solver_mlmtiluc2.cpp 149 KB
Newer Older
Kirill Terekhov's avatar
Kirill Terekhov committed
1
2
3
4
5
6
#define _CRT_SECURE_NO_WARNINGS
#include "inmost_solver.h"
#if defined(USE_SOLVER)
#include "solver_mlmtiluc2.hpp"
#include <sstream>
#include <deque>
7
#include <iomanip>
8
//#define REPORT_ILU
Kirill Terekhov's avatar
Kirill Terekhov committed
9
//#undef REPORT_ILU
10
11
12
//#define REPORT_ILU_PROGRESS
//#define REPORT_ILU_END
//#define REPORT_ILU_SUMMARY
Kirill Terekhov's avatar
Kirill Terekhov committed
13
//#undef REPORT_ILU_PROGRESS
14
#include "../../../Misc/utils.h"
Kirill Terekhov's avatar
Kirill Terekhov committed
15
16
17
18
19
20
21
22
23
24
25
26
//#define USE_OMP

using namespace INMOST;

#ifndef DEFAULT_TAU
#define DEFAULT_TAU 0.01
#endif


#define REORDER_RCM
//#define REORDER_NNZ
#if defined(USE_SOLVER_METIS)
Kirill Terekhov's avatar
Kirill Terekhov committed
27
#define REORDER_METIS_ND
Kirill Terekhov's avatar
Kirill Terekhov committed
28
29
30
31
32
33
34
35
36
37
38
39
#endif
#if defined(USE_SOLVER_MONDRIAAN)
//#define REORDER_MONDRIAAN
#endif
//#define REORDER_ZOLTAN_HUND


static bool run_mpt = true;
static bool rescale_b = true;
static bool allow_pivot = true;

#define ESTIMATOR
40
//#define ESTIMATOR_REFINE
Kirill Terekhov's avatar
Kirill Terekhov committed
41
42
43
44

//#define PREMATURE_DROPPING

//#define EQUALIZE_1NORM
45
46
//#define EQUALIZE_2NORM
#define EQUALIZE_IDOMINANCE
Kirill Terekhov's avatar
Kirill Terekhov committed
47

Kirill Terekhov's avatar
Kirill Terekhov committed
48
#define PIVOT_THRESHOLD
Kirill Terekhov's avatar
Kirill Terekhov committed
49
50
51
52
53
#define PIVOT_THRESHOLD_VALUE 1.0e-12
//#define DIAGONAL_PERTURBATION
#define DIAGONAL_PERTURBATION_REL 1.0e-7
#define DIAGONAL_PERTURBATION_ABS 1.0e-10
#define ILUC2
Kirill Terekhov's avatar
Kirill Terekhov committed
54
55
#define ILUC2_SCHUR
//#define TRACK_DIAGONAL
Kirill Terekhov's avatar
Kirill Terekhov committed
56
//#define PIVOT_COND_DEFAULT 0.1/tau
57
#define PIVOT_COND_DEFAULT 1.0e+2
Kirill Terekhov's avatar
Kirill Terekhov committed
58
#define PIVOT_DIAG_DEFAULT 1.0e+5
59
60
#define SCHUR_DROPPING_LF
#define SCHUR_DROPPING_EU
61
#define SCHUR_DROPPING_S
62
#define DIAGONAL_PIVOT
Kirill Terekhov's avatar
Kirill Terekhov committed
63
64
65
#define CONDITION_PIVOT

#if defined(REORDER_METIS_ND)
Kirill Terekhov's avatar
Kirill Terekhov committed
66
#define METIS_EXPORT
Kirill Terekhov's avatar
Kirill Terekhov committed
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
#include "metis.h"
#endif
#if defined(REORDER_ZOLTAN_HUND)
#include <zoltan.h>
#endif
#if defined(REORDER_MONDRIAAN)
#include <Mondriaan.h>
#endif




	void MLMTILUC_preconditioner::ReorderEF(INMOST_DATA_ENUM_TYPE wbeg,
											INMOST_DATA_ENUM_TYPE wend,
											interval<INMOST_DATA_ENUM_TYPE, bool> & donePQ,
											interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & localP,
											interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & localQ)
	{
		INMOST_DATA_ENUM_TYPE i, k, l;
		if( !E_Address.empty() )
		{
88
			for(k = wbeg; k < wend; ++k) donePQ[k] = false;
Kirill Terekhov's avatar
Kirill Terekhov committed
89
90
91
92
93
94
95
96
97
98
99
100
			i = wbeg;
			while (i < wend)
			{
				if (donePQ[i]) i++;
				else
				{
					if (localP[i] != i)
					{
						k = i;
						do
						{
							for(l = 0; l < (int)E_Address.size(); ++l)
101
102
103
104
105
							{
								Interval t = E_Address[l]->at(i);
								E_Address[l]->at(i) = E_Address[l]->at(localP[k]);
								E_Address[l]->at(localP[k]) = t;
							}
Kirill Terekhov's avatar
Kirill Terekhov committed
106
107
108
109
110
111
112
113
114
115
							donePQ[localP[k]] = true;
							k = localP[k];
						} while (k != i);
					}
					donePQ[i] = true;
				}
			}
		}
		if( !F_Address.empty() )
		{
116
			for(k = wbeg; k < wend; ++k) donePQ[k] = false;
Kirill Terekhov's avatar
Kirill Terekhov committed
117
118
119
120
121
122
123
124
125
126
127
128
			i = wbeg;
			while (i < wend)
			{
				if (donePQ[i]) i++;
				else
				{
					if (localQ[i] != i)
					{
						k = i;
						do
						{
							for(l = 0; l < (int)F_Address.size(); ++l)
129
130
131
132
133
							{
								Interval t = F_Address[l]->at(i);
								F_Address[l]->at(i) = F_Address[l]->at(localQ[k]);
								F_Address[l]->at(localQ[k]) = t;
							}
Kirill Terekhov's avatar
Kirill Terekhov committed
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
							donePQ[localQ[k]] = true;
							k = localQ[k];
						} while (k != i);
					}
					donePQ[i] = true;
				}
			}
		}
	}


	void MLMTILUC_preconditioner::DumpMatrix(interval<INMOST_DATA_ENUM_TYPE, Interval> & Address,
											 std::vector<Sparse::Row::entry> & Entries,
											 INMOST_DATA_ENUM_TYPE wmbeg, INMOST_DATA_ENUM_TYPE wmend,
											 std::string file_name)
	{
		INMOST_DATA_REAL_TYPE norm1 = 0, norm2 = 0, max = -1.0e54, min = 1.0e54, minabs = 1.0e54;
		INMOST_DATA_REAL_TYPE vrowmax, diag, mindiag = 1.0e54, maxdiag = -1.0e54, maxabsdiag = -1.0e54, minabsdiag = 1.0e54;
		INMOST_DATA_ENUM_TYPE nnz = 0, dominant_rows = 0, dominant_cols = 0, irowmax = 0, nodiag = 0;
		INMOST_DATA_ENUM_TYPE addrbeg = Address.get_interval_beg();
		INMOST_DATA_ENUM_TYPE addrend = Address.get_interval_end();
		interval<INMOST_DATA_ENUM_TYPE,INMOST_DATA_REAL_TYPE> vcolmax(wmbeg,wmend,0);
		interval<INMOST_DATA_ENUM_TYPE,INMOST_DATA_ENUM_TYPE> icolmax(wmbeg,wmend,ENUMUNDEF);
		for (INMOST_DATA_ENUM_TYPE k = wmbeg; k < wmend; ++k) 
		{
			if( k < addrbeg || k >= addrend) continue;
			nnz += Address[k].Size();
			vrowmax = 0;

			bool diag_found = false;
			diag = 0;
165
			for (INMOST_DATA_ENUM_TYPE it = Address[k].first; it < Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
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
			{
				norm1 += fabs(Entries[it].second);
				norm2 += Entries[it].second*Entries[it].second;
				if( fabs(Entries[it].second) > vrowmax ) 
				{
					vrowmax = fabs(Entries[it].second);
					irowmax = Entries[it].first;
				}

				if( fabs(Entries[it].second) > vcolmax[Entries[it].first] ) 
				{
					vcolmax[Entries[it].first] = fabs(Entries[it].second);
					icolmax[Entries[it].first] = k;
				}
				if( Entries[it].second > max ) max = Entries[it].second;
				if( Entries[it].second < min ) min = Entries[it].second;
				if( fabs(Entries[it].second) < fabs(minabs) ) minabs = Entries[it].second;

				if( Entries[it].first == k )
				{
					diag_found = true;
					diag = Entries[it].second;
				}
			}

			if( diag_found )
			{
				if( mindiag > diag ) mindiag = diag;
				if( maxdiag < diag ) maxdiag = diag;
				if( minabsdiag > fabs(diag) ) minabsdiag = fabs(diag);
				if( maxabsdiag < fabs(diag) ) maxabsdiag = fabs(diag);
			}
			else nodiag++;

			if( irowmax == k ) ++dominant_rows;
		}

		for (INMOST_DATA_ENUM_TYPE k = wmbeg; k < wmend; ++k) if( icolmax[k] == k ) ++dominant_cols;

		std::cout << "Writing matrix to " << file_name.c_str() << std::endl;
		std::fstream fout(file_name.c_str(),std::ios::out);
		fout << "%%MatrixMarket matrix coordinate real general" << std::endl;
		fout << "% maximum " << max << std::endl;
		fout << "% minimum " << min << std::endl;
		fout << "% absolute minimum " << minabs << std::endl;
		fout << "% A 1-norm  " << norm1 << std::endl;
		fout << "% A 2-norm  " << sqrt(norm2) << std::endl;
		fout << "% mean 1-norm  " << norm1/(wmend-wmbeg) << std::endl;
		fout << "% mean 2-norm  " << sqrt(norm2/(wmend-wmbeg)) << std::endl;
		fout << "% dominant rows  " << dominant_rows << std::endl;
		fout << "% dominant cols  " << dominant_cols << std::endl;
		fout << "% maximal diagonal value " << maxdiag << std::endl;
		fout << "% minimal diagonal value " << mindiag << std::endl;
		fout << "% absolute maximal diagonal value " << maxabsdiag << std::endl;
		fout << "% absolute minimal diagonal value " << minabsdiag << std::endl;
		fout << "% no diagonal value  " << nodiag << std::endl;
		fout << "% true matrix indices interval " << wmbeg << ":" << wmend << std::endl;
		fout << std::scientific;
		
225
226
		//fout.close(); return;
		
Kirill Terekhov's avatar
Kirill Terekhov committed
227
228
229
230
		fout << wmend-wmbeg << " " << wmend-wmbeg << " " << nnz << std::endl;;
		for (INMOST_DATA_ENUM_TYPE k = wmbeg; k < wmend; ++k)
		{
			if( k < addrbeg || k >= addrend) continue;
231
			for (INMOST_DATA_ENUM_TYPE it = Address[k].first; it < Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
232
233
234
235
236
237
238
239
240
				fout << (k-wmbeg+1) << " " << (Entries[it].first-wmbeg+1) << " " << Entries[it].second << std::endl;
		}
		fout.close();
	}
	void MLMTILUC_preconditioner::CheckOrder(interval<INMOST_DATA_ENUM_TYPE, Interval> & Address,
											 std::vector<Sparse::Row::entry> & Entries,
											 INMOST_DATA_ENUM_TYPE rbeg, INMOST_DATA_ENUM_TYPE rend)
	{
		INMOST_DATA_ENUM_TYPE i,r;
241
		for (i = rbeg; i < rend; i++) if( Address[i].first < Address[i].last )
Kirill Terekhov's avatar
Kirill Terekhov committed
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
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
295
		{
			//check ordered on entry
			bool ordered = true;
			for (r = Address[i].first; r < Address[i].last-1; r++)
			{
				if( !(Entries[r].first < Entries[r+1].first) )
				{
					ordered = false;
					break;
				}
			}
			if( !ordered )
			{
				std::cout << "Row " << i << " not ordered: " << std::endl;
				std::cout << "Interval: " << Address[i].first << ":" << Address[i].last << std::endl;
				for (r = Address[i].first; r < Address[i].last; r++)
				{
					std::cout << "(" << Entries[r].first << "," << Entries[r].second << ") ";
				}
				std::cout << std::endl;
				throw -1;
			}
			bool nan = false;
			for (r = Address[i].first; r < Address[i].last; r++)
			{
				if( Entries[r].second != Entries[r].second )
				{
					nan = true;
					break;
				}
			}
			if( nan )
			{
				std::cout << "Row " << i << " contains nan: " << std::endl;
				std::cout << "Interval: " << Address[i].first << ":" << Address[i].last << std::endl;
				for (r = Address[i].first; r < Address[i].last; r++)
				{
					std::cout << "(" << Entries[r].first << "," << Entries[r].second << ") ";
				}
				std::cout << std::endl;
				throw -1;
			}
		}
	}
	
	void MLMTILUC_preconditioner::inversePQ(INMOST_DATA_ENUM_TYPE wbeg,
											INMOST_DATA_ENUM_TYPE wend,
											interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & localP,
											interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & localQ,
											interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & invP,
											interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & invQ)
	{
		//inverse reordering
		// in invPQ numbers indicate where to get current column
296
297
		for (INMOST_DATA_ENUM_TYPE k = wbeg; k < wend; ++k) invP[localP[k]] = k;
		for (INMOST_DATA_ENUM_TYPE k = wbeg; k < wend; ++k) invQ[localQ[k]] = k;
Kirill Terekhov's avatar
Kirill Terekhov committed
298
299
300
301
302
303
304
305
306
307
	}
	void MLMTILUC_preconditioner::applyPQ(INMOST_DATA_ENUM_TYPE wbeg,
										  INMOST_DATA_ENUM_TYPE wend,
										  interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & localP,
										  interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & localQ,
										  interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & invP,
										  interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & invQ)
	{
		INMOST_DATA_ENUM_TYPE k;
		// compute reordering in global P,Q, we need it to compute reordering in vector during solve phase
308
		for (k = wbeg; k < wend; ++k)
Kirill Terekhov's avatar
Kirill Terekhov committed
309
310
311
312
313
		{
			localP[k] = ddP[invP[k]];
			localQ[k] = ddQ[invQ[k]];
		}
		// update reordering in global P,Q
314
		for (k = wbeg; k < wend; ++k)
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
		{
			ddP[k] = localP[k];
			ddQ[k] = localQ[k];
		}
	}
	INMOST_DATA_ENUM_TYPE & MLMTILUC_preconditioner::EnumParameter(std::string name)
	{
		if (name == "scale_iters") return sciters;
		else if( name == "estimator" ) return estimator;
		throw - 1;
	}
	INMOST_DATA_REAL_TYPE & MLMTILUC_preconditioner::RealParameter(std::string name)
	{
		if (name == "tau") return tau;
		else if( name == "tau2" ) return iluc2_tau;
		else if( name == "pivot_cond" ) return pivot_cond;
		else if( name == "pivot_diag" ) return pivot_diag;
		else if( name == "condition_number_L" ) return condestL;
		else if( name == "condition_number_U" ) return condestU;
		throw - 1;
	}
	void MLMTILUC_preconditioner::Copy(const Method * other)
	{
		const MLMTILUC_preconditioner * b = dynamic_cast<const MLMTILUC_preconditioner *>(other);
		assert(b != NULL);
		tau = b->tau;
		iluc2_tau = b->iluc2_tau;
		pivot_cond = b->pivot_cond;
		pivot_diag = b->pivot_diag;
		Alink = b->Alink;
		info = b->info;
		sciters = b->sciters;
		eps = b->eps;
	}
	MLMTILUC_preconditioner::MLMTILUC_preconditioner(const MLMTILUC_preconditioner & other) :Method(other)
	{
		Copy(&other);
	}
	MLMTILUC_preconditioner & MLMTILUC_preconditioner::operator =(MLMTILUC_preconditioner const & other)
	{
		Copy(&other);
		return *this;
	}
	MLMTILUC_preconditioner::MLMTILUC_preconditioner(Solver::OrderInfo & info)
		:tau(DEFAULT_TAU),info(&info)
	{
		Alink = NULL;
		init = false;
		sciters = 8;
		eps = 1e-54;
#if defined(ESTIMATOR)
		estimator = 1;
#else
		estimator = 0;
#endif
		tau = 1.0e-3;
		iluc2_tau = tau*tau;
		pivot_cond = PIVOT_COND_DEFAULT;
		pivot_diag = PIVOT_DIAG_DEFAULT;
	}
	bool MLMTILUC_preconditioner::isInitialized() { return init; }
	bool MLMTILUC_preconditioner::isFinalized() { return !init; }
	bool MLMTILUC_preconditioner::Initialize()
	{
		
380
        const INMOST_DATA_REAL_TYPE subst = 1.0; (void)subst;
Kirill Terekhov's avatar
Kirill Terekhov committed
381
382
		const INMOST_DATA_REAL_TYPE tol_modif = PIVOT_THRESHOLD_VALUE;
		const INMOST_DATA_ENUM_TYPE UNDEF = ENUMUNDEF, EOL = ENUMUNDEF - 1;
383
		bool block_pivot = false;
Kirill Terekhov's avatar
Kirill Terekhov committed
384
385
386
387
388
389
390
391
		
		INMOST_DATA_ENUM_TYPE wbeg, wend; //working interval
		INMOST_DATA_ENUM_TYPE mobeg, moend; // total interval
		INMOST_DATA_ENUM_TYPE vbeg, vend; // vector interval
		
		INMOST_DATA_ENUM_TYPE k, i, j, Li, Ui, curr, next;
		INMOST_DATA_REAL_TYPE l,u,udiag, abs_udiag, max_diag = 0, min_diag = 0;
		INMOST_DATA_REAL_TYPE max_diag_old, min_diag_old;
392
		INMOST_DATA_ENUM_TYPE nzA, nzLU = 0, nzA0;
Kirill Terekhov's avatar
Kirill Terekhov committed
393
		Sparse::Vector DL, DR;
394
		Sparse::Vector DL0,DR0;
Kirill Terekhov's avatar
Kirill Terekhov committed
395
396
397
398
399
400
401
402
		info->GetOverlapRegion(info->GetRank(), mobeg, moend);
		info->GetVectorRegion(vbeg,vend);
		
		//prepare temporal array
		temp.set_interval_beg(vbeg);
		temp.set_interval_end(vend);
		//for(interval<INMOST_DATA_ENUM_TYPE,INMOST_DATA_REAL_TYPE>::iterator rit = temp.begin();
		//	rit != temp.end(); ++rit) *rit = 0.0;
403
		for(k = vbeg; k < vend; ++k) temp[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
404
405
406
407
408
409
410
411
412
413
414
415
416

		
		//prepare reordering vectors
		ddP.set_interval_beg(mobeg);
		ddP.set_interval_end(moend);
		ddQ.set_interval_beg(mobeg);
		ddQ.set_interval_end(moend);

		

		//prepare rescaling vectors
		DL.SetInterval(mobeg, moend);
		DR.SetInterval(mobeg, moend);
417
418
		DL0.SetInterval(mobeg, moend);
		DR0.SetInterval(mobeg, moend);
419
420
421
422
423
		for(k = mobeg; k < moend; ++k) DL[k] = DR[k] = DL0[k] = DR0[k] = 1.0;
		//for(Sparse::Vector::iterator ri = DL.Begin(); ri != DL.End(); ++ri) *ri = 1.0;
		//for(Sparse::Vector::iterator ri = DR.Begin(); ri != DR.End(); ++ri) *ri = 1.0;
		//for(Sparse::Vector::iterator ri = DL0.Begin(); ri != DL0.End(); ++ri) *ri = 1.0;
		//for(Sparse::Vector::iterator ri = DR0.Begin(); ri != DR0.End(); ++ri) *ri = 1.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
424
425
426

		

427
		for (k = mobeg; k < moend; k++) ddP[k] = ddQ[k] = k;
Kirill Terekhov's avatar
Kirill Terekhov committed
428
		// supplementary data for ddPQ reordering
429
		interval<INMOST_DATA_ENUM_TYPE, bool> donePQ(mobeg, moend, false);
Kirill Terekhov's avatar
Kirill Terekhov committed
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> invP(mobeg, moend), invQ(mobeg,moend), localP(mobeg, moend,ENUMUNDEF), localQ(mobeg,moend,ENUMUNDEF);		
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> Bstart(mobeg,moend);
		interval<INMOST_DATA_ENUM_TYPE, bool> Pivot(mobeg,moend,false);

		//supplimentary data structures for ILUC
		INMOST_DATA_ENUM_TYPE LU_Beg, Sbeg;
		U_Address.set_interval_beg(mobeg);
		U_Address.set_interval_end(moend);
		L_Address.set_interval_beg(mobeg);
		L_Address.set_interval_end(moend);
		B_Address.set_interval_beg(mobeg);
		B_Address.set_interval_end(moend);
		LU_Diag.set_interval_beg(mobeg);
		LU_Diag.set_interval_end(moend);
		std::vector<Sparse::Row::entry> A_Entries, S_Entries, LF_Entries, LFt_Entries;
445
		std::vector<Sparse::Row::entry> EU_Entries;
Kirill Terekhov's avatar
Kirill Terekhov committed
446
447
448
		//std::vector<INMOST_DATA_ENUM_TYPE> sort_indeces;
		interval<INMOST_DATA_ENUM_TYPE, Interval> A_Address(mobeg, moend), S_Address(mobeg,moend);
		interval<INMOST_DATA_ENUM_TYPE, Interval> LF_Address(mobeg,moend), LFt_Address(mobeg,moend);
449
		interval<INMOST_DATA_ENUM_TYPE, Interval> EU_Address(mobeg,moend);
Kirill Terekhov's avatar
Kirill Terekhov committed
450
451
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> Ulist(mobeg, moend), Llist(mobeg, moend), Blist(mobeg,moend), Flist(mobeg,moend);
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> Ubeg(mobeg, moend,EOL), Lbeg(mobeg, moend,EOL), Bbeg(mobeg,moend,EOL), Fbeg(mobeg,moend);
452
453
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> Scolnorm(mobeg,moend,0.0);
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> Scolnum(mobeg,moend,0.0);
Kirill Terekhov's avatar
Kirill Terekhov committed
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

		INMOST_DATA_REAL_TYPE NuU = 1, NuL = 1, NuD = 1, NuU_max = 1.0, NuL_max = 1.0;
		INMOST_DATA_REAL_TYPE NuU_acc = 1, NuL_acc = 1, NuD_acc = 1;
#if defined(ESTIMATOR)
		//supplimentary data structures for condition estimates of L^{-1}, U^{-1}
		INMOST_DATA_REAL_TYPE mup, mum, smup, smum, NuL1 = 1, NuL2 = 1, NuU1 = 1, NuU2 = 1;
		INMOST_DATA_REAL_TYPE NuU1_old = 1, NuL1_old = 1, NuU2_old = 1, NuL2_old = 1, NuD_old = 1;
		INMOST_DATA_REAL_TYPE NuU1_new = 1, NuL1_new = 1, vp, vm, v;
#if defined(ESTIMATOR_REFINE)
		INMOST_DATA_ENUM_TYPE np, nm;
		INMOST_DATA_REAL_TYPE NuU2_new, NuL2_new;
#endif
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> EstL1(mobeg, moend,0.0), EstU1(mobeg, moend,0.0);
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> EstL2(mobeg, moend,0.0), EstU2(mobeg, moend,0.0);
#endif
#if defined(ESTIMATOR)
		INMOST_DATA_REAL_TYPE NuU_tmp, NuL_tmp;
#endif
		//supplimentary data structures for returning values of dropped elements
		//INMOST_DATA_REAL_TYPE DropLk, DropUk;
		//interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> DropU(mobeg,moend,0.0), DropL(mobeg,moend,0.0);
		//data structure for linked list

		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> U(mobeg,moend,std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> V(mobeg,moend,std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
479
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> Dist(mobeg,moend+1,std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
Kirill Terekhov's avatar
Kirill Terekhov committed
480
481
482
		
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> LineValuesU(mobeg, moend,0.0), LineValuesL(mobeg,moend,0.0);
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> LineIndecesU(mobeg, moend+1,UNDEF), LineIndecesL(mobeg,moend+1,UNDEF);
483
484
		
		std::vector<INMOST_DATA_ENUM_TYPE> indicesU, indicesL, indicesS;
485
        double tfactor = 0.0, trescale = 0.0, treorder = 0.0, ttransversal = 0.0, treassamble = 0.0, ttotal, tt, testimator = 0.0, tschur = 0.0, tlocal;
Kirill Terekhov's avatar
Kirill Terekhov committed
486
487
488
489
490
491
492
493
494
#if defined(REORDER_METIS_ND)
		double tmetisgraph = 0, tmetisnd = 0;
#endif
#if defined(REORDER_RCM)
		double trcmgraph = 0, trcmorder = 0;
#endif
        double tlfactor, tlrescale, tlreorder, tlreassamble;
		ttotal = Timer();

495
		//(*Alink).Save("M.mtx");
Kirill Terekhov's avatar
Kirill Terekhov committed
496
497
498
499
500
501
502
503
		
		//calculate number of nonzeros
		nzA = 0;
		for (k = mobeg; k < moend; ++k)
		{
			for (Sparse::Row::iterator r = (*Alink)[k].Begin(); r != (*Alink)[k].End(); ++r)
				if (r->first >= mobeg && r->first < moend && fabs(r->second) > 0.0) nzA++;
		}
504
		nzA0 = nzA;
Kirill Terekhov's avatar
Kirill Terekhov committed
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
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593

		//sort_indeces.reserve(256);
		A_Entries.resize(nzA);
		//B_Entries.reserve(nzA);
		//LU_Entries.reserve(nzA*4);

#if defined(REORDER_NNZ)
		wgt_coords sort_wgts(2*(moend - mobeg));
#endif

#if defined(REORDER_ZOLTAN_HUND)
#endif

#if defined(REORDER_MONDRIAAN)
		{
			sparsematrix sp;
			int numparts = 8;
			int * cols = new int[nzA];
			int * rows = new int[Alink->Size()+1];
			double * values = new double[nzA];
			int cnt = 0;
			rows[0] = 0;
			for (k = mobeg; k < moend; ++k)
			{
				for (Sparse::Row::iterator r = (*Alink)[k].Begin(); r != (*Alink)[k].End(); ++r)
				{
					if (r->first >= mobeg && r->first < moend && fabs(r->second) > 0.0)
					{
						cols[cnt] = r->first;
						values[cnt] = r->second;
						++cnt;
					}
				}
				rows[k-mobeg+1] = cnt;
			}
			CRSSparseMatrixInit(&sp,Alink->Size(),Alink->Size(),nzA,cols,rows,values,0);
			
			PstartInit(&sp,numparts);
			opts options;
			
			SetDefaultOptions(&options);
			//SetOption(&options,"SplitMethod","KLFM");
			SetOption(&options,"Permute","BBD");
			SetOption(&options,"SplitStrategy","onedimrow");
			DistributeMatrixMondriaan(&sp,numparts,0.1,&options,NULL);
			sp.

			j = 0;
			for (k = mobeg; k < moend; ++k)
			{
				A_Address[k].first = j;
				for (Sparse::Row::iterator r = (*Alink)[sp.row_perm[k-mobeg]].Begin(); r != (*Alink)[sp.row_perm[k-mobeg]].End(); ++r)
				{
					if (r->first >= mobeg && r->first < moend && r->second != 0.0)
						A_Entries[j++] = Sparse::Row::make_entry(sp.col_perm_inv[r->first-mobeg], r->second);
				}
				A_Address[k].last = j;
				assert(A_Address[k].Size() != 0); //singular matrix
			}
			//DumpMatrix(A_Address,A_Entries,mobeg,moend,"mondriaan.mtx");
			/*
			sp.MMTypeCode[0] = 'M';
			FILE * f = fopen("mondriaan.mtx","w");
			MMWriteSparseMatrix(&sp,f,"mondriaan",&options);
			fclose(f);
			*/
			MMDeleteSparseMatrix(&sp);
			delete [] cols;
			delete [] rows;
			delete [] values;
		}
#else
		
		j = 0;
		for (k = mobeg; k < moend; ++k)
		{
			A_Address[k].first = j;
			for (Sparse::Row::iterator r = (*Alink)[k].Begin(); r != (*Alink)[k].End(); ++r)
			{
				if (r->first >= mobeg && r->first < moend && r->second != 0.0)
					A_Entries[j++] = Sparse::Row::make_entry(r->first, r->second);
			}
			A_Address[k].last = j;
			//assert(A_Address[k].Size() != 0); //singular matrix
		}
#endif
		//DumpMatrix(A_Address, A_Entries, mobeg, moend, "A.mtx");

		std::vector<INMOST_DATA_REAL_TYPE> C_Entries(A_Entries.size());
594
		INMOST_DATA_REAL_TYPE Unorm, Unum, Lnorm, Lnum;
595
596
597
		INMOST_DATA_REAL_TYPE LFnorm, LFnum, LFmax, LFmin, LFtau, LFdrop;
		INMOST_DATA_REAL_TYPE EUnorm, EUnum, EUmax, EUmin, EUtau, EUdrop;
		INMOST_DATA_REAL_TYPE Snorm, Snum, Smax, Smin, Stau, Sdrop;
Kirill Terekhov's avatar
Kirill Terekhov committed
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619

#if defined(ILUC2)
		INMOST_DATA_ENUM_TYPE nzLU2 = 0, nzLU2tot = 0, ndrops = 0;
		INMOST_DATA_REAL_TYPE tau2 = iluc2_tau;
		std::vector<Sparse::Row::entry> LU2_Entries;
		interval<INMOST_DATA_ENUM_TYPE, Interval> L2_Address(mobeg, moend+1), U2_Address(mobeg, moend+1);
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> U2list(mobeg, moend, UNDEF), L2list(mobeg, moend, UNDEF);
		interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> U2beg(mobeg, moend, EOL), L2beg(mobeg, moend, EOL);
		//LU2_Entries.reserve(nzA*4);
#else
		INMOST_DATA_REAL_TYPE tau2 = tau;
#endif

		for(k = mobeg; k < moend; ++k)
		{
			U_Address[k].first = U_Address[k].last = ENUMUNDEF;
			L_Address[k].first = L_Address[k].last = ENUMUNDEF;
			U2_Address[k].first = U2_Address[k].last = ENUMUNDEF;
			L2_Address[k].first = L2_Address[k].last = ENUMUNDEF;
		}

#if defined(REPORT_ILU)
620
		std::cout << "nonzeros in A " << nzA << " pivot cond " << pivot_cond << " diag " << pivot_diag << " tau " << tau << " tau2 " << tau2 << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
621
622
623
624
625
626
627
628
629
630
#endif
		
		int swaps = 0, totswaps = 0;
		//set up working interval
		wbeg = mobeg;
		wend = moend;
		while (wbeg < wend) //iterate into levels until all matrix is factored
		{
			//ddPQ reordering on current Schur complement
			INMOST_DATA_ENUM_TYPE cbeg = wbeg, cend = wend; //next size of factored B block
631
632
			
		//DumpMatrix(A_Address,A_Entries,cbeg,cend,"mat"+to_string(level_size.size())+".mtx");
Kirill Terekhov's avatar
Kirill Terekhov committed
633
634
635
636
637
///////////////////////////////////////////////////////////////////////////////////
///   MAXIMUM TRANSVERSE REORDERING                                             ///
///////////////////////////////////////////////////////////////////////////////////
			if( run_mpt )
			{
638
639
640
#if defined(REPORT_ILU)
				printf("Reordering with MPT\n");
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
				ttransversal = Timer();
				INMOST_DATA_ENUM_TYPE ColumnBegin;
				INMOST_DATA_ENUM_TYPE pop_heap_pos;
				INMOST_DATA_REAL_TYPE ShortestPath, AugmentPath;
				INMOST_DATA_ENUM_TYPE PathEnd, Trace, IPermPrev;
				//Sparse::Vector & U = DL;
				//Sparse::Vector & V = DR;
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_REAL_TYPE> Cmax(wbeg,wend,0.0);
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & Perm = localQ;
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & IPerm = localP;
				//interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & UpdateStack = Ulist;
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & ColumnList = Llist;//(wbeg,wend,ENUMUNDEF);
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> & Parent = Blist;
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> AugmentPosition(wbeg,wend,ENUMUNDEF);
				interval<INMOST_DATA_ENUM_TYPE, INMOST_DATA_ENUM_TYPE> ColumnPosition(wbeg,wend,ENUMUNDEF);
				
657
658
				
				BinaryHeap Heap(&Dist[wbeg],wend-wbeg);
Kirill Terekhov's avatar
Kirill Terekhov committed
659
660
661
///////////////////////////////////////////////////////////////////////////////////
///  Arrays initialization                                                      ///
///////////////////////////////////////////////////////////////////////////////////
662
663
664
				// arrays U,V,Dist are cleared at the end of schur complement calculation
				//std::fill(U.begin() + wbeg - mobeg, U.begin() + wend - mobeg, std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
				//std::fill(V.begin() + wbeg - mobeg, V.begin() + wend - mobeg, std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
Kirill Terekhov's avatar
Kirill Terekhov committed
665
				//std::fill(Cmax.begin() + wbeg - mobeg, Cmax.begin() + wend - mobeg, 0.0);
666
				//std::fill(Dist.begin() + wbeg - mobeg, Dist.begin() + wend + 1 - mobeg, std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
667
668
669
670
671
672
673
				for(k = wbeg; k < wend; ++k)
				{
					Perm[k] = ENUMUNDEF;
					IPerm[k] = ENUMUNDEF;
					Parent[k] = ENUMUNDEF;
					ColumnList[k] = ENUMUNDEF;
				}
Kirill Terekhov's avatar
Kirill Terekhov committed
674
675
676
677
678
679
680
				C_Entries.resize(A_Entries.size());
///////////////////////////////////////////////////////////////////////////////////
///       Initial LOG transformation to dual problem and initial extreme match  ///
///////////////////////////////////////////////////////////////////////////////////
				//double T = Timer();
				for(k = wbeg; k < wend; ++k)
				{
681
					for (INMOST_DATA_ENUM_TYPE it = A_Address[k].first; it < A_Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
682
683
684
685
686
687
688
689
690
691
					{
						i = A_Entries[it].first;
						u = C_Entries[it] = fabs(A_Entries[it].second);
						if( u > Cmax[i] ) Cmax[i] = u;
						//C_Entries.push_back(Sparse::Row::make_entry(i,u));
					}
				}

				for(k = wbeg; k < wend; ++k)
				{
692
					for (INMOST_DATA_ENUM_TYPE it = A_Address[k].first; it < A_Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
					{
						i = A_Entries[it].first;
						if( Cmax[i] == 0 || C_Entries[it] == 0 )
							C_Entries[it] = std::numeric_limits<INMOST_DATA_REAL_TYPE>::max();
						else
						{
							C_Entries[it] = log(Cmax[i])-log(C_Entries[it]);
							//C_Entries[it] = log10(Cmax[i])-log10(C_Entries[it]);
							//C_Entries[it] = fabs(log10(Cmax[i]/C_Entries[it]));
							//C_Entries[it] = fabs(log(Cmax[i]/C_Entries[it]));
							if( C_Entries[it] < U[i] ) U[i] = C_Entries[it];
						}
					}
				}
				for(k = wbeg; k < wend; ++k)
				{
709
					for (INMOST_DATA_ENUM_TYPE it = A_Address[k].first; it < A_Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
710
711
712
713
714
715
716
717
718
					{
						u = C_Entries[it] - U[A_Entries[it].first];
						if( u < V[k] ) V[k] = u;
					}
				}

///////////////////////////////////////////////////////////////////////////////////
///                  Update cost and match                                      ///
///////////////////////////////////////////////////////////////////////////////////
719
				
Kirill Terekhov's avatar
Kirill Terekhov committed
720
721
				for(k = wbeg; k < wend; ++k)
				{
722
					for (INMOST_DATA_ENUM_TYPE it = A_Address[k].first; it < A_Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
723
724
					{
						u = fabs(C_Entries[it] - V[k] - U[A_Entries[it].first]);
725
						if( u < 1.0e-8 && Perm[A_Entries[it].first] == ENUMUNDEF && IPerm[k] == ENUMUNDEF )
Kirill Terekhov's avatar
Kirill Terekhov committed
726
727
728
729
730
731
732
						{
							 Perm[A_Entries[it].first] = k;
							 IPerm[k] = A_Entries[it].first;
							 ColumnPosition[k] = it;
						}
					}
				}
733
				
Kirill Terekhov's avatar
Kirill Terekhov committed
734
735
736
737
738
739
740
741
///////////////////////////////////////////////////////////////////////////////////
/// 1-step augmentation                                                         ///
///////////////////////////////////////////////////////////////////////////////////

				for(k = wbeg; k < wend; ++k)
				{
					if( IPerm[k] == ENUMUNDEF ) //unmatched row
					{
742
						for (INMOST_DATA_ENUM_TYPE it = A_Address[k].first; it < A_Address[k].last && IPerm[k] == ENUMUNDEF; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
743
744
						{
							u = fabs(C_Entries[it] - V[k] - U[A_Entries[it].first]);
745
							if( u <= 1.0e-8 )
Kirill Terekhov's avatar
Kirill Terekhov committed
746
747
748
749
							{
								Li = Perm[A_Entries[it].first];
								assert(Li != ENUMUNDEF);
								// Search other row in C for 0
750
								for (INMOST_DATA_ENUM_TYPE Lit = A_Address[Li].first; Lit < A_Address[Li].last; ++Lit)
Kirill Terekhov's avatar
Kirill Terekhov committed
751
752
								{
									u = fabs(C_Entries[Lit]- V[Li] - U[A_Entries[Lit].first]);
753
									if( u <= 1.0e-8 && Perm[A_Entries[Lit].first] == ENUMUNDEF )
Kirill Terekhov's avatar
Kirill Terekhov committed
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
									{
										Perm[A_Entries[it].first] = k;
										IPerm[k] = A_Entries[it].first;
										ColumnPosition[k] = it;
										Perm[A_Entries[Lit].first] = Li;
										IPerm[Li] = A_Entries[Lit].first;
										ColumnPosition[Li] = Lit;
										break;
									}
								}
							}
						}
					}
				}

///////////////////////////////////////////////////////////////////////////////////
///             Weighted bipartite matching                                     ///
///////////////////////////////////////////////////////////////////////////////////
				for(k = wbeg; k < wend; ++k)
				{
					if( IPerm[k] != ENUMUNDEF )
						continue;
					Li = k;
					ColumnBegin = EOL;
					Parent[Li] = ENUMUNDEF;
					PathEnd = ENUMUNDEF;
					Trace = k;
					ShortestPath = 0;
					AugmentPath = std::numeric_limits<INMOST_DATA_REAL_TYPE>::max();
					while(true)
					{
785
						for (INMOST_DATA_ENUM_TYPE Lit = A_Address[Li].first; Lit < A_Address[Li].last; ++Lit)
Kirill Terekhov's avatar
Kirill Terekhov committed
786
787
788
789
						{
							Ui = A_Entries[Lit].first;
							//if( ColumnList[Ui] == k ) continue;
							if( ColumnList[Ui] != ENUMUNDEF ) continue;
790
							l = fabs(ShortestPath + C_Entries[Lit] - V[Li] - U[Ui]);
Kirill Terekhov's avatar
Kirill Terekhov committed
791
							//if( l < 0.0 ) printf("row %d col %d negative l %g Augment %lf Shortest %lf C %lf V %lf U %lf\n",k,Ui,l,AugmentPath,ShortestPath,C_Entries[Lit],V[Li],U[Ui]);
792
							if( l < 0.0 && l > -1.0e-8 ) l = 0;
Kirill Terekhov's avatar
Kirill Terekhov committed
793
794
795
796
797
798
799
800
801
802
803
804
805
806
							if( l < 0.0 ) continue;
							if( l < AugmentPath )
							{
								if( Perm[Ui] == ENUMUNDEF )
								{
									PathEnd = Ui;
									Trace = Li;
									AugmentPath = l;
									AugmentPosition[Ui] = Lit;
								}
								else if( l < Dist[Ui] )
								{
									Parent[Perm[Ui]] = Li;
									AugmentPosition[Ui] = Lit;
807
808
809
810
									if( Heap.Contains(Ui-wbeg) )
										Heap.DecreaseKey(Ui-wbeg,l);
									else
										Heap.PushHeap(Ui-wbeg,l);
Kirill Terekhov's avatar
Kirill Terekhov committed
811
812
813
814
815
816
817
818
819
820
821
822
823
								}
							}
						}

						pop_heap_pos = Heap.PopHeap();
						if( pop_heap_pos == ENUMUNDEF ) break;
					
						Ui = pop_heap_pos+wbeg;
						ShortestPath = Dist[Ui];

						if( AugmentPath <= ShortestPath ) 
						{
							Dist[Ui] = std::numeric_limits<INMOST_DATA_REAL_TYPE>::max();
824
							//Heap.increaseKey(Ui,Dist[Ui]);
Kirill Terekhov's avatar
Kirill Terekhov committed
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
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
							break;
						}


						ColumnList[Ui] = ColumnBegin;
						ColumnBegin = Ui;

						Li = Perm[Ui];
						
					}
					if( PathEnd != ENUMUNDEF )
					{
						Ui = ColumnBegin;
						while(Ui != EOL)
						{
							U[Ui] += Dist[Ui] - AugmentPath;
							if( Perm[Ui] != ENUMUNDEF ) V[Perm[Ui]] = C_Entries[ColumnPosition[Perm[Ui]]] - U[Ui];
							Dist[Ui] = std::numeric_limits<INMOST_DATA_REAL_TYPE>::max();
							Li = ColumnList[Ui];
							ColumnList[Ui] = ENUMUNDEF;
							Ui = Li;
						}

						Ui = PathEnd;
						while(Trace != ENUMUNDEF)
						{
							IPermPrev = IPerm[Trace];
							Perm[Ui] = Trace;
							IPerm[Trace] = Ui;

							ColumnPosition[Trace] = AugmentPosition[Ui];
							V[Trace] = C_Entries[ColumnPosition[Trace]] - U[Ui];

							Ui = IPermPrev;
							Trace = Parent[Trace];

						}
						Heap.Clear();
					}
				}
				//printf("Maximum product transversal %lf\n",Timer()-T);
				//fclose(rec);

				//T = Timer();
				for (k = cbeg; k < cend; ++k)
				{
					if( V[k] == std::numeric_limits<INMOST_DATA_REAL_TYPE>::max() ) l = 1;
					else l = exp(V[k]);
					if( U[k] == std::numeric_limits<INMOST_DATA_REAL_TYPE>::max() || Cmax[k] == 0 ) u = 1;
					else u = exp(U[k])/Cmax[k];
					//if( l != l || fabs(l) < 1.0e-12 || isnan(l) || isinf(l) ) std::cout << "k " << k << " l is " << l << " V " << V[k] << std::endl;
					//if( u != u || fabs(u) < 1.0e-12 || isnan(u) || isinf(u) ) std::cout << "k " << k << " u is " << u << " U " << U[k] << " Cmax " << Cmax[k] << std::endl;
					DL[k] = l;
					DR[k] = u;
					

					bool flip_sign = false;
882
					for (INMOST_DATA_ENUM_TYPE jt = A_Address[k].first; jt < A_Address[k].last; ++jt)
Kirill Terekhov's avatar
Kirill Terekhov committed
883
884
885
					{
						i = A_Entries[jt].first;
						j = Perm[A_Entries[jt].first];
886
						
Kirill Terekhov's avatar
Kirill Terekhov committed
887
888
						if( U[i] == std::numeric_limits<INMOST_DATA_REAL_TYPE>::max() || Cmax[i] == 0 ) u = 1;
						else u = exp(U[i])/Cmax[i];
889
890
891
892
893
894
895
896
						
						if( fabs(l*A_Entries[jt].second*u) > 1 + 1.0e-7 )
							std::cout << "element on row " << k << " col " << A_Entries[jt].first << " value " << A_Entries[jt].second << " u " << u << " l " << l << " U " << U[i] << " V " << V[k] << " Cmax " << Cmax[k] << " scaled " << l*A_Entries[jt].second*u << std::endl;
						if( j == k )
						{
							
							if( l*A_Entries[jt].second*u < 0.0 ) flip_sign = true;
						}
Kirill Terekhov's avatar
Kirill Terekhov committed
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
					}

					if( flip_sign ) DL[k] *= -1;
				}
				//printf("Reorder matrix %lf\n",Timer()-T);
				/*
				{
					std::stringstream s;
					s << "DL" << level_size.size() << ".mtx";
					DL.Save(s.str());
				}
				{
					std::stringstream s;
					s << "DR" << level_size.size() << ".mtx";
					DR.Save(s.str());
				}
				 */
				/*
				{
					std::stringstream s;
					s << "MPT" << level_size.size() << ".txt";
					std::fstream fout(s.str().c_str(),std::ios::out);
					for(k = cbeg; k < cend; ++k)
					{
						fout << k << " " << DL[k] << " " << DR[k] << " " << localQ[k] << std::endl;
					}
					fout.close();
				}
				*/
926
927
				//std::fill(localP.begin() + (wbeg - mobeg), localP.begin() + (wend - mobeg), ENUMUNDEF);
				for(k = wbeg; k < wend; ++k) localP[k] = ENUMUNDEF;
Kirill Terekhov's avatar
Kirill Terekhov committed
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
                
                { //check that there are no gaps in Perm
                    for(k = cbeg; k < cend; ++k)
                    {
                        if( Perm[k] != ENUMUNDEF )
						{
							assert(localP[Perm[k]] == ENUMUNDEF);
                            localP[Perm[k]] = 0;
						}
                    }
					std::vector<INMOST_DATA_ENUM_TYPE> gaps;
                    for(k = cbeg; k < cend; ++k)
                        if( localP[k] == ENUMUNDEF )
                            gaps.push_back(k);
					
					//std::cout << "@ gaps: " << gaps.size() << std::endl;
                    
                    for(k = cbeg; k < cend; ++k)
                        if( Perm[k] == ENUMUNDEF )
                        {
                            Perm[k] = gaps.back();
                            gaps.pop_back();
                        }
951
952
                    //std::fill(localP.begin() + (wbeg - mobeg), localP.begin() + (wend - mobeg), ENUMUNDEF);
                    for(k = wbeg; k < wend; ++k) localP[k] = ENUMUNDEF;
Kirill Terekhov's avatar
Kirill Terekhov committed
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
                }
				
				
				
				//exit(-1);

				ttransversal = Timer() - ttransversal;

				treorder += ttransversal;
			}
			else
			{
				for(k = wbeg; k < wend; ++k)
				{
					localQ[k] = k;
					DL[k] = DR[k] = 1;
				}
			}
///////////////////////////////////////////////////////////////////////////////////
///                  END MAXIMUM TRANSVERSE REORDERING                          ///
///////////////////////////////////////////////////////////////////////////////////
			tt = Timer();
#if defined(REORDER_METIS_ND)
			tt = Timer();
			idx_t nvtxs = wend-wbeg;
			std::vector<idx_t> xadj(nvtxs+1), adjncy, perm(nvtxs),iperm(nvtxs);
			//adjncy.reserve(nzA*2);
980
981
982
#if defined(REPORT_ILU)
			printf("Reordering with METIS\n");
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
983
984
985
986
987
988
			
			for (k = cbeg; k < cend; ++k) if( localQ[k] == ENUMUNDEF ) printf("%s:%d No column permutation for row %d. Matrix is structurally singular\n",__FILE__,__LINE__,k);

			for (k = cbeg; k < cend; ++k) localP[k] = k;
			for (k = cbeg; k < cend; ++k) V[localQ[k]] = DR[k]; //if you expirience a bug here then the matrix is structurally singular
			for (k = cbeg; k < cend; ++k) DR[k] = V[k];
989
			
990
991
992
993
994
995
996
			for (k = cbeg; k < cend; ++k)
			{
				if( !(localQ[k] >= cbeg && localQ[k] < cend) )
					std::cout << "Bad permutation: " << localQ[k] << " interval [" << cbeg << ":" << cend << "]" << std::endl;
				V[localQ[k]] = DR0[k]; //if you expirience a bug here then the matrix is structurally singular
			}
			for (k = cbeg; k < cend; ++k) DR0[k] = V[k];
Kirill Terekhov's avatar
Kirill Terekhov committed
997
998
999
1000
1001

			
			
			for(k = wbeg; k < wend; ++k)
			{
1002
				for (INMOST_DATA_ENUM_TYPE jt = A_Address[k].first; jt < A_Address[k].last; ++jt)
Kirill Terekhov's avatar
Kirill Terekhov committed
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
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
				{
					A_Entries[jt].first = localQ[A_Entries[jt].first];
					//A_Entries[jt].second *= DL[k]*DR[A_Entries[jt].first];
					
				}
				std::sort(A_Entries.begin()+A_Address[k].first,A_Entries.begin()+A_Address[k].last);
			}
			ReorderEF(wbeg, wend, donePQ, localP, localQ);
			inversePQ(wbeg,wend,localP,localQ, invP,invQ);
			applyPQ(wbeg, wend, localP, localQ, invP, invQ);

			//DumpMatrix(A_Address,A_Entries,wbeg,wend,"a.mtx");

///////////////////////////////////////////////////////////////////////////////////
//       setup indices for transposed traversal of B block                       //
///////////////////////////////////////////////////////////////////////////////////
			
			tmetisgraph = Timer();

			for (k = wend; k > wbeg; --k)
			{
				//vwgt[k-1] = 1;
				if (A_Address[k-1].Size() > 0)
				{
					i = A_Entries[A_Address[k-1].first].first;
					Blist[k-1] = Bbeg[i];
					Bbeg[i] = k-1;
				}
				Ulist[k-1] = A_Address[k-1].first;
			}
			

			xadj[0] = 0;
			for(i = wbeg; i < wend; ++i)
			{
1038
				for (INMOST_DATA_ENUM_TYPE jt = A_Address[i].first; jt < A_Address[i].last; ++jt)
Kirill Terekhov's avatar
Kirill Terekhov committed
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
				{
					if( A_Entries[jt].first != i )
						adjncy.push_back(static_cast<idx_t>(A_Entries[jt].first-wbeg));
				}
				
				Li = Bbeg[i];
				while (Li != EOL)
				{
					if( Li != i )
						adjncy.push_back(static_cast<idx_t>(Li-wbeg));
					Li = Blist[Li];
				}
				
				Li = Bbeg[i];
				while (Li != EOL)
				{
					Ui = Blist[Li];
					Ulist[Li]++;
					if (A_Address[Li].last - Ulist[Li] > 0)
					{
						k = A_Entries[Ulist[Li]].first;
						Blist[Li] = Bbeg[k];
						Bbeg[k] = Li;
					}

					Li = Ui;
				}
				
				std::sort(adjncy.begin()+xadj[i-wbeg],adjncy.end());
				adjncy.resize(std::unique(adjncy.begin()+xadj[i-wbeg],adjncy.end())-adjncy.begin());
				
				xadj[i-wbeg+1] = static_cast<idx_t>(adjncy.size());
			}

1073
1074
			//std::fill(Bbeg.begin(),Bbeg.end(),EOL);
			for(i = mobeg; i < moend; ++i) Bbeg[i] = EOL;
Kirill Terekhov's avatar
Kirill Terekhov committed
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

			tmetisgraph = Timer() - tmetisgraph;
			/*
			{
				FILE * f = fopen("metis.mtx","w");
				fprintf(f,"%%MatrixMarket matrix coordinate real general\n");
				fprintf(f,"%d %d %d\n",nvtxs,nvtxs,adjncy.size());
				for(i = wbeg; i < wend; ++i)
				{
					for(j = xadj[i]; j < xadj[i+1]; ++j)
						fprintf(f,"%d %d 1\n",i+1,adjncy[j]+1);
				}
				fclose(f);
			}
			*/
			//A_Address[0].first = 0;
			//for(k = wbeg+1; k < wend; ++k)
			//	A_Address[k].first = A_Address[k-1].last;
			if( !adjncy.empty() )
			{
				idx_t options[METIS_NOPTIONS];
				METIS_SetDefaultOptions(options);
				options[METIS_OPTION_NUMBERING] = 0;
				
				//options[METIS_OPTION_DBGLVL] = METIS_DBG_INFO | METIS_DBG_TIME | METIS_DBG_COARSEN | METIS_DBG_CONNINFO | METIS_DBG_CONTIGINFO | METIS_DBG_IPART | METIS_DBG_MEMORY | METIS_DBG_MOVEINFO | METIS_DBG_REFINE | METIS_DBG_SEPINFO;
				//options[METIS_OPTION_CTYPE] = METIS_CTYPE_RM;
				//options[METIS_OPTION_RTYPE] = METIS_RTYPE_SEP2SIDED;
				//options[METIS_OPTION_IPTYPE] = METIS_IPTYPE_NODE;
				//options[METIS_OPTION_PTYPE] = METIS_PTYPE_RB;
				//options[METIS_OPTION_NO2HOP] = 0;
				//options[METIS_OPTION_MINCONN] = 0;
				//options[METIS_OPTION_NITER] = 4;
				//printf("before metis\n");
				//METIS_NodeNDP(nvtxs,&xadj[0],&adjncy[0],NULL,4,options,&perm[0],&iperm[0],&sizes[0]);
				tmetisnd = Timer();
				METIS_NodeND(&nvtxs,&xadj[0],&adjncy[0],NULL,options,&perm[0],&iperm[0]);
				tmetisnd = Timer()-tmetisnd;
				//printf("after metis\n");
				for(k = wbeg; k < wend; ++k)
				{
					//localP[k] = iperm[k-wbeg]+wbeg;
					//localQ[k] = iperm[k-wbeg]+wbeg;
					localP[perm[k-wbeg]+wbeg] = k;
					localQ[perm[k-wbeg]+wbeg] = k;
				}
			}
			else
			{
				for(k = wbeg; k < wend; ++k)
				{
					localP[k] = k;
					localQ[k] = k;
				}
			}
			cend = wend;
			i = wend;

#elif defined(REORDER_NNZ)
			//for (k = cbeg; k < cend; ++k) V[k] = DR[localQ[k]];
			//for (k = cbeg; k < cend; ++k) DR[k] = V[k];

			//DumpMatrix(A_Address,A_Entries,cbeg,cend,"mat_original.mtx");
1137
1138
			//std::fill(U.begin() + wbeg - mobeg, U.begin() + wend - mobeg, 0.0);
			//std::fill(V.begin() + wbeg - mobeg, V.begin() + wend - mobeg, 0.0);
Kirill Terekhov's avatar
Kirill Terekhov committed
1139
1140
			

1141
1142
1143
1144
1145
1146
1147
1148
			//std::fill(Ulist.begin() + wbeg - mobeg, Ulist.begin() + wend - mobeg, 0);
			//std::fill(Llist.begin() + wbeg - mobeg, Llist.begin() + wend - mobeg, 0);
			
			for(k = wbeg; k < wend; ++k)
			{
				U[k] = V[k] = 0.0;
				Ulist[k] = Llist[k] = 0;
			}
Kirill Terekhov's avatar
Kirill Terekhov committed
1149
1150
1151
1152
1153

			for (k = wbeg; k < wend; ++k) Blist[localQ[k]] = k;

			for (k = wbeg; k < wend; ++k)
			{
1154
				for (INMOST_DATA_ENUM_TYPE it = A_Address[k].first; it < A_Address[k].last; ++it)
Kirill Terekhov's avatar
Kirill Terekhov committed
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
				{
					i = A_Entries[it].first;
					u = fabs(A_Entries[it].second);
					//row-col norms
					U[k] += u; //row norm
					V[i] += u; //col norm
					//nonzeros
					Ulist[k]++; //row nnz
					Llist[i]++; //col nnz

					if( Blist[i] == k ) temp[k] = u;
				}
			}

			sort_wgts.clear();
			for (k = wbeg; k < wend; ++k)
                sort_wgts.push_back(wgt_coord(Ulist[k]+Llist[Blist[k]]-1, coord(k, Blist[k])));
				//sort_wgts.push_back(wgt_coord((Ulist[k]+Llist[Blist[k]]-1)*(U[k]+V[Blist[k]]), coord(k, Blist[k])));
				//sort_wgts.push_back(wgt_coord((Ulist[k]+Llist[Blist[k]]-1)*(U[k]+V[Blist[k]]-temp[k])/temp[k], coord(k, Blist[k])));
				//sort_wgts.push_back(wgt_coord(1.0/temp[k], coord(k, Blist[k])));
				//sort_wgts.push_back(wgt_coord((U[k]+V[Blist[k]])*(U[k]+V[Blist[k]])/(temp[k]), coord(k, Blist[k])));
			std::sort(sort_wgts.begin(), sort_wgts.end());

			i = wbeg;
			for (wgt_coords::iterator it = sort_wgts.begin(); it != sort_wgts.end(); ++it)
			{
				localP[it->second.first] = i;
				localQ[it->second.second] = i;
				++i;
			}
			cend = i;
#elif defined(REORDER_RCM)
			{
1188
1189
1190
#if defined(REPORT_ILU)
				printf("Reordering with RCM\n");
#endif
Kirill Terekhov's avatar
Kirill Terekhov committed
1191
1192
1193
1194
1195
1196
1197
1198
1199
				//create a symmetric graph of the matrix A + A^T
				std::vector<INMOST_DATA_ENUM_TYPE> xadj(wend-wbeg+1), adjncy;
				//adjncy.reserve(nzA*2);

				for (k = cbeg; k < cend; ++k) if( localQ[k] == ENUMUNDEF ) printf("%s:%d No column permutation for row %d. Matrix is structurally singular\n",__FILE__,__LINE__,k);

				for (k = cbeg; k < cend; ++k) localP[k] = k;
				for (k = cbeg; k < cend; ++k) V[localQ[k]] = DR[k]; //if you expirience a bug here then the matrix is structurally singular
				for (k = cbeg; k < cend; ++k) DR[k] = V[k];
1200
				
1201
1202
				for (k = cbeg; k < cend; ++k) V[localQ[k]] = DR0[k]; //if you expirience a bug here then the matrix is structurally singular
				for (k = cbeg; k < cend; ++k) DR0[k] = V[k];
Kirill Terekhov's avatar
Kirill Terekhov committed
1203
1204
1205

				for(k = wbeg; k < wend; ++k)
				{
1206
					for (INMOST_DATA_ENUM_TYPE jt = A_Address[k].first; jt < A_Address[k].last; ++jt)
Kirill Terekhov's avatar
Kirill Terekhov committed
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
					{
						A_Entries[jt].first = localQ[A_Entries[jt].first];
						//A_Entries[jt].second *= DL[k]*DR[A_Entries[jt].first];
					}
					std::sort(A_Entries.begin()+A_Address[k].first,A_Entries.begin()+A_Address[k].last);
				}
				
				//DumpMatrix(A_Address,A_Entries,cbeg,cend,"A.mtx");
				
				ReorderEF(wbeg, wend, donePQ, localP, localQ);
				inversePQ(wbeg,wend,localP,localQ, invP,invQ);
				applyPQ(wbeg, wend, localP, localQ, invP, invQ);

				trcmgraph = Timer();
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
				
				for(k = wbeg; k < wend; ++k)
				{
					Ulist[k] = EOL;
					Bbeg[k] = EOL;
					Blist[k] = EOL;
				}
				//std::fill(Ulist.begin()+wbeg-mobeg, Ulist.begin()+wend-mobeg,EOL);
				//std::fill(Bbeg.begin()+wbeg-mobeg, Bbeg.begin()+wend-mobeg,EOL);
				//std::fill(Blist.begin()+wbeg-mobeg, Blist.begin()+wend-mobeg,EOL);
Kirill Terekhov's avatar
Kirill Terekhov committed
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
				for (k = wend; k > wbeg; --k)
				{
					//vwgt[k-1] = 1;
					if (A_Address[k-1].Size() > 0)
					{
						i = A_Entries[A_Address[k-1].first].first;
						Blist[k-1] = Bbeg[i];
						Bbeg[i] = k-1;
					}
					Ulist[k-1] = A_Address[k-1].first;
				}
				xadj[0] = 0;
				for(i = wbeg; i < wend; ++i)
				{
1245
					for (INMOST_DATA_ENUM_TYPE jt = A_Address[i].first; jt < A_Address[i].last; ++jt)
Kirill Terekhov's avatar
Kirill Terekhov committed
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
					{
						if( A_Entries[jt].first != i )
							adjncy.push_back(static_cast<INMOST_DATA_ENUM_TYPE>(A_Entries[jt].first));
					}

					Li = Bbeg[i];
					while (Li != EOL)
					{
						if( Li != i )
							adjncy.push_back(static_cast<INMOST_DATA_ENUM_TYPE>(Li));
						Li = Blist[Li];
					}

					Li = Bbeg[i];
					while (Li != EOL)
					{
						Ui = Blist[Li];
						Ulist[Li]++;
						if (A_Address[Li].Size() && A_Address[Li].last - Ulist[Li] > 0)
						{
							k = A_Entries[Ulist[Li]].first;
							Blist[Li] = Bbeg[k];
							Bbeg[k] = Li;
						}

						Li = Ui;
					}

					std::sort(adjncy.begin()+xadj[i-wbeg],adjncy.end());
					adjncy.resize(std::unique(adjncy.begin()+xadj[i-wbeg],adjncy.end())-adjncy.begin());

					xadj[i-wbeg+1] = static_cast<INMOST_DATA_ENUM_TYPE>(adjncy.size());
				}

1280
1281
				//std::fill(Bbeg.begin()+wbeg - mobeg, Bbeg.begin()+wend - mobeg,EOL);
				for(k = wbeg; k < wend; ++k) Bbeg[k] = EOL;
Kirill Terekhov's avatar
Kirill Terekhov committed
1282
1283
1284
1285

				trcmgraph = Timer()-trcmgraph;

				trcmorder = Timer();
1286
1287
				//std::fill(Ulist.begin() + wbeg - mobeg, Ulist.begin() + wend - mobeg, ENUMUNDEF);
				for(k = wbeg; k < wend; ++k) Ulist[k] = ENUMUNDEF;
Kirill Terekhov's avatar
Kirill Terekhov committed
1288
				//find node with the lowest order
1289
                INMOST_DATA_ENUM_TYPE index = wbeg;
Kirill Terekhov's avatar
Kirill Terekhov committed
1290
1291
1292
				INMOST_DATA_ENUM_TYPE cur = ENUMUNDEF;
				std::deque<INMOST_DATA_ENUM_TYPE> q;
				std::vector<INMOST_DATA_ENUM_TYPE> conns;
1293
                do
Kirill Terekhov's avatar
Kirill Terekhov committed
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
				{
					cur = ENUMUNDEF;
					for(k = wbeg; k < wend && cur == ENUMUNDEF; ++k)
					{
						if( Ulist[k] == ENUMUNDEF )
							cur = k;
					}
					assert(cur != ENUMUNDEF);
					for(k = cur+1; k < wend; ++k) if( Ulist[k] == ENUMUNDEF )
					{
						if( RCM_Comparator(wbeg,xadj)(k,cur) )
							cur = k;
					}
					q.push_back(cur);
					Ulist[cur] = index++;
					while(!q.empty())
					{
						cur = q.front();
						q.pop_front();
						for (INMOST_DATA_ENUM_TYPE it = xadj[cur-wbeg]; it < xadj[cur-wbeg+1]; ++it)
							if( Ulist[adjncy[it]] == ENUMUNDEF )
								conns.push_back(adjncy[it]);
						std::sort(conns.begin(),conns.end(),RCM_Comparator(wbeg,xadj));
						for (k = 0; k < static_cast<INMOST_DATA_ENUM_TYPE>(conns.size()); ++k)
						{
							Ulist[conns[k]] = index++;
							q.push_back(conns[k]);
						}
						conns.clear();
					}
				}
				while( index < wend );

				for(k = wbeg; k < wend; ++k)
					Ulist[k] = wend-(Ulist[k]-wbeg)-1;

				for(k = wbeg; k < wend; ++k)
				{
					localP[k] = Ulist[k];//wend - Ulist[k] + 1;
					localQ[k] = Ulist[k];//wend - Ulist[k] + 1;
					//localP[Ulist[k]] = k;
					//localQ[Ulist[k]] = k;
				}
				cend = wend;
				i = wend;


				trcmorder = Timer() - trcmorder;
			}
#else
			cend = wend;
			i = wbeg;
#endif
			tt = Timer() - tt;
			treorder += tt;
			if (cbeg == cend && cbeg != wend)
			{
				//std::cout << __FILE__ << ":" << __LINE__ << " singular matrix, factored " << mobeg << ".." << cend << " out of " << mobeg << ".." << moend << std::endl;
				for (k = cbeg; k < wend; ++k)
					LU_Diag[k] = tol_modif;
				break;
			}
			tt = Timer();
			//finish reordering
			if (i < wend)
			{
				j = i;
				for (k = wbeg; k < wend; ++k)
				{
					if (localP[k] == ENUMUNDEF) localP[k] = i++;
					if (localQ[k] == ENUMUNDEF) localQ[k] = j++;
				}
			}

			for (k = cbeg; k < cend; ++k) 
			{
				U[localP[k]] = DL[k];
				V[localQ[k]] = DR[k];
			}
			for (k = cbeg; k < cend; ++k) 
			{
				DL[k] = U[k];
				DR[k] = V[k];
			}
1378
			
1379
1380
			for (k = cbeg; k < cend; ++k) 
			{
1381
1382
1383
				if( !(localQ[k] >= cbeg && localQ[k] < cend) ||
				    !(localP[k] >= cbeg && localP[k] < cend))
					std::cout << "Bad permutations P: " << localP[k] << " Q: " << localQ[k] << " interval [" << cbeg << ":" << cend << "]" << std::endl;
1384
1385
1386
1387
1388
1389
1390
1391
				U[localP[k]] = DL0[k];
				V[localQ[k]] = DR0[k];
			}
			for (k = cbeg; k < cend; ++k) 
			{
				DL0[k] = U[k];
				DR0[k] = V[k];
			}
1392
			
Kirill Terekhov's avatar
Kirill Terekhov committed
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
			tlreorder = Timer() - tt;
			treorder += tlreorder;
///////////////////////////////////////////////////////////////////////////////////
///                  REASSAMBLE                                                 ///
///////////////////////////////////////////////////////////////////////////////////
			tt = Timer();
			double tt1, tt2,ttt;
			//in localPQ numbers indicate where to put current row/column
			//reorder E,F blocks by swaps
			tt1 = Timer();
			//inverse ordering
			ReorderEF(wbeg,wend, donePQ, localP, localQ);
			inversePQ(wbeg,wend,localP,localQ, invP,invQ);
			tt1 = Timer() - tt1;

			//std::cout << "reorder: " << tt1 << std::endl;
			tt2 = Timer();
			nzA = 0;
			B_Entries.clear();
			for (k = cbeg; k < cend; ++k)
			{
				B_Address[k].first = static_cast<INMOST_DATA_ENUM_TYPE>(B_Entries.size());
1415
				for (INMOST_DATA_ENUM_TYPE jt = A_Address[invP[k]].first; jt < A_Address[invP[k]].last; ++jt)
Kirill Terekhov's avatar
Kirill Terekhov committed
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
				{
					i = localQ[A_Entries[jt].first];
					u = A_Entries[jt].second;
					B_Entries.push_back(Sparse::Row::make_entry(i, u));
				}
				B_Address[k].last = static_cast<INMOST_DATA_ENUM_TYPE>(B_Entries.size());
				std::sort(B_Entries.begin() + B_Address[k].first, B_Entries.end());
				nzA += A_Address[k].Size();
			}
			B_Entries.push_back(Sparse::Row::make_entry(-1,0.0));
			
			//std::stringstream s;
			//s << "B" << level_size.size() << ".mtx";
			//DumpMatrix(B_Address,B_Entries,cbeg,cend,s.str());
			/*
			{
				int rank = 0;
#if defined(USE_MPI)
				MPI_Comm_rank(MPI_COMM_WORLD,&rank);
#endif
				std::stringstream str;
				str << "mat_reorder" << rank << ".mtx";
				DumpMatrix(B_Address,B_Entries,cbeg,cend,str.str());
			}
#if defined(USE_MPI)
			MPI_Barrier(MPI_COMM_WORLD);
#endif
			exit(0);
			*/
			//Setup column addressing for B,F, in descending order to keep list ordered
			for (k = cend; k > cbeg; --k)
			{
				if (B_Address[k - 1].Size() > 0)
				{
					i = B_Entries[B_Address[k - 1].first].first;
					if (i < k-1)
					{
						Blist[k-1] = Bbeg[i];
						Bbeg[i] = k-1;
					}
				}
			}

			tt2 = Timer() - tt2;

			ttt = Timer();
			applyPQ(wbeg, wend, localP, localQ, invP, invQ);
			tt1 += Timer() - ttt;
			
			//reset localPQ
			
			//for(int k = mobeg; k < moend; ++k) localP[k] = localQ[k] = k;
			/*
			{
				std::stringstream s;
				s << "C" << level_size.size() << ".mtx";
				DumpMatrix(B_Address,B_Entries,cbeg,cend,s.str());
			}
			*/
			tlreassamble = Timer() - tt;
			treassamble += tlreassamble;
///////////////////////////////////////////////////////////////////////////////////
///                  RESCALING                                                  ///
///////////////////////////////////////////////////////////////////////////////////
			//Rescale current B block by Row-Column alternating scaling
			tt = Timer();
			if( rescale_b )
			{
#if defined(REPORT_ILU)
				std::cout << " rescaling block B " << std::endl;
#endif
				
#if defined(EQUALIZE_1NORM) || defined(EQUALIZE_2NORM)
				for (k = cbeg; k < cend; k++) DL[k] = DR[k] = 1.0;
#elif defined(EQUALIZE_IDOMINANCE)
				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						B_Entries[r].second *= (DL[k] * DR[B_Entries[r].first]);
					
					U[k] = DL[k];
					V[k] = DR[k];
				}
#else
				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						B_Entries[r].second *= (DL[k] * DR[B_Entries[r].first]);
				}
#endif
				/*
				{
					std::stringstream s;
					s << "CC" << level_size.size() << ".mtx";
					DumpMatrix(B_Address,B_Entries,cbeg,cend,s.str());
				}
				*/
				//DumpMatrix(B_Address,B_Entries,cbeg,cend,"MC64_MPTILUC2.mtx");
				/*
				{
					std::stringstream name;
					name << "mc64_" << info->GetRank() << ".mtx";
					DumpMatrix(B_Address,B_Entries,cbeg,cend,name.str());
				}
				*/
				
///////////////////////////////////////////////////////////////////////////////////
///                COMPUTE GIRSCHGORIN RADIUS                                   ///
///////////////////////////////////////////////////////////////////////////////////
				/*
				INMOST_DATA_REAL_TYPE radii = 0;
				for (k = cbeg; k < cend; k++)
				{
					INMOST_DATA_REAL_TYPE local_radii = 0;
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						if( B_Entries[r].first != k ) local_radii += fabs(B_Entries[r].second);
					if( radii < local_radii ) radii = local_radii;
				}

				printf("Gershgorin's radius after mc64: %e\n",radii);
				*/


				//DumpMatrix(B_Address,B_Entries,cbeg,cend,"mat_mc64.mtx");
#if defined(EQUALIZE_1NORM)
///////////////////////////////////////////////////////////////////////////////////
///         ROW-COLUMN ALTERNATING SCALING FOR 1-NORM BALANCING                 ///
///////////////////////////////////////////////////////////////////////////////////
1544
1545
				//std::fill(DL.Begin() + cbeg - mobeg, DL.Begin() + cend - mobeg, 0.0);
				for(k = cbeg; k < cend; ++k) DL[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1546
1547
1548
1549
1550
1551
1552
1553
				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						DL[k] += fabs(B_Entries[r].second);//*B_Entries[r].second;
				}
				for (k = cbeg; k < cend; k++) if (DL[k] < eps) DL[k] = 1.0 / subst; else DL[k] = 1.0 / DL[k];
				for (INMOST_DATA_ENUM_TYPE iter = 0; iter < sciters; iter++)
				{
1554
1555
					//std::fill(DR.Begin()+cbeg-mobeg, DR.Begin()+cend-mobeg, 0.0);
					for(k = cbeg; k < cend; ++k) DR[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1556
1557
1558
1559
1560
1561
					for (k = cbeg; k < cend; k++)
					{
						for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
							DR[B_Entries[r].first] += DL[k] * fabs(B_Entries[r].second);//*B_Entries[r].second;
					}
					for (k = cbeg; k < cend; k++) if (DR[k] < eps) DR[k] = 1.0 / subst; else DR[k] = 1.0 / DR[k];
1562
1563
					//std::fill(DL.Begin()+cbeg-mobeg, DL.Begin()+cend-mobeg, 0.0);
					for(k = cbeg; k < cend; ++k) DL[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
					for (k = cbeg; k < cend; k++)
					{
						for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
							DL[k] += DR[B_Entries[r].first] * fabs(B_Entries[r].second);//*B_Entries[r].second;
					}
					for (k = cbeg; k < cend; k++) if (DL[k] < eps) DL[k] = 1.0 / subst; else DL[k] = 1.0 / DL[k];
				}
				//for (k = cbeg; k < cend; k++) DL[k] = sqrt(DL[k]);
				//for (k = cbeg; k < cend; k++) DR[k] = sqrt(DR[k]);
				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						B_Entries[r].second *= DL[k] * DR[B_Entries[r].first];
				}
#elif defined(EQUALIZE_2NORM)
1579
1580
				//std::fill(DL.Begin() + cbeg - mobeg, DL.Begin() + cend - mobeg, 0.0);
				for(k = cbeg; k < cend; ++k) DL[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1581
1582
1583
1584
1585
1586
1587
1588
				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						DL[k] += B_Entries[r].second*B_Entries[r].second;
				}
				for (k = cbeg; k < cend; k++) if (DL[k] < eps) DL[k] = 1.0 / subst; else DL[k] = 1.0 / DL[k];
				for (INMOST_DATA_ENUM_TYPE iter = 0; iter < sciters; iter++)
				{
1589
1590
					//std::fill(DR.Begin()+cbeg-mobeg, DR.Begin()+cend-mobeg, 0.0);
					for(k = cbeg; k < cend; ++k) DR[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1591
1592
1593
1594
1595
1596
					for (k = cbeg; k < cend; k++)
					{
						for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
							DR[B_Entries[r].first] += DL[k] * B_Entries[r].second*B_Entries[r].second;
					}
					for (k = cbeg; k < cend; k++) if (DR[k] < eps) DR[k] = 1.0 / subst; else DR[k] = 1.0 / DR[k];
1597
1598
					//std::fill(DL.Begin()+cbeg-mobeg, DL.Begin()+cend-mobeg, 0.0);
					for(k = cbeg; k < cend; ++k) DL[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
					for (k = cbeg; k < cend; k++)
					{
						for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
							DL[k] += DR[B_Entries[r].first] * B_Entries[r].second*B_Entries[r].second;
					}
					for (k = cbeg; k < cend; k++) if (DL[k] < eps) DL[k] = 1.0 / subst; else DL[k] = 1.0 / DL[k];
				}
				//for (k = cbeg; k < cend; k++) DL[k] = sqrt(DL[k]);
				//for (k = cbeg; k < cend; k++) DR[k] = sqrt(DR[k]);
				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						B_Entries[r].second *= DL[k] * DR[B_Entries[r].first];
				}
#elif defined(EQUALIZE_IDOMINANCE)
///////////////////////////////////////////////////////////////////////////////////
///        THIS VERSION OF RESCALING INCREASES DIAGONAL DOMINANCE               ///
///////////////////////////////////////////////////////////////////////////////////
				if( C_Entries.size() < B_Entries.size() )
					C_Entries.resize(B_Entries.size());
				
				for(k = cbeg; k < cend; ++k)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
					{
						u = fabs(B_Entries[r].second);
						if( u > 1.0e-12 )
							C_Entries[r] = -log(u);
						else
							C_Entries[r] = std::numeric_limits<INMOST_DATA_REAL_TYPE>::max();
					}
				}

				
1633
1634
				//std::fill(temp.begin() + cbeg - mobeg, temp.begin() + cend - mobeg, 0.0);
				for(k = cbeg; k < cend; ++k) temp[k] = 0.0;
Kirill Terekhov's avatar
Kirill Terekhov committed
1635
1636
1637

				for (INMOST_DATA_ENUM_TYPE iter = 0; iter < sciters; iter++)
				{
1638
1639
1640
					//std::fill(DL.Begin() + cbeg - mobeg, DL.Begin() + cend - mobeg, std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
					//std::fill(DR.Begin() + cbeg - mobeg, DR.Begin() + cend - mobeg, std::numeric_limits<INMOST_DATA_REAL_TYPE>::max());
					for(k = cbeg; k < cend; ++k) DL[k] = DR[k] = std::numeric_limits<INMOST_DATA_REAL_TYPE>::max();
Kirill Terekhov's avatar
Kirill Terekhov committed
1641
1642
1643
1644
1645
1646
1647
1648
					for (k = cbeg; k < cend; k++) //row number
					{
						for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						{
							i = B_Entries[r].first; //column number
							if( i != k ) //out of diagonal
							{
								u = C_Entries[r] + temp[k] - temp[i];
Kirill Terekhov's avatar
Kirill Terekhov committed
1649
								//if( isnan(u) || u != u ) std::cout << __FILE__ << ":" << __LINE__ << " u is " << u << std::endl;
Kirill Terekhov's avatar
Kirill Terekhov committed
1650
1651
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
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
								DL[k] = std::min(DL[k],u);// update Y1
								DR[i] = std::min(DR[i],u);// update Y2
							}
						}
					}

					for (k = cbeg; k < cend; k++)
					{
						if( DR[k] != std::numeric_limits<INMOST_DATA_REAL_TYPE>::max() &&
							DL[k] != std::numeric_limits<INMOST_DATA_REAL_TYPE>::max() )
							temp[k] += (DR[k]-DL[k])*0.5;
					}

				}

				for (k = cbeg; k < cend; k++)
				{
					DL[k] = exp(-temp[k]);
					DR[k] = exp(temp[k]);
					//if( isnan(DL[k]) || DL[k] != DL[k] || fabs(DL[k]) < 1.0e-12 ) std::cout << __FILE__ << ":" << __LINE__ << " DL[" << k << "] is " << DL[k] << std::endl;
					//if( isnan(DR[k]) || DR[k] != DR[k] || fabs(DR[k]) < 1.0e-12 ) std::cout << __FILE__ << ":" << __LINE__ << " DR[" << k << "] is " << DR[k] << std::endl;
				}

				for (k = cbeg; k < cend; k++)
				{
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						B_Entries[r].second *= (DL[k] * DR[B_Entries[r].first]);
				}
#endif
///////////////////////////////////////////////////////////////////////////////////
///              COMPUTE GIRSCHGORIN RADIUS                                     ///
///////////////////////////////////////////////////////////////////////////////////
				/*
				radii = 0;
				for (k = cbeg; k < cend; k++)
				{
					INMOST_DATA_REAL_TYPE local_radii = 0;
					for (INMOST_DATA_ENUM_TYPE r = B_Address[k].first; r < B_Address[k].last; ++r)
						if( B_Entries[r].first != k ) local_radii += fabs(B_Entries[r].second);
					if( radii < local_radii ) radii = local_radii;
				}

				printf("Gershgorin's radius after equilibration: %e\n",radii);
				*/
#if defined(EQUALIZE_IDOMINANCE)
				for (k = cbeg; k < cend; k++)
				{
					DL[k] *= U[k];
					DR[k] *= V[k];
				}
#endif
				//DumpMatrix(B_Address,B_Entries,cbeg,cend,"mat_equilibration.mtx");
///////////////////////////////////////////////////////////////////////////////////
///               RESCALING DONE                                                ///
///////////////////////////////////////////////////////////////////////////////////
				//End rescale B block
				trescale += Timer() - tt;
1707
				//stack scaling
1708
				
1709
1710
1711
1712
1713
				for (k = cbeg; k < cend; k++)
				{
					DL0[k] *= DL[k];
					DR0[k] *= DR[k];
				}
1714
1715
1716
1717
1718
1719
1720
				/*
				DL.Save("DL_"+to_string(level_size.size())+".mtx");
				DR.Save("DR_"+to_string(level_size.size())+".mtx");
				DL0.Save("DL0_"+to_string(level_size.size())+".mtx");
				DR0.Save("DR0_"+to_string(level_size.size())+".mtx");
				DumpMatrix(B_Address,B_Entries,cbeg,cend,"mat_equil"+to_string(level_size.size())+".mtx");
				 */