solver.cpp 95.5 KB
Newer Older
Kirill Terekhov's avatar
Kirill Terekhov committed
1
2
3
4
5
6
#include "inmost_solver.h"
#if defined(USE_SOLVER)
#include "solver_petsc.h"
#include "solver_ani.h"
#include "solver_ilu2.hpp"
#include "solver_ddpqiluc2.hpp"
Kirill Terekhov's avatar
Kirill Terekhov committed
7
8
9
#if defined(HAVE_SOLVER_MPTILUC2)
#include "solver_mtiluc2.hpp"
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
10
11
12
#if defined(HAVE_SOLVER_MPTILU2)
#include "solver_mtilu2.hpp"
#endif
13
14
15
16
17
18
#if defined(HAVE_SOLVER_FCBIILU2)
#include "solver_fcbiilu2.h"
#endif
#if defined(HAVE_SOLVER_K3BIILU2)
#include "solver_k3biilu2.h"
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
19
20
21
22
23
24
#include "solver_bcgsl.hpp"
#include <fstream>
#include <sstream>
#include <bitset>
#include <stddef.h>

Kirill Terekhov's avatar
Kirill Terekhov committed
25
26
27
28
29
30
31
32
33
34
35
36
37
38
#if defined(USE_SOLVER_TRILINOS)
#if defined(USE_MPI)
#include "Epetra_MpiComm.h"
#else
#include "Epetra_SerialComm.h"
#endif
#include "Epetra_Map.h"
#include "Epetra_Vector.h"
#include "Epetra_CrsMatrix.h"
#include "AztecOO.h"
#include "Ifpack.h"
#include "ml_include.h"
#include "ml_MultiLevelPreconditioner.h"
#include "BelosEpetraOperator.h"
39
40
41
42
43
44
#include "Teuchos_Comm.hpp"
#include "Teuchos_DefaultMpiComm.hpp"
#include "Teuchos_OpaqueWrapper.hpp"
#include "Teuchos_ParameterList.hpp"
#include "Teuchos_TestForException.hpp"
#include "Teuchos_XMLParameterListHelpers.hpp"
Kirill Terekhov's avatar
Kirill Terekhov committed
45
46
#endif

Kirill Terekhov's avatar
Kirill Terekhov committed
47
48
//#define USE_OMP

Kirill Terekhov's avatar
Kirill Terekhov committed
49
#define KSOLVER BCGSL_solver
Kirill Terekhov's avatar
Kirill Terekhov committed
50
51
//#define KSOLVER BCGS_solver

Kirill Terekhov's avatar
Kirill Terekhov committed
52
53
54

namespace INMOST
{
55
56
57
58
59
60
	static std::string petsc_database_file = "";
	static std::string trilinos_aztec_database_file = "";
	static std::string trilinos_ifpack_database_file = "";
	static std::string trilinos_belos_database_file = "";
	static std::string trilinos_ml_database_file = "";
	static std::string ani_database_file = "";
61
62
	static std::string fcbiilu2_database_file = "";
	static std::string k3biilu2_database_file = "";
63
64


Kirill Terekhov's avatar
Kirill Terekhov committed
65
66
67
68
69
70
71
72
73
#if defined(USE_MPI)
	INMOST_MPI_Type Solver::RowEntryType = MPI_DATATYPE_NULL;
#else
	INMOST_MPI_Type Solver::RowEntryType = 0;
#endif

#define GUARD_MPI(x) {ierr = x; if( ierr != MPI_SUCCESS ) {std::cout << #x << " not successfull " << std::endl; MPI_Abort(comm,-1000);}}
#define HASH_TABLE_SIZE 2048

Kirill Terekhov's avatar
Kirill Terekhov committed
74
75
	bool Solver::is_initialized = false;
	bool Solver::is_finalized = false;
Kirill Terekhov's avatar
Kirill Terekhov committed
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95

	int comparator(const void * pa, const void *pb)
	{
		INMOST_DATA_ENUM_TYPE * a = (INMOST_DATA_ENUM_TYPE *)pa, * b = (INMOST_DATA_ENUM_TYPE *)pb;
		return a[0] - b[0];
	}

	INMOST_DATA_ENUM_TYPE binary_search_pairs(INMOST_DATA_ENUM_TYPE * link, INMOST_DATA_ENUM_TYPE size, INMOST_DATA_ENUM_TYPE find)
	{
		INMOST_DATA_ENUM_TYPE rcur = size >> 1, lcur = 0, mid, chk;
		while( rcur >= lcur )
		{
			mid = lcur + ((rcur-lcur) >> 1);
			chk = mid << 1;
			if( link[chk] < find ) lcur = mid+1;
			else if( link[chk] > find ) rcur = mid-1;
			else return chk;
		}
		return size;
	}
Kirill Terekhov's avatar
Kirill Terekhov committed
96
	void Solver::OrderInfo::Integrate(INMOST_DATA_REAL_TYPE * inout, INMOST_DATA_ENUM_TYPE num) const
Kirill Terekhov's avatar
Kirill Terekhov committed
97
98
	{
#if defined(USE_MPI)
Kirill Terekhov's avatar
Kirill Terekhov committed
99
		if( GetSize() == 1 ) return;
Kirill Terekhov's avatar
Kirill Terekhov committed
100
101
102
103
104
105
106
107
108
#if defined(USE_OMP)
#pragma omp single
#endif
		{
			int ierr = 0;
			dynarray<INMOST_DATA_REAL_TYPE,1024> temp(num);
			memcpy(temp.data(),inout,sizeof(INMOST_DATA_REAL_TYPE)*num);
			GUARD_MPI(MPI_Allreduce(temp.data(),inout,num,INMOST_MPI_DATA_REAL_TYPE,MPI_SUM,comm));
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
109
110
111
#else
		(void) inout;
		(void) num;
Kirill Terekhov's avatar
Kirill Terekhov committed
112
113
114
#endif
	}

Kirill Terekhov's avatar
Kirill Terekhov committed
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
	Solver::RowMerger::RowMerger() : Sorted(true), Nonzeros(0) {}

	Solver::RowMerger::RowMerger(INMOST_DATA_ENUM_TYPE interval_begin, INMOST_DATA_ENUM_TYPE interval_end, bool Sorted) 
				: Sorted(Sorted), Nonzeros(0), LinkedList(interval_begin,interval_end+1,Row::make_entry(UNDEF,0.0)) 
	{
		LinkedList.begin()->first = EOL;
	}

	void Solver::RowMerger::Resize(INMOST_DATA_ENUM_TYPE interval_begin, INMOST_DATA_ENUM_TYPE interval_end, bool _Sorted) 
	{
		LinkedList.set_interval_beg(interval_begin);
		LinkedList.set_interval_end(interval_end+1);
		std::fill(LinkedList.begin(),LinkedList.end(),Row::make_entry(UNDEF,0.0));
		LinkedList.begin()->first = EOL;
		Nonzeros = 0;
		Sorted = _Sorted;
	}

	void Solver::RowMerger::Resize(Matrix & A, bool _Sorted)
	{
		INMOST_DATA_ENUM_TYPE mbeg, mend;
		A.GetInterval(mbeg,mend);
		Resize(mbeg,mend,_Sorted);
	}

	Solver::RowMerger::RowMerger(Matrix & A, bool Sorted) : Sorted(Sorted), Nonzeros(0)
	{
		INMOST_DATA_ENUM_TYPE mbeg, mend;
		A.GetInterval(mbeg,mend);
		LinkedList.set_interval_beg(mbeg);
		LinkedList.set_interval_end(mend+1);
		std::fill(LinkedList.begin(),LinkedList.end(),Row::make_entry(UNDEF,0.0));
		LinkedList.begin()->first = EOL;
	}

	Solver::RowMerger::~RowMerger() {}

	void Solver::RowMerger::Clear()
	{
		INMOST_DATA_ENUM_TYPE i = LinkedList.begin()->first, j;
		LinkedList.begin()->first = EOL;
		while( i != EOL )
		{
			j = LinkedList[i].first;
			LinkedList[i].first = UNDEF; 
			i = j;
		}
		Nonzeros = 0;
	}

	void Solver::RowMerger::PushRow(INMOST_DATA_REAL_TYPE coef, Row & r, bool PreSortRow)
	{
		if( Sorted && PreSortRow ) std::sort(r.Begin(),r.End());
		assert(Nonzeros == 0); //Linked list should be empty
		assert(LinkedList.begin()->first == EOL); //again check that list is empty
		INMOST_DATA_ENUM_TYPE index = LinkedList.get_interval_beg();
		Row::iterator it = r.Begin(), jt;
		while( it != r.End() )
		{
			LinkedList[index].first = it->first+1;
			LinkedList[it->first+1].first = EOL;
			LinkedList[it->first+1].second = it->second*coef;
			index = it->first+1;
			++Nonzeros;
			jt = it;
			++it;
			assert(!Sorted || it == r.End() || jt->first < it->first);
		}
	}

	void Solver::RowMerger::AddRow(INMOST_DATA_REAL_TYPE coef, Row & r, bool PreSortRow)
	{
		if( Sorted && PreSortRow ) std::sort(r.Begin(),r.End());
		INMOST_DATA_ENUM_TYPE index = LinkedList.get_interval_beg(), next;
		Row::iterator it = r.Begin(), jt;
		while( it != r.End() )
		{
			if( LinkedList[it->first+1].first != UNDEF )
				LinkedList[it->first+1].second += coef*it->second;
			else if( Sorted )
			{
				next = index;
				while(next < it->first+1)
				{
					index = next;
					next = LinkedList[index].first;
				}
				assert(index < it->first+1);
				assert(it->first+1 < next);
				LinkedList[index].first = it->first+1;
				LinkedList[it->first+1].first = next;
				LinkedList[it->first+1].second = coef*it->second;
				++Nonzeros;
			}
			else
			{
				LinkedList[it->first+1].first = LinkedList[index].first;
				LinkedList[it->first+1].second = coef*it->second;
				LinkedList[index].first = it->first+1;
				++Nonzeros;
			}
			jt = it;
			++it;
			assert(!Sorted || it == r.End() || jt->first < it->first);
		}
	}

	void Solver::RowMerger::RetriveRow(Row & r)
	{
		r.Resize(Nonzeros);
		INMOST_DATA_ENUM_TYPE i = LinkedList.begin()->first, k = 0;
		while( i != EOL )
		{
Kirill Terekhov's avatar
Kirill Terekhov committed
228
229
230
231
232
233
			if( LinkedList[i].second )
			{
				r.GetIndex(k) = i-1;
				r.GetValue(k) = LinkedList[i].second;
				++k;
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
234
235
			i = LinkedList[i].first;
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
236
		r.Resize(k);
Kirill Terekhov's avatar
Kirill Terekhov committed
237
238
	}

Kirill Terekhov's avatar
Kirill Terekhov committed
239
240
241
242
243
244
245
246
	void Solver::OrderInfo::PrepareMatrix(Matrix & m, INMOST_DATA_ENUM_TYPE overlap)
	{
		have_matrix = true;
		m.isParallel() = true;
		INMOST_DATA_ENUM_TYPE  two[2];
		INMOST_DATA_ENUM_TYPE mbeg,mend;
		int initial_rank;
#if defined(USE_MPI)
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
247
		int ierr = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
248
249
250
251
252
253
254
255
		if( comm != INMOST_MPI_COMM_WORLD )
		{
			MPI_Comm_free(&comm);
			comm = INMOST_MPI_COMM_WORLD;
		}
		if( m.GetCommunicator() == INMOST_MPI_COMM_WORLD )
			comm = INMOST_MPI_COMM_WORLD;
		else MPI_Comm_dup(m.GetCommunicator(), &comm);
Kirill Terekhov's avatar
Kirill Terekhov committed
256
257
258
		MPI_Comm_rank(comm,&rank);
		MPI_Comm_size(comm,&size);
#else
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
259
		(void) overlap;
Kirill Terekhov's avatar
Kirill Terekhov committed
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
		rank = 0;
		size = 1;
#endif
		initial_rank = rank;
		//std::vector<MPI_Request> requests;
		global_overlap.resize(size*2);
		global_to_proc.resize(size+1);
		m.GetInterval(mbeg,mend);
		global_to_proc[0] = 0;
		initial_matrix_begin = mbeg;
		initial_matrix_end = mend;
		two[0] = mbeg;
		two[1] = mend;
#if defined(USE_MPI)
		GUARD_MPI(MPI_Allgather(two,2,INMOST_MPI_DATA_ENUM_TYPE,&global_overlap[0],2,INMOST_MPI_DATA_ENUM_TYPE,comm));
#else
		local_vector_begin = initial_matrix_begin = local_matrix_begin = global_overlap[0] = mbeg;
		local_vector_end   = initial_matrix_end   = local_matrix_end   = global_overlap[1] = mend;
		global_to_proc[1] = mend;
#endif
#if defined(USE_MPI)
		//reorder processors if needed
		{
			//starts of local indexes should appear in asscending order
			bool reorder = false;
			for(int k = 0; k < size-1; k++)
				if( global_overlap[2*k] > global_overlap[2*(k+1)] )
				{
					reorder = true;
					break;
				}
			if( reorder )
			{
				storage_type temp(size*2);
				//assemble array that includes rank
Kirill Terekhov's avatar
Kirill Terekhov committed
295
				for(int k = 0; k < size; ++k)
Kirill Terekhov's avatar
Kirill Terekhov committed
296
297
298
299
300
301
302
303
304
305
				{
					temp[2*k+0] = global_overlap[2*k];
					temp[2*k+1] = k;
				}
				//sort array
				qsort(&temp[0],size,sizeof(INMOST_DATA_ENUM_TYPE)*2,comparator);
				//create new group
				MPI_Group oldg, newg;
				MPI_Comm newcomm;
				std::vector<int> ranks(size);
Kirill Terekhov's avatar
Kirill Terekhov committed
306
307
				for(int k = 0; k < size; ++k)
					ranks[k] = temp[2*k+1];
Kirill Terekhov's avatar
Kirill Terekhov committed
308
309
310
				GUARD_MPI(MPI_Comm_group(comm,&oldg));
				GUARD_MPI(MPI_Group_incl(oldg,size,&ranks[0],&newg));
				GUARD_MPI(MPI_Comm_create(comm,newg,&newcomm));
Kirill Terekhov's avatar
Kirill Terekhov committed
311
312
313
314
				if( comm != INMOST_MPI_COMM_WORLD )
				{
					GUARD_MPI(MPI_Comm_free(&comm));
				}
Kirill Terekhov's avatar
Kirill Terekhov committed
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
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
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
				comm = newcomm;
				//compute new rank
				MPI_Comm_rank(comm,&rank);
				//sort array pairs, so we don't need to exchange them again
				qsort(&global_overlap[0],size,sizeof(INMOST_DATA_ENUM_TYPE)*2,comparator);
			}
			//now check that there are no overlaps of local indexes
			//every mend must be equal to every mbeg
			reorder = false;
			for(int k = 0; k < size-1; k++)
				if( global_overlap[2*k+1] != global_overlap[2*(k+1)] )
				{
					//check that end is strictly greater then begin
					if( global_overlap[2*k+1] < global_overlap[2*(k+1)] )
					{
						if( initial_rank == 0 )
						{
							std::cout << __FILE__ << ":" << __LINE__ << " Matrix index intervals are not complete:";
							std::cout << " processor " << k+0 << " interval " << global_overlap[2*(k+0)] << ":" << global_overlap[2*(k+0)+1];
							std::cout << " processor " << k+1 << " interval " << global_overlap[2*(k+1)] << ":" << global_overlap[2*(k+1)+1];
							std::cout << std::endl;
							MPI_Abort(comm,-1000);
						}
					}
					reorder = true;
				}
			if( reorder )
			{
				storage_type old_overlap(global_overlap);
				//move local bounds to get non-overlapping regions
				for(int k = 0; k < size-1; k++)
					while( global_overlap[2*k+1] > global_overlap[2*(k+1)] )
					{
						//move bounds to equalize sizes
						if( global_overlap[2*k+1] - global_overlap[2*k] < global_overlap[2*(k+1)+1] - global_overlap[2*(k+1)] )
							global_overlap[2*k+1]--; //move right bound of the current processor to left
						else
							global_overlap[2*(k+1)]++; //move left bound of the next processor to right
					}
				
				//TODO: if we need to merge overlapping parts of the matrices - do it here
			}
			local_matrix_begin = global_overlap[2*rank+0];
			local_matrix_end   = global_overlap[2*rank+1];
			for(int k = 0; k < size; k++)
				global_to_proc[k+1] = global_overlap[2*k+1];
		}
		MPI_Status stat;
		INMOST_DATA_ENUM_TYPE ext_pos = local_matrix_end;
		//may replace std::map here
		small_hash<INMOST_DATA_ENUM_TYPE,INMOST_DATA_ENUM_TYPE,HASH_TABLE_SIZE> global_to_local;
		std::vector< std::pair<INMOST_DATA_ENUM_TYPE,INMOST_DATA_ENUM_TYPE> > current_global_to_local;
		std::vector< Row::entry > send_row_data, recv_row_data;
		std::vector< INMOST_DATA_ENUM_TYPE > send_row_sizes, recv_row_sizes;
		std::vector<INMOST_DATA_ENUM_TYPE> incoming(4*size);
		std::vector<MPI_Request> requests;
		INMOST_DATA_ENUM_TYPE total_send = 0, total_recv = 0;
		INMOST_DATA_ENUM_TYPE local_start = local_matrix_begin, local_end = local_matrix_end;
		for(INMOST_DATA_ENUM_TYPE it = 0; it < overlap+1; it++)
		{
			total_send = 0, total_recv = 0;
			current_global_to_local.clear();
			for(INMOST_DATA_ENUM_TYPE k = local_start; k < local_end; ++k)
			{
				Solver::Row & r = m[k];
				INMOST_DATA_ENUM_TYPE jend = r.Size(), ind;
				for(INMOST_DATA_ENUM_TYPE j = 0; j < jend; ++j)
				{
					ind = r.GetIndex(j);
					if( ind < local_matrix_begin || ind >= local_matrix_end) 
					{
						INMOST_DATA_ENUM_TYPE & recv_pos = global_to_local[ind];
						if( recv_pos == 0 ) //this number was not assigned yet
						{
							recv_pos = ext_pos++;
							if( it < overlap ) current_global_to_local.push_back(std::make_pair(ind,recv_pos));
						}
					}
				}
			}
			if( it == overlap ) 
				current_global_to_local = global_to_local.serialize();
			std::sort(current_global_to_local.begin(),current_global_to_local.end());
			//if( !current_global_to_local.empty() )
			{
				//check all the indexes that comes from other processors
				//for every processor we need arrays:
				// processor -> (array of index positions where to receive))
				// processor -> (array of index positions from where to send)
				memset(&incoming[0],0,sizeof(INMOST_DATA_ENUM_TYPE)*size*2);
				vector_exchange_recv.clear();
				vector_exchange_recv.push_back(0);
				if( !current_global_to_local.empty() )
				{
					INMOST_DATA_ENUM_TYPE proc_beg = GetProcessor(current_global_to_local.begin()->first), proc_end = GetProcessor(current_global_to_local.rbegin()->first)+1;
					INMOST_DATA_ENUM_TYPE current_ind = 0;
					for(INMOST_DATA_ENUM_TYPE proc = proc_beg; proc < proc_end; proc++)
					{
						bool first = true;
						INMOST_DATA_ENUM_TYPE numind = static_cast<INMOST_DATA_ENUM_TYPE>(vector_exchange_recv.size() + 1);
						while( current_ind < current_global_to_local.size() && current_global_to_local[current_ind].first < global_to_proc[proc+1] )
						{
							INMOST_DATA_ENUM_TYPE k = current_global_to_local[current_ind].first;
							if( first )
							{
								vector_exchange_recv.push_back(proc);
								vector_exchange_recv.push_back(1);
								vector_exchange_recv.push_back(k);
								first = false;
							}
							else
							{
								vector_exchange_recv[numind]++;
								vector_exchange_recv.push_back(k);
							}
							current_ind++;
						}
						if( !first ) 
						{
							incoming[proc]++;
							incoming[proc+size] += vector_exchange_recv[numind];
							vector_exchange_recv[0]++;
						}
					}
				}
				
				GUARD_MPI(MPI_Allreduce(&incoming[0],&incoming[2*size],size*2,INMOST_MPI_DATA_ENUM_TYPE,MPI_SUM,comm));
				//std::cout << GetRank() << " MPI_Allreduce " << __FILE__ << ":" << __LINE__ << " incoming " << incoming[size*2+rank] << " size " << incoming[size*3+rank] << std::endl;
				//prepare array that helps exchanging vector values
				requests.resize(2*vector_exchange_recv[0] + incoming[size*2+rank]);
				INMOST_DATA_ENUM_TYPE j = 1;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_recv[0]; k++) //send rows that i want to receive
				{
					total_recv += vector_exchange_recv[j+1];
					GUARD_MPI(MPI_Isend(&vector_exchange_recv[j+1],1,INMOST_MPI_DATA_ENUM_TYPE,vector_exchange_recv[j],size+vector_exchange_recv[j],comm,&requests[k])); //send number of rows
					GUARD_MPI(MPI_Isend(&vector_exchange_recv[j+2],vector_exchange_recv[j+1],INMOST_MPI_DATA_ENUM_TYPE,vector_exchange_recv[j],2*size+vector_exchange_recv[j],comm,&requests[k+vector_exchange_recv[0]])); //send row positions
					j += vector_exchange_recv[j+1] + 2;
				}

				recv_row_sizes.resize(incoming[size*3+rank]);
				vector_exchange_send.resize(1+incoming[size*2+rank]*2+incoming[size*3+rank]);
				vector_exchange_send[0] = 0;
				j = 1;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < incoming[size*2+rank]; k++) //receive rows that others want from me
				{
					INMOST_DATA_ENUM_TYPE msgsize;
					GUARD_MPI(MPI_Recv(&msgsize,1,INMOST_MPI_DATA_ENUM_TYPE,MPI_ANY_SOURCE,size+rank,comm,&stat)); //recv number of rows
					vector_exchange_send[j++] = stat.MPI_SOURCE;
					vector_exchange_send[j++] = msgsize;
					//std::cout << GetRank() << " MPI_Irecv size " << msgsize << " rank " << stat.MPI_SOURCE << " tag " << 2*size+rank << __FILE__ << ":" << __LINE__ << std::endl;
					GUARD_MPI(MPI_Irecv(&vector_exchange_send[j],msgsize,INMOST_MPI_DATA_ENUM_TYPE,stat.MPI_SOURCE,2*size+rank,comm,&requests[2*vector_exchange_recv[0]+k])); //recv rows
					j += msgsize;
					total_send += msgsize;
					vector_exchange_send[0]++;
				}
				assert(total_send == incoming[size*3+rank]);
				assert(vector_exchange_send[0] == incoming[size*2+rank]);
				if( 2*vector_exchange_recv[0] + incoming[size*2+rank] > 0 )
					GUARD_MPI(MPI_Waitall(2*vector_exchange_recv[0] + incoming[size*2+rank],&requests[0],MPI_STATUSES_IGNORE));
			}
			/*
			else
			{
				vector_exchange_recv.resize(1,0);
				vector_exchange_send.resize(1,0);
			}
			*/
			if( it == overlap )
			{
				//std::cout << rank << " reorder " << std::endl;
				//now we need to reorder off-diagonal parts of the matrix
				for(INMOST_DATA_ENUM_TYPE k = local_matrix_begin; k < local_end; ++k)
					for(Solver::Row::iterator jt = m[k].Begin(); jt != m[k].End(); ++jt)
						if( global_to_local.is_present(jt->first) ) 
							jt->first = global_to_local[jt->first];
						else
						{
							assert(jt->first >= local_matrix_begin);
							assert(jt->first < local_matrix_end);
						}
				local_vector_begin = local_matrix_begin;
				local_vector_end = ext_pos;
				{ 
					// change indexes for recv array
					INMOST_DATA_ENUM_TYPE i,j = 1,k;
					//for(k = 0; k < GetRank(); k++) MPI_Barrier(comm);
					//std::cout << "rank " << GetRank() << std::endl;
					//std::cout << "recv:" << std::endl;
					for(i = 0; i < vector_exchange_recv[0]; i++)
					{
						//std::cout << "proc " << vector_exchange_recv[j] << " size " << vector_exchange_recv[j+1] << std::endl;
						j++; //skip processor number
						for(k = 0; k < vector_exchange_recv[j]; ++k)
						{
							assert(global_to_local.is_present(vector_exchange_recv[j+k+1]));
							vector_exchange_recv[j+k+1] = global_to_local[vector_exchange_recv[j+k+1]];
							assert(vector_exchange_recv[j+k+1] >= local_matrix_end);
						}
						j+=vector_exchange_recv[j]+1; //add vector length + size position
					}
					//check that indexes in send array are in local matrix bounds
					//std::cout << "send:" << std::endl;
#ifndef NDEBUG
					j = 1;
					for(i = 0; i < vector_exchange_send[0]; i++)
					{
						//std::cout << "proc " << vector_exchange_send[j] << " size " << vector_exchange_send[j+1] << std::endl;
						j++; //skip processor number
						for(k = 0; k < vector_exchange_send[j]; ++k)
						{
							assert(vector_exchange_send[j+k+1] >= local_matrix_begin);
							assert(vector_exchange_send[j+k+1] < local_matrix_end);
						}
						j+=vector_exchange_send[j]+1; //add vector length + size position
					}
#endif
					//for(k = GetRank(); k < GetSize(); k++) MPI_Barrier(comm);
				}
				//prepare array local->global
				extended_indexes.resize(local_vector_end-local_matrix_end);
				for(std::vector< std::pair<INMOST_DATA_ENUM_TYPE,INMOST_DATA_ENUM_TYPE> >::iterator jt = current_global_to_local.begin(); jt != current_global_to_local.end(); ++jt)
					extended_indexes[jt->second-local_matrix_end] = jt->first;

				send_storage.resize(total_send);
				recv_storage.resize(total_recv);
				send_requests.resize(vector_exchange_send[0]);
				recv_requests.resize(vector_exchange_recv[0]);
				
			}
			else
			{
				send_row_sizes.resize(total_send);
				recv_row_sizes.resize(total_recv);
Kirill Terekhov's avatar
Kirill Terekhov committed
548

Kirill Terekhov's avatar
Kirill Terekhov committed
549
550
551
552
553
554
555
556
557
				INMOST_DATA_ENUM_TYPE j = 1, q = 0, f = 0, total_rows_send = 0, total_rows_recv = 0;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_recv[0]; k++) //recv sizes of rows
				{
					GUARD_MPI(MPI_Irecv(&recv_row_sizes[q],vector_exchange_recv[j+1],INMOST_MPI_DATA_ENUM_TYPE,vector_exchange_recv[j],3*size+vector_exchange_recv[j],comm,&requests[k]));
					q += vector_exchange_recv[j+1];
					j += vector_exchange_recv[j+1]+2;
				}
				j = 1;
				q = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
558

Kirill Terekhov's avatar
Kirill Terekhov committed
559
560
561
562
563
564
565
566
567
568
569
570
571
572
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_send[0]; k++) //send sizes of rows
				{
					for(INMOST_DATA_ENUM_TYPE r = 0; r < vector_exchange_send[j+1]; r++)
					{
						send_row_sizes[q+r] = m[vector_exchange_send[j+2+r]].Size();
						total_rows_send += m[vector_exchange_send[j+2+r]].Size();
					}
					GUARD_MPI(MPI_Isend(&send_row_sizes[q],vector_exchange_send[j+1],INMOST_MPI_DATA_ENUM_TYPE,vector_exchange_send[j],3*size+rank,comm,&requests[vector_exchange_recv[0]+k])); //recv rows
					//remember processor numbers here
					q += vector_exchange_send[j+1];
					j += vector_exchange_send[j+1]+2;
				}
				send_row_data.clear();
				send_row_data.reserve(total_rows_send);
Kirill Terekhov's avatar
Kirill Terekhov committed
573
574

				
Kirill Terekhov's avatar
Kirill Terekhov committed
575
576
577
578
579
580
581
				j = 1;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_send[0]; k++) //accumulate data in array
				{
					for(INMOST_DATA_ENUM_TYPE r = 0; r < vector_exchange_send[j+1]; r++)
						send_row_data.insert(send_row_data.end(),m[vector_exchange_send[j+2+r]].Begin(),m[vector_exchange_send[j+2+r]].End());
					j += vector_exchange_send[j+1]+2;
				}
Kirill Terekhov's avatar
Kirill Terekhov committed
582
583

				
Kirill Terekhov's avatar
Kirill Terekhov committed
584
585
586
				//replace by mpi_waitsome
				if( vector_exchange_recv[0]+vector_exchange_send[0] > 0 )
					GUARD_MPI(MPI_Waitall(vector_exchange_recv[0]+vector_exchange_send[0],&requests[0],MPI_STATUSES_IGNORE));
Kirill Terekhov's avatar
Kirill Terekhov committed
587
588

				
Kirill Terekhov's avatar
Kirill Terekhov committed
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
				j = 1;
				q = 0;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_recv[0]; k++) //compute total size of data to receive
				{
					for(INMOST_DATA_ENUM_TYPE r = 0; r < vector_exchange_recv[j+1]; r++)
						total_rows_recv += recv_row_sizes[q+r];
					q += vector_exchange_recv[j+1];
					j += vector_exchange_recv[j+1]+2;
				}
				recv_row_data.resize(total_rows_recv);
				j = 1;
				q = 0;
				f = 0;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_recv[0]; k++) //receive row data
				{
					INMOST_DATA_ENUM_TYPE local_size = 0;
					for(INMOST_DATA_ENUM_TYPE r = 0; r < vector_exchange_recv[j+1]; r++)
						local_size += recv_row_sizes[q+r];
					GUARD_MPI(MPI_Irecv(&recv_row_data[f],local_size,Solver::GetRowEntryType(),vector_exchange_recv[j],4*size+vector_exchange_recv[j],comm,&requests[k]));
					q += vector_exchange_recv[j+1];
					j += vector_exchange_recv[j+1]+2;
					f += local_size;
				}
Kirill Terekhov's avatar
Kirill Terekhov committed
612
613

	
Kirill Terekhov's avatar
Kirill Terekhov committed
614
615
616
617
618
619
620
621
622
623
624
625
626
				j = 1;
				q = 0;
				f = 0;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_send[0]; k++) //receive row data
				{
					INMOST_DATA_ENUM_TYPE local_size = 0;
					for(INMOST_DATA_ENUM_TYPE r = 0; r < vector_exchange_send[j+1]; r++)
						local_size += send_row_sizes[q+r];
					GUARD_MPI(MPI_Isend(&send_row_data[f],local_size,Solver::GetRowEntryType(),vector_exchange_send[j],4*size+rank,comm,&requests[k+vector_exchange_recv[0]]));
					q += vector_exchange_send[j+1];
					j += vector_exchange_send[j+1]+2;
					f += local_size;
				}
Kirill Terekhov's avatar
Kirill Terekhov committed
627
628

	
Kirill Terekhov's avatar
Kirill Terekhov committed
629
630
631
632
633
634
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
				local_start = local_end;
				m.SetInterval(local_matrix_begin,ext_pos);
				local_end = ext_pos;
				if( vector_exchange_recv[0]+vector_exchange_send[0] > 0 )
					GUARD_MPI(MPI_Waitall(vector_exchange_recv[0]+vector_exchange_send[0],&requests[0],MPI_STATUSES_IGNORE));
				j = 1;
				q = 0;
				f = 0;
				for(INMOST_DATA_ENUM_TYPE k = 0; k < vector_exchange_recv[0]; k++) //extend matrix
				{
					for(INMOST_DATA_ENUM_TYPE r = 0; r < vector_exchange_recv[j+1]; r++)
					{
						m[global_to_local[vector_exchange_recv[j+2+r]]] = Solver::Row(&recv_row_data[f],&recv_row_data[f]+recv_row_sizes[q+r]);
						f += recv_row_sizes[q+r];
					}
					q += vector_exchange_recv[j+1];
					j += vector_exchange_recv[j+1]+2;
				}
			}
			//std::cout << it << "/" << overlap << " done" << std::endl;
		}
		two[0] = local_matrix_begin;
		two[1] = local_end;
		GUARD_MPI(MPI_Allgather(two,2,INMOST_MPI_DATA_ENUM_TYPE,&global_overlap[0],2,INMOST_MPI_DATA_ENUM_TYPE,comm));
		//std::cout << __FUNCTION__ << " done" << std::endl;
#endif
	}

	void Solver::OrderInfo::RestoreMatrix(Solver::Matrix & m)
	{
		//restore matrix size
		m.SetInterval(initial_matrix_begin,initial_matrix_end);
		//restore indexes
		for(Solver::Matrix::iterator it = m.Begin(); it != m.End(); ++it)
			for(Solver::Row::iterator jt = it->Begin(); jt != it->End(); ++jt)
				if( jt->first >= initial_matrix_end )
					jt->first = extended_indexes[jt->first-initial_matrix_end];
		m.isParallel() = false;
		have_matrix = false;
Kirill Terekhov's avatar
Kirill Terekhov committed
668
669
670
671
672
673
674
#if defined(USE_MPI)
		if( comm != INMOST_MPI_COMM_WORLD )
		{
			MPI_Comm_free(&comm);
			comm = INMOST_MPI_COMM_WORLD;
		}
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
675
676
677
678
679
680
		//std::cout << __FUNCTION__ << std::endl;
	}

	Solver::OrderInfo::~OrderInfo()
	{
#if defined(USE_MPI)
Kirill Terekhov's avatar
Kirill Terekhov committed
681
682
		if( comm != INMOST_MPI_COMM_WORLD )
			MPI_Comm_free(&comm);
Kirill Terekhov's avatar
Kirill Terekhov committed
683
684
685
#endif
	}
	
Kirill Terekhov's avatar
Kirill Terekhov committed
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
	void Solver::OrderInfo::Clear()
	{
		global_to_proc.clear();
		global_overlap.clear();
		vector_exchange_recv.clear();
		vector_exchange_send.clear();
		send_storage.clear();
		recv_storage.clear();
		send_requests.clear();
		recv_requests.clear();
		extended_indexes.clear();
		local_vector_begin = local_vector_end = 0;
		initial_matrix_begin = initial_matrix_end = 0;
		local_matrix_begin = local_matrix_end = 0;
		have_matrix = false;
	}
Kirill Terekhov's avatar
Kirill Terekhov committed
702
	
Kirill Terekhov's avatar
Kirill Terekhov committed
703
	void Solver::OrderInfo::PrepareVector(Vector & v) const
Kirill Terekhov's avatar
Kirill Terekhov committed
704
705
706
707
708
709
	{
		if( !have_matrix ) throw PrepareMatrixFirst;
		v.SetInterval(local_vector_begin,local_vector_end);
		v.isParallel() = true;
	}
	
Kirill Terekhov's avatar
Kirill Terekhov committed
710
	void Solver::OrderInfo::RestoreVector(Vector & v) const
Kirill Terekhov's avatar
Kirill Terekhov committed
711
712
713
714
715
716
717
718
719
	{
		assert(have_matrix);
		if( v.isParallel() )
		{
			v.SetInterval(initial_matrix_begin,initial_matrix_end);
			v.isParallel() = false;
		}
	}
	
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
720
721
722
723
724
725
726
727
728
729
	Solver::OrderInfo::OrderInfo() : 
	global_to_proc(), 
	global_overlap(),
	vector_exchange_recv(), 
	vector_exchange_send(), 
	send_storage(), 
	recv_storage(), 
	send_requests(), 
	recv_requests(),
	extended_indexes()
Kirill Terekhov's avatar
Kirill Terekhov committed
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
	{
		comm = INMOST_MPI_COMM_WORLD;
		rank = 0;
		size = 1;
		initial_matrix_begin = 0;
		initial_matrix_end = 0;
		local_matrix_begin = 0;
		local_matrix_end = 0;
		local_vector_begin = 0;
		local_vector_end = 0;
		have_matrix = false;
	}
	
	Solver::OrderInfo::OrderInfo(const OrderInfo & other) 
		:global_to_proc(other.global_to_proc), global_overlap(other.global_overlap),
		vector_exchange_recv(other.vector_exchange_recv), vector_exchange_send(other.vector_exchange_send), 
		extended_indexes(other.extended_indexes)
	{
#if defined(USE_MPI)
Kirill Terekhov's avatar
Kirill Terekhov committed
749
750
751
		if( other.comm == INMOST_MPI_COMM_WORLD )
			comm = INMOST_MPI_COMM_WORLD;
		else MPI_Comm_dup(other.comm,&comm);
Kirill Terekhov's avatar
Kirill Terekhov committed
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
#else
		comm = other.comm;
#endif
		rank = other.rank;
		size = other.size;
		initial_matrix_begin = other.initial_matrix_begin;
		initial_matrix_end = other.initial_matrix_end;
		local_vector_begin = other.local_vector_begin;
		local_vector_end = other.local_vector_end;
		local_matrix_begin = other.local_matrix_begin;
		local_matrix_end = other.local_matrix_end;
		have_matrix = other.have_matrix; 
		send_storage.resize(other.send_storage.size());
		recv_storage.resize(other.recv_storage.size());
		send_requests.resize(other.send_requests.size());
		recv_requests.resize(other.recv_requests.size());
	}
	
	Solver::OrderInfo & Solver::OrderInfo::operator =(OrderInfo const & other) 
	{
#if defined(USE_MPI)
Kirill Terekhov's avatar
Kirill Terekhov committed
773
774
775
		if( other.comm == INMOST_MPI_COMM_WORLD )
			comm = INMOST_MPI_COMM_WORLD;
		else MPI_Comm_dup(other.comm,&comm);
Kirill Terekhov's avatar
Kirill Terekhov committed
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
#else
		comm = other.comm;
#endif
		global_to_proc = other.global_to_proc; 
		global_overlap = other.global_overlap; 
		vector_exchange_recv = other.vector_exchange_recv;
		vector_exchange_send = other.vector_exchange_send; 
		extended_indexes = other.extended_indexes;
		rank = other.rank;
		size = other.size;
		initial_matrix_begin = other.initial_matrix_begin;
		initial_matrix_end = other.initial_matrix_end;
		local_vector_begin = other.local_vector_begin;
		local_vector_end = other.local_vector_end;
		local_matrix_begin = other.local_matrix_begin;
		local_matrix_end = other.local_matrix_end;
		have_matrix = other.have_matrix; 
		send_storage.resize(other.send_storage.size());
		recv_storage.resize(other.recv_storage.size());
		send_requests.resize(other.send_requests.size());
		recv_requests.resize(other.recv_requests.size());
		return *this;
	}


	INMOST_DATA_ENUM_TYPE Solver::OrderInfo::GetProcessor(INMOST_DATA_ENUM_TYPE gind) const
	{
		assert(have_matrix);
		storage_type::const_iterator find = std::lower_bound(global_to_proc.begin(),global_to_proc.end(),gind);
		assert(find != global_to_proc.end());
		if( (*find) == gind && find+1 != global_to_proc.end()) return static_cast<INMOST_DATA_ENUM_TYPE>(find - global_to_proc.begin());
		else return static_cast<INMOST_DATA_ENUM_TYPE>(find - global_to_proc.begin())-1;
	}
	void Solver::OrderInfo::GetOverlapRegion(INMOST_DATA_ENUM_TYPE proc, INMOST_DATA_ENUM_TYPE & mbeg, INMOST_DATA_ENUM_TYPE & mend) const
	{
		assert(have_matrix); 
		mbeg = global_overlap[proc*2+0];
		mend = global_overlap[proc*2+1];
	}
	void Solver::OrderInfo::GetLocalRegion(INMOST_DATA_ENUM_TYPE proc, INMOST_DATA_ENUM_TYPE & mbeg, INMOST_DATA_ENUM_TYPE & mend) const
	{
		assert(have_matrix);
		mbeg = global_to_proc[proc+0];
		mend = global_to_proc[proc+1];
	}
	
	Solver::Vector::Vector(std::string _name, INMOST_DATA_ENUM_TYPE start, INMOST_DATA_ENUM_TYPE end, INMOST_MPI_Comm _comm) :data(start,end) 
	{
		comm = _comm;
		name = _name;
		is_parallel = false;
	}
	
	Solver::Vector::Vector(const Vector & other) : data(other.data) 
	{
		comm = other.comm;
		name = other.name;
		is_parallel = other.is_parallel;
	}
	
	Solver::Vector & Solver::Vector::operator =(Vector const & other) 
	{
		comm = other.comm;
		data = other.data; 
		name = other.name; 
		is_parallel = other.is_parallel;
		return *this;
	}
	
	Solver::Vector::~Vector() 
	{
	}
	
	void Solver::OrderInfo::Update(Vector & x)
	{
		//std::cout << __FUNCTION__ << " start" << std::endl;
#if defined(USE_MPI)
Kirill Terekhov's avatar
Kirill Terekhov committed
853
		if( GetSize() == 1 ) return;
Kirill Terekhov's avatar
Kirill Terekhov committed
854
855
856
#if defined(USE_OMP)
#pragma omp single
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
857
		{
Kirill Terekhov's avatar
Kirill Terekhov committed
858
859
860
861
			//use MPI_Put/MPI_Get to update vector
			assert(x.isParallel()); //the vector was prepared
			INMOST_DATA_ENUM_TYPE i, j = 1, k, l = 0;
			int ierr;
Kirill Terekhov's avatar
Kirill Terekhov committed
862
863
			for(i = 0; i < vector_exchange_recv[0]; i++)
			{
Kirill Terekhov's avatar
Kirill Terekhov committed
864
865
				//std::cout << GetRank() << " MPI_Irecv size " << vector_exchange_recv[j+1] << " dest " << vector_exchange_recv[j] << " tag " << vector_exchange_recv[j]*size+rank << std::endl;
				GUARD_MPI(MPI_Irecv(&recv_storage[l],vector_exchange_recv[j+1],INMOST_MPI_DATA_REAL_TYPE,vector_exchange_recv[j],vector_exchange_recv[j]*size+rank,comm,&recv_requests[i]));
Kirill Terekhov's avatar
Kirill Terekhov committed
866
867
868
				l += vector_exchange_recv[j+1];
				j += vector_exchange_recv[j+1] + 2;
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
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
			j = 1, l = 0;
			for(i = 0; i < vector_exchange_send[0]; i++)
			{
				//std::cout << GetRank() << " MPI_Isend size " << vector_exchange_send[j+1] << " dest " << vector_exchange_send[j] << " tag " << rank*size+vector_exchange_send[j] << std::endl;
				for(k = 0; k < vector_exchange_send[j+1]; k++)
					send_storage[l+k] = x[vector_exchange_send[k+j+2]];
				GUARD_MPI(MPI_Isend(&send_storage[l],vector_exchange_send[j+1],INMOST_MPI_DATA_REAL_TYPE,vector_exchange_send[j],rank*size+vector_exchange_send[j],comm,&send_requests[i]));
				l += vector_exchange_send[j+1];
				j += vector_exchange_send[j+1] + 2;
			}
			if( vector_exchange_recv[0] > 0 )
			{
				GUARD_MPI(MPI_Waitall(static_cast<int>(recv_requests.size()),&recv_requests[0],MPI_STATUSES_IGNORE));
				j = 1, l = 0;
				for(i = 0; i < vector_exchange_recv[0]; i++)
				{
					for(k = 0; k < vector_exchange_recv[j+1]; k++)
						x[vector_exchange_recv[k+j+2]] = recv_storage[l+k];
					l += vector_exchange_recv[j+1];
					j += vector_exchange_recv[j+1] + 2;
				}
			}
			if( vector_exchange_send[0] > 0 ) 
			{
				GUARD_MPI(MPI_Waitall(static_cast<int>(send_requests.size()),&send_requests[0],MPI_STATUSES_IGNORE));
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
895
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
896
897
#else
		(void) x;
Kirill Terekhov's avatar
Kirill Terekhov committed
898
899
900
901
902
903
904
#endif
		//std::cout << __FUNCTION__ << " end" << std::endl;
	}
	void Solver::OrderInfo::Accumulate(Vector & x)
	{
		//std::cout << __FUNCTION__ << " start" << std::endl;
#if defined(USE_MPI)
Kirill Terekhov's avatar
Kirill Terekhov committed
905
		if( GetSize() == 1 ) return;
Kirill Terekhov's avatar
Kirill Terekhov committed
906
907
908
#if defined(USE_OMP)
#pragma omp single
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
909
		{
Kirill Terekhov's avatar
Kirill Terekhov committed
910
911
912
913
			//use MPI_Put/MPI_Get to update vector
			assert(x.isParallel()); //the vector was prepared
			INMOST_DATA_ENUM_TYPE i, j = 1, k, l = 0;
			int ierr;
Kirill Terekhov's avatar
Kirill Terekhov committed
914
915
			for(i = 0; i < vector_exchange_send[0]; i++)
			{
Kirill Terekhov's avatar
Kirill Terekhov committed
916
917
				//std::cout << GetRank() << " MPI_Irecv size " << vector_exchange_send[j+1] << " dest " << vector_exchange_send[j] << " tag " << vector_exchange_send[j]*size+rank << std::endl;
				GUARD_MPI(MPI_Irecv(&send_storage[l],vector_exchange_send[j+1],INMOST_MPI_DATA_REAL_TYPE,vector_exchange_send[j],vector_exchange_send[j]*size+rank,comm,&send_requests[i]));
Kirill Terekhov's avatar
Kirill Terekhov committed
918
919
920
				l += vector_exchange_send[j+1];
				j += vector_exchange_send[j+1] + 2;
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
			j = 1, l = 0;
			for(i = 0; i < vector_exchange_recv[0]; i++)
			{
				for(k = 0; k < vector_exchange_recv[j+1]; k++)
					recv_storage[l+k] = x[vector_exchange_recv[k+j+2]];
				//std::cout << GetRank() << " MPI_Isend size " << vector_exchange_recv[j+1] << " dest " << vector_exchange_recv[j] << " tag " << rank*size+vector_exchange_recv[j] << std::endl;
				GUARD_MPI(MPI_Isend(&recv_storage[l],vector_exchange_recv[j+1],INMOST_MPI_DATA_REAL_TYPE,vector_exchange_recv[j],rank*size+vector_exchange_recv[j],comm,&recv_requests[i]));
				l += vector_exchange_recv[j+1];
				j += vector_exchange_recv[j+1] + 2;
			}
			if( vector_exchange_send[0] > 0 )
			{
				//std::cout << GetRank() << " Waitall send " << send_requests.size() << std::endl;
				GUARD_MPI(MPI_Waitall(static_cast<int>(send_requests.size()),&send_requests[0],MPI_STATUSES_IGNORE));
				j = 1, l = 0;
				for(i = 0; i < vector_exchange_send[0]; i++)
				{
					for(k = 0; k < vector_exchange_send[j+1]; k++)
						x[vector_exchange_send[k+j+2]] += send_storage[l+k];
					l += vector_exchange_send[j+1];
					j += vector_exchange_send[j+1] + 2;
				}
			}
			if( vector_exchange_recv[0] > 0 ) 
			{
				//std::cout << GetRank() << " Waitall recv " << recv_requests.size() << std::endl;
				GUARD_MPI(MPI_Waitall(static_cast<int>(recv_requests.size()),&recv_requests[0],MPI_STATUSES_IGNORE));
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
949
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
950
951
#else
		(void) x;
Kirill Terekhov's avatar
Kirill Terekhov committed
952
953
954
#endif
		//std::cout << __FUNCTION__ << " end" << std::endl;
	}
Kirill Terekhov's avatar
Kirill Terekhov committed
955
956
	/*
	void Solver::OrderInfo::ScalarProd(Vector const & left, Vector const & right, INMOST_DATA_ENUM_TYPE index_begin, INMOST_DATA_ENUM_TYPE index_end, INMOST_DATA_REAL_TYPE & sum) const
Kirill Terekhov's avatar
Kirill Terekhov committed
957
	{
Kirill Terekhov's avatar
Kirill Terekhov committed
958
959
960
961
962
963
964
965
966
		INMOST_DATA_INTEGER_TYPE ibeg = index_begin, iend = index_end;
#if defined(USE_OMP)
#pragma omp for reduction(+:sum)
#endif
		for(INMOST_DATA_INTEGER_TYPE i = ibeg; i < iend; ++i)
		{
			sum += left[i]*right[i];
		}
		Integrate(&sum,1);
Kirill Terekhov's avatar
Kirill Terekhov committed
967
	}
Kirill Terekhov's avatar
Kirill Terekhov committed
968
	*/
Kirill Terekhov's avatar
Kirill Terekhov committed
969
970
971
972
	
	INMOST_DATA_REAL_TYPE   Solver::Row::RowVec(Solver::Vector & x) const
	{
		INMOST_DATA_REAL_TYPE ret = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
973
974
		INMOST_DATA_ENUM_TYPE end = Size();
		for(INMOST_DATA_ENUM_TYPE i = 0; i < end; i++) ret = ret + x[GetIndex(i)]*GetValue(i);
Kirill Terekhov's avatar
Kirill Terekhov committed
975
976
977
978
979
		return ret;
	}
	
	void Solver::Matrix::MatVec(INMOST_DATA_REAL_TYPE alpha, Solver::Vector & x, INMOST_DATA_REAL_TYPE beta, Solver::Vector & out) const //y = alpha*A*x + beta * y
	{
Kirill Terekhov's avatar
Kirill Terekhov committed
980
981
		INMOST_DATA_ENUM_TYPE mbeg, mend;
		INMOST_DATA_INTEGER_TYPE ind, imbeg, imend;
Kirill Terekhov's avatar
Kirill Terekhov committed
982
983
984
985
986
987
988
989
990
991
		if( out.Empty() )
		{
			INMOST_DATA_ENUM_TYPE vbeg,vend;
			GetInterval(vbeg,vend);
			out.SetInterval(vbeg,vend);
		}
		//CHECK SOMEHOW FOR DEBUG THAT PROVIDED VECTORS ARE OK
		//~ assert(GetFirstIndex() == out.GetFirstIndex());
		//~ assert(Size() == out.Size());
		GetInterval(mbeg,mend);
Kirill Terekhov's avatar
Kirill Terekhov committed
992
993
		imbeg = mbeg;
		imend = mend;
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
994
#if defined(USE_OMP)
Kirill Terekhov's avatar
Kirill Terekhov committed
995
#pragma omp for private(ind)
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
996
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
997
		for(ind = imbeg; ind < imend; ++ind) //iterate rows of matrix
Kirill Terekhov's avatar
Kirill Terekhov committed
998
			out[ind] = beta * out[ind] + alpha * (*this)[ind].RowVec(x);
Kirill Terekhov's avatar
Kirill Terekhov committed
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
		// outer procedure should update out vector, if needed
	}

	void Solver::Matrix::MatVecTranspose(INMOST_DATA_REAL_TYPE alpha, Solver::Vector & x, INMOST_DATA_REAL_TYPE beta, Solver::Vector & out) const //y = alpha*A*x + beta * y
	{
		INMOST_DATA_ENUM_TYPE mbeg, mend;
		INMOST_DATA_INTEGER_TYPE ind, imbeg, imend;
		if( out.Empty() )
		{
			INMOST_DATA_ENUM_TYPE vbeg,vend;
			GetInterval(vbeg,vend);
			out.SetInterval(vbeg,vend);
		}
		//CHECK SOMEHOW FOR DEBUG THAT PROVIDED VECTORS ARE OK
		//~ assert(GetFirstIndex() == out.GetFirstIndex());
		//~ assert(Size() == out.Size());
		GetInterval(mbeg,mend);
		imbeg = mbeg;
		imend = mend;
		if( beta ) for(Vector::iterator it = out.Begin(); it != out.End(); ++it) (*it) *= beta;
#if defined(USE_OMP)
#pragma omp for private(ind)
#endif
		for(ind = imbeg; ind < imend; ++ind)
		{
			for(Row::const_iterator it = (*this)[ind].Begin(); it != (*this)[ind].End(); ++it)
				out[it->first] += alpha * x[ind] * it->second;
		}
		// outer procedure should update out vector, if needed
Kirill Terekhov's avatar
Kirill Terekhov committed
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
	}
	
	Solver::Matrix::Matrix(std::string _name, INMOST_DATA_ENUM_TYPE start, INMOST_DATA_ENUM_TYPE end, INMOST_MPI_Comm _comm) 
	:data(start,end) 
	{
		is_parallel = false;
		comm = _comm;
		SetInterval(start,end);
		name = _name;
	}
	
	Solver::Matrix::Matrix(const Matrix & other) :data(other.data) 
	{
		comm = other.comm;
		name = other.name;
	}
	
	Solver::Matrix & Solver::Matrix::operator =(Matrix const & other) 
	{
		comm = other.comm;
		data = other.data; 
		name = other.name; return *this;
	}
	
	Solver::Matrix::~Matrix() 
	{
	}
	
	
	void      Solver::Matrix::MoveRows(INMOST_DATA_ENUM_TYPE from, INMOST_DATA_ENUM_TYPE to, INMOST_DATA_ENUM_TYPE size)
	{
		INMOST_DATA_ENUM_TYPE i = to + size, j = from + size;
		if( size > 0 && to != from )
			while( j != from ) data[--j].MoveRow(data[--i]);
	}

	void     Solver::Matrix::Load(std::string file, INMOST_DATA_ENUM_TYPE mbeg, INMOST_DATA_ENUM_TYPE mend)
	{
		char str[16384];
		std::ifstream input(file.c_str());
		if( input.fail() ) throw -1;
		int state = 0, k;
		INMOST_DATA_ENUM_TYPE mat_size, max_lines, row, col, mat_block;
		INMOST_DATA_REAL_TYPE val;
		int size = 1, rank = 0;
#if defined(USE_MPI)
		if( mend == ENUMUNDEF && mbeg == ENUMUNDEF )
		{
			MPI_Comm_rank(GetCommunicator(),&rank);
			MPI_Comm_size(GetCommunicator(),&size);
		}
#endif
		int line = 0;
		while( !input.getline(str,16384).eof() )
		{
			line++;
			k = 0; while( isspace(str[k]) ) k++;
			if( str[k] == '%' || str[k] == '\0' ) continue;
			std::istringstream istr(str+k);
			switch(state)
			{
			case 0: 
				istr >> mat_size >> mat_size >> max_lines; state = 1; 
				mat_block = mat_size/size;
				if( mbeg == ENUMUNDEF ) mbeg = rank*mat_block;
				if( mend == ENUMUNDEF ) 
				{
					if( rank == size-1 ) mend = mat_size;
					else mend = mbeg+mat_block;
				}
				SetInterval(mbeg,mend);
				//~ std::cout << rank << " my interval " << mbeg << ":" << mend << std::endl;
			break;
			case 1: 
				istr >> row >> col >> val;
				row--; col--;
				if( row >= mbeg && row < mend ) data[row][col] = val; 
			break;
			}
		}
		int nonzero = 0;
		for(iterator it = Begin(); it != End(); ++it) nonzero += it->Size();
		//~ std::cout << rank << " total nonzero " << max_lines << " my nonzero " << nonzero << std::endl;
		input.close();
	}


	void     Solver::Vector::Load(std::string file, INMOST_DATA_ENUM_TYPE mbeg, INMOST_DATA_ENUM_TYPE mend)
	{
		char str[16384];
		std::ifstream input(file.c_str());
		if( input.fail() ) throw -1;
		int state = 0, k;
		INMOST_DATA_ENUM_TYPE vec_size, vec_block, ind = 0;
		INMOST_DATA_REAL_TYPE val;
		int size = 1, rank = 0;
#if defined(USE_MPI)
		if( mend == ENUMUNDEF && mbeg == ENUMUNDEF )
		{
			MPI_Comm_rank(GetCommunicator(),&rank);
			MPI_Comm_size(GetCommunicator(),&size);
		}
#endif
		while( !input.getline(str,16384).eof() )
		{
			k = 0; while( isspace(str[k]) ) k++;
			if( str[k] == '%' || str[k] == '\0' ) continue;
			std::istringstream istr(str+k);
			switch(state)
			{
			case 0: 
				istr >> vec_size; state = 1; 
				vec_block = vec_size/size;
				if( mbeg == ENUMUNDEF ) mbeg = rank*vec_block;
				if( mend == ENUMUNDEF ) 
				{
					if( rank == size-1 ) mend = vec_size;
					else mend = mbeg+vec_block;
				}
				SetInterval(mbeg,mend);
			break;
			case 1: 
				istr >> val;
				if( ind >= mbeg && ind < mend ) data[ind] = val;
				ind++;
			break;
			}
		}
		input.close();
	}

	void     Solver::Vector::Save(std::string file)
	{
		INMOST_DATA_ENUM_TYPE vecsize = Size();
		
#if defined(USE_MPI)
		int rank = 0, size = 1;
		{
			MPI_Comm_rank(GetCommunicator(),&rank);
			MPI_Comm_size(GetCommunicator(),&size);
			INMOST_DATA_ENUM_TYPE temp = vecsize;
			MPI_Allreduce(&temp,&vecsize,1,INMOST_MPI_DATA_ENUM_TYPE,MPI_SUM,GetCommunicator());
		}
#endif
		std::stringstream rhs(std::ios::in | std::ios::out);
		rhs << std::scientific;
		rhs.precision(15);
		for(iterator it = Begin(); it != End(); ++it) rhs << *it << std::endl;
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1176
1177
#if defined(USE_MPI) && defined(USE_MPI_FILE) // Use mpi files
		{ 
Kirill Terekhov's avatar
Kirill Terekhov committed
1178
1179
1180
			int ierr;
			MPI_File fh;
			MPI_Status stat;
Kirill Terekhov's avatar
Kirill Terekhov committed
1181
1182
1183
1184
			ierr = MPI_File_open(GetCommunicator(),const_cast<char *>(file.c_str()), MPI_MODE_CREATE | MPI_MODE_DELETE_ON_CLOSE | MPI_MODE_WRONLY, MPI_INFO_NULL, &fh);
      if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			ierr = MPI_File_close(&fh);
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
Kirill Terekhov's avatar
Kirill Terekhov committed
1185
1186
1187
1188
1189
1190
1191
1192
1193
			ierr = MPI_File_open(GetCommunicator(),const_cast<char *>(file.c_str()),MPI_MODE_WRONLY | MPI_MODE_CREATE,MPI_INFO_NULL,&fh);
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			if( rank == 0 )
			{
				std::stringstream header;
				//header << "% vector " << name << std::endl;
				//header << "% is written by INMOST" << std::endl;
				//header << "% by MPI_File_* api" << std::endl;
				header << vecsize << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
1194
				ierr = MPI_File_write_shared(fh,const_cast<char *>(header.str().c_str()),static_cast<int>(header.str().size()),MPI_CHAR,&stat);
Kirill Terekhov's avatar
Kirill Terekhov committed
1195
1196
				if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
1197
			ierr = MPI_File_write_ordered(fh,const_cast<char *>(rhs.str().c_str()),static_cast<int>(rhs.str().size()),MPI_CHAR,&stat);
Kirill Terekhov's avatar
Kirill Terekhov committed
1198
1199
1200
1201
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			ierr = MPI_File_close(&fh);
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
#elif defined(USE_MPI) //USE_MPI alternative
		std::string senddata = rhs.str(), recvdata;
		int sendsize = static_cast<int>(senddata.size());
		std::vector<int> recvsize(size), displ(size);
		MPI_Gather(&sendsize,1,MPI_INT,&recvsize[0],1,MPI_INT,0,GetCommunicator());
		if( rank == 0 )
		{
			int totsize = recvsize[0];
			
			displ[0] = 0;
			for(int i = 1; i < size; i++) 
			{
				totsize += recvsize[i];
				displ[i] = displ[i-1]+recvsize[i-1];
			}
			recvdata.resize(totsize);
		}
		else recvdata.resize(1); //protect from dereferencing null
1220
		MPI_Gatherv(&senddata[0],sendsize,MPI_CHAR,&recvdata[0],&recvsize[0],&displ[0],MPI_CHAR,0,GetCommunicator());
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1221
1222
1223
1224
1225
1226
		if( rank == 0 )
		{
			std::fstream output(file.c_str(),std::ios::out);
			output << vecsize << std::endl;
			output << recvdata;
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
#else
		std::fstream output(file.c_str(),std::ios::out);
		//output << "% vector " << name << std::endl;
		//output << "% is written by INMOST" << std::endl;
		//output << "% by sequential write" << std::endl;
		output << vecsize << std::endl;
		output << rhs.rdbuf();
#endif
	}

	void     Solver::Matrix::Save(std::string file)
	{
		INMOST_DATA_ENUM_TYPE matsize = Size(), nonzero = 0, row = GetFirstIndex()+1;
		
		for(iterator it = Begin(); it != End(); ++it) nonzero += it->Size();
#if defined(USE_MPI)
		int rank = 0, size = 1;
		{
			MPI_Comm_rank(GetCommunicator(),&rank);
			MPI_Comm_size(GetCommunicator(),&size);
			INMOST_DATA_ENUM_TYPE temp_two[2] = {matsize,nonzero}, two[2];
			MPI_Allreduce(temp_two,two,2,INMOST_MPI_DATA_ENUM_TYPE,MPI_SUM,GetCommunicator());
			matsize = two[0];
			nonzero = two[1];
		}
#endif
		std::stringstream mtx(std::ios::in | std::ios::out);
		mtx << std::scientific;
		mtx.precision(15);
		for(iterator it = Begin(); it != End(); ++it)
		{
			for(Row::iterator jt = it->Begin(); jt != it->End(); ++jt)
				mtx << row << " " << jt->first+1 << " " << jt->second << std::endl;
			++row;
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1262
1263
#if defined(USE_MPI) && defined(USE_MPI_FILE) // USE_MPI2?
		{
Kirill Terekhov's avatar
Kirill Terekhov committed
1264
1265
1266
			int ierr;
			MPI_File fh;
			MPI_Status stat;
Kirill Terekhov's avatar
Kirill Terekhov committed
1267
1268
1269
1270
			ierr = MPI_File_open(GetCommunicator(),const_cast<char *>(file.c_str()), MPI_MODE_CREATE | MPI_MODE_DELETE_ON_CLOSE | MPI_MODE_WRONLY, MPI_INFO_NULL, &fh);
      if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			ierr = MPI_File_close(&fh);
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
Kirill Terekhov's avatar
Kirill Terekhov committed
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
			ierr = MPI_File_open(GetCommunicator(),const_cast<char *>(file.c_str()),MPI_MODE_WRONLY | MPI_MODE_CREATE,MPI_INFO_NULL,&fh);
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			if( rank == 0 )
			{
				std::stringstream header;
				header << "%%MatrixMarket matrix coordinate real general" << std::endl;
				header << "% matrix " << name << std::endl;
				header << "% is written by INMOST" << std::endl;
				header << "% by MPI_File_* api" << std::endl;
				header << matsize << " " << matsize << " " << nonzero << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
1281
1282
				//std::string header_data(header.str());
				ierr = MPI_File_write_shared(fh,const_cast<char *>(header.str().c_str()),static_cast<int>(header.str().size()),MPI_CHAR,&stat);
Kirill Terekhov's avatar
Kirill Terekhov committed
1283
1284
				if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
1285
			ierr = MPI_File_write_ordered(fh,const_cast<char *>(mtx.str().c_str()),static_cast<int>(mtx.str().size()),MPI_CHAR,&stat);
Kirill Terekhov's avatar
Kirill Terekhov committed
1286
1287
1288
1289
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
			ierr = MPI_File_close(&fh);
			if( ierr != MPI_SUCCESS ) MPI_Abort(GetCommunicator(),-1);
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
#elif defined(USE_MPI)//USE_MPI alternative
		std::string senddata = mtx.str(), recvdata;
		int sendsize = static_cast<int>(senddata.size());
		std::vector<int> recvsize(size), displ(size);
		MPI_Gather(&sendsize,1,MPI_INT,&recvsize[0],1,MPI_INT,0,GetCommunicator());
		if( rank == 0 )
		{
			int totsize = recvsize[0];
			
			displ[0] = 0;
			for(int i = 1; i < size; i++) 
			{
				totsize += recvsize[i];
				displ[i] = displ[i-1]+recvsize[i-1];
			}
			recvdata.resize(totsize);
		}
		else recvdata.resize(1); //protect from dereferencing null
1308
		MPI_Gatherv(&senddata[0],sendsize,MPI_CHAR,&recvdata[0],&recvsize[0],&displ[0],MPI_CHAR,0,GetCommunicator());
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
		if( rank == 0 )
		{
			std::fstream output(file.c_str(),std::ios::out);
			output << "%%MatrixMarket matrix coordinate real general" << std::endl;
			output << "% matrix " << name << std::endl;
			output << "% is written by INMOST" << std::endl;
			output << "% by MPI_Gather* api and sequential write" << std::endl;
			output << matsize << " " << matsize << " " << nonzero << std::endl;
			output << recvdata;
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
#else
		std::fstream output(file.c_str(),std::ios::out);
		output << "%%MatrixMarket matrix coordinate real general" << std::endl;
		output << "% matrix " << name << std::endl;
		output << "% is written by INMOST" << std::endl;
		output << "% by sequential write " << std::endl;
		output << matsize << " " << matsize << " " << nonzero << std::endl;
		output << mtx.rdbuf();
#endif
	}

1330
	//======================================================================
Kirill Terekhov's avatar
Kirill Terekhov committed
1331
	//SOLVER SOURCE CODE
1332
	//======================================================================
Kirill Terekhov's avatar
Kirill Terekhov committed
1333
1334
	void Solver::Initialize(int * argc, char *** argv, const char * database)
	{
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1335
1336
1337
		(void)database;
		(void)argc;
		(void)argv;
1338
1339
1340
1341
1342
1343
1344
		if( database != NULL )
		{
			std::fstream file(database,std::ios::in);
			char str[4096];
			while( !file.eof() && file.getline(str,4096) )
			{
				int k = 0, l;
1345
				for(k = 0; k < (int)strlen(str); ++k)
1346
1347
1348
1349
				{
					if( str[k] == ':' ) break;
				}
				if( k == strlen(str) ) continue; //invalid line
Kirill Terekhov's avatar
Kirill Terekhov committed
1350
				for(l = 0; l < k; ++l) str[l] = tolower(str[l]);
1351
				l = k+1;
1352
				while(l < (int)strlen(str) && isspace(str[l]) ) ++l;
1353
				if( l == strlen(str) ) continue; //skip empty entry
Kirill Terekhov's avatar
Kirill Terekhov committed
1354
				if( !strncmp(str,"petsc",k) )
1355
					petsc_database_file = std::string(str+l);
Kirill Terekhov's avatar
Kirill Terekhov committed
1356
				else if( !strncmp(str,"trilinos_ifpack",k) )
1357
					trilinos_ifpack_database_file = std::string(str+l);
Kirill Terekhov's avatar
Kirill Terekhov committed
1358
				else if( !strncmp(str,"trilinos_aztec",k) )
1359
					trilinos_aztec_database_file = std::string(str+l);
Kirill Terekhov's avatar
Kirill Terekhov committed
1360
				else if( !strncmp(str,"trilinos_ml",k) )
1361
					trilinos_ml_database_file = std::string(str+l);
Kirill Terekhov's avatar
Kirill Terekhov committed
1362
				else if( !strncmp(str,"trilinos_belos",k) )
1363
					trilinos_belos_database_file = std::string(str+l);
Kirill Terekhov's avatar
Kirill Terekhov committed
1364
				else if( !strncmp(str,"ani",k) )
1365
					ani_database_file = std::string(str+l);
1366
1367
1368
1369
				else if( !strncmp(str,"fcbiilu2",k) )
					fcbiilu2_database_file = std::string(str+l);
				else if( !strncmp(str,"k3biilu2",k) )
					k3biilu2_database_file = std::string(str+l);
1370
1371
1372
1373
			}
		}
		//std::cout << "PETSc \"" << petsc_database_file << "\"" << std::endl;
		//std::cout << "Trilinos_Ifpack \"" << trilinos_ifpack_database_file << "\"" << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
1374
#if defined(USE_SOLVER_PETSC)
1375
		SolverInitializePetsc(argc,argv,petsc_database_file.c_str());
Kirill Terekhov's avatar
Kirill Terekhov committed
1376
1377
#endif
#if defined(USE_SOLVER_ANI)
1378
		SolverInitializeAni(argc,argv,ani_database_file.c_str());
Kirill Terekhov's avatar
Kirill Terekhov committed
1379
#endif
1380
1381
1382
1383
1384
1385
#if defined(HAVE_SOLVER_FCBIILU2)
		SolverInitializeFcbiilu2(argc,argv,fcbiilu2_database_file.c_str());
#endif
#if defined(HAVE_SOLVER_K3BIILU2)
		SolverInitializeK3biilu2(argc,argv,k3biilu2_database_file.c_str());
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1386
1387
1388
#if defined(USE_MPI)
		{
			int flag = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1389
			int ierr = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1390
			MPI_Initialized(&flag);
Kirill Terekhov's avatar
Kirill Terekhov committed
1391
1392
1393
1394
1395
1396
1397
1398
			if( !flag ) 
			{
				ierr = MPI_Init(argc,argv);
				if( ierr != MPI_SUCCESS )
				{
					std::cout << __FILE__ << ":" << __LINE__ << "problem in MPI_Init" << std::endl;
				}
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
1399
1400
1401
1402
1403
1404
			MPI_Datatype type[3] = { INMOST_MPI_DATA_ENUM_TYPE, INMOST_MPI_DATA_REAL_TYPE, MPI_UB};
			int blocklen[3] = { 1, 1, 1 };
			MPI_Aint disp[3];
			disp[0] = offsetof(Solver::Row::entry,first);
			disp[1] = offsetof(Solver::Row::entry,second);
			disp[2] = sizeof(Solver::Row::entry);
Kirill Terekhov's avatar
Kirill Terekhov committed
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
			ierr = MPI_Type_create_struct(3, blocklen, disp, type, &RowEntryType);
			if( ierr != MPI_SUCCESS )
			{
				std::cout << __FILE__ << ":" << __LINE__ << "problem in MPI_Type_create_struct" << std::endl;
			}
			ierr = MPI_Type_commit(&RowEntryType);
			if( ierr != MPI_SUCCESS )
			{
				std::cout << __FILE__ << ":" << __LINE__ << "problem in MPI_Type_commit" << std::endl;
			}
			is_initialized = true;
Kirill Terekhov's avatar
Kirill Terekhov committed
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
		}
#endif
	}
	
	void Solver::Finalize()
	{
#if defined(USE_MPI)
		MPI_Type_free(&RowEntryType);
#endif
#if defined(USE_SOLVER_PETSC)
		SolverFinalizePetsc();
#endif
#if defined(USE_SOLVER_ANI)
		SolverFinalizeAni();
#endif
1431
1432
1433
1434
1435
1436
#if defined(HAVE_SOLVER_FCBIILU2)
		SolverFinalizeFcbiilu2();
#endif
#if defined(HAVE_SOLVER_K3BIILU2)
		SolverFinalizeK3biilu2();
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1437
1438
1439
1440
1441
1442
#if defined(USE_MPI)
		{
			int flag = 0;
			MPI_Finalized(&flag);
			if( !flag ) 
				MPI_Finalize();
Kirill Terekhov's avatar
Kirill Terekhov committed
1443
1444
			is_initialized = false;
			is_finalized = true;
Kirill Terekhov's avatar
Kirill Terekhov committed
1445
1446
1447
1448
1449
		}
#endif
	}
	void Solver::SetParameterEnum(std::string name, INMOST_DATA_ENUM_TYPE val)
	{
Kirill Terekhov's avatar
Kirill Terekhov committed
1450
1451
1452
1453
1454
1455
1456
1457
		if( name == "maximum_iterations" )
			maximum_iterations = val;
		else if( name == "rescale_iterations" )
			preconditioner_rescale_iterations = val;
		else if( name == "condition_estimation" )
			preconditioner_condition_estimation = val;
		else if( name == "adapt_ddpq_tolerance" )
			preconditioner_adapt_ddpq_tolerance = val;
Kirill Terekhov's avatar
Fix    
Kirill Terekhov committed
1458
		else if( name == "schwartz_overlap" )
Kirill Terekhov's avatar
Kirill Terekhov committed
1459
1460
1461
1462
1463
			additive_schwartz_overlap = val;
		else if( name == "gmres_substeps" )
			solver_gmres_substeps = val;
		else if( name == "reorder_nonzeros" )
			preconditioner_reorder_nonzero = val;
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1464
		/* Block below leads to confusion - parameters would be reset by preset values during setup
Kirill Terekhov's avatar
Kirill Terekhov committed
1465
1466
1467
1468
		else if( _pack == INNER_ILU2 || _pack == INNER_MLILUC )
		{
			try
			{
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1469
				//This leads to confusion - 
Kirill Terekhov's avatar
Kirill Terekhov committed
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
				IterativeMethod * method = (IterativeMethod *)solver_data;
				if (name[0] == ':')
					 method->EnumParameter(name.substr(1, name.size() - 1)) = val;
				else if (name == "overlap") additive_schwartz_overlap = val;
				else method->EnumParameter(name) = val;
			} 
			catch(...)
			{
				std::cout << "Parameter " << name << " of intergral type is unknown" << std::endl;
			}
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1481
		*/
Kirill Terekhov's avatar
Kirill Terekhov committed
1482
		else std::cout << "Parameter " << name << " of integral type is unknown" << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
1483
1484
1485
	}
	void Solver::SetParameterReal(std::string name, INMOST_DATA_REAL_TYPE val)
	{
Kirill Terekhov's avatar
Kirill Terekhov committed
1486
1487
1488
1489
		if( name == "absolute_tolerance" )
			absolute_tolerance = val;
		else if( name == "relative_tolerance" )
			relative_tolerance = val;
Kirill Terekhov's avatar
fixes    
Kirill Terekhov committed
1490
1491
		else if( name == "divergence_tolerance" )
			divergence_tolerance = val;
Kirill Terekhov's avatar
Kirill Terekhov committed
1492
1493
1494
1495
1496
1497
		else if( name == "drop_tolerance" )
			preconditioner_drop_tolerance = val;
		else if( name == "reuse_tolerance" )
			preconditioner_reuse_tolerance = val;
		else if( name == "ddpq_tolerance" )
			preconditioner_ddpq_tolerance = val;
Kirill Terekhov's avatar
Fix    
Kirill Terekhov committed
1498
1499
		else if( name == "fill_level" )
			preconditioner_fill_level = val;
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1500
		/* Block below leads to confusion - parameters would be reset by preset values during setup
Kirill Terekhov's avatar
Kirill Terekhov committed
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
		else if( _pack == INNER_ILU2 || _pack == INNER_MLILUC )
		{
			try
			{
				IterativeMethod * method = (IterativeMethod *)solver_data;
				if (name[0] == ':')	method->RealParameter(name.substr(1, name.size() - 1)) = val;
				else method->RealParameter(name) = val;
			} 
			catch(...)
			{
				std::cout << "Parameter " << name << " of real type is unknown" << std::endl;
			}
		}
Kirill Terekhov's avatar
Fixes    
Kirill Terekhov committed
1514
		*/
Kirill Terekhov's avatar
Kirill Terekhov committed
1515
		else std::cout << "Parameter " << name << " of real type is unknown" << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
1516
1517
1518
	}
	Solver::Solver(Type pack, std::string _name, INMOST_MPI_Comm _comm)
	{
Kirill Terekhov's avatar
Kirill Terekhov committed
1519
1520
1521
1522
		additive_schwartz_overlap = DEFAULT_ADDITIVE_SCHWARTZ_OVERLAP;
		maximum_iterations = DEFAULT_MAXIMUM_ITERATIONS;
		absolute_tolerance = DEFAULT_ABSOLUTE_TOLERANCE;
		relative_tolerance = DEFAULT_RELATIVE_TOLERANCE;
Kirill Terekhov's avatar
fixes    
Kirill Terekhov committed
1523
		divergence_tolerance = DEFAULT_DIVERGENCE_TOLERANCE;
Kirill Terekhov's avatar
Kirill Terekhov committed
1524
1525
1526
1527
1528
1529
1530
1531
1532
		preconditioner_ddpq_tolerance = DEFAULT_PRECONDITIONER_DDPQ_TOLERANCE;
		preconditioner_drop_tolerance = DEFAULT_PRECONDITIONER_DROP_TOLERANCE;
		preconditioner_reuse_tolerance = DEFAULT_PRECONDITIONER_REUSE_TOLERANCE;
		preconditioner_reorder_nonzero = DEFAULT_PRECONDITIONER_REORDER_NONZEROS;
		preconditioner_rescale_iterations = DEFAULT_PRECONDITIONER_RESCALE_ITERS;
		preconditioner_fill_level = DEFAULT_PRECONDITIONER_FILL_LEVEL;
		preconditioner_adapt_ddpq_tolerance = DEFAULT_PRECONDITIONER_ADAPT_DDPQ_TOLERANCE;
		preconditioner_condition_estimation = DEFAULT_PRECONDITIONER_CONDITION_ESTIMATION;
		solver_gmres_substeps = DEFAULT_SOLVER_GMRES_SUBSTEPS;
Kirill Terekhov's avatar
Kirill Terekhov committed
1533
1534
1535
1536
		comm = _comm;
		_pack = pack;
		name = _name;
		solver_data = NULL;
Kirill Terekhov's avatar
Kirill Terekhov committed
1537
1538
1539
1540
		local_size = global_size = 0;
		last_it = 0;
		last_resid = 0;
		matrix_data = rhs_data = solution_data = precond_data =  NULL;
Kirill Terekhov's avatar
Kirill Terekhov committed
1541
#if defined(USE_SOLVER_PETSC)
1542
		if( _pack == PETSc )
Kirill Terekhov's avatar
Kirill Terekhov committed
1543
1544
1545
1546
1547
1548
1549
1550
1551
		{
			SolverInitDataPetsc(&solver_data,_comm,name.c_str());
		}
#endif
#if defined(USE_SOLVER_ANI)
		if( _pack == ANI )
		{
			SolverInitDataAni(&solver_data,_comm,name.c_str());
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1552
#endif
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
#if defined(HAVE_SOLVER_FCBIILU2)
		if( _pack == FCBIILU2 )
		{
			SolverInitDataFcbiilu2(&solver_data,_comm,name.c_str());
		}
#endif
#if defined(HAVE_SOLVER_K3BIILU2)
		if( _pack == K3BIILU2 )
		{
			SolverInitDataK3biilu2(&solver_data,_comm,name.c_str());
		}
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1565
1566
1567
1568
1569
1570
#if defined(USE_SOLVER_TRILINOS)
		if( _pack == Trilinos_Aztec || 
			_pack == Trilinos_Belos ||
			_pack == Trilinos_ML ||
			_pack == Trilinos_Ifpack)
		{
1571
1572
1573
1574
			std::pair<std::string,Epetra_LinearProblem> * problem = 
				new std::pair<std::string,Epetra_LinearProblem>;
			problem->first = _name;
			solver_data = static_cast<void *>(problem);
Kirill Terekhov's avatar
Kirill Terekhov committed
1575
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1576
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1577
		if( _pack == INNER_ILU2 || _pack == INNER_DDPQILUC || _pack == INNER_MPTILUC || _pack == INNER_MPTILU2)
Kirill Terekhov's avatar
Kirill Terekhov committed
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
		{
			Method * prec;
			if (_pack == INNER_ILU2)
			{
#if defined(__SOLVER_ILU2__)
				prec = new ILU2_preconditioner(info);
#else
				std::cout << "Sorry, ILU2 preconditioner is not included in this release" << std::endl;
				prec = NULL;
#endif
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
1589
			else if( _pack == INNER_DDPQILUC )
Kirill Terekhov's avatar
Kirill Terekhov committed
1590
1591
1592
1593
1594
1595
			{
#if defined(__SOLVER_DDPQILUC2__)
				prec = new ILUC_preconditioner(info);
#else
				std::cout << "Sorry, multilevel condition estimation crout-ilu preconditioner with reordering for diagonal dominance is not included in this release" << std::endl;
				prec = NULL;
Kirill Terekhov's avatar
Kirill Terekhov committed
1596
1597
1598
1599
1600
1601
1602
1603
1604
#endif
			}
			else if( _pack == INNER_MPTILUC )
			{
#if defined(__SOLVER_MTILUC2__)
				prec = new MTILUC_preconditioner(info);
#else
				std::cout << "Sorry, maximum product transverse condition estimation crout-ilu preconditioner is not included in this release" << std::endl;
				prec = NULL;
Kirill Terekhov's avatar
Kirill Terekhov committed
1605
1606
1607
1608
1609
1610
1611
1612
1613
#endif
			}
			else if( _pack == INNER_MPTILU2 )
			{
#if defined(__SOLVER_MTILU2__)
				prec = new MTILU2_preconditioner(info);
#else
				std::cout << "Sorry, maximum product transverse ilu2 preconditioner is not included in this release" << std::endl;
				prec = NULL;
Kirill Terekhov's avatar
Kirill Terekhov committed
1614
1615
1616
1617
1618
1619
1620
1621
1622
#endif
			}
			solver_data = new KSOLVER(prec, info);
		}
	}
	Solver::Solver(const Solver & other)
	{
		comm = other.comm;
#if defined(USE_SOLVER_PETSC)
1623
		if( _pack == PETSc )
Kirill Terekhov's avatar
Kirill Terekhov committed
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
		{
			SolverCopyDataPetsc(&solver_data,other.solver_data,comm);
			if( other.matrix_data != NULL ) 
			{
				MatrixCopyDataPetsc(&matrix_data,other.matrix_data);
				SolverSetMatrixPetsc(solver_data,matrix_data,false,false);
			}
			if( other.rhs_data != NULL ) 
				VectorCopyDataPetsc(&rhs_data,other.rhs_data);
			if( other.solution_data != NULL ) 
				VectorCopyDataPetsc(&solution_data,other.solution_data);
		}
#endif
#if defined(USE_SOLVER_ANI)
		if( _pack == ANI )
		{
			SolverCopyDataAni(&solver_data,other.solver_data,comm);
			if( other.matrix_data != NULL ) 
			{
				MatrixCopyDataAni(&matrix_data,other.matrix_data);
				SolverSetMatrixAni(solver_data,matrix_data,false,false);
			}
			if( other.rhs_data != NULL ) 
				VectorCopyDataAni(&rhs_data,other.rhs_data);
			if( other.solution_data != NULL ) 
				VectorCopyDataAni(&solution_data,other.solution_data);
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1651
#endif
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
#if defined(HAVE_SOLVER_FCBIILU2)
		if( _pack == FCBIILU2 )
		{
			SolverCopyDataFcbiilu2(&solver_data,other.solver_data,comm);
			if( other.matrix_data != NULL ) 
			{
				MatrixCopyDataFcbiilu2(&matrix_data,other.matrix_data);
				SolverSetMatrixFcbiilu2(solver_data,matrix_data,false,false);
			}
			if( other.rhs_data != NULL ) 
				VectorCopyDataFcbiilu2(&rhs_data,other.rhs_data);
			if( other.solution_data != NULL ) 
				VectorCopyDataFcbiilu2(&solution_data,other.solution_data);
		}
#endif
#if defined(HAVE_SOLVER_K3BIILU2)
		if( _pack == K3BIILU2 )
		{
			SolverCopyDataK3biilu2(&solver_data,other.solver_data,comm);
			if( other.matrix_data != NULL ) 
			{
				MatrixCopyDataK3biilu2(&matrix_data,other.matrix_data);
				SolverSetMatrixK3biilu2(solver_data,matrix_data,false,false);
			}
			if( other.rhs_data != NULL ) 
				VectorCopyDataK3biilu2(&rhs_data,other.rhs_data);
			if( other.solution_data != NULL ) 
				VectorCopyDataK3biilu2(&solution_data,other.solution_data);
		}
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1682
1683
1684
1685
1686
1687
1688
1689
#if defined(USE_SOLVER_TRILINOS)
		if( _pack == Trilinos_Aztec || 
			_pack == Trilinos_Belos ||
			_pack == Trilinos_ML ||
			_pack == Trilinos_Ifpack)
		{
			throw - 1; //You should not really want to copy solver's information
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1690
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1691
		if (_pack == INNER_ILU2 || _pack == INNER_DDPQILUC || _pack == INNER_MPTILUC || _pack == INNER_MPTILU2)
Kirill Terekhov's avatar
Kirill Terekhov committed
1692
		{
Kirill Terekhov's avatar
Kirill Terekhov committed
1693
			throw - 1; //You should not really want to copy solver's information
Kirill Terekhov's avatar
Kirill Terekhov committed
1694
1695
		}
	}
Kirill Terekhov's avatar
Kirill Terekhov committed
1696
	void Solver::Clear()
Kirill Terekhov's avatar
Kirill Terekhov committed
1697
	{
Kirill Terekhov's avatar
Kirill Terekhov committed
1698
		local_size = global_size = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1699
		info.Clear();
Kirill Terekhov's avatar
Kirill Terekhov committed
1700
#if defined(USE_SOLVER_PETSC)
1701
		if( _pack == PETSc )
Kirill Terekhov's avatar
Kirill Terekhov committed
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
		{
			if( matrix_data != NULL ) 
				MatrixDestroyDataPetsc(&matrix_data);
			if( rhs_data != NULL )
				VectorDestroyDataPetsc(&rhs_data);
			if( solution_data != NULL )
				VectorDestroyDataPetsc(&solution_data);
			SolverDestroyDataPetsc(&solver_data);
		}
#endif
#if defined(USE_SOLVER_ANI)
		if( _pack == ANI )
		{
			if( matrix_data != NULL ) 
				MatrixDestroyDataAni(&matrix_data);
			if( rhs_data != NULL )
				VectorDestroyDataAni(&rhs_data);
			if( solution_data != NULL )
				VectorDestroyDataAni(&solution_data);
			SolverDestroyDataAni(&solver_data);
		}
Kirill Terekhov's avatar
Kirill Terekhov committed
1723
#endif