; j<d; ++j)
dist += std::abs(delta[j]);
return dist;
}
//! Return true when two iterators point to same member
bool operator== (const ProcListIterator& iter)
{
return i == iter.i;
}
//! Return true when two iterators do not point to same member
bool operator!= (const ProcListIterator& iter)
{
return i != iter.i;
}
//! Increment iterator to next cell.
ProcListIterator& operator++ ()
{
++i;
return *this;
}
private:
typename std::deque<CommPartner>::const_iterator i;
};
//! first process in send list
ProcListIterator sendbegin () const
{
return ProcListIterator(_sendlist.begin());
}
//! end of send list
ProcListIterator sendend () const
{
return ProcListIterator(_sendlist.end());
}
//! first process in receive list
ProcListIterator recvbegin () const
{
return ProcListIterator(_recvlist.begin());
}
//! last process in receive list
ProcListIterator recvend () const
{
return ProcListIterator(_recvlist.end());
}
//! store a send request; buffers are sent in order; handles also local requests with memcpy
void send (int rank, void* buffer, int size) const
{
CommTask task;
task.rank = rank;
task.buffer = buffer;
task.size = size;
if (rank!=_comm.rank())
_sendrequests.push_back(task);
else
_localsendrequests.push_back(task);
}
//! store a receive request; buffers are received in order; handles also local requests with memcpy
void recv (int rank, void* buffer, int size) const
{
CommTask task;
task.rank = rank;
task.buffer = buffer;
task.size = size;
if (rank!=_comm.rank())
_recvrequests.push_back(task);
else
_localrecvrequests.push_back(task);
}
//! exchange messages stored in request buffers; clear request buffers afterwards
void exchange () const
{
// handle local requests first
if (_localsendrequests.size()!=_localrecvrequests.size())
{
std::cout << "[" << rank() << "]: ERROR: local sends/receives do not match in exchange!" << std::endl;
return;
}
for (unsigned int i=0; i<_localsendrequests.size(); i++)
{
if (_localsendrequests[i].size!=_localrecvrequests[i].size)
{
std::cout << "[" << rank() << "]: ERROR: size in local sends/receive does not match in exchange!" << std::endl;
return;
}
memcpy(_localrecvrequests[i].buffer,_localsendrequests[i].buffer,_localsendrequests[i].size);
}
_localsendrequests.clear();
_localrecvrequests.clear();
#if HAVE_MPI
// handle foreign requests
int sends=0;
int recvs=0;
// issue sends to foreign processes
for (unsigned int i=0; i<_sendrequests.size(); i++)
if (_sendrequests[i].rank!=rank())
{
// std::cout << "[" << rank() << "]" << " send " << _sendrequests[i].size << " bytes "
// << "to " << _sendrequests[i].rank << " p=" << _sendrequests[i].buffer << std::endl;
MPI_Isend(_sendrequests[i].buffer, _sendrequests[i].size, MPI_BYTE,
_sendrequests[i].rank, _tag, _comm, &(_sendrequests[i].request));
_sendrequests[i].flag = false;
sends++;
}
// issue receives from foreign processes
for (unsigned int i=0; i<_recvrequests.size(); i++)
if (_recvrequests[i].rank!=rank())
{
// std::cout << "[" << rank() << "]" << " recv " << _recvrequests[i].size << " bytes "
// << "fm " << _recvrequests[i].rank << " p=" << _recvrequests[i].buffer << std::endl;
MPI_Irecv(_recvrequests[i].buffer, _recvrequests[i].size, MPI_BYTE,
_recvrequests[i].rank, _tag, _comm, &(_recvrequests[i].request));
_recvrequests[i].flag = false;
recvs++;
}
// poll sends
while (sends>0)
{
for (unsigned int i=0; i<_sendrequests.size(); i++)
if (!_sendrequests[i].flag)
{
MPI_Status status;
MPI_Test( &(_sendrequests[i].request), &(_sendrequests[i].flag), &status);
if (_sendrequests[i].flag)
{
sends--;
// std::cout << "[" << rank() << "]" << " send to " << _sendrequests[i].rank << " OK" << std::endl;
}
}
}
// poll receives
while (recvs>0)
{
for (unsigned int i=0; i<_recvrequests.size(); i++)
if (!_recvrequests[i].flag)
{
MPI_Status status;
MPI_Test( &(_recvrequests[i].request), &(_recvrequests[i].flag), &status);
if (_recvrequests[i].flag)
{
recvs--;
// std::cout << "[" << rank() << "]" << " recv fm " << _recvrequests[i].rank << " OK" << std::endl;
}
}
}
// clear request buffers
_sendrequests.clear();
_recvrequests.clear();
#endif
}
//! global max
double global_max (double x) const
{
double res = 0.0;
_comm.template allreduce<Dune::Max<double>,double>(&x, &res, 1);
return res;
}
//! print contents of torus object
void print (std::ostream& s) const
{
s << "[" << rank() << "]: Torus " << procs() << " processor(s) arranged as " << dims() << std::endl;
for (ProcListIterator i=sendbegin(); i!=sendend(); ++i)
{
s << "[" << rank() << "]: send to "
<< "rank=" << i.rank()
<< " index=" << i.index()
<< " delta=" << i.delta() << " dist=" << i.distance() << std::endl;
}
for (ProcListIterator i=recvbegin(); i!=recvend(); ++i)
{
s << "[" << rank() << "]: recv from "
<< "rank=" << i.rank()
<< " index=" << i.index()
<< " delta=" << i.delta() << " dist=" << i.distance() << std::endl;
}
}
private:
void proclists ()
{
// compile the full neighbor list
CommPartner cp;
iTupel delta;
std::fill(delta.begin(), delta.end(), -1);
bool ready = false;
iTupel me, nb;
me = rank_to_coord(_comm.rank());
int index = 0;
int last = neighbors()-1;
while (!ready)
{
// find neighbors coordinates
for (int i=0; i<d; i++)
nb[i] = ( me[i]+_dims[i]+delta[i] ) % _dims[i];
// find neighbors rank
int nbrank = coord_to_rank(nb);
// check if delta is not zero
for (int i=0; i<d; i++)
if (delta[i]!=0)
{
cp.rank = nbrank;
cp.delta = delta;
cp.index = index;
_recvlist.push_back(cp);
cp.index = last-index;
_sendlist.push_front(cp);
index++;
break;
}
// next neighbor
ready = true;
for (int i=0; i<d; i++)
if (delta[i]<1)
{
(delta[i])++;
ready=false;
break;
}
else
{
delta[i] = -1;
}
}
}
CollectiveCommunication _comm;
iTupel _dims;
iTupel _increment;
int _tag;
std::deque<CommPartner> _sendlist;
std::deque<CommPartner> _recvlist;
mutable std::vector<CommTask> _sendrequests;
mutable std::vector<CommTask> _recvrequests;
mutable std::vector<CommTask> _localsendrequests;
mutable std::vector<CommTask> _localrecvrequests;
};
//! Output operator for Torus
template <class CollectiveCommunication, int d>
inline std::ostream& operator<< (std::ostream& s, const Torus<CollectiveCommunication, d> & t)
{
t.print(s);
return s;
}
}
#endif
|