xref: /petsc/src/mat/impls/aij/seq/superlu/superlu.c (revision 2d30e087755efd99e28fdfe792ffbeb2ee1ea928)
1 
2 /*  --------------------------------------------------------------------
3 
4      This file implements a subclass of the SeqAIJ matrix class that uses
5      the SuperLU sparse solver.
6 */
7 
8 /*
9      Defines the data structure for the base matrix type (SeqAIJ)
10 */
11 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/
12 
13 /*
14      SuperLU include files
15 */
16 EXTERN_C_BEGIN
17 #if defined(PETSC_USE_COMPLEX)
18 #if defined(PETSC_USE_REAL_SINGLE)
19 #include <slu_cdefs.h>
20 #else
21 #include <slu_zdefs.h>
22 #endif
23 #else
24 #if defined(PETSC_USE_REAL_SINGLE)
25 #include <slu_sdefs.h>
26 #else
27 #include <slu_ddefs.h>
28 #endif
29 #endif
30 #include <slu_util.h>
31 EXTERN_C_END
32 
33 /*
34      This is the data that defines the SuperLU factored matrix type
35 */
36 typedef struct {
37   SuperMatrix       A, L, U, B, X;
38   superlu_options_t options;
39   PetscInt         *perm_c; /* column permutation vector */
40   PetscInt         *perm_r; /* row permutations from partial pivoting */
41   PetscInt         *etree;
42   PetscReal        *R, *C;
43   char              equed[1];
44   PetscInt          lwork;
45   void             *work;
46   PetscReal         rpg, rcond;
47   mem_usage_t       mem_usage;
48   MatStructure      flg;
49   SuperLUStat_t     stat;
50   Mat               A_dup;
51   PetscScalar      *rhs_dup;
52   GlobalLU_t        Glu;
53   PetscBool         needconversion;
54 
55   /* Flag to clean up (non-global) SuperLU objects during Destroy */
56   PetscBool CleanUpSuperLU;
57 } Mat_SuperLU;
58 
59 /*
60     Utility function
61 */
62 static PetscErrorCode MatView_Info_SuperLU(Mat A, PetscViewer viewer) {
63   Mat_SuperLU      *lu = (Mat_SuperLU *)A->data;
64   superlu_options_t options;
65 
66   PetscFunctionBegin;
67   options = lu->options;
68 
69   PetscCall(PetscViewerASCIIPrintf(viewer, "SuperLU run parameters:\n"));
70   PetscCall(PetscViewerASCIIPrintf(viewer, "  Equil: %s\n", (options.Equil != NO) ? "YES" : "NO"));
71   PetscCall(PetscViewerASCIIPrintf(viewer, "  ColPerm: %" PetscInt_FMT "\n", options.ColPerm));
72   PetscCall(PetscViewerASCIIPrintf(viewer, "  IterRefine: %" PetscInt_FMT "\n", options.IterRefine));
73   PetscCall(PetscViewerASCIIPrintf(viewer, "  SymmetricMode: %s\n", (options.SymmetricMode != NO) ? "YES" : "NO"));
74   PetscCall(PetscViewerASCIIPrintf(viewer, "  DiagPivotThresh: %g\n", options.DiagPivotThresh));
75   PetscCall(PetscViewerASCIIPrintf(viewer, "  PivotGrowth: %s\n", (options.PivotGrowth != NO) ? "YES" : "NO"));
76   PetscCall(PetscViewerASCIIPrintf(viewer, "  ConditionNumber: %s\n", (options.ConditionNumber != NO) ? "YES" : "NO"));
77   PetscCall(PetscViewerASCIIPrintf(viewer, "  RowPerm: %" PetscInt_FMT "\n", options.RowPerm));
78   PetscCall(PetscViewerASCIIPrintf(viewer, "  ReplaceTinyPivot: %s\n", (options.ReplaceTinyPivot != NO) ? "YES" : "NO"));
79   PetscCall(PetscViewerASCIIPrintf(viewer, "  PrintStat: %s\n", (options.PrintStat != NO) ? "YES" : "NO"));
80   PetscCall(PetscViewerASCIIPrintf(viewer, "  lwork: %" PetscInt_FMT "\n", lu->lwork));
81   if (A->factortype == MAT_FACTOR_ILU) {
82     PetscCall(PetscViewerASCIIPrintf(viewer, "  ILU_DropTol: %g\n", options.ILU_DropTol));
83     PetscCall(PetscViewerASCIIPrintf(viewer, "  ILU_FillTol: %g\n", options.ILU_FillTol));
84     PetscCall(PetscViewerASCIIPrintf(viewer, "  ILU_FillFactor: %g\n", options.ILU_FillFactor));
85     PetscCall(PetscViewerASCIIPrintf(viewer, "  ILU_DropRule: %" PetscInt_FMT "\n", options.ILU_DropRule));
86     PetscCall(PetscViewerASCIIPrintf(viewer, "  ILU_Norm: %" PetscInt_FMT "\n", options.ILU_Norm));
87     PetscCall(PetscViewerASCIIPrintf(viewer, "  ILU_MILU: %" PetscInt_FMT "\n", options.ILU_MILU));
88   }
89   PetscFunctionReturn(0);
90 }
91 
92 PetscErrorCode MatSolve_SuperLU_Private(Mat A, Vec b, Vec x) {
93   Mat_SuperLU       *lu = (Mat_SuperLU *)A->data;
94   const PetscScalar *barray;
95   PetscScalar       *xarray;
96   PetscInt           info, i, n;
97   PetscReal          ferr, berr;
98   static PetscBool   cite = PETSC_FALSE;
99 
100   PetscFunctionBegin;
101   if (lu->lwork == -1) PetscFunctionReturn(0);
102   PetscCall(PetscCitationsRegister("@article{superlu99,\n  author  = {James W. Demmel and Stanley C. Eisenstat and\n             John R. Gilbert and Xiaoye S. Li and Joseph W. H. Liu},\n  title = {A supernodal approach to sparse partial "
103                                    "pivoting},\n  journal = {SIAM J. Matrix Analysis and Applications},\n  year = {1999},\n  volume  = {20},\n  number = {3},\n  pages = {720-755}\n}\n",
104                                    &cite));
105 
106   PetscCall(VecGetLocalSize(x, &n));
107   lu->B.ncol = 1; /* Set the number of right-hand side */
108   if (lu->options.Equil && !lu->rhs_dup) {
109     /* superlu overwrites b when Equil is used, thus create rhs_dup to keep user's b unchanged */
110     PetscCall(PetscMalloc1(n, &lu->rhs_dup));
111   }
112   if (lu->options.Equil) {
113     /* Copy b into rsh_dup */
114     PetscCall(VecGetArrayRead(b, &barray));
115     PetscCall(PetscArraycpy(lu->rhs_dup, barray, n));
116     PetscCall(VecRestoreArrayRead(b, &barray));
117     barray = lu->rhs_dup;
118   } else {
119     PetscCall(VecGetArrayRead(b, &barray));
120   }
121   PetscCall(VecGetArray(x, &xarray));
122 
123 #if defined(PETSC_USE_COMPLEX)
124 #if defined(PETSC_USE_REAL_SINGLE)
125   ((DNformat *)lu->B.Store)->nzval = (singlecomplex *)barray;
126   ((DNformat *)lu->X.Store)->nzval = (singlecomplex *)xarray;
127 #else
128   ((DNformat *)lu->B.Store)->nzval = (doublecomplex *)barray;
129   ((DNformat *)lu->X.Store)->nzval = (doublecomplex *)xarray;
130 #endif
131 #else
132   ((DNformat *)lu->B.Store)->nzval = (void *)barray;
133   ((DNformat *)lu->X.Store)->nzval = xarray;
134 #endif
135 
136   lu->options.Fact = FACTORED; /* Indicate the factored form of A is supplied. */
137   if (A->factortype == MAT_FACTOR_LU) {
138 #if defined(PETSC_USE_COMPLEX)
139 #if defined(PETSC_USE_REAL_SINGLE)
140     PetscStackCallExternalVoid("SuperLU:cgssvx", cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
141 #else
142     PetscStackCallExternalVoid("SuperLU:zgssvx", zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
143 #endif
144 #else
145 #if defined(PETSC_USE_REAL_SINGLE)
146     PetscStackCallExternalVoid("SuperLU:sgssvx", sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
147 #else
148     PetscStackCallExternalVoid("SuperLU:dgssvx", dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
149 #endif
150 #endif
151   } else if (A->factortype == MAT_FACTOR_ILU) {
152 #if defined(PETSC_USE_COMPLEX)
153 #if defined(PETSC_USE_REAL_SINGLE)
154     PetscStackCallExternalVoid("SuperLU:cgsisx", cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
155 #else
156     PetscStackCallExternalVoid("SuperLU:zgsisx", zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
157 #endif
158 #else
159 #if defined(PETSC_USE_REAL_SINGLE)
160     PetscStackCallExternalVoid("SuperLU:sgsisx", sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
161 #else
162     PetscStackCallExternalVoid("SuperLU:dgsisx", dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info));
163 #endif
164 #endif
165   } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported");
166   if (!lu->options.Equil) PetscCall(VecRestoreArrayRead(b, &barray));
167   PetscCall(VecRestoreArray(x, &xarray));
168 
169   if (!info || info == lu->A.ncol + 1) {
170     if (lu->options.IterRefine) {
171       PetscCall(PetscPrintf(PETSC_COMM_SELF, "Iterative Refinement:\n"));
172       PetscCall(PetscPrintf(PETSC_COMM_SELF, "  %8s%8s%16s%16s\n", "rhs", "Steps", "FERR", "BERR"));
173       for (i = 0; i < 1; ++i) PetscCall(PetscPrintf(PETSC_COMM_SELF, "  %8d%8d%16e%16e\n", i + 1, lu->stat.RefineSteps, ferr, berr));
174     }
175   } else if (info > 0) {
176     if (lu->lwork == -1) {
177       PetscCall(PetscPrintf(PETSC_COMM_SELF, "  ** Estimated memory: %" PetscInt_FMT " bytes\n", info - lu->A.ncol));
178     } else {
179       PetscCall(PetscPrintf(PETSC_COMM_SELF, "  Warning: gssvx() returns info %" PetscInt_FMT "\n", info));
180     }
181   } else PetscCheck(info >= 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "info = %" PetscInt_FMT ", the %" PetscInt_FMT "-th argument in gssvx() had an illegal value", info, -info);
182 
183   if (lu->options.PrintStat) {
184     PetscCall(PetscPrintf(PETSC_COMM_SELF, "MatSolve__SuperLU():\n"));
185     PetscStackCallExternalVoid("SuperLU:StatPrint", StatPrint(&lu->stat));
186   }
187   PetscFunctionReturn(0);
188 }
189 
190 PetscErrorCode MatSolve_SuperLU(Mat A, Vec b, Vec x) {
191   Mat_SuperLU *lu = (Mat_SuperLU *)A->data;
192 
193   PetscFunctionBegin;
194   if (A->factorerrortype) {
195     PetscCall(PetscInfo(A, "MatSolve is called with singular matrix factor, skip\n"));
196     PetscCall(VecSetInf(x));
197     PetscFunctionReturn(0);
198   }
199 
200   lu->options.Trans = TRANS;
201   PetscCall(MatSolve_SuperLU_Private(A, b, x));
202   PetscFunctionReturn(0);
203 }
204 
205 PetscErrorCode MatSolveTranspose_SuperLU(Mat A, Vec b, Vec x) {
206   Mat_SuperLU *lu = (Mat_SuperLU *)A->data;
207 
208   PetscFunctionBegin;
209   if (A->factorerrortype) {
210     PetscCall(PetscInfo(A, "MatSolve is called with singular matrix factor, skip\n"));
211     PetscCall(VecSetInf(x));
212     PetscFunctionReturn(0);
213   }
214 
215   lu->options.Trans = NOTRANS;
216   PetscCall(MatSolve_SuperLU_Private(A, b, x));
217   PetscFunctionReturn(0);
218 }
219 
220 static PetscErrorCode MatLUFactorNumeric_SuperLU(Mat F, Mat A, const MatFactorInfo *info) {
221   Mat_SuperLU *lu = (Mat_SuperLU *)F->data;
222   Mat_SeqAIJ  *aa;
223   PetscInt     sinfo;
224   PetscReal    ferr, berr;
225   NCformat    *Ustore;
226   SCformat    *Lstore;
227 
228   PetscFunctionBegin;
229   if (lu->flg == SAME_NONZERO_PATTERN) { /* successing numerical factorization */
230     lu->options.Fact = SamePattern;
231     /* Ref: ~SuperLU_3.0/EXAMPLE/dlinsolx2.c */
232     Destroy_SuperMatrix_Store(&lu->A);
233     if (lu->A_dup) PetscCall(MatCopy_SeqAIJ(A, lu->A_dup, SAME_NONZERO_PATTERN));
234 
235     if (lu->lwork >= 0) {
236       PetscStackCallExternalVoid("SuperLU:Destroy_SuperNode_Matrix", Destroy_SuperNode_Matrix(&lu->L));
237       PetscStackCallExternalVoid("SuperLU:Destroy_CompCol_Matrix", Destroy_CompCol_Matrix(&lu->U));
238       lu->options.Fact = SamePattern;
239     }
240   }
241 
242   /* Create the SuperMatrix for lu->A=A^T:
243        Since SuperLU likes column-oriented matrices,we pass it the transpose,
244        and then solve A^T X = B in MatSolve(). */
245   if (lu->A_dup) {
246     aa = (Mat_SeqAIJ *)(lu->A_dup)->data;
247   } else {
248     aa = (Mat_SeqAIJ *)(A)->data;
249   }
250 #if defined(PETSC_USE_COMPLEX)
251 #if defined(PETSC_USE_REAL_SINGLE)
252   PetscStackCallExternalVoid("SuperLU:cCreate_CompCol_Matrix", cCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, (singlecomplex *)aa->a, aa->j, aa->i, SLU_NC, SLU_C, SLU_GE));
253 #else
254   PetscStackCallExternalVoid("SuperLU:zCreate_CompCol_Matrix", zCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, (doublecomplex *)aa->a, aa->j, aa->i, SLU_NC, SLU_Z, SLU_GE));
255 #endif
256 #else
257 #if defined(PETSC_USE_REAL_SINGLE)
258   PetscStackCallExternalVoid("SuperLU:sCreate_CompCol_Matrix", sCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, aa->a, aa->j, aa->i, SLU_NC, SLU_S, SLU_GE));
259 #else
260   PetscStackCallExternalVoid("SuperLU:dCreate_CompCol_Matrix", dCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, aa->a, aa->j, aa->i, SLU_NC, SLU_D, SLU_GE));
261 #endif
262 #endif
263 
264   /* Numerical factorization */
265   lu->B.ncol = 0; /* Indicate not to solve the system */
266   if (F->factortype == MAT_FACTOR_LU) {
267 #if defined(PETSC_USE_COMPLEX)
268 #if defined(PETSC_USE_REAL_SINGLE)
269     PetscStackCallExternalVoid("SuperLU:cgssvx", cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
270 #else
271     PetscStackCallExternalVoid("SuperLU:zgssvx", zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
272 #endif
273 #else
274 #if defined(PETSC_USE_REAL_SINGLE)
275     PetscStackCallExternalVoid("SuperLU:sgssvx", sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
276 #else
277     PetscStackCallExternalVoid("SuperLU:dgssvx", dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
278 #endif
279 #endif
280   } else if (F->factortype == MAT_FACTOR_ILU) {
281     /* Compute the incomplete factorization, condition number and pivot growth */
282 #if defined(PETSC_USE_COMPLEX)
283 #if defined(PETSC_USE_REAL_SINGLE)
284     PetscStackCallExternalVoid("SuperLU:cgsisx", cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
285 #else
286     PetscStackCallExternalVoid("SuperLU:zgsisx", zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
287 #endif
288 #else
289 #if defined(PETSC_USE_REAL_SINGLE)
290     PetscStackCallExternalVoid("SuperLU:sgsisx", sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
291 #else
292     PetscStackCallExternalVoid("SuperLU:dgsisx", dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
293 #endif
294 #endif
295   } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported");
296   if (!sinfo || sinfo == lu->A.ncol + 1) {
297     if (lu->options.PivotGrowth) PetscCall(PetscPrintf(PETSC_COMM_SELF, "  Recip. pivot growth = %e\n", lu->rpg));
298     if (lu->options.ConditionNumber) PetscCall(PetscPrintf(PETSC_COMM_SELF, "  Recip. condition number = %e\n", lu->rcond));
299   } else if (sinfo > 0) {
300     if (A->erroriffailure) {
301       SETERRQ(PETSC_COMM_SELF, PETSC_ERR_MAT_LU_ZRPVT, "Zero pivot in row %" PetscInt_FMT, sinfo);
302     } else {
303       if (sinfo <= lu->A.ncol) {
304         if (lu->options.ILU_FillTol == 0.0) F->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
305         PetscCall(PetscInfo(F, "Number of zero pivots %" PetscInt_FMT ", ILU_FillTol %g\n", sinfo, lu->options.ILU_FillTol));
306       } else if (sinfo == lu->A.ncol + 1) {
307         /*
308          U is nonsingular, but RCOND is less than machine
309                       precision, meaning that the matrix is singular to
310                       working precision. Nevertheless, the solution and
311                       error bounds are computed because there are a number
312                       of situations where the computed solution can be more
313                       accurate than the value of RCOND would suggest.
314          */
315         PetscCall(PetscInfo(F, "Matrix factor U is nonsingular, but is singular to working precision. The solution is computed. info %" PetscInt_FMT, sinfo));
316       } else { /* sinfo > lu->A.ncol + 1 */
317         F->factorerrortype = MAT_FACTOR_OUTMEMORY;
318         PetscCall(PetscInfo(F, "Number of bytes allocated when memory allocation fails %" PetscInt_FMT "\n", sinfo));
319       }
320     }
321   } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "info = %" PetscInt_FMT ", the %" PetscInt_FMT "-th argument in gssvx() had an illegal value", sinfo, -sinfo);
322 
323   if (lu->options.PrintStat) {
324     PetscCall(PetscPrintf(PETSC_COMM_SELF, "MatLUFactorNumeric_SuperLU():\n"));
325     PetscStackCallExternalVoid("SuperLU:StatPrint", StatPrint(&lu->stat));
326     Lstore = (SCformat *)lu->L.Store;
327     Ustore = (NCformat *)lu->U.Store;
328     PetscCall(PetscPrintf(PETSC_COMM_SELF, "  No of nonzeros in factor L = %" PetscInt_FMT "\n", Lstore->nnz));
329     PetscCall(PetscPrintf(PETSC_COMM_SELF, "  No of nonzeros in factor U = %" PetscInt_FMT "\n", Ustore->nnz));
330     PetscCall(PetscPrintf(PETSC_COMM_SELF, "  No of nonzeros in L+U = %" PetscInt_FMT "\n", Lstore->nnz + Ustore->nnz - lu->A.ncol));
331     PetscCall(PetscPrintf(PETSC_COMM_SELF, "  L\\U MB %.3f\ttotal MB needed %.3f\n", lu->mem_usage.for_lu / 1e6, lu->mem_usage.total_needed / 1e6));
332   }
333 
334   lu->flg                = SAME_NONZERO_PATTERN;
335   F->ops->solve          = MatSolve_SuperLU;
336   F->ops->solvetranspose = MatSolveTranspose_SuperLU;
337   F->ops->matsolve       = NULL;
338   PetscFunctionReturn(0);
339 }
340 
341 static PetscErrorCode MatDestroy_SuperLU(Mat A) {
342   Mat_SuperLU *lu = (Mat_SuperLU *)A->data;
343 
344   PetscFunctionBegin;
345   if (lu->CleanUpSuperLU) { /* Free the SuperLU datastructures */
346     PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->A));
347     PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->B));
348     PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->X));
349     PetscStackCallExternalVoid("SuperLU:StatFree", StatFree(&lu->stat));
350     if (lu->lwork >= 0) {
351       PetscStackCallExternalVoid("SuperLU:Destroy_SuperNode_Matrix", Destroy_SuperNode_Matrix(&lu->L));
352       PetscStackCallExternalVoid("SuperLU:Destroy_CompCol_Matrix", Destroy_CompCol_Matrix(&lu->U));
353     }
354   }
355   PetscCall(PetscFree(lu->etree));
356   PetscCall(PetscFree(lu->perm_r));
357   PetscCall(PetscFree(lu->perm_c));
358   PetscCall(PetscFree(lu->R));
359   PetscCall(PetscFree(lu->C));
360   PetscCall(PetscFree(lu->rhs_dup));
361   PetscCall(MatDestroy(&lu->A_dup));
362   PetscCall(PetscFree(A->data));
363 
364   /* clear composed functions */
365   PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL));
366   PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSuperluSetILUDropTol_C", NULL));
367   PetscFunctionReturn(0);
368 }
369 
370 static PetscErrorCode MatView_SuperLU(Mat A, PetscViewer viewer) {
371   PetscBool         iascii;
372   PetscViewerFormat format;
373 
374   PetscFunctionBegin;
375   PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
376   if (iascii) {
377     PetscCall(PetscViewerGetFormat(viewer, &format));
378     if (format == PETSC_VIEWER_ASCII_INFO) PetscCall(MatView_Info_SuperLU(A, viewer));
379   }
380   PetscFunctionReturn(0);
381 }
382 
383 PetscErrorCode MatMatSolve_SuperLU(Mat A, Mat B, Mat X) {
384   Mat_SuperLU *lu = (Mat_SuperLU *)A->data;
385   PetscBool    flg;
386 
387   PetscFunctionBegin;
388   PetscCall(PetscObjectTypeCompareAny((PetscObject)B, &flg, MATSEQDENSE, MATMPIDENSE, NULL));
389   PetscCheck(flg, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Matrix B must be MATDENSE matrix");
390   PetscCall(PetscObjectTypeCompareAny((PetscObject)X, &flg, MATSEQDENSE, MATMPIDENSE, NULL));
391   PetscCheck(flg, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Matrix X must be MATDENSE matrix");
392   lu->options.Trans = TRANS;
393   SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "MatMatSolve_SuperLU() is not implemented yet");
394   PetscFunctionReturn(0);
395 }
396 
397 static PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat F, Mat A, IS r, IS c, const MatFactorInfo *info) {
398   Mat_SuperLU *lu = (Mat_SuperLU *)(F->data);
399   PetscInt     indx;
400   PetscBool    flg, set;
401   PetscReal    real_input;
402   const char  *colperm[]    = {"NATURAL", "MMD_ATA", "MMD_AT_PLUS_A", "COLAMD"}; /* MY_PERMC - not supported by the petsc interface yet */
403   const char  *iterrefine[] = {"NOREFINE", "SINGLE", "DOUBLE", "EXTRA"};
404   const char  *rowperm[]    = {"NOROWPERM", "LargeDiag"}; /* MY_PERMC - not supported by the petsc interface yet */
405 
406   PetscFunctionBegin;
407   /* Set options to F */
408   PetscOptionsBegin(PetscObjectComm((PetscObject)F), ((PetscObject)F)->prefix, "SuperLU Options", "Mat");
409   PetscCall(PetscOptionsBool("-mat_superlu_equil", "Equil", "None", (PetscBool)lu->options.Equil, (PetscBool *)&lu->options.Equil, NULL));
410   PetscCall(PetscOptionsEList("-mat_superlu_colperm", "ColPerm", "None", colperm, 4, colperm[3], &indx, &flg));
411   if (flg) lu->options.ColPerm = (colperm_t)indx;
412   PetscCall(PetscOptionsEList("-mat_superlu_iterrefine", "IterRefine", "None", iterrefine, 4, iterrefine[0], &indx, &flg));
413   if (flg) lu->options.IterRefine = (IterRefine_t)indx;
414   PetscCall(PetscOptionsBool("-mat_superlu_symmetricmode", "SymmetricMode", "None", (PetscBool)lu->options.SymmetricMode, &flg, &set));
415   if (set && flg) lu->options.SymmetricMode = YES;
416   PetscCall(PetscOptionsReal("-mat_superlu_diagpivotthresh", "DiagPivotThresh", "None", lu->options.DiagPivotThresh, &real_input, &flg));
417   if (flg) lu->options.DiagPivotThresh = (double)real_input;
418   PetscCall(PetscOptionsBool("-mat_superlu_pivotgrowth", "PivotGrowth", "None", (PetscBool)lu->options.PivotGrowth, &flg, &set));
419   if (set && flg) lu->options.PivotGrowth = YES;
420   PetscCall(PetscOptionsBool("-mat_superlu_conditionnumber", "ConditionNumber", "None", (PetscBool)lu->options.ConditionNumber, &flg, &set));
421   if (set && flg) lu->options.ConditionNumber = YES;
422   PetscCall(PetscOptionsEList("-mat_superlu_rowperm", "rowperm", "None", rowperm, 2, rowperm[lu->options.RowPerm], &indx, &flg));
423   if (flg) lu->options.RowPerm = (rowperm_t)indx;
424   PetscCall(PetscOptionsBool("-mat_superlu_replacetinypivot", "ReplaceTinyPivot", "None", (PetscBool)lu->options.ReplaceTinyPivot, &flg, &set));
425   if (set && flg) lu->options.ReplaceTinyPivot = YES;
426   PetscCall(PetscOptionsBool("-mat_superlu_printstat", "PrintStat", "None", (PetscBool)lu->options.PrintStat, &flg, &set));
427   if (set && flg) lu->options.PrintStat = YES;
428   PetscCall(PetscOptionsInt("-mat_superlu_lwork", "size of work array in bytes used by factorization", "None", lu->lwork, &lu->lwork, NULL));
429   if (lu->lwork > 0) {
430     /* lwork is in bytes, hence PetscMalloc() is used here, not PetscMalloc1()*/
431     PetscCall(PetscMalloc(lu->lwork, &lu->work));
432   } else if (lu->lwork != 0 && lu->lwork != -1) {
433     PetscCall(PetscPrintf(PETSC_COMM_SELF, "   Warning: lwork %" PetscInt_FMT " is not supported by SUPERLU. The default lwork=0 is used.\n", lu->lwork));
434     lu->lwork = 0;
435   }
436   /* ilu options */
437   PetscCall(PetscOptionsReal("-mat_superlu_ilu_droptol", "ILU_DropTol", "None", lu->options.ILU_DropTol, &real_input, &flg));
438   if (flg) lu->options.ILU_DropTol = (double)real_input;
439   PetscCall(PetscOptionsReal("-mat_superlu_ilu_filltol", "ILU_FillTol", "None", lu->options.ILU_FillTol, &real_input, &flg));
440   if (flg) lu->options.ILU_FillTol = (double)real_input;
441   PetscCall(PetscOptionsReal("-mat_superlu_ilu_fillfactor", "ILU_FillFactor", "None", lu->options.ILU_FillFactor, &real_input, &flg));
442   if (flg) lu->options.ILU_FillFactor = (double)real_input;
443   PetscCall(PetscOptionsInt("-mat_superlu_ilu_droprull", "ILU_DropRule", "None", lu->options.ILU_DropRule, &lu->options.ILU_DropRule, NULL));
444   PetscCall(PetscOptionsInt("-mat_superlu_ilu_norm", "ILU_Norm", "None", lu->options.ILU_Norm, &indx, &flg));
445   if (flg) lu->options.ILU_Norm = (norm_t)indx;
446   PetscCall(PetscOptionsInt("-mat_superlu_ilu_milu", "ILU_MILU", "None", lu->options.ILU_MILU, &indx, &flg));
447   if (flg) lu->options.ILU_MILU = (milu_t)indx;
448   PetscOptionsEnd();
449 
450   lu->flg                 = DIFFERENT_NONZERO_PATTERN;
451   lu->CleanUpSuperLU      = PETSC_TRUE;
452   F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU;
453 
454   /* if we are here, the nonzero pattern has changed unless the user explicitly called MatLUFactorSymbolic */
455   PetscCall(MatDestroy(&lu->A_dup));
456   if (lu->needconversion) PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &lu->A_dup));
457   if (lu->options.Equil == YES && !lu->A_dup) { /* superlu overwrites input matrix and rhs when Equil is used, thus create A_dup to keep user's A unchanged */
458     PetscCall(MatDuplicate_SeqAIJ(A, MAT_COPY_VALUES, &lu->A_dup));
459   }
460   PetscFunctionReturn(0);
461 }
462 
463 static PetscErrorCode MatSuperluSetILUDropTol_SuperLU(Mat F, PetscReal dtol) {
464   Mat_SuperLU *lu = (Mat_SuperLU *)F->data;
465 
466   PetscFunctionBegin;
467   lu->options.ILU_DropTol = dtol;
468   PetscFunctionReturn(0);
469 }
470 
471 /*@
472   MatSuperluSetILUDropTol - Set SuperLU ILU drop tolerance
473 
474    Logically Collective on F
475 
476    Input Parameters:
477 +  F - the factored matrix obtained by calling `MatGetFactor()` from PETSc-SuperLU interface
478 -  dtol - drop tolerance
479 
480   Options Database:
481 .   -mat_superlu_ilu_droptol <dtol> - the drop tolerance
482 
483    Level: beginner
484 
485    References:
486 .  * - SuperLU Users' Guide
487 
488 .seealso: `MatGetFactor()`
489 @*/
490 PetscErrorCode MatSuperluSetILUDropTol(Mat F, PetscReal dtol) {
491   PetscFunctionBegin;
492   PetscValidHeaderSpecific(F, MAT_CLASSID, 1);
493   PetscValidLogicalCollectiveReal(F, dtol, 2);
494   PetscTryMethod(F, "MatSuperluSetILUDropTol_C", (Mat, PetscReal), (F, dtol));
495   PetscFunctionReturn(0);
496 }
497 
498 PetscErrorCode MatFactorGetSolverType_seqaij_superlu(Mat A, MatSolverType *type) {
499   PetscFunctionBegin;
500   *type = MATSOLVERSUPERLU;
501   PetscFunctionReturn(0);
502 }
503 
504 /*MC
505   MATSOLVERSUPERLU = "superlu" - A solver package providing solvers LU and ILU for sequential matrices
506   via the external package SuperLU.
507 
508   Use ./configure --download-superlu to have PETSc installed with SuperLU
509 
510   Use -pc_type lu -pc_factor_mat_solver_type superlu to use this direct solver
511 
512   Options Database Keys:
513 + -mat_superlu_equil <FALSE>            - Equil (None)
514 . -mat_superlu_colperm <COLAMD>         - (choose one of) NATURAL MMD_ATA MMD_AT_PLUS_A COLAMD
515 . -mat_superlu_iterrefine <NOREFINE>    - (choose one of) NOREFINE SINGLE DOUBLE EXTRA
516 . -mat_superlu_symmetricmode: <FALSE>   - SymmetricMode (None)
517 . -mat_superlu_diagpivotthresh <1>      - DiagPivotThresh (None)
518 . -mat_superlu_pivotgrowth <FALSE>      - PivotGrowth (None)
519 . -mat_superlu_conditionnumber <FALSE>  - ConditionNumber (None)
520 . -mat_superlu_rowperm <NOROWPERM>      - (choose one of) NOROWPERM LargeDiag
521 . -mat_superlu_replacetinypivot <FALSE> - ReplaceTinyPivot (None)
522 . -mat_superlu_printstat <FALSE>        - PrintStat (None)
523 . -mat_superlu_lwork <0>                - size of work array in bytes used by factorization (None)
524 . -mat_superlu_ilu_droptol <0>          - ILU_DropTol (None)
525 . -mat_superlu_ilu_filltol <0>          - ILU_FillTol (None)
526 . -mat_superlu_ilu_fillfactor <0>       - ILU_FillFactor (None)
527 . -mat_superlu_ilu_droprull <0>         - ILU_DropRule (None)
528 . -mat_superlu_ilu_norm <0>             - ILU_Norm (None)
529 - -mat_superlu_ilu_milu <0>             - ILU_MILU (None)
530 
531    Notes:
532     Do not confuse this with `MATSOLVERSUPERLU_DIST` which is for parallel sparse solves
533 
534     Cannot use ordering provided by PETSc, provides its own.
535 
536    Level: beginner
537 
538 .seealso: `PCLU`, `PCILU`, `MATSOLVERSUPERLU_DIST`, `MATSOLVERMUMPS`, `PCFactorSetMatSolverType()`, `MatSolverType`
539 M*/
540 
541 static PetscErrorCode MatGetFactor_seqaij_superlu(Mat A, MatFactorType ftype, Mat *F) {
542   Mat          B;
543   Mat_SuperLU *lu;
544   PetscInt     m = A->rmap->n, n = A->cmap->n;
545 
546   PetscFunctionBegin;
547   PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
548   PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, PETSC_DETERMINE, PETSC_DETERMINE));
549   PetscCall(PetscStrallocpy("superlu", &((PetscObject)B)->type_name));
550   PetscCall(MatSetUp(B));
551   B->trivialsymbolic = PETSC_TRUE;
552   if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU) {
553     B->ops->lufactorsymbolic  = MatLUFactorSymbolic_SuperLU;
554     B->ops->ilufactorsymbolic = MatLUFactorSymbolic_SuperLU;
555   } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported");
556 
557   PetscCall(PetscFree(B->solvertype));
558   PetscCall(PetscStrallocpy(MATSOLVERSUPERLU, &B->solvertype));
559 
560   B->ops->getinfo = MatGetInfo_External;
561   B->ops->destroy = MatDestroy_SuperLU;
562   B->ops->view    = MatView_SuperLU;
563   B->factortype   = ftype;
564   B->assembled    = PETSC_TRUE; /* required by -ksp_view */
565   B->preallocated = PETSC_TRUE;
566 
567   PetscCall(PetscNewLog(B, &lu));
568 
569   if (ftype == MAT_FACTOR_LU) {
570     set_default_options(&lu->options);
571     /* Comments from SuperLU_4.0/SRC/dgssvx.c:
572       "Whether or not the system will be equilibrated depends on the
573        scaling of the matrix A, but if equilibration is used, A is
574        overwritten by diag(R)*A*diag(C) and B by diag(R)*B
575        (if options->Trans=NOTRANS) or diag(C)*B (if options->Trans = TRANS or CONJ)."
576      We set 'options.Equil = NO' as default because additional space is needed for it.
577     */
578     lu->options.Equil = NO;
579   } else if (ftype == MAT_FACTOR_ILU) {
580     /* Set the default input options of ilu: */
581     PetscStackCallExternalVoid("SuperLU:ilu_set_default_options", ilu_set_default_options(&lu->options));
582   }
583   lu->options.PrintStat = NO;
584 
585   /* Initialize the statistics variables. */
586   PetscStackCallExternalVoid("SuperLU:StatInit", StatInit(&lu->stat));
587   lu->lwork = 0; /* allocate space internally by system malloc */
588 
589   /* Allocate spaces (notice sizes are for the transpose) */
590   PetscCall(PetscMalloc1(m, &lu->etree));
591   PetscCall(PetscMalloc1(n, &lu->perm_r));
592   PetscCall(PetscMalloc1(m, &lu->perm_c));
593   PetscCall(PetscMalloc1(n, &lu->R));
594   PetscCall(PetscMalloc1(m, &lu->C));
595 
596   /* create rhs and solution x without allocate space for .Store */
597 #if defined(PETSC_USE_COMPLEX)
598 #if defined(PETSC_USE_REAL_SINGLE)
599   PetscStackCallExternalVoid("SuperLU:cCreate_Dense_Matrix(", cCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE));
600   PetscStackCallExternalVoid("SuperLU:cCreate_Dense_Matrix(", cCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE));
601 #else
602   PetscStackCallExternalVoid("SuperLU:zCreate_Dense_Matrix", zCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE));
603   PetscStackCallExternalVoid("SuperLU:zCreate_Dense_Matrix", zCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE));
604 #endif
605 #else
606 #if defined(PETSC_USE_REAL_SINGLE)
607   PetscStackCallExternalVoid("SuperLU:sCreate_Dense_Matrix", sCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE));
608   PetscStackCallExternalVoid("SuperLU:sCreate_Dense_Matrix", sCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE));
609 #else
610   PetscStackCallExternalVoid("SuperLU:dCreate_Dense_Matrix", dCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE));
611   PetscStackCallExternalVoid("SuperLU:dCreate_Dense_Matrix", dCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE));
612 #endif
613 #endif
614 
615   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_seqaij_superlu));
616   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSuperluSetILUDropTol_C", MatSuperluSetILUDropTol_SuperLU));
617   B->data = lu;
618 
619   *F = B;
620   PetscFunctionReturn(0);
621 }
622 
623 static PetscErrorCode MatGetFactor_seqsell_superlu(Mat A, MatFactorType ftype, Mat *F) {
624   Mat_SuperLU *lu;
625 
626   PetscFunctionBegin;
627   PetscCall(MatGetFactor_seqaij_superlu(A, ftype, F));
628   lu                 = (Mat_SuperLU *)((*F)->data);
629   lu->needconversion = PETSC_TRUE;
630   PetscFunctionReturn(0);
631 }
632 
633 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_SuperLU(void) {
634   PetscFunctionBegin;
635   PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_superlu));
636   PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQAIJ, MAT_FACTOR_ILU, MatGetFactor_seqaij_superlu));
637   PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQSELL, MAT_FACTOR_LU, MatGetFactor_seqsell_superlu));
638   PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQSELL, MAT_FACTOR_ILU, MatGetFactor_seqsell_superlu));
639   PetscFunctionReturn(0);
640 }
641