CartDecomposition.hpp 43.3 KB
Newer Older
incardon's avatar
incardon committed
1 2 3
/*
 * CartDecomposition.hpp
 *
4
 *  Created on: Oct 07, 2015
5
 *      Author: Pietro Incardona, Antonio Leo
incardon's avatar
incardon committed
6 7 8 9 10 11
 */

#ifndef CARTDECOMPOSITION_HPP
#define CARTDECOMPOSITION_HPP

#include "config.h"
12
#include <cmath>
13 14
#include "VCluster.hpp"
#include "Graph/CartesianGraphFactory.hpp"
tonynsyde's avatar
tonynsyde committed
15
#include "Graph/DistCartesianGraphFactory.hpp"
incardon's avatar
incardon committed
16
#include "Decomposition.hpp"
incardon's avatar
incardon committed
17
#include "Vector/map_vector.hpp"
incardon's avatar
incardon committed
18 19 20 21 22
#include <vector>
#include <initializer_list>
#include "SubdomainGraphNodes.hpp"
#include "dec_optimizer.hpp"
#include "Space/Shape/Box.hpp"
incardon's avatar
incardon committed
23
#include "Space/Shape/Point.hpp"
incardon's avatar
incardon committed
24
#include "NN/CellList/CellDecomposer.hpp"
incardon's avatar
incardon committed
25 26
#include <unordered_map>
#include "NN/CellList/CellList.hpp"
incardon's avatar
incardon committed
27
#include "Space/Ghost.hpp"
incardon's avatar
incardon committed
28 29
#include "common.hpp"
#include "ie_loc_ghost.hpp"
30 31
#include "ie_ghost.hpp"
#include "nn_processor.hpp"
32
#include "GraphMLWriter/GraphMLWriter.hpp"
33
#include "ParMetisDistribution.hpp"
tonynsyde's avatar
tonynsyde committed
34
#include "DistParMetisDistribution.hpp"
35 36
#include "MetisDistribution.hpp"
#include "DLB.hpp"
37
#include "util/se_util.hpp"
38
#include "util/mathutil.hpp"
incardon's avatar
incardon committed
39

40 41
#define CARTDEC_ERROR 2000lu

incardon's avatar
incardon committed
42 43 44 45 46 47
/**
 * \brief This class decompose a space into subspaces
 *
 * \tparam dim is the dimensionality of the physical domain we are going to decompose.
 * \tparam T type of the space we decompose, Real, Integer, Complex ...
 * \tparam Memory Memory factory used to allocate memory
tonynsyde's avatar
tonynsyde committed
48
 * \tparam Distribution type of distribution, can be ParMetisDistribution or MetisDistribution
incardon's avatar
incardon committed
49
 *
50
 * Given an N-dimensional space, this class decompose the space into a Cartesian grid of small
incardon's avatar
incardon committed
51 52 53 54
 * sub-sub-domain. To each sub-sub-domain is assigned an id that identify at which processor is
 * assigned (in general the union of all the sub-sub-domain assigned to a processor is
 * simply connected space), a second step merge several sub-sub-domain with same id into bigger region
 *  sub-domain. Each sub-domain has an extended space called ghost part
55 56 57 58
 *
 * Assuming that VCluster.getProcessUnitID(), equivalent to the MPI processor rank, return the processor local
 * processor id, we define
 *
59 60
 * * local processor: processor rank
 * * local sub-domain: sub-domain given to the local processor
61 62 63 64 65 66 67 68 69 70 71 72
 * * external ghost box: (or ghost box) are the boxes that compose the ghost space of the processor, or the
 *   boxes produced expanding every local sub-domain by the ghost extension and intersecting with the sub-domain
 *   of the other processors
 * * Near processors are the processors adjacent to the local processor, where with adjacent we mean all the processor
 *   that has a non-zero intersection with the ghost part of the local processor, or all the processors that
 *   produce non-zero external boxes with the local processor, or all the processor that should communicate
 *   in case of ghost data synchronization
 * * internal ghost box: is the part of ghost of the near processor that intersect the space of the
 *       processor, or the boxes produced expanding the sub-domain of the near processors with the local sub-domain
 * * Near processor sub-domain: is a sub-domain that live in the a near (or contiguous) processor
 * * Near processor list: the list of all the near processor of the local processor (each processor has a list
 *                        of the near processor)
incardon's avatar
incardon committed
73
 * * Local ghosts interal or external are all the ghosts that does not involve inter-processor communications
74 75
 *
 * \see calculateGhostBoxes() for a visualization of internal and external ghost boxes
incardon's avatar
incardon committed
76
 *
77 78 79
 * ### Create a Cartesian decomposition object on a Box space, distribute, calculate internal and external ghost boxes
 * \snippet CartDecomposition_unit_test.hpp Create CartDecomposition
 *
incardon's avatar
incardon committed
80 81
 */

82
template<unsigned int dim, typename T, typename Memory = HeapMemory, typename Distribution = ParMetisDistribution<dim, T>>
83
class CartDecomposition: public ie_loc_ghost<dim, T>, public nn_prcs<dim, T>, public ie_ghost<dim, T>
incardon's avatar
incardon committed
84
{
incardon's avatar
incardon committed
85

incardon's avatar
incardon committed
86
public:
incardon's avatar
incardon committed
87

incardon's avatar
incardon committed
88 89 90 91
	//! Type of the domain we are going to decompose
	typedef T domain_type;

	//! It simplify to access the SpaceBox element
92
	typedef SpaceBox<dim, T> Box;
incardon's avatar
incardon committed
93 94 95

private:

incardon's avatar
incardon committed
96
	//! This is the key type to access  data_s, for example in the case of vector
incardon's avatar
incardon committed
97
	//! acc_key is size_t
tonynsyde's avatar
tonynsyde committed
98
	typedef typename openfpm::vector<SpaceBox<dim, T>, Memory, openfpm::vector_grow_policy_default, openfpm::vect_isel<SpaceBox<dim, T>>::value>::access_key acc_key;
incardon's avatar
incardon committed
99 100

	//! the set of all local sub-domain as vector
101
	openfpm::vector<SpaceBox<dim, T>> sub_domains;
incardon's avatar
incardon committed
102

incardon's avatar
incardon committed
103
	//! for each sub-domain, contain the list of the neighborhood processors
incardon's avatar
incardon committed
104 105
	openfpm::vector<openfpm::vector<long unsigned int> > box_nn_processor;

incardon's avatar
incardon committed
106
	//! Structure that contain for each sub-sub-domain box the processor id
incardon's avatar
Add ORB  
incardon committed
107
	//! exist for efficient global communication
incardon's avatar
incardon committed
108 109
	openfpm::vector<size_t> fine_s;

incardon's avatar
incardon committed
110
	//! Structure that store the cartesian grid information
111
	grid_sm<dim, void> gr;
incardon's avatar
incardon committed
112

incardon's avatar
incardon committed
113 114
	//! Structure that decompose your structure into cell without creating them
	//! useful to convert positions to CellId or sub-domain id in this case
115
	CellDecomposer_sm<dim, T> cd;
incardon's avatar
incardon committed
116 117

	//! rectangular domain to decompose
Pietro Incardona's avatar
Pietro Incardona committed
118
	::Box<dim,T> domain;
incardon's avatar
incardon committed
119 120 121 122 123 124 125

	//! Box Spacing
	T spacing[dim];

	//! Runtime virtual cluster machine
	Vcluster & v_cl;

126
	//! Create distribution
127 128
	Distribution dist;

incardon's avatar
incardon committed
129 130 131 132
	// Smallest subdivision on each direction
	::Box<dim,T> ss_box;

	::Box<dim,T> bbox;
incardon's avatar
incardon committed
133

134 135 136 137 138 139 140 141 142
	// Heap memory receiver
	HeapMemory hp_recv;

	// vector v_proc
	openfpm::vector<size_t> v_proc;

	// reference counter of the object in case is shared between object
	long int ref_cnt;

Pietro Incardona's avatar
Pietro Incardona committed
143
	// ghost info
144 145
	Ghost<dim,T> ghost;

Pietro Incardona's avatar
Pietro Incardona committed
146 147
	// Boundary condition info
	size_t bc[dim];
incardon's avatar
incardon committed
148

149
	/*! \brief Constructor, it decompose and distribute the sub-domains across the processors
incardon's avatar
incardon committed
150
	 *
151 152
	 * \param v_cl Virtual cluster, used internally for communications
	 *
incardon's avatar
incardon committed
153
	 */
Pietro Incardona's avatar
Pietro Incardona committed
154
	void createSubdomains(Vcluster & v_cl, const size_t (& bc)[dim])
incardon's avatar
incardon committed
155
	{
156 157 158 159 160 161 162
#ifdef SE_CLASS1
		if (&v_cl == NULL)
		{
			std::cerr << __FILE__ << ":" << __LINE__ << " error VCluster instance is null, check that you ever initialized it \n";
			ACTION_ON_ERROR()
		}
#endif
163 164 165

		int p_id = v_cl.getProcessUnitID();

incardon's avatar
incardon committed
166 167 168
		// Calculate the total number of box and and the spacing
		// on each direction
		// Get the box containing the domain
169
		SpaceBox<dim, T> bs = domain.getBox();
incardon's avatar
incardon committed
170

171
		for (unsigned int i = 0; i < dim; i++)
incardon's avatar
incardon committed
172 173
		{
			// Calculate the spacing
incardon's avatar
incardon committed
174
			spacing[i] = (bs.getHigh(i) - bs.getLow(i)) / gr.size(i);
incardon's avatar
incardon committed
175 176
		}

incardon's avatar
Add ORB  
incardon committed
177
		// fill the structure that store the processor id for each sub-domain
incardon's avatar
incardon committed
178
		fine_s.resize(gr.size());
incardon's avatar
Add ORB  
incardon committed
179

incardon's avatar
incardon committed
180 181
		// Optimize the decomposition creating bigger spaces
		// And reducing Ghost over-stress
182
		dec_optimizer<dim, Graph_CSR<nm_v, nm_e>> d_o(dist.getGraph(), gr.getSize());
incardon's avatar
incardon committed
183 184

		// set of Boxes produced by the decomposition optimizer
185
		openfpm::vector<::Box<dim, size_t>> loc_box;
incardon's avatar
incardon committed
186

incardon's avatar
incardon committed
187
		// optimize the decomposition
188
		d_o.template optimize<nm_v::sub_id, nm_v::proc_id>(dist.getGraph(), p_id, loc_box, box_nn_processor,bc);
incardon's avatar
incardon committed
189

incardon's avatar
incardon committed
190 191 192
		// Initialize ss_box and bbox
		if (loc_box.size() >= 0)
		{
193 194
			SpaceBox<dim, size_t> sub_dc = loc_box.get(0);
			SpaceBox<dim, T> sub_d(sub_dc);
incardon's avatar
incardon committed
195 196 197
			sub_d.mul(spacing);
			sub_d.expand(spacing);

198 199 200 201 202
			// Fixing sub-domains to cover all the domain

			// Fixing sub_d
			// if (loc_box) is a the boundary we have to ensure that the box span the full
			// domain (avoiding rounding off error)
203
			for (size_t i = 0; i < dim; i++)
204 205 206
			{
				if (sub_dc.getHigh(i) == cd.getGrid().size(i) - 1)
				{
207
					sub_d.setHigh(i, domain.getHigh(i));
208 209 210
				}
			}

incardon's avatar
incardon committed
211 212 213 214
			// add the sub-domain
			sub_domains.add(sub_d);

			ss_box = sub_d;
incardon's avatar
incardon committed
215
			ss_box -= ss_box.getP1();
incardon's avatar
incardon committed
216 217 218
			bbox = sub_d;
		}

incardon's avatar
incardon committed
219
		// convert into sub-domain
220
		for (size_t s = 1; s < loc_box.size(); s++)
incardon's avatar
incardon committed
221
		{
222 223
			SpaceBox<dim, size_t> sub_dc = loc_box.get(s);
			SpaceBox<dim, T> sub_d(sub_dc);
incardon's avatar
incardon committed
224

incardon's avatar
incardon committed
225 226 227
			// re-scale and add spacing (the end is the starting point of the next domain + spacing)
			sub_d.mul(spacing);
			sub_d.expand(spacing);
incardon's avatar
incardon committed
228

229 230 231 232 233
			// Fixing sub-domains to cover all the domain

			// Fixing sub_d
			// if (loc_box) is a the boundary we have to ensure that the box span the full
			// domain (avoiding rounding off error)
234
			for (size_t i = 0; i < dim; i++)
235 236 237
			{
				if (sub_dc.getHigh(i) == cd.getGrid().size(i) - 1)
				{
238
					sub_d.setHigh(i, domain.getHigh(i));
239 240 241
				}
			}

incardon's avatar
incardon committed
242 243
			// add the sub-domain
			sub_domains.add(sub_d);
incardon's avatar
incardon committed
244 245 246 247 248 249

			// Calculate the bound box
			bbox.enclose(sub_d);

			// Create the smallest box contained in all sub-domain
			ss_box.contained(sub_d);
incardon's avatar
incardon committed
250
		}
incardon's avatar
incardon committed
251

252
		nn_prcs<dim,T>::create(box_nn_processor, sub_domains);
incardon's avatar
incardon committed
253
		nn_prcs<dim,T>::refine_ss_box(ss_box);
Pietro Incardona's avatar
Pietro Incardona committed
254
		nn_prcs<dim,T>::applyBC(domain,ghost,bc);
255

incardon's avatar
incardon committed
256
		// fill fine_s structure
incardon's avatar
incardon committed
257 258 259
		// fine_s structure contain the processor id for each sub-sub-domain
		// with sub-sub-domain we mean the sub-domain decomposition before
		// running dec_optimizer (before merging sub-domains)
260
		auto it = dist.getGraph().getVertexIterator();
incardon's avatar
incardon committed
261 262 263 264 265 266

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

			// fill with the fine decomposition
267
			fine_s.get(key) = dist.getGraph().template vertex_p<nm_v::proc_id>(key);
incardon's avatar
incardon committed
268 269 270

			++it;
		}
271

incardon's avatar
incardon committed
272 273 274 275 276 277 278 279 280 281
		Initialize_geo_cell_lists();
	}

	/*! \brief Initialize geo_cell lists
	 *
	 *
	 *
	 */
	void Initialize_geo_cell_lists()
	{
282
		// Get the smallest sub-division on each direction
283
		::Box<dim, T> unit = getSmallestSubdivision();
284 285
		// Get the processor bounding Box
		::Box<dim,T> bound = getProcessorBounds();
Pietro Incardona's avatar
Pietro Incardona committed
286 287
		// Not necessary, but I prefer
		bound.enlarge(ghost);
288

incardon's avatar
incardon committed
289
		// calculate the sub-divisions
290
		size_t div[dim];
291 292
		for (size_t i = 0; i < dim; i++)
			div[i] = (size_t) ((bound.getHigh(i) - bound.getLow(i)) / unit.getHigh(i));
293 294

		// Create shift
295
		Point<dim, T> orig;
296 297

		// p1 point of the Processor bound box is the shift
298
		for (size_t i = 0; i < dim; i++)
299 300 301
			orig.get(i) = bound.getLow(i);

		// Initialize the geo_cell structure
Pietro Incardona's avatar
Pietro Incardona committed
302
		ie_ghost<dim,T>::Initialize_geo_cell(bound,div,orig);
303

304 305
		// Initialize shift vectors
		ie_ghost<dim,T>::generateShiftVectors(domain);
incardon's avatar
incardon committed
306 307
	}

308 309 310 311 312 313
	/*! \brief Calculate communication and migration costs
	 *
	 * \param ts how many timesteps have passed since last calculation, used to approximate the cost
	 */
	void computeCommunicationAndMigrationCosts(size_t ts)
	{
tonynsyde's avatar
tonynsyde committed
314
		float migration = 0;
315 316 317 318 319 320 321 322 323

		SpaceBox<dim, T> cellBox = cd.getCellBox();
		float b_s = cellBox.getHigh(0);
		float gh_s = ghost.getHigh(0);

		// compute the gh_area for 2 dim case
		float gh_v = (gh_s * b_s);

		// multiply for sub-sub-domain side for each domain
324
		for (size_t i = 2; i < dim; i++)
325 326 327 328 329 330 331 332 333 334 335 336 337 338
			gh_v *= b_s;

		size_t norm = (size_t) (1.0 / gh_v);

		migration = pow(b_s, dim);

		size_t prev = 0;

		for (size_t i = 0; i < dist.getNSubSubDomains(); i++)
		{
			dist.setMigrationCost(i, norm * migration * dist.getVertexWeight(i));

			for (size_t s = 0; s < dist.getNSubSubDomainNeighbors(i); s++)
			{
tonynsyde's avatar
tonynsyde committed
339
				dist.setCommunicationCost(i, s, 1 * dist.getVertexWeight(i) * ts);
340 341 342 343
			}
			prev += dist.getNSubSubDomainNeighbors(i);
		}
	}
incardon's avatar
incardon committed
344

incardon's avatar
incardon committed
345 346 347 348 349 350
	/*! \brief Create the subspaces that decompose your domain
	 *
	 */
	void CreateSubspaces()
	{
		// Create a grid where each point is a space
351
		grid_sm<dim, void> g(div);
incardon's avatar
incardon committed
352 353 354 355 356 357 358 359 360 361 362

		// create a grid_key_dx iterator
		grid_key_dx_iterator<dim> gk_it(g);

		// Divide the space into subspaces
		while (gk_it.isNext())
		{
			//! iterate through all subspaces
			grid_key_dx<dim> key = gk_it.get();

			//! Create a new subspace
363
			SpaceBox<dim, T> tmp;
incardon's avatar
incardon committed
364 365

			//! fill with the Margin of the box
366
			for (int i = 0; i < dim; i++)
incardon's avatar
incardon committed
367
			{
368 369
				tmp.setHigh(i, (key.get(i) + 1) * spacing[i]);
				tmp.setLow(i, key.get(i) * spacing[i]);
incardon's avatar
incardon committed
370 371 372 373 374 375 376 377 378 379
			}

			//! add the space box
			sub_domains.add(tmp);

			// add the iterator
			++gk_it;
		}
	}

incardon's avatar
incardon committed
380

381 382 383 384 385 386 387 388 389 390 391 392
	/*! \brief It copy the sub-domains into another CartesianDecomposition object extending them
	 *
	 * \see duplicate (in case of extended domain)
	 *
	 * \param cart Cartesian decomposition object
	 * \param box Extended domain
	 *
	 */
	void extend_subdomains(CartDecomposition<dim,T> & cart, const ::Box<dim,T> & ext_dom) const
	{
		// Box
		typedef ::Box<dim,T> b;
incardon's avatar
incardon committed
393

394 395
		cart.bbox = ext_dom;
		cart.ss_box = ext_dom;
incardon's avatar
incardon committed
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
		for (size_t i = 0 ; i < sub_domains.size() ; i++)
		{
			::Box<dim,T> box;

			// Calculate the extended box
			for (size_t j = 0 ; j < dim ; j++)
			{
				if (sub_domains.template get<b::p1>(i)[j] == domain.getLow(j))
					box.setLow(j,ext_dom.getLow(j));
				else
					box.setLow(j,sub_domains.template get<b::p1>(i)[j]);

				if (sub_domains.template get<b::p2>(i)[j] == domain.getHigh(j))
					box.setHigh(j,ext_dom.getHigh(j));
				else
					box.setHigh(j,sub_domains.template get<b::p2>(i)[j]);
			}

			// add the subdomain
			cart.sub_domains.add(box);

			// Calculate the bound box
			cart.bbox.enclose(box);

			// Create the smallest box contained in all sub-domain
			cart.ss_box.contained(box);
		}
	}

	/*! \brief Extend the fines for the new Cartesian decomposition
	 *
	 * \param new_fines extended fine_s
	 * \param old_fines old fine_s
	 *
	 */
	void extend_fines(CartDecomposition<dim,T> & cart) const
	{
		// Extension, first we calculate the extensions of the new domain compared
		// to the old one in cell units (each cell unit is a sub-sub-domain)
		::Box<dim,size_t> ext;
		// Extension of the new fines structure
		::Box<dim,size_t> n_fines_ext;
		// Extension of the old fines structure
		::Box<dim,size_t> o_fines_ext;

		size_t sz_new[dim];
		size_t sz_old[dim];

		for (size_t i = 0; i < dim ; i++)
		{
			size_t p1 = (domain.getLow(i) - this->domain.getLow(i)) / cd.getCellBox().getHigh(i) + 1;
			size_t p2 = (domain.getLow(i) - this->domain.getLow(i)) / cd.getCellBox().getHigh(i) + 1;

			ext.setLow(i,p1);
			ext.setHigh(i,p2);
			sz_new[i] = p1+p2+cd.getGrid().size(i);
			sz_old[i] = cd.getGrid().size(i);
		}

		grid_sm<dim,void> info_new(sz_new);
		grid_sm<dim,void> info_old(sz_old);

		// resize the new fines
		cart.fine_s.resize(info_new.size());

		// we create an iterator that iterate across the full new fines
		grid_key_dx_iterator<dim> fines_t(info_new);

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

			// new_fines is bigger than old_fines structure
			// out of bound key must be adjusted
			// The adjustment produce a natural extension
			// a representation can be seen in the figure of
			// CartDecomposition duplicate function with extended domains

			grid_key_dx<dim> key_old;
			for (size_t i = 0 ; i < dim ; i++)
			{
				key_old.set_d(i,(long int)key.get(i) - ext.getLow(i));
				if (key_old.get(i) < 0)
					key_old.set_d(i,0);
				else if(key_old.get(i) >= (long int)info_old.size(i) )
					key_old.set_d(i,info_old.size(i)-1);
			}

			cart.fine_s.get(info_new.LinId(key)) = fine_s.get(info_old.LinId(key_old));

			++fines_t;
		}

		cart.gr.setDimensions(sz_new);

		// the new extended CellDecomposer must be consistent with the old cellDecomposer.
		cart.cd.setDimensions(cd,ext);
	}
incardon's avatar
incardon committed
495

incardon's avatar
incardon committed
496 497
public:

498 499 500 501
	static constexpr int dims = dim;

	typedef T stype;

incardon's avatar
incardon committed
502 503 504 505 506 507 508 509 510 511 512 513 514 515
	//! Increment the reference counter
	void incRef()
	{ref_cnt++;}

	//! Decrement the reference counter
	void decRef()
	{ref_cnt--;}

	//! Return the reference counter
	long int ref()
	{
		return ref_cnt;
	}

incardon's avatar
incardon committed
516 517
	/*! \brief Cartesian decomposition constructor
	 *
518
	 * \param v_cl Virtual cluster, used internally to handle or pipeline communication
incardon's avatar
incardon committed
519 520
	 *
	 */
521
	CartDecomposition(Vcluster & v_cl) :
522
			nn_prcs<dim, T>(v_cl), v_cl(v_cl), dist(v_cl),ref_cnt(0)
incardon's avatar
incardon committed
523 524 525 526
	{
		// Reset the box to zero
		bbox.zero();
	}
incardon's avatar
incardon committed
527

528 529 530 531 532
	/*! \brief Cartesian decomposition copy constructor
	 *
     * \param cart object to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
533
	CartDecomposition(const CartDecomposition<dim,T,Memory> & cart)
534
	:nn_prcs<dim,T>(cart.v_cl),v_cl(cart.v_cl),dist(v_cl),ref_cnt(0)
535 536 537 538 539 540 541 542 543
	{
		this->operator=(cart);
	}

	/*! \brief Cartesian decomposition copy constructor
	 *
     * \param cart object to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
544
	CartDecomposition(CartDecomposition<dim,T,Memory> && cart)
545
	:nn_prcs<dim,T>(cart.v_cl),v_cl(cart.v_cl),dist(v_cl),ref_cnt(0)
546 547 548 549
	{
		this->operator=(cart);
	}

incardon's avatar
incardon committed
550 551
	//! Cartesian decomposition destructor
	~CartDecomposition()
552 553
	{
	}
incardon's avatar
incardon committed
554

555 556 557 558 559 560 561 562 563 564 565 566 567 568
	/*! \brief class to select the returned id by ghost_processorID
	 *
	 */
	class box_id
	{
	public:
		/*! \brief Return the box id
		 *
		 * \param p structure containing the id informations
		 * \param b_id box_id
		 *
		 * \return box id
		 *
		 */
569
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
		{
			return b_id;
		}
	};

	/*! \brief class to select the returned id by ghost_processorID
	 *
	 */
	class processor_id
	{
	public:
		/*! \brief Return the processor id
		 *
		 * \param p structure containing the id informations
		 * \param b_id box_id
		 *
		 * \return processor id
		 *
		 */
589
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
		{
			return p.proc;
		}
	};

	/*! \brief class to select the returned id by ghost_processorID
	 *
	 */
	class lc_processor_id
	{
	public:
		/*! \brief Return the near processor id
		 *
		 * \param p structure containing the id informations
		 * \param b_id box_id
		 *
		 * \return local processor id
		 *
		 */
609
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
610 611 612 613 614
		{
			return p.lc_proc;
		}
	};

615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
	/*! \brief class to select the returned id by ghost_processorID
	 *
	 */
	class shift_id
	{
	public:
		/*! \brief Return the shift id
		 *
		 * \param p structure containing the id informations
		 * \param b_id box_id
		 *
		 * \return shift_id id
		 *
		 */
		inline static size_t id(p_box<dim,T> & p, size_t b_id)
		{
			return p.shift_id;
		}
	};

635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676
	/*! \brief Apply boundary condition to the point
	 *
	 * \param p Point to apply the boundary condition
	 *
	 */
	void applyPointBC(float (& pt)[dim]) const
	{
		for (size_t i = 0 ; i < dim ; i++)
		{
			if (bc[i] == PERIODIC)
				pt[i] = openfpm::math::periodic_l(pt[i],domain.getHigh(i),domain.getLow(i));
		}
	}

	/*! \brief Apply boundary condition to the point
	 *
	 * \param p Point to apply the boundary condition
	 *
	 */
	void applyPointBC(Point<dim,T> & pt) const
	{
		for (size_t i = 0 ; i < dim ; i++)
		{
			if (bc[i] == PERIODIC)
				pt.get(i) = openfpm::math::periodic_l(pt.get(i),domain.getHigh(i),domain.getLow(i));
		}
	}

	/*! \brief Apply boundary condition to the point
	 *
	 * \param encapsulated object
	 *
	 */
	template<typename Mem> void applyPointBC(encapc<1,Point<dim,T>,Mem> && pt) const
	{
		for (size_t i = 0 ; i < dim ; i++)
		{
			if (bc[i] == PERIODIC)
				pt.template get<0>()[i] = openfpm::math::periodic_l(pt.template get<0>()[i],domain.getHigh(i),domain.getLow(i));
		}
	}

incardon's avatar
incardon committed
677
	/*! It calculate the internal ghost boxes
incardon's avatar
incardon committed
678 679 680 681 682
	 *
	 * Example: Processor 10 calculate
	 * B8_0 B9_0 B9_1 and B5_0
	 *
	 *
683 684
	 *
	 \verbatim
incardon's avatar
incardon committed
685 686 687
+----------------------------------------------------+
|                                                    |
|                 Processor 8                        |
incardon's avatar
incardon committed
688
|                 Sub+domain 0                       +-----------------------------------+
incardon's avatar
incardon committed
689 690 691 692 693 694
|                                                    |                                   |
|                                                    |                                   |
++--------------+---+---------------------------+----+        Processor 9                |
 |              |   |     B8_0                  |    |        Subdomain 0                |
 |              +------------------------------------+                                   |
 |              |   |                           |    |                                   |
incardon's avatar
incardon committed
695 696 697 698 699
 |              |   |                           |B9_0|                                   |
 |              | B |    Local processor        |    |                                   |
 | Processor 5  | 5 |    Subdomain 0            |    |                                   |
 | Subdomain 0  | _ |                           +----------------------------------------+
 |              | 0 |                           |    |                                   |
incardon's avatar
incardon committed
700 701 702 703 704 705 706 707 708 709
 |              |   |                           |    |                                   |
 |              |   |                           |    |        Processor 9                |
 |              |   |                           |B9_1|        Subdomain 1                |
 |              |   |                           |    |                                   |
 |              |   |                           |    |                                   |
 |              |   |                           |    |                                   |
 +--------------+---+---------------------------+----+                                   |
                                                     |                                   |
                                                     +-----------------------------------+

incardon's avatar
incardon committed
710

711 712
 \endverbatim

incardon's avatar
incardon committed
713
       and also
incardon's avatar
incardon committed
714
       G8_0 G9_0 G9_1 G5_0 (External ghost boxes)
incardon's avatar
incardon committed
715

incardon's avatar
incardon committed
716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739
      +----------------------------------------------------+
      |                 Processor 8                        |
      |                 Subdomain 0                        +-----------------------------------+
      |                                                    |                                   |
      |           +---------------------------------------------+                              |
      |           |         G8_0                           |    |                              |
+-----+---------------+------------------------------------+    |   Processor 9                |
|                 |   |                                    |    |   Subdomain 0                |
|                 |   |                                    |G9_0|                              |
|                 |   |                                    |    |                              |
|                 |   |                                    |    |                              |
|                 |   |        Local processor             |    |                              |
|  Processor 5    |   |        Sub+domain 0                |    |                              |
|  Subdomain 0    |   |                                    +-----------------------------------+
|                 |   |                                    |    |                              |
|                 | G |                                    |    |                              |
|                 | 5 |                                    |    |   Processor 9                |
|                 | | |                                    |    |   Subdomain 1                |
|                 | 0 |                                    |G9_1|                              |
|                 |   |                                    |    |                              |
|                 |   |                                    |    |                              |
+---------------------+------------------------------------+    |                              |
                  |                                        |    |                              |
                  +----------------------------------------+----+------------------------------+
740
	 \endverbatim
741

incardon's avatar
incardon committed
742 743 744 745 746
	 *
	 *
	 *
	 * \param ghost margins for each dimensions (p1 negative part) (p2 positive part)
	 *
747 748
	 *
	 \verbatim
749 750 751 752 753 754
	 	 	 	 	 ^ p2[1]
	 	 	 	 	 |
	 	 	 	 	 |
	 	 	 	+----+----+
	 	 	 	|         |
	 	 	 	|         |
755
	 p1[0]<-----+         +----> p2[0]
756 757 758 759 760
	 	 	 	|         |
	 	 	 	|         |
	 	 	 	+----+----+
	 	 	 	 	 |
	 	 	 	 	 v  p1[1]
incardon's avatar
incardon committed
761

762
	 \endverbatim
763

incardon's avatar
incardon committed
764 765 766
	 *
	 *
	 */
incardon's avatar
incardon committed
767
	void calculateGhostBoxes()
incardon's avatar
incardon committed
768 769 770 771 772
	{
#ifdef DEBUG
		// the ghost margins are assumed to be smaller
		// than one sub-domain

773
		for (size_t i = 0; i < dim; i++)
incardon's avatar
incardon committed
774
		{
incardon's avatar
incardon committed
775
			if (fabs(ghost.template getLow(i)) >= ss_box.getHigh(i) || ghost.template getHigh(i) >= ss_box.getHigh(i))
incardon's avatar
incardon committed
776
			{
777
				std::cerr << "Error " << __FILE__ << ":" << __LINE__  << " : Ghost are bigger than one sub-domain" << "\n";
incardon's avatar
incardon committed
778 779 780 781 782 783
			}
		}
#endif

		// Intersect all the local sub-domains with the sub-domains of the contiguous processors

incardon's avatar
incardon committed
784
		// create the internal structures that store ghost information
785 786
		ie_ghost<dim, T>::create_box_nn_processor_ext(v_cl, ghost, sub_domains, box_nn_processor, *this);
		ie_ghost<dim, T>::create_box_nn_processor_int(v_cl, ghost, sub_domains, box_nn_processor, *this);
incardon's avatar
incardon committed
787

788
		ie_loc_ghost<dim,T>::create(sub_domains,domain,ghost,bc);
incardon's avatar
incardon committed
789 790

		// get the smallest sub-domain dimension on each direction
791
		for (size_t i = 0; i < dim; i++)
incardon's avatar
incardon committed
792
		{
incardon's avatar
incardon committed
793
			if (fabs(ghost.template getLow(i)) >= ss_box.getHigh(i) || ghost.template getHigh(i) >= ss_box.getHigh(i))
incardon's avatar
incardon committed
794
			{
795
				std::cerr << "Error " << __FILE__ << ":" << __LINE__  << " : Ghost are bigger than one sub-domain" << "\n";
incardon's avatar
incardon committed
796 797
			}
		}
incardon's avatar
incardon committed
798 799
	}

incardon's avatar
incardon committed
800 801 802 803 804 805 806
	/*! \brief It create another object that contain the same decomposition information but with different ghost boxes
	 *
	 * \param g ghost
	 *
	 * \return a duplicated decomposition with different ghost boxes
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
807
	CartDecomposition<dim,T,Memory> duplicate(const Ghost<dim,T> & g) const
incardon's avatar
incardon committed
808
	{
Pietro Incardona's avatar
Pietro Incardona committed
809
		CartDecomposition<dim,T,Memory> cart(v_cl);
incardon's avatar
incardon committed
810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826

		cart.box_nn_processor = box_nn_processor;
		cart.sub_domains = sub_domains;
		cart.fine_s = fine_s;

		cart.gr = gr;
		cart.cd = cd;
		cart.domain = domain;
		std::copy(spacing,spacing+3,cart.spacing);

		//! Runtime virtual cluster
		cart.v_cl = v_cl;

		cart.bbox = bbox;
		cart.ss_box = ss_box;
		cart.ghost = g;

827 828
		cart.dist = dist;

829 830 831
		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = bc[i];

832 833
		(static_cast<nn_prcs<dim,T> &>(cart)).create(box_nn_processor, sub_domains);
		(static_cast<nn_prcs<dim,T> &>(cart)).applyBC(domain,ghost,bc);
incardon's avatar
incardon committed
834

835 836
		cart.Initialize_geo_cell_lists();
		cart.calculateGhostBoxes();
incardon's avatar
incardon committed
837 838 839 840

		return cart;
	}

841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904
	/*! \brief It create another object that contain the same decomposition information but with different ghost boxes and an extended domain
	 *
	 * The domain extension is produced extending the boxes at the border like in figure
	 *
	 * \verbatim
	 *
+--------------^--------^----------^----------+
|              |        |          |          |
|        A     |    E   |     F    |    N     |
|    +-----------------------------------+---->
|    |         |        |          |     |    |
|  A |   A     |        |     F    |     |    |
|    |         |        |          |     |    |
|    |         |    E   +----------+  N  |  N |
<--------------+        |          |     |    |
|    |         |        |          |     |    |
|    |         |        |     G    |     |    |
|    |         |        |          +---------->
|  B |   B     |        +----------+     |    |
|    |         +--------+          |  M  |  M |
|    |         |        |     H    |     |    |
|    |         |        +-----+----+---------->
<--------------+    D   |     |          |    |
|    |         |        |  I  |     L    |  L |
|  C |   C     |        |     |          |    |
|    |         |        |     |          |    |
|    +-----------------------------------+    |
|              |        |     |               |
|        C     |    D   |  I  |     L         |
+--------------v--------v-----v---------------+

	 *
	 * \endverbatim
	 *
	 * \param g ghost
	 * \param domain extended domain (MUST be extended)
	 *
	 * \return a duplicated decomposition with different ghost boxes and an extended domain
	 *
	 */
	CartDecomposition<dim,T,Memory> duplicate(const Ghost<dim,T> & g, const ::Box<dim,T> & ext_domain) const
	{
		CartDecomposition<dim,T,Memory> cart(v_cl);

		cart.box_nn_processor = box_nn_processor;

		// Calculate new sub-domains for extended domain
		extend_subdomains(cart,ext_domain);

		// Calculate fine_s structure for the extended domain
		// update the cell decomposer and gr
		extend_fines(cart);

		// Get the old sub-sub-domain grid extension

		cart.domain = ext_domain;

		// spacing does not change
		std::copy(spacing,spacing+3,cart.spacing);

		//! Runtime virtual cluster
		cart.v_cl = v_cl;

		cart.ghost = g;
905
		cart.dist = dist;
906 907 908 909 910 911 912 913 914 915 916 917 918

		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = bc[i];

		(static_cast<nn_prcs<dim,T> &>(cart)).create(cart.box_nn_processor, cart.sub_domains);
		(static_cast<nn_prcs<dim,T> &>(cart)).applyBC(ext_domain,ghost,bc);

		cart.Initialize_geo_cell_lists();
		cart.calculateGhostBoxes();

		return cart;
	}

incardon's avatar
incardon committed
919
	/*! \brief It create another object that contain the same information and act in the same way
920 921 922 923
	 *
	 * \return a duplicated decomposition
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
924
	CartDecomposition<dim,T,Memory> duplicate() const
925
	{
Pietro Incardona's avatar
Pietro Incardona committed
926
		CartDecomposition<dim,T,Memory> cart(v_cl);
927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944

		(static_cast<ie_loc_ghost<dim,T>*>(&cart))->operator=(static_cast<ie_loc_ghost<dim,T>>(*this));
		(static_cast<nn_prcs<dim,T>*>(&cart))->operator=(static_cast<nn_prcs<dim,T>>(*this));
		(static_cast<ie_ghost<dim,T>*>(&cart))->operator=(static_cast<ie_ghost<dim,T>>(*this));

		cart.sub_domains = sub_domains;
		cart.box_nn_processor = box_nn_processor;
		cart.fine_s = fine_s;
		cart.gr = gr;
		cart.cd = cd;
		cart.domain = domain;
		std::copy(spacing,spacing+3,cart.spacing);

		//! Runtime virtual cluster
		cart.v_cl = v_cl;

		cart.ghost = ghost;

incardon's avatar
incardon committed
945 946 947
		cart.bbox = bbox;
		cart.ss_box = ss_box;

948 949 950
		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = this->bc[i];

951 952 953 954 955 956 957 958
		return cart;
	}

	/*! \brief Copy the element
	 *
	 * \param cart element to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
959
	CartDecomposition<dim,T,Memory> & operator=(const CartDecomposition & cart)
960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977
	{
		static_cast<ie_loc_ghost<dim,T>*>(this)->operator=(static_cast<ie_loc_ghost<dim,T>>(cart));
		static_cast<nn_prcs<dim,T>*>(this)->operator=(static_cast<nn_prcs<dim,T>>(cart));
		static_cast<ie_ghost<dim,T>*>(this)->operator=(static_cast<ie_ghost<dim,T>>(cart));

		sub_domains = cart.sub_domains;
		box_nn_processor = cart.box_nn_processor;
		fine_s = cart.fine_s;
		gr = cart.gr;
		cd = cart.cd;
		domain = cart.domain;
		std::copy(cart.spacing,cart.spacing+3,spacing);

		//! Runtime virtual cluster
		v_cl = cart.v_cl;

		ghost = cart.ghost;

incardon's avatar
incardon committed
978 979 980
		bbox = cart.bbox;
		ss_box = cart.ss_box;

981 982 983
		for (size_t i = 0 ; i < dim ; i++)
			bc[i] = cart.bc[i];

984 985 986 987 988 989 990 991
		return *this;
	}

	/*! \brief Copy the element, move semantic
	 *
	 * \param cart element to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
992
	CartDecomposition<dim,T,Memory> & operator=(CartDecomposition && cart)
993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
	{
		static_cast<ie_loc_ghost<dim,T>*>(this)->operator=(static_cast<ie_loc_ghost<dim,T>*>(cart));
		static_cast<nn_prcs<dim,T>*>(this)->operator=(static_cast<nn_prcs<dim,T>*>(cart));
		static_cast<ie_ghost<dim,T>*>(this)->operator=(static_cast<ie_ghost<dim,T>*>(cart));

		sub_domains.swap(cart.sub_domains);
		box_nn_processor.swap(cart.box_nn_processor);
		fine_s.swap(cart.fine_s);
		gr = cart.gr;
		cd = cart.cd;
		domain = cart.domain;
		std::copy(cart.spacing,cart.spacing+3,spacing);

		//! Runtime virtual cluster
		v_cl = cart.v_cl;

		ghost = cart.ghost;

incardon's avatar
incardon committed
1011 1012 1013
		cart.bbox = bbox;
		cart.ss_box = ss_box;

1014 1015 1016
		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = bc[i];

1017 1018 1019
		return *this;
	}

1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030
	/*! \brief The default grid size
	 *
	 *  The default grid is always an isotropic grid that adapt with the number of processors,
	 *  it define in how many cell it will be divided the space for a particular required minimum
	 *  number of sub-domain
	 *
	 */
	static size_t getDefaultGrid(size_t n_sub)
	{
		// Calculate the number of sub-sub-domain on
		// each dimension
1031
		return openfpm::math::round_big_2(pow(n_sub, 1.0 / dim));
1032 1033
	}

1034
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
1035 1036 1037 1038
	 *
	 * \return processorID
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
1039
	template<typename Mem, typename ofb> size_t inline processorID(encapc<1, Point<dim,T>, Mem> p)
incardon's avatar
incardon committed
1040
	{
Pietro Incardona's avatar
Pietro Incardona committed
1041
		return fine_s.get(cd.template getCell<ofb>(p));
incardon's avatar
incardon committed
1042 1043
	}

1044
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
1045
	 *
1046
	 * \return processorID
incardon's avatar
incardon committed
1047 1048
	 *
	 */
1049
	size_t inline processorID(const Point<dim,T> &p) const
incardon's avatar
incardon committed
1050
	{
incardon's avatar
incardon committed
1051
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
1052 1053
	}

1054
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
1055 1056 1057 1058
	 *
	 * \return processorID
	 *
	 */
1059
	size_t inline processorID(const T (&p)[dim]) const
incardon's avatar
incardon committed
1060
	{
incardon's avatar
incardon committed
1061
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
1062
	}
incardon's avatar
incardon committed
1063

1064
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
1065
	 *
1066 1067 1068
	 * Boundary conditions are considered
	 *
	 * \return processorID
incardon's avatar
incardon committed
1069 1070
	 *
	 */
1071
	template<typename Mem> size_t inline processorIDBC(encapc<1, Point<dim,T>, Mem> p)
incardon's avatar
incardon committed
1072
	{
1073 1074 1075 1076
		Point<dim,T> pt = p;
		applyPointBC(pt);

		return fine_s.get(cd.getCell(pt));
incardon's avatar
incardon committed
1077 1078
	}

1079
	/*! \brief Given a point return in which processor the particle should go
1080 1081
	 *
	 * Boundary conditions are considered
incardon's avatar
incardon committed
1082 1083 1084 1085
	 *
	 * \return processorID
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
1086
	template<typename ofb> size_t inline processorIDBC(const Point<dim,T> &p) const
1087 1088 1089
	{
		Point<dim,T> pt = p;
		applyPointBC(pt);
incardon's avatar
incardon committed
1090