Eigen_Colamd.h
Go to the documentation of this file.
1 // // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2012 Desire Nuentsa Wakam <desire.nuentsa_wakam@inria.fr>
5 //
6 // This Source Code Form is subject to the terms of the Mozilla
7 // Public License v. 2.0. If a copy of the MPL was not distributed
8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9 
10 // This file is modified from the colamd/symamd library. The copyright is below
11 
12 // The authors of the code itself are Stefan I. Larimore and Timothy A.
13 // Davis (davis@cise.ufl.edu), University of Florida. The algorithm was
14 // developed in collaboration with John Gilbert, Xerox PARC, and Esmond
15 // Ng, Oak Ridge National Laboratory.
16 //
17 // Date:
18 //
19 // September 8, 2003. Version 2.3.
20 //
21 // Acknowledgements:
22 //
23 // This work was supported by the National Science Foundation, under
24 // grants DMS-9504974 and DMS-9803599.
25 //
26 // Notice:
27 //
28 // Copyright (c) 1998-2003 by the University of Florida.
29 // All Rights Reserved.
30 //
31 // THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY
32 // EXPRESSED OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
33 //
34 // Permission is hereby granted to use, copy, modify, and/or distribute
35 // this program, provided that the Copyright, this License, and the
36 // Availability of the original version is retained on all copies and made
37 // accessible to the end-user of any code or package that includes COLAMD
38 // or any modified version of COLAMD.
39 //
40 // Availability:
41 //
42 // The colamd/symamd library is available at
43 //
44 // http://www.suitesparse.com
45 
46 
47 #ifndef EIGEN_COLAMD_H
48 #define EIGEN_COLAMD_H
49 
50 namespace internal {
51 
52 namespace Colamd {
53 
54 /* Ensure that debugging is turned off: */
55 #ifndef COLAMD_NDEBUG
56 #define COLAMD_NDEBUG
57 #endif /* NDEBUG */
58 
59 
60 /* ========================================================================== */
61 /* === Knob and statistics definitions ====================================== */
62 /* ========================================================================== */
63 
64 /* size of the knobs [ ] array. Only knobs [0..1] are currently used. */
65 const int NKnobs = 20;
66 
67 /* number of output statistics. Only stats [0..6] are currently used. */
68 const int NStats = 20;
69 
70 /* Indices into knobs and stats array. */
72  /* knobs [0] and stats [0]: dense row knob and output statistic. */
73  DenseRow = 0,
74 
75  /* knobs [1] and stats [1]: dense column knob and output statistic. */
76  DenseCol = 1,
77 
78  /* stats [2]: memory defragmentation count output statistic */
80 
81  /* stats [3]: colamd status: zero OK, > 0 warning or notice, < 0 error */
82  Status = 3,
83 
84  /* stats [4..6]: error info, or info on jumbled columns */
85  Info1 = 4,
86  Info2 = 5,
87  Info3 = 6
88 };
89 
90 /* error codes returned in stats [3]: */
91 enum Status {
92  Ok = 0,
105 };
106 /* ========================================================================== */
107 /* === Definitions ========================================================== */
108 /* ========================================================================== */
109 
110 template <typename IndexType>
111 IndexType ones_complement(const IndexType r) {
112  return (-(r)-1);
113 }
114 
115 /* -------------------------------------------------------------------------- */
116 const int Empty = -1;
117 
118 /* Row and column status */
120  Alive = 0,
121  Dead = -1
122 };
123 
124 /* Column status */
128 };
129 
130 /* ========================================================================== */
131 /* === Colamd reporting mechanism =========================================== */
132 /* ========================================================================== */
133 
134 // == Row and Column structures ==
135 template <typename IndexType>
137 {
138  IndexType start ; /* index for A of first row in this column, or Dead */
139  /* if column is dead */
140  IndexType length ; /* number of rows in this column */
141  union
142  {
143  IndexType thickness ; /* number of original columns represented by this */
144  /* col, if the column is alive */
145  IndexType parent ; /* parent in parent tree super-column structure, if */
146  /* the column is dead */
147  } shared1 ;
148  union
149  {
150  IndexType score ; /* the score used to maintain heap, if col is alive */
151  IndexType order ; /* pivot ordering of this column, if col is dead */
152  } shared2 ;
153  union
154  {
155  IndexType headhash ; /* head of a hash bucket, if col is at the head of */
156  /* a degree list */
157  IndexType hash ; /* hash value, if col is not in a degree list */
158  IndexType prev ; /* previous column in degree list, if col is in a */
159  /* degree list (but not at the head of a degree list) */
160  } shared3 ;
161  union
162  {
163  IndexType degree_next ; /* next column, if col is in a degree list */
164  IndexType hash_next ; /* next column, if col is in a hash list */
165  } shared4 ;
166 
167  inline bool is_dead() const { return start < Alive; }
168 
169  inline bool is_alive() const { return start >= Alive; }
170 
171  inline bool is_dead_principal() const { return start == DeadPrincipal; }
172 
173  inline void kill_principal() { start = DeadPrincipal; }
174 
176 
177 };
178 
179 template <typename IndexType>
181 {
182  IndexType start ; /* index for A of first col in this row */
183  IndexType length ; /* number of principal columns in this row */
184  union
185  {
186  IndexType degree ; /* number of principal & non-principal columns in row */
187  IndexType p ; /* used as a row pointer in init_rows_cols () */
188  } shared1 ;
189  union
190  {
191  IndexType mark ; /* for computing set differences and marking dead rows*/
192  IndexType first_column ;/* first column in row (used in garbage collection) */
193  } shared2 ;
194 
195  inline bool is_dead() const { return shared2.mark < Alive; }
196 
197  inline bool is_alive() const { return shared2.mark >= Alive; }
198 
199  inline void kill() { shared2.mark = Dead; }
200 
201 };
202 
203 /* ========================================================================== */
204 /* === Colamd recommended memory size ======================================= */
205 /* ========================================================================== */
206 
207 /*
208  The recommended length Alen of the array A passed to colamd is given by
209  the COLAMD_RECOMMENDED (nnz, n_row, n_col) macro. It returns -1 if any
210  argument is negative. 2*nnz space is required for the row and column
211  indices of the matrix. colamd_c (n_col) + colamd_r (n_row) space is
212  required for the Col and Row arrays, respectively, which are internal to
213  colamd. An additional n_col space is the minimal amount of "elbow room",
214  and nnz/5 more space is recommended for run time efficiency.
215 
216  This macro is not needed when using symamd.
217 
218  Explicit typecast to IndexType added Sept. 23, 2002, COLAMD version 2.2, to avoid
219  gcc -pedantic warning messages.
220 */
221 template <typename IndexType>
222 inline IndexType colamd_c(IndexType n_col)
223 { return IndexType( ((n_col) + 1) * sizeof (ColStructure<IndexType>) / sizeof (IndexType) ) ; }
224 
225 template <typename IndexType>
226 inline IndexType colamd_r(IndexType n_row)
227 { return IndexType(((n_row) + 1) * sizeof (RowStructure<IndexType>) / sizeof (IndexType)); }
228 
229 // Prototypes of non-user callable routines
230 template <typename IndexType>
231 static IndexType init_rows_cols (IndexType n_row, IndexType n_col, RowStructure<IndexType> Row [], ColStructure<IndexType> col [], IndexType A [], IndexType p [], IndexType stats[NStats] );
232 
233 template <typename IndexType>
234 static void init_scoring (IndexType n_row, IndexType n_col, RowStructure<IndexType> Row [], ColStructure<IndexType> Col [], IndexType A [], IndexType head [], double knobs[NKnobs], IndexType *p_n_row2, IndexType *p_n_col2, IndexType *p_max_deg);
235 
236 template <typename IndexType>
237 static IndexType find_ordering (IndexType n_row, IndexType n_col, IndexType Alen, RowStructure<IndexType> Row [], ColStructure<IndexType> Col [], IndexType A [], IndexType head [], IndexType n_col2, IndexType max_deg, IndexType pfree);
238 
239 template <typename IndexType>
240 static void order_children (IndexType n_col, ColStructure<IndexType> Col [], IndexType p []);
241 
242 template <typename IndexType>
243 static void detect_super_cols (ColStructure<IndexType> Col [], IndexType A [], IndexType head [], IndexType row_start, IndexType row_length ) ;
244 
245 template <typename IndexType>
246 static IndexType garbage_collection (IndexType n_row, IndexType n_col, RowStructure<IndexType> Row [], ColStructure<IndexType> Col [], IndexType A [], IndexType *pfree) ;
247 
248 template <typename IndexType>
249 static inline IndexType clear_mark (IndexType n_row, RowStructure<IndexType> Row [] ) ;
250 
251 /* === No debugging ========================================================= */
252 
253 #define COLAMD_DEBUG0(params) ;
254 #define COLAMD_DEBUG1(params) ;
255 #define COLAMD_DEBUG2(params) ;
256 #define COLAMD_DEBUG3(params) ;
257 #define COLAMD_DEBUG4(params) ;
258 
259 #define COLAMD_ASSERT(expression) ((void) 0)
260 
261 
276 template <typename IndexType>
277 inline IndexType recommended ( IndexType nnz, IndexType n_row, IndexType n_col)
278 {
279  if ((nnz) < 0 || (n_row) < 0 || (n_col) < 0)
280  return (-1);
281  else
282  return (2 * (nnz) + colamd_c (n_col) + colamd_r (n_row) + (n_col) + ((nnz) / 5));
283 }
284 
306 static inline void set_defaults(double knobs[NKnobs])
307 {
308  /* === Local variables ================================================== */
309 
310  int i ;
311 
312  if (!knobs)
313  {
314  return ; /* no knobs to initialize */
315  }
316  for (i = 0 ; i < NKnobs ; i++)
317  {
318  knobs [i] = 0 ;
319  }
320  knobs [Colamd::DenseRow] = 0.5 ; /* ignore rows over 50% dense */
321  knobs [Colamd::DenseCol] = 0.5 ; /* ignore columns over 50% dense */
322 }
323 
341 template <typename IndexType>
342 static bool compute_ordering(IndexType n_row, IndexType n_col, IndexType Alen, IndexType *A, IndexType *p, double knobs[NKnobs], IndexType stats[NStats])
343 {
344  /* === Local variables ================================================== */
345 
346  IndexType i ; /* loop index */
347  IndexType nnz ; /* nonzeros in A */
348  IndexType Row_size ; /* size of Row [], in integers */
349  IndexType Col_size ; /* size of Col [], in integers */
350  IndexType need ; /* minimum required length of A */
351  Colamd::RowStructure<IndexType> *Row ; /* pointer into A of Row [0..n_row] array */
352  Colamd::ColStructure<IndexType> *Col ; /* pointer into A of Col [0..n_col] array */
353  IndexType n_col2 ; /* number of non-dense, non-empty columns */
354  IndexType n_row2 ; /* number of non-dense, non-empty rows */
355  IndexType ngarbage ; /* number of garbage collections performed */
356  IndexType max_deg ; /* maximum row degree */
357  double default_knobs [NKnobs] ; /* default knobs array */
358 
359 
360  /* === Check the input arguments ======================================== */
361 
362  if (!stats)
363  {
364  COLAMD_DEBUG0 (("colamd: stats not present\n")) ;
365  return (false) ;
366  }
367  for (i = 0 ; i < NStats ; i++)
368  {
369  stats [i] = 0 ;
370  }
372  stats [Colamd::Info1] = -1 ;
373  stats [Colamd::Info2] = -1 ;
374 
375  if (!A) /* A is not present */
376  {
378  COLAMD_DEBUG0 (("colamd: A not present\n")) ;
379  return (false) ;
380  }
381 
382  if (!p) /* p is not present */
383  {
385  COLAMD_DEBUG0 (("colamd: p not present\n")) ;
386  return (false) ;
387  }
388 
389  if (n_row < 0) /* n_row must be >= 0 */
390  {
392  stats [Colamd::Info1] = n_row ;
393  COLAMD_DEBUG0 (("colamd: nrow negative %d\n", n_row)) ;
394  return (false) ;
395  }
396 
397  if (n_col < 0) /* n_col must be >= 0 */
398  {
400  stats [Colamd::Info1] = n_col ;
401  COLAMD_DEBUG0 (("colamd: ncol negative %d\n", n_col)) ;
402  return (false) ;
403  }
404 
405  nnz = p [n_col] ;
406  if (nnz < 0) /* nnz must be >= 0 */
407  {
409  stats [Colamd::Info1] = nnz ;
410  COLAMD_DEBUG0 (("colamd: number of entries negative %d\n", nnz)) ;
411  return (false) ;
412  }
413 
414  if (p [0] != 0)
415  {
417  stats [Colamd::Info1] = p [0] ;
418  COLAMD_DEBUG0 (("colamd: p[0] not zero %d\n", p [0])) ;
419  return (false) ;
420  }
421 
422  /* === If no knobs, set default knobs =================================== */
423 
424  if (!knobs)
425  {
426  set_defaults (default_knobs) ;
427  knobs = default_knobs ;
428  }
429 
430  /* === Allocate the Row and Col arrays from array A ===================== */
431 
432  Col_size = colamd_c (n_col) ;
433  Row_size = colamd_r (n_row) ;
434  need = 2*nnz + n_col + Col_size + Row_size ;
435 
436  if (need > Alen)
437  {
438  /* not enough space in array A to perform the ordering */
440  stats [Colamd::Info1] = need ;
441  stats [Colamd::Info2] = Alen ;
442  COLAMD_DEBUG0 (("colamd: Need Alen >= %d, given only Alen = %d\n", need,Alen));
443  return (false) ;
444  }
445 
446  Alen -= Col_size + Row_size ;
447  Col = (ColStructure<IndexType> *) &A [Alen] ;
448  Row = (RowStructure<IndexType> *) &A [Alen + Col_size] ;
449 
450  /* === Construct the row and column data structures ===================== */
451 
452  if (!Colamd::init_rows_cols (n_row, n_col, Row, Col, A, p, stats))
453  {
454  /* input matrix is invalid */
455  COLAMD_DEBUG0 (("colamd: Matrix invalid\n")) ;
456  return (false) ;
457  }
458 
459  /* === Initialize scores, kill dense rows/columns ======================= */
460 
461  Colamd::init_scoring (n_row, n_col, Row, Col, A, p, knobs,
462  &n_row2, &n_col2, &max_deg) ;
463 
464  /* === Order the supercolumns =========================================== */
465 
466  ngarbage = Colamd::find_ordering (n_row, n_col, Alen, Row, Col, A, p,
467  n_col2, max_deg, 2*nnz) ;
468 
469  /* === Order the non-principal columns ================================== */
470 
471  Colamd::order_children (n_col, Col, p) ;
472 
473  /* === Return statistics in stats ======================================= */
474 
475  stats [Colamd::DenseRow] = n_row - n_row2 ;
476  stats [Colamd::DenseCol] = n_col - n_col2 ;
477  stats [Colamd::DefragCount] = ngarbage ;
478  COLAMD_DEBUG0 (("colamd: done.\n")) ;
479  return (true) ;
480 }
481 
482 /* ========================================================================== */
483 /* === NON-USER-CALLABLE ROUTINES: ========================================== */
484 /* ========================================================================== */
485 
486 /* There are no user-callable routines beyond this point in the file */
487 
488 /* ========================================================================== */
489 /* === init_rows_cols ======================================================= */
490 /* ========================================================================== */
491 
492 /*
493  Takes the column form of the matrix in A and creates the row form of the
494  matrix. Also, row and column attributes are stored in the Col and Row
495  structs. If the columns are un-sorted or contain duplicate row indices,
496  this routine will also sort and remove duplicate row indices from the
497  column form of the matrix. Returns false if the matrix is invalid,
498  true otherwise. Not user-callable.
499 */
500 template <typename IndexType>
501 static IndexType init_rows_cols /* returns true if OK, or false otherwise */
502  (
503  /* === Parameters ======================================================= */
504 
505  IndexType n_row, /* number of rows of A */
506  IndexType n_col, /* number of columns of A */
507  RowStructure<IndexType> Row [], /* of size n_row+1 */
508  ColStructure<IndexType> Col [], /* of size n_col+1 */
509  IndexType A [], /* row indices of A, of size Alen */
510  IndexType p [], /* pointers to columns in A, of size n_col+1 */
511  IndexType stats [NStats] /* colamd statistics */
512  )
513 {
514  /* === Local variables ================================================== */
515 
516  IndexType col ; /* a column index */
517  IndexType row ; /* a row index */
518  IndexType *cp ; /* a column pointer */
519  IndexType *cp_end ; /* a pointer to the end of a column */
520  IndexType *rp ; /* a row pointer */
521  IndexType *rp_end ; /* a pointer to the end of a row */
522  IndexType last_row ; /* previous row */
523 
524  /* === Initialize columns, and check column pointers ==================== */
525 
526  for (col = 0 ; col < n_col ; col++)
527  {
528  Col [col].start = p [col] ;
529  Col [col].length = p [col+1] - p [col] ;
530 
531  if ((Col [col].length) < 0) // extra parentheses to work-around gcc bug 10200
532  {
533  /* column pointers must be non-decreasing */
535  stats [Colamd::Info1] = col ;
536  stats [Colamd::Info2] = Col [col].length ;
537  COLAMD_DEBUG0 (("colamd: col %d length %d < 0\n", col, Col [col].length)) ;
538  return (false) ;
539  }
540 
541  Col [col].shared1.thickness = 1 ;
542  Col [col].shared2.score = 0 ;
543  Col [col].shared3.prev = Empty ;
544  Col [col].shared4.degree_next = Empty ;
545  }
546 
547  /* p [0..n_col] no longer needed, used as "head" in subsequent routines */
548 
549  /* === Scan columns, compute row degrees, and check row indices ========= */
550 
551  stats [Info3] = 0 ; /* number of duplicate or unsorted row indices*/
552 
553  for (row = 0 ; row < n_row ; row++)
554  {
555  Row [row].length = 0 ;
556  Row [row].shared2.mark = -1 ;
557  }
558 
559  for (col = 0 ; col < n_col ; col++)
560  {
561  last_row = -1 ;
562 
563  cp = &A [p [col]] ;
564  cp_end = &A [p [col+1]] ;
565 
566  while (cp < cp_end)
567  {
568  row = *cp++ ;
569 
570  /* make sure row indices within range */
571  if (row < 0 || row >= n_row)
572  {
574  stats [Colamd::Info1] = col ;
575  stats [Colamd::Info2] = row ;
576  stats [Colamd::Info3] = n_row ;
577  COLAMD_DEBUG0 (("colamd: row %d col %d out of bounds\n", row, col)) ;
578  return (false) ;
579  }
580 
581  if (row <= last_row || Row [row].shared2.mark == col)
582  {
583  /* row index are unsorted or repeated (or both), thus col */
584  /* is jumbled. This is a notice, not an error condition. */
586  stats [Colamd::Info1] = col ;
587  stats [Colamd::Info2] = row ;
588  (stats [Colamd::Info3]) ++ ;
589  COLAMD_DEBUG1 (("colamd: row %d col %d unsorted/duplicate\n",row,col));
590  }
591 
592  if (Row [row].shared2.mark != col)
593  {
594  Row [row].length++ ;
595  }
596  else
597  {
598  /* this is a repeated entry in the column, */
599  /* it will be removed */
600  Col [col].length-- ;
601  }
602 
603  /* mark the row as having been seen in this column */
604  Row [row].shared2.mark = col ;
605 
606  last_row = row ;
607  }
608  }
609 
610  /* === Compute row pointers ============================================= */
611 
612  /* row form of the matrix starts directly after the column */
613  /* form of matrix in A */
614  Row [0].start = p [n_col] ;
615  Row [0].shared1.p = Row [0].start ;
616  Row [0].shared2.mark = -1 ;
617  for (row = 1 ; row < n_row ; row++)
618  {
619  Row [row].start = Row [row-1].start + Row [row-1].length ;
620  Row [row].shared1.p = Row [row].start ;
621  Row [row].shared2.mark = -1 ;
622  }
623 
624  /* === Create row form ================================================== */
625 
626  if (stats [Status] == OkButJumbled)
627  {
628  /* if cols jumbled, watch for repeated row indices */
629  for (col = 0 ; col < n_col ; col++)
630  {
631  cp = &A [p [col]] ;
632  cp_end = &A [p [col+1]] ;
633  while (cp < cp_end)
634  {
635  row = *cp++ ;
636  if (Row [row].shared2.mark != col)
637  {
638  A [(Row [row].shared1.p)++] = col ;
639  Row [row].shared2.mark = col ;
640  }
641  }
642  }
643  }
644  else
645  {
646  /* if cols not jumbled, we don't need the mark (this is faster) */
647  for (col = 0 ; col < n_col ; col++)
648  {
649  cp = &A [p [col]] ;
650  cp_end = &A [p [col+1]] ;
651  while (cp < cp_end)
652  {
653  A [(Row [*cp++].shared1.p)++] = col ;
654  }
655  }
656  }
657 
658  /* === Clear the row marks and set row degrees ========================== */
659 
660  for (row = 0 ; row < n_row ; row++)
661  {
662  Row [row].shared2.mark = 0 ;
663  Row [row].shared1.degree = Row [row].length ;
664  }
665 
666  /* === See if we need to re-create columns ============================== */
667 
668  if (stats [Status] == OkButJumbled)
669  {
670  COLAMD_DEBUG0 (("colamd: reconstructing column form, matrix jumbled\n")) ;
671 
672 
673  /* === Compute col pointers ========================================= */
674 
675  /* col form of the matrix starts at A [0]. */
676  /* Note, we may have a gap between the col form and the row */
677  /* form if there were duplicate entries, if so, it will be */
678  /* removed upon the first garbage collection */
679  Col [0].start = 0 ;
680  p [0] = Col [0].start ;
681  for (col = 1 ; col < n_col ; col++)
682  {
683  /* note that the lengths here are for pruned columns, i.e. */
684  /* no duplicate row indices will exist for these columns */
685  Col [col].start = Col [col-1].start + Col [col-1].length ;
686  p [col] = Col [col].start ;
687  }
688 
689  /* === Re-create col form =========================================== */
690 
691  for (row = 0 ; row < n_row ; row++)
692  {
693  rp = &A [Row [row].start] ;
694  rp_end = rp + Row [row].length ;
695  while (rp < rp_end)
696  {
697  A [(p [*rp++])++] = row ;
698  }
699  }
700  }
701 
702  /* === Done. Matrix is not (or no longer) jumbled ====================== */
703 
704  return (true) ;
705 }
706 
707 
708 /* ========================================================================== */
709 /* === init_scoring ========================================================= */
710 /* ========================================================================== */
711 
712 /*
713  Kills dense or empty columns and rows, calculates an initial score for
714  each column, and places all columns in the degree lists. Not user-callable.
715 */
716 template <typename IndexType>
717 static void init_scoring
718  (
719  /* === Parameters ======================================================= */
720 
721  IndexType n_row, /* number of rows of A */
722  IndexType n_col, /* number of columns of A */
723  RowStructure<IndexType> Row [], /* of size n_row+1 */
724  ColStructure<IndexType> Col [], /* of size n_col+1 */
725  IndexType A [], /* column form and row form of A */
726  IndexType head [], /* of size n_col+1 */
727  double knobs [NKnobs],/* parameters */
728  IndexType *p_n_row2, /* number of non-dense, non-empty rows */
729  IndexType *p_n_col2, /* number of non-dense, non-empty columns */
730  IndexType *p_max_deg /* maximum row degree */
731  )
732 {
733  /* === Local variables ================================================== */
734 
735  IndexType c ; /* a column index */
736  IndexType r, row ; /* a row index */
737  IndexType *cp ; /* a column pointer */
738  IndexType deg ; /* degree of a row or column */
739  IndexType *cp_end ; /* a pointer to the end of a column */
740  IndexType *new_cp ; /* new column pointer */
741  IndexType col_length ; /* length of pruned column */
742  IndexType score ; /* current column score */
743  IndexType n_col2 ; /* number of non-dense, non-empty columns */
744  IndexType n_row2 ; /* number of non-dense, non-empty rows */
745  IndexType dense_row_count ; /* remove rows with more entries than this */
746  IndexType dense_col_count ; /* remove cols with more entries than this */
747  IndexType min_score ; /* smallest column score */
748  IndexType max_deg ; /* maximum row degree */
749  IndexType next_col ; /* Used to add to degree list.*/
750 
751 
752  /* === Extract knobs ==================================================== */
753 
754  dense_row_count = numext::maxi(IndexType(0), numext::mini(IndexType(knobs [Colamd::DenseRow] * n_col), n_col)) ;
755  dense_col_count = numext::maxi(IndexType(0), numext::mini(IndexType(knobs [Colamd::DenseCol] * n_row), n_row)) ;
756  COLAMD_DEBUG1 (("colamd: densecount: %d %d\n", dense_row_count, dense_col_count)) ;
757  max_deg = 0 ;
758  n_col2 = n_col ;
759  n_row2 = n_row ;
760 
761  /* === Kill empty columns =============================================== */
762 
763  /* Put the empty columns at the end in their natural order, so that LU */
764  /* factorization can proceed as far as possible. */
765  for (c = n_col-1 ; c >= 0 ; c--)
766  {
767  deg = Col [c].length ;
768  if (deg == 0)
769  {
770  /* this is a empty column, kill and order it last */
771  Col [c].shared2.order = --n_col2 ;
772  Col[c].kill_principal() ;
773  }
774  }
775  COLAMD_DEBUG1 (("colamd: null columns killed: %d\n", n_col - n_col2)) ;
776 
777  /* === Kill dense columns =============================================== */
778 
779  /* Put the dense columns at the end, in their natural order */
780  for (c = n_col-1 ; c >= 0 ; c--)
781  {
782  /* skip any dead columns */
783  if (Col[c].is_dead())
784  {
785  continue ;
786  }
787  deg = Col [c].length ;
788  if (deg > dense_col_count)
789  {
790  /* this is a dense column, kill and order it last */
791  Col [c].shared2.order = --n_col2 ;
792  /* decrement the row degrees */
793  cp = &A [Col [c].start] ;
794  cp_end = cp + Col [c].length ;
795  while (cp < cp_end)
796  {
797  Row [*cp++].shared1.degree-- ;
798  }
799  Col[c].kill_principal() ;
800  }
801  }
802  COLAMD_DEBUG1 (("colamd: Dense and null columns killed: %d\n", n_col - n_col2)) ;
803 
804  /* === Kill dense and empty rows ======================================== */
805 
806  for (r = 0 ; r < n_row ; r++)
807  {
808  deg = Row [r].shared1.degree ;
809  COLAMD_ASSERT (deg >= 0 && deg <= n_col) ;
810  if (deg > dense_row_count || deg == 0)
811  {
812  /* kill a dense or empty row */
813  Row[r].kill() ;
814  --n_row2 ;
815  }
816  else
817  {
818  /* keep track of max degree of remaining rows */
819  max_deg = numext::maxi(max_deg, deg) ;
820  }
821  }
822  COLAMD_DEBUG1 (("colamd: Dense and null rows killed: %d\n", n_row - n_row2)) ;
823 
824  /* === Compute initial column scores ==================================== */
825 
826  /* At this point the row degrees are accurate. They reflect the number */
827  /* of "live" (non-dense) columns in each row. No empty rows exist. */
828  /* Some "live" columns may contain only dead rows, however. These are */
829  /* pruned in the code below. */
830 
831  /* now find the initial matlab score for each column */
832  for (c = n_col-1 ; c >= 0 ; c--)
833  {
834  /* skip dead column */
835  if (Col[c].is_dead())
836  {
837  continue ;
838  }
839  score = 0 ;
840  cp = &A [Col [c].start] ;
841  new_cp = cp ;
842  cp_end = cp + Col [c].length ;
843  while (cp < cp_end)
844  {
845  /* get a row */
846  row = *cp++ ;
847  /* skip if dead */
848  if (Row[row].is_dead())
849  {
850  continue ;
851  }
852  /* compact the column */
853  *new_cp++ = row ;
854  /* add row's external degree */
855  score += Row [row].shared1.degree - 1 ;
856  /* guard against integer overflow */
857  score = numext::mini(score, n_col) ;
858  }
859  /* determine pruned column length */
860  col_length = (IndexType) (new_cp - &A [Col [c].start]) ;
861  if (col_length == 0)
862  {
863  /* a newly-made null column (all rows in this col are "dense" */
864  /* and have already been killed) */
865  COLAMD_DEBUG2 (("Newly null killed: %d\n", c)) ;
866  Col [c].shared2.order = --n_col2 ;
867  Col[c].kill_principal() ;
868  }
869  else
870  {
871  /* set column length and set score */
872  COLAMD_ASSERT (score >= 0) ;
873  COLAMD_ASSERT (score <= n_col) ;
874  Col [c].length = col_length ;
875  Col [c].shared2.score = score ;
876  }
877  }
878  COLAMD_DEBUG1 (("colamd: Dense, null, and newly-null columns killed: %d\n",
879  n_col-n_col2)) ;
880 
881  /* At this point, all empty rows and columns are dead. All live columns */
882  /* are "clean" (containing no dead rows) and simplicial (no supercolumns */
883  /* yet). Rows may contain dead columns, but all live rows contain at */
884  /* least one live column. */
885 
886  /* === Initialize degree lists ========================================== */
887 
888 
889  /* clear the hash buckets */
890  for (c = 0 ; c <= n_col ; c++)
891  {
892  head [c] = Empty ;
893  }
894  min_score = n_col ;
895  /* place in reverse order, so low column indices are at the front */
896  /* of the lists. This is to encourage natural tie-breaking */
897  for (c = n_col-1 ; c >= 0 ; c--)
898  {
899  /* only add principal columns to degree lists */
900  if (Col[c].is_alive())
901  {
902  COLAMD_DEBUG4 (("place %d score %d minscore %d ncol %d\n",
903  c, Col [c].shared2.score, min_score, n_col)) ;
904 
905  /* === Add columns score to DList =============================== */
906 
907  score = Col [c].shared2.score ;
908 
909  COLAMD_ASSERT (min_score >= 0) ;
910  COLAMD_ASSERT (min_score <= n_col) ;
911  COLAMD_ASSERT (score >= 0) ;
912  COLAMD_ASSERT (score <= n_col) ;
913  COLAMD_ASSERT (head [score] >= Empty) ;
914 
915  /* now add this column to dList at proper score location */
916  next_col = head [score] ;
917  Col [c].shared3.prev = Empty ;
918  Col [c].shared4.degree_next = next_col ;
919 
920  /* if there already was a column with the same score, set its */
921  /* previous pointer to this new column */
922  if (next_col != Empty)
923  {
924  Col [next_col].shared3.prev = c ;
925  }
926  head [score] = c ;
927 
928  /* see if this score is less than current min */
929  min_score = numext::mini(min_score, score) ;
930 
931 
932  }
933  }
934 
935 
936  /* === Return number of remaining columns, and max row degree =========== */
937 
938  *p_n_col2 = n_col2 ;
939  *p_n_row2 = n_row2 ;
940  *p_max_deg = max_deg ;
941 }
942 
943 
944 /* ========================================================================== */
945 /* === find_ordering ======================================================== */
946 /* ========================================================================== */
947 
948 /*
949  Order the principal columns of the supercolumn form of the matrix
950  (no supercolumns on input). Uses a minimum approximate column minimum
951  degree ordering method. Not user-callable.
952 */
953 template <typename IndexType>
954 static IndexType find_ordering /* return the number of garbage collections */
955  (
956  /* === Parameters ======================================================= */
957 
958  IndexType n_row, /* number of rows of A */
959  IndexType n_col, /* number of columns of A */
960  IndexType Alen, /* size of A, 2*nnz + n_col or larger */
961  RowStructure<IndexType> Row [], /* of size n_row+1 */
962  ColStructure<IndexType> Col [], /* of size n_col+1 */
963  IndexType A [], /* column form and row form of A */
964  IndexType head [], /* of size n_col+1 */
965  IndexType n_col2, /* Remaining columns to order */
966  IndexType max_deg, /* Maximum row degree */
967  IndexType pfree /* index of first free slot (2*nnz on entry) */
968  )
969 {
970  /* === Local variables ================================================== */
971 
972  IndexType k ; /* current pivot ordering step */
973  IndexType pivot_col ; /* current pivot column */
974  IndexType *cp ; /* a column pointer */
975  IndexType *rp ; /* a row pointer */
976  IndexType pivot_row ; /* current pivot row */
977  IndexType *new_cp ; /* modified column pointer */
978  IndexType *new_rp ; /* modified row pointer */
979  IndexType pivot_row_start ; /* pointer to start of pivot row */
980  IndexType pivot_row_degree ; /* number of columns in pivot row */
981  IndexType pivot_row_length ; /* number of supercolumns in pivot row */
982  IndexType pivot_col_score ; /* score of pivot column */
983  IndexType needed_memory ; /* free space needed for pivot row */
984  IndexType *cp_end ; /* pointer to the end of a column */
985  IndexType *rp_end ; /* pointer to the end of a row */
986  IndexType row ; /* a row index */
987  IndexType col ; /* a column index */
988  IndexType max_score ; /* maximum possible score */
989  IndexType cur_score ; /* score of current column */
990  unsigned int hash ; /* hash value for supernode detection */
991  IndexType head_column ; /* head of hash bucket */
992  IndexType first_col ; /* first column in hash bucket */
993  IndexType tag_mark ; /* marker value for mark array */
994  IndexType row_mark ; /* Row [row].shared2.mark */
995  IndexType set_difference ; /* set difference size of row with pivot row */
996  IndexType min_score ; /* smallest column score */
997  IndexType col_thickness ; /* "thickness" (no. of columns in a supercol) */
998  IndexType max_mark ; /* maximum value of tag_mark */
999  IndexType pivot_col_thickness ; /* number of columns represented by pivot col */
1000  IndexType prev_col ; /* Used by Dlist operations. */
1001  IndexType next_col ; /* Used by Dlist operations. */
1002  IndexType ngarbage ; /* number of garbage collections performed */
1003 
1004 
1005  /* === Initialization and clear mark ==================================== */
1006 
1007  max_mark = INT_MAX - n_col ; /* INT_MAX defined in <limits.h> */
1008  tag_mark = Colamd::clear_mark (n_row, Row) ;
1009  min_score = 0 ;
1010  ngarbage = 0 ;
1011  COLAMD_DEBUG1 (("colamd: Ordering, n_col2=%d\n", n_col2)) ;
1012 
1013  /* === Order the columns ================================================ */
1014 
1015  for (k = 0 ; k < n_col2 ; /* 'k' is incremented below */)
1016  {
1017 
1018  /* === Select pivot column, and order it ============================ */
1019 
1020  /* make sure degree list isn't empty */
1021  COLAMD_ASSERT (min_score >= 0) ;
1022  COLAMD_ASSERT (min_score <= n_col) ;
1023  COLAMD_ASSERT (head [min_score] >= Empty) ;
1024 
1025  /* get pivot column from head of minimum degree list */
1026  while (min_score < n_col && head [min_score] == Empty)
1027  {
1028  min_score++ ;
1029  }
1030  pivot_col = head [min_score] ;
1031  COLAMD_ASSERT (pivot_col >= 0 && pivot_col <= n_col) ;
1032  next_col = Col [pivot_col].shared4.degree_next ;
1033  head [min_score] = next_col ;
1034  if (next_col != Empty)
1035  {
1036  Col [next_col].shared3.prev = Empty ;
1037  }
1038 
1039  COLAMD_ASSERT (Col[pivot_col].is_alive()) ;
1040  COLAMD_DEBUG3 (("Pivot col: %d\n", pivot_col)) ;
1041 
1042  /* remember score for defrag check */
1043  pivot_col_score = Col [pivot_col].shared2.score ;
1044 
1045  /* the pivot column is the kth column in the pivot order */
1046  Col [pivot_col].shared2.order = k ;
1047 
1048  /* increment order count by column thickness */
1049  pivot_col_thickness = Col [pivot_col].shared1.thickness ;
1050  k += pivot_col_thickness ;
1051  COLAMD_ASSERT (pivot_col_thickness > 0) ;
1052 
1053  /* === Garbage_collection, if necessary ============================= */
1054 
1055  needed_memory = numext::mini(pivot_col_score, n_col - k) ;
1056  if (pfree + needed_memory >= Alen)
1057  {
1058  pfree = Colamd::garbage_collection (n_row, n_col, Row, Col, A, &A [pfree]) ;
1059  ngarbage++ ;
1060  /* after garbage collection we will have enough */
1061  COLAMD_ASSERT (pfree + needed_memory < Alen) ;
1062  /* garbage collection has wiped out the Row[].shared2.mark array */
1063  tag_mark = Colamd::clear_mark (n_row, Row) ;
1064 
1065  }
1066 
1067  /* === Compute pivot row pattern ==================================== */
1068 
1069  /* get starting location for this new merged row */
1070  pivot_row_start = pfree ;
1071 
1072  /* initialize new row counts to zero */
1073  pivot_row_degree = 0 ;
1074 
1075  /* tag pivot column as having been visited so it isn't included */
1076  /* in merged pivot row */
1077  Col [pivot_col].shared1.thickness = -pivot_col_thickness ;
1078 
1079  /* pivot row is the union of all rows in the pivot column pattern */
1080  cp = &A [Col [pivot_col].start] ;
1081  cp_end = cp + Col [pivot_col].length ;
1082  while (cp < cp_end)
1083  {
1084  /* get a row */
1085  row = *cp++ ;
1086  COLAMD_DEBUG4 (("Pivot col pattern %d %d\n", Row[row].is_alive(), row)) ;
1087  /* skip if row is dead */
1088  if (Row[row].is_dead())
1089  {
1090  continue ;
1091  }
1092  rp = &A [Row [row].start] ;
1093  rp_end = rp + Row [row].length ;
1094  while (rp < rp_end)
1095  {
1096  /* get a column */
1097  col = *rp++ ;
1098  /* add the column, if alive and untagged */
1099  col_thickness = Col [col].shared1.thickness ;
1100  if (col_thickness > 0 && Col[col].is_alive())
1101  {
1102  /* tag column in pivot row */
1103  Col [col].shared1.thickness = -col_thickness ;
1104  COLAMD_ASSERT (pfree < Alen) ;
1105  /* place column in pivot row */
1106  A [pfree++] = col ;
1107  pivot_row_degree += col_thickness ;
1108  }
1109  }
1110  }
1111 
1112  /* clear tag on pivot column */
1113  Col [pivot_col].shared1.thickness = pivot_col_thickness ;
1114  max_deg = numext::maxi(max_deg, pivot_row_degree) ;
1115 
1116 
1117  /* === Kill all rows used to construct pivot row ==================== */
1118 
1119  /* also kill pivot row, temporarily */
1120  cp = &A [Col [pivot_col].start] ;
1121  cp_end = cp + Col [pivot_col].length ;
1122  while (cp < cp_end)
1123  {
1124  /* may be killing an already dead row */
1125  row = *cp++ ;
1126  COLAMD_DEBUG3 (("Kill row in pivot col: %d\n", row)) ;
1127  Row[row].kill() ;
1128  }
1129 
1130  /* === Select a row index to use as the new pivot row =============== */
1131 
1132  pivot_row_length = pfree - pivot_row_start ;
1133  if (pivot_row_length > 0)
1134  {
1135  /* pick the "pivot" row arbitrarily (first row in col) */
1136  pivot_row = A [Col [pivot_col].start] ;
1137  COLAMD_DEBUG3 (("Pivotal row is %d\n", pivot_row)) ;
1138  }
1139  else
1140  {
1141  /* there is no pivot row, since it is of zero length */
1142  pivot_row = Empty ;
1143  COLAMD_ASSERT (pivot_row_length == 0) ;
1144  }
1145  COLAMD_ASSERT (Col [pivot_col].length > 0 || pivot_row_length == 0) ;
1146 
1147  /* === Approximate degree computation =============================== */
1148 
1149  /* Here begins the computation of the approximate degree. The column */
1150  /* score is the sum of the pivot row "length", plus the size of the */
1151  /* set differences of each row in the column minus the pattern of the */
1152  /* pivot row itself. The column ("thickness") itself is also */
1153  /* excluded from the column score (we thus use an approximate */
1154  /* external degree). */
1155 
1156  /* The time taken by the following code (compute set differences, and */
1157  /* add them up) is proportional to the size of the data structure */
1158  /* being scanned - that is, the sum of the sizes of each column in */
1159  /* the pivot row. Thus, the amortized time to compute a column score */
1160  /* is proportional to the size of that column (where size, in this */
1161  /* context, is the column "length", or the number of row indices */
1162  /* in that column). The number of row indices in a column is */
1163  /* monotonically non-decreasing, from the length of the original */
1164  /* column on input to colamd. */
1165 
1166  /* === Compute set differences ====================================== */
1167 
1168  COLAMD_DEBUG3 (("** Computing set differences phase. **\n")) ;
1169 
1170  /* pivot row is currently dead - it will be revived later. */
1171 
1172  COLAMD_DEBUG3 (("Pivot row: ")) ;
1173  /* for each column in pivot row */
1174  rp = &A [pivot_row_start] ;
1175  rp_end = rp + pivot_row_length ;
1176  while (rp < rp_end)
1177  {
1178  col = *rp++ ;
1179  COLAMD_ASSERT (Col[col].is_alive() && col != pivot_col) ;
1180  COLAMD_DEBUG3 (("Col: %d\n", col)) ;
1181 
1182  /* clear tags used to construct pivot row pattern */
1183  col_thickness = -Col [col].shared1.thickness ;
1184  COLAMD_ASSERT (col_thickness > 0) ;
1185  Col [col].shared1.thickness = col_thickness ;
1186 
1187  /* === Remove column from degree list =========================== */
1188 
1189  cur_score = Col [col].shared2.score ;
1190  prev_col = Col [col].shared3.prev ;
1191  next_col = Col [col].shared4.degree_next ;
1192  COLAMD_ASSERT (cur_score >= 0) ;
1193  COLAMD_ASSERT (cur_score <= n_col) ;
1194  COLAMD_ASSERT (cur_score >= Empty) ;
1195  if (prev_col == Empty)
1196  {
1197  head [cur_score] = next_col ;
1198  }
1199  else
1200  {
1201  Col [prev_col].shared4.degree_next = next_col ;
1202  }
1203  if (next_col != Empty)
1204  {
1205  Col [next_col].shared3.prev = prev_col ;
1206  }
1207 
1208  /* === Scan the column ========================================== */
1209 
1210  cp = &A [Col [col].start] ;
1211  cp_end = cp + Col [col].length ;
1212  while (cp < cp_end)
1213  {
1214  /* get a row */
1215  row = *cp++ ;
1216  /* skip if dead */
1217  if (Row[row].is_dead())
1218  {
1219  continue ;
1220  }
1221  row_mark = Row [row].shared2.mark ;
1222  COLAMD_ASSERT (row != pivot_row) ;
1223  set_difference = row_mark - tag_mark ;
1224  /* check if the row has been seen yet */
1225  if (set_difference < 0)
1226  {
1227  COLAMD_ASSERT (Row [row].shared1.degree <= max_deg) ;
1228  set_difference = Row [row].shared1.degree ;
1229  }
1230  /* subtract column thickness from this row's set difference */
1231  set_difference -= col_thickness ;
1232  COLAMD_ASSERT (set_difference >= 0) ;
1233  /* absorb this row if the set difference becomes zero */
1234  if (set_difference == 0)
1235  {
1236  COLAMD_DEBUG3 (("aggressive absorption. Row: %d\n", row)) ;
1237  Row[row].kill() ;
1238  }
1239  else
1240  {
1241  /* save the new mark */
1242  Row [row].shared2.mark = set_difference + tag_mark ;
1243  }
1244  }
1245  }
1246 
1247 
1248  /* === Add up set differences for each column ======================= */
1249 
1250  COLAMD_DEBUG3 (("** Adding set differences phase. **\n")) ;
1251 
1252  /* for each column in pivot row */
1253  rp = &A [pivot_row_start] ;
1254  rp_end = rp + pivot_row_length ;
1255  while (rp < rp_end)
1256  {
1257  /* get a column */
1258  col = *rp++ ;
1259  COLAMD_ASSERT (Col[col].is_alive() && col != pivot_col) ;
1260  hash = 0 ;
1261  cur_score = 0 ;
1262  cp = &A [Col [col].start] ;
1263  /* compact the column */
1264  new_cp = cp ;
1265  cp_end = cp + Col [col].length ;
1266 
1267  COLAMD_DEBUG4 (("Adding set diffs for Col: %d.\n", col)) ;
1268 
1269  while (cp < cp_end)
1270  {
1271  /* get a row */
1272  row = *cp++ ;
1273  COLAMD_ASSERT(row >= 0 && row < n_row) ;
1274  /* skip if dead */
1275  if (Row [row].is_dead())
1276  {
1277  continue ;
1278  }
1279  row_mark = Row [row].shared2.mark ;
1280  COLAMD_ASSERT (row_mark > tag_mark) ;
1281  /* compact the column */
1282  *new_cp++ = row ;
1283  /* compute hash function */
1284  hash += row ;
1285  /* add set difference */
1286  cur_score += row_mark - tag_mark ;
1287  /* integer overflow... */
1288  cur_score = numext::mini(cur_score, n_col) ;
1289  }
1290 
1291  /* recompute the column's length */
1292  Col [col].length = (IndexType) (new_cp - &A [Col [col].start]) ;
1293 
1294  /* === Further mass elimination ================================= */
1295 
1296  if (Col [col].length == 0)
1297  {
1298  COLAMD_DEBUG4 (("further mass elimination. Col: %d\n", col)) ;
1299  /* nothing left but the pivot row in this column */
1300  Col[col].kill_principal() ;
1301  pivot_row_degree -= Col [col].shared1.thickness ;
1302  COLAMD_ASSERT (pivot_row_degree >= 0) ;
1303  /* order it */
1304  Col [col].shared2.order = k ;
1305  /* increment order count by column thickness */
1306  k += Col [col].shared1.thickness ;
1307  }
1308  else
1309  {
1310  /* === Prepare for supercolumn detection ==================== */
1311 
1312  COLAMD_DEBUG4 (("Preparing supercol detection for Col: %d.\n", col)) ;
1313 
1314  /* save score so far */
1315  Col [col].shared2.score = cur_score ;
1316 
1317  /* add column to hash table, for supercolumn detection */
1318  hash %= n_col + 1 ;
1319 
1320  COLAMD_DEBUG4 ((" Hash = %d, n_col = %d.\n", hash, n_col)) ;
1321  COLAMD_ASSERT (hash <= n_col) ;
1322 
1323  head_column = head [hash] ;
1324  if (head_column > Empty)
1325  {
1326  /* degree list "hash" is non-empty, use prev (shared3) of */
1327  /* first column in degree list as head of hash bucket */
1328  first_col = Col [head_column].shared3.headhash ;
1329  Col [head_column].shared3.headhash = col ;
1330  }
1331  else
1332  {
1333  /* degree list "hash" is empty, use head as hash bucket */
1334  first_col = - (head_column + 2) ;
1335  head [hash] = - (col + 2) ;
1336  }
1337  Col [col].shared4.hash_next = first_col ;
1338 
1339  /* save hash function in Col [col].shared3.hash */
1340  Col [col].shared3.hash = (IndexType) hash ;
1341  COLAMD_ASSERT (Col[col].is_alive()) ;
1342  }
1343  }
1344 
1345  /* The approximate external column degree is now computed. */
1346 
1347  /* === Supercolumn detection ======================================== */
1348 
1349  COLAMD_DEBUG3 (("** Supercolumn detection phase. **\n")) ;
1350 
1351  Colamd::detect_super_cols (Col, A, head, pivot_row_start, pivot_row_length) ;
1352 
1353  /* === Kill the pivotal column ====================================== */
1354 
1355  Col[pivot_col].kill_principal() ;
1356 
1357  /* === Clear mark =================================================== */
1358 
1359  tag_mark += (max_deg + 1) ;
1360  if (tag_mark >= max_mark)
1361  {
1362  COLAMD_DEBUG2 (("clearing tag_mark\n")) ;
1363  tag_mark = Colamd::clear_mark (n_row, Row) ;
1364  }
1365 
1366  /* === Finalize the new pivot row, and column scores ================ */
1367 
1368  COLAMD_DEBUG3 (("** Finalize scores phase. **\n")) ;
1369 
1370  /* for each column in pivot row */
1371  rp = &A [pivot_row_start] ;
1372  /* compact the pivot row */
1373  new_rp = rp ;
1374  rp_end = rp + pivot_row_length ;
1375  while (rp < rp_end)
1376  {
1377  col = *rp++ ;
1378  /* skip dead columns */
1379  if (Col[col].is_dead())
1380  {
1381  continue ;
1382  }
1383  *new_rp++ = col ;
1384  /* add new pivot row to column */
1385  A [Col [col].start + (Col [col].length++)] = pivot_row ;
1386 
1387  /* retrieve score so far and add on pivot row's degree. */
1388  /* (we wait until here for this in case the pivot */
1389  /* row's degree was reduced due to mass elimination). */
1390  cur_score = Col [col].shared2.score + pivot_row_degree ;
1391 
1392  /* calculate the max possible score as the number of */
1393  /* external columns minus the 'k' value minus the */
1394  /* columns thickness */
1395  max_score = n_col - k - Col [col].shared1.thickness ;
1396 
1397  /* make the score the external degree of the union-of-rows */
1398  cur_score -= Col [col].shared1.thickness ;
1399 
1400  /* make sure score is less or equal than the max score */
1401  cur_score = numext::mini(cur_score, max_score) ;
1402  COLAMD_ASSERT (cur_score >= 0) ;
1403 
1404  /* store updated score */
1405  Col [col].shared2.score = cur_score ;
1406 
1407  /* === Place column back in degree list ========================= */
1408 
1409  COLAMD_ASSERT (min_score >= 0) ;
1410  COLAMD_ASSERT (min_score <= n_col) ;
1411  COLAMD_ASSERT (cur_score >= 0) ;
1412  COLAMD_ASSERT (cur_score <= n_col) ;
1413  COLAMD_ASSERT (head [cur_score] >= Empty) ;
1414  next_col = head [cur_score] ;
1415  Col [col].shared4.degree_next = next_col ;
1416  Col [col].shared3.prev = Empty ;
1417  if (next_col != Empty)
1418  {
1419  Col [next_col].shared3.prev = col ;
1420  }
1421  head [cur_score] = col ;
1422 
1423  /* see if this score is less than current min */
1424  min_score = numext::mini(min_score, cur_score) ;
1425 
1426  }
1427 
1428  /* === Resurrect the new pivot row ================================== */
1429 
1430  if (pivot_row_degree > 0)
1431  {
1432  /* update pivot row length to reflect any cols that were killed */
1433  /* during super-col detection and mass elimination */
1434  Row [pivot_row].start = pivot_row_start ;
1435  Row [pivot_row].length = (IndexType) (new_rp - &A[pivot_row_start]) ;
1436  Row [pivot_row].shared1.degree = pivot_row_degree ;
1437  Row [pivot_row].shared2.mark = 0 ;
1438  /* pivot row is no longer dead */
1439  }
1440  }
1441 
1442  /* === All principal columns have now been ordered ====================== */
1443 
1444  return (ngarbage) ;
1445 }
1446 
1447 
1448 /* ========================================================================== */
1449 /* === order_children ======================================================= */
1450 /* ========================================================================== */
1451 
1452 /*
1453  The find_ordering routine has ordered all of the principal columns (the
1454  representatives of the supercolumns). The non-principal columns have not
1455  yet been ordered. This routine orders those columns by walking up the
1456  parent tree (a column is a child of the column which absorbed it). The
1457  final permutation vector is then placed in p [0 ... n_col-1], with p [0]
1458  being the first column, and p [n_col-1] being the last. It doesn't look
1459  like it at first glance, but be assured that this routine takes time linear
1460  in the number of columns. Although not immediately obvious, the time
1461  taken by this routine is O (n_col), that is, linear in the number of
1462  columns. Not user-callable.
1463 */
1464 template <typename IndexType>
1465 static inline void order_children
1467  /* === Parameters ======================================================= */
1468 
1469  IndexType n_col, /* number of columns of A */
1470  ColStructure<IndexType> Col [], /* of size n_col+1 */
1471  IndexType p [] /* p [0 ... n_col-1] is the column permutation*/
1472  )
1473 {
1474  /* === Local variables ================================================== */
1475 
1476  IndexType i ; /* loop counter for all columns */
1477  IndexType c ; /* column index */
1478  IndexType parent ; /* index of column's parent */
1479  IndexType order ; /* column's order */
1480 
1481  /* === Order each non-principal column ================================== */
1482 
1483  for (i = 0 ; i < n_col ; i++)
1484  {
1485  /* find an un-ordered non-principal column */
1486  COLAMD_ASSERT (col_is_dead(Col, i)) ;
1487  if (!Col[i].is_dead_principal() && Col [i].shared2.order == Empty)
1488  {
1489  parent = i ;
1490  /* once found, find its principal parent */
1491  do
1492  {
1493  parent = Col [parent].shared1.parent ;
1494  } while (!Col[parent].is_dead_principal()) ;
1495 
1496  /* now, order all un-ordered non-principal columns along path */
1497  /* to this parent. collapse tree at the same time */
1498  c = i ;
1499  /* get order of parent */
1500  order = Col [parent].shared2.order ;
1501 
1502  do
1503  {
1504  COLAMD_ASSERT (Col [c].shared2.order == Empty) ;
1505 
1506  /* order this column */
1507  Col [c].shared2.order = order++ ;
1508  /* collaps tree */
1509  Col [c].shared1.parent = parent ;
1510 
1511  /* get immediate parent of this column */
1512  c = Col [c].shared1.parent ;
1513 
1514  /* continue until we hit an ordered column. There are */
1515  /* guaranteed not to be anymore unordered columns */
1516  /* above an ordered column */
1517  } while (Col [c].shared2.order == Empty) ;
1518 
1519  /* re-order the super_col parent to largest order for this group */
1520  Col [parent].shared2.order = order ;
1521  }
1522  }
1523 
1524  /* === Generate the permutation ========================================= */
1525 
1526  for (c = 0 ; c < n_col ; c++)
1527  {
1528  p [Col [c].shared2.order] = c ;
1529  }
1530 }
1531 
1532 
1533 /* ========================================================================== */
1534 /* === detect_super_cols ==================================================== */
1535 /* ========================================================================== */
1536 
1537 /*
1538  Detects supercolumns by finding matches between columns in the hash buckets.
1539  Check amongst columns in the set A [row_start ... row_start + row_length-1].
1540  The columns under consideration are currently *not* in the degree lists,
1541  and have already been placed in the hash buckets.
1542 
1543  The hash bucket for columns whose hash function is equal to h is stored
1544  as follows:
1545 
1546  if head [h] is >= 0, then head [h] contains a degree list, so:
1547 
1548  head [h] is the first column in degree bucket h.
1549  Col [head [h]].headhash gives the first column in hash bucket h.
1550 
1551  otherwise, the degree list is empty, and:
1552 
1553  -(head [h] + 2) is the first column in hash bucket h.
1554 
1555  For a column c in a hash bucket, Col [c].shared3.prev is NOT a "previous
1556  column" pointer. Col [c].shared3.hash is used instead as the hash number
1557  for that column. The value of Col [c].shared4.hash_next is the next column
1558  in the same hash bucket.
1559 
1560  Assuming no, or "few" hash collisions, the time taken by this routine is
1561  linear in the sum of the sizes (lengths) of each column whose score has
1562  just been computed in the approximate degree computation.
1563  Not user-callable.
1564 */
1565 template <typename IndexType>
1566 static void detect_super_cols
1568  /* === Parameters ======================================================= */
1569 
1570  ColStructure<IndexType> Col [], /* of size n_col+1 */
1571  IndexType A [], /* row indices of A */
1572  IndexType head [], /* head of degree lists and hash buckets */
1573  IndexType row_start, /* pointer to set of columns to check */
1574  IndexType row_length /* number of columns to check */
1575 )
1576 {
1577  /* === Local variables ================================================== */
1578 
1579  IndexType hash ; /* hash value for a column */
1580  IndexType *rp ; /* pointer to a row */
1581  IndexType c ; /* a column index */
1582  IndexType super_c ; /* column index of the column to absorb into */
1583  IndexType *cp1 ; /* column pointer for column super_c */
1584  IndexType *cp2 ; /* column pointer for column c */
1585  IndexType length ; /* length of column super_c */
1586  IndexType prev_c ; /* column preceding c in hash bucket */
1587  IndexType i ; /* loop counter */
1588  IndexType *rp_end ; /* pointer to the end of the row */
1589  IndexType col ; /* a column index in the row to check */
1590  IndexType head_column ; /* first column in hash bucket or degree list */
1591  IndexType first_col ; /* first column in hash bucket */
1592 
1593  /* === Consider each column in the row ================================== */
1594 
1595  rp = &A [row_start] ;
1596  rp_end = rp + row_length ;
1597  while (rp < rp_end)
1598  {
1599  col = *rp++ ;
1600  if (Col[col].is_dead())
1601  {
1602  continue ;
1603  }
1604 
1605  /* get hash number for this column */
1606  hash = Col [col].shared3.hash ;
1607  COLAMD_ASSERT (hash <= n_col) ;
1608 
1609  /* === Get the first column in this hash bucket ===================== */
1610 
1611  head_column = head [hash] ;
1612  if (head_column > Empty)
1613  {
1614  first_col = Col [head_column].shared3.headhash ;
1615  }
1616  else
1617  {
1618  first_col = - (head_column + 2) ;
1619  }
1620 
1621  /* === Consider each column in the hash bucket ====================== */
1622 
1623  for (super_c = first_col ; super_c != Empty ;
1624  super_c = Col [super_c].shared4.hash_next)
1625  {
1626  COLAMD_ASSERT (Col [super_c].is_alive()) ;
1627  COLAMD_ASSERT (Col [super_c].shared3.hash == hash) ;
1628  length = Col [super_c].length ;
1629 
1630  /* prev_c is the column preceding column c in the hash bucket */
1631  prev_c = super_c ;
1632 
1633  /* === Compare super_c with all columns after it ================ */
1634 
1635  for (c = Col [super_c].shared4.hash_next ;
1636  c != Empty ; c = Col [c].shared4.hash_next)
1637  {
1638  COLAMD_ASSERT (c != super_c) ;
1639  COLAMD_ASSERT (Col[c].is_alive()) ;
1640  COLAMD_ASSERT (Col [c].shared3.hash == hash) ;
1641 
1642  /* not identical if lengths or scores are different */
1643  if (Col [c].length != length ||
1644  Col [c].shared2.score != Col [super_c].shared2.score)
1645  {
1646  prev_c = c ;
1647  continue ;
1648  }
1649 
1650  /* compare the two columns */
1651  cp1 = &A [Col [super_c].start] ;
1652  cp2 = &A [Col [c].start] ;
1653 
1654  for (i = 0 ; i < length ; i++)
1655  {
1656  /* the columns are "clean" (no dead rows) */
1657  COLAMD_ASSERT ( cp1->is_alive() );
1658  COLAMD_ASSERT ( cp2->is_alive() );
1659  /* row indices will same order for both supercols, */
1660  /* no gather scatter necessary */
1661  if (*cp1++ != *cp2++)
1662  {
1663  break ;
1664  }
1665  }
1666 
1667  /* the two columns are different if the for-loop "broke" */
1668  if (i != length)
1669  {
1670  prev_c = c ;
1671  continue ;
1672  }
1673 
1674  /* === Got it! two columns are identical =================== */
1675 
1676  COLAMD_ASSERT (Col [c].shared2.score == Col [super_c].shared2.score) ;
1677 
1678  Col [super_c].shared1.thickness += Col [c].shared1.thickness ;
1679  Col [c].shared1.parent = super_c ;
1680  Col[c].kill_non_principal() ;
1681  /* order c later, in order_children() */
1682  Col [c].shared2.order = Empty ;
1683  /* remove c from hash bucket */
1684  Col [prev_c].shared4.hash_next = Col [c].shared4.hash_next ;
1685  }
1686  }
1687 
1688  /* === Empty this hash bucket ======================================= */
1689 
1690  if (head_column > Empty)
1691  {
1692  /* corresponding degree list "hash" is not empty */
1693  Col [head_column].shared3.headhash = Empty ;
1694  }
1695  else
1696  {
1697  /* corresponding degree list "hash" is empty */
1698  head [hash] = Empty ;
1699  }
1700  }
1701 }
1702 
1703 
1704 /* ========================================================================== */
1705 /* === garbage_collection =================================================== */
1706 /* ========================================================================== */
1707 
1708 /*
1709  Defragments and compacts columns and rows in the workspace A. Used when
1710  all available memory has been used while performing row merging. Returns
1711  the index of the first free position in A, after garbage collection. The
1712  time taken by this routine is linear is the size of the array A, which is
1713  itself linear in the number of nonzeros in the input matrix.
1714  Not user-callable.
1715 */
1716 template <typename IndexType>
1717 static IndexType garbage_collection /* returns the new value of pfree */
1718  (
1719  /* === Parameters ======================================================= */
1720 
1721  IndexType n_row, /* number of rows */
1722  IndexType n_col, /* number of columns */
1723  RowStructure<IndexType> Row [], /* row info */
1724  ColStructure<IndexType> Col [], /* column info */
1725  IndexType A [], /* A [0 ... Alen-1] holds the matrix */
1726  IndexType *pfree /* &A [0] ... pfree is in use */
1727  )
1728 {
1729  /* === Local variables ================================================== */
1730 
1731  IndexType *psrc ; /* source pointer */
1732  IndexType *pdest ; /* destination pointer */
1733  IndexType j ; /* counter */
1734  IndexType r ; /* a row index */
1735  IndexType c ; /* a column index */
1736  IndexType length ; /* length of a row or column */
1737 
1738  /* === Defragment the columns =========================================== */
1739 
1740  pdest = &A[0] ;
1741  for (c = 0 ; c < n_col ; c++)
1742  {
1743  if (Col[c].is_alive())
1744  {
1745  psrc = &A [Col [c].start] ;
1746 
1747  /* move and compact the column */
1748  COLAMD_ASSERT (pdest <= psrc) ;
1749  Col [c].start = (IndexType) (pdest - &A [0]) ;
1750  length = Col [c].length ;
1751  for (j = 0 ; j < length ; j++)
1752  {
1753  r = *psrc++ ;
1754  if (Row[r].is_alive())
1755  {
1756  *pdest++ = r ;
1757  }
1758  }
1759  Col [c].length = (IndexType) (pdest - &A [Col [c].start]) ;
1760  }
1761  }
1762 
1763  /* === Prepare to defragment the rows =================================== */
1764 
1765  for (r = 0 ; r < n_row ; r++)
1766  {
1767  if (Row[r].is_alive())
1768  {
1769  if (Row [r].length == 0)
1770  {
1771  /* this row is of zero length. cannot compact it, so kill it */
1772  COLAMD_DEBUG3 (("Defrag row kill\n")) ;
1773  Row[r].kill() ;
1774  }
1775  else
1776  {
1777  /* save first column index in Row [r].shared2.first_column */
1778  psrc = &A [Row [r].start] ;
1779  Row [r].shared2.first_column = *psrc ;
1780  COLAMD_ASSERT (Row[r].is_alive()) ;
1781  /* flag the start of the row with the one's complement of row */
1782  *psrc = ones_complement(r) ;
1783 
1784  }
1785  }
1786  }
1787 
1788  /* === Defragment the rows ============================================== */
1789 
1790  psrc = pdest ;
1791  while (psrc < pfree)
1792  {
1793  /* find a negative number ... the start of a row */
1794  if (*psrc++ < 0)
1795  {
1796  psrc-- ;
1797  /* get the row index */
1798  r = ones_complement(*psrc) ;
1799  COLAMD_ASSERT (r >= 0 && r < n_row) ;
1800  /* restore first column index */
1801  *psrc = Row [r].shared2.first_column ;
1802  COLAMD_ASSERT (Row[r].is_alive()) ;
1803 
1804  /* move and compact the row */
1805  COLAMD_ASSERT (pdest <= psrc) ;
1806  Row [r].start = (IndexType) (pdest - &A [0]) ;
1807  length = Row [r].length ;
1808  for (j = 0 ; j < length ; j++)
1809  {
1810  c = *psrc++ ;
1811  if (Col[c].is_alive())
1812  {
1813  *pdest++ = c ;
1814  }
1815  }
1816  Row [r].length = (IndexType) (pdest - &A [Row [r].start]) ;
1817 
1818  }
1819  }
1820  /* ensure we found all the rows */
1821  COLAMD_ASSERT (debug_rows == 0) ;
1822 
1823  /* === Return the new value of pfree ==================================== */
1824 
1825  return ((IndexType) (pdest - &A [0])) ;
1826 }
1827 
1828 
1829 /* ========================================================================== */
1830 /* === clear_mark =========================================================== */
1831 /* ========================================================================== */
1832 
1833 /*
1834  Clears the Row [].shared2.mark array, and returns the new tag_mark.
1835  Return value is the new tag_mark. Not user-callable.
1836 */
1837 template <typename IndexType>
1838 static inline IndexType clear_mark /* return the new value for tag_mark */
1839  (
1840  /* === Parameters ======================================================= */
1841 
1842  IndexType n_row, /* number of rows in A */
1843  RowStructure<IndexType> Row [] /* Row [0 ... n_row-1].shared2.mark is set to zero */
1844  )
1845 {
1846  /* === Local variables ================================================== */
1847 
1848  IndexType r ;
1849 
1850  for (r = 0 ; r < n_row ; r++)
1851  {
1852  if (Row[r].is_alive())
1853  {
1854  Row [r].shared2.mark = 0 ;
1855  }
1856  }
1857  return (1) ;
1858 }
1859 
1860 } // namespace Colamd
1861 
1862 } // namespace internal
1863 #endif
internal::Colamd::Status
Status
Definition: Eigen_Colamd.h:91
internal::Colamd::Alive
@ Alive
Definition: Eigen_Colamd.h:120
internal::Colamd::ColStructure::is_dead
bool is_dead() const
Definition: Eigen_Colamd.h:167
internal::Colamd::ErrorNrowNegative
@ ErrorNrowNegative
Definition: Eigen_Colamd.h:96
internal::Colamd::ColStructure::score
IndexType score
Definition: Eigen_Colamd.h:150
internal::Colamd::ColStructure::headhash
IndexType headhash
Definition: Eigen_Colamd.h:155
internal::Colamd::KnobsStatsIndex
KnobsStatsIndex
Definition: Eigen_Colamd.h:71
col
m col(1)
internal::Colamd::order_children
static void order_children(IndexType n_col, ColStructure< IndexType > Col[], IndexType p[])
Definition: Eigen_Colamd.h:1466
internal::Colamd::RowStructure::mark
IndexType mark
Definition: Eigen_Colamd.h:191
internal::Colamd::ColStructure::is_alive
bool is_alive() const
Definition: Eigen_Colamd.h:169
internal::Colamd::ErrorANotPresent
@ ErrorANotPresent
Definition: Eigen_Colamd.h:94
internal::Colamd::OkButJumbled
@ OkButJumbled
Definition: Eigen_Colamd.h:93
c
Scalar Scalar * c
Definition: benchVecAdd.cpp:17
internal::Colamd::DeadPrincipal
@ DeadPrincipal
Definition: Eigen_Colamd.h:126
internal::Colamd::RowStructure::is_alive
bool is_alive() const
Definition: Eigen_Colamd.h:197
internal::Colamd::Info3
@ Info3
Definition: Eigen_Colamd.h:87
internal::Colamd::DenseCol
@ DenseCol
Definition: Eigen_Colamd.h:76
internal::Colamd::ColStructure::shared4
union internal::Colamd::ColStructure::@785 shared4
internal::Colamd::DefragCount
@ DefragCount
Definition: Eigen_Colamd.h:79
internal::Colamd::ErrorPNotPresent
@ ErrorPNotPresent
Definition: Eigen_Colamd.h:95
gtsam::Row
std::vector< double > Row
Definition: SignatureParser.cpp:9
internal::Colamd::ColStructure::kill_non_principal
void kill_non_principal()
Definition: Eigen_Colamd.h:175
internal::Colamd::ColStructure
Definition: Eigen_Colamd.h:136
internal::Colamd::ErrorP0Nonzero
@ ErrorP0Nonzero
Definition: Eigen_Colamd.h:99
internal::Colamd::ErrorATooSmall
@ ErrorATooSmall
Definition: Eigen_Colamd.h:100
internal::Colamd::RowStructure::p
IndexType p
Definition: Eigen_Colamd.h:187
internal::Colamd::ColStructure::hash_next
IndexType hash_next
Definition: Eigen_Colamd.h:164
COLAMD_DEBUG0
#define COLAMD_DEBUG0(params)
Definition: Eigen_Colamd.h:253
internal::Colamd::recommended
IndexType recommended(IndexType nnz, IndexType n_row, IndexType n_col)
Returns the recommended value of Alen.
Definition: Eigen_Colamd.h:277
internal::Colamd::ColStructure::shared3
union internal::Colamd::ColStructure::@784 shared3
internal::Colamd::RowStructure
Definition: Eigen_Colamd.h:180
internal::Colamd::Info1
@ Info1
Definition: Eigen_Colamd.h:85
internal::Colamd::ColStructure::order
IndexType order
Definition: Eigen_Colamd.h:151
hash
ssize_t hash(handle obj)
Definition: pytypes.h:917
COLAMD_DEBUG2
#define COLAMD_DEBUG2(params)
Definition: Eigen_Colamd.h:255
internal::Colamd::ColStructure::shared2
union internal::Colamd::ColStructure::@783 shared2
internal::Colamd::garbage_collection
static IndexType garbage_collection(IndexType n_row, IndexType n_col, RowStructure< IndexType > Row[], ColStructure< IndexType > Col[], IndexType A[], IndexType *pfree)
Definition: Eigen_Colamd.h:1718
internal::Colamd::Dead
@ Dead
Definition: Eigen_Colamd.h:121
internal::Colamd::Ok
@ Ok
Definition: Eigen_Colamd.h:92
internal::Colamd::RowStructure::length
IndexType length
Definition: Eigen_Colamd.h:183
internal::Colamd::colamd_r
IndexType colamd_r(IndexType n_row)
Definition: Eigen_Colamd.h:226
A
Definition: test_numpy_dtypes.cpp:298
j
std::ptrdiff_t j
Definition: tut_arithmetic_redux_minmax.cpp:2
internal::Colamd::RowStructure::degree
IndexType degree
Definition: Eigen_Colamd.h:186
stats
bool stats
Definition: SolverComparer.cpp:100
internal::Colamd::Empty
const int Empty
Definition: Eigen_Colamd.h:116
internal::Colamd::RowStructure::kill
void kill()
Definition: Eigen_Colamd.h:199
internal::Colamd::compute_ordering
static bool compute_ordering(IndexType n_row, IndexType n_col, IndexType Alen, IndexType *A, IndexType *p, double knobs[NKnobs], IndexType stats[NStats])
Computes a column ordering using the column approximate minimum degree ordering.
Definition: Eigen_Colamd.h:342
internal::Colamd::ColumnStatus
ColumnStatus
Definition: Eigen_Colamd.h:125
internal::Colamd::ErrorColLengthNegative
@ ErrorColLengthNegative
Definition: Eigen_Colamd.h:101
Eigen::numext::mini
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE T mini(const T &x, const T &y)
Definition: Eigen/src/Core/MathFunctions.h:1085
internal::Colamd::ColStructure::thickness
IndexType thickness
Definition: Eigen_Colamd.h:143
internal::Colamd::RowStructure::is_dead
bool is_dead() const
Definition: Eigen_Colamd.h:195
internal::Colamd::ColStructure::shared1
union internal::Colamd::ColStructure::@782 shared1
internal::Colamd::clear_mark
static IndexType clear_mark(IndexType n_row, RowStructure< IndexType > Row[])
Definition: Eigen_Colamd.h:1839
internal::Colamd::ColStructure::prev
IndexType prev
Definition: Eigen_Colamd.h:158
internal::Colamd::ErrorNnzNegative
@ ErrorNnzNegative
Definition: Eigen_Colamd.h:98
internal::Colamd::RowStructure::first_column
IndexType first_column
Definition: Eigen_Colamd.h:192
internal::Colamd::ColStructure::degree_next
IndexType degree_next
Definition: Eigen_Colamd.h:163
internal::Colamd::RowStructure::shared1
union internal::Colamd::RowStructure::@786 shared1
COLAMD_DEBUG3
#define COLAMD_DEBUG3(params)
Definition: Eigen_Colamd.h:256
head
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE FixedSegmentReturnType< internal::get_fixed_value< NType >::value >::Type head(NType n)
Definition: BlockMethods.h:1208
internal::Colamd::ColStructure::start
IndexType start
Definition: Eigen_Colamd.h:138
internal::Colamd::NKnobs
const int NKnobs
Definition: Eigen_Colamd.h:65
internal::Colamd::set_defaults
static void set_defaults(double knobs[NKnobs])
set default parameters The use of this routine is optional.
Definition: Eigen_Colamd.h:306
internal::Colamd::RowColumnStatus
RowColumnStatus
Definition: Eigen_Colamd.h:119
internal::Colamd::ErrorOutOfMemory
@ ErrorOutOfMemory
Definition: Eigen_Colamd.h:103
internal::Colamd::find_ordering
static IndexType find_ordering(IndexType n_row, IndexType n_col, IndexType Alen, RowStructure< IndexType > Row[], ColStructure< IndexType > Col[], IndexType A[], IndexType head[], IndexType n_col2, IndexType max_deg, IndexType pfree)
Definition: Eigen_Colamd.h:955
internal::Colamd::NStats
const int NStats
Definition: Eigen_Colamd.h:68
internal::Colamd::colamd_c
IndexType colamd_c(IndexType n_col)
Definition: Eigen_Colamd.h:222
COLAMD_ASSERT
#define COLAMD_ASSERT(expression)
Definition: Eigen_Colamd.h:259
return
if n return
Definition: level1_cplx_impl.h:33
row
m row(1)
internal::Colamd::Info2
@ Info2
Definition: Eigen_Colamd.h:86
p
float * p
Definition: Tutorial_Map_using.cpp:9
internal::Colamd::ErrorNcolNegative
@ ErrorNcolNegative
Definition: Eigen_Colamd.h:97
internal::Colamd::RowStructure::start
IndexType start
Definition: Eigen_Colamd.h:182
internal::Colamd::ColStructure::kill_principal
void kill_principal()
Definition: Eigen_Colamd.h:173
internal::Colamd::ErrorInternalError
@ ErrorInternalError
Definition: Eigen_Colamd.h:104
internal::Colamd::ColStructure::parent
IndexType parent
Definition: Eigen_Colamd.h:145
internal::Colamd::ColStructure::is_dead_principal
bool is_dead_principal() const
Definition: Eigen_Colamd.h:171
COLAMD_DEBUG4
#define COLAMD_DEBUG4(params)
Definition: Eigen_Colamd.h:257
internal
Definition: BandTriangularSolver.h:13
Eigen::numext::maxi
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE T maxi(const T &x, const T &y)
Definition: Eigen/src/Core/MathFunctions.h:1093
internal::Colamd::RowStructure::shared2
union internal::Colamd::RowStructure::@787 shared2
internal::Colamd::ColStructure::length
IndexType length
Definition: Eigen_Colamd.h:140
internal::Colamd::DeadNonPrincipal
@ DeadNonPrincipal
Definition: Eigen_Colamd.h:127
COLAMD_DEBUG1
#define COLAMD_DEBUG1(params)
Definition: Eigen_Colamd.h:254
internal::Colamd::init_rows_cols
static IndexType init_rows_cols(IndexType n_row, IndexType n_col, RowStructure< IndexType > Row[], ColStructure< IndexType > col[], IndexType A[], IndexType p[], IndexType stats[NStats])
Definition: Eigen_Colamd.h:502
internal::Colamd::DenseRow
@ DenseRow
Definition: Eigen_Colamd.h:73
internal::Colamd::init_scoring
static void init_scoring(IndexType n_row, IndexType n_col, RowStructure< IndexType > Row[], ColStructure< IndexType > Col[], IndexType A[], IndexType head[], double knobs[NKnobs], IndexType *p_n_row2, IndexType *p_n_col2, IndexType *p_max_deg)
Definition: Eigen_Colamd.h:718
i
int i
Definition: BiCGSTAB_step_by_step.cpp:9
internal::Colamd::ColStructure::hash
IndexType hash
Definition: Eigen_Colamd.h:157
internal::Colamd::ErrorRowIndexOutOfBounds
@ ErrorRowIndexOutOfBounds
Definition: Eigen_Colamd.h:102
internal::Colamd::detect_super_cols
static void detect_super_cols(ColStructure< IndexType > Col[], IndexType A[], IndexType head[], IndexType row_start, IndexType row_length)
Definition: Eigen_Colamd.h:1567
internal::Colamd::ones_complement
IndexType ones_complement(const IndexType r)
Definition: Eigen_Colamd.h:111


gtsam
Author(s):
autogenerated on Tue Jun 25 2024 03:00:50