vector_dist_comm.hpp 32.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13
/*
 * vector_dist_comm.hpp
 *
 *  Created on: Aug 18, 2016
 *      Author: i-bird
 */

#ifndef SRC_VECTOR_VECTOR_DIST_COMM_HPP_
#define SRC_VECTOR_VECTOR_DIST_COMM_HPP_

#define V_SUB_UNIT_FACTOR 64

#define SKIP_LABELLING 512
14
#define KEEP_PROPERTIES 512
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

#define NO_POSITION 1
#define WITH_POSITION 2

/*! \brief This class is an helper for the communication of vector_dist
 *
 * \tparam dim Dimensionality of the space where the elements lives
 * \tparam St type of space float, double ...
 * \tparam prop properties the vector element store in OpenFPM data structure format
 * \tparam Decomposition Decomposition strategy to use CartDecomposition ...
 * \tparam Memory Memory pool where store the information HeapMemory ...
 *
 * \see vector_dist
 *
 */

template<unsigned int dim, typename St, typename prop, typename Decomposition = CartDecomposition<dim,St>, typename Memory = HeapMemory>
class vector_dist_comm
{
34 35 36
	//! definition of the send vector for position
	typedef openfpm::vector<Point<dim, St>, Memory> send_pos_vector;

37 38 39 40 41 42 43 44 45
	//! VCluster
	Vcluster & v_cl;

	//! Domain decomposition
	Decomposition dec;

	//! It map the processor id with the communication request into map procedure
	openfpm::vector<size_t> p_map_req;

Pietro Incardona's avatar
Pietro Incardona committed
46
	//! For each near processor, outgoing particle id
47 48 49 50 51
	//! \warning opart is assumed to be an ordered list
	//! first id particle id
	//! second id shift id
	//! third id is the processor id
	openfpm::vector<aggregate<size_t,size_t,size_t>> m_opart;
52

incardon's avatar
incardon committed
53
	//! Per processor ordered particles id for ghost_get (see prc_g_opart)
54 55 56
	//! For each processor the internal vector store the id of the
	//! particles that must be communicated to the other processors
	openfpm::vector<openfpm::vector<aggregate<size_t,size_t>>> g_opart;
57

incardon's avatar
incardon committed
58 59 60
	// processor rank list of g_opart
	openfpm::vector<size_t> prc_g_opart;

61
	//! Sending buffer for the ghost particles position
62
	openfpm::vector<send_pos_vector> g_pos_send;
63

64 65 66
	//! It store the list of processor that communicate with us (local processor)
	//! from the last ghost get
	openfpm::vector<size_t> prc_recv_get;
Pietro Incardona's avatar
Pietro Incardona committed
67

68
	//! the same as prc_recv_get but for put
Pietro Incardona's avatar
Pietro Incardona committed
69 70
	openfpm::vector<size_t> prc_recv_put;

71 72
	//! the same as prc_recv_get but for map
	openfpm::vector<size_t> prc_recv_map;
Pietro Incardona's avatar
Pietro Incardona committed
73

74 75 76
	//! It store the size of the elements added for each processor that communicate with us (local processor)
	//! from the last ghost get
	openfpm::vector<size_t> recv_sz_get;
77

78
	//! The same as recv_sz_get but for put
Pietro Incardona's avatar
Pietro Incardona committed
79 80
	openfpm::vector<size_t> recv_sz_put;

81 82
	//! The same as recv_sz_get but for map
	openfpm::vector<size_t> recv_sz_map;
83

Pietro Incardona's avatar
Pietro Incardona committed
84 85 86 87
	//! Local ghost marker (across the ghost particles it mark from where we have the)
	//! replicated ghost particles that are local
	size_t lg_m;

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
	//! process the particle with properties
	struct proc_without_prp
	{
		template<typename T1, typename T2> inline static void proc(size_t lbl, size_t cnt, size_t id, T1 & v_prp, T2 & m_prp)
		{
			m_prp.get(lbl).set(cnt, v_prp.get(id));
		}
	};

	template<typename prp_object, int ... prp>
	struct proc_with_prp
	{
		template<typename T1, typename T2> inline static void proc(size_t lbl, size_t cnt, size_t id, T1 & v_prp, T2 & m_prp)
		{
			// source object type
			typedef encapc<1, prop, typename openfpm::vector<prop>::layout_type> encap_src;
			// destination object type
			typedef encapc<1, prp_object, typename openfpm::vector<prp_object>::layout_type> encap_dst;

			// Copy only the selected properties
			object_si_d<encap_src, encap_dst, OBJ_ENCAP, prp...>(v_prp.get(id), m_prp.get(lbl).get(cnt));
		}
	};

	//! It process one particle
	template<typename proc_class, typename T1, typename T2, typename T3, typename T4> inline void process_map_particle(size_t i, long int & end, long int & id_end, T1 & m_pos, T2 & m_prp, T3 & v_pos, T4 & v_prp, openfpm::vector<size_t> & cnt)
	{
		long int prc_id = m_opart.template get<2>(i);
		size_t id = m_opart.template get<0>(i);

		if (prc_id >= 0)
		{
			size_t lbl = p_map_req.get(prc_id);

			m_pos.get(lbl).set(cnt.get(lbl), v_pos.get(id));
			proc_class::proc(lbl,cnt.get(lbl),id,v_prp,m_prp);

			cnt.get(lbl)++;

			// swap the particle
			long int id_valid = get_end_valid(end,id_end);

			if (id_valid > 0 && (long int)id < id_valid)
			{
				v_pos.set(id,v_pos.get(id_valid));
				v_prp.set(id,v_prp.get(id_valid));
			}
		}
		else
		{
			// swap the particle
			long int id_valid = get_end_valid(end,id_end);

			if (id_valid > 0 && (long int)id < id_valid)
			{
				v_pos.set(id,v_pos.get(id_valid));
				v_prp.set(id,v_prp.get(id_valid));
			}
		}
	}

149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
	/*! \brief Return a valid particle starting from end and tracing back
	 *
	 * \param end actual opart particle pointer
	 * \param actual end particle point
	 *
	 * \return a valid particle
	 *
	 */
	inline size_t get_end_valid(long int & end, long int & end_id)
	{
		end_id--;

		while (end >= 0 && end_id >= 0 && (long int)m_opart.template get<0>(end) == end_id)
		{
			end_id--;
			end--;
		}

		return end_id;
	}

Pietro Incardona's avatar
Pietro Incardona committed
170
	//! Flags that indicate that the function createShiftBox() has been called
171 172
	bool is_shift_box_created = false;

Pietro Incardona's avatar
Pietro Incardona committed
173
	//! this map is used to check if a combination is already present
174 175
	std::unordered_map<size_t, size_t> map_cmb;

Pietro Incardona's avatar
Pietro Incardona committed
176 177
	//! The boxes touching the border of the domain are divided in groups (first vector)
	//! each group contain internal ghost coming from sub-domains of the same section
178 179
	openfpm::vector_std<openfpm::vector_std<Box<dim, St>>>box_f;

Pietro Incardona's avatar
Pietro Incardona committed
180
	//! Store the sector for each group (previous vector)
181 182
	openfpm::vector_std<comb<dim>> box_cmb;

Pietro Incardona's avatar
Pietro Incardona committed
183
	//! Id of the local particle to replicate for ghost_get
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
	openfpm::vector<aggregate<size_t,size_t>> o_part_loc;

	/*! \brief For every internal ghost box we create a structure that order such internal local ghost box in
	 *         shift vectors
	 *
	 */
	void createShiftBox()
	{
		if (is_shift_box_created == true)
			return;

		// Add local particles coming from periodic boundary, the only boxes that count are the one
		// touching the border, filter them
		for (size_t i = 0; i < dec.getNLocalSub(); i++)
		{
			size_t Nl = dec.getLocalNIGhost(i);

			for (size_t j = 0; j < Nl; j++)
			{
				// If the ghost does not come from the intersection with an out of
				// border sub-domain the combination is all zero and n_zero return dim
				if (dec.getLocalIGhostPos(i, j).n_zero() == dim)
					continue;

				// Check if we already have boxes with such combination
				auto it = map_cmb.find(dec.getLocalIGhostPos(i, j).lin());
				if (it == map_cmb.end())
				{
					// we do not have it
					box_f.add();
					box_f.last().add(dec.getLocalIGhostBox(i, j));
					box_cmb.add(dec.getLocalIGhostPos(i, j));
					map_cmb[dec.getLocalIGhostPos(i, j).lin()] = box_f.size() - 1;
				}
				else
				{
					// we have it
					box_f.get(it->second).add(dec.getLocalIGhostBox(i, j));
				}

			}
		}

		is_shift_box_created = true;
	}

	/*! \brief Local ghost from labeled particles
	 *
Pietro Incardona's avatar
Pietro Incardona committed
232 233
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particles properties
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
	 *
	 */
	void local_ghost_from_opart(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp)
	{
		// get the shift vectors
		const openfpm::vector<Point<dim, St>> & shifts = dec.getShiftVectors();

		for (size_t i = 0 ; i < o_part_loc.size() ; i++)
		{
			size_t lin_id = o_part_loc.get<1>(i);
			size_t key = o_part_loc.template get<0>(i);

			Point<dim, St> p = v_pos.get(key);
			// shift
			p -= shifts.get(lin_id);

			// add this particle shifting its position
			v_pos.add(p);
incardon's avatar
incardon committed
252
			v_prp.get(lg_m+i) = v_prp.get(key);
253 254 255 256 257 258 259
		}
	}

	/*! \brief Local ghost from decomposition
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particle properties
Pietro Incardona's avatar
Pietro Incardona committed
260
	 * \param g_m ghost marker
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
	 *
	 */
	void local_ghost_from_dec(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, size_t g_m)
	{
		o_part_loc.clear();

		// get the shift vectors
		const openfpm::vector<Point<dim, St>> & shifts = dec.getShiftVectors();

		// Label the internal (assigned) particles
		auto it = v_pos.getIteratorTo(g_m);

		while (it.isNext())
		{
			auto key = it.get();

			// If particles are inside these boxes
			for (size_t i = 0; i < box_f.size(); i++)
			{
				for (size_t j = 0; j < box_f.get(i).size(); j++)
				{
					if (box_f.get(i).get(j).isInside(v_pos.get(key)) == true)
					{
						size_t lin_id = box_cmb.get(i).lin();

						o_part_loc.add();
						o_part_loc.template get<0>(o_part_loc.size()-1) = key;
						o_part_loc.template get<1>(o_part_loc.size()-1) = lin_id;

						Point<dim, St> p = v_pos.get(key);
						// shift
						p -= shifts.get(lin_id);

						// add this particle shifting its position
						v_pos.add(p);
						v_prp.add();
						v_prp.last() = v_prp.get(key);

						// boxes in one group can be overlapping
						// we do not have to search for the other
						// boxes otherwise we will have duplicate particles
						//
						// A small note overlap of boxes across groups is fine
						// (and needed) because each group has different shift
						// producing non overlapping particles
						//
						break;
					}
				}
			}

			++it;
		}
	}

	/*! \brief Add local particles based on the boundary conditions
	 *
	 * In order to understand what this function use the following
	 *
	 \verbatim

	 [1,1]
	 +---------+------------------------+---------+
	 | (1,-1)  |                        | (1,1)   |
	 |   |     |    (1,0) --> 7         |   |     |
	 |   v     |                        |   v     |
	 |   6     |                        |   8     |
	 +--------------------------------------------+
	 |         |                        |         |
	 |         |                        |         |
	 |         |                        |         |
	 | (-1,0)  |                        | (1,0)   |
	 |    |    |                        |   |     |
	 |    v    |      (0,0) --> 4       |   v     |
	 |    3    |                        |   5     |
	 |         |                        |         |
 B	 |         |                        |     A   |
 *	 |         |                        |    *    |
	 |         |                        |         |
	 |         |                        |         |
	 |         |                        |         |
	 +--------------------------------------------+
	 | (-1,-1) |                        | (-1,1)  |
	 |    |    |   (-1,0) --> 1         |    |    |
	 |    v    |                        |    v    |
	 |    0    |                        |    2    |
	 +---------+------------------------+---------+


	 \endverbatim

	 *
	 *  The box is the domain, while all boxes at the border (so not (0,0) ) are the
	 *  ghost part at the border of the domain. If a particle A is in the position in figure
	 *  a particle B must be created. This function duplicate the particle A, if A and B are
	 *  local
	 *
	 * \param v_pos vector of particle of positions
	 * \param v_prp vector of particle properties
	 * \param g_m ghost marker
	 * \param opt options
	 *
	 */
	void add_loc_particles_bc(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp ,size_t & g_m, size_t opt)
	{
		// Create the shift boxes
		createShiftBox();

incardon's avatar
incardon committed
369 370 371 372 373
		if (opt != SKIP_LABELLING)
			lg_m = v_prp.size();

//		v_pos.resize(lg_m);
//		v_prp.resize(lg_m);
Pietro Incardona's avatar
Pietro Incardona committed
374

375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
		if (box_f.size() == 0)
			return;
		else
		{
			if (opt & SKIP_LABELLING)
				local_ghost_from_opart(v_pos,v_prp);
			else
				local_ghost_from_dec(v_pos,v_prp,g_m);
		}
	}

	/*! \brief This function fill the send buffer for the particle position after the particles has been label with labelParticles
	 *
	 * \param v_pos vector of particle positions
	 * \param g_pos_send Send buffer to fill
	 * \param prAlloc_pos Memory object for the send buffer
	 *
	 */
393
	void fill_send_ghost_pos_buf(openfpm::vector<Point<dim, St>> & v_pos,openfpm::vector<send_pos_vector> & g_pos_send)
394 395 396 397 398
	{
		// get the shift vectors
		const openfpm::vector<Point<dim, St>> & shifts = dec.getShiftVectors();

		// create a number of send buffers equal to the near processors
399
		g_pos_send.resize(g_opart.size());
400 401 402
		for (size_t i = 0; i < g_pos_send.size(); i++)
		{
			// resize the sending vector (No allocation is produced)
403
			g_pos_send.get(i).resize(g_opart.get(i).size());
404 405 406
		}

		// Fill the send buffer
407
		for (size_t i = 0; i < g_opart.size(); i++)
408
		{
409
			for (size_t j = 0; j < g_opart.get(i).size(); j++)
410
			{
411 412
				Point<dim, St> s = v_pos.get(g_opart.get(i).template get<0>(j));
				s -= shifts.get(g_opart.get(i).template get<1>(j));
413 414 415 416 417
				g_pos_send.get(i).set(j, s);
			}
		}
	}

Pietro Incardona's avatar
Pietro Incardona committed
418 419 420 421 422 423 424 425 426 427 428 429
	/*! \brief This function fill the send buffer for ghost_put
	 *
	 * \tparam send_vector type used to send data
	 * \tparam prp_object object containing only the properties to send
	 * \tparam prp set of properties to send
	 *
	 * \param v_prp vector of particle properties
	 * \param g_send_prp Send buffer to fill
	 * \param prAlloc_prp Memory object for the send buffer
	 * \param g_m ghost marker
	 *
	 */
430
	template<typename send_vector, typename prp_object, int ... prp> void fill_send_ghost_put_prp_buf(openfpm::vector<prop> & v_prp, openfpm::vector<send_vector> & g_send_prp, size_t & g_m)
Pietro Incardona's avatar
Pietro Incardona committed
431 432 433
	{
		// create a number of send buffers equal to the near processors
		// from which we received
434
		g_send_prp.resize(prc_recv_get.size());
Pietro Incardona's avatar
Pietro Incardona committed
435 436 437
		for (size_t i = 0; i < g_send_prp.size(); i++)
		{
			// resize the sending vector (No allocation is produced)
438
			g_send_prp.get(i).resize(recv_sz_get.get(i));
Pietro Incardona's avatar
Pietro Incardona committed
439 440 441 442 443
		}

		size_t accum = g_m;

		// Fill the send buffer
444
		for (size_t i = 0; i < prc_recv_get.size(); i++)
Pietro Incardona's avatar
Pietro Incardona committed
445 446
		{
			size_t j2 = 0;
447
			for (size_t j = accum; j < accum + recv_sz_get.get(i); j++)
Pietro Incardona's avatar
Pietro Incardona committed
448 449 450 451 452 453 454 455 456 457 458 459
			{
				// source object type
				typedef encapc<1, prop, typename openfpm::vector<prop>::layout_type> encap_src;
				// destination object type
				typedef encapc<1, prp_object, typename openfpm::vector<prp_object>::layout_type> encap_dst;

				// Copy only the selected properties
				object_si_d<encap_src, encap_dst, OBJ_ENCAP, prp...>(v_prp.get(j), g_send_prp.get(i).get(j2));

				j2++;
			}

460
			accum = accum + recv_sz_get.get(i);
Pietro Incardona's avatar
Pietro Incardona committed
461 462 463
		}
	}

464 465 466 467 468 469 470 471 472 473 474
	/*! \brief This function fill the send buffer for properties after the particles has been label with labelParticles
	 *
	 * \tparam send_vector type used to send data
	 * \tparam prp_object object containing only the properties to send
	 * \tparam prp set of properties to send
	 *
	 * \param v_prp vector of particle properties
	 * \param g_send_prp Send buffer to fill
	 * \param prAlloc_prp Memory object for the send buffer
	 *
	 */
475
	template<typename send_vector, typename prp_object, int ... prp> void fill_send_ghost_prp_buf(openfpm::vector<prop> & v_prp, openfpm::vector<send_vector> & g_send_prp)
476 477
	{
		// create a number of send buffers equal to the near processors
478
		g_send_prp.resize(g_opart.size());
479 480 481
		for (size_t i = 0; i < g_send_prp.size(); i++)
		{
			// resize the sending vector (No allocation is produced)
482
			g_send_prp.get(i).resize(g_opart.get(i).size());
483 484 485
		}

		// Fill the send buffer
486
		for (size_t i = 0; i < g_opart.size(); i++)
487
		{
488
			for (size_t j = 0; j < g_opart.get(i).size(); j++)
489 490 491 492 493 494 495
			{
				// source object type
				typedef encapc<1, prop, typename openfpm::vector<prop>::layout_type> encap_src;
				// destination object type
				typedef encapc<1, prp_object, typename openfpm::vector<prp_object>::layout_type> encap_dst;

				// Copy only the selected properties
496
				object_si_d<encap_src, encap_dst, OBJ_ENCAP, prp...>(v_prp.get(g_opart.get(i).template get<0>(j)), g_send_prp.get(i).get(j));
497 498 499 500 501 502 503 504 505 506 507 508 509
			}
		}
	}

	/*! \brief allocate and fill the send buffer for the map function
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particles properties
	 * \param prc_r List of processor rank involved in the send
	 * \param prc_sz_r For each processor in the list the size of the message to send
	 * \param pb send buffer
	 *
	 */
510
	void fill_send_map_buf(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, openfpm::vector<size_t> & prc_sz_r, openfpm::vector<openfpm::vector<Point<dim,St>>> & m_pos, openfpm::vector<openfpm::vector<prop>> & m_prp)
511
	{
512 513 514
		m_prp.resize(prc_sz_r.size());
		m_pos.resize(prc_sz_r.size());
		openfpm::vector<size_t> cnt(prc_sz_r.size());
515

516
		for (size_t i = 0; i < prc_sz_r.size() ; i++)
517 518
		{
			// set the size and allocate, using mem warant that pos and prp is contiguous
519 520 521
			m_pos.get(i).resize(prc_sz_r.get(i));
			m_prp.get(i).resize(prc_sz_r.get(i));
			cnt.get(i) = 0;
522 523
		}

524 525
		// end vector point
		long int id_end = v_pos.size();
526

527 528 529 530 531
		// end opart point
		long int end = m_opart.size()-1;

		// Run through all the particles and fill the sending buffer
		for (size_t i = 0; i < m_opart.size(); i++)
532
		{
533
			process_map_particle<proc_without_prp>(i,end,id_end,m_pos,m_prp,v_pos,v_prp,cnt);
534
		}
535 536 537

		v_pos.resize(v_pos.size() - m_opart.size());
		v_prp.resize(v_prp.size() - m_opart.size());
538 539
	}

540

541 542 543 544 545 546 547 548 549
	/*! \brief allocate and fill the send buffer for the map function
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particle properties
	 * \param prc_r List of processor rank involved in the send
	 * \param prc_sz_r For each processor in the list the size of the message to send
	 * \param pb send buffer
	 *
	 */
550
	template<typename prp_object,int ... prp> void fill_send_map_buf_list(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, openfpm::vector<size_t> & prc_sz_r, openfpm::vector<openfpm::vector<Point<dim,St>>> & m_pos, openfpm::vector<openfpm::vector<prp_object>> & m_prp)
551
	{
552 553 554
		m_prp.resize(prc_sz_r.size());
		m_pos.resize(prc_sz_r.size());
		openfpm::vector<size_t> cnt(prc_sz_r.size());
555

556
		for (size_t i = 0; i < prc_sz_r.size(); i++)
557 558
		{
			// set the size and allocate, using mem warant that pos and prp is contiguous
559 560
			m_pos.get(i).resize(prc_sz_r.get(i));
			m_prp.get(i).resize(prc_sz_r.get(i));
561
			cnt.get(i) = 0;
562 563
		}

564 565 566 567 568
		// end vector point
		long int id_end = v_pos.size();

		// end opart point
		long int end = m_opart.size()-1;
569

570 571
		// Run through all the particles and fill the sending buffer
		for (size_t i = 0; i < m_opart.size(); i++)
572
		{
573
			process_map_particle<proc_with_prp<prp_object,prp...>>(i,end,id_end,m_pos,m_prp,v_pos,v_prp,cnt);
574
		}
575

576 577
		v_pos.resize(v_pos.size() - m_opart.size());
		v_prp.resize(v_prp.size() - m_opart.size());
578 579 580 581 582 583 584 585 586 587
	}

	/*! \brief Label particles for mappings
	 *
	 * \param v_pos vector of particle positions
	 * \param lbl_p Particle labeled
	 * \param prc_sz For each processor the number of particles to send
	 * \param opart id of the particles to send
	 *
	 */
588
	template<typename obp> void labelParticleProcessor(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<aggregate<size_t,size_t,size_t>> & lbl_p, openfpm::vector<size_t> & prc_sz)
589 590
	{
		// reset lbl_p
591
		lbl_p.clear();
592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618

		// resize the label buffer
		prc_sz.resize(v_cl.getProcessingUnits());

		auto it = v_pos.getIterator();

		// Label all the particles with the processor id where they should go
		while (it.isNext())
		{
			auto key = it.get();

			// Apply the boundary conditions
			dec.applyPointBC(v_pos.get(key));

			size_t p_id = 0;

			// Check if the particle is inside the domain
			if (dec.getDomain().isInside(v_pos.get(key)) == true)
				p_id = dec.processorIDBC(v_pos.get(key));
			else
				p_id = obp::out(key, v_cl.getProcessUnitID());

			// Particle to move
			if (p_id != v_cl.getProcessUnitID())
			{
				if ((long int) p_id != -1)
				{
Pietro Incardona's avatar
Pietro Incardona committed
619
					prc_sz.get(p_id)++;
620 621 622
					lbl_p.add();
					lbl_p.last().template get<0>() = key;
					lbl_p.last().template get<2>() = p_id;
623
				}
624 625 626 627 628 629
				else
				{
					lbl_p.add();
					lbl_p.last().template get<0>() = key;
					lbl_p.last().template get<2>() = p_id;
				}
630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
			}

			// Add processors and add size

			++it;
		}
	}

	/*! \brief Label the particles
	 *
	 * It count the number of particle to send to each processors and save its ids
	 *
	 * \see nn_prcs::getShiftvectors()
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particle properties
	 * \param g_m ghost marker
	 *
	 */
649
	void labelParticlesGhost(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, openfpm::vector<size_t> & prc, size_t & g_m)
650 651
	{
		// Buffer that contain for each processor the id of the particle to send
652 653
		g_opart.clear();
		g_opart.resize(dec.getNNProcessors());
incardon's avatar
incardon committed
654
		prc_g_opart.clear();
655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671

		// Iterate over all particles
		auto it = v_pos.getIteratorTo(g_m);
		while (it.isNext())
		{
			auto key = it.get();

			// Given a particle, it return which processor require it (first id) and shift id, second id
			// For an explanation about shifts vectors please consult getShiftVector in ie_ghost
			const openfpm::vector<std::pair<size_t, size_t>> & vp_id = dec.template ghost_processorID_pair<typename Decomposition::lc_processor_id, typename Decomposition::shift_id>(v_pos.get(key), UNIQUE);

			for (size_t i = 0; i < vp_id.size(); i++)
			{
				// processor id
				size_t p_id = vp_id.get(i).first;

				// add particle to communicate
672 673 674
				g_opart.get(p_id).add();
				g_opart.get(p_id).last().template get<0>() = key;
				g_opart.get(p_id).last().template get<1>() = vp_id.get(i).second;
675 676 677 678
			}

			++it;
		}
679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694

		// remove all zero entry and construct prc (the list of the sending processors)
		openfpm::vector<openfpm::vector<aggregate<size_t,size_t>>> g_opart_f;

		// count the non zero element
		for (size_t i = 0 ; i < g_opart.size() ; i++)
		{
			if (g_opart.get(i).size() != 0)
			{
				g_opart_f.add();
				g_opart.get(i).swap(g_opart_f.last());
				prc.add(dec.IDtoProc(i));
			}
		}

		g_opart.swap(g_opart_f);
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722
	}

	/*! \brief Call-back to allocate buffer to receive incoming elements (particles)
	 *
	 * \param msg_i size required to receive the message from i
	 * \param total_msg total size to receive from all the processors
	 * \param total_p the total number of processor that want to communicate with you
	 * \param i processor id
	 * \param ri request id (it is an id that goes from 0 to total_p, and is unique
	 *           every time message_alloc is called)
	 * \param ptr a pointer to the vector_dist structure
	 *
	 * \return the pointer where to store the message for the processor i
	 *
	 */
	static void * message_alloc_map(size_t msg_i, size_t total_msg, size_t total_p, size_t i, size_t ri, void * ptr)
	{
		// cast the pointer
		vector_dist_comm<dim, St, prop, Decomposition, Memory> * vd = static_cast<vector_dist_comm<dim, St, prop, Decomposition, Memory> *>(ptr);

		vd->recv_mem_gm.resize(vd->v_cl.getProcessingUnits());
		vd->recv_mem_gm.get(i).resize(msg_i);

		return vd->recv_mem_gm.get(i).getPointer();
	}

public:

incardon's avatar
incardon committed
723 724 725 726 727 728
	/*! \brief Copy Constructor
	 *
	 * \param v vector to copy
	 *
	 */
	vector_dist_comm(const vector_dist_comm<dim,St,prop,Decomposition,Memory> & v)
incardon's avatar
incardon committed
729
	:v_cl(create_vcluster()),dec(create_vcluster()),lg_m(0)
incardon's avatar
incardon committed
730 731 732 733 734
	{
		this->operator=(v);
	}


735 736 737 738 739 740
	/*! \brief Constructor
	 *
	 * \param dec Domain decompositon
	 *
	 */
	vector_dist_comm(const Decomposition & dec)
incardon's avatar
incardon committed
741
	:v_cl(create_vcluster()),dec(dec),lg_m(0)
742 743 744 745 746 747 748 749 750 751
	{

	}

	/*! \brief Constructor
	 *
	 * \param dec Domain decompositon
	 *
	 */
	vector_dist_comm(Decomposition && dec)
Pietro Incardona's avatar
Pietro Incardona committed
752
	:v_cl(create_vcluster()),dec(dec)
753 754 755 756 757 758 759 760
	{

	}

	/*! \brief Constructor
	 *
	 */
	vector_dist_comm()
incardon's avatar
incardon committed
761
	:v_cl(create_vcluster()),dec(create_vcluster()),lg_m(0)
762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815
	{
	}

	/*! \brief Get the number of minimum sub-domain
	 *
	 * \return minimum number
	 *
	 */
	static size_t getDefaultNsubsub()
	{
		return V_SUB_UNIT_FACTOR;
	}

	/*! \brief Initialize the decomposition
	 *
	 * \param box domain
	 * \param bc boundary conditions
	 * \param g ghost extension
	 *
	 */
	void init_decomposition(Box<dim,St> & box, const size_t (& bc)[dim],const Ghost<dim,St> & g)
	{
		// Create a valid decomposition of the space
		// Get the number of processor and calculate the number of sub-domain
		// for decomposition
		size_t n_proc = v_cl.getProcessingUnits();
		size_t n_sub = n_proc * getDefaultNsubsub();

		// Calculate the maximum number (before merging) of sub-domain on
		// each dimension
		size_t div[dim];
		for (size_t i = 0; i < dim; i++)
		{
			div[i] = openfpm::math::round_big_2(pow(n_sub, 1.0 / dim));
		}

		// Create the sub-domains
		dec.setParameters(div, box, bc, g);
		dec.decompose();
	}

	/*! \brief It synchronize the properties and position of the ghost particles
	 *
	 * \tparam prp list of properties to get synchronize
	 *
	 * \param opt options WITH_POSITION, it send also the positional information of the particles
	 * \param v_pos vector of position to update
	 * \param v_prp vector of properties to update
	 * \param g_m marker between real and ghost particles
	 *
	 */
	template<int ... prp> inline void ghost_get_(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, size_t & g_m, size_t opt = WITH_POSITION)
	{
		// Unload receive buffer
816 817
		for (size_t i = 0 ; i < recv_sz_get.size() ; i++)
			recv_sz_get.get(i) = 0;
818 819 820 821 822

		// Sending property object
		typedef object<typename object_creator<typename prop::type, prp...>::type> prp_object;

		// send vector for each processor
823
		typedef openfpm::vector<prp_object> send_vector;
824

incardon's avatar
incardon committed
825 826
		v_pos.resize(g_m);

827
		// reset the ghost part
incardon's avatar
incardon committed
828 829 830 831
		if (opt != SKIP_LABELLING)
		{
			v_prp.resize(g_m);
		}
832 833 834

		// Label all the particles
		if ((opt & SKIP_LABELLING) == false)
incardon's avatar
incardon committed
835
			labelParticlesGhost(v_pos,v_prp,prc_g_opart,g_m);
836

837
		// Send and receive ghost particle information
838
		openfpm::vector<send_vector> g_send_prp;
839
		fill_send_ghost_prp_buf<send_vector, prp_object, prp...>(v_prp,g_send_prp);
840 841 842

		// Create and fill the send buffer for the particle position
		if (opt != NO_POSITION)
843
			fill_send_ghost_pos_buf(v_pos,g_pos_send);
844

845 846
		prc_recv_get.clear();
		recv_sz_get.clear();
incardon's avatar
incardon committed
847 848 849 850 851 852 853 854 855

		if (opt == SKIP_LABELLING)
		{
			op_ssend_gg_recv_merge opm(g_m);
//			v_cl.SSendRecvP_op<op_ssend_recv_merge<op>
			v_cl.SSendRecvP_op<op_ssend_gg_recv_merge,send_vector,decltype(v_prp),prp...>(g_send_prp,v_prp,prc_g_opart,opm,prc_recv_get,recv_sz_get);
		}
		else
			v_cl.SSendRecvP<send_vector,decltype(v_prp),prp...>(g_send_prp,v_prp,prc_g_opart,prc_recv_get,recv_sz_get);
856 857 858

		if (opt != NO_POSITION)
		{
859 860
			prc_recv_get.clear();
			recv_sz_get.clear();
incardon's avatar
incardon committed
861
			v_cl.SSendRecv(g_pos_send,v_pos,prc_g_opart,prc_recv_get,recv_sz_get);
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897
		}

		add_loc_particles_bc(v_pos,v_prp,g_m,opt);
	}


	/*! \brief It move all the particles that does not belong to the local processor to the respective processor
	 *
	 * \tparam out of bound policy it specify what to do when the particles are detected out of bound
	 *
	 * In general this function is called after moving the particles to move the
	 * elements out the local processor. Or just after initialization if each processor
	 * contain non local particles
	 *
	 * \tparam prp properties to communicate
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particle properties
	 * \param g_m ghost marker
	 *
	 */
	template<unsigned int ... prp> void map_list_(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, size_t & g_m)
	{
		typedef KillParticle obp;

		// Processor communication size
		openfpm::vector<size_t> prc_sz(v_cl.getProcessingUnits());

		// It contain the list of the processors this processor should to communicate with
		openfpm::vector<size_t> p_list;

		// map completely reset the ghost part
		v_pos.resize(g_m);
		v_prp.resize(g_m);

		// Contain the processor id of each particle (basically where they have to go)
898
		labelParticleProcessor<obp>(v_pos,m_opart, prc_sz);
899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918

		// Calculate the sending buffer size for each processor, put this information in
		// a contiguous buffer
		p_map_req.resize(v_cl.getProcessingUnits());
		openfpm::vector<size_t> prc_sz_r;
		openfpm::vector<size_t> prc_r;

		for (size_t i = 0; i < v_cl.getProcessingUnits(); i++)
		{
			if (prc_sz.get(i) != 0)
			{
				p_map_req.get(i) = prc_r.size();
				prc_r.add(i);
				prc_sz_r.add(prc_sz.get(i));
			}
		}

		// Sending property object
		typedef object<typename object_creator<typename prop::type, prp...>::type> prp_object;

919 920 921 922
		//! position vector
		openfpm::vector<openfpm::vector<Point<dim, St>>> m_pos;
		//! properties vector
		openfpm::vector<openfpm::vector<prp_object>> m_prp;
923

924
		fill_send_map_buf_list<prp_object,prp...>(v_pos,v_prp,prc_sz_r, m_pos, m_prp);
925

926 927
		v_cl.SSendRecv(m_pos,v_pos,prc_r,prc_recv_map,recv_sz_map);
		v_cl.SSendRecvP<openfpm::vector<prp_object>,decltype(v_prp),prp...>(m_prp,v_prp,prc_r,prc_recv_map,recv_sz_map);
928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959

		// mark the ghost part

		g_m = v_pos.size();
	}

	/*! \brief It move all the particles that does not belong to the local processor to the respective processor
	 *
	 * \tparam out of bound policy it specify what to do when the particles are detected out of bound
	 *
	 * In general this function is called after moving the particles to move the
	 * elements out the local processor. Or just after initialization if each processor
	 * contain non local particles
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector of particle properties
	 * \param g_m ghost marker
	 *
	 */
	template<typename obp = KillParticle> void map_(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, size_t & g_m)
	{
		// Processor communication size
		openfpm::vector<size_t> prc_sz(v_cl.getProcessingUnits());

		// It contain the list of the processors this processor should to communicate with
		openfpm::vector<size_t> p_list;

		// map completely reset the ghost part
		v_pos.resize(g_m);
		v_prp.resize(g_m);

		// Contain the processor id of each particle (basically where they have to go)
960
		labelParticleProcessor<obp>(v_pos,m_opart, prc_sz);
961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977

		// Calculate the sending buffer size for each processor, put this information in
		// a contiguous buffer
		p_map_req.resize(v_cl.getProcessingUnits());
		openfpm::vector<size_t> prc_sz_r;
		openfpm::vector<size_t> prc_r;

		for (size_t i = 0; i < v_cl.getProcessingUnits(); i++)
		{
			if (prc_sz.get(i) != 0)
			{
				p_map_req.get(i) = prc_r.size();
				prc_r.add(i);
				prc_sz_r.add(prc_sz.get(i));
			}
		}

978 979 980 981
		//! position vector
		openfpm::vector<openfpm::vector<Point<dim, St>>> m_pos;
		//! properties vector
		openfpm::vector<openfpm::vector<prop>> m_prp;
982

983
		fill_send_map_buf(v_pos,v_prp, prc_sz_r, m_pos, m_prp);
984

985 986
		v_cl.SSendRecv(m_pos,v_pos,prc_r,prc_recv_map,recv_sz_map);
		v_cl.SSendRecv(m_prp,v_prp,prc_r,prc_recv_map,recv_sz_map);
987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002

		// mark the ghost part

		g_m = v_pos.size();
	}

	/*! \brief Get the decomposition
	 *
	 * \return
	 *
	 */
	inline Decomposition & getDecomposition()
	{
		return dec;
	}

incardon's avatar
incardon committed
1003 1004 1005 1006 1007 1008 1009 1010 1011 1012
	/*! \brief Get the decomposition
	 *
	 * \return
	 *
	 */
	inline const Decomposition & getDecomposition() const
	{
		return dec;
	}

1013 1014 1015 1016
	/*! \brief Copy a vector
	 *
	 * \param vc vector to copy
	 *
Pietro Incardona's avatar
Pietro Incardona committed
1017 1018
	 * \return iteself
	 *
1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030
	 */
	vector_dist_comm<dim,St,prop,Decomposition,Memory> & operator=(const vector_dist_comm<dim,St,prop,Decomposition,Memory> & vc)
	{
		dec = vc.dec;

		return *this;
	}

	/*! \brief Copy a vector
	 *
	 * \param vc vector to copy
	 *
Pietro Incardona's avatar
Pietro Incardona committed
1031 1032
	 * \return itself
	 *
1033 1034 1035
	 */
	vector_dist_comm<dim,St,prop,Decomposition,Memory> & operator=(vector_dist_comm<dim,St,prop,Decomposition,Memory> && vc)
	{
1036
		dec = vc.dec;
1037 1038 1039

		return *this;
	}
Pietro Incardona's avatar
Pietro Incardona committed
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056

	/*! \brief Ghost put
	 *
	 * \tparam op operation to apply
	 * \tparam prp set of properties
	 *
	 * \param v_pos vector of particle positions
	 * \param v_prp vector od particle properties
	 * \param g_m ghost marker
	 *
	 */
	template<template<typename,typename> class op, int ... prp> void ghost_put_(openfpm::vector<Point<dim, St>> & v_pos, openfpm::vector<prop> & v_prp, size_t & g_m)
	{
		// Sending property object
		typedef object<typename object_creator<typename prop::type, prp...>::type> prp_object;

		// send vector for each processor
1057
		typedef openfpm::vector<prp_object> send_vector;
Pietro Incardona's avatar
Pietro Incardona committed
1058 1059

		openfpm::vector<send_vector> g_send_prp;
1060
		fill_send_ghost_put_prp_buf<send_vector, prp_object, prp...>(v_prp,g_send_prp,g_m);
Pietro Incardona's avatar
Pietro Incardona committed
1061

1062 1063 1064
		// Send and receive ghost particle information
		op_ssend_recv_merge<op> opm(g_opart);
		v_cl.SSendRecvP_op<op_ssend_recv_merge<op>,send_vector,decltype(v_prp),prp...>(g_send_prp,v_prp,prc_recv_get,opm,prc_recv_put,recv_sz_put);
1065 1066 1067 1068 1069 1070 1071

		// process also the local replicated particles

		size_t i2 = 0;


		if (v_prp.size() - lg_m != o_part_loc.size())
incardon's avatar
incardon committed
1072 1073 1074 1075
		{
			std::cerr << "Error: " << __FILE__ << ":" << __LINE__ << " Local ghost particles = " << v_prp.size() - lg_m << " != " << o_part_loc.size() << std::endl;
			std::cerr << "Error: " << __FILE__ << ":" << __LINE__ << "Check that you did a ghost_get before a ghost_put" << std::endl;
		}
1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086


		for (size_t i = lg_m ; i < v_prp.size() ; i++)
		{
			auto dst = v_prp.get(o_part_loc.template get<0>(i2));
			auto src = v_prp.get(i);
			copy_cpu_encap_encap_op_prp<op,decltype(v_prp.get(0)),decltype(v_prp.get(0)),prp...> cp(src,dst);

			boost::mpl::for_each_ref< boost::mpl::range_c<int,0,sizeof...(prp)> >(cp);
			i2++;
		}
Pietro Incardona's avatar
Pietro Incardona committed
1087
	}
1088 1089 1090 1091
};


#endif /* SRC_VECTOR_VECTOR_DIST_COMM_HPP_ */