Actual source code: superlu.c
petsc-3.5.4 2015-05-23
2: /* --------------------------------------------------------------------
4: This file implements a subclass of the SeqAIJ matrix class that uses
5: the SuperLU sparse solver. You can use this as a starting point for
6: implementing your own subclass of a PETSc matrix class.
8: This demonstrates a way to make an implementation inheritence of a PETSc
9: matrix type. This means constructing a new matrix type (SuperLU) by changing some
10: of the methods of the previous type (SeqAIJ), adding additional data, and possibly
11: additional method. (See any book on object oriented programming).
12: */
14: /*
15: Defines the data structure for the base matrix type (SeqAIJ)
16: */
17: #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/
19: /*
20: SuperLU include files
21: */
22: EXTERN_C_BEGIN
23: #if defined(PETSC_USE_COMPLEX)
24: #if defined(PETSC_USE_REAL_SINGLE)
25: #include <slu_cdefs.h>
26: #else
27: #include <slu_zdefs.h>
28: #endif
29: #else
30: #if defined(PETSC_USE_REAL_SINGLE)
31: #include <slu_sdefs.h>
32: #else
33: #include <slu_ddefs.h>
34: #endif
35: #endif
36: #include <slu_util.h>
37: EXTERN_C_END
39: /*
40: This is the data we are "ADDING" to the SeqAIJ matrix type to get the SuperLU matrix type
41: */
42: typedef struct {
43: SuperMatrix A,L,U,B,X;
44: superlu_options_t options;
45: PetscInt *perm_c; /* column permutation vector */
46: PetscInt *perm_r; /* row permutations from partial pivoting */
47: PetscInt *etree;
48: PetscReal *R, *C;
49: char equed[1];
50: PetscInt lwork;
51: void *work;
52: PetscReal rpg, rcond;
53: mem_usage_t mem_usage;
54: MatStructure flg;
55: SuperLUStat_t stat;
56: Mat A_dup;
57: PetscScalar *rhs_dup;
59: /* Flag to clean up (non-global) SuperLU objects during Destroy */
60: PetscBool CleanUpSuperLU;
61: } Mat_SuperLU;
63: extern PetscErrorCode MatFactorInfo_SuperLU(Mat,PetscViewer);
64: extern PetscErrorCode MatLUFactorNumeric_SuperLU(Mat,Mat,const MatFactorInfo*);
65: extern PetscErrorCode MatDestroy_SuperLU(Mat);
66: extern PetscErrorCode MatView_SuperLU(Mat,PetscViewer);
67: extern PetscErrorCode MatAssemblyEnd_SuperLU(Mat,MatAssemblyType);
68: extern PetscErrorCode MatSolve_SuperLU(Mat,Vec,Vec);
69: extern PetscErrorCode MatMatSolve_SuperLU(Mat,Mat,Mat);
70: extern PetscErrorCode MatSolveTranspose_SuperLU(Mat,Vec,Vec);
71: extern PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat,Mat,IS,IS,const MatFactorInfo*);
72: extern PetscErrorCode MatDuplicate_SuperLU(Mat, MatDuplicateOption, Mat*);
74: /*
75: Utility function
76: */
79: PetscErrorCode MatFactorInfo_SuperLU(Mat A,PetscViewer viewer)
80: {
81: Mat_SuperLU *lu= (Mat_SuperLU*)A->spptr;
82: PetscErrorCode ierr;
83: superlu_options_t options;
86: /* check if matrix is superlu_dist type */
87: if (A->ops->solve != MatSolve_SuperLU) return(0);
89: options = lu->options;
91: PetscViewerASCIIPrintf(viewer,"SuperLU run parameters:\n");
92: PetscViewerASCIIPrintf(viewer," Equil: %s\n",(options.Equil != NO) ? "YES" : "NO");
93: PetscViewerASCIIPrintf(viewer," ColPerm: %D\n",options.ColPerm);
94: PetscViewerASCIIPrintf(viewer," IterRefine: %D\n",options.IterRefine);
95: PetscViewerASCIIPrintf(viewer," SymmetricMode: %s\n",(options.SymmetricMode != NO) ? "YES" : "NO");
96: PetscViewerASCIIPrintf(viewer," DiagPivotThresh: %g\n",options.DiagPivotThresh);
97: PetscViewerASCIIPrintf(viewer," PivotGrowth: %s\n",(options.PivotGrowth != NO) ? "YES" : "NO");
98: PetscViewerASCIIPrintf(viewer," ConditionNumber: %s\n",(options.ConditionNumber != NO) ? "YES" : "NO");
99: PetscViewerASCIIPrintf(viewer," RowPerm: %D\n",options.RowPerm);
100: PetscViewerASCIIPrintf(viewer," ReplaceTinyPivot: %s\n",(options.ReplaceTinyPivot != NO) ? "YES" : "NO");
101: PetscViewerASCIIPrintf(viewer," PrintStat: %s\n",(options.PrintStat != NO) ? "YES" : "NO");
102: PetscViewerASCIIPrintf(viewer," lwork: %D\n",lu->lwork);
103: if (A->factortype == MAT_FACTOR_ILU) {
104: PetscViewerASCIIPrintf(viewer," ILU_DropTol: %g\n",options.ILU_DropTol);
105: PetscViewerASCIIPrintf(viewer," ILU_FillTol: %g\n",options.ILU_FillTol);
106: PetscViewerASCIIPrintf(viewer," ILU_FillFactor: %g\n",options.ILU_FillFactor);
107: PetscViewerASCIIPrintf(viewer," ILU_DropRule: %D\n",options.ILU_DropRule);
108: PetscViewerASCIIPrintf(viewer," ILU_Norm: %D\n",options.ILU_Norm);
109: PetscViewerASCIIPrintf(viewer," ILU_MILU: %D\n",options.ILU_MILU);
110: }
111: return(0);
112: }
114: /*
115: These are the methods provided to REPLACE the corresponding methods of the
116: SeqAIJ matrix class. Other methods of SeqAIJ are not replaced
117: */
120: PetscErrorCode MatLUFactorNumeric_SuperLU(Mat F,Mat A,const MatFactorInfo *info)
121: {
122: Mat_SuperLU *lu = (Mat_SuperLU*)F->spptr;
123: Mat_SeqAIJ *aa;
125: PetscInt sinfo;
126: PetscReal ferr, berr;
127: NCformat *Ustore;
128: SCformat *Lstore;
131: if (lu->flg == SAME_NONZERO_PATTERN) { /* successing numerical factorization */
132: lu->options.Fact = SamePattern;
133: /* Ref: ~SuperLU_3.0/EXAMPLE/dlinsolx2.c */
134: Destroy_SuperMatrix_Store(&lu->A);
135: if (lu->options.Equil) {
136: MatCopy_SeqAIJ(A,lu->A_dup,SAME_NONZERO_PATTERN);
137: }
138: if (lu->lwork >= 0) {
139: PetscStackCall("SuperLU:Destroy_SuperNode_Matrix",Destroy_SuperNode_Matrix(&lu->L));
140: PetscStackCall("SuperLU:Destroy_CompCol_Matrix",Destroy_CompCol_Matrix(&lu->U));
141: lu->options.Fact = SamePattern;
142: }
143: }
145: /* Create the SuperMatrix for lu->A=A^T:
146: Since SuperLU likes column-oriented matrices,we pass it the transpose,
147: and then solve A^T X = B in MatSolve(). */
148: if (lu->options.Equil) {
149: aa = (Mat_SeqAIJ*)(lu->A_dup)->data;
150: } else {
151: aa = (Mat_SeqAIJ*)(A)->data;
152: }
153: #if defined(PETSC_USE_COMPLEX)
154: #if defined(PETSC_USE_REAL_SINGLE)
155: 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));
156: #else
157: 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));
158: #endif
159: #else
160: #if defined(PETSC_USE_REAL_SINGLE)
161: 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));
162: #else
163: 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));
164: #endif
165: #endif
167: /* Numerical factorization */
168: lu->B.ncol = 0; /* Indicate not to solve the system */
169: if (F->factortype == MAT_FACTOR_LU) {
170: #if defined(PETSC_USE_COMPLEX)
171: #if defined(PETSC_USE_REAL_SINGLE)
172: PetscStackCall("SuperLU:cgssvx",cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
173: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
174: &lu->mem_usage, &lu->stat, &sinfo));
175: #else
176: PetscStackCall("SuperLU:zgssvx",zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
177: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
178: &lu->mem_usage, &lu->stat, &sinfo));
179: #endif
180: #else
181: #if defined(PETSC_USE_REAL_SINGLE)
182: PetscStackCall("SuperLU:sgssvx",sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
183: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
184: &lu->mem_usage, &lu->stat, &sinfo));
185: #else
186: PetscStackCall("SuperLU:dgssvx",dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
187: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
188: &lu->mem_usage, &lu->stat, &sinfo));
189: #endif
190: #endif
191: } else if (F->factortype == MAT_FACTOR_ILU) {
192: /* Compute the incomplete factorization, condition number and pivot growth */
193: #if defined(PETSC_USE_COMPLEX)
194: #if defined(PETSC_USE_REAL_SINGLE)
195: PetscStackCall("SuperLU:cgsisx",cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r,lu->etree, lu->equed, lu->R, lu->C,
196: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
197: &lu->mem_usage, &lu->stat, &sinfo));
198: #else
199: PetscStackCall("SuperLU:zgsisx",zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r,lu->etree, lu->equed, lu->R, lu->C,
200: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
201: &lu->mem_usage, &lu->stat, &sinfo));
202: #endif
203: #else
204: #if defined(PETSC_USE_REAL_SINGLE)
205: PetscStackCall("SuperLU:sgsisx",sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
206: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
207: &lu->mem_usage, &lu->stat, &sinfo));
208: #else
209: PetscStackCall("SuperLU:dgsisx",dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
210: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
211: &lu->mem_usage, &lu->stat, &sinfo));
212: #endif
213: #endif
214: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
215: if (!sinfo || sinfo == lu->A.ncol+1) {
216: if (lu->options.PivotGrowth) {
217: PetscPrintf(PETSC_COMM_SELF," Recip. pivot growth = %e\n", lu->rpg);
218: }
219: if (lu->options.ConditionNumber) {
220: PetscPrintf(PETSC_COMM_SELF," Recip. condition number = %e\n", lu->rcond);
221: }
222: } else if (sinfo > 0) {
223: if (lu->lwork == -1) {
224: PetscPrintf(PETSC_COMM_SELF," ** Estimated memory: %D bytes\n", sinfo - lu->A.ncol);
225: } else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot in row %D",sinfo);
226: } else SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %D, the %D-th argument in gssvx() had an illegal value", sinfo,-sinfo);
228: if (lu->options.PrintStat) {
229: PetscPrintf(PETSC_COMM_SELF,"MatLUFactorNumeric_SuperLU():\n");
230: PetscStackCall("SuperLU:StatPrint",StatPrint(&lu->stat));
231: Lstore = (SCformat*) lu->L.Store;
232: Ustore = (NCformat*) lu->U.Store;
233: PetscPrintf(PETSC_COMM_SELF," No of nonzeros in factor L = %D\n", Lstore->nnz);
234: PetscPrintf(PETSC_COMM_SELF," No of nonzeros in factor U = %D\n", Ustore->nnz);
235: PetscPrintf(PETSC_COMM_SELF," No of nonzeros in L+U = %D\n", Lstore->nnz + Ustore->nnz - lu->A.ncol);
236: PetscPrintf(PETSC_COMM_SELF," L\\U MB %.3f\ttotal MB needed %.3f\n",
237: lu->mem_usage.for_lu/1e6, lu->mem_usage.total_needed/1e6);
238: }
240: lu->flg = SAME_NONZERO_PATTERN;
241: F->ops->solve = MatSolve_SuperLU;
242: F->ops->solvetranspose = MatSolveTranspose_SuperLU;
243: F->ops->matsolve = MatMatSolve_SuperLU;
244: return(0);
245: }
249: PetscErrorCode MatDestroy_SuperLU(Mat A)
250: {
252: Mat_SuperLU *lu=(Mat_SuperLU*)A->spptr;
255: if (lu && lu->CleanUpSuperLU) { /* Free the SuperLU datastructures */
256: PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->A));
257: PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->B));
258: PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->X));
259: PetscStackCall("SuperLU:StatFree",StatFree(&lu->stat));
260: if (lu->lwork >= 0) {
261: PetscStackCall("SuperLU:Destroy_SuperNode_Matrix",Destroy_SuperNode_Matrix(&lu->L));
262: PetscStackCall("SuperLU:Destroy_CompCol_Matrix",Destroy_CompCol_Matrix(&lu->U));
263: }
264: }
265: if (lu) {
266: PetscFree(lu->etree);
267: PetscFree(lu->perm_r);
268: PetscFree(lu->perm_c);
269: PetscFree(lu->R);
270: PetscFree(lu->C);
271: PetscFree(lu->rhs_dup);
272: MatDestroy(&lu->A_dup);
273: }
274: PetscFree(A->spptr);
276: /* clear composed functions */
277: PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverPackage_C",NULL);
278: PetscObjectComposeFunction((PetscObject)A,"MatSuperluSetILUDropTol_C",NULL);
280: MatDestroy_SeqAIJ(A);
281: return(0);
282: }
286: PetscErrorCode MatView_SuperLU(Mat A,PetscViewer viewer)
287: {
288: PetscErrorCode ierr;
289: PetscBool iascii;
290: PetscViewerFormat format;
293: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
294: if (iascii) {
295: PetscViewerGetFormat(viewer,&format);
296: if (format == PETSC_VIEWER_ASCII_INFO) {
297: MatFactorInfo_SuperLU(A,viewer);
298: }
299: }
300: return(0);
301: }
306: PetscErrorCode MatSolve_SuperLU_Private(Mat A,Vec b,Vec x)
307: {
308: Mat_SuperLU *lu = (Mat_SuperLU*)A->spptr;
309: PetscScalar *barray,*xarray;
310: PetscErrorCode ierr;
311: PetscInt info,i,n;
312: PetscReal ferr,berr;
313: static PetscBool cite = PETSC_FALSE;
316: if (lu->lwork == -1) return(0);
317: 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);
319: VecGetLocalSize(x,&n);
320: lu->B.ncol = 1; /* Set the number of right-hand side */
321: if (lu->options.Equil && !lu->rhs_dup) {
322: /* superlu overwrites b when Equil is used, thus create rhs_dup to keep user's b unchanged */
323: PetscMalloc1(n,&lu->rhs_dup);
324: }
325: if (lu->options.Equil) {
326: /* Copy b into rsh_dup */
327: VecGetArray(b,&barray);
328: PetscMemcpy(lu->rhs_dup,barray,n*sizeof(PetscScalar));
329: VecRestoreArray(b,&barray);
330: barray = lu->rhs_dup;
331: } else {
332: VecGetArray(b,&barray);
333: }
334: VecGetArray(x,&xarray);
336: #if defined(PETSC_USE_COMPLEX)
337: #if defined(PETSC_USE_REAL_SINGLE)
338: ((DNformat*)lu->B.Store)->nzval = (singlecomplex*)barray;
339: ((DNformat*)lu->X.Store)->nzval = (singlecomplex*)xarray;
340: #else
341: ((DNformat*)lu->B.Store)->nzval = (doublecomplex*)barray;
342: ((DNformat*)lu->X.Store)->nzval = (doublecomplex*)xarray;
343: #endif
344: #else
345: ((DNformat*)lu->B.Store)->nzval = barray;
346: ((DNformat*)lu->X.Store)->nzval = xarray;
347: #endif
349: lu->options.Fact = FACTORED; /* Indicate the factored form of A is supplied. */
350: if (A->factortype == MAT_FACTOR_LU) {
351: #if defined(PETSC_USE_COMPLEX)
352: #if defined(PETSC_USE_REAL_SINGLE)
353: PetscStackCall("SuperLU:cgssvx",cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
354: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
355: &lu->mem_usage, &lu->stat, &info));
356: #else
357: PetscStackCall("SuperLU:zgssvx",zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
358: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
359: &lu->mem_usage, &lu->stat, &info));
360: #endif
361: #else
362: #if defined(PETSC_USE_REAL_SINGLE)
363: PetscStackCall("SuperLU:sgssvx",sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
364: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
365: &lu->mem_usage, &lu->stat, &info));
366: #else
367: PetscStackCall("SuperLU:dgssvx",dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
368: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr,
369: &lu->mem_usage, &lu->stat, &info));
370: #endif
371: #endif
372: } else if (A->factortype == MAT_FACTOR_ILU) {
373: #if defined(PETSC_USE_COMPLEX)
374: #if defined(PETSC_USE_REAL_SINGLE)
375: PetscStackCall("SuperLU:cgsisx",cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
376: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
377: &lu->mem_usage, &lu->stat, &info));
378: #else
379: PetscStackCall("SuperLU:zgsisx",zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
380: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
381: &lu->mem_usage, &lu->stat, &info));
382: #endif
383: #else
384: #if defined(PETSC_USE_REAL_SINGLE)
385: PetscStackCall("SuperLU:sgsisx",sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
386: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
387: &lu->mem_usage, &lu->stat, &info));
388: #else
389: PetscStackCall("SuperLU:dgsisx",dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C,
390: &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond,
391: &lu->mem_usage, &lu->stat, &info));
392: #endif
393: #endif
394: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
395: if (!lu->options.Equil) {
396: VecRestoreArray(b,&barray);
397: }
398: VecRestoreArray(x,&xarray);
400: if (!info || info == lu->A.ncol+1) {
401: if (lu->options.IterRefine) {
402: PetscPrintf(PETSC_COMM_SELF,"Iterative Refinement:\n");
403: PetscPrintf(PETSC_COMM_SELF," %8s%8s%16s%16s\n", "rhs", "Steps", "FERR", "BERR");
404: for (i = 0; i < 1; ++i) {
405: PetscPrintf(PETSC_COMM_SELF," %8d%8d%16e%16e\n", i+1, lu->stat.RefineSteps, ferr, berr);
406: }
407: }
408: } else if (info > 0) {
409: if (lu->lwork == -1) {
410: PetscPrintf(PETSC_COMM_SELF," ** Estimated memory: %D bytes\n", info - lu->A.ncol);
411: } else {
412: PetscPrintf(PETSC_COMM_SELF," Warning: gssvx() returns info %D\n",info);
413: }
414: } 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);
416: if (lu->options.PrintStat) {
417: PetscPrintf(PETSC_COMM_SELF,"MatSolve__SuperLU():\n");
418: PetscStackCall("SuperLU:StatPrint",StatPrint(&lu->stat));
419: }
420: return(0);
421: }
425: PetscErrorCode MatSolve_SuperLU(Mat A,Vec b,Vec x)
426: {
427: Mat_SuperLU *lu = (Mat_SuperLU*)A->spptr;
431: lu->options.Trans = TRANS;
433: MatSolve_SuperLU_Private(A,b,x);
434: return(0);
435: }
439: PetscErrorCode MatSolveTranspose_SuperLU(Mat A,Vec b,Vec x)
440: {
441: Mat_SuperLU *lu = (Mat_SuperLU*)A->spptr;
445: lu->options.Trans = NOTRANS;
447: MatSolve_SuperLU_Private(A,b,x);
448: return(0);
449: }
453: PetscErrorCode MatMatSolve_SuperLU(Mat A,Mat B,Mat X)
454: {
455: Mat_SuperLU *lu = (Mat_SuperLU*)A->spptr;
456: PetscBool flg;
460: PetscObjectTypeCompareAny((PetscObject)B,&flg,MATSEQDENSE,MATMPIDENSE,NULL);
461: if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix");
462: PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,NULL);
463: if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix");
464: lu->options.Trans = TRANS;
465: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatSolve_SuperLU() is not implemented yet");
466: return(0);
467: }
469: /*
470: Note the r permutation is ignored
471: */
474: PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info)
475: {
476: Mat_SuperLU *lu = (Mat_SuperLU*)(F->spptr);
479: lu->flg = DIFFERENT_NONZERO_PATTERN;
480: lu->CleanUpSuperLU = PETSC_TRUE;
481: F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU;
482: return(0);
483: }
487: static PetscErrorCode MatSuperluSetILUDropTol_SuperLU(Mat F,PetscReal dtol)
488: {
489: Mat_SuperLU *lu= (Mat_SuperLU*)F->spptr;
492: lu->options.ILU_DropTol = dtol;
493: return(0);
494: }
498: /*@
499: MatSuperluSetILUDropTol - Set SuperLU ILU drop tolerance
500: Logically Collective on Mat
502: Input Parameters:
503: + F - the factored matrix obtained by calling MatGetFactor() from PETSc-SuperLU interface
504: - dtol - drop tolerance
506: Options Database:
507: . -mat_superlu_ilu_droptol <dtol>
509: Level: beginner
511: References: SuperLU Users' Guide
513: .seealso: MatGetFactor()
514: @*/
515: PetscErrorCode MatSuperluSetILUDropTol(Mat F,PetscReal dtol)
516: {
522: PetscTryMethod(F,"MatSuperluSetILUDropTol_C",(Mat,PetscReal),(F,dtol));
523: return(0);
524: }
528: PetscErrorCode MatFactorGetSolverPackage_seqaij_superlu(Mat A,const MatSolverPackage *type)
529: {
531: *type = MATSOLVERSUPERLU;
532: return(0);
533: }
536: /*MC
537: MATSOLVERSUPERLU = "superlu" - A solver package providing solvers LU and ILU for sequential matrices
538: via the external package SuperLU.
540: Use ./configure --download-superlu to have PETSc installed with SuperLU
542: Options Database Keys:
543: + -mat_superlu_equil <FALSE> - Equil (None)
544: . -mat_superlu_colperm <COLAMD> - (choose one of) NATURAL MMD_ATA MMD_AT_PLUS_A COLAMD
545: . -mat_superlu_iterrefine <NOREFINE> - (choose one of) NOREFINE SINGLE DOUBLE EXTRA
546: . -mat_superlu_symmetricmode: <FALSE> - SymmetricMode (None)
547: . -mat_superlu_diagpivotthresh <1> - DiagPivotThresh (None)
548: . -mat_superlu_pivotgrowth <FALSE> - PivotGrowth (None)
549: . -mat_superlu_conditionnumber <FALSE> - ConditionNumber (None)
550: . -mat_superlu_rowperm <NOROWPERM> - (choose one of) NOROWPERM LargeDiag
551: . -mat_superlu_replacetinypivot <FALSE> - ReplaceTinyPivot (None)
552: . -mat_superlu_printstat <FALSE> - PrintStat (None)
553: . -mat_superlu_lwork <0> - size of work array in bytes used by factorization (None)
554: . -mat_superlu_ilu_droptol <0> - ILU_DropTol (None)
555: . -mat_superlu_ilu_filltol <0> - ILU_FillTol (None)
556: . -mat_superlu_ilu_fillfactor <0> - ILU_FillFactor (None)
557: . -mat_superlu_ilu_droprull <0> - ILU_DropRule (None)
558: . -mat_superlu_ilu_norm <0> - ILU_Norm (None)
559: - -mat_superlu_ilu_milu <0> - ILU_MILU (None)
561: Notes: Do not confuse this with MATSOLVERSUPERLU_DIST which is for parallel sparse solves
563: Level: beginner
565: .seealso: PCLU, PCILU, MATSOLVERSUPERLU_DIST, MATSOLVERMUMPS, PCFactorSetMatSolverPackage(), MatSolverPackage
566: M*/
570: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat A,MatFactorType ftype,Mat *F)
571: {
572: Mat B;
573: Mat_SuperLU *lu;
575: PetscInt indx,m=A->rmap->n,n=A->cmap->n;
576: PetscBool flg;
577: PetscReal real_input;
578: const char *colperm[] ={"NATURAL","MMD_ATA","MMD_AT_PLUS_A","COLAMD"}; /* MY_PERMC - not supported by the petsc interface yet */
579: const char *iterrefine[]={"NOREFINE", "SINGLE", "DOUBLE", "EXTRA"};
580: const char *rowperm[] ={"NOROWPERM", "LargeDiag"}; /* MY_PERMC - not supported by the petsc interface yet */
583: MatCreate(PetscObjectComm((PetscObject)A),&B);
584: MatSetSizes(B,A->rmap->n,A->cmap->n,PETSC_DETERMINE,PETSC_DETERMINE);
585: MatSetType(B,((PetscObject)A)->type_name);
586: MatSeqAIJSetPreallocation(B,0,NULL);
588: if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU) {
589: B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU;
590: B->ops->ilufactorsymbolic = MatLUFactorSymbolic_SuperLU;
591: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
593: B->ops->destroy = MatDestroy_SuperLU;
594: B->ops->view = MatView_SuperLU;
595: B->factortype = ftype;
596: B->assembled = PETSC_TRUE; /* required by -ksp_view */
597: B->preallocated = PETSC_TRUE;
599: PetscNewLog(B,&lu);
601: if (ftype == MAT_FACTOR_LU) {
602: set_default_options(&lu->options);
603: /* Comments from SuperLU_4.0/SRC/dgssvx.c:
604: "Whether or not the system will be equilibrated depends on the
605: scaling of the matrix A, but if equilibration is used, A is
606: overwritten by diag(R)*A*diag(C) and B by diag(R)*B
607: (if options->Trans=NOTRANS) or diag(C)*B (if options->Trans = TRANS or CONJ)."
608: We set 'options.Equil = NO' as default because additional space is needed for it.
609: */
610: lu->options.Equil = NO;
611: } else if (ftype == MAT_FACTOR_ILU) {
612: /* Set the default input options of ilu: */
613: PetscStackCall("SuperLU:ilu_set_default_options",ilu_set_default_options(&lu->options));
614: }
615: lu->options.PrintStat = NO;
617: /* Initialize the statistics variables. */
618: PetscStackCall("SuperLU:StatInit",StatInit(&lu->stat));
619: lu->lwork = 0; /* allocate space internally by system malloc */
621: PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"SuperLU Options","Mat");
622: PetscOptionsBool("-mat_superlu_equil","Equil","None",(PetscBool)lu->options.Equil,(PetscBool*)&lu->options.Equil,0);
623: PetscOptionsEList("-mat_superlu_colperm","ColPerm","None",colperm,4,colperm[3],&indx,&flg);
624: if (flg) lu->options.ColPerm = (colperm_t)indx;
625: PetscOptionsEList("-mat_superlu_iterrefine","IterRefine","None",iterrefine,4,iterrefine[0],&indx,&flg);
626: if (flg) lu->options.IterRefine = (IterRefine_t)indx;
627: PetscOptionsBool("-mat_superlu_symmetricmode","SymmetricMode","None",(PetscBool)lu->options.SymmetricMode,&flg,0);
628: if (flg) lu->options.SymmetricMode = YES;
629: PetscOptionsReal("-mat_superlu_diagpivotthresh","DiagPivotThresh","None",lu->options.DiagPivotThresh,&real_input,&flg);
630: if (flg) lu->options.DiagPivotThresh = (double) real_input;
631: PetscOptionsBool("-mat_superlu_pivotgrowth","PivotGrowth","None",(PetscBool)lu->options.PivotGrowth,&flg,0);
632: if (flg) lu->options.PivotGrowth = YES;
633: PetscOptionsBool("-mat_superlu_conditionnumber","ConditionNumber","None",(PetscBool)lu->options.ConditionNumber,&flg,0);
634: if (flg) lu->options.ConditionNumber = YES;
635: PetscOptionsEList("-mat_superlu_rowperm","rowperm","None",rowperm,2,rowperm[lu->options.RowPerm],&indx,&flg);
636: if (flg) lu->options.RowPerm = (rowperm_t)indx;
637: PetscOptionsBool("-mat_superlu_replacetinypivot","ReplaceTinyPivot","None",(PetscBool)lu->options.ReplaceTinyPivot,&flg,0);
638: if (flg) lu->options.ReplaceTinyPivot = YES;
639: PetscOptionsBool("-mat_superlu_printstat","PrintStat","None",(PetscBool)lu->options.PrintStat,&flg,0);
640: if (flg) lu->options.PrintStat = YES;
641: PetscOptionsInt("-mat_superlu_lwork","size of work array in bytes used by factorization","None",lu->lwork,&lu->lwork,NULL);
642: if (lu->lwork > 0) {
643: PetscMalloc(lu->lwork,&lu->work);
644: } else if (lu->lwork != 0 && lu->lwork != -1) {
645: PetscPrintf(PETSC_COMM_SELF," Warning: lwork %D is not supported by SUPERLU. The default lwork=0 is used.\n",lu->lwork);
646: lu->lwork = 0;
647: }
648: /* ilu options */
649: PetscOptionsReal("-mat_superlu_ilu_droptol","ILU_DropTol","None",lu->options.ILU_DropTol,&real_input,&flg);
650: if (flg) lu->options.ILU_DropTol = (double) real_input;
651: PetscOptionsReal("-mat_superlu_ilu_filltol","ILU_FillTol","None",lu->options.ILU_FillTol,&real_input,&flg);
652: if (flg) lu->options.ILU_FillTol = (double) real_input;
653: PetscOptionsReal("-mat_superlu_ilu_fillfactor","ILU_FillFactor","None",lu->options.ILU_FillFactor,&real_input,&flg);
654: if (flg) lu->options.ILU_FillFactor = (double) real_input;
655: PetscOptionsInt("-mat_superlu_ilu_droprull","ILU_DropRule","None",lu->options.ILU_DropRule,&lu->options.ILU_DropRule,NULL);
656: PetscOptionsInt("-mat_superlu_ilu_norm","ILU_Norm","None",lu->options.ILU_Norm,&indx,&flg);
657: if (flg) lu->options.ILU_Norm = (norm_t)indx;
658: PetscOptionsInt("-mat_superlu_ilu_milu","ILU_MILU","None",lu->options.ILU_MILU,&indx,&flg);
659: if (flg) lu->options.ILU_MILU = (milu_t)indx;
660: PetscOptionsEnd();
661: if (lu->options.Equil == YES) {
662: /* superlu overwrites input matrix and rhs when Equil is used, thus create A_dup to keep user's A unchanged */
663: MatDuplicate_SeqAIJ(A,MAT_COPY_VALUES,&lu->A_dup);
664: }
666: /* Allocate spaces (notice sizes are for the transpose) */
667: PetscMalloc1(m,&lu->etree);
668: PetscMalloc1(n,&lu->perm_r);
669: PetscMalloc1(m,&lu->perm_c);
670: PetscMalloc1(n,&lu->R);
671: PetscMalloc1(m,&lu->C);
673: /* create rhs and solution x without allocate space for .Store */
674: #if defined(PETSC_USE_COMPLEX)
675: #if defined(PETSC_USE_REAL_SINGLE)
676: PetscStackCall("SuperLU:cCreate_Dense_Matrix(",cCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE));
677: PetscStackCall("SuperLU:cCreate_Dense_Matrix(",cCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE));
678: #else
679: PetscStackCall("SuperLU:zCreate_Dense_Matrix",zCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE));
680: PetscStackCall("SuperLU:zCreate_Dense_Matrix",zCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE));
681: #endif
682: #else
683: #if defined(PETSC_USE_REAL_SINGLE)
684: PetscStackCall("SuperLU:sCreate_Dense_Matrix",sCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE));
685: PetscStackCall("SuperLU:sCreate_Dense_Matrix",sCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE));
686: #else
687: PetscStackCall("SuperLU:dCreate_Dense_Matrix",dCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE));
688: PetscStackCall("SuperLU:dCreate_Dense_Matrix",dCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE));
689: #endif
690: #endif
692: PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_seqaij_superlu);
693: PetscObjectComposeFunction((PetscObject)B,"MatSuperluSetILUDropTol_C",MatSuperluSetILUDropTol_SuperLU);
694: B->spptr = lu;
695: *F = B;
696: return(0);
697: }