Actual source code: precon.c
1: /*
2: The PC (preconditioner) interface routines, callable by users.
3: */
4: #include <petsc/private/pcimpl.h>
5: #include <petscdm.h>
7: /* Logging support */
8: PetscClassId PC_CLASSID;
9: PetscLogEvent PC_SetUp, PC_SetUpOnBlocks, PC_Apply, PC_MatApply, PC_ApplyCoarse, PC_ApplyMultiple, PC_ApplySymmetricLeft;
10: PetscLogEvent PC_ApplySymmetricRight, PC_ModifySubMatrices, PC_ApplyOnBlocks, PC_ApplyTransposeOnBlocks;
11: PetscInt PetscMGLevelId;
12: PetscLogStage PCMPIStage;
14: PETSC_INTERN PetscErrorCode PCGetDefaultType_Private(PC pc, const char *type[])
15: {
16: PetscMPIInt size;
17: PetscBool hasopblock, hasopsolve, flg1, flg2, set, flg3, isnormal;
19: PetscFunctionBegin;
20: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)pc), &size));
21: if (pc->pmat) {
22: PetscCall(MatHasOperation(pc->pmat, MATOP_GET_DIAGONAL_BLOCK, &hasopblock));
23: PetscCall(MatHasOperation(pc->pmat, MATOP_SOLVE, &hasopsolve));
24: if (size == 1) {
25: PetscCall(MatGetFactorAvailable(pc->pmat, "petsc", MAT_FACTOR_ICC, &flg1));
26: PetscCall(MatGetFactorAvailable(pc->pmat, "petsc", MAT_FACTOR_ILU, &flg2));
27: PetscCall(MatIsSymmetricKnown(pc->pmat, &set, &flg3));
28: PetscCall(PetscObjectTypeCompareAny((PetscObject)pc->pmat, &isnormal, MATNORMAL, MATNORMALHERMITIAN, NULL));
29: if (flg1 && (!flg2 || (set && flg3))) {
30: *type = PCICC;
31: } else if (flg2) {
32: *type = PCILU;
33: } else if (isnormal) {
34: *type = PCNONE;
35: } else if (hasopblock) { /* likely is a parallel matrix run on one processor */
36: *type = PCBJACOBI;
37: } else if (hasopsolve) {
38: *type = PCMAT;
39: } else {
40: *type = PCNONE;
41: }
42: } else {
43: if (hasopblock) {
44: *type = PCBJACOBI;
45: } else if (hasopsolve) {
46: *type = PCMAT;
47: } else {
48: *type = PCNONE;
49: }
50: }
51: } else *type = NULL;
52: PetscFunctionReturn(PETSC_SUCCESS);
53: }
55: /* do not log solves, setup and applications of preconditioners while constructing preconditioners; perhaps they should be logged separately from the regular solves */
56: PETSC_EXTERN PetscLogEvent KSP_Solve, KSP_SetUp;
58: static PetscErrorCode PCLogEventsDeactivatePush(void)
59: {
60: PetscFunctionBegin;
61: PetscCall(KSPInitializePackage());
62: PetscCall(PetscLogEventDeactivatePush(KSP_Solve));
63: PetscCall(PetscLogEventDeactivatePush(KSP_SetUp));
64: PetscCall(PetscLogEventDeactivatePush(PC_Apply));
65: PetscCall(PetscLogEventDeactivatePush(PC_SetUp));
66: PetscCall(PetscLogEventDeactivatePush(PC_SetUpOnBlocks));
67: PetscFunctionReturn(PETSC_SUCCESS);
68: }
70: static PetscErrorCode PCLogEventsDeactivatePop(void)
71: {
72: PetscFunctionBegin;
73: PetscCall(KSPInitializePackage());
74: PetscCall(PetscLogEventDeactivatePop(KSP_Solve));
75: PetscCall(PetscLogEventDeactivatePop(KSP_SetUp));
76: PetscCall(PetscLogEventDeactivatePop(PC_Apply));
77: PetscCall(PetscLogEventDeactivatePop(PC_SetUp));
78: PetscCall(PetscLogEventDeactivatePop(PC_SetUpOnBlocks));
79: PetscFunctionReturn(PETSC_SUCCESS);
80: }
82: /*@
83: PCReset - Resets a `PC` context to the state it was in before `PCSetUp()` was called, and removes any allocated `Vec` and `Mat` from its data structure
85: Collective
87: Input Parameter:
88: . pc - the `PC` preconditioner context
90: Level: developer
92: Notes:
93: Any options set, including those set with `KSPSetFromOptions()` remain.
95: This allows a `PC` to be reused for a different sized linear system but using the same options that have been previously set in `pc`
97: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCSetUp()`
98: @*/
99: PetscErrorCode PCReset(PC pc)
100: {
101: PetscFunctionBegin;
103: PetscTryTypeMethod(pc, reset);
104: PetscCall(VecDestroy(&pc->diagonalscaleright));
105: PetscCall(VecDestroy(&pc->diagonalscaleleft));
106: PetscCall(MatDestroy(&pc->pmat));
107: PetscCall(MatDestroy(&pc->mat));
109: pc->setupcalled = 0;
110: PetscFunctionReturn(PETSC_SUCCESS);
111: }
113: /*@
114: PCDestroy - Destroys `PC` context that was created with `PCCreate()`.
116: Collective
118: Input Parameter:
119: . pc - the `PC` preconditioner context
121: Level: developer
123: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCSetUp()`
124: @*/
125: PetscErrorCode PCDestroy(PC *pc)
126: {
127: PetscFunctionBegin;
128: if (!*pc) PetscFunctionReturn(PETSC_SUCCESS);
130: if (--((PetscObject)*pc)->refct > 0) {
131: *pc = NULL;
132: PetscFunctionReturn(PETSC_SUCCESS);
133: }
135: PetscCall(PCReset(*pc));
137: /* if memory was published with SAWs then destroy it */
138: PetscCall(PetscObjectSAWsViewOff((PetscObject)*pc));
139: PetscTryTypeMethod(*pc, destroy);
140: PetscCall(DMDestroy(&(*pc)->dm));
141: PetscCall(PetscHeaderDestroy(pc));
142: PetscFunctionReturn(PETSC_SUCCESS);
143: }
145: /*@
146: PCGetDiagonalScale - Indicates if the preconditioner applies an additional left and right
147: scaling as needed by certain time-stepping codes.
149: Logically Collective
151: Input Parameter:
152: . pc - the `PC` preconditioner context
154: Output Parameter:
155: . flag - `PETSC_TRUE` if it applies the scaling
157: Level: developer
159: Note:
160: If this returns `PETSC_TRUE` then the system solved via the Krylov method is, for left and right preconditioning,
162: $$
163: \begin{align*}
164: D M A D^{-1} y = D M b \\
165: D A M D^{-1} z = D b.
166: \end{align*}
167: $$
169: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCSetUp()`, `PCDiagonalScaleLeft()`, `PCDiagonalScaleRight()`, `PCSetDiagonalScale()`
170: @*/
171: PetscErrorCode PCGetDiagonalScale(PC pc, PetscBool *flag)
172: {
173: PetscFunctionBegin;
175: PetscAssertPointer(flag, 2);
176: *flag = pc->diagonalscale;
177: PetscFunctionReturn(PETSC_SUCCESS);
178: }
180: /*@
181: PCSetDiagonalScale - Indicates the left scaling to use to apply an additional left and right
182: scaling as needed by certain time-stepping codes.
184: Logically Collective
186: Input Parameters:
187: + pc - the `PC` preconditioner context
188: - s - scaling vector
190: Level: intermediate
192: Notes:
193: The system solved via the Krylov method is, for left and right preconditioning,
194: $$
195: \begin{align*}
196: D M A D^{-1} y = D M b \\
197: D A M D^{-1} z = D b.
198: \end{align*}
199: $$
201: `PCDiagonalScaleLeft()` scales a vector by $D$. `PCDiagonalScaleRight()` scales a vector by $D^{-1}$.
203: .seealso: [](ch_ksp), `PCCreate()`, `PCSetUp()`, `PCDiagonalScaleLeft()`, `PCDiagonalScaleRight()`, `PCGetDiagonalScale()`
204: @*/
205: PetscErrorCode PCSetDiagonalScale(PC pc, Vec s)
206: {
207: PetscFunctionBegin;
210: pc->diagonalscale = PETSC_TRUE;
212: PetscCall(PetscObjectReference((PetscObject)s));
213: PetscCall(VecDestroy(&pc->diagonalscaleleft));
215: pc->diagonalscaleleft = s;
217: PetscCall(VecDuplicate(s, &pc->diagonalscaleright));
218: PetscCall(VecCopy(s, pc->diagonalscaleright));
219: PetscCall(VecReciprocal(pc->diagonalscaleright));
220: PetscFunctionReturn(PETSC_SUCCESS);
221: }
223: /*@
224: PCDiagonalScaleLeft - Scales a vector by the left scaling as needed by certain time-stepping codes.
226: Logically Collective
228: Input Parameters:
229: + pc - the `PC` preconditioner context
230: . in - input vector
231: - out - scaled vector (maybe the same as in)
233: Level: intermediate
235: Notes:
236: The system solved via the Krylov method is, for left and right preconditioning,
238: $$
239: \begin{align*}
240: D M A D^{-1} y = D M b \\
241: D A M D^{-1} z = D b.
242: \end{align*}
243: $$
245: `PCDiagonalScaleLeft()` scales a vector by $D$. `PCDiagonalScaleRight()` scales a vector by $D^{-1}$.
247: If diagonal scaling is turned off and `in` is not `out` then `in` is copied to `out`
249: .seealso: [](ch_ksp), `PCCreate()`, `PCSetUp()`, `PCSetDiagonalScale()`, `PCDiagonalScaleRight()`, `MatDiagonalScale()`
250: @*/
251: PetscErrorCode PCDiagonalScaleLeft(PC pc, Vec in, Vec out)
252: {
253: PetscFunctionBegin;
257: if (pc->diagonalscale) {
258: PetscCall(VecPointwiseMult(out, pc->diagonalscaleleft, in));
259: } else if (in != out) {
260: PetscCall(VecCopy(in, out));
261: }
262: PetscFunctionReturn(PETSC_SUCCESS);
263: }
265: /*@
266: PCDiagonalScaleRight - Scales a vector by the right scaling as needed by certain time-stepping codes.
268: Logically Collective
270: Input Parameters:
271: + pc - the `PC` preconditioner context
272: . in - input vector
273: - out - scaled vector (maybe the same as in)
275: Level: intermediate
277: Notes:
278: The system solved via the Krylov method is, for left and right preconditioning,
280: $$
281: \begin{align*}
282: D M A D^{-1} y = D M b \\
283: D A M D^{-1} z = D b.
284: \end{align*}
285: $$
287: `PCDiagonalScaleLeft()` scales a vector by $D$. `PCDiagonalScaleRight()` scales a vector by $D^{-1}$.
289: If diagonal scaling is turned off and `in` is not `out` then `in` is copied to `out`
291: .seealso: [](ch_ksp), `PCCreate()`, `PCSetUp()`, `PCDiagonalScaleLeft()`, `PCSetDiagonalScale()`, `MatDiagonalScale()`
292: @*/
293: PetscErrorCode PCDiagonalScaleRight(PC pc, Vec in, Vec out)
294: {
295: PetscFunctionBegin;
299: if (pc->diagonalscale) {
300: PetscCall(VecPointwiseMult(out, pc->diagonalscaleright, in));
301: } else if (in != out) {
302: PetscCall(VecCopy(in, out));
303: }
304: PetscFunctionReturn(PETSC_SUCCESS);
305: }
307: /*@
308: PCSetUseAmat - Sets a flag to indicate that when the preconditioner needs to apply (part of) the
309: operator during the preconditioning process it applies the Amat provided to `TSSetRHSJacobian()`,
310: `TSSetIJacobian()`, `SNESSetJacobian()`, `KSPSetOperators()` or `PCSetOperators()` not the Pmat.
312: Logically Collective
314: Input Parameters:
315: + pc - the `PC` preconditioner context
316: - flg - `PETSC_TRUE` to use the Amat, `PETSC_FALSE` to use the Pmat (default is false)
318: Options Database Key:
319: . -pc_use_amat <true,false> - use the amat argument to `KSPSetOperators()` or `PCSetOperators()` to apply the operator
321: Level: intermediate
323: Note:
324: For the common case in which the linear system matrix and the matrix used to construct the
325: preconditioner are identical, this routine has no affect.
327: .seealso: [](ch_ksp), `PC`, `PCGetUseAmat()`, `PCBJACOBI`, `PCMG`, `PCFIELDSPLIT`, `PCCOMPOSITE`,
328: `KSPSetOperators()`, `PCSetOperators()`
329: @*/
330: PetscErrorCode PCSetUseAmat(PC pc, PetscBool flg)
331: {
332: PetscFunctionBegin;
334: pc->useAmat = flg;
335: PetscFunctionReturn(PETSC_SUCCESS);
336: }
338: /*@
339: PCSetErrorIfFailure - Causes `PC` to generate an error if a floating point exception, for example a zero pivot, is detected.
341: Logically Collective
343: Input Parameters:
344: + pc - iterative context obtained from `PCCreate()`
345: - flg - `PETSC_TRUE` indicates you want the error generated
347: Level: advanced
349: Notes:
350: Normally PETSc continues if a linear solver fails due to a failed setup of a preconditioner, you can call `KSPGetConvergedReason()` after a `KSPSolve()`
351: to determine if it has converged or failed. Or use -ksp_error_if_not_converged to cause the program to terminate as soon as lack of convergence is
352: detected.
354: This is propagated into `KSP`s used by this `PC`, which then propagate it into `PC`s used by those `KSP`s
356: .seealso: [](ch_ksp), `PC`, `KSPSetErrorIfNotConverged()`, `PCGetInitialGuessNonzero()`, `PCSetInitialGuessKnoll()`, `PCGetInitialGuessKnoll()`
357: @*/
358: PetscErrorCode PCSetErrorIfFailure(PC pc, PetscBool flg)
359: {
360: PetscFunctionBegin;
363: pc->erroriffailure = flg;
364: PetscFunctionReturn(PETSC_SUCCESS);
365: }
367: /*@
368: PCGetUseAmat - Gets a flag to indicate that when the preconditioner needs to apply (part of) the
369: operator during the preconditioning process it applies the Amat provided to `TSSetRHSJacobian()`,
370: `TSSetIJacobian()`, `SNESSetJacobian()`, `KSPSetOperators()` or `PCSetOperators()` not the Pmat.
372: Logically Collective
374: Input Parameter:
375: . pc - the `PC` preconditioner context
377: Output Parameter:
378: . flg - `PETSC_TRUE` to use the Amat, `PETSC_FALSE` to use the Pmat (default is false)
380: Level: intermediate
382: Note:
383: For the common case in which the linear system matrix and the matrix used to construct the
384: preconditioner are identical, this routine is does nothing.
386: .seealso: [](ch_ksp), `PC`, `PCSetUseAmat()`, `PCBJACOBI`, `PCMG`, `PCFIELDSPLIT`, `PCCOMPOSITE`
387: @*/
388: PetscErrorCode PCGetUseAmat(PC pc, PetscBool *flg)
389: {
390: PetscFunctionBegin;
392: *flg = pc->useAmat;
393: PetscFunctionReturn(PETSC_SUCCESS);
394: }
396: /*@
397: PCSetKSPNestLevel - sets the amount of nesting the `KSP` that contains this `PC` has
399: Collective
401: Input Parameters:
402: + pc - the `PC`
403: - level - the nest level
405: Level: developer
407: .seealso: [](ch_ksp), `KSPSetUp()`, `KSPSolve()`, `KSPDestroy()`, `KSP`, `KSPGMRES`, `KSPType`, `KSPGetNestLevel()`, `PCGetKSPNestLevel()`, `KSPSetNestLevel()`
408: @*/
409: PetscErrorCode PCSetKSPNestLevel(PC pc, PetscInt level)
410: {
411: PetscFunctionBegin;
414: pc->kspnestlevel = level;
415: PetscFunctionReturn(PETSC_SUCCESS);
416: }
418: /*@
419: PCGetKSPNestLevel - gets the amount of nesting the `KSP` that contains this `PC` has
421: Not Collective
423: Input Parameter:
424: . pc - the `PC`
426: Output Parameter:
427: . level - the nest level
429: Level: developer
431: .seealso: [](ch_ksp), `KSPSetUp()`, `KSPSolve()`, `KSPDestroy()`, `KSP`, `KSPGMRES`, `KSPType`, `KSPSetNestLevel()`, `PCSetKSPNestLevel()`, `KSPGetNestLevel()`
432: @*/
433: PetscErrorCode PCGetKSPNestLevel(PC pc, PetscInt *level)
434: {
435: PetscFunctionBegin;
437: PetscAssertPointer(level, 2);
438: *level = pc->kspnestlevel;
439: PetscFunctionReturn(PETSC_SUCCESS);
440: }
442: /*@
443: PCCreate - Creates a preconditioner context, `PC`
445: Collective
447: Input Parameter:
448: . comm - MPI communicator
450: Output Parameter:
451: . newpc - location to put the `PC` preconditioner context
453: Level: developer
455: Notes:
456: This is rarely called directly by users since `KSP` manages the `PC` objects it uses. Use `KSPGetPC()` to access the `PC` used by a `KSP`.
458: Use `PCSetType()` or `PCSetFromOptions()` with the option `-pc_type pctype` to set the `PCType` for this `PC`
460: The default preconditioner type `PCType` for sparse matrices is `PCILU` or `PCICC` with 0 fill on one process and block Jacobi (`PCBJACOBI`) with `PCILU` or `PCICC`
461: in parallel. For dense matrices it is always `PCNONE`.
463: .seealso: [](ch_ksp), `PC`, `PCType`, `PCSetType`, `PCSetUp()`, `PCApply()`, `PCDestroy()`, `KSP`, `KSPGetPC()`
464: @*/
465: PetscErrorCode PCCreate(MPI_Comm comm, PC *newpc)
466: {
467: PC pc;
469: PetscFunctionBegin;
470: PetscAssertPointer(newpc, 2);
471: PetscCall(PCInitializePackage());
473: PetscCall(PetscHeaderCreate(pc, PC_CLASSID, "PC", "Preconditioner", "PC", comm, PCDestroy, PCView));
474: pc->mat = NULL;
475: pc->pmat = NULL;
476: pc->setupcalled = 0;
477: pc->setfromoptionscalled = 0;
478: pc->data = NULL;
479: pc->diagonalscale = PETSC_FALSE;
480: pc->diagonalscaleleft = NULL;
481: pc->diagonalscaleright = NULL;
483: pc->modifysubmatrices = NULL;
484: pc->modifysubmatricesP = NULL;
486: *newpc = pc;
487: PetscFunctionReturn(PETSC_SUCCESS);
488: }
490: /*@
491: PCApply - Applies the preconditioner to a vector.
493: Collective
495: Input Parameters:
496: + pc - the `PC` preconditioner context
497: - x - input vector
499: Output Parameter:
500: . y - output vector
502: Level: developer
504: .seealso: [](ch_ksp), `PC`, `PCApplyTranspose()`, `PCApplyBAorAB()`
505: @*/
506: PetscErrorCode PCApply(PC pc, Vec x, Vec y)
507: {
508: PetscInt m, n, mv, nv;
510: PetscFunctionBegin;
514: PetscCheck(x != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "x and y must be different vectors");
515: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
516: /* use pmat to check vector sizes since for KSPLSQR the pmat may be of a different size than mat */
517: PetscCall(MatGetLocalSize(pc->pmat, &m, &n));
518: PetscCall(VecGetLocalSize(x, &mv));
519: PetscCall(VecGetLocalSize(y, &nv));
520: /* check pmat * y = x is feasible */
521: PetscCheck(mv == m, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Preconditioner number of local rows %" PetscInt_FMT " does not equal input vector size %" PetscInt_FMT, m, mv);
522: PetscCheck(nv == n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Preconditioner number of local columns %" PetscInt_FMT " does not equal output vector size %" PetscInt_FMT, n, nv);
523: PetscCall(VecSetErrorIfLocked(y, 3));
525: PetscCall(PCSetUp(pc));
526: PetscCall(VecLockReadPush(x));
527: PetscCall(PetscLogEventBegin(PC_Apply, pc, x, y, 0));
528: PetscUseTypeMethod(pc, apply, x, y);
529: PetscCall(PetscLogEventEnd(PC_Apply, pc, x, y, 0));
530: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
531: PetscCall(VecLockReadPop(x));
532: PetscFunctionReturn(PETSC_SUCCESS);
533: }
535: /*@
536: PCMatApply - Applies the preconditioner to multiple vectors stored as a `MATDENSE`. Like `PCApply()`, `Y` and `X` must be different matrices.
538: Collective
540: Input Parameters:
541: + pc - the `PC` preconditioner context
542: - X - block of input vectors
544: Output Parameter:
545: . Y - block of output vectors
547: Level: developer
549: .seealso: [](ch_ksp), `PC`, `PCApply()`, `KSPMatSolve()`
550: @*/
551: PetscErrorCode PCMatApply(PC pc, Mat X, Mat Y)
552: {
553: Mat A;
554: Vec cy, cx;
555: PetscInt m1, M1, m2, M2, n1, N1, n2, N2, m3, M3, n3, N3;
556: PetscBool match;
558: PetscFunctionBegin;
562: PetscCheckSameComm(pc, 1, X, 2);
563: PetscCheckSameComm(pc, 1, Y, 3);
564: PetscCheck(Y != X, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "Y and X must be different matrices");
565: PetscCall(PCGetOperators(pc, NULL, &A));
566: PetscCall(MatGetLocalSize(A, &m3, &n3));
567: PetscCall(MatGetLocalSize(X, &m2, &n2));
568: PetscCall(MatGetLocalSize(Y, &m1, &n1));
569: PetscCall(MatGetSize(A, &M3, &N3));
570: PetscCall(MatGetSize(X, &M2, &N2));
571: PetscCall(MatGetSize(Y, &M1, &N1));
572: PetscCheck(n1 == n2 && N1 == N2, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Incompatible number of columns between block of input vectors (n,N) = (%" PetscInt_FMT ",%" PetscInt_FMT ") and block of output vectors (n,N) = (%" PetscInt_FMT ",%" PetscInt_FMT ")", n2, N2, n1, N1);
573: PetscCheck(m2 == m3 && M2 == M3, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Incompatible layout between block of input vectors (m,M) = (%" PetscInt_FMT ",%" PetscInt_FMT ") and Pmat (m,M)x(n,N) = (%" PetscInt_FMT ",%" PetscInt_FMT ")x(%" PetscInt_FMT ",%" PetscInt_FMT ")", m2, M2, m3, M3, n3, N3);
574: PetscCheck(m1 == n3 && M1 == N3, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Incompatible layout between block of output vectors (m,M) = (%" PetscInt_FMT ",%" PetscInt_FMT ") and Pmat (m,M)x(n,N) = (%" PetscInt_FMT ",%" PetscInt_FMT ")x(%" PetscInt_FMT ",%" PetscInt_FMT ")", m1, M1, m3, M3, n3, N3);
575: PetscCall(PetscObjectBaseTypeCompareAny((PetscObject)Y, &match, MATSEQDENSE, MATMPIDENSE, ""));
576: PetscCheck(match, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Provided block of output vectors not stored in a dense Mat");
577: PetscCall(PetscObjectBaseTypeCompareAny((PetscObject)X, &match, MATSEQDENSE, MATMPIDENSE, ""));
578: PetscCheck(match, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Provided block of input vectors not stored in a dense Mat");
579: PetscCall(PCSetUp(pc));
580: if (pc->ops->matapply) {
581: PetscCall(PetscLogEventBegin(PC_MatApply, pc, X, Y, 0));
582: PetscUseTypeMethod(pc, matapply, X, Y);
583: PetscCall(PetscLogEventEnd(PC_MatApply, pc, X, Y, 0));
584: } else {
585: PetscCall(PetscInfo(pc, "PC type %s applying column by column\n", ((PetscObject)pc)->type_name));
586: for (n1 = 0; n1 < N1; ++n1) {
587: PetscCall(MatDenseGetColumnVecRead(X, n1, &cx));
588: PetscCall(MatDenseGetColumnVecWrite(Y, n1, &cy));
589: PetscCall(PCApply(pc, cx, cy));
590: PetscCall(MatDenseRestoreColumnVecWrite(Y, n1, &cy));
591: PetscCall(MatDenseRestoreColumnVecRead(X, n1, &cx));
592: }
593: }
594: PetscFunctionReturn(PETSC_SUCCESS);
595: }
597: /*@
598: PCApplySymmetricLeft - Applies the left part of a symmetric preconditioner to a vector.
600: Collective
602: Input Parameters:
603: + pc - the `PC` preconditioner context
604: - x - input vector
606: Output Parameter:
607: . y - output vector
609: Level: developer
611: Note:
612: Currently, this routine is implemented only for `PCICC` and `PCJACOBI` preconditioners.
614: .seealso: [](ch_ksp), `PC`, `PCApply()`, `PCApplySymmetricRight()`
615: @*/
616: PetscErrorCode PCApplySymmetricLeft(PC pc, Vec x, Vec y)
617: {
618: PetscFunctionBegin;
622: PetscCheck(x != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "x and y must be different vectors");
623: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
624: PetscCall(PCSetUp(pc));
625: PetscCall(VecLockReadPush(x));
626: PetscCall(PetscLogEventBegin(PC_ApplySymmetricLeft, pc, x, y, 0));
627: PetscUseTypeMethod(pc, applysymmetricleft, x, y);
628: PetscCall(PetscLogEventEnd(PC_ApplySymmetricLeft, pc, x, y, 0));
629: PetscCall(VecLockReadPop(x));
630: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
631: PetscFunctionReturn(PETSC_SUCCESS);
632: }
634: /*@
635: PCApplySymmetricRight - Applies the right part of a symmetric preconditioner to a vector.
637: Collective
639: Input Parameters:
640: + pc - the `PC` preconditioner context
641: - x - input vector
643: Output Parameter:
644: . y - output vector
646: Level: developer
648: Note:
649: Currently, this routine is implemented only for `PCICC` and `PCJACOBI` preconditioners.
651: .seealso: [](ch_ksp), `PC`, `PCApply()`, `PCApplySymmetricLeft()`
652: @*/
653: PetscErrorCode PCApplySymmetricRight(PC pc, Vec x, Vec y)
654: {
655: PetscFunctionBegin;
659: PetscCheck(x != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "x and y must be different vectors");
660: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
661: PetscCall(PCSetUp(pc));
662: PetscCall(VecLockReadPush(x));
663: PetscCall(PetscLogEventBegin(PC_ApplySymmetricRight, pc, x, y, 0));
664: PetscUseTypeMethod(pc, applysymmetricright, x, y);
665: PetscCall(PetscLogEventEnd(PC_ApplySymmetricRight, pc, x, y, 0));
666: PetscCall(VecLockReadPop(x));
667: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
668: PetscFunctionReturn(PETSC_SUCCESS);
669: }
671: /*@
672: PCApplyTranspose - Applies the transpose of preconditioner to a vector.
674: Collective
676: Input Parameters:
677: + pc - the `PC` preconditioner context
678: - x - input vector
680: Output Parameter:
681: . y - output vector
683: Level: developer
685: Note:
686: For complex numbers this applies the non-Hermitian transpose.
688: Developer Note:
689: We need to implement a `PCApplyHermitianTranspose()`
691: .seealso: [](ch_ksp), `PC`, `PCApply()`, `PCApplyBAorAB()`, `PCApplyBAorABTranspose()`, `PCApplyTransposeExists()`
692: @*/
693: PetscErrorCode PCApplyTranspose(PC pc, Vec x, Vec y)
694: {
695: PetscFunctionBegin;
699: PetscCheck(x != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "x and y must be different vectors");
700: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
701: PetscCall(PCSetUp(pc));
702: PetscCall(VecLockReadPush(x));
703: PetscCall(PetscLogEventBegin(PC_Apply, pc, x, y, 0));
704: PetscUseTypeMethod(pc, applytranspose, x, y);
705: PetscCall(PetscLogEventEnd(PC_Apply, pc, x, y, 0));
706: PetscCall(VecLockReadPop(x));
707: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
708: PetscFunctionReturn(PETSC_SUCCESS);
709: }
711: /*@
712: PCApplyTransposeExists - Test whether the preconditioner has a transpose apply operation
714: Collective
716: Input Parameter:
717: . pc - the `PC` preconditioner context
719: Output Parameter:
720: . flg - `PETSC_TRUE` if a transpose operation is defined
722: Level: developer
724: .seealso: [](ch_ksp), `PC`, `PCApplyTranspose()`
725: @*/
726: PetscErrorCode PCApplyTransposeExists(PC pc, PetscBool *flg)
727: {
728: PetscFunctionBegin;
730: PetscAssertPointer(flg, 2);
731: if (pc->ops->applytranspose) *flg = PETSC_TRUE;
732: else *flg = PETSC_FALSE;
733: PetscFunctionReturn(PETSC_SUCCESS);
734: }
736: /*@
737: PCApplyBAorAB - Applies the preconditioner and operator to a vector. $y = B*A*x $ or $ y = A*B*x$.
739: Collective
741: Input Parameters:
742: + pc - the `PC` preconditioner context
743: . side - indicates the preconditioner side, one of `PC_LEFT`, `PC_RIGHT`, or `PC_SYMMETRIC`
744: . x - input vector
745: - work - work vector
747: Output Parameter:
748: . y - output vector
750: Level: developer
752: Note:
753: If the `PC` has had `PCSetDiagonalScale()` set then $ D M A D^{-1} $ for left preconditioning or $ D A M D^{-1} $ is actually applied.
754: The specific `KSPSolve()` method must also be written to handle the post-solve "correction" for the diagonal scaling.
756: .seealso: [](ch_ksp), `PC`, `PCApply()`, `PCApplyTranspose()`, `PCApplyBAorABTranspose()`
757: @*/
758: PetscErrorCode PCApplyBAorAB(PC pc, PCSide side, Vec x, Vec y, Vec work)
759: {
760: PetscFunctionBegin;
766: PetscCheckSameComm(pc, 1, x, 3);
767: PetscCheckSameComm(pc, 1, y, 4);
768: PetscCheckSameComm(pc, 1, work, 5);
769: PetscCheck(x != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "x and y must be different vectors");
770: PetscCheck(side == PC_LEFT || side == PC_SYMMETRIC || side == PC_RIGHT, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_OUTOFRANGE, "Side must be right, left, or symmetric");
771: PetscCheck(!pc->diagonalscale || side != PC_SYMMETRIC, PetscObjectComm((PetscObject)pc), PETSC_ERR_SUP, "Cannot include diagonal scaling with symmetric preconditioner application");
772: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(x, 3, PETSC_TRUE));
774: PetscCall(PCSetUp(pc));
775: if (pc->diagonalscale) {
776: if (pc->ops->applyBA) {
777: Vec work2; /* this is expensive, but to fix requires a second work vector argument to PCApplyBAorAB() */
778: PetscCall(VecDuplicate(x, &work2));
779: PetscCall(PCDiagonalScaleRight(pc, x, work2));
780: PetscUseTypeMethod(pc, applyBA, side, work2, y, work);
781: PetscCall(PCDiagonalScaleLeft(pc, y, y));
782: PetscCall(VecDestroy(&work2));
783: } else if (side == PC_RIGHT) {
784: PetscCall(PCDiagonalScaleRight(pc, x, y));
785: PetscCall(PCApply(pc, y, work));
786: PetscCall(MatMult(pc->mat, work, y));
787: PetscCall(PCDiagonalScaleLeft(pc, y, y));
788: } else if (side == PC_LEFT) {
789: PetscCall(PCDiagonalScaleRight(pc, x, y));
790: PetscCall(MatMult(pc->mat, y, work));
791: PetscCall(PCApply(pc, work, y));
792: PetscCall(PCDiagonalScaleLeft(pc, y, y));
793: } else PetscCheck(side != PC_SYMMETRIC, PetscObjectComm((PetscObject)pc), PETSC_ERR_SUP, "Cannot provide diagonal scaling with symmetric application of preconditioner");
794: } else {
795: if (pc->ops->applyBA) {
796: PetscUseTypeMethod(pc, applyBA, side, x, y, work);
797: } else if (side == PC_RIGHT) {
798: PetscCall(PCApply(pc, x, work));
799: PetscCall(MatMult(pc->mat, work, y));
800: } else if (side == PC_LEFT) {
801: PetscCall(MatMult(pc->mat, x, work));
802: PetscCall(PCApply(pc, work, y));
803: } else if (side == PC_SYMMETRIC) {
804: /* There's an extra copy here; maybe should provide 2 work vectors instead? */
805: PetscCall(PCApplySymmetricRight(pc, x, work));
806: PetscCall(MatMult(pc->mat, work, y));
807: PetscCall(VecCopy(y, work));
808: PetscCall(PCApplySymmetricLeft(pc, work, y));
809: }
810: }
811: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 4, PETSC_FALSE));
812: PetscFunctionReturn(PETSC_SUCCESS);
813: }
815: /*@
816: PCApplyBAorABTranspose - Applies the transpose of the preconditioner
817: and operator to a vector. That is, applies $B^T * A^T$ with left preconditioning,
818: NOT $(B*A)^T = A^T*B^T$.
820: Collective
822: Input Parameters:
823: + pc - the `PC` preconditioner context
824: . side - indicates the preconditioner side, one of `PC_LEFT`, `PC_RIGHT`, or `PC_SYMMETRIC`
825: . x - input vector
826: - work - work vector
828: Output Parameter:
829: . y - output vector
831: Level: developer
833: Note:
834: This routine is used internally so that the same Krylov code can be used to solve $A x = b$ and $A^T x = b$, with a preconditioner
835: defined by $B^T$. This is why this has the funny form that it computes $B^T * A^T$
837: .seealso: [](ch_ksp), `PC`, `PCApply()`, `PCApplyTranspose()`, `PCApplyBAorAB()`
838: @*/
839: PetscErrorCode PCApplyBAorABTranspose(PC pc, PCSide side, Vec x, Vec y, Vec work)
840: {
841: PetscFunctionBegin;
846: PetscCheck(x != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "x and y must be different vectors");
847: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(x, 3, PETSC_TRUE));
848: if (pc->ops->applyBAtranspose) {
849: PetscUseTypeMethod(pc, applyBAtranspose, side, x, y, work);
850: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 4, PETSC_FALSE));
851: PetscFunctionReturn(PETSC_SUCCESS);
852: }
853: PetscCheck(side == PC_LEFT || side == PC_RIGHT, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_OUTOFRANGE, "Side must be right or left");
855: PetscCall(PCSetUp(pc));
856: if (side == PC_RIGHT) {
857: PetscCall(PCApplyTranspose(pc, x, work));
858: PetscCall(MatMultTranspose(pc->mat, work, y));
859: } else if (side == PC_LEFT) {
860: PetscCall(MatMultTranspose(pc->mat, x, work));
861: PetscCall(PCApplyTranspose(pc, work, y));
862: }
863: /* add support for PC_SYMMETRIC */
864: if (pc->erroriffailure) PetscCall(VecValidValues_Internal(y, 4, PETSC_FALSE));
865: PetscFunctionReturn(PETSC_SUCCESS);
866: }
868: /*@
869: PCApplyRichardsonExists - Determines whether a particular preconditioner has a
870: built-in fast application of Richardson's method.
872: Not Collective
874: Input Parameter:
875: . pc - the preconditioner
877: Output Parameter:
878: . exists - `PETSC_TRUE` or `PETSC_FALSE`
880: Level: developer
882: .seealso: [](ch_ksp), `PC`, `KSPRICHARDSON`, `PCApplyRichardson()`
883: @*/
884: PetscErrorCode PCApplyRichardsonExists(PC pc, PetscBool *exists)
885: {
886: PetscFunctionBegin;
888: PetscAssertPointer(exists, 2);
889: if (pc->ops->applyrichardson) *exists = PETSC_TRUE;
890: else *exists = PETSC_FALSE;
891: PetscFunctionReturn(PETSC_SUCCESS);
892: }
894: /*@
895: PCApplyRichardson - Applies several steps of Richardson iteration with
896: the particular preconditioner. This routine is usually used by the
897: Krylov solvers and not the application code directly.
899: Collective
901: Input Parameters:
902: + pc - the `PC` preconditioner context
903: . b - the right-hand side
904: . w - one work vector
905: . rtol - relative decrease in residual norm convergence criteria
906: . abstol - absolute residual norm convergence criteria
907: . dtol - divergence residual norm increase criteria
908: . its - the number of iterations to apply.
909: - guesszero - if the input x contains nonzero initial guess
911: Output Parameters:
912: + outits - number of iterations actually used (for SOR this always equals its)
913: . reason - the reason the apply terminated
914: - y - the solution (also contains initial guess if guesszero is `PETSC_FALSE`
916: Level: developer
918: Notes:
919: Most preconditioners do not support this function. Use the command
920: `PCApplyRichardsonExists()` to determine if one does.
922: Except for the `PCMG` this routine ignores the convergence tolerances
923: and always runs for the number of iterations
925: .seealso: [](ch_ksp), `PC`, `PCApplyRichardsonExists()`
926: @*/
927: PetscErrorCode PCApplyRichardson(PC pc, Vec b, Vec y, Vec w, PetscReal rtol, PetscReal abstol, PetscReal dtol, PetscInt its, PetscBool guesszero, PetscInt *outits, PCRichardsonConvergedReason *reason)
928: {
929: PetscFunctionBegin;
934: PetscCheck(b != y, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_IDN, "b and y must be different vectors");
935: PetscCall(PCSetUp(pc));
936: PetscUseTypeMethod(pc, applyrichardson, b, y, w, rtol, abstol, dtol, its, guesszero, outits, reason);
937: PetscFunctionReturn(PETSC_SUCCESS);
938: }
940: /*@
941: PCSetFailedReason - Sets the reason a `PCSetUp()` failed or `PC_NOERROR` if it did not fail
943: Logically Collective
945: Input Parameters:
946: + pc - the `PC` preconditioner context
947: - reason - the reason it failed
949: Level: advanced
951: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCApply()`, `PCDestroy()`, `PCFailedReason`
952: @*/
953: PetscErrorCode PCSetFailedReason(PC pc, PCFailedReason reason)
954: {
955: PetscFunctionBegin;
957: pc->failedreason = reason;
958: PetscFunctionReturn(PETSC_SUCCESS);
959: }
961: /*@
962: PCGetFailedReason - Gets the reason a `PCSetUp()` failed or `PC_NOERROR` if it did not fail
964: Not Collective
966: Input Parameter:
967: . pc - the `PC` preconditioner context
969: Output Parameter:
970: . reason - the reason it failed
972: Level: advanced
974: Note:
975: After a call to `KSPCheckDot()` or `KSPCheckNorm()` inside a `KSPSolve()` or a call to `PCReduceFailedReason()`
976: this is the maximum reason over all MPI processes in the `PC` communicator and hence logically collective.
977: Otherwise it returns the local value.
979: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCApply()`, `PCDestroy()`, `PCSetFailedReason()`, `PCFailedReason`
980: @*/
981: PetscErrorCode PCGetFailedReason(PC pc, PCFailedReason *reason)
982: {
983: PetscFunctionBegin;
985: if (pc->setupcalled < 0) *reason = (PCFailedReason)pc->setupcalled;
986: else *reason = pc->failedreason;
987: PetscFunctionReturn(PETSC_SUCCESS);
988: }
990: /*@
991: PCReduceFailedReason - Reduce the failed reason among the MPI processes that share the `PC`
993: Collective
995: Input Parameter:
996: . pc - the `PC` preconditioner context
998: Level: advanced
1000: Note:
1001: Different MPI processes may have different reasons or no reason, see `PCGetFailedReason()`. This routine
1002: makes them have a common value (failure if any MPI process had a failure).
1004: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCApply()`, `PCDestroy()`, `PCGetFailedReason()`, `PCSetFailedReason()`, `PCFailedReason`
1005: @*/
1006: PetscErrorCode PCReduceFailedReason(PC pc)
1007: {
1008: PetscInt buf;
1010: PetscFunctionBegin;
1012: buf = (PetscInt)pc->failedreason;
1013: PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &buf, 1, MPIU_INT, MPI_MAX, PetscObjectComm((PetscObject)pc)));
1014: pc->failedreason = (PCFailedReason)buf;
1015: PetscFunctionReturn(PETSC_SUCCESS);
1016: }
1018: /*
1019: a setupcall of 0 indicates never setup,
1020: 1 indicates has been previously setup
1021: -1 indicates a PCSetUp() was attempted and failed
1022: */
1023: /*@
1024: PCSetUp - Prepares for the use of a preconditioner. Performs all the one-time operations needed before the preconditioner
1025: can be used with `PCApply()`
1027: Collective
1029: Input Parameter:
1030: . pc - the `PC` preconditioner context
1032: Level: developer
1034: Notes:
1035: For example, for `PCLU` this will compute the factorization.
1037: This is called automatically by `KSPSetUp()` or `PCApply()` so rarely needs to be called directly.
1039: For nested preconditioners, such as `PCFIELDSPLIT` or `PCBJACOBI` this may not finish the construction of the preconditioner
1040: on the inner levels, the routine `PCSetUpOnBlocks()` may compute more of the preconditioner in those situations.
1042: .seealso: [](ch_ksp), `PC`, `PCCreate()`, `PCApply()`, `PCDestroy()`, `KSPSetUp()`, `PCSetUpOnBlocks()`
1043: @*/
1044: PetscErrorCode PCSetUp(PC pc)
1045: {
1046: const char *def;
1047: PetscObjectState matstate, matnonzerostate;
1049: PetscFunctionBegin;
1051: PetscCheck(pc->mat, PetscObjectComm((PetscObject)pc), PETSC_ERR_ARG_WRONGSTATE, "Matrix must be set first");
1053: if (pc->setupcalled && pc->reusepreconditioner) {
1054: PetscCall(PetscInfo(pc, "Leaving PC with identical preconditioner since reuse preconditioner is set\n"));
1055: PetscFunctionReturn(PETSC_SUCCESS);
1056: }
1058: PetscCall(PetscObjectStateGet((PetscObject)pc->pmat, &matstate));
1059: PetscCall(MatGetNonzeroState(pc->pmat, &matnonzerostate));
1060: if (!pc->setupcalled) {
1061: //PetscCall(PetscInfo(pc, "Setting up PC for first time\n"));
1062: pc->flag = DIFFERENT_NONZERO_PATTERN;
1063: } else if (matstate == pc->matstate) PetscFunctionReturn(PETSC_SUCCESS);
1064: else {
1065: if (matnonzerostate != pc->matnonzerostate) {
1066: PetscCall(PetscInfo(pc, "Setting up PC with different nonzero pattern\n"));
1067: pc->flag = DIFFERENT_NONZERO_PATTERN;
1068: } else {
1069: //PetscCall(PetscInfo(pc, "Setting up PC with same nonzero pattern\n"));
1070: pc->flag = SAME_NONZERO_PATTERN;
1071: }
1072: }
1073: pc->matstate = matstate;
1074: pc->matnonzerostate = matnonzerostate;
1076: if (!((PetscObject)pc)->type_name) {
1077: PetscCall(PCGetDefaultType_Private(pc, &def));
1078: PetscCall(PCSetType(pc, def));
1079: }
1081: PetscCall(MatSetErrorIfFailure(pc->pmat, pc->erroriffailure));
1082: PetscCall(MatSetErrorIfFailure(pc->mat, pc->erroriffailure));
1083: PetscCall(PetscLogEventBegin(PC_SetUp, pc, 0, 0, 0));
1084: if (pc->ops->setup) {
1085: PetscCall(PCLogEventsDeactivatePush());
1086: PetscUseTypeMethod(pc, setup);
1087: PetscCall(PCLogEventsDeactivatePop());
1088: }
1089: PetscCall(PetscLogEventEnd(PC_SetUp, pc, 0, 0, 0));
1090: if (!pc->setupcalled) pc->setupcalled = 1;
1091: PetscFunctionReturn(PETSC_SUCCESS);
1092: }
1094: /*@
1095: PCSetUpOnBlocks - Sets up the preconditioner for each block in
1096: the block Jacobi, overlapping Schwarz, and fieldsplit methods.
1098: Collective
1100: Input Parameter:
1101: . pc - the `PC` preconditioner context
1103: Level: developer
1105: Notes:
1106: For nested preconditioners such as `PCBJACOBI`, `PCSetUp()` is not called on each sub-`KSP` when `PCSetUp()` is
1107: called on the outer `PC`, this routine ensures it is called.
1109: It calls `PCSetUp()` if not yet called.
1111: .seealso: [](ch_ksp), `PC`, `PCSetUp()`, `PCCreate()`, `PCApply()`, `PCDestroy()`
1112: @*/
1113: PetscErrorCode PCSetUpOnBlocks(PC pc)
1114: {
1115: PetscFunctionBegin;
1117: if (!pc->setupcalled) PetscCall(PCSetUp(pc)); /* "if" to prevent -info extra prints */
1118: if (!pc->ops->setuponblocks) PetscFunctionReturn(PETSC_SUCCESS);
1119: PetscCall(PetscLogEventBegin(PC_SetUpOnBlocks, pc, 0, 0, 0));
1120: PetscCall(PCLogEventsDeactivatePush());
1121: PetscUseTypeMethod(pc, setuponblocks);
1122: PetscCall(PCLogEventsDeactivatePop());
1123: PetscCall(PetscLogEventEnd(PC_SetUpOnBlocks, pc, 0, 0, 0));
1124: PetscFunctionReturn(PETSC_SUCCESS);
1125: }
1127: /*@C
1128: PCSetModifySubMatrices - Sets a user-defined routine for modifying the
1129: submatrices that arise within certain subdomain-based preconditioners such as `PCASM`
1131: Logically Collective
1133: Input Parameters:
1134: + pc - the `PC` preconditioner context
1135: . func - routine for modifying the submatrices
1136: - ctx - optional user-defined context (may be `NULL`)
1138: Calling sequence of `func`:
1139: + pc - the `PC` preconditioner context
1140: . nsub - number of index sets
1141: . row - an array of index sets that contain the global row numbers
1142: that comprise each local submatrix
1143: . col - an array of index sets that contain the global column numbers
1144: that comprise each local submatrix
1145: . submat - array of local submatrices
1146: - ctx - optional user-defined context for private data for the
1147: user-defined func routine (may be `NULL`)
1149: Level: advanced
1151: Notes:
1152: The basic submatrices are extracted from the preconditioner matrix as
1153: usual; the user can then alter these (for example, to set different boundary
1154: conditions for each submatrix) before they are used for the local solves.
1156: `PCSetModifySubMatrices()` MUST be called before `KSPSetUp()` and
1157: `KSPSolve()`.
1159: A routine set by `PCSetModifySubMatrices()` is currently called within
1160: the block Jacobi (`PCBJACOBI`) and additive Schwarz (`PCASM`)
1161: preconditioners. All other preconditioners ignore this routine.
1163: .seealso: [](ch_ksp), `PC`, `PCBJACOBI`, `PCASM`, `PCModifySubMatrices()`
1164: @*/
1165: PetscErrorCode PCSetModifySubMatrices(PC pc, PetscErrorCode (*func)(PC pc, PetscInt nsub, const IS row[], const IS col[], Mat submat[], void *ctx), void *ctx)
1166: {
1167: PetscFunctionBegin;
1169: pc->modifysubmatrices = func;
1170: pc->modifysubmatricesP = ctx;
1171: PetscFunctionReturn(PETSC_SUCCESS);
1172: }
1174: /*@C
1175: PCModifySubMatrices - Calls an optional user-defined routine within
1176: certain preconditioners if one has been set with `PCSetModifySubMatrices()`.
1178: Collective
1180: Input Parameters:
1181: + pc - the `PC` preconditioner context
1182: . nsub - the number of local submatrices
1183: . row - an array of index sets that contain the global row numbers
1184: that comprise each local submatrix
1185: . col - an array of index sets that contain the global column numbers
1186: that comprise each local submatrix
1187: . submat - array of local submatrices
1188: - ctx - optional user-defined context for private data for the
1189: user-defined routine (may be `NULL`)
1191: Output Parameter:
1192: . submat - array of local submatrices (the entries of which may
1193: have been modified)
1195: Level: developer
1197: Note:
1198: The user should NOT generally call this routine, as it will
1199: automatically be called within certain preconditioners.
1201: .seealso: [](ch_ksp), `PC`, `PCSetModifySubMatrices()`
1202: @*/
1203: PetscErrorCode PCModifySubMatrices(PC pc, PetscInt nsub, const IS row[], const IS col[], Mat submat[], void *ctx)
1204: {
1205: PetscFunctionBegin;
1207: if (!pc->modifysubmatrices) PetscFunctionReturn(PETSC_SUCCESS);
1208: PetscCall(PetscLogEventBegin(PC_ModifySubMatrices, pc, 0, 0, 0));
1209: PetscCall((*pc->modifysubmatrices)(pc, nsub, row, col, submat, ctx));
1210: PetscCall(PetscLogEventEnd(PC_ModifySubMatrices, pc, 0, 0, 0));
1211: PetscFunctionReturn(PETSC_SUCCESS);
1212: }
1214: /*@
1215: PCSetOperators - Sets the matrix associated with the linear system and
1216: a (possibly) different one associated with the preconditioner.
1218: Logically Collective
1220: Input Parameters:
1221: + pc - the `PC` preconditioner context
1222: . Amat - the matrix that defines the linear system
1223: - Pmat - the matrix to be used in constructing the preconditioner, usually the same as Amat.
1225: Level: intermediate
1227: Notes:
1228: Passing a `NULL` for `Amat` or `Pmat` removes the matrix that is currently used.
1230: If you wish to replace either `Amat` or `Pmat` but leave the other one untouched then
1231: first call `KSPGetOperators()` to get the one you wish to keep, call `PetscObjectReference()`
1232: on it and then pass it back in in your call to `KSPSetOperators()`.
1234: More Notes about Repeated Solution of Linear Systems:
1235: PETSc does NOT reset the matrix entries of either `Amat` or `Pmat`
1236: to zero after a linear solve; the user is completely responsible for
1237: matrix assembly. See the routine `MatZeroEntries()` if desiring to
1238: zero all elements of a matrix.
1240: .seealso: [](ch_ksp), `PC`, `PCGetOperators()`, `MatZeroEntries()`
1241: @*/
1242: PetscErrorCode PCSetOperators(PC pc, Mat Amat, Mat Pmat)
1243: {
1244: PetscInt m1, n1, m2, n2;
1246: PetscFunctionBegin;
1250: if (Amat) PetscCheckSameComm(pc, 1, Amat, 2);
1251: if (Pmat) PetscCheckSameComm(pc, 1, Pmat, 3);
1252: if (pc->setupcalled && pc->mat && pc->pmat && Amat && Pmat) {
1253: PetscCall(MatGetLocalSize(Amat, &m1, &n1));
1254: PetscCall(MatGetLocalSize(pc->mat, &m2, &n2));
1255: PetscCheck(m1 == m2 && n1 == n2, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Cannot change local size of Amat after use old sizes %" PetscInt_FMT " %" PetscInt_FMT " new sizes %" PetscInt_FMT " %" PetscInt_FMT, m2, n2, m1, n1);
1256: PetscCall(MatGetLocalSize(Pmat, &m1, &n1));
1257: PetscCall(MatGetLocalSize(pc->pmat, &m2, &n2));
1258: PetscCheck(m1 == m2 && n1 == n2, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Cannot change local size of Pmat after use old sizes %" PetscInt_FMT " %" PetscInt_FMT " new sizes %" PetscInt_FMT " %" PetscInt_FMT, m2, n2, m1, n1);
1259: }
1261: if (Pmat != pc->pmat) {
1262: /* changing the operator that defines the preconditioner thus reneed to clear current states so new preconditioner is built */
1263: pc->matnonzerostate = -1;
1264: pc->matstate = -1;
1265: }
1267: /* reference first in case the matrices are the same */
1268: if (Amat) PetscCall(PetscObjectReference((PetscObject)Amat));
1269: PetscCall(MatDestroy(&pc->mat));
1270: if (Pmat) PetscCall(PetscObjectReference((PetscObject)Pmat));
1271: PetscCall(MatDestroy(&pc->pmat));
1272: pc->mat = Amat;
1273: pc->pmat = Pmat;
1274: PetscFunctionReturn(PETSC_SUCCESS);
1275: }
1277: /*@
1278: PCSetReusePreconditioner - reuse the current preconditioner even if the operator in the preconditioner `PC` has changed.
1280: Logically Collective
1282: Input Parameters:
1283: + pc - the `PC` preconditioner context
1284: - flag - `PETSC_TRUE` do not compute a new preconditioner, `PETSC_FALSE` do compute a new preconditioner
1286: Level: intermediate
1288: Note:
1289: Normally if a matrix inside a `PC` changes the `PC` automatically updates itself using information from the changed matrix. This option
1290: prevents this.
1292: .seealso: [](ch_ksp), `PC`, `PCGetOperators()`, `MatZeroEntries()`, `PCGetReusePreconditioner()`, `KSPSetReusePreconditioner()`
1293: @*/
1294: PetscErrorCode PCSetReusePreconditioner(PC pc, PetscBool flag)
1295: {
1296: PetscFunctionBegin;
1299: pc->reusepreconditioner = flag;
1300: PetscFunctionReturn(PETSC_SUCCESS);
1301: }
1303: /*@
1304: PCGetReusePreconditioner - Determines if the `PC` reuses the current preconditioner even if the operator in the preconditioner has changed.
1306: Not Collective
1308: Input Parameter:
1309: . pc - the `PC` preconditioner context
1311: Output Parameter:
1312: . flag - `PETSC_TRUE` do not compute a new preconditioner, `PETSC_FALSE` do compute a new preconditioner
1314: Level: intermediate
1316: .seealso: [](ch_ksp), `PC`, `PCGetOperators()`, `MatZeroEntries()`, `PCSetReusePreconditioner()`
1317: @*/
1318: PetscErrorCode PCGetReusePreconditioner(PC pc, PetscBool *flag)
1319: {
1320: PetscFunctionBegin;
1322: PetscAssertPointer(flag, 2);
1323: *flag = pc->reusepreconditioner;
1324: PetscFunctionReturn(PETSC_SUCCESS);
1325: }
1327: /*@
1328: PCGetOperators - Gets the matrix associated with the linear system and
1329: possibly a different one which is used to construct the preconditioner.
1331: Not Collective, though parallel `Mat`s are returned if `pc` is parallel
1333: Input Parameter:
1334: . pc - the `PC` preconditioner context
1336: Output Parameters:
1337: + Amat - the matrix defining the linear system
1338: - Pmat - the matrix from which the preconditioner is constructed, usually the same as Amat.
1340: Level: intermediate
1342: Note:
1343: Does not increase the reference count of the matrices, so you should not destroy them
1345: Alternative usage: If the operators have NOT been set with `KSPSetOperators()`/`PCSetOperators()` then the operators
1346: are created in `PC` and returned to the user. In this case, if both operators
1347: mat and pmat are requested, two DIFFERENT operators will be returned. If
1348: only one is requested both operators in the PC will be the same (i.e. as
1349: if one had called `KSPSetOperators()`/`PCSetOperators()` with the same argument for both Mats).
1350: The user must set the sizes of the returned matrices and their type etc just
1351: as if the user created them with `MatCreate()`. For example,
1353: .vb
1354: KSP/PCGetOperators(ksp/pc,&Amat,NULL); is equivalent to
1355: set size, type, etc of Amat
1357: MatCreate(comm,&mat);
1358: KSP/PCSetOperators(ksp/pc,Amat,Amat);
1359: PetscObjectDereference((PetscObject)mat);
1360: set size, type, etc of Amat
1361: .ve
1363: and
1365: .vb
1366: KSP/PCGetOperators(ksp/pc,&Amat,&Pmat); is equivalent to
1367: set size, type, etc of Amat and Pmat
1369: MatCreate(comm,&Amat);
1370: MatCreate(comm,&Pmat);
1371: KSP/PCSetOperators(ksp/pc,Amat,Pmat);
1372: PetscObjectDereference((PetscObject)Amat);
1373: PetscObjectDereference((PetscObject)Pmat);
1374: set size, type, etc of Amat and Pmat
1375: .ve
1377: The rationale for this support is so that when creating a `TS`, `SNES`, or `KSP` the hierarchy
1378: of underlying objects (i.e. `SNES`, `KSP`, `PC`, `Mat`) and their lifespans can be completely
1379: managed by the top most level object (i.e. the `TS`, `SNES`, or `KSP`). Another way to look
1380: at this is when you create a `SNES` you do not NEED to create a `KSP` and attach it to
1381: the `SNES` object (the `SNES` object manages it for you). Similarly when you create a KSP
1382: you do not need to attach a `PC` to it (the `KSP` object manages the `PC` object for you).
1383: Thus, why should YOU have to create the `Mat` and attach it to the `SNES`/`KSP`/`PC`, when
1384: it can be created for you?
1386: .seealso: [](ch_ksp), `PC`, `PCSetOperators()`, `KSPGetOperators()`, `KSPSetOperators()`, `PCGetOperatorsSet()`
1387: @*/
1388: PetscErrorCode PCGetOperators(PC pc, Mat *Amat, Mat *Pmat)
1389: {
1390: PetscFunctionBegin;
1392: if (Amat) {
1393: if (!pc->mat) {
1394: if (pc->pmat && !Pmat) { /* Pmat has been set, but user did not request it, so use for Amat */
1395: pc->mat = pc->pmat;
1396: PetscCall(PetscObjectReference((PetscObject)pc->mat));
1397: } else { /* both Amat and Pmat are empty */
1398: PetscCall(MatCreate(PetscObjectComm((PetscObject)pc), &pc->mat));
1399: if (!Pmat) { /* user did NOT request Pmat, so make same as Amat */
1400: pc->pmat = pc->mat;
1401: PetscCall(PetscObjectReference((PetscObject)pc->pmat));
1402: }
1403: }
1404: }
1405: *Amat = pc->mat;
1406: }
1407: if (Pmat) {
1408: if (!pc->pmat) {
1409: if (pc->mat && !Amat) { /* Amat has been set but was not requested, so use for pmat */
1410: pc->pmat = pc->mat;
1411: PetscCall(PetscObjectReference((PetscObject)pc->pmat));
1412: } else {
1413: PetscCall(MatCreate(PetscObjectComm((PetscObject)pc), &pc->pmat));
1414: if (!Amat) { /* user did NOT request Amat, so make same as Pmat */
1415: pc->mat = pc->pmat;
1416: PetscCall(PetscObjectReference((PetscObject)pc->mat));
1417: }
1418: }
1419: }
1420: *Pmat = pc->pmat;
1421: }
1422: PetscFunctionReturn(PETSC_SUCCESS);
1423: }
1425: /*@
1426: PCGetOperatorsSet - Determines if the matrix associated with the linear system and
1427: possibly a different one associated with the preconditioner have been set in the `PC`.
1429: Not Collective, though the results on all processes should be the same
1431: Input Parameter:
1432: . pc - the `PC` preconditioner context
1434: Output Parameters:
1435: + mat - the matrix associated with the linear system was set
1436: - pmat - matrix associated with the preconditioner was set, usually the same
1438: Level: intermediate
1440: .seealso: [](ch_ksp), `PC`, `PCSetOperators()`, `KSPGetOperators()`, `KSPSetOperators()`, `PCGetOperators()`
1441: @*/
1442: PetscErrorCode PCGetOperatorsSet(PC pc, PetscBool *mat, PetscBool *pmat)
1443: {
1444: PetscFunctionBegin;
1446: if (mat) *mat = (pc->mat) ? PETSC_TRUE : PETSC_FALSE;
1447: if (pmat) *pmat = (pc->pmat) ? PETSC_TRUE : PETSC_FALSE;
1448: PetscFunctionReturn(PETSC_SUCCESS);
1449: }
1451: /*@
1452: PCFactorGetMatrix - Gets the factored matrix from the
1453: preconditioner context. This routine is valid only for the `PCLU`,
1454: `PCILU`, `PCCHOLESKY`, and `PCICC` methods.
1456: Not Collective though `mat` is parallel if `pc` is parallel
1458: Input Parameter:
1459: . pc - the `PC` preconditioner context
1461: Output Parameters:
1462: . mat - the factored matrix
1464: Level: advanced
1466: Note:
1467: Does not increase the reference count for `mat` so DO NOT destroy it
1469: .seealso: [](ch_ksp), `PC`, `PCLU`, `PCILU`, `PCCHOLESKY`, `PCICC`
1470: @*/
1471: PetscErrorCode PCFactorGetMatrix(PC pc, Mat *mat)
1472: {
1473: PetscFunctionBegin;
1475: PetscAssertPointer(mat, 2);
1476: PetscCall(PCFactorSetUpMatSolverType(pc));
1477: PetscUseTypeMethod(pc, getfactoredmatrix, mat);
1478: PetscFunctionReturn(PETSC_SUCCESS);
1479: }
1481: /*@
1482: PCSetOptionsPrefix - Sets the prefix used for searching for all
1483: `PC` options in the database.
1485: Logically Collective
1487: Input Parameters:
1488: + pc - the `PC` preconditioner context
1489: - prefix - the prefix string to prepend to all `PC` option requests
1491: Note:
1492: A hyphen (-) must NOT be given at the beginning of the prefix name.
1493: The first character of all runtime options is AUTOMATICALLY the
1494: hyphen.
1496: Level: advanced
1498: .seealso: [](ch_ksp), `PC`, `PCSetFromOptions`, `PCAppendOptionsPrefix()`, `PCGetOptionsPrefix()`
1499: @*/
1500: PetscErrorCode PCSetOptionsPrefix(PC pc, const char prefix[])
1501: {
1502: PetscFunctionBegin;
1504: PetscCall(PetscObjectSetOptionsPrefix((PetscObject)pc, prefix));
1505: PetscFunctionReturn(PETSC_SUCCESS);
1506: }
1508: /*@
1509: PCAppendOptionsPrefix - Appends to the prefix used for searching for all
1510: `PC` options in the database.
1512: Logically Collective
1514: Input Parameters:
1515: + pc - the `PC` preconditioner context
1516: - prefix - the prefix string to prepend to all `PC` option requests
1518: Note:
1519: A hyphen (-) must NOT be given at the beginning of the prefix name.
1520: The first character of all runtime options is AUTOMATICALLY the
1521: hyphen.
1523: Level: advanced
1525: .seealso: [](ch_ksp), `PC`, `PCSetFromOptions`, `PCSetOptionsPrefix()`, `PCGetOptionsPrefix()`
1526: @*/
1527: PetscErrorCode PCAppendOptionsPrefix(PC pc, const char prefix[])
1528: {
1529: PetscFunctionBegin;
1531: PetscCall(PetscObjectAppendOptionsPrefix((PetscObject)pc, prefix));
1532: PetscFunctionReturn(PETSC_SUCCESS);
1533: }
1535: /*@
1536: PCGetOptionsPrefix - Gets the prefix used for searching for all
1537: PC options in the database.
1539: Not Collective
1541: Input Parameter:
1542: . pc - the `PC` preconditioner context
1544: Output Parameter:
1545: . prefix - pointer to the prefix string used, is returned
1547: Level: advanced
1549: Fortran Note:
1550: The user should pass in a string `prefix` of
1551: sufficient length to hold the prefix.
1553: .seealso: [](ch_ksp), `PC`, `PCSetFromOptions`, `PCSetOptionsPrefix()`, `PCAppendOptionsPrefix()`
1554: @*/
1555: PetscErrorCode PCGetOptionsPrefix(PC pc, const char *prefix[])
1556: {
1557: PetscFunctionBegin;
1559: PetscAssertPointer(prefix, 2);
1560: PetscCall(PetscObjectGetOptionsPrefix((PetscObject)pc, prefix));
1561: PetscFunctionReturn(PETSC_SUCCESS);
1562: }
1564: /*
1565: Indicates the right-hand side will be changed by KSPSolve(), this occurs for a few
1566: preconditioners including BDDC and Eisentat that transform the equations before applying
1567: the Krylov methods
1568: */
1569: PETSC_INTERN PetscErrorCode PCPreSolveChangeRHS(PC pc, PetscBool *change)
1570: {
1571: PetscFunctionBegin;
1573: PetscAssertPointer(change, 2);
1574: *change = PETSC_FALSE;
1575: PetscTryMethod(pc, "PCPreSolveChangeRHS_C", (PC, PetscBool *), (pc, change));
1576: PetscFunctionReturn(PETSC_SUCCESS);
1577: }
1579: /*@
1580: PCPreSolve - Optional pre-solve phase, intended for any preconditioner-specific actions that must be performed before
1581: the iterative solve itself. Used in conjunction with `PCPostSolve()`
1583: Collective
1585: Input Parameters:
1586: + pc - the `PC` preconditioner context
1587: - ksp - the Krylov subspace context
1589: Level: developer
1591: Notes:
1592: `KSPSolve()` calls this directly, so is rarely called by the user.
1594: Certain preconditioners, such as the `PCType` of `PCEISENSTAT`, change the formulation of the linear system to be solved iteratively.
1595: This function performs that transformation. `PCPostSolve()` then transforms the system back to its original form after the solve.
1596: `PCPostSolve()` also transforms the resulting solution of the transformed system to the solution of the original problem.
1598: `KSPSetPreSolve()` and `KSPSetPostSolve()` provide an alternative way to provide such transformations.
1600: .seealso: [](ch_ksp), `PC`, `PCPostSolve()`, `KSP`, `PCSetPreSolve()`, `KSPSetPreSolve()`, `KSPSetPostSolve()`
1601: @*/
1602: PetscErrorCode PCPreSolve(PC pc, KSP ksp)
1603: {
1604: Vec x, rhs;
1606: PetscFunctionBegin;
1609: pc->presolvedone++;
1610: PetscCheck(pc->presolvedone <= 2, PetscObjectComm((PetscObject)pc), PETSC_ERR_SUP, "Cannot embed PCPreSolve() more than twice");
1611: PetscCall(KSPGetSolution(ksp, &x));
1612: PetscCall(KSPGetRhs(ksp, &rhs));
1614: if (pc->ops->presolve) PetscUseTypeMethod(pc, presolve, ksp, rhs, x);
1615: else if (pc->presolve) PetscCall(pc->presolve(pc, ksp));
1616: PetscFunctionReturn(PETSC_SUCCESS);
1617: }
1619: /*@C
1620: PCSetPreSolve - Sets function used by `PCPreSolve()` which is intended for any
1621: preconditioner-specific actions that must be performed before
1622: the iterative solve itself.
1624: Logically Collective
1626: Input Parameters:
1627: + pc - the preconditioner object
1628: - presolve - the function to call before the solve
1630: Calling sequence of `presolve`:
1631: + pc - the `PC` context
1632: - ksp - the `KSP` context
1634: Level: developer
1636: .seealso: [](ch_ksp), `PC`, `PCSetUp()`, `PCPreSolve()`
1637: @*/
1638: PetscErrorCode PCSetPreSolve(PC pc, PetscErrorCode (*presolve)(PC pc, KSP ksp))
1639: {
1640: PetscFunctionBegin;
1642: pc->presolve = presolve;
1643: PetscFunctionReturn(PETSC_SUCCESS);
1644: }
1646: /*@
1647: PCPostSolve - Optional post-solve phase, intended for any
1648: preconditioner-specific actions that must be performed after
1649: the iterative solve itself.
1651: Collective
1653: Input Parameters:
1654: + pc - the `PC` preconditioner context
1655: - ksp - the `KSP` Krylov subspace context
1657: Example Usage:
1658: .vb
1659: PCPreSolve(pc,ksp);
1660: KSPSolve(ksp,b,x);
1661: PCPostSolve(pc,ksp);
1662: .ve
1664: Level: developer
1666: Note:
1667: `KSPSolve()` calls this routine directly, so it is rarely called by the user.
1669: .seealso: [](ch_ksp), `PC`, `PCSetPreSolve()`, `KSPSetPostSolve()`, `KSPSetPreSolve()`, `PCPreSolve()`, `KSPSolve()`
1670: @*/
1671: PetscErrorCode PCPostSolve(PC pc, KSP ksp)
1672: {
1673: Vec x, rhs;
1675: PetscFunctionBegin;
1678: pc->presolvedone--;
1679: PetscCall(KSPGetSolution(ksp, &x));
1680: PetscCall(KSPGetRhs(ksp, &rhs));
1681: PetscTryTypeMethod(pc, postsolve, ksp, rhs, x);
1682: PetscFunctionReturn(PETSC_SUCCESS);
1683: }
1685: /*@
1686: PCLoad - Loads a `PC` that has been stored in binary with `PCView()`.
1688: Collective
1690: Input Parameters:
1691: + newdm - the newly loaded `PC`, this needs to have been created with `PCCreate()` or
1692: some related function before a call to `PCLoad()`.
1693: - viewer - binary file viewer `PETSCVIEWERBINARY`, obtained from `PetscViewerBinaryOpen()`
1695: Level: intermediate
1697: Note:
1698: The type is determined by the data in the file, any `PCType` set into the `PC` before this call is ignored.
1700: .seealso: [](ch_ksp), `PC`, `PetscViewerBinaryOpen()`, `PCView()`, `MatLoad()`, `VecLoad()`, `PETSCVIEWERBINARY`
1701: @*/
1702: PetscErrorCode PCLoad(PC newdm, PetscViewer viewer)
1703: {
1704: PetscBool isbinary;
1705: PetscInt classid;
1706: char type[256];
1708: PetscFunctionBegin;
1711: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary));
1712: PetscCheck(isbinary, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Invalid viewer; open viewer with PetscViewerBinaryOpen()");
1714: PetscCall(PetscViewerBinaryRead(viewer, &classid, 1, NULL, PETSC_INT));
1715: PetscCheck(classid == PC_FILE_CLASSID, PetscObjectComm((PetscObject)newdm), PETSC_ERR_ARG_WRONG, "Not PC next in file");
1716: PetscCall(PetscViewerBinaryRead(viewer, type, 256, NULL, PETSC_CHAR));
1717: PetscCall(PCSetType(newdm, type));
1718: PetscTryTypeMethod(newdm, load, viewer);
1719: PetscFunctionReturn(PETSC_SUCCESS);
1720: }
1722: #include <petscdraw.h>
1723: #if defined(PETSC_HAVE_SAWS)
1724: #include <petscviewersaws.h>
1725: #endif
1727: /*@
1728: PCViewFromOptions - View (print or provide information about) the `PC`, based on options in the options database
1730: Collective
1732: Input Parameters:
1733: + A - the `PC` context
1734: . obj - Optional object that provides the options prefix
1735: - name - command line option name
1737: Level: developer
1739: .seealso: [](ch_ksp), `PC`, `PCView`, `PetscObjectViewFromOptions()`, `PCCreate()`
1740: @*/
1741: PetscErrorCode PCViewFromOptions(PC A, PetscObject obj, const char name[])
1742: {
1743: PetscFunctionBegin;
1745: PetscCall(PetscObjectViewFromOptions((PetscObject)A, obj, name));
1746: PetscFunctionReturn(PETSC_SUCCESS);
1747: }
1749: /*@
1750: PCView - Prints information about the `PC`
1752: Collective
1754: Input Parameters:
1755: + pc - the `PC` preconditioner context
1756: - viewer - optional `PetscViewer` visualization context
1758: Level: intermediate
1760: Notes:
1761: The available visualization contexts include
1762: + `PETSC_VIEWER_STDOUT_SELF` - standard output (default)
1763: - `PETSC_VIEWER_STDOUT_WORLD` - synchronized standard
1764: output where only the first processor opens
1765: the file. All other processors send their
1766: data to the first processor to print.
1768: The user can open an alternative visualization contexts with
1769: `PetscViewerASCIIOpen()` (output to a specified file).
1771: .seealso: [](ch_ksp), `PC`, `PetscViewer`, `PetscViewerType`, `KSPView()`, `PetscViewerASCIIOpen()`
1772: @*/
1773: PetscErrorCode PCView(PC pc, PetscViewer viewer)
1774: {
1775: PCType cstr;
1776: PetscViewerFormat format;
1777: PetscBool iascii, isstring, isbinary, isdraw, pop = PETSC_FALSE;
1778: #if defined(PETSC_HAVE_SAWS)
1779: PetscBool issaws;
1780: #endif
1782: PetscFunctionBegin;
1784: if (!viewer) PetscCall(PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)pc), &viewer));
1786: PetscCheckSameComm(pc, 1, viewer, 2);
1788: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
1789: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSTRING, &isstring));
1790: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary));
1791: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERDRAW, &isdraw));
1792: #if defined(PETSC_HAVE_SAWS)
1793: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSAWS, &issaws));
1794: #endif
1796: if (iascii) {
1797: PetscCall(PetscObjectPrintClassNamePrefixType((PetscObject)pc, viewer));
1798: if (!pc->setupcalled) PetscCall(PetscViewerASCIIPrintf(viewer, " PC has not been set up so information may be incomplete\n"));
1799: PetscCall(PetscViewerASCIIPushTab(viewer));
1800: PetscTryTypeMethod(pc, view, viewer);
1801: PetscCall(PetscViewerASCIIPopTab(viewer));
1802: if (pc->mat) {
1803: PetscCall(PetscViewerGetFormat(viewer, &format));
1804: if (format != PETSC_VIEWER_ASCII_INFO_DETAIL) {
1805: PetscCall(PetscViewerPushFormat(viewer, PETSC_VIEWER_ASCII_INFO));
1806: pop = PETSC_TRUE;
1807: }
1808: if (pc->pmat == pc->mat) {
1809: PetscCall(PetscViewerASCIIPrintf(viewer, " linear system matrix = precond matrix:\n"));
1810: PetscCall(PetscViewerASCIIPushTab(viewer));
1811: PetscCall(MatView(pc->mat, viewer));
1812: PetscCall(PetscViewerASCIIPopTab(viewer));
1813: } else {
1814: if (pc->pmat) {
1815: PetscCall(PetscViewerASCIIPrintf(viewer, " linear system matrix followed by preconditioner matrix:\n"));
1816: } else {
1817: PetscCall(PetscViewerASCIIPrintf(viewer, " linear system matrix:\n"));
1818: }
1819: PetscCall(PetscViewerASCIIPushTab(viewer));
1820: PetscCall(MatView(pc->mat, viewer));
1821: if (pc->pmat) PetscCall(MatView(pc->pmat, viewer));
1822: PetscCall(PetscViewerASCIIPopTab(viewer));
1823: }
1824: if (pop) PetscCall(PetscViewerPopFormat(viewer));
1825: }
1826: } else if (isstring) {
1827: PetscCall(PCGetType(pc, &cstr));
1828: PetscCall(PetscViewerStringSPrintf(viewer, " PCType: %-7.7s", cstr));
1829: PetscTryTypeMethod(pc, view, viewer);
1830: if (pc->mat) PetscCall(MatView(pc->mat, viewer));
1831: if (pc->pmat && pc->pmat != pc->mat) PetscCall(MatView(pc->pmat, viewer));
1832: } else if (isbinary) {
1833: PetscInt classid = PC_FILE_CLASSID;
1834: MPI_Comm comm;
1835: PetscMPIInt rank;
1836: char type[256];
1838: PetscCall(PetscObjectGetComm((PetscObject)pc, &comm));
1839: PetscCallMPI(MPI_Comm_rank(comm, &rank));
1840: if (rank == 0) {
1841: PetscCall(PetscViewerBinaryWrite(viewer, &classid, 1, PETSC_INT));
1842: PetscCall(PetscStrncpy(type, ((PetscObject)pc)->type_name, 256));
1843: PetscCall(PetscViewerBinaryWrite(viewer, type, 256, PETSC_CHAR));
1844: }
1845: PetscTryTypeMethod(pc, view, viewer);
1846: } else if (isdraw) {
1847: PetscDraw draw;
1848: char str[25];
1849: PetscReal x, y, bottom, h;
1850: PetscInt n;
1852: PetscCall(PetscViewerDrawGetDraw(viewer, 0, &draw));
1853: PetscCall(PetscDrawGetCurrentPoint(draw, &x, &y));
1854: if (pc->mat) {
1855: PetscCall(MatGetSize(pc->mat, &n, NULL));
1856: PetscCall(PetscSNPrintf(str, 25, "PC: %s (%" PetscInt_FMT ")", ((PetscObject)pc)->type_name, n));
1857: } else {
1858: PetscCall(PetscSNPrintf(str, 25, "PC: %s", ((PetscObject)pc)->type_name));
1859: }
1860: PetscCall(PetscDrawStringBoxed(draw, x, y, PETSC_DRAW_RED, PETSC_DRAW_BLACK, str, NULL, &h));
1861: bottom = y - h;
1862: PetscCall(PetscDrawPushCurrentPoint(draw, x, bottom));
1863: PetscTryTypeMethod(pc, view, viewer);
1864: PetscCall(PetscDrawPopCurrentPoint(draw));
1865: #if defined(PETSC_HAVE_SAWS)
1866: } else if (issaws) {
1867: PetscMPIInt rank;
1869: PetscCall(PetscObjectName((PetscObject)pc));
1870: PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
1871: if (!((PetscObject)pc)->amsmem && rank == 0) PetscCall(PetscObjectViewSAWs((PetscObject)pc, viewer));
1872: if (pc->mat) PetscCall(MatView(pc->mat, viewer));
1873: if (pc->pmat && pc->pmat != pc->mat) PetscCall(MatView(pc->pmat, viewer));
1874: #endif
1875: }
1876: PetscFunctionReturn(PETSC_SUCCESS);
1877: }
1879: /*@C
1880: PCRegister - Adds a method (`PCType`) to the PETSc preconditioner package.
1882: Not collective. No Fortran Support
1884: Input Parameters:
1885: + sname - name of a new user-defined solver
1886: - function - routine to create the method context which will be stored in a `PC` when `PCSetType()` is called
1888: Example Usage:
1889: .vb
1890: PCRegister("my_solver", MySolverCreate);
1891: .ve
1893: Then, your solver can be chosen with the procedural interface via
1894: $ PCSetType(pc, "my_solver")
1895: or at runtime via the option
1896: $ -pc_type my_solver
1898: Level: advanced
1900: Note:
1901: A simpler alternative to using `PCRegister()` for an application specific preconditioner is to use a `PC` of `PCType` `PCSHELL` and
1902: provide your customizations with `PCShellSetContext()` and `PCShellSetApply()`
1904: `PCRegister()` may be called multiple times to add several user-defined preconditioners.
1906: .seealso: [](ch_ksp), `PC`, `PCType`, `PCRegisterAll()`, `PCSetType()`, `PCShellSetContext()`, `PCShellSetApply()`, `PCSHELL`
1907: @*/
1908: PetscErrorCode PCRegister(const char sname[], PetscErrorCode (*function)(PC))
1909: {
1910: PetscFunctionBegin;
1911: PetscCall(PCInitializePackage());
1912: PetscCall(PetscFunctionListAdd(&PCList, sname, function));
1913: PetscFunctionReturn(PETSC_SUCCESS);
1914: }
1916: static PetscErrorCode MatMult_PC(Mat A, Vec X, Vec Y)
1917: {
1918: PC pc;
1920: PetscFunctionBegin;
1921: PetscCall(MatShellGetContext(A, &pc));
1922: PetscCall(PCApply(pc, X, Y));
1923: PetscFunctionReturn(PETSC_SUCCESS);
1924: }
1926: /*@
1927: PCComputeOperator - Computes the explicit preconditioned operator as a matrix `Mat`.
1929: Collective
1931: Input Parameters:
1932: + pc - the `PC` preconditioner object
1933: - mattype - the `MatType` to be used for the operator
1935: Output Parameter:
1936: . mat - the explicit preconditioned operator
1938: Level: advanced
1940: Note:
1941: This computation is done by applying the operators to columns of the identity matrix.
1942: This routine is costly in general, and is recommended for use only with relatively small systems.
1943: Currently, this routine uses a dense matrix format when `mattype` == `NULL`
1945: Developer Note:
1946: This should be called `PCCreateExplicitOperator()`
1948: .seealso: [](ch_ksp), `PC`, `KSPComputeOperator()`, `MatType`
1949: @*/
1950: PetscErrorCode PCComputeOperator(PC pc, MatType mattype, Mat *mat)
1951: {
1952: PetscInt N, M, m, n;
1953: Mat A, Apc;
1955: PetscFunctionBegin;
1957: PetscAssertPointer(mat, 3);
1958: PetscCall(PCGetOperators(pc, &A, NULL));
1959: PetscCall(MatGetLocalSize(A, &m, &n));
1960: PetscCall(MatGetSize(A, &M, &N));
1961: PetscCall(MatCreateShell(PetscObjectComm((PetscObject)pc), m, n, M, N, pc, &Apc));
1962: PetscCall(MatShellSetOperation(Apc, MATOP_MULT, (void (*)(void))MatMult_PC));
1963: PetscCall(MatComputeOperator(Apc, mattype, mat));
1964: PetscCall(MatDestroy(&Apc));
1965: PetscFunctionReturn(PETSC_SUCCESS);
1966: }
1968: /*@
1969: PCSetCoordinates - sets the coordinates of all the nodes (degrees of freedom in the vector) on the local process
1971: Collective
1973: Input Parameters:
1974: + pc - the `PC` preconditioner context
1975: . dim - the dimension of the coordinates 1, 2, or 3
1976: . nloc - the blocked size of the coordinates array
1977: - coords - the coordinates array
1979: Level: intermediate
1981: Notes:
1982: `coords` is an array of the dim coordinates for the nodes on
1983: the local processor, of size `dim`*`nloc`.
1984: If there are 108 equations (dofs) on a processor
1985: for a 3d displacement finite element discretization of elasticity (so
1986: that there are nloc = 36 = 108/3 nodes) then the array must have 108
1987: double precision values (ie, 3 * 36). These x y z coordinates
1988: should be ordered for nodes 0 to N-1 like so: [ 0.x, 0.y, 0.z, 1.x,
1989: ... , N-1.z ].
1991: The information provided here can be used by some preconditioners, such as `PCGAMG`, to produce a better preconditioner.
1992: See also `MatSetNearNullSpace()`.
1994: .seealso: [](ch_ksp), `PC`, `MatSetNearNullSpace()`
1995: @*/
1996: PetscErrorCode PCSetCoordinates(PC pc, PetscInt dim, PetscInt nloc, PetscReal coords[])
1997: {
1998: PetscFunctionBegin;
2001: PetscTryMethod(pc, "PCSetCoordinates_C", (PC, PetscInt, PetscInt, PetscReal[]), (pc, dim, nloc, coords));
2002: PetscFunctionReturn(PETSC_SUCCESS);
2003: }
2005: /*@
2006: PCGetInterpolations - Gets interpolation matrices for all levels (except level 0)
2008: Logically Collective
2010: Input Parameter:
2011: . pc - the precondition context
2013: Output Parameters:
2014: + num_levels - the number of levels
2015: - interpolations - the interpolation matrices (size of `num_levels`-1)
2017: Level: advanced
2019: Developer Note:
2020: Why is this here instead of in `PCMG` etc?
2022: .seealso: [](ch_ksp), `PC`, `PCMG`, `PCMGGetRestriction()`, `PCMGSetInterpolation()`, `PCMGGetInterpolation()`, `PCGetCoarseOperators()`
2023: @*/
2024: PetscErrorCode PCGetInterpolations(PC pc, PetscInt *num_levels, Mat *interpolations[])
2025: {
2026: PetscFunctionBegin;
2028: PetscAssertPointer(num_levels, 2);
2029: PetscAssertPointer(interpolations, 3);
2030: PetscUseMethod(pc, "PCGetInterpolations_C", (PC, PetscInt *, Mat *[]), (pc, num_levels, interpolations));
2031: PetscFunctionReturn(PETSC_SUCCESS);
2032: }
2034: /*@
2035: PCGetCoarseOperators - Gets coarse operator matrices for all levels (except the finest level)
2037: Logically Collective
2039: Input Parameter:
2040: . pc - the precondition context
2042: Output Parameters:
2043: + num_levels - the number of levels
2044: - coarseOperators - the coarse operator matrices (size of `num_levels`-1)
2046: Level: advanced
2048: Developer Note:
2049: Why is this here instead of in `PCMG` etc?
2051: .seealso: [](ch_ksp), `PC`, `PCMG`, `PCMGGetRestriction()`, `PCMGSetInterpolation()`, `PCMGGetRScale()`, `PCMGGetInterpolation()`, `PCGetInterpolations()`
2052: @*/
2053: PetscErrorCode PCGetCoarseOperators(PC pc, PetscInt *num_levels, Mat *coarseOperators[])
2054: {
2055: PetscFunctionBegin;
2057: PetscAssertPointer(num_levels, 2);
2058: PetscAssertPointer(coarseOperators, 3);
2059: PetscUseMethod(pc, "PCGetCoarseOperators_C", (PC, PetscInt *, Mat *[]), (pc, num_levels, coarseOperators));
2060: PetscFunctionReturn(PETSC_SUCCESS);
2061: }