Actual source code: pastix.c

  1: /*
  2:  Provides an interface to the PaStiX sparse solver
  3:  */
  4: #include <../src/mat/impls/aij/seq/aij.h>
  5: #include <../src/mat/impls/aij/mpi/mpiaij.h>
  6: #include <../src/mat/impls/sbaij/seq/sbaij.h>
  7: #include <../src/mat/impls/sbaij/mpi/mpisbaij.h>

  9: #if defined(PETSC_USE_COMPLEX)
 10:   #define _H_COMPLEX
 11: #endif

 13: EXTERN_C_BEGIN
 14: #include <pastix.h>
 15: EXTERN_C_END

 17: #if defined(PETSC_USE_COMPLEX)
 18:   #if defined(PETSC_USE_REAL_SINGLE)
 19:     #define PASTIX_CALL c_pastix
 20:   #else
 21:     #define PASTIX_CALL z_pastix
 22:   #endif

 24: #else /* PETSC_USE_COMPLEX */

 26:   #if defined(PETSC_USE_REAL_SINGLE)
 27:     #define PASTIX_CALL s_pastix
 28:   #else
 29:     #define PASTIX_CALL d_pastix
 30:   #endif

 32: #endif /* PETSC_USE_COMPLEX */

 34: #define PetscPastixCall(...) \
 35:   do { \
 36:     PetscCall(PetscFPTrapPush(PETSC_FP_TRAP_OFF)); \
 37:     PetscStackCallExternalVoid(PetscStringize(PASTIX_CALL), PASTIX_CALL(__VA_ARGS__)); \
 38:     PetscCall(PetscFPTrapPop()); \
 39:   } while (0)

 41: typedef PetscScalar PastixScalar;

 43: typedef struct Mat_Pastix_ {
 44:   pastix_data_t *pastix_data; /* Pastix data storage structure                        */
 45:   MatStructure   matstruc;
 46:   PetscInt       n;                 /* Number of columns in the matrix                      */
 47:   PetscInt      *colptr;            /* Index of first element of each column in row and val */
 48:   PetscInt      *row;               /* Row of each element of the matrix                    */
 49:   PetscScalar   *val;               /* Value of each element of the matrix                  */
 50:   PetscInt      *perm;              /* Permutation tabular                                  */
 51:   PetscInt      *invp;              /* Reverse permutation tabular                          */
 52:   PetscScalar   *rhs;               /* Rhight-hand-side member                              */
 53:   PetscInt       rhsnbr;            /* Rhight-hand-side number (must be 1)                  */
 54:   PetscInt       iparm[IPARM_SIZE]; /* Integer parameters                                   */
 55:   double         dparm[DPARM_SIZE]; /* Floating point parameters                            */
 56:   MPI_Comm       pastix_comm;       /* PaStiX MPI communicator                              */
 57:   PetscMPIInt    commRank;          /* MPI rank                                             */
 58:   PetscMPIInt    commSize;          /* MPI communicator size                                */
 59:   PetscBool      CleanUpPastix;     /* Boolean indicating if we call PaStiX clean step      */
 60:   VecScatter     scat_rhs;
 61:   VecScatter     scat_sol;
 62:   Vec            b_seq;
 63: } Mat_Pastix;

 65: extern PetscErrorCode MatDuplicate_Pastix(Mat, MatDuplicateOption, Mat *);

 67: /*
 68:    convert Petsc seqaij matrix to CSC: colptr[n], row[nz], val[nz]

 70:   input:
 71:     A       - matrix in seqaij or mpisbaij (bs=1) format
 72:     valOnly - FALSE: spaces are allocated and values are set for the CSC
 73:               TRUE:  Only fill values
 74:   output:
 75:     n       - Size of the matrix
 76:     colptr  - Index of first element of each column in row and val
 77:     row     - Row of each element of the matrix
 78:     values  - Value of each element of the matrix
 79:  */
 80: static PetscErrorCode MatConvertToCSC(Mat A, PetscBool valOnly, PetscInt *n, PetscInt **colptr, PetscInt **row, PetscScalar **values)
 81: {
 82:   Mat_SeqAIJ  *aa      = (Mat_SeqAIJ *)A->data;
 83:   PetscInt    *rowptr  = aa->i;
 84:   PetscInt    *col     = aa->j;
 85:   PetscScalar *rvalues = aa->a;
 86:   PetscInt     m       = A->rmap->N;
 87:   PetscInt     nnz;
 88:   PetscInt     i, j, k;
 89:   PetscInt     base = 1;
 90:   PetscInt     idx;
 91:   PetscInt     colidx;
 92:   PetscInt    *colcount;
 93:   PetscBool    isSBAIJ;
 94:   PetscBool    isSeqSBAIJ;
 95:   PetscBool    isMpiSBAIJ;
 96:   PetscBool    isSym;

 98:   PetscFunctionBegin;
 99:   PetscCall(MatIsSymmetric(A, 0.0, &isSym));
100:   PetscCall(PetscObjectTypeCompare((PetscObject)A, MATSBAIJ, &isSBAIJ));
101:   PetscCall(PetscObjectTypeCompare((PetscObject)A, MATSEQSBAIJ, &isSeqSBAIJ));
102:   PetscCall(PetscObjectTypeCompare((PetscObject)A, MATMPISBAIJ, &isMpiSBAIJ));

104:   *n = A->cmap->N;

106:   /* PaStiX only needs triangular matrix if matrix is symmetric
107:    */
108:   if (isSym && !(isSBAIJ || isSeqSBAIJ || isMpiSBAIJ)) nnz = (aa->nz - *n) / 2 + *n;
109:   else nnz = aa->nz;

111:   if (!valOnly) {
112:     PetscCall(PetscMalloc1((*n) + 1, colptr));
113:     PetscCall(PetscMalloc1(nnz, row));
114:     PetscCall(PetscMalloc1(nnz, values));

116:     if (isSBAIJ || isSeqSBAIJ || isMpiSBAIJ) {
117:       PetscCall(PetscArraycpy(*colptr, rowptr, (*n) + 1));
118:       for (i = 0; i < *n + 1; i++) (*colptr)[i] += base;
119:       PetscCall(PetscArraycpy(*row, col, nnz));
120:       for (i = 0; i < nnz; i++) (*row)[i] += base;
121:       PetscCall(PetscArraycpy(*values, rvalues, nnz));
122:     } else {
123:       PetscCall(PetscMalloc1(*n, &colcount));

125:       for (i = 0; i < m; i++) colcount[i] = 0;
126:       /* Fill-in colptr */
127:       for (i = 0; i < m; i++) {
128:         for (j = rowptr[i]; j < rowptr[i + 1]; j++) {
129:           if (!isSym || col[j] <= i) colcount[col[j]]++;
130:         }
131:       }

133:       (*colptr)[0] = base;
134:       for (j = 0; j < *n; j++) {
135:         (*colptr)[j + 1] = (*colptr)[j] + colcount[j];
136:         /* in next loop we fill starting from (*colptr)[colidx] - base */
137:         colcount[j] = -base;
138:       }

140:       /* Fill-in rows and values */
141:       for (i = 0; i < m; i++) {
142:         for (j = rowptr[i]; j < rowptr[i + 1]; j++) {
143:           if (!isSym || col[j] <= i) {
144:             colidx         = col[j];
145:             idx            = (*colptr)[colidx] + colcount[colidx];
146:             (*row)[idx]    = i + base;
147:             (*values)[idx] = rvalues[j];
148:             colcount[colidx]++;
149:           }
150:         }
151:       }
152:       PetscCall(PetscFree(colcount));
153:     }
154:   } else {
155:     /* Fill-in only values */
156:     for (i = 0; i < m; i++) {
157:       for (j = rowptr[i]; j < rowptr[i + 1]; j++) {
158:         colidx = col[j];
159:         if ((isSBAIJ || isSeqSBAIJ || isMpiSBAIJ) || !isSym || col[j] <= i) {
160:           /* look for the value to fill */
161:           for (k = (*colptr)[colidx] - base; k < (*colptr)[colidx + 1] - base; k++) {
162:             if (((*row)[k] - base) == i) {
163:               (*values)[k] = rvalues[j];
164:               break;
165:             }
166:           }
167:           /* data structure of sparse matrix has changed */
168:           PetscCheck(k != (*colptr)[colidx + 1] - base, PETSC_COMM_SELF, PETSC_ERR_PLIB, "overflow on k %" PetscInt_FMT, k);
169:         }
170:       }
171:     }
172:   }
173:   PetscFunctionReturn(PETSC_SUCCESS);
174: }

176: /*
177:   Call clean step of PaStiX if lu->CleanUpPastix == true.
178:   Free the CSC matrix.
179:  */
180: static PetscErrorCode MatDestroy_Pastix(Mat A)
181: {
182:   Mat_Pastix *lu = (Mat_Pastix *)A->data;

184:   PetscFunctionBegin;
185:   if (lu->CleanUpPastix) {
186:     /* Terminate instance, deallocate memories */
187:     PetscCall(VecScatterDestroy(&lu->scat_rhs));
188:     PetscCall(VecDestroy(&lu->b_seq));
189:     PetscCall(VecScatterDestroy(&lu->scat_sol));

191:     lu->iparm[IPARM_START_TASK] = API_TASK_CLEAN;
192:     lu->iparm[IPARM_END_TASK]   = API_TASK_CLEAN;

194:     PetscPastixCall(&lu->pastix_data, lu->pastix_comm, lu->n, lu->colptr, lu->row, (PastixScalar *)lu->val, lu->perm, lu->invp, (PastixScalar *)lu->rhs, lu->rhsnbr, lu->iparm, lu->dparm);
195:     PetscCheck(lu->iparm[IPARM_ERROR_NUMBER] == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "Error reported by PaStiX in destroy: iparm(IPARM_ERROR_NUMBER)=%" PetscInt_FMT, lu->iparm[IPARM_ERROR_NUMBER]);
196:     PetscCall(PetscFree(lu->colptr));
197:     PetscCall(PetscFree(lu->row));
198:     PetscCall(PetscFree(lu->val));
199:     PetscCall(PetscFree(lu->perm));
200:     PetscCall(PetscFree(lu->invp));
201:     PetscCallMPI(MPI_Comm_free(&lu->pastix_comm));
202:   }
203:   PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL));
204:   PetscCall(PetscFree(A->data));
205:   PetscFunctionReturn(PETSC_SUCCESS);
206: }

208: /*
209:   Gather right-hand side.
210:   Call for Solve step.
211:   Scatter solution.
212:  */
213: static PetscErrorCode MatSolve_PaStiX(Mat A, Vec b, Vec x)
214: {
215:   Mat_Pastix  *lu = (Mat_Pastix *)A->data;
216:   PetscScalar *array;
217:   Vec          x_seq;

219:   PetscFunctionBegin;
220:   lu->rhsnbr = 1;
221:   x_seq      = lu->b_seq;
222:   if (lu->commSize > 1) {
223:     /* PaStiX only supports centralized rhs. Scatter b into a sequential rhs vector */
224:     PetscCall(VecScatterBegin(lu->scat_rhs, b, x_seq, INSERT_VALUES, SCATTER_FORWARD));
225:     PetscCall(VecScatterEnd(lu->scat_rhs, b, x_seq, INSERT_VALUES, SCATTER_FORWARD));
226:     PetscCall(VecGetArray(x_seq, &array));
227:   } else { /* size == 1 */
228:     PetscCall(VecCopy(b, x));
229:     PetscCall(VecGetArray(x, &array));
230:   }
231:   lu->rhs = array;
232:   if (lu->commSize == 1) {
233:     PetscCall(VecRestoreArray(x, &array));
234:   } else {
235:     PetscCall(VecRestoreArray(x_seq, &array));
236:   }

238:   /* solve phase */
239:   lu->iparm[IPARM_START_TASK] = API_TASK_SOLVE;
240:   lu->iparm[IPARM_END_TASK]   = API_TASK_REFINE;
241:   lu->iparm[IPARM_RHS_MAKING] = API_RHS_B;

243:   PetscPastixCall(&lu->pastix_data, lu->pastix_comm, lu->n, lu->colptr, lu->row, (PastixScalar *)lu->val, lu->perm, lu->invp, (PastixScalar *)lu->rhs, lu->rhsnbr, lu->iparm, lu->dparm);
244:   PetscCheck(lu->iparm[IPARM_ERROR_NUMBER] == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "Error reported by PaStiX in solve phase: lu->iparm[IPARM_ERROR_NUMBER] = %" PetscInt_FMT, lu->iparm[IPARM_ERROR_NUMBER]);

246:   if (lu->commSize == 1) {
247:     PetscCall(VecRestoreArray(x, &lu->rhs));
248:   } else {
249:     PetscCall(VecRestoreArray(x_seq, &lu->rhs));
250:   }

252:   if (lu->commSize > 1) { /* convert PaStiX centralized solution to petsc mpi x */
253:     PetscCall(VecScatterBegin(lu->scat_sol, x_seq, x, INSERT_VALUES, SCATTER_FORWARD));
254:     PetscCall(VecScatterEnd(lu->scat_sol, x_seq, x, INSERT_VALUES, SCATTER_FORWARD));
255:   }
256:   PetscFunctionReturn(PETSC_SUCCESS);
257: }

259: /*
260:   Numeric factorisation using PaStiX solver.

262:  */
263: static PetscErrorCode MatFactorNumeric_PaStiX(Mat F, Mat A, const MatFactorInfo *info)
264: {
265:   Mat_Pastix *lu = (Mat_Pastix *)F->data;
266:   Mat        *tseq;
267:   PetscInt    icntl;
268:   PetscInt    M = A->rmap->N;
269:   PetscBool   valOnly, flg, isSym;
270:   IS          is_iden;
271:   Vec         b;
272:   IS          isrow;
273:   PetscBool   isSeqAIJ, isSeqSBAIJ, isMPIAIJ;

275:   PetscFunctionBegin;
276:   PetscCall(PetscObjectTypeCompare((PetscObject)A, MATSEQAIJ, &isSeqAIJ));
277:   PetscCall(PetscObjectTypeCompare((PetscObject)A, MATMPIAIJ, &isMPIAIJ));
278:   PetscCall(PetscObjectTypeCompare((PetscObject)A, MATSEQSBAIJ, &isSeqSBAIJ));
279:   if (lu->matstruc == DIFFERENT_NONZERO_PATTERN) {
280:     F->ops->solve = MatSolve_PaStiX;

282:     /* Initialize a PASTIX instance */
283:     PetscCallMPI(MPI_Comm_dup(PetscObjectComm((PetscObject)A), &lu->pastix_comm));
284:     PetscCallMPI(MPI_Comm_rank(lu->pastix_comm, &lu->commRank));
285:     PetscCallMPI(MPI_Comm_size(lu->pastix_comm, &lu->commSize));

287:     /* Set pastix options */
288:     lu->iparm[IPARM_MODIFY_PARAMETER] = API_NO;
289:     lu->iparm[IPARM_START_TASK]       = API_TASK_INIT;
290:     lu->iparm[IPARM_END_TASK]         = API_TASK_INIT;

292:     lu->rhsnbr = 1;

294:     /* Call to set default pastix options */
295:     PetscPastixCall(&lu->pastix_data, lu->pastix_comm, lu->n, lu->colptr, lu->row, (PastixScalar *)lu->val, lu->perm, lu->invp, (PastixScalar *)lu->rhs, lu->rhsnbr, lu->iparm, lu->dparm);
296:     PetscCheck(lu->iparm[IPARM_ERROR_NUMBER] == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "Error reported by PaStiX in MatFactorNumeric: iparm(IPARM_ERROR_NUMBER)=%" PetscInt_FMT, lu->iparm[IPARM_ERROR_NUMBER]);

298:     PetscOptionsBegin(PetscObjectComm((PetscObject)F), ((PetscObject)F)->prefix, "PaStiX Options", "Mat");
299:     icntl                    = -1;
300:     lu->iparm[IPARM_VERBOSE] = API_VERBOSE_NOT;
301:     PetscCall(PetscOptionsInt("-mat_pastix_verbose", "iparm[IPARM_VERBOSE] : level of printing (0 to 2)", "None", lu->iparm[IPARM_VERBOSE], &icntl, &flg));
302:     if ((flg && icntl >= 0) || PetscLogPrintInfo) lu->iparm[IPARM_VERBOSE] = icntl;
303:     icntl = -1;
304:     PetscCall(PetscOptionsInt("-mat_pastix_threadnbr", "iparm[IPARM_THREAD_NBR] : Number of thread by MPI node", "None", lu->iparm[IPARM_THREAD_NBR], &icntl, &flg));
305:     if (flg && icntl > 0) lu->iparm[IPARM_THREAD_NBR] = icntl;
306:     PetscOptionsEnd();
307:     valOnly = PETSC_FALSE;
308:   } else {
309:     if (isSeqAIJ || isMPIAIJ) {
310:       PetscCall(PetscFree(lu->colptr));
311:       PetscCall(PetscFree(lu->row));
312:       PetscCall(PetscFree(lu->val));
313:       valOnly = PETSC_FALSE;
314:     } else valOnly = PETSC_TRUE;
315:   }

317:   lu->iparm[IPARM_MATRIX_VERIFICATION] = API_YES;

319:   /* convert mpi A to seq mat A */
320:   PetscCall(ISCreateStride(PETSC_COMM_SELF, M, 0, 1, &isrow));
321:   PetscCall(MatCreateSubMatrices(A, 1, &isrow, &isrow, MAT_INITIAL_MATRIX, &tseq));
322:   PetscCall(ISDestroy(&isrow));

324:   PetscCall(MatConvertToCSC(*tseq, valOnly, &lu->n, &lu->colptr, &lu->row, &lu->val));
325:   PetscCall(MatIsSymmetric(*tseq, 0.0, &isSym));
326:   PetscCall(MatDestroyMatrices(1, &tseq));

328:   if (!lu->perm) {
329:     PetscCall(PetscMalloc1(lu->n, &lu->perm));
330:     PetscCall(PetscMalloc1(lu->n, &lu->invp));
331:   }

333:   if (isSym) {
334:     /* On symmetric matrix, LLT */
335:     lu->iparm[IPARM_SYM]           = API_SYM_YES;
336:     lu->iparm[IPARM_FACTORIZATION] = API_FACT_LDLT;
337:   } else {
338:     /* On unsymmetric matrix, LU */
339:     lu->iparm[IPARM_SYM]           = API_SYM_NO;
340:     lu->iparm[IPARM_FACTORIZATION] = API_FACT_LU;
341:   }

343:   if (lu->matstruc == DIFFERENT_NONZERO_PATTERN) {
344:     if (!(isSeqAIJ || isSeqSBAIJ) && !lu->b_seq) {
345:       /* PaStiX only supports centralized rhs. Create scatter scat_rhs for repeated use in MatSolve() */
346:       PetscCall(VecCreateSeq(PETSC_COMM_SELF, A->cmap->N, &lu->b_seq));
347:       PetscCall(ISCreateStride(PETSC_COMM_SELF, A->cmap->N, 0, 1, &is_iden));
348:       PetscCall(MatCreateVecs(A, NULL, &b));
349:       PetscCall(VecScatterCreate(b, is_iden, lu->b_seq, is_iden, &lu->scat_rhs));
350:       PetscCall(VecScatterCreate(lu->b_seq, is_iden, b, is_iden, &lu->scat_sol));
351:       PetscCall(ISDestroy(&is_iden));
352:       PetscCall(VecDestroy(&b));
353:     }
354:     lu->iparm[IPARM_START_TASK] = API_TASK_ORDERING;
355:     lu->iparm[IPARM_END_TASK]   = API_TASK_NUMFACT;

357:     PetscPastixCall(&lu->pastix_data, lu->pastix_comm, lu->n, lu->colptr, lu->row, (PastixScalar *)lu->val, lu->perm, lu->invp, (PastixScalar *)lu->rhs, lu->rhsnbr, lu->iparm, lu->dparm);
358:     PetscCheck(lu->iparm[IPARM_ERROR_NUMBER] == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "Error reported by PaStiX in analysis phase: iparm(IPARM_ERROR_NUMBER)=%" PetscInt_FMT, lu->iparm[IPARM_ERROR_NUMBER]);
359:   } else {
360:     lu->iparm[IPARM_START_TASK] = API_TASK_NUMFACT;
361:     lu->iparm[IPARM_END_TASK]   = API_TASK_NUMFACT;
362:     PetscPastixCall(&lu->pastix_data, lu->pastix_comm, lu->n, lu->colptr, lu->row, (PastixScalar *)lu->val, lu->perm, lu->invp, (PastixScalar *)lu->rhs, lu->rhsnbr, lu->iparm, lu->dparm);
363:     PetscCheck(lu->iparm[IPARM_ERROR_NUMBER] == 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "Error reported by PaStiX in analysis phase: iparm(IPARM_ERROR_NUMBER)=%" PetscInt_FMT, lu->iparm[IPARM_ERROR_NUMBER]);
364:   }

366:   F->assembled      = PETSC_TRUE;
367:   lu->matstruc      = SAME_NONZERO_PATTERN;
368:   lu->CleanUpPastix = PETSC_TRUE;
369:   PetscFunctionReturn(PETSC_SUCCESS);
370: }

372: /* Note the Petsc r and c permutations are ignored */
373: static PetscErrorCode MatLUFactorSymbolic_AIJPASTIX(Mat F, Mat A, IS r, IS c, const MatFactorInfo *info)
374: {
375:   Mat_Pastix *lu = (Mat_Pastix *)F->data;

377:   PetscFunctionBegin;
378:   lu->iparm[IPARM_FACTORIZATION] = API_FACT_LU;
379:   lu->iparm[IPARM_SYM]           = API_SYM_YES;
380:   lu->matstruc                   = DIFFERENT_NONZERO_PATTERN;
381:   F->ops->lufactornumeric        = MatFactorNumeric_PaStiX;
382:   PetscFunctionReturn(PETSC_SUCCESS);
383: }

385: static PetscErrorCode MatCholeskyFactorSymbolic_SBAIJPASTIX(Mat F, Mat A, IS r, const MatFactorInfo *info)
386: {
387:   Mat_Pastix *lu = (Mat_Pastix *)F->data;

389:   PetscFunctionBegin;
390:   lu->iparm[IPARM_FACTORIZATION] = API_FACT_LLT;
391:   lu->iparm[IPARM_SYM]           = API_SYM_NO;
392:   lu->matstruc                   = DIFFERENT_NONZERO_PATTERN;
393:   F->ops->choleskyfactornumeric  = MatFactorNumeric_PaStiX;
394:   PetscFunctionReturn(PETSC_SUCCESS);
395: }

397: static PetscErrorCode MatView_PaStiX(Mat A, PetscViewer viewer)
398: {
399:   PetscBool         iascii;
400:   PetscViewerFormat format;

402:   PetscFunctionBegin;
403:   PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
404:   if (iascii) {
405:     PetscCall(PetscViewerGetFormat(viewer, &format));
406:     if (format == PETSC_VIEWER_ASCII_INFO) {
407:       Mat_Pastix *lu = (Mat_Pastix *)A->data;

409:       PetscCall(PetscViewerASCIIPrintf(viewer, "PaStiX run parameters:\n"));
410:       PetscCall(PetscViewerASCIIPrintf(viewer, "  Matrix type :                      %s \n", (lu->iparm[IPARM_SYM] == API_SYM_YES) ? "Symmetric" : "Unsymmetric"));
411:       PetscCall(PetscViewerASCIIPrintf(viewer, "  Level of printing (0,1,2):         %" PetscInt_FMT " \n", lu->iparm[IPARM_VERBOSE]));
412:       PetscCall(PetscViewerASCIIPrintf(viewer, "  Number of refinements iterations : %" PetscInt_FMT " \n", lu->iparm[IPARM_NBITER]));
413:       PetscCall(PetscPrintf(PETSC_COMM_SELF, "  Error :                        %g \n", lu->dparm[DPARM_RELATIVE_ERROR]));
414:     }
415:   }
416:   PetscFunctionReturn(PETSC_SUCCESS);
417: }

419: /*MC
420:      MATSOLVERPASTIX  - A solver package providing direct solvers (LU) for distributed
421:   and sequential matrices via the external package PaStiX.

423:   Use `./configure` `--download-pastix` `--download-ptscotch`  to have PETSc installed with PasTiX

425:   Use `-pc_type lu` `-pc_factor_mat_solver_type pastix` to use this direct solver

427:   Options Database Keys:
428: + -mat_pastix_verbose   <0,1,2>   - print level of information messages from PaStiX
429: - -mat_pastix_threadnbr <integer> - Set the number of threads for each MPI process

431:   Notes:
432:     This only works for matrices with symmetric nonzero structure, if you pass it a matrix with
433:    nonsymmetric structure PasTiX, and hence, PETSc return with an error.

435:   Level: beginner

437: .seealso: [](ch_matrices), `Mat`, `PCFactorSetMatSolverType()`, `MatSolverType`, `MatGetFactor()`
438: M*/

440: static PetscErrorCode MatGetInfo_PaStiX(Mat A, MatInfoType flag, MatInfo *info)
441: {
442:   Mat_Pastix *lu = (Mat_Pastix *)A->data;

444:   PetscFunctionBegin;
445:   info->block_size        = 1.0;
446:   info->nz_allocated      = lu->iparm[IPARM_NNZEROS];
447:   info->nz_used           = lu->iparm[IPARM_NNZEROS];
448:   info->nz_unneeded       = 0.0;
449:   info->assemblies        = 0.0;
450:   info->mallocs           = 0.0;
451:   info->memory            = 0.0;
452:   info->fill_ratio_given  = 0;
453:   info->fill_ratio_needed = 0;
454:   info->factor_mallocs    = 0;
455:   PetscFunctionReturn(PETSC_SUCCESS);
456: }

458: static PetscErrorCode MatFactorGetSolverType_pastix(Mat A, MatSolverType *type)
459: {
460:   PetscFunctionBegin;
461:   *type = MATSOLVERPASTIX;
462:   PetscFunctionReturn(PETSC_SUCCESS);
463: }

465: /*
466:     The seq and mpi versions of this function are the same
467: */
468: static PetscErrorCode MatGetFactor_seqaij_pastix(Mat A, MatFactorType ftype, Mat *F)
469: {
470:   Mat         B;
471:   Mat_Pastix *pastix;

473:   PetscFunctionBegin;
474:   PetscCheck(ftype == MAT_FACTOR_LU, PETSC_COMM_SELF, PETSC_ERR_SUP, "Cannot use PETSc AIJ matrices with PaStiX Cholesky, use SBAIJ matrix");
475:   /* Create the factorization matrix */
476:   PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
477:   PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, A->rmap->N, A->cmap->N));
478:   PetscCall(PetscStrallocpy("pastix", &((PetscObject)B)->type_name));
479:   PetscCall(MatSetUp(B));

481:   B->trivialsymbolic       = PETSC_TRUE;
482:   B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJPASTIX;
483:   B->ops->view             = MatView_PaStiX;
484:   B->ops->getinfo          = MatGetInfo_PaStiX;

486:   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_pastix));

488:   B->factortype = MAT_FACTOR_LU;

490:   /* set solvertype */
491:   PetscCall(PetscFree(B->solvertype));
492:   PetscCall(PetscStrallocpy(MATSOLVERPASTIX, &B->solvertype));

494:   PetscCall(PetscNew(&pastix));

496:   pastix->CleanUpPastix = PETSC_FALSE;
497:   pastix->scat_rhs      = NULL;
498:   pastix->scat_sol      = NULL;
499:   B->ops->getinfo       = MatGetInfo_External;
500:   B->ops->destroy       = MatDestroy_Pastix;
501:   B->data               = (void *)pastix;

503:   *F = B;
504:   PetscFunctionReturn(PETSC_SUCCESS);
505: }

507: static PetscErrorCode MatGetFactor_mpiaij_pastix(Mat A, MatFactorType ftype, Mat *F)
508: {
509:   Mat         B;
510:   Mat_Pastix *pastix;

512:   PetscFunctionBegin;
513:   PetscCheck(ftype == MAT_FACTOR_LU, PETSC_COMM_SELF, PETSC_ERR_SUP, "Cannot use PETSc AIJ matrices with PaStiX Cholesky, use SBAIJ matrix");
514:   /* Create the factorization matrix */
515:   PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
516:   PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, A->rmap->N, A->cmap->N));
517:   PetscCall(PetscStrallocpy("pastix", &((PetscObject)B)->type_name));
518:   PetscCall(MatSetUp(B));

520:   B->trivialsymbolic       = PETSC_TRUE;
521:   B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJPASTIX;
522:   B->ops->view             = MatView_PaStiX;
523:   B->ops->getinfo          = MatGetInfo_PaStiX;
524:   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_pastix));

526:   B->factortype = MAT_FACTOR_LU;

528:   /* set solvertype */
529:   PetscCall(PetscFree(B->solvertype));
530:   PetscCall(PetscStrallocpy(MATSOLVERPASTIX, &B->solvertype));

532:   PetscCall(PetscNew(&pastix));

534:   pastix->CleanUpPastix = PETSC_FALSE;
535:   pastix->scat_rhs      = NULL;
536:   pastix->scat_sol      = NULL;
537:   B->ops->getinfo       = MatGetInfo_External;
538:   B->ops->destroy       = MatDestroy_Pastix;
539:   B->data               = (void *)pastix;

541:   *F = B;
542:   PetscFunctionReturn(PETSC_SUCCESS);
543: }

545: static PetscErrorCode MatGetFactor_seqsbaij_pastix(Mat A, MatFactorType ftype, Mat *F)
546: {
547:   Mat         B;
548:   Mat_Pastix *pastix;

550:   PetscFunctionBegin;
551:   PetscCheck(ftype == MAT_FACTOR_CHOLESKY, PETSC_COMM_SELF, PETSC_ERR_SUP, "Cannot use PETSc SBAIJ matrices with PaStiX LU, use AIJ matrix");
552:   /* Create the factorization matrix */
553:   PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
554:   PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, A->rmap->N, A->cmap->N));
555:   PetscCall(PetscStrallocpy("pastix", &((PetscObject)B)->type_name));
556:   PetscCall(MatSetUp(B));

558:   B->trivialsymbolic             = PETSC_TRUE;
559:   B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SBAIJPASTIX;
560:   B->ops->view                   = MatView_PaStiX;
561:   B->ops->getinfo                = MatGetInfo_PaStiX;
562:   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_pastix));

564:   B->factortype = MAT_FACTOR_CHOLESKY;

566:   /* set solvertype */
567:   PetscCall(PetscFree(B->solvertype));
568:   PetscCall(PetscStrallocpy(MATSOLVERPASTIX, &B->solvertype));

570:   PetscCall(PetscNew(&pastix));

572:   pastix->CleanUpPastix = PETSC_FALSE;
573:   pastix->scat_rhs      = NULL;
574:   pastix->scat_sol      = NULL;
575:   B->ops->getinfo       = MatGetInfo_External;
576:   B->ops->destroy       = MatDestroy_Pastix;
577:   B->data               = (void *)pastix;
578:   *F                    = B;
579:   PetscFunctionReturn(PETSC_SUCCESS);
580: }

582: static PetscErrorCode MatGetFactor_mpisbaij_pastix(Mat A, MatFactorType ftype, Mat *F)
583: {
584:   Mat         B;
585:   Mat_Pastix *pastix;

587:   PetscFunctionBegin;
588:   PetscCheck(ftype == MAT_FACTOR_CHOLESKY, PETSC_COMM_SELF, PETSC_ERR_SUP, "Cannot use PETSc SBAIJ matrices with PaStiX LU, use AIJ matrix");

590:   /* Create the factorization matrix */
591:   PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
592:   PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, A->rmap->N, A->cmap->N));
593:   PetscCall(PetscStrallocpy("pastix", &((PetscObject)B)->type_name));
594:   PetscCall(MatSetUp(B));

596:   B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SBAIJPASTIX;
597:   B->ops->view                   = MatView_PaStiX;
598:   B->ops->getinfo                = MatGetInfo_PaStiX;
599:   B->ops->destroy                = MatDestroy_Pastix;
600:   PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_pastix));

602:   B->factortype = MAT_FACTOR_CHOLESKY;

604:   /* set solvertype */
605:   PetscCall(PetscFree(B->solvertype));
606:   PetscCall(PetscStrallocpy(MATSOLVERPASTIX, &B->solvertype));

608:   PetscCall(PetscNew(&pastix));

610:   pastix->CleanUpPastix = PETSC_FALSE;
611:   pastix->scat_rhs      = NULL;
612:   pastix->scat_sol      = NULL;
613:   B->data               = (void *)pastix;

615:   *F = B;
616:   PetscFunctionReturn(PETSC_SUCCESS);
617: }

619: PETSC_INTERN PetscErrorCode MatSolverTypeRegister_Pastix(void)
620: {
621:   PetscFunctionBegin;
622:   PetscCall(MatSolverTypeRegister(MATSOLVERPASTIX, MATMPIAIJ, MAT_FACTOR_LU, MatGetFactor_mpiaij_pastix));
623:   PetscCall(MatSolverTypeRegister(MATSOLVERPASTIX, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_pastix));
624:   PetscCall(MatSolverTypeRegister(MATSOLVERPASTIX, MATMPISBAIJ, MAT_FACTOR_CHOLESKY, MatGetFactor_mpisbaij_pastix));
625:   PetscCall(MatSolverTypeRegister(MATSOLVERPASTIX, MATSEQSBAIJ, MAT_FACTOR_CHOLESKY, MatGetFactor_seqsbaij_pastix));
626:   PetscFunctionReturn(PETSC_SUCCESS);
627: }