| 1 | /* |
| 2 | * nn_processor.hpp |
| 3 | * |
| 4 | * Created on: Aug 9, 2015 |
| 5 | * Author: i-bird |
| 6 | */ |
| 7 | |
| 8 | #ifndef SRC_DECOMPOSITION_NN_PROCESSOR_HPP_ |
| 9 | #define SRC_DECOMPOSITION_NN_PROCESSOR_HPP_ |
| 10 | |
| 11 | #include "common.hpp" |
| 12 | #include <unordered_map> |
| 13 | |
| 14 | /*! \brief This class store the adjacent processors and the adjacent sub_domains |
| 15 | * |
| 16 | * \tparam dim is the dimensionality of the physical domain we are going to decompose. |
| 17 | * \tparam T type of the space we decompose, Real, Integer, Complex ... |
| 18 | * |
| 19 | * \see CartDecomposition |
| 20 | * |
| 21 | */ |
| 22 | template<unsigned int dim, typename T, template <typename> class layout_base, typename Memory> |
| 23 | class nn_prcs |
| 24 | { |
| 25 | //! Virtual cluster |
| 26 | Vcluster<> & v_cl; |
| 27 | |
| 28 | //! List of adjacent processors |
| 29 | openfpm::vector<size_t> nn_processors; |
| 30 | |
| 31 | //! for each near processor store the sub-domains of the near processors |
| 32 | std::unordered_map<size_t, N_box<dim,T>> nn_processor_subdomains; |
| 33 | |
| 34 | //! when we add new boxes, are added here |
| 35 | std::unordered_map<size_t, N_box<dim,T>> nn_processor_subdomains_tmp; |
| 36 | |
| 37 | //! contain the same information as the member boxes with the difference that |
| 38 | //! instead of the Box itself, it contain the sub-domain id in the list of the |
| 39 | //! local sub-domains |
| 40 | openfpm::vector<openfpm::vector<size_t>> proc_adj_box; |
| 41 | |
| 42 | //! contain the set of sub-domains sent to the other processors |
| 43 | openfpm::vector< openfpm::vector< ::SpaceBox<dim,T>> > boxes; |
| 44 | |
| 45 | //! Receive counter |
| 46 | size_t recv_cnt; |
| 47 | |
| 48 | //! applyBC function is suppose to be called only one time |
| 49 | bool aBC; |
| 50 | |
| 51 | /*! \brief It shift a box but it does consistently |
| 52 | * |
| 53 | * In calculating internal and external ghost boxes, domains are shifted by periodicity. |
| 54 | * In particular, consider a box touching with the left bolder the left border of the domain |
| 55 | * |
| 56 | |
| 57 | before shift after shift |
| 58 | |
| 59 | +-----------------------------+ +------------------------------+ |
| 60 | | | | | |
| 61 | | domain | | domain | |
| 62 | | | | | |
| 63 | | | | | |
| 64 | +---------+ | | +---------+ |
| 65 | | | | | | | |
| 66 | | | | | | | |
| 67 | | box | | | | box | |
| 68 | | | | | | | |
| 69 | | | | | | | |
| 70 | +---------+ | | +---------+ |
| 71 | | | | | |
| 72 | | | | | |
| 73 | | | | | |
| 74 | | | | | |
| 75 | | | | | |
| 76 | +-----------------------------+ +------------------------------+ |
| 77 | |
| 78 | * |
| 79 | * |
| 80 | * |
| 81 | * |
| 82 | * |
| 83 | * shifting the box on the right by the size of the domain, we expect to have a box touching with |
| 84 | * the left side the right side of the domain. Because of rounding off problem this is not possible |
| 85 | * with a simple shift. This function ensure consistency like ensuring the previous condition, with |
| 86 | * the assumption that the shift is +/- the domain size |
| 87 | * |
| 88 | * \param box to shift |
| 89 | * \param domain |
| 90 | * \param shift |
| 91 | * |
| 92 | */ |
| 93 | inline void consistent_shift(Box<dim,T> & box, const Box<dim,T> & domain, const Point<dim,T> & shift) |
| 94 | { |
| 95 | for (size_t k = 0 ; k < dim ; k++) |
| 96 | { |
| 97 | // if it touch on the left and shift on the right |
| 98 | if (box.getLow(k) == domain.getLow(k) && shift.get(k) > 0) |
| 99 | { |
| 100 | box.setLow(k,domain.getHigh(k)); |
| 101 | box.setHigh(k,box.getHigh(k) + shift.get(k)); |
| 102 | } |
| 103 | else if (box.getLow(k) == domain.getHigh(k) && shift.get(k) < 0) |
| 104 | { |
| 105 | box.setLow(k,domain.getLow(k)); |
| 106 | box.setHigh(k,box.getHigh(k) + shift.get(k)); |
| 107 | } |
| 108 | else if (box.getHigh(k) == domain.getHigh(k) && shift.get(k) < 0) |
| 109 | { |
| 110 | box.setHigh(k,domain.getLow(k)); |
| 111 | box.setLow(k,box.getLow(k) + shift.get(k)); |
| 112 | } |
| 113 | else if (box.getHigh(k) == domain.getLow(k) && shift.get(k) > 0) |
| 114 | { |
| 115 | box.setHigh(k,domain.getHigh(k)); |
| 116 | box.setLow(k,box.getLow(k) + shift.get(k)); |
| 117 | } |
| 118 | else |
| 119 | { |
| 120 | box.setHigh(k,box.getHigh(k) + shift.get(k)); |
| 121 | box.setLow(k,box.getLow(k) + shift.get(k)); |
| 122 | } |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | /*! \brief Message allocation |
| 127 | * |
| 128 | * \param msg_i message size required to receive from i |
| 129 | * \param total_msg total message size to receive from all the processors |
| 130 | * \param total_p the total number of processor want to communicate with you |
| 131 | * \param i processor id from which we receive |
| 132 | * \param ri request id (it is an id that goes from 0 to total_p, and is unique |
| 133 | * every time message_alloc is called) |
| 134 | * \param ptr a pointer to the vector_dist structure |
| 135 | * |
| 136 | * \return the pointer where to store the message |
| 137 | * |
| 138 | */ |
| 139 | static void * message_alloc(size_t msg_i ,size_t total_msg, size_t total_p, size_t i, size_t ri, size_t tag, void * ptr) |
| 140 | { |
| 141 | // cast the pointer |
| 142 | nn_prcs<dim,T,layout_base,Memory> * cd = static_cast< nn_prcs<dim,T,layout_base,Memory> *>(ptr); |
| 143 | |
| 144 | cd->nn_processor_subdomains[i].bx.resize(msg_i / sizeof(::Box<dim,T>) ); |
| 145 | |
| 146 | // Return the receive pointer |
| 147 | return cd->nn_processor_subdomains[i].bx.getPointer(); |
| 148 | } |
| 149 | |
| 150 | /*! \brief add sub-domains to processor for a near processor i |
| 151 | * |
| 152 | * \param i near processor |
| 153 | * \param r_sub real sub-domain id |
| 154 | * \param bx Box to add |
| 155 | * \param c from which sector the sub-domain come from |
| 156 | * |
| 157 | */ |
| 158 | inline void add_nn_subdomain(size_t i, size_t r_sub, const Box<dim,T> & bx, const comb<dim> & c) |
| 159 | { |
| 160 | N_box<dim,T> & nnpst = nn_processor_subdomains_tmp[i]; |
| 161 | nnpst.bx.add(bx); |
| 162 | nnpst.pos.add(c); |
| 163 | nnpst.r_sub.add(r_sub); |
| 164 | } |
| 165 | |
| 166 | /*! \brief In case of periodic boundary conditions we replicate the sub-domains at the border |
| 167 | * |
| 168 | * \param domain Domain |
| 169 | * \param ghost ghost part |
| 170 | * \param bc boundary boundary conditions |
| 171 | * |
| 172 | */ |
| 173 | void add_box_periodic(const Box<dim,T> & domain, const Ghost<dim,T> & ghost, const size_t (&bc)[dim]) |
| 174 | { |
| 175 | HyperCube<dim> hyp; |
| 176 | |
| 177 | // first we create boxes at the border of the domain used to detect the sub-domain |
| 178 | // that must be adjusted, each of this boxes define a shift in case of periodic boundary condition |
| 179 | for (long int i = dim-1 ; i >= 0 ; i--) |
| 180 | { |
| 181 | std::vector<comb<dim>> cmbs = hyp.getCombinations_R_bc(i,bc); |
| 182 | |
| 183 | for (size_t j = 0 ; j < cmbs.size() ; j++) |
| 184 | { |
| 185 | if (check_valid(cmbs[j],bc) == false) |
| 186 | continue; |
| 187 | |
| 188 | // Calculate the sector box |
| 189 | Box<dim,T> bp; |
| 190 | Point<dim,T> shift; |
| 191 | |
| 192 | for (size_t k = 0 ; k < dim ; k++) |
| 193 | { |
| 194 | switch (cmbs[j][k]) |
| 195 | { |
| 196 | case 1: |
| 197 | bp.setLow(k,domain.getHigh(k)+ghost.getLow(k)); |
| 198 | bp.setHigh(k,domain.getHigh(k)); |
| 199 | shift.get(k) = -domain.getHigh(k)+domain.getLow(k); |
| 200 | break; |
| 201 | case 0: |
| 202 | bp.setLow(k,domain.getLow(k)); |
| 203 | bp.setHigh(k,domain.getHigh(k)); |
| 204 | shift.get(k) = 0; |
| 205 | break; |
| 206 | case -1: |
| 207 | bp.setLow(k,domain.getLow(k)); |
| 208 | bp.setHigh(k,domain.getLow(k)+ghost.getHigh(k)); |
| 209 | shift.get(k) = domain.getHigh(k)-domain.getLow(k); |
| 210 | break; |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | // Detect all the sub-domain involved, shift them and add to the list |
| 215 | // Detection is performed intersecting the sub-domains with the ghost |
| 216 | // parts near the domain borders |
| 217 | for (size_t k = 0 ; k < getNNProcessors() ; k++) |
| 218 | { |
| 219 | // sub-domains of the near processor |
| 220 | const openfpm::vector< ::Box<dim,T> > & nn_sub = getNearSubdomains(IDtoProc(k)); |
| 221 | |
| 222 | for (size_t l = 0 ; l < nn_sub.size(); l++) |
| 223 | { |
| 224 | Box<dim,T> sub = nn_sub.get(l); |
| 225 | Box<dim,T> b_int; |
| 226 | |
| 227 | if (sub.Intersect(bp,b_int) == true) |
| 228 | { |
| 229 | Box<dim,T> sub2 = sub; |
| 230 | sub2 += shift; |
| 231 | |
| 232 | // Here we have to be careful of rounding off problems, in particular if any part |
| 233 | // of the sub-domain touch the border of the domain |
| 234 | |
| 235 | consistent_shift(sub,domain,shift); |
| 236 | |
| 237 | add_nn_subdomain(IDtoProc(k),l,sub,cmbs[j]); |
| 238 | } |
| 239 | } |
| 240 | } |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | flush(); |
| 245 | } |
| 246 | |
| 247 | /*! \brief Flush the temporal added sub-domain to the processor sub-domain |
| 248 | * |
| 249 | * |
| 250 | */ |
| 251 | void flush() |
| 252 | { |
| 253 | for ( auto it = nn_processor_subdomains_tmp.begin(); it != nn_processor_subdomains_tmp.end(); ++it ) |
| 254 | { |
| 255 | const N_box<dim,T> & nnp_bx = it->second; |
| 256 | |
| 257 | for (size_t i = 0 ; i < nnp_bx.bx.size() ; i++) |
| 258 | { |
| 259 | N_box<dim,T> & nnps = nn_processor_subdomains[it->first]; |
| 260 | const N_box<dim,T> & nnps_tmp = nn_processor_subdomains_tmp[it->first]; |
| 261 | |
| 262 | nnps.bx.add(nnps_tmp.bx.get(i)); |
| 263 | nnps.pos.add(nnps_tmp.pos.get(i)); |
| 264 | nnps.r_sub.add(nnps_tmp.r_sub.get(i)); |
| 265 | } |
| 266 | } |
| 267 | |
| 268 | nn_processor_subdomains_tmp.clear(); |
| 269 | } |
| 270 | |
| 271 | public: |
| 272 | |
| 273 | //! Constructor require Vcluster |
| 274 | nn_prcs(Vcluster<> & v_cl) |
| 275 | :v_cl(v_cl),recv_cnt(0),aBC(false) |
| 276 | {} |
| 277 | |
| 278 | //! Constructor from another nn_prcs |
| 279 | nn_prcs(const nn_prcs<dim,T,layout_base,Memory> & ilg) |
| 280 | :v_cl(ilg.v_cl),recv_cnt(0),aBC(false) |
| 281 | { |
| 282 | this->operator=(ilg); |
| 283 | }; |
| 284 | |
| 285 | //! Constructor from temporal ie_loc_ghost |
| 286 | nn_prcs(nn_prcs<dim,T,layout_base,Memory> && ilg) |
| 287 | :v_cl(ilg.v_cl),recv_cnt(0),aBC(false) |
| 288 | { |
| 289 | this->operator=(ilg); |
| 290 | } |
| 291 | |
| 292 | /*! Check that the combination is valid |
| 293 | * |
| 294 | * Is a function that is used in otder to understand if a sub-domain |
| 295 | * must be mirrored because of boundary conditions |
| 296 | * |
| 297 | * \param cmb combination |
| 298 | * \param bc boundary conditions |
| 299 | * |
| 300 | * \return true if the combination is valid |
| 301 | * |
| 302 | */ |
| 303 | static bool inline check_valid(comb<dim> cmb,const size_t (& bc)[dim]) |
| 304 | { |
| 305 | // the combination 0 is not valid |
| 306 | if (cmb.n_zero() == dim) |
| 307 | return false; |
| 308 | |
| 309 | for (size_t i = 0 ; i < dim ; i++) |
| 310 | { |
| 311 | if (bc[i] == NON_PERIODIC && cmb.getComb()[i] != 0) |
| 312 | return false; |
| 313 | } |
| 314 | return true; |
| 315 | } |
| 316 | |
| 317 | /*! \brief Copy the object |
| 318 | * |
| 319 | * \param nnp object to copy |
| 320 | * |
| 321 | * \return itself |
| 322 | * |
| 323 | */ |
| 324 | nn_prcs<dim,T,layout_base,Memory> & operator=(const nn_prcs<dim,T,layout_base,Memory> & nnp) |
| 325 | { |
| 326 | nn_processors = nnp.nn_processors; |
| 327 | nn_processor_subdomains = nnp.nn_processor_subdomains; |
| 328 | proc_adj_box = nnp.proc_adj_box; |
| 329 | boxes = nnp.boxes; |
| 330 | |
| 331 | return *this; |
| 332 | } |
| 333 | |
| 334 | /*! \brief Copy the object |
| 335 | * |
| 336 | * \param nnp object to copy |
| 337 | * |
| 338 | * \return itself |
| 339 | * |
| 340 | */ |
| 341 | nn_prcs<dim,T,layout_base,Memory> & operator=(nn_prcs<dim,T,layout_base,Memory> && nnp) |
| 342 | { |
| 343 | nn_processors.swap(nnp.nn_processors); |
| 344 | nn_processor_subdomains.swap(nnp.nn_processor_subdomains); |
| 345 | proc_adj_box.swap(nnp.proc_adj_box); |
| 346 | boxes = nnp.boxes; |
| 347 | |
| 348 | return *this; |
| 349 | } |
| 350 | |
| 351 | /*! \brief Copy the object |
| 352 | * |
| 353 | * \param nnp object to copy |
| 354 | * |
| 355 | * \return itself |
| 356 | * |
| 357 | */ |
| 358 | template<typename Memory2, template <typename> class layout_base2> |
| 359 | nn_prcs<dim,T,layout_base,Memory> & operator=(const nn_prcs<dim,T,layout_base2,Memory2> & nnp) |
| 360 | { |
| 361 | nn_processors = nnp.private_get_nn_processors(); |
| 362 | nn_processor_subdomains = nnp.private_get_nn_processor_subdomains(); |
| 363 | proc_adj_box = nnp.private_get_proc_adj_box(); |
| 364 | boxes = nnp.private_get_boxes(); |
| 365 | |
| 366 | return *this; |
| 367 | } |
| 368 | |
| 369 | /*! \brief Return the internal nn_processor struct |
| 370 | * |
| 371 | * \return the internal nn_processor struct |
| 372 | * |
| 373 | */ |
| 374 | openfpm::vector<size_t> & private_get_nn_processors() |
| 375 | { |
| 376 | return nn_processors; |
| 377 | } |
| 378 | |
| 379 | /*! \brief Return the internal nn_processor_subdomains |
| 380 | * |
| 381 | * \return the internal nn_processor_subdomains |
| 382 | * |
| 383 | */ |
| 384 | std::unordered_map<size_t, N_box<dim,T>> & private_get_nn_processor_subdomains() |
| 385 | { |
| 386 | return nn_processor_subdomains; |
| 387 | } |
| 388 | |
| 389 | /*! \brief Return the internal proc_adj_box |
| 390 | * |
| 391 | * \return the internal proc_adj_box |
| 392 | * |
| 393 | */ |
| 394 | openfpm::vector<openfpm::vector<size_t>> & private_get_proc_adj_box() |
| 395 | { |
| 396 | return proc_adj_box; |
| 397 | } |
| 398 | |
| 399 | /*! \brief Return the internal boxes structure |
| 400 | * |
| 401 | * \return the internal boxes structure |
| 402 | * |
| 403 | */ |
| 404 | openfpm::vector< openfpm::vector< ::SpaceBox<dim,T>> > & private_get_boxes() |
| 405 | { |
| 406 | return boxes; |
| 407 | } |
| 408 | |
| 409 | /*! \brief Copy the object |
| 410 | * |
| 411 | * \param nnp object to copy |
| 412 | * |
| 413 | * \return itself |
| 414 | * |
| 415 | */ |
| 416 | template<typename Memory2, template <typename> class layout_base2> |
| 417 | nn_prcs<dim,T,layout_base,Memory> & operator=(nn_prcs<dim,T,layout_base2,Memory2> && nnp) |
| 418 | { |
| 419 | nn_processors.swap(nnp.private_get_nn_processors()); |
| 420 | nn_processor_subdomains.swap(nnp.private_get_nn_processor_subdomains()); |
| 421 | proc_adj_box.swap(nnp.private_get_proc_adj_box()); |
| 422 | boxes = nnp.private_get_boxes(); |
| 423 | |
| 424 | return *this; |
| 425 | } |
| 426 | |
| 427 | /*! \brief Create the list of adjacent processors and the list of adjacent sub-domains |
| 428 | * |
| 429 | * \param box_nn_processor list of adjacent processors for each sub-domain |
| 430 | * \param sub_domains list of local sub-domains |
| 431 | * |
| 432 | */ |
| 433 | void create(const openfpm::vector<openfpm::vector<long unsigned int> > & box_nn_processor, |
| 434 | const openfpm::vector<SpaceBox<dim,T>,Memory,layout_base> & sub_domains) |
| 435 | { |
| 436 | // produce the list of the adjacent processor (nn_processors) list |
| 437 | for (size_t i = 0 ; i < box_nn_processor.size() ; i++) |
| 438 | { |
| 439 | for (size_t j = 0 ; j < box_nn_processor.get(i).size() ; j++) |
| 440 | { |
| 441 | nn_processors.add(box_nn_processor.get(i).get(j)); |
| 442 | } |
| 443 | } |
| 444 | |
| 445 | // make the list of the processor sort and unique |
| 446 | std::sort(nn_processors.begin(), nn_processors.end()); |
| 447 | auto last = std::unique(nn_processors.begin(), nn_processors.end()); |
| 448 | nn_processors.erase(last, nn_processors.end()); |
| 449 | |
| 450 | // link nn_processor_subdomains to nn_processors |
| 451 | // it is used to quickly convert the Processor rank to the position in the list of the |
| 452 | // near processors |
| 453 | for (size_t i = 0 ; i < box_nn_processor.size() ; i++) |
| 454 | { |
| 455 | for (size_t j = 0 ; j < box_nn_processor.get(i).size() ; j++) |
| 456 | { |
| 457 | // processor id adjacent to this sub-domain |
| 458 | size_t proc_id = box_nn_processor.get(i).get(j); |
| 459 | |
| 460 | size_t k = 0; |
| 461 | // search inside near processor list |
| 462 | for (k = 0 ; k < nn_processors.size() ; k++) |
| 463 | if (nn_processors.get(k) == proc_id) break; |
| 464 | |
| 465 | nn_processor_subdomains[proc_id].id = k; |
| 466 | } |
| 467 | } |
| 468 | |
| 469 | // create a buffer with the sub-domains that can have an intersection with |
| 470 | // the near processors |
| 471 | proc_adj_box.resize(getNNProcessors()); |
| 472 | boxes.resize(getNNProcessors()); |
| 473 | |
| 474 | for (size_t b = 0 ; b < box_nn_processor.size() ; b++) |
| 475 | { |
| 476 | for (size_t p = 0 ; p < box_nn_processor.get(b).size() ; p++) |
| 477 | { |
| 478 | size_t prc = box_nn_processor.get(b).get(p); |
| 479 | |
| 480 | // id of the processor in the processor list |
| 481 | // [value between 0 and the number of the near processors] |
| 482 | size_t id = ProctoID(prc); |
| 483 | |
| 484 | boxes.get(id).add(sub_domains.get(b)); |
| 485 | proc_adj_box.get(id).add(b); |
| 486 | } |
| 487 | } |
| 488 | |
| 489 | nn_processor_subdomains.reserve(nn_processors.size()); |
| 490 | |
| 491 | // Get the sub-domains of the near processors |
| 492 | v_cl.sendrecvMultipleMessagesNBX(nn_processors,boxes,nn_prcs<dim,T,layout_base,Memory>::message_alloc, this ,NEED_ALL_SIZE); |
| 493 | |
| 494 | // Add to all the received sub-domains the information that they live in the central sector |
| 495 | for ( auto it = nn_processor_subdomains.begin(); it != nn_processor_subdomains.end(); ++it ) |
| 496 | { |
| 497 | const N_box<dim,T> & nnp_bx = it->second; |
| 498 | |
| 499 | for (size_t i = 0 ; i < nnp_bx.bx.size() ; i++) |
| 500 | { |
| 501 | comb<dim> c; |
| 502 | c.zero(); |
| 503 | |
| 504 | N_box<dim,T> & nnps = nn_processor_subdomains[it->first]; |
| 505 | |
| 506 | nnps.pos.add(c); |
| 507 | nnps.r_sub.add(i); |
| 508 | nnps.n_real_sub = nnps.bx.size(); |
| 509 | } |
| 510 | } |
| 511 | } |
| 512 | |
| 513 | /*! \brief Get the number of Near processors |
| 514 | * |
| 515 | * \return the number of near processors |
| 516 | * |
| 517 | */ |
| 518 | inline size_t getNNProcessors() const |
| 519 | { |
| 520 | return nn_processors.size(); |
| 521 | } |
| 522 | |
| 523 | /*! \brief Return the processor id of the near processor list at place id |
| 524 | * |
| 525 | * \param id |
| 526 | * |
| 527 | * \return return the processor rank |
| 528 | * |
| 529 | */ |
| 530 | inline size_t IDtoProc(size_t id) const |
| 531 | { |
| 532 | return nn_processors.get(id); |
| 533 | } |
| 534 | |
| 535 | /*! \brief Get the real-id of the sub-domains of a near processor |
| 536 | * |
| 537 | * \param p_id near processor rank |
| 538 | * |
| 539 | * \return the sub-domains real id |
| 540 | * |
| 541 | */ |
| 542 | inline const openfpm::vector< size_t > & getNearSubdomainsRealId(size_t p_id) const |
| 543 | { |
| 544 | auto key = nn_processor_subdomains.find(p_id); |
| 545 | #ifdef SE_CLASS1 |
| 546 | if (key == nn_processor_subdomains.end()) |
| 547 | { |
| 548 | std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " error this process rank is not adjacent to the local processor" ; |
| 549 | } |
| 550 | #endif |
| 551 | |
| 552 | return key->second.r_sub; |
| 553 | } |
| 554 | |
| 555 | /*! \brief Get the sub-domains of a near processor |
| 556 | * |
| 557 | * \param p_id near processor rank |
| 558 | * |
| 559 | * \return the sub-domains |
| 560 | * |
| 561 | */ |
| 562 | inline const openfpm::vector< ::Box<dim,T> > & getNearSubdomains(size_t p_id) const |
| 563 | { |
| 564 | auto key = nn_processor_subdomains.find(p_id); |
| 565 | #ifdef SE_CLASS1 |
| 566 | if (key == nn_processor_subdomains.end()) |
| 567 | { |
| 568 | std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " error this process rank is not adjacent to the local processor" ; |
| 569 | } |
| 570 | #endif |
| 571 | |
| 572 | return key->second.bx; |
| 573 | } |
| 574 | |
| 575 | /*! \brief Get the number of real sub-domains of a near processor |
| 576 | * |
| 577 | * \note the real sub-domain are the subdomain in the central sector, or any sub-domain that has not been create because of boundary conditions |
| 578 | * |
| 579 | * \param p_id near processor rank |
| 580 | * |
| 581 | * \return the number of real sub-domains |
| 582 | * |
| 583 | */ |
| 584 | inline size_t getNRealSubdomains(size_t p_id) const |
| 585 | { |
| 586 | auto key = nn_processor_subdomains.find(p_id); |
| 587 | #ifdef SE_CLASS1 |
| 588 | if (key == nn_processor_subdomains.end()) |
| 589 | { |
| 590 | std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " error this process rank is not adjacent to the local processor" ; |
| 591 | } |
| 592 | #endif |
| 593 | |
| 594 | return key->second.n_real_sub; |
| 595 | } |
| 596 | |
| 597 | /*! \brief Get the sub-domains sector position of a near processor |
| 598 | * |
| 599 | * \param p_id near processor rank |
| 600 | * |
| 601 | * \return the sub-domains positions |
| 602 | * |
| 603 | */ |
| 604 | inline const openfpm::vector< comb<dim> > & getNearSubdomainsPos(size_t p_id) const |
| 605 | { |
| 606 | auto key = nn_processor_subdomains.find(p_id); |
| 607 | #ifdef SE_CLASS1 |
| 608 | if (key == nn_processor_subdomains.end()) |
| 609 | { |
| 610 | std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " error this process rank is not adjacent to the local processor" ; |
| 611 | } |
| 612 | #endif |
| 613 | return key->second.pos; |
| 614 | } |
| 615 | |
| 616 | /*! \brief Get the near processor id |
| 617 | * |
| 618 | * \param p_id adjacent processor rank |
| 619 | * |
| 620 | * \return the processor rank |
| 621 | * |
| 622 | */ |
| 623 | inline size_t getNearProcessor(size_t p_id) const |
| 624 | { |
| 625 | auto key = nn_processor_subdomains.find(p_id); |
| 626 | #ifdef SE_CLASS1 |
| 627 | if (key == nn_processor_subdomains.end()) |
| 628 | { |
| 629 | std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " error this process rank is not adjacent to the local processor" ; |
| 630 | } |
| 631 | #endif |
| 632 | return key->second.id; |
| 633 | } |
| 634 | |
| 635 | |
| 636 | /*! \brief For each near processor it give a vector with the id |
| 637 | * of the local sub-domain sent to that processor |
| 638 | * |
| 639 | * \param p_id adjacent processor (id from 0 to getNNProcessors()) |
| 640 | * |
| 641 | * \return a vector of sub-domains id |
| 642 | * |
| 643 | */ |
| 644 | inline const openfpm::vector<size_t> & getSentSubdomains(size_t p_id) const |
| 645 | { |
| 646 | return proc_adj_box.get(p_id); |
| 647 | } |
| 648 | |
| 649 | /*! \brief Convert the processor rank to the id in the list |
| 650 | * |
| 651 | * \param p processor rank |
| 652 | * |
| 653 | * \return the id |
| 654 | * |
| 655 | */ |
| 656 | inline size_t ProctoID(size_t p) const |
| 657 | { |
| 658 | auto key = nn_processor_subdomains.find(p); |
| 659 | #ifdef SE_CLASS1 |
| 660 | if (key == nn_processor_subdomains.end()) |
| 661 | { |
| 662 | std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " error this process rank is not adjacent to the local processor" ; |
| 663 | } |
| 664 | #endif |
| 665 | |
| 666 | return key->second.id; |
| 667 | } |
| 668 | |
| 669 | /*! \brief Write the decomposition as VTK file |
| 670 | * |
| 671 | * The function generate several files |
| 672 | * |
| 673 | * 1) subdomains_adjacent_X.vtk sub-domains adjacent to the local processor (X) |
| 674 | * |
| 675 | * where X is the local processor rank |
| 676 | * |
| 677 | * \param output directory where to write the files |
| 678 | * |
| 679 | * \return true if the write procedure succeed |
| 680 | * |
| 681 | */ |
| 682 | bool write(std::string output) const |
| 683 | { |
| 684 | //! subdomains_adjacent_X.vtk sub-domains adjacent to the local processor (X) |
| 685 | VTKWriter<openfpm::vector<::Box<dim,T>>,VECTOR_BOX> vtk_box2; |
| 686 | for (size_t p = 0 ; p < nn_processors.size() ; p++) |
| 687 | { |
| 688 | size_t prc = nn_processors.get(p); |
| 689 | auto it = nn_processor_subdomains.find(prc); |
| 690 | if (it != nn_processor_subdomains.end()) |
| 691 | vtk_box2.add(nn_processor_subdomains.at(prc).bx); |
| 692 | } |
| 693 | vtk_box2.write(output + std::string("subdomains_adjacent_" ) + std::to_string(v_cl.getProcessUnitID()) + std::string(".vtk" )); |
| 694 | |
| 695 | return true; |
| 696 | } |
| 697 | |
| 698 | /*! \brief Apply boundary conditions |
| 699 | * |
| 700 | * \param domain The simulation domain |
| 701 | * \param ghost ghost part |
| 702 | * \param bc Boundary conditions |
| 703 | * |
| 704 | */ |
| 705 | void applyBC(const Box<dim,T> & domain, const Ghost<dim,T> & ghost, const size_t (&bc)[dim]) |
| 706 | { |
| 707 | if (aBC == true) |
| 708 | { |
| 709 | std::cerr << "Warning " << __FILE__ << ":" << __LINE__ << " apply BC is suppose to be called only one time\n" ; |
| 710 | return; |
| 711 | } |
| 712 | |
| 713 | aBC=true; |
| 714 | |
| 715 | add_box_periodic(domain,ghost,bc); |
| 716 | } |
| 717 | |
| 718 | /*! \brief Check if the nn_prcs contain the same information |
| 719 | * |
| 720 | * \param np Element to check |
| 721 | * |
| 722 | * \return true if they are equal |
| 723 | * |
| 724 | */ |
| 725 | bool is_equal(nn_prcs<dim,T,layout_base,Memory> & np) |
| 726 | { |
| 727 | if (np.getNNProcessors() != getNNProcessors()) |
| 728 | return false; |
| 729 | |
| 730 | for (size_t p = 0 ; p < getNNProcessors() ; p++) |
| 731 | { |
| 732 | if (getNearSubdomains(IDtoProc(p)) != np.getNearSubdomains(IDtoProc(p))) |
| 733 | return false; |
| 734 | if (getNearProcessor(IDtoProc(p)) != np.getNearProcessor(IDtoProc(p))) |
| 735 | return false; |
| 736 | if (getSentSubdomains(p) != np.getSentSubdomains(p)) |
| 737 | return false; |
| 738 | } |
| 739 | |
| 740 | return true; |
| 741 | } |
| 742 | |
| 743 | /*! \brief Reset the nn_prcs structure |
| 744 | * |
| 745 | */ |
| 746 | void reset() |
| 747 | { |
| 748 | nn_processors.clear(); |
| 749 | nn_processor_subdomains.clear(); |
| 750 | nn_processor_subdomains_tmp.clear(); |
| 751 | proc_adj_box.clear(); |
| 752 | boxes.clear(); |
| 753 | recv_cnt = 0; |
| 754 | aBC = false; |
| 755 | } |
| 756 | |
| 757 | //! Used for testing porpose do not use |
| 758 | std::unordered_map<size_t, N_box<dim,T>> & get_nn_processor_subdomains() |
| 759 | { |
| 760 | return nn_processor_subdomains; |
| 761 | } |
| 762 | |
| 763 | //! Used for testing porpose do not use |
| 764 | openfpm::vector<size_t> & get_nn_processors() |
| 765 | { |
| 766 | return nn_processors; |
| 767 | } |
| 768 | }; |
| 769 | |
| 770 | |
| 771 | #endif /* SRC_DECOMPOSITION_NN_PROCESSOR_HPP_ */ |
| 772 | |