CartDecomposition.hpp 25.9 KB
Newer Older
incardon's avatar
incardon committed
1 2 3
/*
 * CartDecomposition.hpp
 *
4 5
 *  Created on: Oct 07, 2015
 *      Author: 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"
incardon's avatar
incardon committed
15
#include "Decomposition.hpp"
incardon's avatar
incardon committed
16
#include "Vector/map_vector.hpp"
incardon's avatar
incardon committed
17 18 19 20 21
#include <vector>
#include <initializer_list>
#include "SubdomainGraphNodes.hpp"
#include "dec_optimizer.hpp"
#include "Space/Shape/Box.hpp"
incardon's avatar
incardon committed
22
#include "Space/Shape/Point.hpp"
incardon's avatar
incardon committed
23
#include "NN/CellList/CellDecomposer.hpp"
incardon's avatar
incardon committed
24 25
#include <unordered_map>
#include "NN/CellList/CellList.hpp"
incardon's avatar
incardon committed
26
#include "Space/Ghost.hpp"
incardon's avatar
incardon committed
27 28
#include "common.hpp"
#include "ie_loc_ghost.hpp"
29 30
#include "ie_ghost.hpp"
#include "nn_processor.hpp"
31 32 33 34
#include "GraphMLWriter.hpp"
#include "ParMetisDistribution.hpp"
#include "MetisDistribution.hpp"
#include "DLB.hpp"
incardon's avatar
incardon committed
35

36 37 38 39 40 41 42 43 44 45 46
#define CARTDEC_ERROR 2000lu

// Macro that decide what to do in case of error
#ifdef STOP_ON_ERROR
#define ACTION_ON_ERROR() exit(1);
#elif defined(THROW_ON_ERROR)
#define ACTION_ON_ERROR() throw CARTDEC_ERROR;
#else
#define ACTION_ON_ERROR()
#endif

incardon's avatar
incardon committed
47 48 49 50 51 52 53 54
/**
 * \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
 * \tparam Domain Structure that contain the information of your physical domain
 *
55 56
 * Given an N-dimensional space, this class decompose the space into a Cartesian grid of small
 * sub-sub-domain. At each sub-sub-domain is assigned  an id that identify which processor is
incardon's avatar
incardon committed
57
 * going to take care of that part of space (in general the space assigned to a processor is
58 59 60 61 62 63
 * simply connected), a second step merge several sub-sub-domain with same id into bigger region
 *  sub-domain with the id. Each sub-domain has an extended space called ghost part
 *
 * Assuming that VCluster.getProcessUnitID(), equivalent to the MPI processor rank, return the processor local
 * processor id, we define
 *
64 65
 * * local processor: processor rank
 * * local sub-domain: sub-domain given to the local processor
66 67 68 69 70 71 72 73 74 75 76 77
 * * 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
78
 * * Local ghosts interal or external are all the ghosts that does not involve inter-processor communications
79 80
 *
 * \see calculateGhostBoxes() for a visualization of internal and external ghost boxes
incardon's avatar
incardon committed
81
 *
82 83 84
 * ### 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
85 86
 */

87 88 89
template<unsigned int dim, typename T, typename Memory = HeapMemory,
		template<unsigned int, typename > class Domain = Box, typename Distribution = ParMetisDistribution<dim, T>>
class CartDecomposition: public ie_loc_ghost<dim, T>, public nn_prcs<dim, T>, public ie_ghost<dim, T>
incardon's avatar
incardon committed
90
{
incardon's avatar
incardon committed
91

incardon's avatar
incardon committed
92
public:
incardon's avatar
incardon committed
93

incardon's avatar
incardon committed
94 95 96 97
	//! Type of the domain we are going to decompose
	typedef T domain_type;

	//! It simplify to access the SpaceBox element
98
	typedef SpaceBox<dim, T> Box;
incardon's avatar
incardon committed
99 100 101

private:

incardon's avatar
incardon committed
102
	//! This is the key type to access  data_s, for example in the case of vector
incardon's avatar
incardon committed
103
	//! acc_key is size_t
104 105
	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
106 107

	//! the set of all local sub-domain as vector
108
	openfpm::vector<SpaceBox<dim, T>> sub_domains;
incardon's avatar
incardon committed
109

incardon's avatar
incardon committed
110
	//! for each sub-domain, contain the list of the neighborhood processors
incardon's avatar
incardon committed
111 112
	openfpm::vector<openfpm::vector<long unsigned int> > box_nn_processor;

incardon's avatar
incardon committed
113
	//! Structure that contain for each sub-sub-domain box the processor id
incardon's avatar
Add ORB  
incardon committed
114
	//! exist for efficient global communication
incardon's avatar
incardon committed
115 116
	openfpm::vector<size_t> fine_s;

incardon's avatar
incardon committed
117
	//! Structure that store the cartesian grid information
118
	grid_sm<dim, void> gr;
incardon's avatar
incardon committed
119

incardon's avatar
incardon committed
120 121
	//! Structure that decompose your structure into cell without creating them
	//! useful to convert positions to CellId or sub-domain id in this case
122
	CellDecomposer_sm<dim, T> cd;
incardon's avatar
incardon committed
123 124

	//! rectangular domain to decompose
125
	Domain<dim, T> domain;
incardon's avatar
incardon committed
126 127 128 129 130 131 132

	//! Box Spacing
	T spacing[dim];

	//! Runtime virtual cluster machine
	Vcluster & v_cl;

133 134 135
	//! Create ditribution
	Distribution dist;

incardon's avatar
incardon committed
136
	//! Cell-list that store the geometrical information of the local internal ghost boxes
137
	CellList<dim, T, FAST> lgeo_cell;
incardon's avatar
incardon committed
138

139
	/*! \brief Constructor, it decompose and distribute the sub-domains across the processors
incardon's avatar
incardon committed
140
	 *
141 142
	 * \param v_cl Virtual cluster, used internally for communications
	 *
incardon's avatar
incardon committed
143 144 145
	 */
	void CreateDecomposition(Vcluster & v_cl)
	{
146 147 148 149 150 151 152
#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
153 154 155

		int p_id = v_cl.getProcessUnitID();

incardon's avatar
incardon committed
156 157 158
		// Calculate the total number of box and and the spacing
		// on each direction
		// Get the box containing the domain
159
		SpaceBox<dim, T> bs = domain.getBox();
incardon's avatar
incardon committed
160

161
		for (unsigned int i = 0; i < dim; i++)
incardon's avatar
incardon committed
162 163
		{
			// Calculate the spacing
incardon's avatar
incardon committed
164
			spacing[i] = (bs.getHigh(i) - bs.getLow(i)) / gr.size(i);
incardon's avatar
incardon committed
165 166
		}

incardon's avatar
Add ORB  
incardon committed
167
		// fill the structure that store the processor id for each sub-domain
incardon's avatar
incardon committed
168
		fine_s.resize(gr.size());
incardon's avatar
Add ORB  
incardon committed
169

incardon's avatar
incardon committed
170 171
		// Optimize the decomposition creating bigger spaces
		// And reducing Ghost over-stress
172
		dec_optimizer<dim, Graph_CSR<nm_v, nm_e>> d_o(dist.getGraph(), gr.getSize());
incardon's avatar
incardon committed
173 174

		// set of Boxes produced by the decomposition optimizer
175
		openfpm::vector<::Box<dim, size_t>> loc_box;
incardon's avatar
incardon committed
176

incardon's avatar
incardon committed
177
		// optimize the decomposition
178
		d_o.template optimize<nm_v::sub_id, nm_v::proc_id>(dist.getGraph(), p_id, loc_box, box_nn_processor);
incardon's avatar
incardon committed
179

incardon's avatar
incardon committed
180 181 182
		// Initialize ss_box and bbox
		if (loc_box.size() >= 0)
		{
183 184
			SpaceBox<dim, size_t> sub_dc = loc_box.get(0);
			SpaceBox<dim, T> sub_d(sub_dc);
incardon's avatar
incardon committed
185 186 187
			sub_d.mul(spacing);
			sub_d.expand(spacing);

188 189 190 191 192
			// 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)
193
			for (size_t i = 0; i < dim; i++)
194 195 196
			{
				if (sub_dc.getHigh(i) == cd.getGrid().size(i) - 1)
				{
197
					sub_d.setHigh(i, domain.getHigh(i));
198 199 200
				}
			}

incardon's avatar
incardon committed
201 202 203 204
			// add the sub-domain
			sub_domains.add(sub_d);

			ss_box = sub_d;
incardon's avatar
incardon committed
205
			ss_box -= ss_box.getP1();
incardon's avatar
incardon committed
206 207 208
			bbox = sub_d;
		}

incardon's avatar
incardon committed
209
		// convert into sub-domain
210
		for (size_t s = 1; s < loc_box.size(); s++)
incardon's avatar
incardon committed
211
		{
212 213
			SpaceBox<dim, size_t> sub_dc = loc_box.get(s);
			SpaceBox<dim, T> sub_d(sub_dc);
incardon's avatar
incardon committed
214

incardon's avatar
incardon committed
215 216 217
			// 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
218

219 220 221 222 223
			// 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)
224
			for (size_t i = 0; i < dim; i++)
225 226 227
			{
				if (sub_dc.getHigh(i) == cd.getGrid().size(i) - 1)
				{
228
					sub_d.setHigh(i, domain.getHigh(i));
229 230 231
				}
			}

incardon's avatar
incardon committed
232 233
			// add the sub-domain
			sub_domains.add(sub_d);
incardon's avatar
incardon committed
234 235 236 237 238 239

			// 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
240
		}
incardon's avatar
incardon committed
241

242
		nn_prcs<dim, T>::create(box_nn_processor, sub_domains);
243

incardon's avatar
incardon committed
244
		// fill fine_s structure
incardon's avatar
incardon committed
245 246 247
		// 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)
248
		auto it = dist.getGraph().getVertexIterator();
incardon's avatar
incardon committed
249 250 251 252 253 254

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

			// fill with the fine decomposition
255
			fine_s.get(key) = dist.getGraph().template vertex_p<nm_v::proc_id>(key);
incardon's avatar
incardon committed
256 257 258

			++it;
		}
259 260

		// Get the smallest sub-division on each direction
261
		::Box<dim, T> unit = getSmallestSubdivision();
262
		// Get the processor bounding Box
263
		::Box<dim, T> bound = getProcessorBounds();
264

incardon's avatar
incardon committed
265
		// calculate the sub-divisions
266
		size_t div[dim];
267 268
		for (size_t i = 0; i < dim; i++)
			div[i] = (size_t) ((bound.getHigh(i) - bound.getLow(i)) / unit.getHigh(i));
269 270

		// Create shift
271
		Point<dim, T> orig;
272 273

		// p1 point of the Processor bound box is the shift
274
		for (size_t i = 0; i < dim; i++)
275 276 277
			orig.get(i) = bound.getLow(i);

		// Initialize the geo_cell structure
278 279 280
		ie_ghost<dim, T>::Initialize_geo_cell(domain, div, orig);
		lgeo_cell.Initialize(domain, div, orig);

incardon's avatar
incardon committed
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
	/*! \brief Calculate communication and migration costs
	 *
	 * \param gh_s ghost thickness
	 * \param ts how many timesteps have passed since last calculation, used to approximate the cost
	 */
	void computeCommunicationAndMigrationCosts(size_t ts)
	{

		size_t p_id = v_cl.getProcessUnitID();
		float migration;

		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
		for (int i = 2; i < dim; i++)
			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++)
			{
				dist.setCommunicationCost(prev + s, 1 * dist.getVertexWeight(i) * ts);
			}
			prev += dist.getNSubSubDomainNeighbors(i);
		}
	}
incardon's avatar
incardon committed
322

323 324
	// Save the ghost boundaries
	Ghost<dim, T> ghost;
incardon's avatar
incardon committed
325

incardon's avatar
incardon committed
326 327 328 329 330 331
	/*! \brief Create the subspaces that decompose your domain
	 *
	 */
	void CreateSubspaces()
	{
		// Create a grid where each point is a space
332
		grid_sm<dim, void> g(div);
incardon's avatar
incardon committed
333 334 335 336 337 338 339 340 341 342 343

		// 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
344
			SpaceBox<dim, T> tmp;
incardon's avatar
incardon committed
345 346

			//! fill with the Margin of the box
347
			for (int i = 0; i < dim; i++)
incardon's avatar
incardon committed
348
			{
349 350
				tmp.setHigh(i, (key.get(i) + 1) * spacing[i]);
				tmp.setLow(i, key.get(i) * spacing[i]);
incardon's avatar
incardon committed
351 352 353 354 355 356 357 358 359 360
			}

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

			// add the iterator
			++gk_it;
		}
	}

incardon's avatar
incardon committed
361 362 363 364 365 366 367 368 369
	// Heap memory receiver
	HeapMemory hp_recv;

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

	// Receive counter
	size_t recv_cnt;

incardon's avatar
incardon committed
370 371 372
	// reference counter of the object in case is shared between object
	long int ref_cnt;

incardon's avatar
incardon committed
373 374
public:

375
	/*! \brief Cart decomposition constructor
incardon's avatar
incardon committed
376
	 *
377
	 * \param v_cl Virtual cluster, used internally to handle or pipeline communication
incardon's avatar
incardon committed
378 379
	 *
	 */
380 381
	CartDecomposition(Vcluster & v_cl) :
			nn_prcs<dim, T>(v_cl), v_cl(v_cl), dist(v_cl)
incardon's avatar
incardon committed
382 383 384 385
	{
		// Reset the box to zero
		bbox.zero();
	}
incardon's avatar
incardon committed
386

387
	//! Cart decomposition destructor
incardon's avatar
incardon committed
388
	~CartDecomposition()
389 390
	{
	}
incardon's avatar
incardon committed
391

392 393 394 395 396 397 398 399 400 401 402 403 404 405
	/*! \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
		 *
		 */
406
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
		{
			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
		 *
		 */
426
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445
		{
			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
		 *
		 */
446
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
447 448 449 450 451
		{
			return p.lc_proc;
		}
	};

incardon's avatar
incardon committed
452
	/*! It calculate the internal ghost boxes
incardon's avatar
incardon committed
453 454 455 456 457
	 *
	 * Example: Processor 10 calculate
	 * B8_0 B9_0 B9_1 and B5_0
	 *
	 *
458 459 460
	 *
	 \verbatim

461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519
	 +----------------------------------------------------+
	 |                                                    |
	 |                 Processor 8                        |
	 |                 Sub-domain 0                       +-----------------------------------+
	 |                                                    |                                   |
	 |                                                    |                                   |
	 ++--------------+---+---------------------------+----+        Processor 9                |
	 |              |   |     B8_0                  |    |        Subdomain 0                |
	 |              +------------------------------------+                                   |
	 |              |   |                           |    |                                   |
	 |              |   |  XXXXXXXXXXXXX XX         |B9_0|                                   |
	 |              | B |  X Processor 10 X         |    |                                   |
	 | Processor 5  | 5 |  X Sub-domain 0 X         |    |                                   |
	 | Subdomain 0  | _ |  X              X         +----------------------------------------+
	 |              | 0 |  XXXXXXXXXXXXXXXX         |    |                                   |
	 |              |   |                           |    |                                   |
	 |              |   |                           |    |        Processor 9                |
	 |              |   |                           |B9_1|        Subdomain 1                |
	 |              |   |                           |    |                                   |
	 |              |   |                           |    |                                   |
	 |              |   |                           |    |                                   |
	 +--------------+---+---------------------------+----+                                   |
	 |                                   |
	 +-----------------------------------+

	 \endverbatim

	 and also
	 G8_0 G9_0 G9_1 G5_0 (External ghost boxes)

	 \verbatim

	 +----------------------------------------------------+
	 |                                                    |
	 |                 Processor 8                        |
	 |                 Sub-domain 0                       +-----------------------------------+
	 |           +---------------------------------------------+                              |
	 |           |         G8_0                           |    |                              |
	 ++--------------+------------------------------------+    |   Processor 9                |
	 |          |   |                                    |    |   Subdomain 0                |
	 |          |   |                                    |G9_0|                              |
	 |          |   |                                    |    |                              |
	 |          |   |      XXXXXXXXXXXXX XX              |    |                              |
	 |          |   |      X Processor 10 X              |    |                              |
	 | Processor|5  |      X Sub-domain 0 X              |    |                              |
	 | Subdomain|0  |      X              X              +-----------------------------------+
	 |          |   |      XXXXXXXXXXXXXXXX              |    |                              |
	 |          | G |                                    |    |                              |
	 |          | 5 |                                    |    |   Processor 9                |
	 |          | | |                                    |    |   Subdomain 1                |
	 |          | 0 |                                    |G9_1|                              |
	 |          |   |                                    |    |                              |
	 |          |   |                                    |    |                              |
	 +--------------+------------------------------------+    |                              |
	 |                                        |    |                              |
	 +----------------------------------------+----+------------------------------+


	 \endverbatim
520

incardon's avatar
incardon committed
521 522 523 524 525
	 *
	 *
	 *
	 * \param ghost margins for each dimensions (p1 negative part) (p2 positive part)
	 *
526 527
	 *
	 \verbatim
528 529 530 531 532 533 534 535 536 537 538 539
	 ^ p2[1]
	 |
	 |
	 +----+----+
	 |         |
	 |         |
	 p1[0]<-----+         +----> p2[0]
	 |         |
	 |         |
	 +----+----+
	 |
	 v  p1[1]
incardon's avatar
incardon committed
540

541
	 \endverbatim
542

incardon's avatar
incardon committed
543 544 545
	 *
	 *
	 */
incardon's avatar
incardon committed
546
	void calculateGhostBoxes()
incardon's avatar
incardon committed
547 548 549 550 551
	{
#ifdef DEBUG
		// the ghost margins are assumed to be smaller
		// than one sub-domain

552
		for (size_t i = 0; i < dim; i++)
incardon's avatar
incardon committed
553
		{
554 555
			if (ghost.template getLow(i) >= domain.template getHigh(i) / gr.size(i)
					|| ghost.template getHigh(i) >= domain.template getHigh(i) / gr.size(i))
incardon's avatar
incardon committed
556
			{
557
				std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " : Ghost are bigger than one domain" << "\n";
incardon's avatar
incardon committed
558 559 560 561 562 563
			}
		}
#endif

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

incardon's avatar
incardon committed
564
		// create the internal structures that store ghost information
565 566
		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
567

incardon's avatar
incardon committed
568
		// ebox must come after ibox (in this case)
incardon's avatar
incardon committed
569

570 571
		ie_loc_ghost<dim, T>::create_loc_ghost_ibox(ghost, sub_domains);
		ie_loc_ghost<dim, T>::create_loc_ghost_ebox(ghost, sub_domains);
incardon's avatar
incardon committed
572 573

		// get the smallest sub-domain dimension on each direction
574
		for (size_t i = 0; i < dim; i++)
incardon's avatar
incardon committed
575
		{
576 577
			if (ghost.template getLow(i) >= ss_box.getHigh(i)
					|| ghost.template getHigh(i) >= domain.template getHigh(i) / gr.size(i))
incardon's avatar
incardon committed
578
			{
579
				std::cerr << "Error " << __FILE__ << ":" << __LINE__ << " : Ghost are bigger than one domain" << "\n";
incardon's avatar
incardon committed
580 581
			}
		}
incardon's avatar
incardon committed
582 583
	}

584 585 586 587 588 589 590 591 592 593 594
	/*! \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
595
		return openfpm::math::round_big_2(pow(n_sub, 1.0 / dim));
596 597
	}

598
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
599 600 601 602
	 *
	 * \return processorID
	 *
	 */
603
	template<typename Mem> size_t inline processorID(encapc<1, Point<dim, T>, Mem> p)
incardon's avatar
incardon committed
604
	{
incardon's avatar
incardon committed
605
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
606 607
	}

incardon's avatar
incardon committed
608
	// Smallest subdivision on each direction
609
	::Box<dim, T> ss_box;
incardon's avatar
incardon committed
610 611 612 613 614 615

	/*! \brief Get the smallest subdivision of the domain on each direction
	 *
	 * \return a box p1 is set to zero
	 *
	 */
616
	const ::Box<dim, T> & getSmallestSubdivision()
incardon's avatar
incardon committed
617 618 619 620
	{
		return ss_box;
	}

621
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
622 623 624 625 626
	 *
	 * \return processorID
	 *
	 */

627
	size_t inline processorID(const T (&p)[dim]) const
incardon's avatar
incardon committed
628
	{
incardon's avatar
incardon committed
629
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
630 631
	}

incardon's avatar
incardon committed
632 633
	/*! \brief Set the parameter of the decomposition
	 *
634 635
	 * \param div_ storing into how many domain to decompose on each dimension
	 * \param domain_ domain to decompose
incardon's avatar
incardon committed
636 637
	 *
	 */
638
	void setParameters(const size_t (&div_)[dim], Domain<dim, T> domain_, Ghost<dim, T> ghost = Ghost<dim, T>())
incardon's avatar
incardon committed
639
	{
incardon's avatar
incardon committed
640 641
		// set the ghost
		this->ghost = ghost;
incardon's avatar
incardon committed
642

643
		// Set the decomposition parameters
incardon's avatar
incardon committed
644
		gr.setDimensions(div_);
incardon's avatar
incardon committed
645
		domain = domain_;
646
		cd.setDimensions(domain, div_, 0);
incardon's avatar
incardon committed
647

648 649 650 651 652 653 654 655 656 657 658 659 660
		// init distribution
		dist.init(gr, domain);

	}

	/*! \brief Start decomposition
	 *
	 */
	void decompose()
	{
		computeCommunicationAndMigrationCosts(1);

		dist.decompose();
incardon's avatar
incardon committed
661 662 663 664

		CreateDecomposition(v_cl);
	}

665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707
	/*! \brief Refine the decomposition, available only for ParMetis distribution, for Metis it is a null call
	 *
	 */
	void rebalance()
	{
		computeCommunicationAndMigrationCosts(1);

		dist.refine();
	}

	/*! \brief Refine the decomposition, available only for ParMetis distribution, for Metis it is a null call
	 *
	 */
	void rebalance(DLB & dlb)
	{
		if (dlb.rebalanceNeeded())
		{
			computeCommunicationAndMigrationCosts(dlb.getNTimeStepSinceDLB());
			dist.refine();
		}
	}

	/*! \brief function that return the position of the cell in the space
	 *
	 * \param id vertex id
	 * \param pos vector that will contain x, y, z
	 *
	 */
	inline void getSubSubDomainPosition(size_t id, openfpm::vector<real_t> &pos)
	{
		dist.getVertexPosition(id, pos);
	}

	/*! \brief function that set the weight of the vertex
	 *
	 * \param id vertex id
	 *
	 */
	inline void setSubSubDomainComputationCost(size_t id, size_t weight)
	{
		dist.setVertexWeight(id, weight);
	}

708
	/*! \brief Get the number of local sub-domains
incardon's avatar
incardon committed
709 710 711 712 713 714 715 716 717
	 *
	 * \return the number of sub-domains
	 *
	 */
	size_t getNLocalHyperCube()
	{
		return sub_domains.size();
	}

incardon's avatar
incardon committed
718
	/*! \brief Get the local sub-domain
incardon's avatar
incardon committed
719
	 *
incardon's avatar
incardon committed
720 721
	 * \param i (each local processor can have more than one sub-domain)
	 * \return the sub-domain
incardon's avatar
incardon committed
722 723
	 *
	 */
724
	SpaceBox<dim, T> getLocalHyperCube(size_t lc)
incardon's avatar
incardon committed
725 726
	{
		// Create a space box
727
		SpaceBox<dim, T> sp;
incardon's avatar
incardon committed
728 729 730

		// fill the space box

731
		for (size_t k = 0; k < dim; k++)
incardon's avatar
incardon committed
732 733
		{
			// create the SpaceBox Low and High
734 735
			sp.setLow(k, sub_domains.template get<Box::p1>(lc)[k]);
			sp.setHigh(k, sub_domains.template get<Box::p2>(lc)[k]);
incardon's avatar
incardon committed
736 737 738 739 740
		}

		return sp;
	}

incardon's avatar
incardon committed
741 742 743 744 745 746
	/*! \brief Get the local sub-domain with ghost extension
	 *
	 * \param i (each local processor can have more than one sub-domain)
	 * \return the sub-domain
	 *
	 */
747
	SpaceBox<dim, T> getSubDomainWithGhost(size_t lc)
incardon's avatar
incardon committed
748 749
	{
		// Create a space box
750
		SpaceBox<dim, T> sp = sub_domains.get(lc);
incardon's avatar
incardon committed
751 752 753 754 755 756 757

		// enlarge with ghost
		sp.enlarge(ghost);

		return sp;
	}

incardon's avatar
incardon committed
758 759 760 761 762
	/*! \brief Return the structure that store the physical domain
	 *
	 * \return The physical domain
	 *
	 */
763
	Domain<dim, T> & getDomain()
incardon's avatar
incardon committed
764 765 766 767
	{
		return domain;
	}

incardon's avatar
incardon committed
768 769 770
	/*! \brief Check if the particle is local
	 *
	 * \param p object position
incardon's avatar
incardon committed
771
	 *
incardon's avatar
incardon committed
772
	 * \return true if it is local
incardon's avatar
incardon committed
773 774
	 *
	 */
775
	template<typename Mem> bool isLocal(const encapc<1, Point<dim, T>, Mem> p) const
incardon's avatar
incardon committed
776
	{
777
		return processorID<Mem>(p) == v_cl.getProcessUnitID();
incardon's avatar
incardon committed
778
	}
incardon's avatar
incardon committed
779

incardon's avatar
incardon committed
780
	/*! \brief Check if the particle is local
incardon's avatar
incardon committed
781
	 *
incardon's avatar
incardon committed
782
	 * \param p object position
incardon's avatar
incardon committed
783
	 *
incardon's avatar
incardon committed
784
	 * \return true if it is local
incardon's avatar
incardon committed
785 786
	 *
	 */
787
	bool isLocal(const T (&pos)[dim]) const
incardon's avatar
incardon committed
788
	{
incardon's avatar
incardon committed
789 790
		return processorID(pos) == v_cl.getProcessUnitID();
	}
incardon's avatar
incardon committed
791

792
	::Box<dim, T> bbox;
incardon's avatar
incardon committed
793

794
	/*! \brief Return the bounding box containing union of all the sub-domains for the local processor
incardon's avatar
incardon committed
795 796 797 798
	 *
	 * \return The bounding box
	 *
	 */
799
	::Box<dim, T> & getProcessorBounds()
incardon's avatar
incardon committed
800 801 802
	{
		return bbox;
	}
incardon's avatar
incardon committed
803

804 805
	////////////// Functions to get decomposition information ///////////////

incardon's avatar
incardon committed
806 807 808 809
	/*! \brief Write the decomposition as VTK file
	 *
	 * The function generate several files
	 *
810 811 812 813 814 815
	 * * subdomains_X.vtk domain for the local processor (X) as union of sub-domain
	 * * subdomains_adjacent_X.vtk sub-domains adjacent to the local processor (X)
	 * * internal_ghost_X.vtk Internal ghost boxes for the local processor (X)
	 * * external_ghost_X.vtk External ghost boxes for the local processor (X)
	 * * local_internal_ghost_X.vtk internal local ghost boxes for the local processor (X)
	 * * local_external_ghost_X.vtk external local ghost boxes for the local processor (X)
incardon's avatar
incardon committed
816
	 *
incardon's avatar
incardon committed
817
	 * where X is the local processor rank
incardon's avatar
incardon committed
818 819 820 821 822 823
	 *
	 * \param output directory where to write the files
	 *
	 */
	bool write(std::string output) const
	{
824
		//! subdomains_X.vtk domain for the local processor (X) as union of sub-domain
825
		VTKWriter<openfpm::vector<::SpaceBox<dim, T>>, VECTOR_BOX> vtk_box1;
incardon's avatar
incardon committed
826
		vtk_box1.add(sub_domains);
827 828
		vtk_box1.write(
				output + std::string("subdomains_") + std::to_string(v_cl.getProcessUnitID()) + std::string(".vtk"));
incardon's avatar
incardon committed
829

830 831 832
		nn_prcs<dim, T>::write(output);
		ie_ghost<dim, T>::write(output, v_cl.getProcessUnitID());
		ie_loc_ghost<dim, T>::write(output, v_cl.getProcessUnitID());
incardon's avatar
incardon committed
833

834
		return true;
incardon's avatar
incardon committed
835
	}
incardon's avatar
incardon committed
836 837 838 839 840 841 842 843

	/*! \brief function to check the consistency of the information of the decomposition
	 *
	 * \return false if is inconsistent
	 *
	 */
	bool check_consistency()
	{
844
		if (ie_loc_ghost<dim, T>::check_consistency(getNLocalHyperCube()) == false)
incardon's avatar
incardon committed
845
			return false;
incardon's avatar
incardon committed
846 847 848

		return true;
	}
849

850 851 852
	/*! \brief Print subdomains, external and internal ghost boxes
	 *
	 */
853 854
	void debugPrint()
	{
855
		std::cout << "Subdomains\n";
856
		for (size_t p = 0; p < sub_domains.size(); p++)
857
		{
858
			std::cout << ::SpaceBox<dim, T>(sub_domains.get(p)).toString() << "\n";
859 860
		}

861
		std::cout << "External ghost box\n";
862

863
		for (size_t p = 0; p<nn_prcs < dim, T>::getNNProcessors(); p++)
864
		{
865
			for (size_t i = 0; i<ie_ghost < dim, T>::getProcessorNEGhost(p); i++)
866
			{
867 868 869
				std::cout << ie_ghost<dim, T>::getProcessorEGhostBox(p, i).toString() << "   prc="
						<< nn_prcs<dim, T>::IDtoProc(p) << "   id=" << ie_ghost<dim, T>::getProcessorEGhostId(p, i)
						<< "\n";
870
			}
871
		}
872

873
		std::cout << "Internal ghost box\n";
874

875
		for (size_t p = 0; p<nn_prcs < dim, T>::getNNProcessors(); p++)
876
		{
877
			for (size_t i = 0; i<ie_ghost < dim, T>::getProcessorNIGhost(p); i++)
878
			{
879 880 881
				std::cout << ie_ghost<dim, T>::getProcessorIGhostBox(p, i).toString() << "   prc="
						<< nn_prcs<dim, T>::IDtoProc(p) << "   id=" << ie_ghost<dim, T>::getProcessorIGhostId(p, i)
						<< "\n";
882 883
			}
		}
884
	}
incardon's avatar
incardon committed
885

886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916
	/*! \brief Print current graph and save it to file with name test_graph_[id]
	 *
	 * \param id to attach to the filename
	 *
	 */
	void printCurrentDecomposition(int id)
	{
		if (v_cl.getProcessUnitID() == 0)
		{
			dist.printCurrentDecomposition(id);
		}
	}

	//! 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
917 918

#endif