frontal_solver.cc
Go to the documentation of this file.
1 // LIC// ====================================================================
2 // LIC// This file forms part of oomph-lib, the object-oriented,
3 // LIC// multi-physics finite-element library, available
4 // LIC// at http://www.oomph-lib.org.
5 // LIC//
6 // LIC// Copyright (C) 2006-2024 Matthias Heil and Andrew Hazel
7 // LIC//
8 // LIC// This library is free software; you can redistribute it and/or
9 // LIC// modify it under the terms of the GNU Lesser General Public
10 // LIC// License as published by the Free Software Foundation; either
11 // LIC// version 2.1 of the License, or (at your option) any later version.
12 // LIC//
13 // LIC// This library is distributed in the hope that it will be useful,
14 // LIC// but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // LIC// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 // LIC// Lesser General Public License for more details.
17 // LIC//
18 // LIC// You should have received a copy of the GNU Lesser General Public
19 // LIC// License along with this library; if not, write to the Free Software
20 // LIC// Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
21 // LIC// 02110-1301 USA.
22 // LIC//
23 // LIC// The authors may be contacted at oomph-lib@maths.man.ac.uk.
24 // LIC//
25 // LIC//====================================================================
26 // Interface to HSL frontal solver (fortran)
27 
28 
29 #ifdef OOMPH_HAS_MPI
30 #include "mpi.h"
31 #endif
32 
33 // oomph-lib headers
34 #include "cfortran.h"
35 #include "frontal.h"
36 #include "frontal_solver.h"
37 #include "problem.h"
38 #include "mesh.h"
39 #include "matrices.h"
40 #include "assembly_handler.h"
41 
42 namespace oomph
43 {
44  //====================================================================
45  /// Special solver for problems with one DOF -- HSL_MA42 can't handle
46  /// that!
47  //====================================================================
48  void HSL_MA42::solve_for_one_dof(Problem* const& problem_pt,
49  DoubleVector& result)
50  {
51  // Find the number of elements
52  unsigned long n_el = problem_pt->mesh_pt()->nelement();
53 
54 #ifdef PARANOID
55  // Find the number of dofs (variables)
56  int n_dof = problem_pt->ndof();
57  // Check
58  if (n_dof != 1)
59  {
60  throw OomphLibError(
61  "You can only call this if the problem has just one dof!",
62  OOMPH_CURRENT_FUNCTION,
63  OOMPH_EXCEPTION_LOCATION);
64  }
65 #endif
66 
67  // Global jac and residuals are scalars!
68  double global_jac = 0.0;
69  double global_res = 0.0;
70 
71  // Locally cache pointer to assembly handler
72  AssemblyHandler* const assembly_handler_pt =
73  problem_pt->assembly_handler_pt();
74 
75 
76  // Do the assembly
77  for (unsigned long e = 0; e < n_el; e++)
78  {
79  // Get pointer to the element
80  GeneralisedElement* elem_pt = problem_pt->mesh_pt()->element_pt(e);
81 
82  // Get number of variables in element
83  int nvar = assembly_handler_pt->ndof(elem_pt);
84 
85  // Only assemble if there's something to be assembled
86  if (nvar > 0)
87  {
88  // Set up matrices and Vector for call to get_residuals
89  Vector<double> residuals(nvar);
90  DenseMatrix<double> jacobian(nvar);
91 
92  // Get the residuals and jacobian
93  assembly_handler_pt->get_jacobian(elem_pt, residuals, jacobian);
94 
95  // Add contribution
96  global_jac += jacobian(0, 0);
97  global_res += residuals[0];
98  }
99  }
100 
101  // "Solve"
102  result[0] = global_res / global_jac;
103 
104  // If we are storing the matrix, allocate the storage
105  if (Enable_resolve)
106  {
107  // If storage has been allocated delete it
108  if (W != 0)
109  {
110  delete[] W;
111  }
112  // Now allocate new storage
113  W = new double[1];
114  // Store the global jacobian
115  W[0] = global_jac;
116  // Set the number of degrees of freedom
117  N_dof = 1;
118  }
119  // Set the sign of the jacobian
120  problem_pt->sign_of_jacobian() =
121  static_cast<int>(std::fabs(global_jac) / global_jac);
122  }
123 
124  //====================================================================
125  /// Wrapper for HSL MA42 frontal solver
126  //====================================================================
127  void HSL_MA42::solve(Problem* const& problem_pt, DoubleVector& result)
128  {
129 #ifdef OOMPH_HAS_MPI
130  if (problem_pt->communicator_pt()->nproc() > 1)
131  {
132  std::ostringstream error_message;
133  error_message
134  << "HSL_MA42 solver cannot be used in parallel.\n"
135  << "Please change to another linear solver.\n"
136  << "If you want to use a frontal solver then try MumpsSolver\n";
137 
138  throw OomphLibError(
139  error_message.str(), OOMPH_CURRENT_FUNCTION, OOMPH_EXCEPTION_LOCATION);
140  }
141 #endif
142 
143 
144  // Find the number of elements
145  unsigned long n_el = problem_pt->mesh_pt()->nelement();
146 
147  // Find the number of dofs (variables) and store for resolves
148  N_dof = problem_pt->ndof();
149 
150  // Build the distribution .. this is a serial solver
151  LinearAlgebraDistribution dist(problem_pt->communicator_pt(), N_dof, false);
152  this->build_distribution(dist);
153 
154  // Cast the number of dofs into an integer for the HSL solver
155  int n_dof = N_dof;
156 
157  // Special case: Just one variable! MA42 can't handle that
158  if (n_dof == 1)
159  {
160  solve_for_one_dof(problem_pt, result);
161  return;
162  }
163 
164  // Reorder?
165  if (this->Reorder_flag)
166  {
167  reorder_elements(problem_pt);
168  }
169 
170  // Set the control flags
171  int ifsize[5];
172  double cntl[2], rinfo[2];
173 
174  // Set up memory for last and for ndf
175  int ndf, nmaxe = 2, nrhs = 1, lrhs = 1;
176 
177  // Memory for last and dx should be allocated dynamically from the heap
178  // rather than the stack, otherwise one gets a nasty segmentation fault
179  int* last = new int[n_dof];
180  // Set up memory for dx
181  double** dx = new double*;
182  *dx = new double[n_dof];
183 
184  // Provide storage for exact values required for lenbuf array
185  int exact_lenbuf[3] = {0, 0, 0};
186  bool exact_lenbuf_available = false;
187 
188  // Storage for recommendations for lenbuf so we can check how
189  // close our factors are
190  int lenbuf0_recommendation = 0;
191  int lenbuf1_recommendation = 0;
192  int lenbuf2_recommendation = 0;
193 
194  // Provide storage
195  int lenbuf[3];
196 
197 
198  // Provide storage for exact values required for lenfle array
199  int exact_lenfle[3] = {0, 0, 0};
200  bool exact_lenfle_available = false;
201 
202  // Storage for recommendation for lenfle so we can check how
203  // close our factor is
204  int lenfle0_recommendation = 0;
205  int lenfle1_recommendation = 0;
206  int lenfle2_recommendation = 0;
207 
208  // Provide storage
209  int lenfle[3];
210 
211 
212  // Storage for recommendations for front size so we can check how
213  // close our factors are
214  double front0_recommendation = 0;
215  double front1_recommendation = 0;
216 
217 
218  // Flag to indicate if exact, required values for nfront are available
219  // from previous unsucessful solve
220  bool exact_nfront_available = false;
221 
222  // Storage for front sizes
223  int nfront[2];
224 
225  // Locally cache pointer to assembly handler
226  AssemblyHandler* const assembly_handler_pt =
227  problem_pt->assembly_handler_pt();
228 
229 
230  // Loop over size allocation/solver until we've made all the arrays
231  //=================================================================
232  // big enough...
233  //==============
234  bool not_done = true;
235  while (not_done)
236  {
237  // Initialise frontal solver
238  //=========================
239  MA42ID(Icntl, cntl, Isave);
240 
241 
242  // Loop over the elements to setup variables associated with elements
243  //==================================================================
244  for (unsigned long e = 0; e < n_el; e++)
245  {
246  // Pointer to the element
247  GeneralisedElement* elem_pt = problem_pt->mesh_pt()->element_pt(e);
248 
249  // MH: changed array to allocateable
250  int* ivar;
251 
252  // Get number of variables in element
253  int nvar = assembly_handler_pt->ndof(elem_pt);
254 
255 
256  // Multiple equations
257  //-------------------
258  if (nvar != 1)
259  {
260  // Copy global equation numbers into local array
261  ivar = new int[nvar];
262 
263  for (int i = 0; i < nvar; i++)
264  {
265  // Need to add one to equation numbers
266  ivar[i] = 1 + assembly_handler_pt->eqn_number(elem_pt, i);
267  }
268  }
269 
270  // Just one equation
271  //------------------
272  else
273  {
274  // Copy global equation numbers into local array - minimum length: 2
275  ivar = new int[2];
276 
277  // Here's the number of the only equation
278  int only_eqn = assembly_handler_pt->eqn_number(elem_pt, 0);
279 
280  // Need to add one to equation numbers
281  ivar[0] = 1 + only_eqn;
282 
283  // Now add a dummy eqn either before or after the current one
284  int dummy_eqn;
285  if (only_eqn != (n_dof - 1))
286  {
287  dummy_eqn = only_eqn + 1;
288  }
289  else
290  {
291  dummy_eqn = only_eqn - 1;
292  }
293  ivar[1] = 1 + dummy_eqn;
294 
295  nvar = 2;
296  }
297 
298 
299  // Now call the frontal routine
300  // GB: added check to exclude case with no dofs
301  if (nvar)
302  {
303  MA42AD(nvar, ivar, ndf, last, n_dof, Icntl, Isave, Info);
304  // ALH : throw an exception if the frontal_solver fails
305  if (Info[0] < 0)
306  {
307  throw NewtonSolverError(true);
308  }
309  }
310 
311  // Cleanup
312  delete[] ivar;
313  ivar = 0;
314  }
315 
316 
317  // Loop over the elements again for symbolic factorisation
318  //=======================================================
319  for (unsigned long e = 0; e < n_el; e++)
320  {
321  // Pointer to the element
322  GeneralisedElement* elem_pt = problem_pt->mesh_pt()->element_pt(e);
323 
324  // MH: changed array to allocateable
325  int* ivar;
326 
327  // Get number of variables in element
328  int nvar = assembly_handler_pt->ndof(elem_pt);
329 
330  // Multiple equations
331  //-------------------
332  if (nvar != 1)
333  {
334  // Copy global equation numbers into local array
335  ivar = new int[nvar];
336 
337  for (int i = 0; i < nvar; i++)
338  {
339  // Need to add one to equation numbers
340  ivar[i] = 1 + assembly_handler_pt->eqn_number(elem_pt, i);
341  }
342  }
343 
344  // Just one equation
345  //------------------
346  else
347  {
348  // Copy global equation numbers into local array - minimum length: 2
349  ivar = new int[2];
350 
351  // Here's the number of the only equation
352  int only_eqn = assembly_handler_pt->eqn_number(elem_pt, 0);
353 
354  // Need to add one to equation numbers
355  ivar[0] = 1 + only_eqn;
356 
357  // Now add a dummy eqn either before or after the current one
358  int dummy_eqn;
359  if (only_eqn != (n_dof - 1))
360  {
361  dummy_eqn = only_eqn + 1;
362  }
363  else
364  {
365  dummy_eqn = only_eqn - 1;
366  }
367  ivar[1] = 1 + dummy_eqn;
368 
369  nvar = 2;
370  }
371 
372  // GB: added check to exclude case with no dofs
373  if (nvar)
374  {
375  MA42JD(nvar, ivar, ndf, last, nmaxe, ifsize, Icntl, Isave, Info);
376 
377  // ALH : throw an exception if the frontal_solver fails
378  if (Info[0] < 0)
379  {
380  throw NewtonSolverError(true);
381  }
382  }
383 
384  // Cleanup
385  delete[] ivar;
386  ivar = 0;
387 
388  } // end of symbolic factorisation
389 
390 
391  // Front sizes: "Somewhat larger than..."
392  //---------------------------------------
393  front0_recommendation = ifsize[0];
394  front1_recommendation = ifsize[1];
395  if (!exact_nfront_available)
396  {
397  nfront[0] = int(ceil(Front_factor * double(front0_recommendation)));
398  nfront[1] = int(ceil(Front_factor * double(front1_recommendation)));
399 
400  if (n_dof < nfront[0]) nfront[0] = n_dof;
401  if (n_dof < nfront[1]) nfront[1] = n_dof;
402  }
403  // We have a problem if the front size is cocked
404 
405 
406  // Sizes for lenbuf[]: "Somewhat larger than..."
407  //----------------------------------------------
408  lenbuf0_recommendation = ifsize[2] + ndf;
409  // If we are storing the matrix,
410  // we need to allocate storage for lenbuf_1
411  if (Enable_resolve)
412  {
413  lenbuf1_recommendation = ifsize[3];
414  }
415  // Otherwise set it to zero
416  else
417  {
418  lenbuf1_recommendation = 0;
419  }
420  lenbuf2_recommendation = ifsize[4];
421 
422 
423  // Enable use of direct access files?
424  //----------------------------------
426  {
427  if (Doc_stats)
428  {
429  oomph_info << "Using direct access files" << std::endl;
430  }
431 
432  // Unit numbers for direct access files (middle one only used for
433  // re-solve; set to zero as this is not used here).
434  int istrm[3];
435  istrm[0] = 17;
436  // If we are storing the matrix, set the stream
437  if (Enable_resolve)
438  {
439  istrm[1] = 19;
440  }
441  // otherwise, set it to zero
442  else
443  {
444  istrm[1] = 0;
445  }
446  istrm[2] = 18;
447 
448  // Set up the memory: Need memory "somewhat larger" than ...
449  double factor = 1.1;
450  lenbuf[0] = int(ceil(factor * double(10 * (nfront[0] + 1))));
451  // If we are storing the matrix, set the value
452  if (Enable_resolve)
453  {
454  lenbuf[1] = int(ceil(factor * double(10 * (nfront[0]))));
455  }
456  // Otherwise, set to zero
457  else
458  {
459  lenbuf[1] = 0;
460  }
461  lenbuf[2] = int(ceil(factor * double(10 * (2 * nfront[0] + 5))));
462 
463 
464  // Initial size allocation: Need memory "somewhat larger" than ...
465  if (exact_lenfle_available)
466  {
467  lenfle[0] = exact_lenfle[0];
468  lenfle[1] = exact_lenfle[1];
469  lenfle[2] = exact_lenfle[2];
470  }
471  else
472  {
473  lenfle0_recommendation = ifsize[2] + ndf;
474  lenfle1_recommendation = ifsize[3];
475  lenfle2_recommendation = ifsize[4];
476  lenfle[0] = int(ceil(Lenfle_factor * double(lenfle0_recommendation)));
477  lenfle[1] = int(ceil(Lenfle_factor * double(lenfle1_recommendation)));
478  lenfle[2] = int(ceil(Lenfle_factor * double(lenfle2_recommendation)));
479  }
480 
481  // Setup direct access files
482  MA42PD(istrm, lenbuf, lenfle, Icntl, Isave, Info);
483  }
484  else
485  {
486  if (Doc_stats)
487  {
488  oomph_info << "Not using direct access files" << std::endl;
489  }
490 
491  // Initial size allocation: Need memory "somewhat larger" than ...
492  if (exact_lenbuf_available)
493  {
494  lenbuf[0] = exact_lenbuf[0];
495  lenbuf[1] = exact_lenbuf[1];
496  lenbuf[2] = exact_lenbuf[2];
497  }
498  else
499  {
500  lenbuf[0] =
501  int(ceil(Lenbuf_factor0 * double(lenbuf0_recommendation)));
502  // If we are storing the matrix, set the buffer size
503  if (Enable_resolve)
504  {
505  lenbuf[1] =
506  int(ceil(Lenbuf_factor1 * double(lenbuf1_recommendation)));
507  }
508  // Otherwise its zero
509  else
510  {
511  lenbuf[1] = 0;
512  }
513  lenbuf[2] =
514  int(ceil(Lenbuf_factor2 * double(lenbuf2_recommendation)));
515  }
516  }
517 
518 
519  if (Doc_stats)
520  {
521  oomph_info << "\n FRONT SIZE (min and actual): " << ifsize[0] << " "
522  << nfront[0] << std::endl;
523  }
524 
525 
526  // Initialise success
527  bool success = true;
528 
529  // Workspace arrays: calculate amount in local variables initiall
530  int lw = 1 + lenbuf[0] + lenbuf[1] + nfront[0] * nfront[1];
531  if (lrhs * nfront[0] > nrhs * nfront[1])
532  {
533  lw += lrhs * nfront[0];
534  }
535  else
536  {
537  lw += nrhs * nfront[1];
538  }
539 
540  int liw = lenbuf[2] + 2 * nfront[0] + 4 * nfront[1];
541 
542  // Set up memory for w and iw
543  // Again allocate dynamically from the heap, rather than the stack
544  // If the required amount of storage has changed, reallocate
545  // and store the values in the object member data
546  if (lw != Lw)
547  {
548  // Set the new value of Lw (member data)
549  Lw = lw;
550  // Delete the previously allocated storage
551  if (W) delete[] W;
552  // Now allocate new storage
553  W = new (std::nothrow) double[Lw];
554  if (!W)
555  {
556  throw OomphLibError("Out of memory in allocation of w\n",
557  OOMPH_CURRENT_FUNCTION,
558  OOMPH_EXCEPTION_LOCATION);
559  }
560  }
561 
562  // If the required amount of storage has changed, reallocate
563  if (liw != Liw)
564  {
565  // Set the new value of Liw (member data)
566  Liw = liw;
567  // Delete the previously allocated storgae
568  if (IW != 0)
569  {
570  delete[] IW;
571  }
572  // Now allocate new storage
573  IW = new (std::nothrow) int[Liw];
574  if (!IW)
575  {
576  throw OomphLibError("Out of memory in allocation of iw\n",
577  OOMPH_CURRENT_FUNCTION,
578  OOMPH_EXCEPTION_LOCATION);
579  }
580  }
581 
582  // Doc
583  if (Doc_stats)
584  {
585  double temp = (lenbuf[0] * sizeof(double) + lenbuf[2] * sizeof(int)) /
586  (1024.0 * 1024.0);
587  oomph_info << "\n ESTIMATED MEMORY USAGE " << temp << "Mb" << std::endl;
588  }
589 
590 
591  // Now do the actual frontal assembly and solve loop
592  //=================================================
593  for (unsigned long e = 0; e < n_el; e++)
594  {
595  // Get pointer to the element
596  GeneralisedElement* elem_pt = problem_pt->mesh_pt()->element_pt(e);
597 
598  // Get number of variables in element
599  int nvar = assembly_handler_pt->ndof(elem_pt);
600 
601  // MH: changed array to allocatable
602  int* ivar;
603 
604  // Multiple equations
605  //-------------------
606  if (nvar != 1)
607  {
608  // Copy global equation numbers into local array
609  ivar = new int[nvar];
610 
611  for (int i = 0; i < nvar; i++)
612  {
613  // Need to add one to equation numbers
614  ivar[i] = 1 + assembly_handler_pt->eqn_number(elem_pt, i);
615  }
616  nmaxe = nvar;
617  }
618 
619  // Just one equation
620  //------------------
621  else
622  {
623  // Copy global equation numbers into local array - minimum length: 2
624  ivar = new int[2];
625 
626  // Here's the number of the only equation
627  int only_eqn = assembly_handler_pt->eqn_number(elem_pt, 0);
628 
629  // Need to add one to equation numbers
630  ivar[0] = 1 + only_eqn;
631 
632  // Now add a dummy eqn either before or after the current one
633  int dummy_eqn;
634  if (only_eqn != (n_dof - 1))
635  {
636  dummy_eqn = only_eqn + 1;
637  }
638  else
639  {
640  dummy_eqn = only_eqn - 1;
641  }
642  ivar[1] = 1 + dummy_eqn;
643 
644  nmaxe = 2;
645  }
646 
647  // Set up matrices and Vector for call to get_residuals
648  Vector<double> residuals(nvar);
649  DenseMatrix<double> jacobian(nvar);
650 
651  // Get the residuals and jacobian
652  assembly_handler_pt->get_jacobian(elem_pt, residuals, jacobian);
653 
654  // Add dummy rows and columns if required
655  if (nvar == 1)
656  {
657  double onlyjac = jacobian(0, 0);
658  jacobian.resize(2, 2);
659  jacobian(0, 0) = onlyjac;
660  jacobian(1, 0) = 0.0;
661  jacobian(0, 1) = 0.0;
662  jacobian(1, 1) = 0.0;
663 
664  double onlyres = residuals[0];
665  residuals.resize(2);
666  residuals[0] = onlyres;
667  residuals[1] = 0.0;
668 
669  nvar = 2;
670  }
671 
672 
673  // GB: added check to exclude case with no dofs
674  if (nvar)
675  {
676  // Now translate the residuals and jacobian into form to be
677  // taken by stupid FORTRAN frontal solver
678  // double avar[nvar][nmaxe], rhs[1][nmaxe];
679  // Assign memory on the heap rather than the stack because the ndofs
680  // could get too large
681  double** avar = new double*[nvar];
682  double* tmp = new double[nvar * nmaxe];
683  for (int i = 0; i < nvar; i++)
684  {
685  avar[i] = &tmp[i * nmaxe];
686  }
687  double** rhs = new double*[1];
688  rhs[0] = new double[nmaxe];
689 
690 
691  for (int i = 0; i < nmaxe; i++)
692  {
693  rhs[0][i] = residuals[i];
694  for (int j = 0; j < nvar; j++)
695  {
696  // Note need to transpose here
697  avar[j][i] = jacobian(i, j);
698  }
699  }
700 
701  // Call the frontal solver
702  MA42BD(nvar,
703  ivar,
704  ndf,
705  last,
706  nmaxe,
707  avar,
708  nrhs,
709  rhs,
710  lrhs,
711  n_dof,
712  dx,
713  nfront,
714  lenbuf,
715  Lw,
716  W,
717  Liw,
718  IW,
719  Icntl,
720  cntl,
721  Isave,
722  Info,
723  rinfo);
724 
725 
726  // Error check:
727  if (Info[0] < 0)
728  {
729  if (Doc_stats)
730  oomph_info << "Error: Info[0]=" << Info[0] << std::endl;
731 
732  // Solve isn't going to be successful
733  success = false;
734 
735  // Error: Entries in lenbuf too small -- this is covered
736  if (Info[0] == -16)
737  {
738  if (Doc_stats)
739  oomph_info << "lenbuf[] too small -- can recover..."
740  << std::endl;
741  }
742  else if (Info[0] == -12)
743  {
744  if (Doc_stats)
745  oomph_info << "nfront[] too small -- can recover..."
746  << std::endl;
747  }
748  else if (Info[0] == -17)
749  {
750  if (Doc_stats)
751  oomph_info << "lenfle[] too small -- can recover..."
752  << std::endl;
753  }
754  else
755  {
756  if (Doc_stats)
757  {
758  oomph_info << "Can't recover from this error" << std::endl;
759  }
760  throw NewtonSolverError(true);
761  }
762  }
763 
764  // Clean up the memory
765  delete[] rhs[0];
766  rhs[0] = 0;
767  delete[] rhs;
768  rhs = 0;
769  delete[] avar[0];
770  avar[0] = 0;
771  delete[] avar;
772  }
773 
774  // Cleanup
775  delete[] ivar;
776  ivar = 0;
777 
778  } // end of loop over elements for assemble/solve
779 
780  // Set the sign of the jacobian matrix
781  problem_pt->sign_of_jacobian() = Info[1];
782 
783  // If we are not storing the matrix, then delete the assigned workspace
784  if (!Enable_resolve)
785  {
786  // Free the workspace memory assigned and set stored values to zero
787  delete[] IW;
788  IW = 0;
789  Liw = 0;
790  delete[] W;
791  W = 0;
792  Lw = 0;
793  // Reset the number of dofs
794  N_dof = 0;
795  }
796 
797  // We've done the elements -- did we encounter an error
798  // that forces us to re-allocate workspace?
799  if (success)
800  {
801  // We're done!
802  not_done = false;
803  }
804  else
805  {
806  // Reset sizes for lenbuf
808  {
809  exact_lenbuf[0] = Info[4];
810  exact_lenbuf[1] = Info[5];
811  exact_lenbuf[2] = Info[6];
812  exact_lenbuf_available = true;
813  }
814 
815  // nfront[] has already been overwritten with required values -- don't
816  // need to update anything. Just indicate that new values are
817  // available so they don't have to be re-assigned.
818  exact_nfront_available = true;
819 
820  // Reset sizes for lenfle
821  exact_lenfle[0] = lenbuf[0] * Info[9];
822  exact_lenfle[1] = lenbuf[1] * Info[10];
823  exact_lenfle[2] = lenbuf[2] * Info[11];
824  exact_lenfle_available = true;
825  }
826 
827  } // end of loop over shuffling of workspace array sizes until it all
828  // fits...
829 
830  result.build(&dist);
831  // Now load the values back into result
832  for (int i = 0; i < n_dof; i++) result[i] = dx[0][i];
833 
834  // Print the actual memory used
835  if (Doc_stats)
836  {
838  {
839  double temp = (Info[4] * sizeof(double) + Info[6] * sizeof(int)) /
840  (1024.0 * 1024.0);
841  oomph_info << " TOTAL MEMORY USED " << temp << "Mb" << std::endl;
842  }
843  oomph_info << std::endl;
844  oomph_info << "lenbuf[]: " << lenbuf[0] << " " << lenbuf[1] << " "
845  << lenbuf[2] << " " << std::endl;
846  oomph_info << "lenbuf[] factors required and initially specified:"
847  << std::endl;
848  oomph_info << "lenbuf[0]: " << Info[4] / double(lenbuf0_recommendation)
849  << " " << Lenbuf_factor0 << std::endl;
850  oomph_info << "lenbuf[1]: " << Info[5] / double(lenbuf1_recommendation)
851  << " " << Lenbuf_factor1 << std::endl;
852 
853  oomph_info << "lenbuf[2]: " << Info[6] / double(lenbuf2_recommendation)
854  << " " << Lenbuf_factor2 << std::endl;
855  oomph_info << "nfront[] factors required and initially specified:"
856  << std::endl;
857  oomph_info << "nfront[0]: " << nfront[0] / double(front0_recommendation)
858  << " " << Front_factor << std::endl;
859  oomph_info << "nfront[1]: " << nfront[1] / double(front1_recommendation)
860  << " " << Front_factor << std::endl;
862  {
863  oomph_info << "lenfle[]: " << lenfle[0] << " " << lenfle[1] << " "
864  << lenfle[2] << " " << std::endl;
865  oomph_info << "lenfle[] factors required and initially specified:"
866  << std::endl;
867  oomph_info << "lenfle[0]: "
868  << Info[9] * lenbuf[0] / double(lenfle0_recommendation)
869  << " " << Lenfle_factor << std::endl;
870  oomph_info << "lenfle[1]: "
871  << Info[10] * lenbuf[1] / double(lenfle1_recommendation)
872  << " " << Lenfle_factor << std::endl;
873  oomph_info << "lenfle[2]: "
874  << Info[11] * lenbuf[2] / double(lenfle2_recommendation)
875  << " " << Lenfle_factor << std::endl;
876  }
877  }
878 
879  // Free the memory assigned
880  delete[] * dx;
881  delete dx;
882  delete[] last;
883  }
884 
885  //====================================================================
886  /// Wrapper for HSL MA42 frontal solver
887  //====================================================================
888  void HSL_MA42::resolve(const DoubleVector& rhs, DoubleVector& result)
889  {
890  // If we haven't stored the factors complain
891  if (W == 0)
892  {
893  throw OomphLibError("Resolve called without first solving",
894  OOMPH_CURRENT_FUNCTION,
895  OOMPH_EXCEPTION_LOCATION);
896  }
897 
898  // Find the number of dofs (variables)
899  int n_dof = N_dof;
900 
901  // Check that the number of DOFs is equal to the size of the RHS vector
902 #ifdef PARANOID
903  if (n_dof != static_cast<int>(rhs.nrow()))
904  {
905  throw OomphLibError(
906  "RHS does not have the same dimension as the linear system",
907  OOMPH_CURRENT_FUNCTION,
908  OOMPH_EXCEPTION_LOCATION);
909  }
910 #endif
911 
912  // Special case: just one variable! MA42 can't handle it
913  // Solve using the stored jacobian (single double)
914  if (n_dof == 1)
915  {
916  result[0] = rhs[0] / W[0];
917  return;
918  }
919 
920  // Otherwise actually call the MA42 routine
921  // We are solving the original matrix (not the transpose)
922  int trans = 0;
923  // There is only one rhs
924  int nrhs = 1;
925  // The size of the vectors is the number of degrees of freedom in the
926  // problem
927  int lx = n_dof;
928 
929  // Allocate storage for the RHS
930  double** b = new double*;
931  *b = new double[n_dof];
932  // Load in the RHS vector
933  for (int i = 0; i < n_dof; i++)
934  {
935  b[0][i] = rhs[i];
936  }
937 
938  // Storage for the results
939  double** x = new double*;
940  *x = new double[n_dof];
941 
942  // Now call the resolver
943  MA42CD(trans, nrhs, lx, b, x, Lw, W, Liw, IW, Icntl, Isave, Info);
944 
945  // If there has been an error throw it
946  if (Info[0] < 0)
947  {
948  throw NewtonSolverError(true);
949  }
950 
951  result.build(this->distribution_pt());
952  // Put the answer into the result array
953  for (int i = 0; i < n_dof; ++i)
954  {
955  result[i] = x[0][i];
956  }
957 
958  // Delete the allocated storage
959  delete[] * x;
960  delete x;
961  delete[] * b;
962  delete b;
963  }
964 
965 
966  //=========================================================================
967  /// Function to reorder the elements according to Sloan's algorithm
968  //=========================================================================
969  void HSL_MA42::reorder_elements(Problem* const& problem_pt)
970  {
971  // Find the number of elements
972  unsigned long n_el = problem_pt->mesh_pt()->nelement();
973 
974  // Find the number of dofs
975  int n_dof = problem_pt->ndof();
976 
977  Vector<int> order(n_el);
978 
979  // Control parameters
980  int icntl[10], info[15], direct, nsup, vars[n_dof], svar[n_dof];
981  int liw, lw, *perm = 0;
982  double wt[3], rinfo[6];
983 
984 
985  // Direct ordering: 1
986  direct = 1;
987 
988  // Initial guess for workspaces (deliberately too small so the
989  // routine fails and returns the required sizes
990  liw = 1;
991  lw = 1;
992 
993  // Initialise things here
994  MC63ID(icntl);
995 
996 
997  // Suppress printing of error and warning messages?
998  if (!Doc_stats)
999  {
1000  icntl[0] = -1;
1001  icntl[1] = -1;
1002  }
1003 
1004  // Set up iw and w
1005  int* iw = new int[liw];
1006  double* w = new double[lw];
1007 
1008  // Set up the vectors that hold the element connectivity information
1009  // Can initialise the first entry of eltptr to 1
1010  Vector<int> eltvar, eltptr(1, 1);
1011 
1012  // Locally cache pointer to assembly handler
1013  AssemblyHandler* const assembly_handler_pt =
1014  problem_pt->assembly_handler_pt();
1015 
1016  // Now loop over all the elements and assemble eltvar and eltptr
1017  for (unsigned long e = 0; e < n_el; e++)
1018  {
1019  // Set up the default order
1020  order[e] = e;
1021 
1022  // Get pointer to the element
1023  GeneralisedElement* elem_pt = problem_pt->mesh_pt()->element_pt(e);
1024 
1025  // Get the number of variables in the element via the assembly handler
1026  int nvar = assembly_handler_pt->ndof(elem_pt);
1027 
1028  // Multiple equations
1029  //-------------------
1030  if (nvar != 1)
1031  {
1032  // Copy the equation numbers into the global array
1033  for (int i = 0; i < nvar; i++)
1034  {
1035  // Need to add one to equation numbers
1036  eltvar.push_back(1 + assembly_handler_pt->eqn_number(elem_pt, i));
1037  }
1038  eltptr.push_back(eltptr.back() + nvar);
1039  }
1040  // Just one equation
1041  //------------------
1042  else
1043  {
1044  // Here's the number of the only equation
1045  int only_eqn = assembly_handler_pt->eqn_number(elem_pt, 0);
1046 
1047  // Need to add one to equation numbers
1048  eltvar.push_back(1 + only_eqn);
1049 
1050  // Now add a dummy eqn either before or after the current one
1051  int dummy_eqn;
1052  if (only_eqn != (n_dof - 1))
1053  {
1054  dummy_eqn = only_eqn + 1;
1055  }
1056  else
1057  {
1058  dummy_eqn = only_eqn - 1;
1059  }
1060 
1061  eltvar.push_back(1 + dummy_eqn);
1062 
1063  eltptr.push_back(eltptr.back() + 2);
1064  }
1065 
1066  } // End of loop over the elements
1067 
1068  // Set the value of n_e (number of entries in the element variable list
1069  int n_e = eltvar.size();
1070 
1071  do
1072  {
1073  // Call the reordering routine
1074  MC63AD(direct,
1075  n_dof,
1076  n_el,
1077  n_e,
1078  &eltvar[0],
1079  &eltptr[0],
1080  &order[0],
1081  perm,
1082  nsup,
1083  vars,
1084  svar,
1085  wt,
1086  liw,
1087  iw,
1088  lw,
1089  w,
1090  icntl,
1091  info,
1092  rinfo);
1093 
1094  // If not enough space in iw or w, allocate enough and try again
1095  if (info[0] == -4)
1096  {
1097  if (Doc_stats)
1098  oomph_info << " Reallocating liw to " << info[4] << std::endl;
1099  delete[] iw;
1100  liw = info[4];
1101  iw = new int[liw];
1102  }
1103 
1104  // If not enough space in w, allocate enough and try again
1105  if (info[0] == -2)
1106  {
1107  if (Doc_stats)
1108  oomph_info << " Reallocating lw to " << info[5] << std::endl;
1109  delete[] w;
1110  lw = info[5];
1111  w = new double[lw];
1112  }
1113 
1114  } while (info[0] < 0);
1115 
1116 
1117  if (Doc_stats)
1118  {
1119  oomph_info << "\n Initial wavefront details :\n max " << rinfo[0]
1120  << "\tmean " << rinfo[1] << "\tprofile " << rinfo[2];
1121  oomph_info << "\n Reordered wavefront details:\n max " << rinfo[3]
1122  << "\tmean " << rinfo[4] << "\tprofile " << rinfo[5];
1123  oomph_info << std::endl;
1124  }
1125 
1126  // Free the memory allocated
1127  delete[] w;
1128  delete[] iw;
1129 
1130 
1131  // Store re-ordered elements in temporary vector
1132  Vector<GeneralisedElement*> tmp_element_pt(n_el);
1133  for (unsigned e = 0; e < n_el; e++)
1134  {
1135  tmp_element_pt[e] = problem_pt->mesh_pt()->element_pt(order[e] - 1);
1136  }
1137  for (unsigned e = 0; e < n_el; e++)
1138  {
1139  problem_pt->mesh_pt()->element_pt(e) = tmp_element_pt[e];
1140  }
1141  }
1142 
1143 } // namespace oomph
e
Definition: cfortran.h:571
cstr elem_len * i
Definition: cfortran.h:603
A class that is used to define the functions used to assemble the elemental contributions to the resi...
virtual unsigned ndof(GeneralisedElement *const &elem_pt)
Return the number of degrees of freedom in the element elem_pt.
virtual unsigned long eqn_number(GeneralisedElement *const &elem_pt, const unsigned &ieqn_local)
Return the global equation number of the local unknown ieqn_local in elem_pt.
virtual void get_jacobian(GeneralisedElement *const &elem_pt, Vector< double > &residuals, DenseMatrix< double > &jacobian)
Calculate the elemental Jacobian matrix "d equation / d variable" for elem_pt.
void resize(const unsigned long &n)
Resize to a square nxn matrix; any values already present will be transfered.
Definition: matrices.h:498
LinearAlgebraDistribution * distribution_pt() const
access to the LinearAlgebraDistribution
unsigned nrow() const
access function to the number of global rows.
void build_distribution(const LinearAlgebraDistribution *const dist_pt)
setup the distribution of this distributable linear algebra object
A vector in the mathematical sense, initially developed for linear algebra type applications....
Definition: double_vector.h:58
void build(const DoubleVector &old_vector)
Just copys the argument DoubleVector.
A Generalised Element class.
Definition: elements.h:73
int Lw
Size of the workspace array, W.
bool Doc_stats
Doc the solver stats or stay quiet?
double Lenfle_factor
Factor to increase size of direct access files; see MA42 documentation for details.
bool Use_direct_access_files
Use direct access files?
bool Reorder_flag
Reorder elements with Sloan's algorithm?
double Lenbuf_factor2
Factor to increase storage for lenbuf[2]; see MA42 documentation for details.
double Lenbuf_factor1
Factor to increase storage for lenbuf[1]; see MA42 documentation for details.
int Icntl[8]
Control flag for MA42; see MA42 documentation for details.
int Isave[45]
Control flag for MA42; see MA42 documentation for details.
int Liw
Size of the integer workspace array.
void reorder_elements(Problem *const &problem_pt)
Function to reorder the elements based on Sloan's algorithm.
void solve_for_one_dof(Problem *const &problem_pt, DoubleVector &result)
Special solver for problems with 1 dof (MA42 can't handle this case so solve() forwards the "solve" t...
double Front_factor
Factor to increase storage for front size; see MA42 documentation for details.
int Info[23]
Control flag for MA42; see MA42 documentation for details.
double * W
Workspace storage for MA42.
void resolve(const DoubleVector &rhs, DoubleVector &result)
Return the solution to the linear system Ax = result, where A is the most recently factorised jacobia...
double Lenbuf_factor0
Factor to increase storage for lenbuf[0]; see MA42 documentation for details.
void solve(Problem *const &problem_pt, DoubleVector &result)
Solver: Takes pointer to problem and returns the results Vector which contains the solution of the li...
int * IW
Integer workspace storage for MA42.
unsigned long N_dof
Size of the linear system.
Describes the distribution of a distributable linear algebra type object. Typically this is a contain...
bool Enable_resolve
Boolean that indicates whether the matrix (or its factors, in the case of direct solver) should be st...
Definition: linear_solver.h:73
GeneralisedElement *& element_pt(const unsigned long &e)
Return pointer to element e.
Definition: mesh.h:448
unsigned long nelement() const
Return number of elements in the mesh.
Definition: mesh.h:590
A class to handle errors in the Newton solver.
Definition: problem.h:3057
An OomphLibError object which should be thrown when an run-time error is encountered....
////////////////////////////////////////////////////////////////// //////////////////////////////////...
Definition: problem.h:154
unsigned long ndof() const
Return the number of dofs.
Definition: problem.h:1704
AssemblyHandler *& assembly_handler_pt()
Return a pointer to the assembly handler object.
Definition: problem.h:1590
OomphCommunicator * communicator_pt()
access function to the oomph-lib communicator
Definition: problem.h:1266
int & sign_of_jacobian()
Access function for the sign of the global jacobian matrix. This will be set by the linear solver,...
Definition: problem.h:2529
Mesh *& mesh_pt()
Return a pointer to the global mesh.
Definition: problem.h:1300
//////////////////////////////////////////////////////////////////// ////////////////////////////////...
OomphInfo oomph_info
Single (global) instantiation of the OomphInfo object – this is used throughout the library as a "rep...