1 | /* $Id: CoinOslFactorization.cpp 1448 2011-06-19 15:34:41Z stefan $ */ |
2 | // Copyright (C) 1987, 2009, International Business Machines |
3 | // Corporation and others. All Rights Reserved. |
4 | // This code is licensed under the terms of the Eclipse Public License (EPL). |
5 | |
6 | #include "CoinUtilsConfig.h" |
7 | |
8 | #include <cassert> |
9 | #include "CoinPragma.hpp" |
10 | #include "CoinOslFactorization.hpp" |
11 | #include "CoinOslC.h" |
12 | #include "CoinIndexedVector.hpp" |
13 | #include "CoinHelperFunctions.hpp" |
14 | #include "CoinPackedMatrix.hpp" |
15 | #include "CoinTypes.hpp" |
16 | #include "CoinFinite.hpp" |
17 | #include <stdio.h> |
18 | static void c_ekksmem(EKKfactinfo *fact,int numberRows,int maximumPivots); |
19 | static void c_ekksmem_copy(EKKfactinfo *fact,const EKKfactinfo * rhsFact); |
20 | static void c_ekksmem_delete(EKKfactinfo *fact); |
21 | //:class CoinOslFactorization. Deals with Factorization and Updates |
22 | // CoinOslFactorization. Constructor |
23 | CoinOslFactorization::CoinOslFactorization ( ) |
24 | : CoinOtherFactorization() |
25 | { |
26 | gutsOfInitialize(); |
27 | } |
28 | |
29 | /// Copy constructor |
30 | CoinOslFactorization::CoinOslFactorization ( const CoinOslFactorization &other) |
31 | : CoinOtherFactorization(other) |
32 | { |
33 | gutsOfInitialize(); |
34 | gutsOfCopy(other); |
35 | } |
36 | // Clone |
37 | CoinOtherFactorization * |
38 | CoinOslFactorization::clone() const |
39 | { |
40 | return new CoinOslFactorization(*this); |
41 | } |
42 | /// The real work of constructors etc |
43 | void CoinOslFactorization::gutsOfDestructor(bool clearFact) |
44 | { |
45 | delete [] elements_; |
46 | delete [] pivotRow_; |
47 | delete [] workArea_; |
48 | elements_ = NULL; |
49 | pivotRow_ = NULL; |
50 | workArea_ = NULL; |
51 | numberRows_ = 0; |
52 | numberColumns_ = 0; |
53 | numberGoodU_ = 0; |
54 | status_ = -1; |
55 | maximumRows_=0; |
56 | maximumSpace_=0; |
57 | solveMode_=0; |
58 | if (clearFact) |
59 | c_ekksmem_delete(&factInfo_); |
60 | } |
61 | void CoinOslFactorization::gutsOfInitialize(bool zapFact) |
62 | { |
63 | pivotTolerance_ = 1.0e-1; |
64 | zeroTolerance_ = 1.0e-13; |
65 | #ifndef COIN_FAST_CODE |
66 | slackValue_ = -1.0; |
67 | #endif |
68 | maximumPivots_=200; |
69 | relaxCheck_=1.0; |
70 | numberRows_ = 0; |
71 | numberColumns_ = 0; |
72 | numberGoodU_ = 0; |
73 | status_ = -1; |
74 | numberPivots_ = 0; |
75 | maximumRows_=0; |
76 | maximumSpace_=0; |
77 | elements_ = NULL; |
78 | pivotRow_ = NULL; |
79 | workArea_ = NULL; |
80 | solveMode_=0; |
81 | if (zapFact) { |
82 | memset(&factInfo_,0,sizeof(factInfo_)); |
83 | factInfo_.maxinv=100; |
84 | factInfo_.drtpiv=1.0e-10; |
85 | factInfo_.zeroTolerance=1.0e-12; |
86 | factInfo_.zpivlu=0.1; |
87 | factInfo_.areaFactor=1.0; |
88 | factInfo_.nbfinv=100; |
89 | } |
90 | } |
91 | // ~CoinOslFactorization. Destructor |
92 | CoinOslFactorization::~CoinOslFactorization ( ) |
93 | { |
94 | gutsOfDestructor(); |
95 | } |
96 | // = |
97 | CoinOslFactorization & CoinOslFactorization::operator = ( const CoinOslFactorization & other ) { |
98 | if (this != &other) { |
99 | bool noGood = factInfo_.nrowmx!=other.factInfo_.nrowmx&& |
100 | factInfo_.eta_size!=other.factInfo_.eta_size; |
101 | gutsOfDestructor(noGood); |
102 | gutsOfInitialize(noGood); |
103 | gutsOfCopy(other); |
104 | } |
105 | return *this; |
106 | } |
107 | #define WORK_MULT 2 |
108 | void CoinOslFactorization::gutsOfCopy(const CoinOslFactorization &other) |
109 | { |
110 | pivotTolerance_ = other.pivotTolerance_; |
111 | zeroTolerance_ = other.zeroTolerance_; |
112 | #ifndef COIN_FAST_CODE |
113 | slackValue_ = other.slackValue_; |
114 | #endif |
115 | relaxCheck_ = other.relaxCheck_; |
116 | numberRows_ = other.numberRows_; |
117 | numberColumns_ = other.numberColumns_; |
118 | maximumRows_ = other.maximumRows_; |
119 | maximumSpace_ = other.maximumSpace_; |
120 | solveMode_ = other.solveMode_; |
121 | numberGoodU_ = other.numberGoodU_; |
122 | maximumPivots_ = other.maximumPivots_; |
123 | numberPivots_ = other.numberPivots_; |
124 | factorElements_ = other.factorElements_; |
125 | status_ = other.status_; |
126 | elements_ = NULL; |
127 | pivotRow_ = NULL; |
128 | workArea_ = NULL; |
129 | c_ekksmem_copy(&factInfo_,&other.factInfo_); |
130 | } |
131 | |
132 | // getAreas. Gets space for a factorization |
133 | //called by constructors |
134 | void |
135 | CoinOslFactorization::getAreas ( int numberOfRows, |
136 | int numberOfColumns, |
137 | CoinBigIndex maximumL, |
138 | CoinBigIndex maximumU ) |
139 | { |
140 | |
141 | numberRows_ = numberOfRows; |
142 | numberColumns_ = numberOfColumns; |
143 | CoinBigIndex size = static_cast<CoinBigIndex>(factInfo_.areaFactor* |
144 | (maximumL+maximumU)); |
145 | factInfo_.zeroTolerance=zeroTolerance_; |
146 | // If wildly out redo |
147 | if (maximumRows_>numberRows_+1000) { |
148 | maximumRows_=0; |
149 | maximumSpace_=0; |
150 | factInfo_.last_eta_size=0; |
151 | } |
152 | if (size>maximumSpace_) { |
153 | //delete [] elements_; |
154 | //elements_ = new CoinFactorizationDouble [size]; |
155 | maximumSpace_ = size; |
156 | } |
157 | factInfo_.lastEtaCount = factInfo_.nnentu+factInfo_.nnentl; |
158 | int oldnnetas=factInfo_.last_eta_size; |
159 | // If we are going to increase then be on safe side |
160 | if (size>oldnnetas) |
161 | size = static_cast<int>(1.1*size); |
162 | factInfo_.eta_size=CoinMax(size,oldnnetas); |
163 | //printf("clp size %d, old %d now %d - iteration %d - last count %d - rows %d,%d,%d\n", |
164 | // size,oldnnetas,factInfo_.eta_size,factInfo_.iterno,factInfo_.lastEtaCount, |
165 | //numberRows_,factInfo_.nrowmx,factInfo_.nrow); |
166 | //if (!factInfo_.iterno) { |
167 | //printf("here\n"); |
168 | //} |
169 | /** Get solve mode e.g. 0 C++ code, 1 Lapack, 2 choose |
170 | If 4 set then values pass |
171 | if 8 set then has iterated |
172 | */ |
173 | solveMode_ &= 4+8; // clear bottom bits |
174 | factInfo_.ifvsol= ((solveMode_&4)!=0) ? 1 : 0; |
175 | if ((solveMode_&8)!=0) { |
176 | factInfo_.ifvsol=0; |
177 | factInfo_.invok=1; |
178 | } else { |
179 | factInfo_.iter0=factInfo_.iterno; |
180 | factInfo_.invok=-1; |
181 | factInfo_.if_sparse_update=0; |
182 | } |
183 | #if 0 |
184 | if (!factInfo_.if_sparse_update && |
185 | factInfo_.iterno>factInfo_.iter0 && |
186 | numberRows_>=C_EKK_GO_SPARSE) { |
187 | printf("count %d rows %d etasize %d\n" , |
188 | factInfo_.lastEtaCount,factInfo_.nrow,factInfo_.eta_size); |
189 | |
190 | } |
191 | #endif |
192 | if (!factInfo_.if_sparse_update && |
193 | factInfo_.iterno>factInfo_.iter0 && |
194 | numberRows_>=C_EKK_GO_SPARSE && |
195 | (factInfo_.lastEtaCount>>2)<factInfo_.nrow&& |
196 | !factInfo_.switch_off_sparse_update) { |
197 | #if PRINT_DEBUG |
198 | printf("**** Switching on sparse update - etacount\n" ); |
199 | #endif |
200 | /* I suspect this can go into c_ekksslvf; |
201 | * if c_ekkshff decides to switch sparse_update off, |
202 | * then it problably always switches it off (?) |
203 | */ |
204 | factInfo_.if_sparse_update=2; |
205 | } |
206 | c_ekksmem(&factInfo_,numberRows_,maximumPivots_); |
207 | if (numberRows_>maximumRows_) { |
208 | maximumRows_ = numberRows_; |
209 | //delete [] pivotRow_; |
210 | //delete [] workArea_; |
211 | //pivotRow_ = new int [2*maximumRows_+maximumPivots_]; |
212 | //workArea_ = new CoinFactorizationDouble [maximumRows_*WORK_MULT]; |
213 | } |
214 | } |
215 | |
216 | // preProcess. |
217 | void |
218 | CoinOslFactorization::preProcess () |
219 | { |
220 | factInfo_.zpivlu=pivotTolerance_; |
221 | // Go to Fortran |
222 | int * hcoli=factInfo_.xecadr+1; |
223 | int * indexRowU = factInfo_.xeradr+1; |
224 | CoinBigIndex * startColumnU=factInfo_.xcsadr+1; |
225 | for (int i=0;i<numberRows_;i++) { |
226 | int start = startColumnU[i]; |
227 | startColumnU[i]++; // to Fortran |
228 | for (int j=start;j<startColumnU[i+1];j++) { |
229 | indexRowU[j]++; // to Fortran |
230 | hcoli[j]=i+1; // to Fortran |
231 | } |
232 | } |
233 | startColumnU[numberRows_]++; // to Fortran |
234 | |
235 | /* can do in column order - no zeros or duplicates */ |
236 | #ifndef NDEBUG |
237 | int ninbas = |
238 | #endif |
239 | c_ekkslcf(&factInfo_); |
240 | assert (ninbas>0); |
241 | } |
242 | |
243 | //Does factorization |
244 | int |
245 | CoinOslFactorization::factor ( ) |
246 | { |
247 | /* Uwe's factorization (sort of) */ |
248 | int irtcod = c_ekklfct(&factInfo_); |
249 | |
250 | /* Check return code */ |
251 | /* 0 - Fine , 1 - Backtrack, 2 - Singularities on initial, 3-Fatal */ |
252 | /* now 5-Need more memory */ |
253 | |
254 | status_= 0; |
255 | if (factInfo_.eta_size>factInfo_.last_eta_size) { |
256 | factInfo_.areaFactor *= factInfo_.eta_size; |
257 | factInfo_.areaFactor /= factInfo_.last_eta_size; |
258 | #ifdef CLP_INVESTIGATE |
259 | printf("areaFactor increased to %g\n" ,factInfo_.areaFactor); |
260 | #endif |
261 | } |
262 | if (irtcod==5) { |
263 | status_=-99; |
264 | assert (factInfo_.eta_size>factInfo_.last_eta_size) ; |
265 | #ifdef CLP_INVESTIGATE |
266 | printf("need more memory\n" ); |
267 | #endif |
268 | } else if (irtcod) { |
269 | status_=-1; |
270 | //printf("singular %d\n",irtcod); |
271 | } |
272 | return status_; |
273 | } |
274 | // Makes a non-singular basis by replacing variables |
275 | void |
276 | CoinOslFactorization::makeNonSingular(int * sequence, int numberColumns) |
277 | { |
278 | const EKKHlink *rlink = factInfo_.kp1adr; |
279 | const EKKHlink *clink = factInfo_.kp2adr; |
280 | int nextRow=0; |
281 | //int * mark = reinterpret_cast<int *>(factInfo_.kw1adr); |
282 | //int nr=0; |
283 | //int nc=0; |
284 | #if 0 |
285 | for (int i=0;i<numberRows_;i++) { |
286 | //mark[i]=-1; |
287 | if (rlink[i].pre>=0||rlink[i].pre==-(numberRows_+1)) { |
288 | nr++; |
289 | printf("%d rl %d cl %d\n" ,i,rlink[i].pre,clink[i].pre); |
290 | } |
291 | if (clink[i].pre>=0||clink[i].pre==-(numberRows_+1)) { |
292 | nc++; |
293 | printf("%d rl %d cl %d\n" ,i,rlink[i].pre,clink[i].pre); |
294 | } |
295 | } |
296 | #endif |
297 | //printf("nr %d nc %d\n",nr,nc); |
298 | bool goodPass=true; |
299 | int numberDone=0; |
300 | for (int i=0;i<numberRows_;i++) { |
301 | int cRow =(-clink[i].pre)-1; |
302 | if (cRow==numberRows_||cRow<0) { |
303 | // throw out |
304 | for (;nextRow<numberRows_;nextRow++) { |
305 | int rRow =(-rlink[nextRow].pre)-1; |
306 | if (rRow==numberRows_||rRow<0) |
307 | break; |
308 | } |
309 | if (nextRow<numberRows_) { |
310 | sequence[i]=nextRow+numberColumns; |
311 | nextRow++; |
312 | numberDone++; |
313 | } else { |
314 | goodPass=false; |
315 | assert(numberDone); |
316 | //printf("BAD singular at row %d\n",i); |
317 | break; |
318 | } |
319 | } |
320 | } |
321 | #ifndef NDEBUG |
322 | if (goodPass) { |
323 | for (;nextRow<numberRows_;nextRow++) { |
324 | int rRow =(-rlink[nextRow].pre)-1; |
325 | assert (!(rRow==numberRows_||rRow<0)); |
326 | } |
327 | } |
328 | #endif |
329 | } |
330 | // Does post processing on valid factorization - putting variables on correct rows |
331 | void |
332 | CoinOslFactorization::postProcess(const int * sequence, int * pivotVariable) |
333 | { |
334 | factInfo_.iterin=factInfo_.iterno; |
335 | factInfo_.npivots=0; |
336 | numberPivots_=0; |
337 | const int * permute3 = factInfo_.mpermu+1; |
338 | assert (permute3==reinterpret_cast<const int *> |
339 | (factInfo_.kadrpm+numberRows_+1)); |
340 | // this is ridiculous - must be better way |
341 | int * permute2 = reinterpret_cast<int *>(factInfo_.kw1adr); |
342 | const int * permute = reinterpret_cast<const int *>(factInfo_.kp2adr); |
343 | for (int i=0;i<numberRows_;i++) { |
344 | permute2[permute[i]-1]=i; |
345 | } |
346 | for (int i=0;i<numberRows_;i++) { |
347 | // the row is i |
348 | // the column is whatever matches k3[i] in k1 |
349 | int look=permute3[i]-1; |
350 | int j=permute2[look]; |
351 | int k = sequence[j]; |
352 | pivotVariable[i]=k; |
353 | } |
354 | #ifdef CLP_REUSE_ETAS |
355 | int * start = factInfo_.xcsadr+1; |
356 | int * putSeq = factInfo_.xrsadr+2*factInfo_.nrowmx+2; |
357 | int * position = putSeq+factInfo_.maxinv; |
358 | int * putStart = position+factInfo_.maxinv; |
359 | memcpy(putStart,start,numberRows_*sizeof(int)); |
360 | int iLast=start[numberRows_-1]; |
361 | putStart[numberRows_]=iLast+factInfo_.xeradr[iLast]+1; |
362 | factInfo_.save_nnentu=factInfo_.nnentu; |
363 | #endif |
364 | #ifndef NDEBUG |
365 | { |
366 | int lstart=numberRows_+factInfo_.maxinv+5; |
367 | int ndo = factInfo_.xnetal-lstart; |
368 | double * dluval=factInfo_.xeeadr; |
369 | int * mcstrt = factInfo_.xcsadr+lstart; |
370 | if (ndo) |
371 | assert (dluval[mcstrt[ndo]+1]<1.0e50); |
372 | } |
373 | #endif |
374 | } |
375 | /* Replaces one Column to basis, |
376 | returns 0=OK, 1=Probably OK, 2=singular, 3=no room |
377 | If checkBeforeModifying is true will do all accuracy checks |
378 | before modifying factorization. Whether to set this depends on |
379 | speed considerations. You could just do this on first iteration |
380 | after factorization and thereafter re-factorize |
381 | partial update already in U */ |
382 | int |
383 | CoinOslFactorization::replaceColumn ( CoinIndexedVector * regionSparse, |
384 | int pivotRow, |
385 | double pivotCheck , |
386 | bool /*checkBeforeModifying*/, |
387 | double acceptablePivot) |
388 | { |
389 | if (numberPivots_+1==maximumPivots_) |
390 | return 3; |
391 | int *regionIndex = regionSparse->getIndices ( ); |
392 | double *region = regionSparse->denseVector ( ); |
393 | int orig_nincol=0; |
394 | double saveTolerance = factInfo_.drtpiv; |
395 | factInfo_.drtpiv=acceptablePivot; |
396 | int returnCode=c_ekketsj(&factInfo_,region-1, |
397 | regionIndex, |
398 | pivotCheck,orig_nincol, |
399 | numberPivots_,&factInfo_.nuspike, |
400 | pivotRow+1, |
401 | reinterpret_cast<int *>(factInfo_.kw1adr)); |
402 | factInfo_.drtpiv=saveTolerance; |
403 | if (returnCode!=2) |
404 | numberPivots_++; |
405 | #ifndef NDEBUG |
406 | { |
407 | int lstart=numberRows_+factInfo_.maxinv+5; |
408 | int ndo = factInfo_.xnetal-lstart; |
409 | double * dluval=factInfo_.xeeadr; |
410 | int * mcstrt = factInfo_.xcsadr+lstart; |
411 | if (ndo) |
412 | assert (dluval[mcstrt[ndo]+1]<1.0e50); |
413 | } |
414 | #endif |
415 | return returnCode; |
416 | } |
417 | /* This version has same effect as above with FTUpdate==false |
418 | so number returned is always >=0 */ |
419 | int |
420 | CoinOslFactorization::updateColumn ( CoinIndexedVector * regionSparse, |
421 | CoinIndexedVector * regionSparse2, |
422 | bool /*noPermute*/) const |
423 | { |
424 | #ifndef NDEBUG |
425 | { |
426 | int lstart=numberRows_+factInfo_.maxinv+5; |
427 | int ndo = factInfo_.xnetal-lstart; |
428 | double * dluval=factInfo_.xeeadr; |
429 | int * mcstrt = factInfo_.xcsadr+lstart; |
430 | if (ndo) |
431 | assert (dluval[mcstrt[ndo]+1]<1.0e50); |
432 | } |
433 | #endif |
434 | assert (numberRows_==numberColumns_); |
435 | double *region2 = regionSparse2->denseVector ( ); |
436 | int *regionIndex2 = regionSparse2->getIndices ( ); |
437 | int numberNonZero = regionSparse2->getNumElements ( ); |
438 | double *region = regionSparse->denseVector ( ); |
439 | //const int * permuteIn = factInfo_.mpermu+1; |
440 | // Stuff is put one up so won't get illegal read |
441 | assert (!region[numberRows_]); |
442 | assert (!regionSparse2->packedMode()); |
443 | #if 0 |
444 | int first=numberRows_; |
445 | for (int j=0;j<numberNonZero;j++) { |
446 | int jRow = regionIndex2[j]; |
447 | int iRow = permuteIn[jRow]; |
448 | region[iRow]=region2[jRow]; |
449 | first=CoinMin(first,iRow); |
450 | region2[jRow]=0.0; |
451 | } |
452 | #endif |
453 | numberNonZero=c_ekkftrn(&factInfo_, |
454 | region2-1,region,regionIndex2,numberNonZero); |
455 | regionSparse2->setNumElements(numberNonZero); |
456 | return 0; |
457 | } |
458 | /* Updates one column (FTRAN) from regionSparse2 |
459 | Tries to do FT update |
460 | number returned is negative if no room |
461 | regionSparse starts as zero and is zero at end. |
462 | Note - if regionSparse2 packed on input - will be packed on output |
463 | */ |
464 | int |
465 | CoinOslFactorization::updateColumnFT ( CoinIndexedVector * regionSparse, |
466 | CoinIndexedVector * regionSparse2, |
467 | bool /*noPermute*/) |
468 | { |
469 | assert (numberRows_==numberColumns_); |
470 | double *region2 = regionSparse2->denseVector ( ); |
471 | int *regionIndex2 = regionSparse2->getIndices ( ); |
472 | int numberNonZero = regionSparse2->getNumElements ( ); |
473 | assert (regionSparse2->packedMode()); |
474 | // packed mode |
475 | //int numberNonInOriginal=numberNonZero; |
476 | //double *dpermu = factInfo_.kadrpm; |
477 | // Use region instead of dpermu |
478 | double * save =factInfo_.kadrpm; |
479 | factInfo_.kadrpm=regionSparse->denseVector()-1; |
480 | int nuspike=c_ekkftrn_ft(&factInfo_, region2,regionIndex2, |
481 | &numberNonZero); |
482 | factInfo_.kadrpm=save; |
483 | regionSparse2->setNumElements(numberNonZero); |
484 | //regionSparse2->print(); |
485 | factInfo_.nuspike=nuspike; |
486 | return nuspike; |
487 | } |
488 | |
489 | |
490 | int |
491 | CoinOslFactorization::updateTwoColumnsFT(CoinIndexedVector * regionSparse1, |
492 | CoinIndexedVector * regionSparse2, |
493 | CoinIndexedVector * regionSparse3, |
494 | bool /*noPermute*/) |
495 | { |
496 | #if 1 |
497 | // probably best to merge on a LU part by part |
498 | // but can try full merge |
499 | double *region2 = regionSparse2->denseVector ( ); |
500 | int *regionIndex2 = regionSparse2->getIndices ( ); |
501 | int numberNonZero2 = regionSparse2->getNumElements ( ); |
502 | assert (regionSparse2->packedMode()); |
503 | |
504 | assert (numberRows_==numberColumns_); |
505 | double *region3 = regionSparse3->denseVector ( ); |
506 | int *regionIndex3 = regionSparse3->getIndices ( ); |
507 | int numberNonZero3 = regionSparse3->getNumElements ( ); |
508 | double *region = regionSparse1->denseVector ( ); |
509 | // Stuff is put one up so won't get illegal read |
510 | assert (!region[numberRows_]); |
511 | assert (!regionSparse3->packedMode()); |
512 | // packed mode |
513 | //double *dpermu = factInfo_.kadrpm; |
514 | #if 0 |
515 | factInfo_.nuspike=c_ekkftrn_ft(&factInfo_, region2,regionIndex2, |
516 | &numberNonZero2); |
517 | numberNonZero3=c_ekkftrn(&factInfo_, |
518 | region3-1,region,regionIndex3,numberNonZero3); |
519 | #else |
520 | c_ekkftrn2(&factInfo_,region3-1,region,regionIndex3,&numberNonZero3, |
521 | region2,regionIndex2,&numberNonZero2); |
522 | #endif |
523 | regionSparse2->setNumElements(numberNonZero2); |
524 | regionSparse3->setNumElements(numberNonZero3); |
525 | return factInfo_.nuspike; |
526 | #else |
527 | // probably best to merge on a LU part by part |
528 | // but can try full merge |
529 | int returnCode= updateColumnFT(regionSparse1, |
530 | regionSparse2); |
531 | updateColumn(regionSparse1, |
532 | regionSparse3, |
533 | noPermute); |
534 | return returnCode; |
535 | #endif |
536 | } |
537 | |
538 | /* Updates one column (BTRAN) from regionSparse2 |
539 | regionSparse starts as zero and is zero at end |
540 | Note - if regionSparse2 packed on input - will be packed on output |
541 | */ |
542 | int |
543 | CoinOslFactorization::updateColumnTranspose ( CoinIndexedVector * regionSparse, |
544 | CoinIndexedVector * regionSparse2) const |
545 | { |
546 | assert (numberRows_==numberColumns_); |
547 | double *region2 = regionSparse2->denseVector ( ); |
548 | int *regionIndex2 = regionSparse2->getIndices ( ); |
549 | int numberNonZero = regionSparse2->getNumElements ( ); |
550 | //double *region = regionSparse->denseVector ( ); |
551 | /*int *regionIndex = regionSparse->getIndices ( );*/ |
552 | const int * permuteIn = factInfo_.mpermu+1; |
553 | factInfo_.packedMode = regionSparse2->packedMode() ? 1 : 0; |
554 | // Use region instead of dpermu |
555 | double * save =factInfo_.kadrpm; |
556 | factInfo_.kadrpm=regionSparse->denseVector()-1; |
557 | // use internal one for now (address is one off) |
558 | double * region = factInfo_.kadrpm; |
559 | if (numberNonZero<2) { |
560 | if (numberNonZero) { |
561 | int ipivrw=regionIndex2[0]; |
562 | if (factInfo_.packedMode) { |
563 | double value=region2[0]; |
564 | region2[0]=0.0; |
565 | region2[ipivrw]=value; |
566 | } |
567 | numberNonZero=c_ekkbtrn_ipivrw(&factInfo_, region2-1, |
568 | regionIndex2-1,ipivrw+1, |
569 | reinterpret_cast<int *>(factInfo_.kp1adr)); |
570 | } |
571 | } else { |
572 | #ifndef NDEBUG |
573 | { |
574 | int *mcstrt = factInfo_.xcsadr; |
575 | int * hpivco_new=factInfo_.kcpadr+1; |
576 | int nrow=factInfo_.nrow; |
577 | int i; |
578 | int ipiv = hpivco_new[0]; |
579 | int last = mcstrt[ipiv]; |
580 | for (i=0;i<nrow-1;i++) { |
581 | ipiv=hpivco_new[ipiv]; |
582 | assert (mcstrt[ipiv]>last); |
583 | last=mcstrt[ipiv]; |
584 | } |
585 | } |
586 | #endif |
587 | int iSmallest = COIN_INT_MAX; |
588 | int iPiv=0; |
589 | const int *mcstrt = factInfo_.xcsadr; |
590 | // permute and save where nonzeros are |
591 | if (!factInfo_.packedMode) { |
592 | if ((numberRows_<200||(numberNonZero<<4)>numberRows_)) { |
593 | for (int j=0;j<numberNonZero;j++) { |
594 | int jRow = regionIndex2[j]; |
595 | int iRow = permuteIn[jRow]; |
596 | regionIndex2[j]=iRow; |
597 | region[iRow]=region2[jRow]; |
598 | region2[jRow]=0.0; |
599 | } |
600 | } else { |
601 | for (int j=0;j<numberNonZero;j++) { |
602 | int jRow = regionIndex2[j]; |
603 | int iRow = permuteIn[jRow]; |
604 | regionIndex2[j]=iRow; |
605 | region[iRow]=region2[jRow]; |
606 | if (mcstrt[iRow]<iSmallest) { |
607 | iPiv=iRow; |
608 | iSmallest=mcstrt[iRow]; |
609 | } |
610 | region2[jRow]=0.0; |
611 | } |
612 | } |
613 | } else { |
614 | for (int j=0;j<numberNonZero;j++) { |
615 | int jRow = regionIndex2[j]; |
616 | int iRow = permuteIn[jRow]; |
617 | regionIndex2[j]=iRow; |
618 | region[iRow]=region2[j]; |
619 | region2[j]=0.0; |
620 | } |
621 | } |
622 | assert (iPiv>=0); |
623 | numberNonZero=c_ekkbtrn(&factInfo_, region2-1,regionIndex2-1,iPiv); |
624 | } |
625 | factInfo_.kadrpm=save; |
626 | factInfo_.packedMode=0; |
627 | regionSparse2->setNumElements(numberNonZero); |
628 | return 0; |
629 | } |
630 | // Number of entries in each row |
631 | int * |
632 | CoinOslFactorization::numberInRow() const |
633 | { return reinterpret_cast<int *> (factInfo_.xrnadr+1);} |
634 | // Number of entries in each column |
635 | int * |
636 | CoinOslFactorization::numberInColumn() const |
637 | { return reinterpret_cast<int *> (factInfo_.xcnadr+1);} |
638 | // Returns array to put basis starts in |
639 | CoinBigIndex * |
640 | CoinOslFactorization::starts() const |
641 | { return reinterpret_cast<CoinBigIndex *> (factInfo_.xcsadr+1);} |
642 | // Returns array to put basis elements in |
643 | CoinFactorizationDouble * |
644 | CoinOslFactorization::elements() const |
645 | { return factInfo_.xeeadr+1;} |
646 | // Returns pivot row |
647 | int * |
648 | CoinOslFactorization::pivotRow() const |
649 | { return factInfo_.krpadr+1;} |
650 | // Returns work area |
651 | CoinFactorizationDouble * |
652 | CoinOslFactorization::workArea() const |
653 | { return factInfo_.kw1adr;} |
654 | // Returns int work area |
655 | int * |
656 | CoinOslFactorization::intWorkArea() const |
657 | { return reinterpret_cast<int *> (factInfo_.kw1adr);} |
658 | // Returns permute back |
659 | int * |
660 | CoinOslFactorization::permuteBack() const |
661 | { return factInfo_.kcpadr+1;} |
662 | // Returns array to put basis indices in |
663 | int * |
664 | CoinOslFactorization::indices() const |
665 | { return factInfo_.xeradr+1;} |
666 | // Returns true if wants tableauColumn in replaceColumn |
667 | bool |
668 | CoinOslFactorization::wantsTableauColumn() const |
669 | { return false;} |
670 | /* Useful information for factorization |
671 | 0 - iteration number |
672 | whereFrom is 0 for factorize and 1 for replaceColumn |
673 | */ |
674 | #ifdef CLP_REUSE_ETAS |
675 | void |
676 | CoinOslFactorization::setUsefulInformation(const int * info,int whereFrom) |
677 | { |
678 | factInfo_.iterno=info[0]; |
679 | if (whereFrom) { |
680 | factInfo_.reintro=-1; |
681 | if( factInfo_.first_dense>=factInfo_.last_dense) { |
682 | int * putSeq = factInfo_.xrsadr+2*factInfo_.nrowmx+2; |
683 | int * position = putSeq+factInfo_.maxinv; |
684 | //int * putStart = position+factInfo_.maxinv; |
685 | int iSequence=info[1]; |
686 | if (whereFrom==1) { |
687 | putSeq[factInfo_.npivots]=iSequence; |
688 | } else { |
689 | int i; |
690 | for (i=factInfo_.npivots-1;i>=0;i--) { |
691 | if (putSeq[i]==iSequence) |
692 | break; |
693 | } |
694 | if (i>=0) { |
695 | factInfo_.reintro=position[i]; |
696 | } else { |
697 | factInfo_.reintro=-1; |
698 | } |
699 | factInfo_.nnentu=factInfo_.save_nnentu; |
700 | } |
701 | } |
702 | } |
703 | } |
704 | #else |
705 | void |
706 | CoinOslFactorization::setUsefulInformation(const int * info,int /*whereFrom*/) |
707 | { factInfo_.iterno=info[0]; } |
708 | #endif |
709 | |
710 | // Get rid of all memory |
711 | void |
712 | CoinOslFactorization::clearArrays() |
713 | { |
714 | factInfo_.nR_etas=0; |
715 | factInfo_.nnentu=0; |
716 | factInfo_.nnentl=0; |
717 | maximumRows_=0; |
718 | maximumSpace_=0; |
719 | factInfo_.last_eta_size=0; |
720 | gutsOfDestructor(false); |
721 | } |
722 | void |
723 | CoinOslFactorization::maximumPivots ( int value ) |
724 | { |
725 | maximumPivots_ = value; |
726 | } |
727 | #define CLP_FILL 15 |
728 | /*#undef NDEBUG*/ |
729 | //#define CLP_DEBUG_MALLOC 1000000 |
730 | #if CLP_DEBUG_MALLOC |
731 | static int malloc_number=0; |
732 | static int malloc_check=-1; |
733 | static int malloc_counts_on=0; |
734 | struct malloc_struct { |
735 | void * previous; |
736 | void * next; |
737 | int size; |
738 | int when; |
739 | int type; |
740 | }; |
741 | static double malloc_times=0.0; |
742 | static double malloc_total=0.0; |
743 | static double malloc_current=0.0; |
744 | static double malloc_max=0.0; |
745 | static int malloc_amount[]={0,32,128,256,1024,4096,16384,65536,262144, |
746 | 2000000000}; |
747 | static int malloc_n=10; |
748 | double malloc_counts[10]={0,0,0,0,0,0,0,0,0,0}; |
749 | typedef struct malloc_struct malloc_struct; |
750 | static malloc_struct startM = {0,0,0,0}; |
751 | static malloc_struct endM = {0,0,0,0}; |
752 | static int extra=4; |
753 | void clp_memory(int type) |
754 | { |
755 | if (type==0) { |
756 | /* switch on */ |
757 | malloc_counts_on=1; |
758 | startM.next=&endM; |
759 | endM.previous=&startM; |
760 | } else { |
761 | /* summary */ |
762 | double average = malloc_total/malloc_times; |
763 | int i; |
764 | malloc_struct * previous = (malloc_struct *) endM.previous; |
765 | printf("count %g bytes %g - average %g\n" ,malloc_times,malloc_total,average); |
766 | printf("current bytes %g - maximum %g\n" ,malloc_current,malloc_max); |
767 | |
768 | for ( i=0;i<malloc_n;i++) |
769 | printf("%g " ,malloc_counts[i]); |
770 | printf("\n" ); |
771 | malloc_counts_on=0; |
772 | if (previous->previous!=&startM) { |
773 | int n=0; |
774 | printf("Allocated blocks\n" ); |
775 | while (previous->previous!=&startM) { |
776 | printf("(%d at %d) " ,previous->size,previous->when); |
777 | n++; |
778 | if ((n%5)==0) |
779 | printf("\n" ); |
780 | previous = (malloc_struct *) previous->previous; |
781 | } |
782 | printf("\n - total %d\n" ,n); |
783 | } |
784 | } |
785 | malloc_number=0; |
786 | malloc_times=0.0; |
787 | malloc_total=0.0; |
788 | malloc_current=0.0; |
789 | malloc_max=0.0; |
790 | memset(malloc_counts,0,sizeof(malloc_counts)); |
791 | } |
792 | static void clp_adjust(void * temp,int size,int type) |
793 | { |
794 | malloc_struct * itemp = (malloc_struct *) temp; |
795 | malloc_struct * first = (malloc_struct *) startM.next; |
796 | int i; |
797 | startM.next=temp; |
798 | first->previous=temp; |
799 | itemp->previous=&startM; |
800 | itemp->next=first; |
801 | malloc_number++; |
802 | if (malloc_number==malloc_check) { |
803 | printf("Allocation of %d bytes at %d (type %d) not freed\n" , |
804 | size,malloc_number,type); |
805 | } |
806 | itemp->when=malloc_number; |
807 | itemp->size=size; |
808 | itemp->type=type; |
809 | malloc_times ++; |
810 | malloc_total += size; |
811 | malloc_current += size; |
812 | malloc_max=CoinMax(malloc_max,malloc_current); |
813 | for (i=0;i<malloc_n;i++) { |
814 | if ((int) size<=malloc_amount[i]) { |
815 | malloc_counts[i]++; |
816 | break; |
817 | } |
818 | } |
819 | } |
820 | #endif |
821 | /* covers */ |
822 | double * clp_double(int number_entries) |
823 | { |
824 | #if CLP_DEBUG_MALLOC==0 |
825 | return reinterpret_cast<double *>( malloc(number_entries*sizeof(double))); |
826 | #else |
827 | double * temp = reinterpret_cast<double *>( malloc((number_entries+extra)*sizeof(double))); |
828 | clp_adjust(temp,number_entries*sizeof(double),1); |
829 | #if CLP_DEBUG_MALLOC>1 |
830 | if (number_entries*sizeof(double)>=CLP_DEBUG_MALLOC) |
831 | printf("WWW %x malloced by double %d - size %d\n" , |
832 | temp+extra,malloc_number,number_entries); |
833 | #endif |
834 | return temp+extra; |
835 | #endif |
836 | } |
837 | int * clp_int(int number_entries) |
838 | { |
839 | #if CLP_DEBUG_MALLOC==0 |
840 | return reinterpret_cast<int *>( malloc(number_entries*sizeof(int))); |
841 | #else |
842 | double * temp = reinterpret_cast<double *>( malloc(((number_entries+1)/2+extra)*sizeof(double))); |
843 | clp_adjust(temp,number_entries*sizeof(int),2); |
844 | #if CLP_DEBUG_MALLOC>1 |
845 | if (number_entries*sizeof(int)>=CLP_DEBUG_MALLOC) |
846 | printf("WWW %x malloced by int %d - size %d\n" , |
847 | temp+extra,malloc_number,number_entries); |
848 | #endif |
849 | return reinterpret_cast<int *>( (temp+extra)); |
850 | #endif |
851 | } |
852 | void * clp_malloc(int number_entries) |
853 | { |
854 | #if CLP_DEBUG_MALLOC==0 |
855 | return malloc(number_entries); |
856 | #else |
857 | double * temp = reinterpret_cast<double *>( malloc(number_entries+extra*sizeof(double))); |
858 | clp_adjust(temp,number_entries,0); |
859 | #if CLP_DEBUG_MALLOC>1 |
860 | if (number_entries>=CLP_DEBUG_MALLOC) |
861 | printf("WWW %x malloced by void %d - size %d\n" , |
862 | temp+extra,malloc_number,number_entries); |
863 | #endif |
864 | return (void *) (temp+extra); |
865 | #endif |
866 | } |
867 | void clp_free(void * oldArray) |
868 | { |
869 | #if CLP_DEBUG_MALLOC==0 |
870 | free(oldArray); |
871 | #else |
872 | if (oldArray) { |
873 | double * temp = (reinterpret_cast<double *>( oldArray)-extra); |
874 | malloc_struct * itemp = (malloc_struct *) temp; |
875 | malloc_struct * next = (malloc_struct *) itemp->next; |
876 | malloc_struct * previous = (malloc_struct *) itemp->previous; |
877 | previous->next=next; |
878 | next->previous=previous; |
879 | malloc_current -= itemp->size; |
880 | #if CLP_DEBUG_MALLOC>1 |
881 | if (itemp->size>=CLP_DEBUG_MALLOC) |
882 | printf("WWW %x freed by free %d - old length %d - type %d\n" , |
883 | oldArray,itemp->when,itemp->size,itemp->type); |
884 | #endif |
885 | free(temp); |
886 | } |
887 | #endif |
888 | } |
889 | /*#define FIX_ADD 4*nrowmx+5 |
890 | #define FIX_ADD2 4*nrowmx+5*/ |
891 | #define FIX_ADD 1*nrowmx+5 |
892 | #define FIX_ADD2 1*nrowmx+5 |
893 | #define ALIGNMENT 32 |
894 | static void * clp_align (void * memory) |
895 | { |
896 | if (sizeof(int)==sizeof(void *)&&ALIGNMENT) { |
897 | CoinInt64 k = reinterpret_cast<CoinInt64> (memory); |
898 | if ((k&(ALIGNMENT-1))!=0) { |
899 | k &= ~(ALIGNMENT-1); |
900 | k += ALIGNMENT; |
901 | memory = reinterpret_cast<void *> (k); |
902 | } |
903 | return memory; |
904 | } else { |
905 | return memory; |
906 | } |
907 | } |
908 | void clp_setup_pointers(EKKfactinfo * fact) |
909 | { |
910 | /* do extra stuff */ |
911 | int nrow=fact->nrow; |
912 | int nrowmx=fact->nrowmx; |
913 | int maxinv=fact->maxinv; |
914 | fact->lstart = nrow + maxinv + 5; |
915 | /* this is the number of L transforms */ |
916 | fact->xnetalval = fact->xnetal - fact->lstart; |
917 | fact->mpermu = (reinterpret_cast<int*> (fact->kadrpm+nrow))+1; |
918 | fact->bitArray = fact->krpadr + ( nrowmx+2); |
919 | fact->back = fact->kcpadr+2*nrow + maxinv + 4; |
920 | fact->hpivcoR = fact->kcpadr+nrow+3; |
921 | fact->nonzero = (reinterpret_cast<char *>( &fact->mpermu[nrow+1]))-1; |
922 | } |
923 | #ifndef NDEBUG |
924 | int ets_count=0; |
925 | int ets_check=-1; |
926 | //static int adjust_count=0; |
927 | //static int adjust_check=-1; |
928 | #endif |
929 | static void clp_adjust_pointers(EKKfactinfo * fact, int adjust) |
930 | { |
931 | #if 0 //ndef NDEBUG |
932 | adjust_count++; |
933 | if (adjust_check>=0&&adjust_count>=adjust_check) { |
934 | printf("trouble\n" ); |
935 | } |
936 | #endif |
937 | if (fact->trueStart) { |
938 | fact->kadrpm += adjust; |
939 | fact->krpadr += adjust; |
940 | fact->kcpadr += adjust; |
941 | fact->xrsadr += adjust; |
942 | fact->xcsadr += adjust; |
943 | fact->xrnadr += adjust; |
944 | fact->xcnadr += adjust; |
945 | } |
946 | if (fact->xeradr) { |
947 | fact->xeradr += adjust; |
948 | fact->xecadr += adjust; |
949 | fact->xeeadr += adjust; |
950 | } |
951 | } |
952 | /* deals with memory for complicated array |
953 | 0 just do addresses |
954 | 1 just get memory */ |
955 | static double * |
956 | clp_alloc_memory(EKKfactinfo * fact,int type, int * length) |
957 | { |
958 | int nDouble=0; |
959 | int nInt=0; |
960 | int nrowmxp; |
961 | int ntot1; |
962 | int ntot2; |
963 | int ntot3; |
964 | int nrowmx; |
965 | int * tempI; |
966 | double * tempD; |
967 | nrowmx=fact->nrowmx; |
968 | nrowmxp = nrowmx + 2; |
969 | ntot1 = nrowmxp; |
970 | ntot2 = 3*nrowmx+5; /* space for three lists */ |
971 | ntot3 = 2*nrowmx; |
972 | if ((ntot1<<1)<ntot2) { |
973 | ntot1=ntot2>>1; |
974 | } |
975 | ntot3=CoinMax(ntot3,ntot1); |
976 | /* Row work regions */ |
977 | /* must be contiguous so allocate as one chunk */ |
978 | /* may only need 2.5 */ |
979 | /* now doing all at once - far too much - reduce later */ |
980 | tempD=fact->kw1adr; |
981 | tempD+=nrowmxp; |
982 | tempD = reinterpret_cast<double *>( clp_align(tempD)); |
983 | fact->kw2adr=tempD; |
984 | tempD+=nrowmxp; |
985 | tempD = reinterpret_cast<double *>( clp_align(tempD)); |
986 | fact->kw3adr=tempD-1; |
987 | tempD+=nrowmxp; |
988 | tempD = reinterpret_cast<double *>( clp_align(tempD)); |
989 | fact->kp1adr=reinterpret_cast<EKKHlink *>(tempD); |
990 | tempD+=nrowmxp; |
991 | tempD = reinterpret_cast<double *>( clp_align(tempD)); |
992 | fact->kp2adr=reinterpret_cast<EKKHlink *>(tempD); |
993 | //tempD+=ntot3; |
994 | tempD+=nrowmxp; |
995 | tempD = reinterpret_cast<double *>( clp_align(tempD)); |
996 | /*printf("zz %x %x\n",tempD,fact->kadrpm);*/ |
997 | fact->kadrpm = tempD; |
998 | /* seems a lot */ |
999 | tempD += ((6*nrowmx +8)*(sizeof(int))/sizeof(double)); |
1000 | /* integer arrays */ |
1001 | tempI = reinterpret_cast<int *>( tempD); |
1002 | tempI = reinterpret_cast<int *>( clp_align(tempI)); |
1003 | fact->xrsadr = tempI; |
1004 | #ifdef CLP_REUSE_ETAS |
1005 | tempI +=( 3*(nrowmx+fact->maxinv+1)); |
1006 | #else |
1007 | tempI +=( (nrowmx<<1)+fact->maxinv+1); |
1008 | #endif |
1009 | tempI = reinterpret_cast<int *>( clp_align(tempI)); |
1010 | fact->xcsadr = tempI; |
1011 | #if 1 //def CLP_REUSE_ETAS |
1012 | tempI += ( 2*nrowmx+8+2*fact->maxinv); |
1013 | #else |
1014 | tempI += ( 2*nrowmx+8+fact->maxinv); |
1015 | #endif |
1016 | tempI += FIX_ADD+FIX_ADD2; |
1017 | tempI = reinterpret_cast<int *>( clp_align(tempI)); |
1018 | fact->xrnadr = tempI; |
1019 | tempI += nrowmx; |
1020 | tempI = reinterpret_cast<int *>( clp_align(tempI)); |
1021 | fact->xcnadr = tempI; |
1022 | tempI += nrowmx; |
1023 | tempI = reinterpret_cast<int *>( clp_align(tempI)); |
1024 | fact->krpadr = tempI; |
1025 | tempI += ( nrowmx+1) +((nrowmx+33)>>5); |
1026 | /*printf("zzz %x %x\n",tempI,fact->kcpadr);*/ |
1027 | tempI = reinterpret_cast<int *>( clp_align(tempI)); |
1028 | fact->kcpadr = tempI; |
1029 | tempI += 3*nrowmx+8+fact->maxinv; |
1030 | fact->R_etas_start = fact->xcsadr+nrowmx+fact->maxinv+4; |
1031 | fact->R_etas_start += FIX_ADD; |
1032 | nInt = static_cast<int>(tempI-(reinterpret_cast<int *>( fact->trueStart))); |
1033 | nDouble = static_cast<int>(sizeof(int)*(nInt+1)/sizeof(double)); |
1034 | *length = nDouble; |
1035 | /*printf("nDouble %d - type %d\n",nDouble,type);*/ |
1036 | nDouble += static_cast<int>((2*ALIGNMENT)/sizeof(double)); |
1037 | if (type) { |
1038 | /*printf("%d allocated\n",nDouble);*/ |
1039 | tempD = reinterpret_cast<double *>( clp_double(nDouble)); |
1040 | #ifndef NDEBUG |
1041 | memset(tempD,CLP_FILL,nDouble*sizeof(double)); |
1042 | #endif |
1043 | } |
1044 | return tempD; |
1045 | } |
1046 | static void c_ekksmem(EKKfactinfo *fact,int nrow,int maximumPivots) |
1047 | { |
1048 | /* space for invert */ |
1049 | int nnetas=fact->eta_size; |
1050 | fact->nrow=nrow; |
1051 | if (!(nnetas>fact->last_eta_size||(!fact->xe2adr&&fact->if_sparse_update)|| |
1052 | nrow>fact->nrowmx||maximumPivots>fact->maxinv)) |
1053 | return; |
1054 | clp_adjust_pointers(fact, +1); |
1055 | if (nrow>fact->nrowmx||maximumPivots>fact->maxinv) { |
1056 | int length; |
1057 | fact->nrowmx=CoinMax(nrow,fact->nrowmx); |
1058 | fact->maxinv=CoinMax(maximumPivots,fact->maxinv); |
1059 | clp_free(fact->trueStart); |
1060 | fact->trueStart=0; |
1061 | fact->kw1adr=0; |
1062 | fact->trueStart=clp_alloc_memory(fact,1,&length); |
1063 | fact->kw1adr=reinterpret_cast<double *>( clp_align(fact->trueStart)); |
1064 | clp_alloc_memory(fact,0,&length); |
1065 | } |
1066 | /*if (!fact->iterno) fact->eta_size+=1000000;*//* TEMP*/ |
1067 | if (nnetas>fact->last_eta_size||(!fact->xe2adr&&fact->if_sparse_update)) { |
1068 | fact->last_eta_size = nnetas; |
1069 | clp_free(reinterpret_cast<char *>(fact->xe2adr)); |
1070 | /* if malloc fails - we have lost memory - start again */ |
1071 | if (!fact->ndenuc &&fact->if_sparse_update) { |
1072 | /* allow second copy of elements */ |
1073 | fact->xe2adr = clp_double(nnetas); |
1074 | #ifndef NDEBUG |
1075 | memset(fact->xe2adr,CLP_FILL,nnetas*sizeof(double)); |
1076 | #endif |
1077 | if (!fact->xe2adr) { |
1078 | fact->maxNNetas=fact->last_eta_size; /* dont allow any increase */ |
1079 | nnetas=fact->last_eta_size; |
1080 | fact->eta_size=nnetas; |
1081 | #ifdef PRINT_DEBUG |
1082 | if (fact->if_sparse_update) { |
1083 | printf("*** Sparse update off due to memory\n" ); |
1084 | } |
1085 | #endif |
1086 | fact->if_sparse_update=0; |
1087 | fact->switch_off_sparse_update=1; |
1088 | } |
1089 | } else { |
1090 | fact->xe2adr = 0; |
1091 | fact->if_sparse_update=0; |
1092 | } |
1093 | clp_free(fact->xeradr); |
1094 | fact->xeradr= clp_int( nnetas); |
1095 | #ifndef NDEBUG |
1096 | memset(fact->xeradr,CLP_FILL,nnetas*sizeof(int)); |
1097 | #endif |
1098 | if (!fact->xeradr) { |
1099 | nnetas=0; |
1100 | } |
1101 | if (nnetas) { |
1102 | clp_free(fact->xecadr); |
1103 | fact->xecadr= clp_int( nnetas); |
1104 | #ifndef NDEBUG |
1105 | memset(fact->xecadr,CLP_FILL,nnetas*sizeof(int)); |
1106 | #endif |
1107 | if (!fact->xecadr) { |
1108 | nnetas=0; |
1109 | } |
1110 | } |
1111 | if (nnetas) { |
1112 | clp_free(fact->xeeadr); |
1113 | fact->xeeadr= clp_double(nnetas); |
1114 | #ifndef NDEBUG |
1115 | memset(fact->xeeadr,CLP_FILL,nnetas*sizeof(double)); |
1116 | #endif |
1117 | if (!fact->xeeadr) { |
1118 | nnetas=0; |
1119 | } |
1120 | } |
1121 | } |
1122 | if (!nnetas) { |
1123 | char msg[100]; |
1124 | sprintf(msg,"Unable to allocate factorization memory for %d elements" , |
1125 | nnetas); |
1126 | throw(msg); |
1127 | } |
1128 | /*c_ekklplp->nnetas=nnetas;*/ |
1129 | fact->nnetas=nnetas; |
1130 | clp_adjust_pointers(fact, -1); |
1131 | } |
1132 | static void c_ekksmem_copy(EKKfactinfo *fact,const EKKfactinfo * rhsFact) |
1133 | { |
1134 | /* space for invert */ |
1135 | int nrowmx=rhsFact->nrowmx,nnetas=rhsFact->nnetas; |
1136 | int canReuseEtas= (fact->eta_size==rhsFact->eta_size) ? 1 : 0; |
1137 | int canReuseArrays = (fact->nrowmx==rhsFact->nrowmx) ? 1 : 0; |
1138 | clp_adjust_pointers(fact, +1); |
1139 | clp_adjust_pointers(const_cast<EKKfactinfo *>(rhsFact), +1); |
1140 | /*memset(fact,0,sizeof(EKKfactinfo));*/ |
1141 | /* copy scalars */ |
1142 | memcpy(&fact->drtpiv,&rhsFact->drtpiv,5*sizeof(double)); |
1143 | memcpy(&fact->nrow,&rhsFact->nrow,((&fact->maxNNetas-&fact->nrow)+1)* |
1144 | sizeof(int)); |
1145 | if (nrowmx) { |
1146 | int length; |
1147 | int kCopyEnd,nCopyEnd,nCopyStart; |
1148 | if (!canReuseEtas) { |
1149 | clp_free(fact->xeradr); |
1150 | clp_free(fact->xecadr); |
1151 | clp_free(fact->xeeadr); |
1152 | clp_free(fact->xe2adr); |
1153 | fact->xeradr = 0; |
1154 | fact->xecadr = 0; |
1155 | fact->xeeadr = 0; |
1156 | fact->xe2adr = 0; |
1157 | } |
1158 | if (!canReuseArrays) { |
1159 | clp_free(fact->trueStart); |
1160 | fact->trueStart=0; |
1161 | fact->kw1adr=0; |
1162 | fact->trueStart=clp_alloc_memory(fact,1,&length); |
1163 | fact->kw1adr=reinterpret_cast<double *>( clp_align(fact->trueStart)); |
1164 | } |
1165 | clp_alloc_memory(fact,0,&length); |
1166 | nnetas=fact->eta_size; |
1167 | assert (nnetas); |
1168 | { |
1169 | int n2 = rhsFact->nR_etas; |
1170 | int n3 = n2 ? rhsFact->R_etas_start[1+n2]: 0; |
1171 | int * startR = rhsFact->R_etas_index+n3; |
1172 | nCopyEnd=static_cast<int>((rhsFact->xeradr+nnetas)-startR); |
1173 | nCopyStart=rhsFact->nnentu; |
1174 | nCopyEnd = CoinMin(nCopyEnd+20,nnetas); |
1175 | kCopyEnd = nnetas-nCopyEnd; |
1176 | nCopyStart = CoinMin(nCopyStart+20,nnetas); |
1177 | if (!n2&&!rhsFact->nnentu&&!rhsFact->nnentl) { |
1178 | nCopyStart=nCopyEnd=0; |
1179 | } |
1180 | } |
1181 | /* copy */ |
1182 | if(nCopyStart||nCopyEnd||true) { |
1183 | #if 1 |
1184 | memcpy(fact->kw1adr,rhsFact->kw1adr,length*sizeof(double)); |
1185 | #else |
1186 | c_ekkscpy((length*sizeof(double))/sizeof(int), |
1187 | reinterpret_cast<int *>( rhsFact->kw1adr,reinterpret_cast<int *>( fact->kw1adr)); |
1188 | #endif |
1189 | } |
1190 | /* if malloc fails - we have lost memory - start again */ |
1191 | if (!fact->ndenuc &&fact->if_sparse_update) { |
1192 | /* allow second copy of elements */ |
1193 | if (!canReuseEtas) |
1194 | fact->xe2adr = clp_double(nnetas); |
1195 | if (!fact->xe2adr) { |
1196 | fact->maxNNetas=nnetas; /* dont allow any increase */ |
1197 | #ifdef PRINT_DEBUG |
1198 | if (fact->if_sparse_update) { |
1199 | printf("*** Sparse update off due to memory\n" ); |
1200 | } |
1201 | #endif |
1202 | fact->if_sparse_update=0; |
1203 | } else { |
1204 | #ifndef NDEBUG |
1205 | memset(fact->xe2adr,CLP_FILL,nnetas*sizeof(double)); |
1206 | #endif |
1207 | } |
1208 | } else { |
1209 | clp_free(fact->xe2adr); |
1210 | fact->xe2adr = 0; |
1211 | fact->if_sparse_update=0; |
1212 | } |
1213 | |
1214 | if (!canReuseEtas) |
1215 | fact->xeradr= clp_int(nnetas); |
1216 | if (!fact->xeradr) { |
1217 | nnetas=0; |
1218 | } else { |
1219 | #ifndef NDEBUG |
1220 | memset(fact->xeradr,CLP_FILL,nnetas*sizeof(int)); |
1221 | #endif |
1222 | |
1223 | /* copy */ |
1224 | if(nCopyStart||nCopyEnd) { |
1225 | #if 0 |
1226 | memcpy(fact->xeradr,rhsFact->xeradr,nCopyStart*sizeof(int)); |
1227 | memcpy(fact->xeradr+kCopyEnd,rhsFact->xeradr+kCopyEnd,nCopyEnd*sizeof(int)); |
1228 | #else |
1229 | c_ekkscpy(nCopyStart,rhsFact->xeradr,fact->xeradr); |
1230 | c_ekkscpy(nCopyEnd,rhsFact->xeradr+kCopyEnd,fact->xeradr+kCopyEnd); |
1231 | #endif |
1232 | } |
1233 | } |
1234 | if (nnetas) { |
1235 | if (!canReuseEtas) |
1236 | fact->xecadr= clp_int(nnetas); |
1237 | if (!fact->xecadr) { |
1238 | nnetas=0; |
1239 | } else { |
1240 | #ifndef NDEBUG |
1241 | memset(fact->xecadr,CLP_FILL,nnetas*sizeof(int)); |
1242 | #endif |
1243 | /* copy */ |
1244 | if (fact->rows_ok&&(nCopyStart||nCopyEnd)) { |
1245 | int i; |
1246 | int * hcoliR=rhsFact->xecadr-1; |
1247 | int * hcoli=fact->xecadr-1; |
1248 | int * mrstrt=fact->xrsadr; |
1249 | int * hinrow=fact->xrnadr; |
1250 | #if 0 |
1251 | memcpy(fact->xecadr+kCopyEnd,rhsFact->xecadr+kCopyEnd, |
1252 | nCopyEnd*sizeof(int)); |
1253 | #else |
1254 | c_ekkscpy(nCopyEnd,rhsFact->xecadr+kCopyEnd,fact->xecadr+kCopyEnd); |
1255 | #endif |
1256 | if (!fact->xe2adr) { |
1257 | for (i=0;i<fact->nrow;i++) { |
1258 | int istart = mrstrt[i]; |
1259 | assert (istart>0&&istart<=nnetas); |
1260 | assert (hinrow[i]>=0&&hinrow[i]<=fact->nrow); |
1261 | memcpy(hcoli+istart,hcoliR+istart,hinrow[i]*sizeof(int)); |
1262 | } |
1263 | } else { |
1264 | double * de2valR=rhsFact->xe2adr-1; |
1265 | double * de2val=fact->xe2adr-1; |
1266 | #if 0 |
1267 | memcpy(fact->xe2adr+kCopyEnd,rhsFact->xe2adr+kCopyEnd, |
1268 | nCopyEnd*sizeof(double)); |
1269 | #else |
1270 | c_ekkdcpy(nCopyEnd,rhsFact->xe2adr+kCopyEnd |
1271 | ,fact->xe2adr+kCopyEnd); |
1272 | #endif |
1273 | for (i=0;i<fact->nrow;i++) { |
1274 | int istart = mrstrt[i]; |
1275 | assert (istart>0&&istart<=nnetas); |
1276 | assert (hinrow[i]>=0&&hinrow[i]<=fact->nrow); |
1277 | memcpy(hcoli+istart,hcoliR+istart,hinrow[i]*sizeof(int)); |
1278 | memcpy(de2val+istart,de2valR+istart,hinrow[i]*sizeof(double)); |
1279 | #ifndef NDEBUG |
1280 | { |
1281 | int j; |
1282 | for (j=istart;j<istart+hinrow[i];j++) |
1283 | assert (fabs(de2val[j])<1.0e50); |
1284 | } |
1285 | #endif |
1286 | } |
1287 | } |
1288 | } |
1289 | } |
1290 | } |
1291 | if (nnetas) { |
1292 | if (!canReuseEtas) |
1293 | fact->xeeadr= clp_double(nnetas); |
1294 | if (!fact->xeeadr) { |
1295 | nnetas=0; |
1296 | } else { |
1297 | #ifndef NDEBUG |
1298 | memset(fact->xeeadr,CLP_FILL,nnetas*sizeof(double)); |
1299 | #endif |
1300 | /* copy */ |
1301 | if(nCopyStart||nCopyEnd) { |
1302 | #if 0 |
1303 | memcpy(fact->xeeadr,rhsFact->xeeadr,nCopyStart*sizeof(double)); |
1304 | memcpy(fact->xeeadr+kCopyEnd,rhsFact->xeeadr+kCopyEnd,nCopyEnd*sizeof(double)); |
1305 | #else |
1306 | c_ekkdcpy(nCopyStart, |
1307 | rhsFact->xeeadr,fact->xeeadr); |
1308 | c_ekkdcpy(nCopyEnd, |
1309 | rhsFact->xeeadr+kCopyEnd, |
1310 | fact->xeeadr+kCopyEnd); |
1311 | #endif |
1312 | } |
1313 | /*fact->R_etas_index = &XERADR1()[kdnspt - 1]; |
1314 | fact->R_etas_element = &XEEADR1()[kdnspt - 1];*/ |
1315 | fact->R_etas_start = fact->xcsadr+ |
1316 | (rhsFact->R_etas_start-rhsFact->xcsadr); |
1317 | fact->R_etas_index = fact->xeradr+ |
1318 | (rhsFact->R_etas_index-rhsFact->xeradr); |
1319 | fact->R_etas_element = fact->xeeadr+ |
1320 | (rhsFact->R_etas_element-rhsFact->xeeadr); |
1321 | } |
1322 | } |
1323 | } |
1324 | assert (nnetas||!nrowmx); |
1325 | fact->nnetas=nnetas; |
1326 | clp_adjust_pointers(fact, -1); |
1327 | clp_setup_pointers(fact); |
1328 | clp_adjust_pointers(const_cast<EKKfactinfo *>(rhsFact), -1); |
1329 | } |
1330 | static void c_ekksmem_delete(EKKfactinfo *fact) |
1331 | { |
1332 | clp_adjust_pointers(fact, +1); |
1333 | clp_free(fact->trueStart); |
1334 | clp_free(fact->xe2adr); |
1335 | clp_free(fact->xecadr); |
1336 | clp_free(fact->xeradr); |
1337 | clp_free(fact->xeeadr); |
1338 | fact->eta_size=0; |
1339 | fact->xrsadr = 0; |
1340 | fact->xcsadr = 0; |
1341 | fact->xrnadr = 0; |
1342 | fact->xcnadr = 0; |
1343 | fact->krpadr = 0; |
1344 | fact->kcpadr = 0; |
1345 | fact->xeradr = 0; |
1346 | fact->xecadr = 0; |
1347 | fact->xeeadr = 0; |
1348 | fact->xe2adr = 0; |
1349 | fact->trueStart = 0; |
1350 | fact->kw2adr = 0; |
1351 | fact->kw3adr = 0; |
1352 | fact->kp1adr = 0; |
1353 | fact->kp2adr = 0; |
1354 | fact->kadrpm = 0; |
1355 | fact->kw1adr = 0; |
1356 | } |
1357 | int c_ekk_IsSet(const int * array,int bit); |
1358 | void c_ekk_Set(int * array,int bit); |
1359 | void c_ekk_Unset(int * array,int bit); |
1360 | int c_ekk_IsSet(const int * array,int bit) |
1361 | { |
1362 | int iWord = bit>>5; |
1363 | int iBit = bit&31; |
1364 | int word = array[iWord]; |
1365 | return (word&(1<<iBit))!=0; |
1366 | } |
1367 | void c_ekk_Set(int * array,int bit) |
1368 | { |
1369 | int iWord = bit>>5; |
1370 | int iBit = bit&31; |
1371 | array[iWord] |= (1<<iBit); |
1372 | } |
1373 | void c_ekk_Unset(int * array,int bit) |
1374 | { |
1375 | int iWord = bit>>5; |
1376 | int iBit = bit&31; |
1377 | array[iWord] &= ~(1<<iBit); |
1378 | } |
1379 | int CoinOslFactorization::factorize ( |
1380 | const CoinPackedMatrix & matrix, |
1381 | int rowIsBasic[], |
1382 | int columnIsBasic[], |
1383 | double areaFactor ) |
1384 | { |
1385 | setSolveMode(10); |
1386 | if (areaFactor) |
1387 | factInfo_.areaFactor = areaFactor; |
1388 | const int * row = matrix.getIndices(); |
1389 | const CoinBigIndex * columnStart = matrix.getVectorStarts(); |
1390 | const int * columnLength = matrix.getVectorLengths(); |
1391 | const double * element = matrix.getElements(); |
1392 | int numberRows=matrix.getNumRows(); |
1393 | int numberColumns=matrix.getNumCols(); |
1394 | int numberBasic = 0; |
1395 | CoinBigIndex numberElements=0; |
1396 | int numberRowBasic=0; |
1397 | |
1398 | // compute how much in basis |
1399 | |
1400 | int i; |
1401 | // Move pivot variables across if they look good |
1402 | int * pivotTemp = new int [numberRows]; |
1403 | |
1404 | for (i=0;i<numberRows;i++) { |
1405 | if (rowIsBasic[i]>=0) |
1406 | pivotTemp[numberRowBasic++]=i; |
1407 | } |
1408 | |
1409 | numberBasic = numberRowBasic; |
1410 | |
1411 | for (i=0;i<numberColumns;i++) { |
1412 | if (columnIsBasic[i]>=0) { |
1413 | pivotTemp[numberBasic++]=i; |
1414 | numberElements += columnLength[i]; |
1415 | } |
1416 | } |
1417 | if ( numberBasic > numberRows ) { |
1418 | return -2; // say too many in basis |
1419 | } |
1420 | numberElements = 3 * numberRows + 3 * numberElements + 20000; |
1421 | setUsefulInformation(&numberRows,0); |
1422 | getAreas ( numberRows, numberRows, numberElements, |
1423 | 2 * numberElements ); |
1424 | //fill |
1425 | numberBasic=0; |
1426 | numberElements=0; |
1427 | // Fill in counts so we can skip part of preProcess |
1428 | double * elementU=elements(); |
1429 | int * indexRowU=indices(); |
1430 | int * startColumnU=starts(); |
1431 | int * numberInRow=this->numberInRow(); |
1432 | int * numberInColumn=this->numberInColumn(); |
1433 | CoinZeroN ( numberInRow, numberRows ); |
1434 | CoinZeroN ( numberInColumn, numberRows ); |
1435 | for (i=0;i<numberRowBasic;i++) { |
1436 | int iRow = pivotTemp[i]; |
1437 | // Change pivotTemp to correct sequence |
1438 | pivotTemp[i]=iRow+numberColumns; |
1439 | indexRowU[i]=iRow; |
1440 | startColumnU[i]=i; |
1441 | elementU[i]=-1.0; |
1442 | numberInRow[iRow]=1; |
1443 | numberInColumn[i]=1; |
1444 | } |
1445 | startColumnU[numberRowBasic]=numberRowBasic; |
1446 | numberElements=numberRowBasic; |
1447 | numberBasic=numberRowBasic; |
1448 | for (i=0;i<numberColumns;i++) { |
1449 | if (columnIsBasic[i]>=0) { |
1450 | CoinBigIndex j; |
1451 | for (j=columnStart[i];j<columnStart[i]+columnLength[i];j++) { |
1452 | int iRow=row[j]; |
1453 | numberInRow[iRow]++; |
1454 | indexRowU[numberElements]=iRow; |
1455 | elementU[numberElements++]=element[j]; |
1456 | } |
1457 | numberInColumn[numberBasic]=columnLength[i]; |
1458 | numberBasic++; |
1459 | startColumnU[numberBasic]=numberElements; |
1460 | } |
1461 | } |
1462 | preProcess ( ); |
1463 | factor ( ); |
1464 | if (status() == 0) { |
1465 | int * pivotVariable = new int [numberRows]; |
1466 | postProcess(pivotTemp,pivotVariable); |
1467 | for (i=0;i<numberRows;i++) { |
1468 | int iPivot=pivotVariable[i]; |
1469 | if (iPivot<numberColumns) { |
1470 | assert (columnIsBasic[iPivot]>=0); |
1471 | columnIsBasic[iPivot]=i; |
1472 | } else { |
1473 | iPivot-=numberColumns; |
1474 | assert (rowIsBasic[iPivot]>=0); |
1475 | rowIsBasic[iPivot]=i; |
1476 | } |
1477 | } |
1478 | delete [] pivotVariable; |
1479 | } |
1480 | delete [] pivotTemp; |
1481 | return status_; |
1482 | } |
1483 | // Condition number - product of pivots after factorization |
1484 | double |
1485 | CoinOslFactorization::conditionNumber() const |
1486 | { |
1487 | double condition = 1.0; |
1488 | const double *dluval = factInfo_.xeeadr+1-1; // stored before |
1489 | const int *mcstrt = factInfo_.xcsadr+1; |
1490 | for (int i=0;i<numberRows_;i++) { |
1491 | const int kx = mcstrt[i]; |
1492 | const double dpiv = dluval[kx]; |
1493 | condition *= dpiv; |
1494 | } |
1495 | condition = CoinMax(fabs(condition),1.0e-50); |
1496 | return 1.0/condition; |
1497 | } |
1498 | |