xref: /petsc/src/mat/impls/aij/seq/superlu/superlu.c (revision c3d89e0aef1fb8977d2f9d367fc4b91cc47aeb54)
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 we are "ADDING" to the SeqAIJ matrix type to get the SuperLU 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 
54   /* Flag to clean up (non-global) SuperLU objects during Destroy */
55   PetscBool CleanUpSuperLU;
56 } Mat_SuperLU;
57 
58 extern PetscErrorCode MatFactorInfo_SuperLU(Mat,PetscViewer);
59 extern PetscErrorCode MatLUFactorNumeric_SuperLU(Mat,Mat,const MatFactorInfo*);
60 extern PetscErrorCode MatDestroy_SuperLU(Mat);
61 extern PetscErrorCode MatView_SuperLU(Mat,PetscViewer);
62 extern PetscErrorCode MatAssemblyEnd_SuperLU(Mat,MatAssemblyType);
63 extern PetscErrorCode MatSolve_SuperLU(Mat,Vec,Vec);
64 extern PetscErrorCode MatMatSolve_SuperLU(Mat,Mat,Mat);
65 extern PetscErrorCode MatSolveTranspose_SuperLU(Mat,Vec,Vec);
66 extern PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat,Mat,IS,IS,const MatFactorInfo*);
67 extern PetscErrorCode MatDuplicate_SuperLU(Mat, MatDuplicateOption, Mat*);
68 
69 /*
70     Utility function
71 */
72 #undef __FUNCT__
73 #define __FUNCT__ "MatFactorInfo_SuperLU"
74 PetscErrorCode MatFactorInfo_SuperLU(Mat A,PetscViewer viewer)
75 {
76   Mat_SuperLU       *lu= (Mat_SuperLU*)A->spptr;
77   PetscErrorCode    ierr;
78   superlu_options_t options;
79 
80   PetscFunctionBegin;
81   /* check if matrix is superlu_dist type */
82   if (A->ops->solve != MatSolve_SuperLU) PetscFunctionReturn(0);
83 
84   options = lu->options;
85 
86   ierr = PetscViewerASCIIPrintf(viewer,"SuperLU run parameters:\n");CHKERRQ(ierr);
87   ierr = PetscViewerASCIIPrintf(viewer,"  Equil: %s\n",(options.Equil != NO) ? "YES" : "NO");CHKERRQ(ierr);
88   ierr = PetscViewerASCIIPrintf(viewer,"  ColPerm: %D\n",options.ColPerm);CHKERRQ(ierr);
89   ierr = PetscViewerASCIIPrintf(viewer,"  IterRefine: %D\n",options.IterRefine);CHKERRQ(ierr);
90   ierr = PetscViewerASCIIPrintf(viewer,"  SymmetricMode: %s\n",(options.SymmetricMode != NO) ? "YES" : "NO");CHKERRQ(ierr);
91   ierr = PetscViewerASCIIPrintf(viewer,"  DiagPivotThresh: %g\n",options.DiagPivotThresh);CHKERRQ(ierr);
92   ierr = PetscViewerASCIIPrintf(viewer,"  PivotGrowth: %s\n",(options.PivotGrowth != NO) ? "YES" : "NO");CHKERRQ(ierr);
93   ierr = PetscViewerASCIIPrintf(viewer,"  ConditionNumber: %s\n",(options.ConditionNumber != NO) ? "YES" : "NO");CHKERRQ(ierr);
94   ierr = PetscViewerASCIIPrintf(viewer,"  RowPerm: %D\n",options.RowPerm);CHKERRQ(ierr);
95   ierr = PetscViewerASCIIPrintf(viewer,"  ReplaceTinyPivot: %s\n",(options.ReplaceTinyPivot != NO) ? "YES" : "NO");CHKERRQ(ierr);
96   ierr = PetscViewerASCIIPrintf(viewer,"  PrintStat: %s\n",(options.PrintStat != NO) ? "YES" : "NO");CHKERRQ(ierr);
97   ierr = PetscViewerASCIIPrintf(viewer,"  lwork: %D\n",lu->lwork);CHKERRQ(ierr);
98   if (A->factortype == MAT_FACTOR_ILU) {
99     ierr = PetscViewerASCIIPrintf(viewer,"  ILU_DropTol: %g\n",options.ILU_DropTol);CHKERRQ(ierr);
100     ierr = PetscViewerASCIIPrintf(viewer,"  ILU_FillTol: %g\n",options.ILU_FillTol);CHKERRQ(ierr);
101     ierr = PetscViewerASCIIPrintf(viewer,"  ILU_FillFactor: %g\n",options.ILU_FillFactor);CHKERRQ(ierr);
102     ierr = PetscViewerASCIIPrintf(viewer,"  ILU_DropRule: %D\n",options.ILU_DropRule);CHKERRQ(ierr);
103     ierr = PetscViewerASCIIPrintf(viewer,"  ILU_Norm: %D\n",options.ILU_Norm);CHKERRQ(ierr);
104     ierr = PetscViewerASCIIPrintf(viewer,"  ILU_MILU: %D\n",options.ILU_MILU);CHKERRQ(ierr);
105   }
106   PetscFunctionReturn(0);
107 }
108 
109 /*
110     These are the methods provided to REPLACE the corresponding methods of the
111    SeqAIJ matrix class. Other methods of SeqAIJ are not replaced
112 */
113 #undef __FUNCT__
114 #define __FUNCT__ "MatLUFactorNumeric_SuperLU"
115 PetscErrorCode MatLUFactorNumeric_SuperLU(Mat F,Mat A,const MatFactorInfo *info)
116 {
117   Mat_SuperLU    *lu = (Mat_SuperLU*)F->spptr;
118   Mat_SeqAIJ     *aa;
119   PetscErrorCode ierr;
120   PetscInt       sinfo;
121   PetscReal      ferr, berr;
122   NCformat       *Ustore;
123   SCformat       *Lstore;
124 
125   PetscFunctionBegin;
126   if (lu->flg == SAME_NONZERO_PATTERN) { /* successing numerical factorization */
127     lu->options.Fact = SamePattern;
128     /* Ref: ~SuperLU_3.0/EXAMPLE/dlinsolx2.c */
129     Destroy_SuperMatrix_Store(&lu->A);
130     if (lu->options.Equil) {
131       ierr = MatCopy_SeqAIJ(A,lu->A_dup,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
132     }
133     if (lu->lwork >= 0) {
134       PetscStackCall("SuperLU:Destroy_SuperNode_Matrix",Destroy_SuperNode_Matrix(&lu->L));
135       PetscStackCall("SuperLU:Destroy_CompCol_Matrix",Destroy_CompCol_Matrix(&lu->U));
136       lu->options.Fact = SamePattern;
137     }
138   }
139 
140   /* Create the SuperMatrix for lu->A=A^T:
141        Since SuperLU likes column-oriented matrices,we pass it the transpose,
142        and then solve A^T X = B in MatSolve(). */
143   if (lu->options.Equil) {
144     aa = (Mat_SeqAIJ*)(lu->A_dup)->data;
145   } else {
146     aa = (Mat_SeqAIJ*)(A)->data;
147   }
148 #if defined(PETSC_USE_COMPLEX)
149 #if defined(PETSC_USE_REAL_SINGLE)
150   PetscStackCall("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));
151 #else
152   PetscStackCall("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));
153 #endif
154 #else
155 #if defined(PETSC_USE_REAL_SINGLE)
156   PetscStackCall("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));
157 #else
158   PetscStackCall("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));
159 #endif
160 #endif
161 
162   /* Numerical factorization */
163   lu->B.ncol = 0;  /* Indicate not to solve the system */
164   if (F->factortype == MAT_FACTOR_LU) {
165 #if defined(PETSC_USE_COMPLEX)
166 #if defined(PETSC_USE_REAL_SINGLE)
167     PetscStackCall("SuperLU:cgssvx",cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
168                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
169                                      &lu->mem_usage, &lu->stat, &sinfo));
170 #else
171     PetscStackCall("SuperLU:zgssvx",zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
172                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
173                                      &lu->mem_usage, &lu->stat, &sinfo));
174 #endif
175 #else
176 #if defined(PETSC_USE_REAL_SINGLE)
177     PetscStackCall("SuperLU:sgssvx",sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
178                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
179                                      &lu->mem_usage, &lu->stat, &sinfo));
180 #else
181     PetscStackCall("SuperLU:dgssvx",dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
182                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
183                                      &lu->Glu,&lu->mem_usage, &lu->stat, &sinfo));
184 #endif
185 #endif
186   } else if (F->factortype == MAT_FACTOR_ILU) {
187     /* Compute the incomplete factorization, condition number and pivot growth */
188 #if defined(PETSC_USE_COMPLEX)
189 #if defined(PETSC_USE_REAL_SINGLE)
190     PetscStackCall("SuperLU:cgsisx",cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r,lu->etree, lu->equed, lu->R, lu->C,
191                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
192                                      &lu->mem_usage, &lu->stat, &sinfo));
193 #else
194     PetscStackCall("SuperLU:zgsisx",zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r,lu->etree, lu->equed, lu->R, lu->C,
195                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
196                                      &lu->mem_usage, &lu->stat, &sinfo));
197 #endif
198 #else
199 #if defined(PETSC_USE_REAL_SINGLE)
200     PetscStackCall("SuperLU:sgsisx",sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
201                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
202                                      &lu->mem_usage, &lu->stat, &sinfo));
203 #else
204     PetscStackCall("SuperLU:dgsisx",dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
205                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
206                                      &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo));
207 #endif
208 #endif
209   } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
210   if (!sinfo || sinfo == lu->A.ncol+1) {
211     if (lu->options.PivotGrowth) {
212       ierr = PetscPrintf(PETSC_COMM_SELF,"  Recip. pivot growth = %e\n", lu->rpg);
213     }
214     if (lu->options.ConditionNumber) {
215       ierr = PetscPrintf(PETSC_COMM_SELF,"  Recip. condition number = %e\n", lu->rcond);
216     }
217   } else if (sinfo > 0) {
218     if (lu->lwork == -1) {
219       ierr = PetscPrintf(PETSC_COMM_SELF,"  ** Estimated memory: %D bytes\n", sinfo - lu->A.ncol);
220     } else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot in row %D",sinfo);
221   } else SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %D, the %D-th argument in gssvx() had an illegal value", sinfo,-sinfo);
222 
223   if (lu->options.PrintStat) {
224     ierr = PetscPrintf(PETSC_COMM_SELF,"MatLUFactorNumeric_SuperLU():\n");
225     PetscStackCall("SuperLU:StatPrint",StatPrint(&lu->stat));
226     Lstore = (SCformat*) lu->L.Store;
227     Ustore = (NCformat*) lu->U.Store;
228     ierr   = PetscPrintf(PETSC_COMM_SELF,"  No of nonzeros in factor L = %D\n", Lstore->nnz);
229     ierr   = PetscPrintf(PETSC_COMM_SELF,"  No of nonzeros in factor U = %D\n", Ustore->nnz);
230     ierr   = PetscPrintf(PETSC_COMM_SELF,"  No of nonzeros in L+U = %D\n", Lstore->nnz + Ustore->nnz - lu->A.ncol);
231     ierr   = PetscPrintf(PETSC_COMM_SELF,"  L\\U MB %.3f\ttotal MB needed %.3f\n",
232                          lu->mem_usage.for_lu/1e6, lu->mem_usage.total_needed/1e6);
233   }
234 
235   lu->flg                = SAME_NONZERO_PATTERN;
236   F->ops->solve          = MatSolve_SuperLU;
237   F->ops->solvetranspose = MatSolveTranspose_SuperLU;
238   F->ops->matsolve       = NULL;
239   PetscFunctionReturn(0);
240 }
241 
242 #undef __FUNCT__
243 #define __FUNCT__ "MatGetDiagonal_SuperLU"
244 PetscErrorCode MatGetDiagonal_SuperLU(Mat A,Vec v)
245 {
246   PetscFunctionBegin;
247   SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Mat type: SuperLU factor");
248   PetscFunctionReturn(0);
249 }
250 
251 #undef __FUNCT__
252 #define __FUNCT__ "MatDestroy_SuperLU"
253 PetscErrorCode MatDestroy_SuperLU(Mat A)
254 {
255   PetscErrorCode ierr;
256   Mat_SuperLU    *lu=(Mat_SuperLU*)A->spptr;
257 
258   PetscFunctionBegin;
259   if (lu && lu->CleanUpSuperLU) { /* Free the SuperLU datastructures */
260     PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->A));
261     PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->B));
262     PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->X));
263     PetscStackCall("SuperLU:StatFree",StatFree(&lu->stat));
264     if (lu->lwork >= 0) {
265       PetscStackCall("SuperLU:Destroy_SuperNode_Matrix",Destroy_SuperNode_Matrix(&lu->L));
266       PetscStackCall("SuperLU:Destroy_CompCol_Matrix",Destroy_CompCol_Matrix(&lu->U));
267     }
268   }
269   if (lu) {
270     ierr = PetscFree(lu->etree);CHKERRQ(ierr);
271     ierr = PetscFree(lu->perm_r);CHKERRQ(ierr);
272     ierr = PetscFree(lu->perm_c);CHKERRQ(ierr);
273     ierr = PetscFree(lu->R);CHKERRQ(ierr);
274     ierr = PetscFree(lu->C);CHKERRQ(ierr);
275     ierr = PetscFree(lu->rhs_dup);CHKERRQ(ierr);
276     ierr = MatDestroy(&lu->A_dup);CHKERRQ(ierr);
277   }
278   ierr = PetscFree(A->spptr);CHKERRQ(ierr);
279 
280   /* clear composed functions */
281   ierr = PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverPackage_C",NULL);CHKERRQ(ierr);
282   ierr = PetscObjectComposeFunction((PetscObject)A,"MatSuperluSetILUDropTol_C",NULL);CHKERRQ(ierr);
283 
284   ierr = MatDestroy_SeqAIJ(A);CHKERRQ(ierr);
285   PetscFunctionReturn(0);
286 }
287 
288 #undef __FUNCT__
289 #define __FUNCT__ "MatView_SuperLU"
290 PetscErrorCode MatView_SuperLU(Mat A,PetscViewer viewer)
291 {
292   PetscErrorCode    ierr;
293   PetscBool         iascii;
294   PetscViewerFormat format;
295 
296   PetscFunctionBegin;
297   ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr);
298   if (iascii) {
299     ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr);
300     if (format == PETSC_VIEWER_ASCII_INFO) {
301       ierr = MatFactorInfo_SuperLU(A,viewer);CHKERRQ(ierr);
302     }
303   }
304   PetscFunctionReturn(0);
305 }
306 
307 
308 #undef __FUNCT__
309 #define __FUNCT__ "MatSolve_SuperLU_Private"
310 PetscErrorCode MatSolve_SuperLU_Private(Mat A,Vec b,Vec x)
311 {
312   Mat_SuperLU       *lu = (Mat_SuperLU*)A->spptr;
313   const PetscScalar *barray;
314   PetscScalar       *xarray;
315   PetscErrorCode    ierr;
316   PetscInt          info,i,n;
317   PetscReal         ferr,berr;
318   static PetscBool  cite = PETSC_FALSE;
319 
320   PetscFunctionBegin;
321   if (lu->lwork == -1) PetscFunctionReturn(0);
322   ierr = 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 pivoting},\n  journal = {SIAM J. Matrix Analysis and Applications},\n  year = {1999},\n  volume  = {20},\n  number = {3},\n  pages = {720-755}\n}\n",&cite);CHKERRQ(ierr);
323 
324   ierr = VecGetLocalSize(x,&n);CHKERRQ(ierr);
325   lu->B.ncol = 1;   /* Set the number of right-hand side */
326   if (lu->options.Equil && !lu->rhs_dup) {
327     /* superlu overwrites b when Equil is used, thus create rhs_dup to keep user's b unchanged */
328     ierr = PetscMalloc1(n,&lu->rhs_dup);CHKERRQ(ierr);
329   }
330   if (lu->options.Equil) {
331     /* Copy b into rsh_dup */
332     ierr   = VecGetArrayRead(b,&barray);CHKERRQ(ierr);
333     ierr   = PetscMemcpy(lu->rhs_dup,barray,n*sizeof(PetscScalar));CHKERRQ(ierr);
334     ierr   = VecRestoreArrayRead(b,&barray);CHKERRQ(ierr);
335     barray = lu->rhs_dup;
336   } else {
337     ierr = VecGetArrayRead(b,&barray);CHKERRQ(ierr);
338   }
339   ierr = VecGetArray(x,&xarray);CHKERRQ(ierr);
340 
341 #if defined(PETSC_USE_COMPLEX)
342 #if defined(PETSC_USE_REAL_SINGLE)
343   ((DNformat*)lu->B.Store)->nzval = (singlecomplex*)barray;
344   ((DNformat*)lu->X.Store)->nzval = (singlecomplex*)xarray;
345 #else
346   ((DNformat*)lu->B.Store)->nzval = (doublecomplex*)barray;
347   ((DNformat*)lu->X.Store)->nzval = (doublecomplex*)xarray;
348 #endif
349 #else
350   ((DNformat*)lu->B.Store)->nzval = (void*)barray;
351   ((DNformat*)lu->X.Store)->nzval = xarray;
352 #endif
353 
354   lu->options.Fact = FACTORED; /* Indicate the factored form of A is supplied. */
355   if (A->factortype == MAT_FACTOR_LU) {
356 #if defined(PETSC_USE_COMPLEX)
357 #if defined(PETSC_USE_REAL_SINGLE)
358     PetscStackCall("SuperLU:cgssvx",cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
359                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
360                                      &lu->mem_usage, &lu->stat, &info));
361 #else
362     PetscStackCall("SuperLU:zgssvx",zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
363                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
364                                      &lu->mem_usage, &lu->stat, &info));
365 #endif
366 #else
367 #if defined(PETSC_USE_REAL_SINGLE)
368     PetscStackCall("SuperLU:sgssvx",sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
369                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
370                                      &lu->mem_usage, &lu->stat, &info));
371 #else
372     PetscStackCall("SuperLU:dgssvx",dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
373                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
374                                      &lu->Glu,&lu->mem_usage, &lu->stat, &info));
375 #endif
376 #endif
377   } else if (A->factortype == MAT_FACTOR_ILU) {
378 #if defined(PETSC_USE_COMPLEX)
379 #if defined(PETSC_USE_REAL_SINGLE)
380     PetscStackCall("SuperLU:cgsisx",cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
381                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
382                                      &lu->mem_usage, &lu->stat, &info));
383 #else
384     PetscStackCall("SuperLU:zgsisx",zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
385                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
386                                      &lu->mem_usage, &lu->stat, &info));
387 #endif
388 #else
389 #if defined(PETSC_USE_REAL_SINGLE)
390     PetscStackCall("SuperLU:sgsisx",sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
391                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
392                                      &lu->mem_usage, &lu->stat, &info));
393 #else
394     PetscStackCall("SuperLU:dgsisx",dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
395                                      &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
396                                      &lu->Glu, &lu->mem_usage, &lu->stat, &info));
397 #endif
398 #endif
399   } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
400   if (!lu->options.Equil) {
401     ierr = VecRestoreArrayRead(b,&barray);CHKERRQ(ierr);
402   }
403   ierr = VecRestoreArray(x,&xarray);CHKERRQ(ierr);
404 
405   if (!info || info == lu->A.ncol+1) {
406     if (lu->options.IterRefine) {
407       ierr = PetscPrintf(PETSC_COMM_SELF,"Iterative Refinement:\n");
408       ierr = PetscPrintf(PETSC_COMM_SELF,"  %8s%8s%16s%16s\n", "rhs", "Steps", "FERR", "BERR");
409       for (i = 0; i < 1; ++i) {
410         ierr = PetscPrintf(PETSC_COMM_SELF,"  %8d%8d%16e%16e\n", i+1, lu->stat.RefineSteps, ferr, berr);
411       }
412     }
413   } else if (info > 0) {
414     if (lu->lwork == -1) {
415       ierr = PetscPrintf(PETSC_COMM_SELF,"  ** Estimated memory: %D bytes\n", info - lu->A.ncol);
416     } else {
417       ierr = PetscPrintf(PETSC_COMM_SELF,"  Warning: gssvx() returns info %D\n",info);
418     }
419   } else if (info < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %D, the %D-th argument in gssvx() had an illegal value", info,-info);
420 
421   if (lu->options.PrintStat) {
422     ierr = PetscPrintf(PETSC_COMM_SELF,"MatSolve__SuperLU():\n");
423     PetscStackCall("SuperLU:StatPrint",StatPrint(&lu->stat));
424   }
425   PetscFunctionReturn(0);
426 }
427 
428 #undef __FUNCT__
429 #define __FUNCT__ "MatSolve_SuperLU"
430 PetscErrorCode MatSolve_SuperLU(Mat A,Vec b,Vec x)
431 {
432   Mat_SuperLU    *lu = (Mat_SuperLU*)A->spptr;
433   PetscErrorCode ierr;
434 
435   PetscFunctionBegin;
436   lu->options.Trans = TRANS;
437 
438   ierr = MatSolve_SuperLU_Private(A,b,x);CHKERRQ(ierr);
439   PetscFunctionReturn(0);
440 }
441 
442 #undef __FUNCT__
443 #define __FUNCT__ "MatSolveTranspose_SuperLU"
444 PetscErrorCode MatSolveTranspose_SuperLU(Mat A,Vec b,Vec x)
445 {
446   Mat_SuperLU    *lu = (Mat_SuperLU*)A->spptr;
447   PetscErrorCode ierr;
448 
449   PetscFunctionBegin;
450   lu->options.Trans = NOTRANS;
451 
452   ierr = MatSolve_SuperLU_Private(A,b,x);CHKERRQ(ierr);
453   PetscFunctionReturn(0);
454 }
455 
456 #undef __FUNCT__
457 #define __FUNCT__ "MatMatSolve_SuperLU"
458 PetscErrorCode MatMatSolve_SuperLU(Mat A,Mat B,Mat X)
459 {
460   Mat_SuperLU    *lu = (Mat_SuperLU*)A->spptr;
461   PetscBool      flg;
462   PetscErrorCode ierr;
463 
464   PetscFunctionBegin;
465   ierr = PetscObjectTypeCompareAny((PetscObject)B,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr);
466   if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix");
467   ierr = PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr);
468   if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix");
469   lu->options.Trans = TRANS;
470   SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatSolve_SuperLU() is not implemented yet");
471   PetscFunctionReturn(0);
472 }
473 
474 /*
475    Note the r permutation is ignored
476 */
477 #undef __FUNCT__
478 #define __FUNCT__ "MatLUFactorSymbolic_SuperLU"
479 PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info)
480 {
481   Mat_SuperLU *lu = (Mat_SuperLU*)(F->spptr);
482 
483   PetscFunctionBegin;
484   lu->flg                 = DIFFERENT_NONZERO_PATTERN;
485   lu->CleanUpSuperLU      = PETSC_TRUE;
486   F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU;
487   PetscFunctionReturn(0);
488 }
489 
490 #undef __FUNCT__
491 #define __FUNCT__ "MatSuperluSetILUDropTol_SuperLU"
492 static PetscErrorCode MatSuperluSetILUDropTol_SuperLU(Mat F,PetscReal dtol)
493 {
494   Mat_SuperLU *lu= (Mat_SuperLU*)F->spptr;
495 
496   PetscFunctionBegin;
497   lu->options.ILU_DropTol = dtol;
498   PetscFunctionReturn(0);
499 }
500 
501 #undef __FUNCT__
502 #define __FUNCT__ "MatSuperluSetILUDropTol"
503 /*@
504   MatSuperluSetILUDropTol - Set SuperLU ILU drop tolerance
505    Logically Collective on Mat
506 
507    Input Parameters:
508 +  F - the factored matrix obtained by calling MatGetFactor() from PETSc-SuperLU interface
509 -  dtol - drop tolerance
510 
511   Options Database:
512 .   -mat_superlu_ilu_droptol <dtol>
513 
514    Level: beginner
515 
516    References: SuperLU Users' Guide
517 
518 .seealso: MatGetFactor()
519 @*/
520 PetscErrorCode MatSuperluSetILUDropTol(Mat F,PetscReal dtol)
521 {
522   PetscErrorCode ierr;
523 
524   PetscFunctionBegin;
525   PetscValidHeaderSpecific(F,MAT_CLASSID,1);
526   PetscValidLogicalCollectiveReal(F,dtol,2);
527   ierr = PetscTryMethod(F,"MatSuperluSetILUDropTol_C",(Mat,PetscReal),(F,dtol));CHKERRQ(ierr);
528   PetscFunctionReturn(0);
529 }
530 
531 #undef __FUNCT__
532 #define __FUNCT__ "MatFactorGetSolverPackage_seqaij_superlu"
533 PetscErrorCode MatFactorGetSolverPackage_seqaij_superlu(Mat A,const MatSolverPackage *type)
534 {
535   PetscFunctionBegin;
536   *type = MATSOLVERSUPERLU;
537   PetscFunctionReturn(0);
538 }
539 
540 
541 /*MC
542   MATSOLVERSUPERLU = "superlu" - A solver package providing solvers LU and ILU for sequential matrices
543   via the external package SuperLU.
544 
545   Use ./configure --download-superlu to have PETSc installed with SuperLU
546 
547   Use -pc_type lu -pc_factor_mat_solver_package superlu to us this direct solver
548 
549   Options Database Keys:
550 + -mat_superlu_equil <FALSE>            - Equil (None)
551 . -mat_superlu_colperm <COLAMD>         - (choose one of) NATURAL MMD_ATA MMD_AT_PLUS_A COLAMD
552 . -mat_superlu_iterrefine <NOREFINE>    - (choose one of) NOREFINE SINGLE DOUBLE EXTRA
553 . -mat_superlu_symmetricmode: <FALSE>   - SymmetricMode (None)
554 . -mat_superlu_diagpivotthresh <1>      - DiagPivotThresh (None)
555 . -mat_superlu_pivotgrowth <FALSE>      - PivotGrowth (None)
556 . -mat_superlu_conditionnumber <FALSE>  - ConditionNumber (None)
557 . -mat_superlu_rowperm <NOROWPERM>      - (choose one of) NOROWPERM LargeDiag
558 . -mat_superlu_replacetinypivot <FALSE> - ReplaceTinyPivot (None)
559 . -mat_superlu_printstat <FALSE>        - PrintStat (None)
560 . -mat_superlu_lwork <0>                - size of work array in bytes used by factorization (None)
561 . -mat_superlu_ilu_droptol <0>          - ILU_DropTol (None)
562 . -mat_superlu_ilu_filltol <0>          - ILU_FillTol (None)
563 . -mat_superlu_ilu_fillfactor <0>       - ILU_FillFactor (None)
564 . -mat_superlu_ilu_droprull <0>         - ILU_DropRule (None)
565 . -mat_superlu_ilu_norm <0>             - ILU_Norm (None)
566 - -mat_superlu_ilu_milu <0>             - ILU_MILU (None)
567 
568    Notes: Do not confuse this with MATSOLVERSUPERLU_DIST which is for parallel sparse solves
569 
570    Level: beginner
571 
572 .seealso: PCLU, PCILU, MATSOLVERSUPERLU_DIST, MATSOLVERMUMPS, PCFactorSetMatSolverPackage(), MatSolverPackage
573 M*/
574 
575 #undef __FUNCT__
576 #define __FUNCT__ "MatGetFactor_seqaij_superlu"
577 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat A,MatFactorType ftype,Mat *F)
578 {
579   Mat            B;
580   Mat_SuperLU    *lu;
581   PetscErrorCode ierr;
582   PetscInt       indx,m=A->rmap->n,n=A->cmap->n;
583   PetscBool      flg,set;
584   PetscReal      real_input;
585   const char     *colperm[]   ={"NATURAL","MMD_ATA","MMD_AT_PLUS_A","COLAMD"}; /* MY_PERMC - not supported by the petsc interface yet */
586   const char     *iterrefine[]={"NOREFINE", "SINGLE", "DOUBLE", "EXTRA"};
587   const char     *rowperm[]   ={"NOROWPERM", "LargeDiag"}; /* MY_PERMC - not supported by the petsc interface yet */
588 
589   PetscFunctionBegin;
590   ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr);
591   ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr);
592   ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr);
593   ierr = MatSeqAIJSetPreallocation(B,0,NULL);CHKERRQ(ierr);
594 
595   if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU) {
596     B->ops->lufactorsymbolic  = MatLUFactorSymbolic_SuperLU;
597     B->ops->ilufactorsymbolic = MatLUFactorSymbolic_SuperLU;
598   } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
599 
600   B->ops->destroy     = MatDestroy_SuperLU;
601   B->ops->view        = MatView_SuperLU;
602   B->ops->getdiagonal = MatGetDiagonal_SuperLU;
603   B->factortype       = ftype;
604   B->assembled        = PETSC_TRUE;           /* required by -ksp_view */
605   B->preallocated     = PETSC_TRUE;
606 
607   ierr = PetscNewLog(B,&lu);CHKERRQ(ierr);
608 
609   if (ftype == MAT_FACTOR_LU) {
610     set_default_options(&lu->options);
611     /* Comments from SuperLU_4.0/SRC/dgssvx.c:
612       "Whether or not the system will be equilibrated depends on the
613        scaling of the matrix A, but if equilibration is used, A is
614        overwritten by diag(R)*A*diag(C) and B by diag(R)*B
615        (if options->Trans=NOTRANS) or diag(C)*B (if options->Trans = TRANS or CONJ)."
616      We set 'options.Equil = NO' as default because additional space is needed for it.
617     */
618     lu->options.Equil = NO;
619   } else if (ftype == MAT_FACTOR_ILU) {
620     /* Set the default input options of ilu: */
621     PetscStackCall("SuperLU:ilu_set_default_options",ilu_set_default_options(&lu->options));
622   }
623   lu->options.PrintStat = NO;
624 
625   /* Initialize the statistics variables. */
626   PetscStackCall("SuperLU:StatInit",StatInit(&lu->stat));
627   lu->lwork = 0;   /* allocate space internally by system malloc */
628 
629   ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"SuperLU Options","Mat");CHKERRQ(ierr);
630   ierr = PetscOptionsBool("-mat_superlu_equil","Equil","None",(PetscBool)lu->options.Equil,(PetscBool*)&lu->options.Equil,NULL);CHKERRQ(ierr);
631   ierr = PetscOptionsEList("-mat_superlu_colperm","ColPerm","None",colperm,4,colperm[3],&indx,&flg);CHKERRQ(ierr);
632   if (flg) lu->options.ColPerm = (colperm_t)indx;
633   ierr = PetscOptionsEList("-mat_superlu_iterrefine","IterRefine","None",iterrefine,4,iterrefine[0],&indx,&flg);CHKERRQ(ierr);
634   if (flg) lu->options.IterRefine = (IterRefine_t)indx;
635   ierr = PetscOptionsBool("-mat_superlu_symmetricmode","SymmetricMode","None",(PetscBool)lu->options.SymmetricMode,&flg,&set);CHKERRQ(ierr);
636   if (set && flg) lu->options.SymmetricMode = YES;
637   ierr = PetscOptionsReal("-mat_superlu_diagpivotthresh","DiagPivotThresh","None",lu->options.DiagPivotThresh,&real_input,&flg);CHKERRQ(ierr);
638   if (flg) lu->options.DiagPivotThresh = (double) real_input;
639   ierr = PetscOptionsBool("-mat_superlu_pivotgrowth","PivotGrowth","None",(PetscBool)lu->options.PivotGrowth,&flg,&set);CHKERRQ(ierr);
640   if (set && flg) lu->options.PivotGrowth = YES;
641   ierr = PetscOptionsBool("-mat_superlu_conditionnumber","ConditionNumber","None",(PetscBool)lu->options.ConditionNumber,&flg,&set);CHKERRQ(ierr);
642   if (set && flg) lu->options.ConditionNumber = YES;
643   ierr = PetscOptionsEList("-mat_superlu_rowperm","rowperm","None",rowperm,2,rowperm[lu->options.RowPerm],&indx,&flg);CHKERRQ(ierr);
644   if (flg) lu->options.RowPerm = (rowperm_t)indx;
645   ierr = PetscOptionsBool("-mat_superlu_replacetinypivot","ReplaceTinyPivot","None",(PetscBool)lu->options.ReplaceTinyPivot,&flg,&set);CHKERRQ(ierr);
646   if (set && flg) lu->options.ReplaceTinyPivot = YES;
647   ierr = PetscOptionsBool("-mat_superlu_printstat","PrintStat","None",(PetscBool)lu->options.PrintStat,&flg,&set);CHKERRQ(ierr);
648   if (set && flg) lu->options.PrintStat = YES;
649   ierr = PetscOptionsInt("-mat_superlu_lwork","size of work array in bytes used by factorization","None",lu->lwork,&lu->lwork,NULL);CHKERRQ(ierr);
650   if (lu->lwork > 0) {
651     /* lwork is in bytes, hence PetscMalloc() is used here, not PetscMalloc1()*/
652     ierr = PetscMalloc(lu->lwork,&lu->work);CHKERRQ(ierr);
653   } else if (lu->lwork != 0 && lu->lwork != -1) {
654     ierr      = PetscPrintf(PETSC_COMM_SELF,"   Warning: lwork %D is not supported by SUPERLU. The default lwork=0 is used.\n",lu->lwork);
655     lu->lwork = 0;
656   }
657   /* ilu options */
658   ierr = PetscOptionsReal("-mat_superlu_ilu_droptol","ILU_DropTol","None",lu->options.ILU_DropTol,&real_input,&flg);CHKERRQ(ierr);
659   if (flg) lu->options.ILU_DropTol = (double) real_input;
660   ierr = PetscOptionsReal("-mat_superlu_ilu_filltol","ILU_FillTol","None",lu->options.ILU_FillTol,&real_input,&flg);CHKERRQ(ierr);
661   if (flg) lu->options.ILU_FillTol = (double) real_input;
662   ierr = PetscOptionsReal("-mat_superlu_ilu_fillfactor","ILU_FillFactor","None",lu->options.ILU_FillFactor,&real_input,&flg);CHKERRQ(ierr);
663   if (flg) lu->options.ILU_FillFactor = (double) real_input;
664   ierr = PetscOptionsInt("-mat_superlu_ilu_droprull","ILU_DropRule","None",lu->options.ILU_DropRule,&lu->options.ILU_DropRule,NULL);CHKERRQ(ierr);
665   ierr = PetscOptionsInt("-mat_superlu_ilu_norm","ILU_Norm","None",lu->options.ILU_Norm,&indx,&flg);CHKERRQ(ierr);
666   if (flg) lu->options.ILU_Norm = (norm_t)indx;
667   ierr = PetscOptionsInt("-mat_superlu_ilu_milu","ILU_MILU","None",lu->options.ILU_MILU,&indx,&flg);CHKERRQ(ierr);
668   if (flg) lu->options.ILU_MILU = (milu_t)indx;
669   PetscOptionsEnd();
670   if (lu->options.Equil == YES) {
671     /* superlu overwrites input matrix and rhs when Equil is used, thus create A_dup to keep user's A unchanged */
672     ierr = MatDuplicate_SeqAIJ(A,MAT_COPY_VALUES,&lu->A_dup);CHKERRQ(ierr);
673   }
674 
675   /* Allocate spaces (notice sizes are for the transpose) */
676   ierr = PetscMalloc1(m,&lu->etree);CHKERRQ(ierr);
677   ierr = PetscMalloc1(n,&lu->perm_r);CHKERRQ(ierr);
678   ierr = PetscMalloc1(m,&lu->perm_c);CHKERRQ(ierr);
679   ierr = PetscMalloc1(n,&lu->R);CHKERRQ(ierr);
680   ierr = PetscMalloc1(m,&lu->C);CHKERRQ(ierr);
681 
682   /* create rhs and solution x without allocate space for .Store */
683 #if defined(PETSC_USE_COMPLEX)
684 #if defined(PETSC_USE_REAL_SINGLE)
685   PetscStackCall("SuperLU:cCreate_Dense_Matrix(",cCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE));
686   PetscStackCall("SuperLU:cCreate_Dense_Matrix(",cCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE));
687 #else
688   PetscStackCall("SuperLU:zCreate_Dense_Matrix",zCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE));
689   PetscStackCall("SuperLU:zCreate_Dense_Matrix",zCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE));
690 #endif
691 #else
692 #if defined(PETSC_USE_REAL_SINGLE)
693   PetscStackCall("SuperLU:sCreate_Dense_Matrix",sCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE));
694   PetscStackCall("SuperLU:sCreate_Dense_Matrix",sCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE));
695 #else
696   PetscStackCall("SuperLU:dCreate_Dense_Matrix",dCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE));
697   PetscStackCall("SuperLU:dCreate_Dense_Matrix",dCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE));
698 #endif
699 #endif
700 
701   ierr     = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_seqaij_superlu);CHKERRQ(ierr);
702   ierr     = PetscObjectComposeFunction((PetscObject)B,"MatSuperluSetILUDropTol_C",MatSuperluSetILUDropTol_SuperLU);CHKERRQ(ierr);
703   B->spptr = lu;
704 
705   *F       = B;
706   PetscFunctionReturn(0);
707 }
708 
709 #undef __FUNCT__
710 #define __FUNCT__ "MatSolverPackageRegister_SuperLU"
711 PETSC_EXTERN PetscErrorCode MatSolverPackageRegister_SuperLU(void)
712 {
713   PetscErrorCode ierr;
714 
715   PetscFunctionBegin;
716   ierr = MatSolverPackageRegister(MATSOLVERSUPERLU,MATSEQAIJ,       MAT_FACTOR_LU,MatGetFactor_seqaij_superlu);CHKERRQ(ierr);
717   ierr = MatSolverPackageRegister(MATSOLVERSUPERLU,MATSEQAIJ,       MAT_FACTOR_ILU,MatGetFactor_seqaij_superlu);CHKERRQ(ierr);
718   PetscFunctionReturn(0);
719 }
720