xref: /petsc/src/mat/impls/aij/seq/lusol/lusol.c (revision f13dfd9ea68e0ddeee984e65c377a1819eab8a8a)
1 /*
2         Provides an interface to the LUSOL package of ....
3 
4 */
5 #include <../src/mat/impls/aij/seq/aij.h>
6 
7 #if defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
8   #define LU1FAC lu1fac_
9   #define LU6SOL lu6sol_
10   #define M1PAGE m1page_
11   #define M5SETX m5setx_
12   #define M6RDEL m6rdel_
13 #elif !defined(PETSC_HAVE_FORTRAN_CAPS)
14   #define LU1FAC lu1fac
15   #define LU6SOL lu6sol
16   #define M1PAGE m1page
17   #define M5SETX m5setx
18   #define M6RDEL m6rdel
19 #endif
20 
21 /*
22     Dummy symbols that the MINOS files mi25bfac.f and mi15blas.f may require
23 */
24 PETSC_EXTERN void M1PAGE()
25 {
26   ;
27 }
28 PETSC_EXTERN void M5SETX()
29 {
30   ;
31 }
32 
33 PETSC_EXTERN void M6RDEL()
34 {
35   ;
36 }
37 
38 PETSC_EXTERN void LU1FAC(int *m, int *n, int *nnz, int *size, int *luparm, double *parmlu, double *data, int *indc, int *indr, int *rowperm, int *colperm, int *collen, int *rowlen, int *colstart, int *rowstart, int *rploc, int *cploc, int *rpinv, int *cpinv, double *w, int *inform);
39 
40 PETSC_EXTERN void LU6SOL(int *mode, int *m, int *n, double *rhs, double *x, int *size, int *luparm, double *parmlu, double *data, int *indc, int *indr, int *rowperm, int *colperm, int *collen, int *rowlen, int *colstart, int *rowstart, int *inform);
41 
42 extern PetscErrorCode MatDuplicate_LUSOL(Mat, MatDuplicateOption, Mat *);
43 
44 typedef struct {
45   double *data;
46   int    *indc;
47   int    *indr;
48 
49   int    *ip;
50   int    *iq;
51   int    *lenc;
52   int    *lenr;
53   int    *locc;
54   int    *locr;
55   int    *iploc;
56   int    *iqloc;
57   int    *ipinv;
58   int    *iqinv;
59   double *mnsw;
60   double *mnsv;
61 
62   double elbowroom;
63   double luroom;     /* Extra space allocated when factor fails   */
64   double parmlu[30]; /* Input/output to LUSOL                     */
65 
66   int n;          /* Number of rows/columns in matrix          */
67   int nz;         /* Number of nonzeros                        */
68   int nnz;        /* Number of nonzeros allocated for factors  */
69   int luparm[30]; /* Input/output to LUSOL                     */
70 
71   PetscBool CleanUpLUSOL;
72 
73 } Mat_LUSOL;
74 
75 /*  LUSOL input/Output Parameters (Description uses C-style indexes
76  *
77  *  Input parameters                                        Typical value
78  *
79  *  luparm(0) = nout     File number for printed messages.         6
80  *  luparm(1) = lprint   Print level.                              0
81  *                    < 0 suppresses output.
82  *                    = 0 gives error messages.
83  *                    = 1 gives debug output from some of the
84  *                        other routines in LUSOL.
85  *                   >= 2 gives the pivot row and column and the
86  *                        no. of rows and columns involved at
87  *                        each elimination step in lu1fac.
88  *  luparm(2) = maxcol   lu1fac: maximum number of columns         5
89  *                        searched allowed in a Markowitz-type
90  *                        search for the next pivot element.
91  *                        For some of the factorization, the
92  *                        number of rows searched is
93  *                        maxrow = maxcol - 1.
94  *
95  *
96  *  Output parameters:
97  *
98  *  luparm(9) = inform   Return code from last call to any LU routine.
99  *  luparm(10) = nsing    No. of singularities marked in the
100  *                        output array w(*).
101  *  luparm(11) = jsing    Column index of last singularity.
102  *  luparm(12) = minlen   Minimum recommended value for  lena.
103  *  luparm(13) = maxlen   ?
104  *  luparm(14) = nupdat   No. of updates performed by the lu8 routines.
105  *  luparm(15) = nrank    No. of nonempty rows of U.
106  *  luparm(16) = ndens1   No. of columns remaining when the density of
107  *                        the matrix being factorized reached dens1.
108  *  luparm(17) = ndens2   No. of columns remaining when the density of
109  *                        the matrix being factorized reached dens2.
110  *  luparm(18) = jumin    The column index associated with dumin.
111  *  luparm(19) = numl0    No. of columns in initial  L.
112  *  luparm(20) = lenl0    Size of initial  L  (no. of nonzeros).
113  *  luparm(21) = lenu0    Size of initial  U.
114  *  luparm(22) = lenl     Size of current  L.
115  *  luparm(23) = lenu     Size of current  U.
116  *  luparm(24) = lrow     Length of row file.
117  *  luparm(25) = ncp      No. of compressions of LU data structures.
118  *  luparm(26) = mersum   lu1fac: sum of Markowitz merit counts.
119  *  luparm(27) = nutri    lu1fac: triangular rows in U.
120  *  luparm(28) = nltri    lu1fac: triangular rows in L.
121  *  luparm(29) =
122  *
123  *
124  *  Input parameters                                        Typical value
125  *
126  *  parmlu(0) = elmax1   Max multiplier allowed in  L           10.0
127  *                        during factor.
128  *  parmlu(1) = elmax2   Max multiplier allowed in  L           10.0
129  *                        during updates.
130  *  parmlu(2) = small    Absolute tolerance for             eps**0.8
131  *                        treating reals as zero.     IBM double: 3.0d-13
132  *  parmlu(3) = utol1    Absolute tol for flagging          eps**0.66667
133  *                        small diagonals of U.       IBM double: 3.7d-11
134  *  parmlu(4) = utol2    Relative tol for flagging          eps**0.66667
135  *                        small diagonals of U.       IBM double: 3.7d-11
136  *  parmlu(5) = uspace   Factor limiting waste space in  U.      3.0
137  *                        In lu1fac, the row or column lists
138  *                        are compressed if their length
139  *                        exceeds uspace times the length of
140  *                        either file after the last compression.
141  *  parmlu(6) = dens1    The density at which the Markowitz      0.3
142  *                        strategy should search maxcol columns
143  *                        and no rows.
144  *  parmlu(7) = dens2    the density at which the Markowitz      0.6
145  *                        strategy should search only 1 column
146  *                        or (preferably) use a dense LU for
147  *                        all the remaining rows and columns.
148  *
149  *
150  *  Output parameters:
151  *
152  *  parmlu(9) = amax     Maximum element in  A.
153  *  parmlu(10) = elmax    Maximum multiplier in current  L.
154  *  parmlu(11) = umax     Maximum element in current  U.
155  *  parmlu(12) = dumax    Maximum diagonal in  U.
156  *  parmlu(13) = dumin    Minimum diagonal in  U.
157  *  parmlu(14) =
158  *  parmlu(15) =
159  *  parmlu(16) =
160  *  parmlu(17) =
161  *  parmlu(18) =
162  *  parmlu(19) = resid    lu6sol: residual after solve with U or U'.
163  *  ...
164  *  parmlu(29) =
165  */
166 
167 #define Factorization_Tolerance       1e-1
168 #define Factorization_Pivot_Tolerance pow(2.2204460492503131E-16, 2.0 / 3.0)
169 #define Factorization_Small_Tolerance 1e-15 /* pow(DBL_EPSILON, 0.8) */
170 
171 static PetscErrorCode MatDestroy_LUSOL(Mat A)
172 {
173   Mat_LUSOL *lusol = (Mat_LUSOL *)A->spptr;
174 
175   PetscFunctionBegin;
176   if (lusol && lusol->CleanUpLUSOL) {
177     PetscCall(PetscFree(lusol->ip));
178     PetscCall(PetscFree(lusol->iq));
179     PetscCall(PetscFree(lusol->lenc));
180     PetscCall(PetscFree(lusol->lenr));
181     PetscCall(PetscFree(lusol->locc));
182     PetscCall(PetscFree(lusol->locr));
183     PetscCall(PetscFree(lusol->iploc));
184     PetscCall(PetscFree(lusol->iqloc));
185     PetscCall(PetscFree(lusol->ipinv));
186     PetscCall(PetscFree(lusol->iqinv));
187     PetscCall(PetscFree(lusol->mnsw));
188     PetscCall(PetscFree(lusol->mnsv));
189     PetscCall(PetscFree3(lusol->data, lusol->indc, lusol->indr));
190   }
191   PetscCall(PetscFree(A->spptr));
192   PetscCall(MatDestroy_SeqAIJ(A));
193   PetscFunctionReturn(PETSC_SUCCESS);
194 }
195 
196 static PetscErrorCode MatSolve_LUSOL(Mat A, Vec b, Vec x)
197 {
198   Mat_LUSOL    *lusol = (Mat_LUSOL *)A->spptr;
199   double       *xx;
200   const double *bb;
201   int           mode = 5;
202   int           i, m, n, nnz, status;
203 
204   PetscFunctionBegin;
205   PetscCall(VecGetArray(x, &xx));
206   PetscCall(VecGetArrayRead(b, &bb));
207 
208   m = n = lusol->n;
209   nnz   = lusol->nnz;
210 
211   for (i = 0; i < m; i++) lusol->mnsv[i] = bb[i];
212 
213   LU6SOL(&mode, &m, &n, lusol->mnsv, xx, &nnz, lusol->luparm, lusol->parmlu, lusol->data, lusol->indc, lusol->indr, lusol->ip, lusol->iq, lusol->lenc, lusol->lenr, lusol->locc, lusol->locr, &status);
214 
215   PetscCheck(!status, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "solve failed, error code %d", status);
216 
217   PetscCall(VecRestoreArray(x, &xx));
218   PetscCall(VecRestoreArrayRead(b, &bb));
219   PetscFunctionReturn(PETSC_SUCCESS);
220 }
221 
222 static PetscErrorCode MatLUFactorNumeric_LUSOL(Mat F, Mat A, const MatFactorInfo *info)
223 {
224   Mat_SeqAIJ *a;
225   Mat_LUSOL  *lusol = (Mat_LUSOL *)F->spptr;
226   int         m, n, nz, nnz, status;
227   int         i, rs, re;
228   int         factorizations;
229 
230   PetscFunctionBegin;
231   PetscCall(MatGetSize(A, &m, &n));
232   a = (Mat_SeqAIJ *)A->data;
233 
234   PetscCheck(m == lusol->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "factorization struct inconsistent");
235 
236   factorizations = 0;
237   do {
238     /*******************************************************************/
239     /* Check the workspace allocation.                                 */
240     /*******************************************************************/
241 
242     nz  = a->nz;
243     nnz = PetscMax(lusol->nnz, (int)(lusol->elbowroom * nz));
244     nnz = PetscMax(nnz, 5 * n);
245 
246     if (nnz < lusol->luparm[12]) {
247       nnz = (int)(lusol->luroom * lusol->luparm[12]);
248     } else if ((factorizations > 0) && (lusol->luroom < 6)) {
249       lusol->luroom += 0.1;
250     }
251 
252     nnz = PetscMax(nnz, (int)(lusol->luroom * (lusol->luparm[22] + lusol->luparm[23])));
253 
254     if (nnz > lusol->nnz) {
255       PetscCall(PetscFree3(lusol->data, lusol->indc, lusol->indr));
256       PetscCall(PetscMalloc3(nnz, &lusol->data, nnz, &lusol->indc, nnz, &lusol->indr));
257       lusol->nnz = nnz;
258     }
259 
260     /*******************************************************************/
261     /* Fill in the data for the problem.      (1-based Fortran style)  */
262     /*******************************************************************/
263 
264     nz = 0;
265     for (i = 0; i < n; i++) {
266       rs = a->i[i];
267       re = a->i[i + 1];
268 
269       while (rs < re) {
270         if (a->a[rs] != 0.0) {
271           lusol->indc[nz] = i + 1;
272           lusol->indr[nz] = a->j[rs] + 1;
273           lusol->data[nz] = a->a[rs];
274           nz++;
275         }
276         rs++;
277       }
278     }
279 
280     /*******************************************************************/
281     /* Do the factorization.                                           */
282     /*******************************************************************/
283 
284     LU1FAC(&m, &n, &nz, &nnz, lusol->luparm, lusol->parmlu, lusol->data, lusol->indc, lusol->indr, lusol->ip, lusol->iq, lusol->lenc, lusol->lenr, lusol->locc, lusol->locr, lusol->iploc, lusol->iqloc, lusol->ipinv, lusol->iqinv, lusol->mnsw, &status);
285 
286     switch (status) {
287     case 0: /* factored */
288       break;
289 
290     case 7: /* insufficient memory */
291       break;
292 
293     case 1:
294     case -1: /* singular */
295       SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "Singular matrix");
296 
297     case 3:
298     case 4: /* error conditions */
299       SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "matrix error");
300 
301     default: /* unknown condition */
302       SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "matrix unknown return code");
303     }
304 
305     factorizations++;
306   } while (status == 7);
307   F->ops->solve   = MatSolve_LUSOL;
308   F->assembled    = PETSC_TRUE;
309   F->preallocated = PETSC_TRUE;
310   PetscFunctionReturn(PETSC_SUCCESS);
311 }
312 
313 static PetscErrorCode MatLUFactorSymbolic_LUSOL(Mat F, Mat A, IS r, IS c, const MatFactorInfo *info)
314 {
315   /************************************************************************/
316   /* Input                                                                */
317   /*     A  - matrix to factor                                            */
318   /*     r  - row permutation (ignored)                                   */
319   /*     c  - column permutation (ignored)                                */
320   /*                                                                      */
321   /* Output                                                               */
322   /*     F  - matrix storing the factorization;                           */
323   /************************************************************************/
324   Mat_LUSOL *lusol;
325   int        i, m, n, nz, nnz;
326 
327   PetscFunctionBegin;
328   /************************************************************************/
329   /* Check the arguments.                                                 */
330   /************************************************************************/
331 
332   PetscCall(MatGetSize(A, &m, &n));
333   nz = ((Mat_SeqAIJ *)A->data)->nz;
334 
335   /************************************************************************/
336   /* Create the factorization.                                            */
337   /************************************************************************/
338 
339   F->ops->lufactornumeric = MatLUFactorNumeric_LUSOL;
340   lusol                   = (Mat_LUSOL *)F->spptr;
341 
342   /************************************************************************/
343   /* Initialize parameters                                                */
344   /************************************************************************/
345 
346   for (i = 0; i < 30; i++) {
347     lusol->luparm[i] = 0;
348     lusol->parmlu[i] = 0;
349   }
350 
351   lusol->luparm[1] = -1;
352   lusol->luparm[2] = 5;
353   lusol->luparm[7] = 1;
354 
355   lusol->parmlu[0] = 1 / Factorization_Tolerance;
356   lusol->parmlu[1] = 1 / Factorization_Tolerance;
357   lusol->parmlu[2] = Factorization_Small_Tolerance;
358   lusol->parmlu[3] = Factorization_Pivot_Tolerance;
359   lusol->parmlu[4] = Factorization_Pivot_Tolerance;
360   lusol->parmlu[5] = 3.0;
361   lusol->parmlu[6] = 0.3;
362   lusol->parmlu[7] = 0.6;
363 
364   /************************************************************************/
365   /* Allocate the workspace needed by LUSOL.                              */
366   /************************************************************************/
367 
368   lusol->elbowroom = PetscMax(lusol->elbowroom, info->fill);
369   nnz              = PetscMax((int)(lusol->elbowroom * nz), 5 * n);
370 
371   lusol->n      = n;
372   lusol->nz     = nz;
373   lusol->nnz    = nnz;
374   lusol->luroom = 1.75;
375 
376   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->ip));
377   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->iq));
378   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->lenc));
379   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->lenr));
380   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->locc));
381   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->locr));
382   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->iploc));
383   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->iqloc));
384   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->ipinv));
385   PetscCall(PetscMalloc(sizeof(int) * n, &lusol->iqinv));
386   PetscCall(PetscMalloc(sizeof(double) * n, &lusol->mnsw));
387   PetscCall(PetscMalloc(sizeof(double) * n, &lusol->mnsv));
388   PetscCall(PetscMalloc3(nnz, &lusol->data, nnz, &lusol->indc, nnz, &lusol->indr));
389 
390   lusol->CleanUpLUSOL     = PETSC_TRUE;
391   F->ops->lufactornumeric = MatLUFactorNumeric_LUSOL;
392   PetscFunctionReturn(PETSC_SUCCESS);
393 }
394 
395 static PetscErrorCode MatFactorGetSolverType_seqaij_lusol(Mat A, MatSolverType *type)
396 {
397   PetscFunctionBegin;
398   *type = MATSOLVERLUSOL;
399   PetscFunctionReturn(PETSC_SUCCESS);
400 }
401 
402 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_lusol(Mat A, MatFactorType ftype, Mat *F)
403 {
404   Mat        B;
405   Mat_LUSOL *lusol;
406   int        m, n;
407 
408   PetscFunctionBegin;
409   PetscCall(MatGetSize(A, &m, &n));
410   PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
411   PetscCall(MatSetSizes(B, PETSC_DECIDE, PETSC_DECIDE, m, n));
412   PetscCall(MatSetType(B, ((PetscObject)A)->type_name));
413   PetscCall(MatSeqAIJSetPreallocation(B, 0, NULL));
414 
415   PetscCall(PetscNew(&lusol));
416   B->spptr = lusol;
417 
418   B->trivialsymbolic       = PETSC_TRUE;
419   B->ops->lufactorsymbolic = MatLUFactorSymbolic_LUSOL;
420   B->ops->destroy          = MatDestroy_LUSOL;
421 
422   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_seqaij_lusol));
423 
424   B->factortype = MAT_FACTOR_LU;
425   PetscCall(PetscFree(B->solvertype));
426   PetscCall(PetscStrallocpy(MATSOLVERLUSOL, &B->solvertype));
427   PetscFunctionReturn(PETSC_SUCCESS);
428 }
429 
430 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_Lusol(void)
431 {
432   PetscFunctionBegin;
433   PetscCall(MatSolverTypeRegister(MATSOLVERLUSOL, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_lusol));
434   PetscFunctionReturn(PETSC_SUCCESS);
435 }
436 
437 /*MC
438   MATSOLVERLUSOL - "lusol" - Provides direct solvers, LU, for sequential matrices
439                          via the external package LUSOL.
440 
441   Works with `MATSEQAIJ` matrices
442 
443    Level: beginner
444 
445 .seealso: [](ch_matrices), `Mat`, `PCLU`, `PCFactorSetMatSolverType()`, `MatSolverType`
446 M*/
447