CartDecomposition.hpp 38.7 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"
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
#include "GraphMLWriter/GraphMLWriter.hpp"
32 33 34 35
#include "Distribution/ParMetisDistribution.hpp"
#include "Distribution/DistParMetisDistribution.hpp"
#include "Distribution/MetisDistribution.hpp"
#include "DLB/DLB.hpp"
36
#include "util/se_util.hpp"
37
#include "util/mathutil.hpp"
Pietro Incardona's avatar
Pietro Incardona committed
38
#include "CartDecomposition_ext.hpp"
incardon's avatar
incardon committed
39

40 41
#define CARTDEC_ERROR 2000lu

incardon's avatar
incardon committed
42
/**
Pietro Incardona's avatar
Pietro Incardona committed
43
 * \brief This class decompose a space into sub-sub-domains and distribute them across processors
incardon's avatar
incardon committed
44 45 46 47
 *
 * \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
 */

Pietro Incardona's avatar
Pietro Incardona committed
82
template<unsigned int dim, typename T, typename Memory, typename Distribution>
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

Pietro Incardona's avatar
Pietro Incardona committed
94 95 96 97 98 99 100
	//! This class is base of itself
	typedef CartDecomposition<dim,T,Memory,Distribution> base_type;

	//! This class admit a class defined on an extended domain
	typedef CartDecomposition_ext<dim,T,Memory,Distribution> extended_type;

protected:
incardon's avatar
incardon committed
101

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
tonynsyde's avatar
tonynsyde committed
104
	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
105 106

	//! the set of all local sub-domain as vector
107
	openfpm::vector<SpaceBox<dim, T>> sub_domains;
incardon's avatar
incardon committed
108

incardon's avatar
incardon committed
109
	//! for each sub-domain, contain the list of the neighborhood processors
incardon's avatar
incardon committed
110 111
	openfpm::vector<openfpm::vector<long unsigned int> > box_nn_processor;

incardon's avatar
incardon committed
112
	//! Structure that contain for each sub-sub-domain box the processor id
incardon's avatar
Add ORB  
incardon committed
113
	//! exist for efficient global communication
incardon's avatar
incardon committed
114 115
	openfpm::vector<size_t> fine_s;

incardon's avatar
incardon committed
116
	//! Structure that store the cartesian grid information
117
	grid_sm<dim, void> gr;
incardon's avatar
incardon committed
118

incardon's avatar
incardon committed
119 120
	//! Structure that decompose your structure into cell without creating them
	//! useful to convert positions to CellId or sub-domain id in this case
121
	CellDecomposer_sm<dim, T> cd;
incardon's avatar
incardon committed
122 123

	//! rectangular domain to decompose
Pietro Incardona's avatar
Pietro Incardona committed
124
	::Box<dim,T> domain;
incardon's avatar
incardon committed
125 126 127 128 129 130 131

	//! Box Spacing
	T spacing[dim];

	//! Runtime virtual cluster machine
	Vcluster & v_cl;

132
	//! Create distribution
133 134
	Distribution dist;

incardon's avatar
incardon committed
135 136 137 138
	// Smallest subdivision on each direction
	::Box<dim,T> ss_box;

	::Box<dim,T> bbox;
incardon's avatar
incardon committed
139

140 141 142
	// 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

149 150 151 152 153 154
	// Heap memory receiver
	HeapMemory hp_recv;

	// Receive counter
	size_t recv_cnt;

Pietro Incardona's avatar
Pietro Incardona committed
155 156 157 158 159 160
protected:



public:

161
	/*! \brief Constructor, it decompose and distribute the sub-domains across the processors
incardon's avatar
incardon committed
162
	 *
163 164
	 * \param v_cl Virtual cluster, used internally for communications
	 *
incardon's avatar
incardon committed
165
	 */
Pietro Incardona's avatar
Pietro Incardona committed
166
	void createSubdomains(Vcluster & v_cl, const size_t (& bc)[dim])
incardon's avatar
incardon committed
167
	{
168 169 170 171 172 173 174
#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
175 176 177

		int p_id = v_cl.getProcessUnitID();

incardon's avatar
incardon committed
178 179 180
		// Calculate the total number of box and and the spacing
		// on each direction
		// Get the box containing the domain
181
		SpaceBox<dim, T> bs = domain.getBox();
incardon's avatar
incardon committed
182

183
		for (unsigned int i = 0; i < dim; i++)
incardon's avatar
incardon committed
184 185
		{
			// Calculate the spacing
incardon's avatar
incardon committed
186
			spacing[i] = (bs.getHigh(i) - bs.getLow(i)) / gr.size(i);
incardon's avatar
incardon committed
187 188
		}

incardon's avatar
Add ORB  
incardon committed
189
		// fill the structure that store the processor id for each sub-domain
incardon's avatar
incardon committed
190
		fine_s.resize(gr.size());
incardon's avatar
Add ORB  
incardon committed
191

incardon's avatar
incardon committed
192 193
		// Optimize the decomposition creating bigger spaces
		// And reducing Ghost over-stress
194
		dec_optimizer<dim, Graph_CSR<nm_v, nm_e>> d_o(dist.getGraph(), gr.getSize());
incardon's avatar
incardon committed
195 196

		// set of Boxes produced by the decomposition optimizer
197
		openfpm::vector<::Box<dim, size_t>> loc_box;
incardon's avatar
incardon committed
198

incardon's avatar
incardon committed
199
		// optimize the decomposition
200
		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
201

incardon's avatar
incardon committed
202 203 204
		// Initialize ss_box and bbox
		if (loc_box.size() >= 0)
		{
205 206
			SpaceBox<dim, size_t> sub_dc = loc_box.get(0);
			SpaceBox<dim, T> sub_d(sub_dc);
incardon's avatar
incardon committed
207 208
			sub_d.mul(spacing);
			sub_d.expand(spacing);
Pietro Incardona's avatar
Pietro Incardona committed
209 210 211
			sub_d += domain.getP1();

			// we add the
incardon's avatar
incardon committed
212

213 214 215
			// Fixing sub-domains to cover all the domain

			// Fixing sub_d
Pietro Incardona's avatar
Pietro Incardona committed
216
			// if (loc_box) is at the boundary we have to ensure that the box span the full
217
			// domain (avoiding rounding off error)
218
			for (size_t i = 0; i < dim; i++)
219 220
			{
				if (sub_dc.getHigh(i) == cd.getGrid().size(i) - 1)
221
					sub_d.setHigh(i, domain.getHigh(i));
222

Pietro Incardona's avatar
Pietro Incardona committed
223 224
				if (sub_dc.getLow(i) == 0)
					sub_d.setLow(i,domain.getLow(i));
225 226
			}

incardon's avatar
incardon committed
227 228 229 230
			// add the sub-domain
			sub_domains.add(sub_d);

			ss_box = sub_d;
incardon's avatar
incardon committed
231
			ss_box -= ss_box.getP1();
incardon's avatar
incardon committed
232 233 234
			bbox = sub_d;
		}

Pietro Incardona's avatar
Pietro Incardona committed
235 236 237 238
/*		if (loc_box.size())
		bbox.zero();
		ss_box = domain;*/

incardon's avatar
incardon committed
239
		// convert into sub-domain
240
		for (size_t s = 1; s < loc_box.size(); s++)
incardon's avatar
incardon committed
241
		{
242 243
			SpaceBox<dim, size_t> sub_dc = loc_box.get(s);
			SpaceBox<dim, T> sub_d(sub_dc);
incardon's avatar
incardon committed
244

incardon's avatar
incardon committed
245 246 247
			// re-scale and add spacing (the end is the starting point of the next domain + spacing)
			sub_d.mul(spacing);
			sub_d.expand(spacing);
Pietro Incardona's avatar
Pietro Incardona committed
248
			sub_d += domain.getP1();
incardon's avatar
incardon committed
249

250 251 252 253 254
			// 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)
255
			for (size_t i = 0; i < dim; i++)
256 257
			{
				if (sub_dc.getHigh(i) == cd.getGrid().size(i) - 1)
258
					sub_d.setHigh(i, domain.getHigh(i));
259

Pietro Incardona's avatar
Pietro Incardona committed
260 261
				if (sub_dc.getLow(i) == 0)
					sub_d.setLow(i,domain.getLow(i));
262 263
			}

incardon's avatar
incardon committed
264 265
			// add the sub-domain
			sub_domains.add(sub_d);
incardon's avatar
incardon committed
266 267 268 269 270 271

			// 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
272
		}
incardon's avatar
incardon committed
273

274
		nn_prcs<dim,T>::create(box_nn_processor, sub_domains);
incardon's avatar
incardon committed
275
		nn_prcs<dim,T>::refine_ss_box(ss_box);
Pietro Incardona's avatar
Pietro Incardona committed
276
		nn_prcs<dim,T>::applyBC(domain,ghost,bc);
277

incardon's avatar
incardon committed
278
		// fill fine_s structure
incardon's avatar
incardon committed
279 280 281
		// 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)
282
		auto it = dist.getGraph().getVertexIterator();
incardon's avatar
incardon committed
283 284 285 286 287 288

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

			// fill with the fine decomposition
289
			fine_s.get(key) = dist.getGraph().template vertex_p<nm_v::proc_id>(key);
incardon's avatar
incardon committed
290 291 292

			++it;
		}
293

incardon's avatar
incardon committed
294 295 296 297 298 299 300 301 302 303
		Initialize_geo_cell_lists();
	}

	/*! \brief Initialize geo_cell lists
	 *
	 *
	 *
	 */
	void Initialize_geo_cell_lists()
	{
304
		// Get the smallest sub-division on each direction
305
		::Box<dim, T> unit = getSmallestSubdivision();
306 307
		// Get the processor bounding Box
		::Box<dim,T> bound = getProcessorBounds();
Pietro Incardona's avatar
Pietro Incardona committed
308 309
		// Not necessary, but I prefer
		bound.enlarge(ghost);
310

incardon's avatar
incardon committed
311
		// calculate the sub-divisions
312
		size_t div[dim];
313 314
		for (size_t i = 0; i < dim; i++)
			div[i] = (size_t) ((bound.getHigh(i) - bound.getLow(i)) / unit.getHigh(i));
315 316

		// Initialize the geo_cell structure
317
		ie_ghost<dim,T>::Initialize_geo_cell(bound,div);
318

319 320
		// Initialize shift vectors
		ie_ghost<dim,T>::generateShiftVectors(domain);
incardon's avatar
incardon committed
321 322
	}

323 324 325 326 327 328
	/*! \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
329
		float migration = 0;
330 331 332 333 334 335 336 337 338

		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
339
		for (size_t i = 2; i < dim; i++)
340 341 342 343 344 345 346 347 348 349
			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++)
		{
350
			dist.setMigrationCost(i, norm * migration * dist.getSubSubDomainComputationCost(i));
351 352 353

			for (size_t s = 0; s < dist.getNSubSubDomainNeighbors(i); s++)
			{
354
				dist.setCommunicationCost(i, s, 1 * dist.getSubSubDomainComputationCost(i) * ts);
355 356 357 358
			}
			prev += dist.getNSubSubDomainNeighbors(i);
		}
	}
incardon's avatar
incardon committed
359

incardon's avatar
incardon committed
360 361 362 363 364 365
	/*! \brief Create the subspaces that decompose your domain
	 *
	 */
	void CreateSubspaces()
	{
		// Create a grid where each point is a space
366
		grid_sm<dim, void> g(div);
incardon's avatar
incardon committed
367 368 369 370 371 372 373 374 375 376 377

		// 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
378
			SpaceBox<dim, T> tmp;
incardon's avatar
incardon committed
379 380

			//! fill with the Margin of the box
381
			for (int i = 0; i < dim; i++)
incardon's avatar
incardon committed
382
			{
383 384
				tmp.setHigh(i, (key.get(i) + 1) * spacing[i]);
				tmp.setLow(i, key.get(i) * spacing[i]);
incardon's avatar
incardon committed
385 386 387 388 389 390 391 392 393 394
			}

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

			// add the iterator
			++gk_it;
		}
	}

395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521

	/*! It calculate the internal ghost boxes
	 *
	 * Example: Processor 10 calculate
	 * B8_0 B9_0 B9_1 and B5_0
	 *
	 *
	 *
	 \verbatim

+----------------------------------------------------+
|                                                    |
|                 Processor 8                        |
|                 Sub+domain 0                       +-----------------------------------+
|                                                    |                                   |
|                                                    |                                   |
++--------------+---+---------------------------+----+        Processor 9                |
 |              |   |     B8_0                  |    |        Subdomain 0                |
 |              +------------------------------------+                                   |
 |              |   |                           |    |                                   |
 |              |   |                           |B9_0|                                   |
 |              | B |    Local processor        |    |                                   |
 | Processor 5  | 5 |    Subdomain 0            |    |                                   |
 | Subdomain 0  | _ |                           +----------------------------------------+
 |              | 0 |                           |    |                                   |
 |              |   |                           |    |                                   |
 |              |   |                           |    |        Processor 9                |
 |              |   |                           |B9_1|        Subdomain 1                |
 |              |   |                           |    |                                   |
 |              |   |                           |    |                                   |
 |              |   |                           |    |                                   |
 +--------------+---+---------------------------+----+                                   |
                                                     |                                   |
                                                     +-----------------------------------+


 \endverbatim

       and also
       G8_0 G9_0 G9_1 G5_0 (External ghost boxes)

      +----------------------------------------------------+
      |                 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|                              |
|                 |   |                                    |    |                              |
|                 |   |                                    |    |                              |
+---------------------+------------------------------------+    |                              |
                  |                                        |    |                              |
                  +----------------------------------------+----+------------------------------+

	 \endverbatim

	 *
	 *
	 *
	 * \param ghost margins for each dimensions (p1 negative part) (p2 positive part)
	 *
	 *
	 \verbatim

	 	 	 	 	 ^ p2[1]
	 	 	 	 	 |
	 	 	 	 	 |
	 	 	 	+----+----+
	 	 	 	|         |
	 	 	 	|         |
	 p1[0]<-----+         +----> p2[0]
	 	 	 	|         |
	 	 	 	|         |
	 	 	 	+----+----+
	 	 	 	 	 |
	 	 	 	 	 v  p1[1]

	 \endverbatim

	 *
	 *
	 */
	void calculateGhostBoxes()
	{
#ifdef DEBUG
		// the ghost margins are assumed to be smaller
		// than one sub-domain

		for (size_t i = 0; i < dim; i++)
		{
			if (fabs(ghost.template getLow(i)) >= ss_box.getHigh(i) || ghost.template getHigh(i) >= ss_box.getHigh(i))
			{
				std::cerr << "Error " << __FILE__ << ":" << __LINE__  << " : Ghost are bigger than one sub-domain" << "\n";
			}
		}
#endif

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

		// create the internal structures that store ghost information
		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);

		ie_loc_ghost<dim,T>::create(sub_domains,domain,ghost,bc);

		// get the smallest sub-domain dimension on each direction
		for (size_t i = 0; i < dim; i++)
		{
			if (fabs(ghost.template getLow(i)) >= ss_box.getHigh(i) || ghost.template getHigh(i) >= ss_box.getHigh(i))
			{
				std::cerr << "Error " << __FILE__ << ":" << __LINE__  << " : Ghost are bigger than one sub-domain" << "\n";
			}
		}
	}

incardon's avatar
incardon committed
522 523
public:

524 525 526 527
	static constexpr int dims = dim;

	typedef T stype;

incardon's avatar
incardon committed
528 529 530 531 532 533 534 535 536 537 538 539 540 541
	//! 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
542 543
	/*! \brief Cartesian decomposition constructor
	 *
544
	 * \param v_cl Virtual cluster, used internally to handle or pipeline communication
incardon's avatar
incardon committed
545 546
	 *
	 */
547
	CartDecomposition(Vcluster & v_cl) :
548
			nn_prcs<dim, T>(v_cl), v_cl(v_cl), dist(v_cl),ref_cnt(0)
incardon's avatar
incardon committed
549 550 551 552
	{
		// Reset the box to zero
		bbox.zero();
	}
incardon's avatar
incardon committed
553

554 555 556 557 558
	/*! \brief Cartesian decomposition copy constructor
	 *
     * \param cart object to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
559
	CartDecomposition(const CartDecomposition<dim,T,Memory> & cart)
560
	:nn_prcs<dim,T>(cart.v_cl),v_cl(cart.v_cl),dist(v_cl),ref_cnt(0)
561 562 563 564 565 566 567 568 569
	{
		this->operator=(cart);
	}

	/*! \brief Cartesian decomposition copy constructor
	 *
     * \param cart object to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
570
	CartDecomposition(CartDecomposition<dim,T,Memory> && cart)
571
	:nn_prcs<dim,T>(cart.v_cl),v_cl(cart.v_cl),dist(v_cl),ref_cnt(0)
572 573 574 575
	{
		this->operator=(cart);
	}

incardon's avatar
incardon committed
576 577
	//! Cartesian decomposition destructor
	~CartDecomposition()
578 579
	{
	}
incardon's avatar
incardon committed
580

581 582 583 584 585 586 587 588 589 590 591 592 593 594
	/*! \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
		 *
		 */
595
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
		{
			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
		 *
		 */
615
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
		{
			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
		 *
		 */
635
		inline static size_t id(p_box<dim, T> & p, size_t b_id)
636 637 638 639 640
		{
			return p.lc_proc;
		}
	};

641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
	/*! \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;
		}
	};

661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702
	/*! \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
703 704 705 706 707 708 709
	/*! \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
710
	CartDecomposition<dim,T,Memory> duplicate(const Ghost<dim,T> & g) const
incardon's avatar
incardon committed
711
	{
Pietro Incardona's avatar
Pietro Incardona committed
712
		CartDecomposition<dim,T,Memory> cart(v_cl);
incardon's avatar
incardon committed
713 714 715 716 717 718 719 720 721 722 723 724 725 726

		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);

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

727 728
		cart.dist = dist;

729 730 731
		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = bc[i];

732 733
		(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
734

735 736
		cart.Initialize_geo_cell_lists();
		cart.calculateGhostBoxes();
incardon's avatar
incardon committed
737 738 739 740 741

		return cart;
	}

	/*! \brief It create another object that contain the same information and act in the same way
742 743 744 745
	 *
	 * \return a duplicated decomposition
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
746
	CartDecomposition<dim,T,Memory> duplicate() const
747
	{
Pietro Incardona's avatar
Pietro Incardona committed
748
		CartDecomposition<dim,T,Memory> cart(v_cl);
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763

		(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);

		cart.ghost = ghost;

incardon's avatar
incardon committed
764 765 766
		cart.bbox = bbox;
		cart.ss_box = ss_box;

767 768 769
		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = this->bc[i];

770 771 772 773 774 775 776 777
		return cart;
	}

	/*! \brief Copy the element
	 *
	 * \param cart element to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
778
	CartDecomposition<dim,T,Memory> & operator=(const CartDecomposition & cart)
779 780 781 782 783 784 785 786 787 788 789 790 791 792 793
	{
		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);

		ghost = cart.ghost;

incardon's avatar
incardon committed
794 795 796
		bbox = cart.bbox;
		ss_box = cart.ss_box;

797 798 799
		for (size_t i = 0 ; i < dim ; i++)
			bc[i] = cart.bc[i];

800 801 802 803 804 805 806 807
		return *this;
	}

	/*! \brief Copy the element, move semantic
	 *
	 * \param cart element to copy
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
808
	CartDecomposition<dim,T,Memory> & operator=(CartDecomposition && cart)
809
	{
Pietro Incardona's avatar
Pietro Incardona committed
810 811 812
		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));
813 814 815 816 817 818 819 820 821 822 823 824 825 826

		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
827 828 829
		cart.bbox = bbox;
		cart.ss_box = ss_box;

830 831 832
		for (size_t i = 0 ; i < dim ; i++)
			cart.bc[i] = bc[i];

833 834 835
		return *this;
	}

836 837 838 839 840 841 842 843 844 845 846
	/*! \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
847
		return openfpm::math::round_big_2(pow(n_sub, 1.0 / dim));
848 849
	}

850
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
851 852 853 854
	 *
	 * \return processorID
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
855
	template<typename Mem, typename ofb> size_t inline processorID(encapc<1, Point<dim,T>, Mem> p)
incardon's avatar
incardon committed
856
	{
Pietro Incardona's avatar
Pietro Incardona committed
857
		return fine_s.get(cd.template getCell<ofb>(p));
incardon's avatar
incardon committed
858 859
	}

860
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
861
	 *
862
	 * \return processorID
incardon's avatar
incardon committed
863 864
	 *
	 */
865
	size_t inline processorID(const Point<dim,T> &p) const
incardon's avatar
incardon committed
866
	{
incardon's avatar
incardon committed
867
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
868 869
	}

870
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
871 872 873 874
	 *
	 * \return processorID
	 *
	 */
875
	size_t inline processorID(const T (&p)[dim]) const
incardon's avatar
incardon committed
876
	{
incardon's avatar
incardon committed
877
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
878
	}
incardon's avatar
incardon committed
879

880
	/*! \brief Given a point return in which processor the particle should go
incardon's avatar
incardon committed
881
	 *
882 883 884
	 * Boundary conditions are considered
	 *
	 * \return processorID
incardon's avatar
incardon committed
885 886
	 *
	 */
887
	template<typename Mem> size_t inline processorIDBC(encapc<1, Point<dim,T>, Mem> p)
incardon's avatar
incardon committed
888
	{
889 890 891 892
		Point<dim,T> pt = p;
		applyPointBC(pt);

		return fine_s.get(cd.getCell(pt));
incardon's avatar
incardon committed
893 894
	}

895
	/*! \brief Given a point return in which processor the particle should go
896 897
	 *
	 * Boundary conditions are considered
incardon's avatar
incardon committed
898 899 900 901
	 *
	 * \return processorID
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
902
	template<typename ofb> size_t inline processorIDBC(const Point<dim,T> &p) const
903 904 905
	{
		Point<dim,T> pt = p;
		applyPointBC(pt);
incardon's avatar
incardon committed
906

907 908 909 910 911 912 913 914 915 916
		return fine_s.get(cd.getCell(p));
	}

	/*! \brief Given a point return in which processor the particle should go
	 *
	 * Boundary consition are considered
	 *
	 * \return processorID
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
917
	template<typename ofb> size_t inline processorIDBC(const T (&p)[dim]) const
incardon's avatar
incardon committed
918
	{
919 920 921
		Point<dim,T> pt = p;
		applyPointBC(pt);

incardon's avatar
incardon committed
922
		return fine_s.get(cd.getCell(p));
incardon's avatar
incardon committed
923 924
	}

925 926 927 928 929 930 931 932 933 934
	/*! \brief Get the smallest subdivision of the domain on each direction
	 *
	 * \return a box p1 is set to zero
	 *
	 */
	const ::Box<dim,T> & getSmallestSubdivision()
	{
		return ss_box;
	}

935 936 937 938 939 940 941 942 943 944 945 946
	/*! \brief Get the periodicity on i dimension
	 *
	 * \param i dimension
	 *
	 * \return the periodicity in direction i
	 *
	 */
	size_t isPeriodic(size_t i)
	{
		return bc[i];
	}

incardon's avatar
incardon committed
947 948
	/*! \brief Set the parameter of the decomposition
	 *
949 950
	 * \param div_ storing into how many domain to decompose on each dimension
	 * \param domain_ domain to decompose
incardon's avatar
incardon committed
951 952
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
953
	void setParameters(const size_t (& div_)[dim], ::Box<dim,T> domain_, const size_t (& bc)[dim] ,const Ghost<dim,T> & ghost)
incardon's avatar
incardon committed
954
	{
Pietro Incardona's avatar
Pietro Incardona committed
955 956 957 958
		// set the boundary conditions
		for (size_t i = 0 ; i < dim ; i++)
			this->bc[i] = bc[i];

incardon's avatar
incardon committed
959 960
		// set the ghost
		this->ghost = ghost;
incardon's avatar
incardon committed
961

962
		// Set the decomposition parameters
incardon's avatar
incardon committed
963
		gr.setDimensions(div_);
incardon's avatar
incardon committed
964
		domain = domain_;
965
		cd.setDimensions(domain, div_, 0);
incardon's avatar
incardon committed
966

967
		// init distribution
968
		dist.createCartGraph(gr, domain);
969 970 971

	}

972 973 974 975 976 977 978 979 980 981
	void reset()
	{
		sub_domains.clear();
		box_nn_processor.clear();
		fine_s.clear();
		nn_prcs<dim, T>::reset();
		ie_ghost<dim, T>::reset();
		ie_loc_ghost<dim, T>::reset();
	}

982 983 984 985 986
	/*! \brief Start decomposition
	 *
	 */
	void decompose()
	{
987 988
		reset();

989 990 991
		computeCommunicationAndMigrationCosts(1);

		dist.decompose();
incardon's avatar
incardon committed
992

993
		createSubdomains(v_cl,bc);
994 995

		calculateGhostBoxes();
incardon's avatar
incardon committed
996 997
	}

998 999 1000
	/*! \brief Refine the decomposition, available only for ParMetis distribution, for Metis it is a null call
	 *
	 */
1001
	void rebalance(size_t ts)
1002
	{
1003 1004 1005
		reset();

		computeCommunicationAndMigrationCosts(ts);
1006 1007

		dist.refine();
1008 1009 1010 1011

		createSubdomains(v_cl,bc);

		calculateGhostBoxes();
1012 1013 1014 1015
	}

	/*! \brief Refine the decomposition, available only for ParMetis distribution, for Metis it is a null call
	 *
tonynsyde's avatar
tonynsyde committed
1016
	 * \return true if the re-balance has been executed, false otherwise
1017
	 */
tonynsyde's avatar
tonynsyde committed
1018
	bool rebalance(DLB & dlb)
1019
	{
tonynsyde's avatar
tonynsyde committed
1020 1021 1022 1023 1024 1025 1026
		// if the DLB heuristic to use is the "Unbalance Threshold" get unbalance percentage
		if (dlb.getHeurisitc() == DLB::Heuristic::UNBALANCE_THRLD)
		{
			float unbalance = dist.getUnbalance();
			dlb.setUnbalance(unbalance);
			if (v_cl.getProcessUnitID() == 0)
			{
1027
				std::cout << std::setprecision(3) << unbalance << "\n";
tonynsyde's avatar
tonynsyde committed
1028 1029 1030
			}
		}

1031 1032
		if (dlb.rebalanceNeeded())
		{
1033 1034
			rebalance(dlb.getNTimeStepSinceDLB());

tonynsyde's avatar
tonynsyde committed
1035
			return true;
1036
		}
tonynsyde's avatar
tonynsyde committed
1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055
		return false;
	}

	/*! \brief Get the current un-balance value
	 *
	 * \return the un-balance percentage value
	 */
	float getUnbalance()
	{
		return dist.getUnbalance();
	}

	/*! \brief Compute the processor load counting the total weights of its vertices
	 *
	 * \return the current processor load
	 */
	size_t getProcessorLoad()
	{
		return dist.getProcessorLoad();
1056 1057 1058 1059 1060 1061 1062 1063
	}

	/*! \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
	 *
	 */
tonynsyde's avatar
tonynsyde committed
1064
	inline void getSubSubDomainPosition(size_t id, T (&pos)[dim])
1065
	{
1066
		dist.getSubSubDomainPosition(id, pos);
1067 1068
	}

1069
	//TODO fix in Parmetis distribution to get only the right amount of vertices
tonynsyde's avatar
tonynsyde committed
1070 1071
	/*! \brief Get the number of sub-sub-domains in this sub-graph
	 *
1072
	 * \return number of sub-sub-domains in this sub-graph
tonynsyde's avatar
tonynsyde committed
1073 1074 1075 1076 1077 1078
	 */
	size_t getNSubSubDomains()
	{
		return dist.getNSubSubDomains();
	}

1079 1080 1081 1082 1083 1084 1085
	/*! \brief function that set the weight of the vertex
	 *
	 * \param id vertex id
	 *
	 */
	inline void setSubSubDomainComputationCost(size_t id, size_t weight)
	{
1086
		dist.setComputationCost(id, weight);
1087 1088
	}

tonynsyde's avatar
tonynsyde committed
1089 1090 1091 1092 1093 1094 1095
	/*! \brief function that set the weight of the vertex
	 *
	 * \param id vertex id
	 *
	 */
	inline size_t getSubSubDomainComputationCost(size_t id)
	{
1096
		return dist.getComputationCost(id);
tonynsyde's avatar
tonynsyde committed
1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107
	}

	/*! \brief Operator to access the size of the sub-graph
	 *
	 * \return the size of the subgraph
	 */
	size_t subSize()
	{
		return dist.subSize();
	}

1108
	/*! \brief Get the number of local sub-domains
incardon's avatar
incardon committed
1109 1110 1111 1112
	 *
	 * \return the number of sub-domains
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
1113
	size_t getNSubDomain()
incardon's avatar
incardon committed
1114 1115 1116 1117
	{
		return sub_domains.size();
	}

incardon's avatar
incardon committed
1118
	/*! \brief Get the local sub-domain
incardon's avatar
incardon committed
1119
	 *
incardon's avatar
incardon committed
1120 1121
	 * \param i (each local processor can have more than one sub-domain)
	 * \return the sub-domain
incardon's avatar
incardon committed
1122 1123
	 *
	 */
Pietro Incardona's avatar
Pietro Incardona committed
1124
	SpaceBox<dim, T> getSubDomain(size_t lc)
incardon's avatar
incardon committed
1125 1126
	{
		// Create a space box
1127
		SpaceBox<dim, T> sp;
incardon's avatar
incardon committed
1128 1129 1130

		// fill the space box

1131
		for (size_t k = 0; k < dim; k++)
incardon's avatar
incardon committed
1132 1133
		{
			// create the SpaceBox Low and High
1134 1135
			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
1136 1137 1138 1139 1140
		}

		return sp;
	}

incardon's avatar
incardon committed
1141 1142 1143 1144 1145 1146
	/*! \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
	 *
	 */
1147
	SpaceBox<dim, T> getSubDomainWithGhost(size_t lc)
incardon's avatar
incardon committed
1148