Actual source code: kspimpl.h
1: #pragma once
3: #include <petscksp.h>
4: #include <petscds.h>
5: #include <petsc/private/petscimpl.h>
7: /* SUBMANSEC = KSP */
9: PETSC_EXTERN PetscBool KSPRegisterAllCalled;
10: PETSC_EXTERN PetscBool KSPMonitorRegisterAllCalled;
11: PETSC_EXTERN PetscErrorCode KSPRegisterAll(void);
12: PETSC_EXTERN PetscErrorCode KSPMonitorRegisterAll(void);
13: PETSC_EXTERN PetscErrorCode KSPGuessRegisterAll(void);
14: PETSC_EXTERN PetscErrorCode KSPMatRegisterAll(void);
16: typedef struct _KSPOps *KSPOps;
18: struct _KSPOps {
19: PetscErrorCode (*buildsolution)(KSP, Vec, Vec *); /* Returns a pointer to the solution, or
20: calculates the solution in a
21: user-provided area. */
22: PetscErrorCode (*buildresidual)(KSP, Vec, Vec, Vec *); /* Returns a pointer to the residual, or
23: calculates the residual in a
24: user-provided area. */
25: PetscErrorCode (*solve)(KSP); /* actual solver */
26: PetscErrorCode (*matsolve)(KSP, Mat, Mat); /* multiple dense RHS solver */
27: PetscErrorCode (*setup)(KSP);
28: PetscErrorCode (*setfromoptions)(KSP, PetscOptionItems *);
29: PetscErrorCode (*publishoptions)(KSP);
30: PetscErrorCode (*computeextremesingularvalues)(KSP, PetscReal *, PetscReal *);
31: PetscErrorCode (*computeeigenvalues)(KSP, PetscInt, PetscReal *, PetscReal *, PetscInt *);
32: PetscErrorCode (*computeritz)(KSP, PetscBool, PetscBool, PetscInt *, Vec[], PetscReal *, PetscReal *);
33: PetscErrorCode (*destroy)(KSP);
34: PetscErrorCode (*view)(KSP, PetscViewer);
35: PetscErrorCode (*reset)(KSP);
36: PetscErrorCode (*load)(KSP, PetscViewer);
37: };
39: typedef struct _KSPGuessOps *KSPGuessOps;
41: struct _KSPGuessOps {
42: PetscErrorCode (*formguess)(KSPGuess, Vec, Vec); /* Form initial guess */
43: PetscErrorCode (*update)(KSPGuess, Vec, Vec); /* Update database */
44: PetscErrorCode (*setfromoptions)(KSPGuess);
45: PetscErrorCode (*settolerance)(KSPGuess, PetscReal);
46: PetscErrorCode (*setup)(KSPGuess);
47: PetscErrorCode (*destroy)(KSPGuess);
48: PetscErrorCode (*view)(KSPGuess, PetscViewer);
49: PetscErrorCode (*reset)(KSPGuess);
50: };
52: /*
53: Defines the KSPGuess data structure.
54: */
55: struct _p_KSPGuess {
56: PETSCHEADER(struct _KSPGuessOps);
57: KSP ksp; /* the parent KSP */
58: Mat A; /* the current linear operator */
59: PetscObjectState omatstate; /* previous linear operator state */
60: void *data; /* pointer to the specific implementation */
61: };
63: PETSC_EXTERN PetscErrorCode KSPGuessCreate_Fischer(KSPGuess);
64: PETSC_EXTERN PetscErrorCode KSPGuessCreate_POD(KSPGuess);
66: /*
67: Maximum number of monitors you can run with a single KSP
68: */
69: #define MAXKSPMONITORS 5
70: #define MAXKSPREASONVIEWS 5
71: typedef enum {
72: KSP_SETUP_NEW = 0,
73: KSP_SETUP_NEWMATRIX,
74: KSP_SETUP_NEWRHS
75: } KSPSetUpStage;
77: /*
78: Defines the KSP data structure.
79: */
80: struct _p_KSP {
81: PETSCHEADER(struct _KSPOps);
82: DM dm;
83: PetscBool dmAuto; /* DM was created automatically by KSP */
84: PetscBool dmActive; /* KSP should use DM for computing operators */
85: /*------------------------- User parameters--------------------------*/
86: PetscInt max_it; /* maximum number of iterations */
87: PetscInt min_it; /* minimum number of iterations */
88: KSPGuess guess;
89: PetscBool guess_zero, /* flag for whether initial guess is 0 */
90: guess_not_read, /* guess is not read, does not need to be zeroed */
91: calc_sings, /* calculate extreme Singular Values */
92: calc_ritz, /* calculate (harmonic) Ritz pairs */
93: guess_knoll; /* use initial guess of PCApply(ksp->B,b */
94: PCSide pc_side; /* flag for left, right, or symmetric preconditioning */
95: PetscInt normsupporttable[KSP_NORM_MAX][PC_SIDE_MAX]; /* Table of supported norms and pc_side, see KSPSetSupportedNorm() */
96: PetscReal rtol, /* relative tolerance */
97: abstol, /* absolute tolerance */
98: ttol, /* (not set by user) */
99: divtol; /* divergence tolerance */
100: PetscReal rnorm0; /* initial residual norm (used for divergence testing) */
101: PetscReal rnorm; /* current residual norm */
102: KSPConvergedReason reason;
103: PetscBool errorifnotconverged; /* create an error if the KSPSolve() does not converge */
105: Vec vec_sol, vec_rhs; /* pointer to where user has stashed
106: the solution and rhs, these are
107: never touched by the code, only
108: passed back to the user */
109: PetscReal *res_hist; /* If !0 stores residual each at iteration */
110: PetscReal *res_hist_alloc; /* If !0 means user did not provide buffer, needs deallocation */
111: size_t res_hist_len; /* current size of residual history array */
112: size_t res_hist_max; /* actual amount of storage in residual history */
113: PetscBool res_hist_reset; /* reset history to length zero for each new solve */
114: PetscReal *err_hist; /* If !0 stores error at each iteration */
115: PetscReal *err_hist_alloc; /* If !0 means user did not provide buffer, needs deallocation */
116: size_t err_hist_len; /* current size of error history array */
117: size_t err_hist_max; /* actual amount of storage in error history */
118: PetscBool err_hist_reset; /* reset history to length zero for each new solve */
120: PetscInt chknorm; /* only compute/check norm if iterations is great than this */
121: PetscBool lagnorm; /* Lag the residual norm calculation so that it is computed as part of the
122: MPI_Allreduce() for computing the inner products for the next iteration. */
124: PetscInt nmax; /* maximum number of right-hand sides to be handled simultaneously */
126: /* --------User (or default) routines (most return -1 on error) --------*/
127: PetscErrorCode (*monitor[MAXKSPMONITORS])(KSP, PetscInt, PetscReal, void *); /* returns control to user after */
128: PetscErrorCode (*monitordestroy[MAXKSPMONITORS])(void **); /* */
129: void *monitorcontext[MAXKSPMONITORS]; /* residual calculation, allows user */
130: PetscInt numbermonitors; /* to, for instance, print residual norm, etc. */
131: PetscBool pauseFinal; /* Pause all drawing monitor at the final iterate */
133: PetscErrorCode (*reasonview[MAXKSPREASONVIEWS])(KSP, void *); /* KSP converged reason view */
134: PetscErrorCode (*reasonviewdestroy[MAXKSPREASONVIEWS])(void **); /* Optional destroy routine */
135: void *reasonviewcontext[MAXKSPREASONVIEWS]; /* User context */
136: PetscInt numberreasonviews; /* Number if reason viewers */
138: PetscErrorCode (*converged)(KSP, PetscInt, PetscReal, KSPConvergedReason *, void *);
139: PetscErrorCode (*convergeddestroy)(void *);
140: void *cnvP;
142: void *user; /* optional user-defined context */
144: PC pc;
146: void *data; /* holder for misc stuff associated
147: with a particular iterative solver */
149: PetscBool view, viewPre, viewRate, viewMat, viewPMat, viewRhs, viewSol, viewMatExp, viewEV, viewSV, viewEVExp, viewFinalRes, viewPOpExp, viewDScale;
150: PetscViewer viewer, viewerPre, viewerRate, viewerMat, viewerPMat, viewerRhs, viewerSol, viewerMatExp, viewerEV, viewerSV, viewerEVExp, viewerFinalRes, viewerPOpExp, viewerDScale;
151: PetscViewerFormat format, formatPre, formatRate, formatMat, formatPMat, formatRhs, formatSol, formatMatExp, formatEV, formatSV, formatEVExp, formatFinalRes, formatPOpExp, formatDScale;
153: /* ----------------Default work-area management -------------------- */
154: PetscInt nwork;
155: Vec *work;
157: KSPSetUpStage setupstage;
158: PetscBool setupnewmatrix; /* true if we need to call ksp->ops->setup with KSP_SETUP_NEWMATRIX */
160: PetscInt its; /* number of iterations so far computed in THIS linear solve*/
161: PetscInt totalits; /* number of iterations used by this KSP object since it was created */
163: PetscBool transpose_solve; /* solve transpose system instead */
164: struct {
165: Mat AT, BT;
166: PetscBool use_explicittranspose; /* transpose the system explicitly in KSPSolveTranspose */
167: PetscBool reuse_transpose; /* reuse the previous transposed system */
168: } transpose;
170: KSPNormType normtype; /* type of norm used for convergence tests */
172: PCSide pc_side_set; /* PC type set explicitly by user */
173: KSPNormType normtype_set; /* Norm type set explicitly by user */
175: /* Allow diagonally scaling the matrix before computing the preconditioner or using
176: the Krylov method. Note this is NOT just Jacobi preconditioning */
178: PetscBool dscale; /* diagonal scale system; used with KSPSetDiagonalScale() */
179: PetscBool dscalefix; /* unscale system after solve */
180: PetscBool dscalefix2; /* system has been unscaled */
181: Vec diagonal; /* 1/sqrt(diag of matrix) */
182: Vec truediagonal;
184: /* Allow declaring convergence when negative curvature is detected */
185: PetscBool converged_neg_curve;
187: PetscInt setfromoptionscalled;
188: PetscBool skippcsetfromoptions; /* if set then KSPSetFromOptions() does not call PCSetFromOptions() */
190: PetscErrorCode (*presolve)(KSP, Vec, Vec, void *);
191: PetscErrorCode (*postsolve)(KSP, Vec, Vec, void *);
192: void *prectx, *postctx;
194: PetscInt nestlevel; /* how many levels of nesting does the KSP have */
195: };
197: typedef struct { /* dummy data structure used in KSPMonitorDynamicTolerance() */
198: PetscReal coef;
199: PetscReal bnrm;
200: } KSPDynTolCtx;
202: typedef struct {
203: PetscBool initialrtol; /* default relative residual decrease is computed from initial residual, not rhs */
204: PetscBool mininitialrtol; /* default relative residual decrease is computed from min of initial residual and rhs */
205: PetscBool convmaxits; /* if true, the convergence test returns KSP_CONVERGED_ITS if the maximum number of iterations is reached */
206: Vec work;
207: } KSPConvergedDefaultCtx;
209: static inline PetscErrorCode KSPLogResidualHistory(KSP ksp, PetscReal norm)
210: {
211: PetscFunctionBegin;
212: PetscCall(PetscObjectSAWsTakeAccess((PetscObject)ksp));
213: if (ksp->res_hist && ksp->res_hist_max > ksp->res_hist_len) ksp->res_hist[ksp->res_hist_len++] = norm;
214: PetscCall(PetscObjectSAWsGrantAccess((PetscObject)ksp));
215: PetscFunctionReturn(PETSC_SUCCESS);
216: }
218: static inline PetscErrorCode KSPLogErrorHistory(KSP ksp)
219: {
220: DM dm;
222: PetscFunctionBegin;
223: PetscCall(PetscObjectSAWsTakeAccess((PetscObject)ksp));
224: PetscCall(KSPGetDM(ksp, &dm));
225: if (dm && ksp->err_hist && ksp->err_hist_max > ksp->err_hist_len) {
226: PetscSimplePointFn *exactSol;
227: void *exactCtx;
228: PetscDS ds;
229: Vec u;
230: PetscReal error;
231: PetscInt Nf;
233: PetscCall(KSPBuildSolution(ksp, NULL, &u));
234: /* TODO Was needed to correct for Newton solution, but I just need to set a solution */
235: //PetscCall(VecScale(u, -1.0));
236: /* TODO Case when I have a solution */
237: if (0) {
238: PetscCall(DMGetDS(dm, &ds));
239: PetscCall(PetscDSGetNumFields(ds, &Nf));
240: PetscCheck(Nf <= 1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Cannot handle number of fields %" PetscInt_FMT " > 1 right now", Nf);
241: PetscCall(PetscDSGetExactSolution(ds, 0, &exactSol, &exactCtx));
242: PetscCall(DMComputeL2FieldDiff(dm, 0.0, &exactSol, &exactCtx, u, &error));
243: } else {
244: /* The null solution A 0 = 0 */
245: PetscCall(VecNorm(u, NORM_2, &error));
246: }
247: ksp->err_hist[ksp->err_hist_len++] = error;
248: }
249: PetscCall(PetscObjectSAWsGrantAccess((PetscObject)ksp));
250: PetscFunctionReturn(PETSC_SUCCESS);
251: }
253: static inline PetscScalar KSPNoisyHash_Private(PetscInt xx)
254: {
255: unsigned int x = (unsigned int)xx;
256: x = ((x >> 16) ^ x) * 0x45d9f3b;
257: x = ((x >> 16) ^ x) * 0x45d9f3b;
258: x = ((x >> 16) ^ x);
259: return (PetscScalar)(((PetscInt64)x - 2147483648) * 5.e-10); /* center around zero, scaled about -1. to 1.*/
260: }
262: static inline PetscErrorCode KSPSetNoisy_Private(Vec v)
263: {
264: PetscScalar *a;
265: PetscInt n, istart;
267: PetscFunctionBegin;
268: PetscCall(VecGetOwnershipRange(v, &istart, NULL));
269: PetscCall(VecGetLocalSize(v, &n));
270: PetscCall(VecGetArrayWrite(v, &a));
271: for (PetscInt i = 0; i < n; ++i) a[i] = KSPNoisyHash_Private(i + istart);
272: PetscCall(VecRestoreArrayWrite(v, &a));
273: PetscFunctionReturn(PETSC_SUCCESS);
274: }
276: PETSC_INTERN PetscErrorCode KSPSetUpNorms_Private(KSP, PetscBool, KSPNormType *, PCSide *);
278: PETSC_INTERN PetscErrorCode KSPPlotEigenContours_Private(KSP, PetscInt, const PetscReal *, const PetscReal *);
280: typedef struct _p_DMKSP *DMKSP;
281: typedef struct _DMKSPOps *DMKSPOps;
282: struct _DMKSPOps {
283: KSPComputeOperatorsFn *computeoperators;
284: KSPComputeRHSFn *computerhs;
285: KSPComputeInitialGuessFn *computeinitialguess;
286: PetscErrorCode (*destroy)(DMKSP *);
287: PetscErrorCode (*duplicate)(DMKSP, DMKSP);
288: };
290: /*S
291: DMKSP - Object held by a `DM` that contains all the callback functions and their contexts needed by a `KSP`
293: Level: developer
295: Notes:
296: Users provides callback functions and their contexts to `KSP` using, for example, `KSPSetComputeRHS()`. These values are stored
297: in a `DMKSP` that is contained in the `DM` associated with the `KSP`. If no `DM` was provided by
298: the user with `KSPSetDM()` it is automatically created by `KSPGetDM()` with `DMShellCreate()`.
300: Users very rarely need to worked directly with the `DMKSP` object, rather they work with the `KSP` and the `DM` they created
302: Multiple `DM` can share a single `DMKSP`, often each `DM` is associated with
303: a grid refinement level. `DMGetDMKSP()` returns the `DMKSP` associated with a `DM`. `DMGetDMKSPWrite()` returns a unique
304: `DMKSP` that is only associated with the current `DM`, making a copy of the shared `DMKSP` if needed (copy-on-write).
306: Developer Notes:
307: It is rather subtle why `DMKSP`, `DMSNES`, and `DMTS` are needed instead of simply storing the user callback functions and contexts in `DM` or `KSP`, `SNES`, or `TS`.
308: It is to support composable solvers such as geometric multigrid. We want, by default, the same callback functions and contexts for all the levels in the computation,
309: but we need to also support different callbacks and contexts on each level. The copy-on-write approach of `DMGetDMKSPWrite()` makes this possible.
311: The `originaldm` inside the `DMKSP` is NOT reference counted (to prevent a reference count loop between a `DM` and a `DMKSP`).
312: The `DM` on which this context was first created is cached here to implement one-way
313: copy-on-write. When `DMGetDMKSPWrite()` sees a request using a different `DM`, it makes a copy of the `TSDM`. Thus, if a user
314: only interacts directly with one level, e.g., using `TSSetIFunction()`, then coarse levels of a multilevel item
315: integrator are built, then the user changes the routine with another call to `TSSetIFunction()`, it automatically
316: propagates to all the levels. If instead, they get out a specific level and set the function on that level,
317: subsequent changes to the original level will no longer propagate to that level.
319: .seealso: [](ch_ts), `KSP`, `KSPCreate()`, `DM`, `DMGetDMKSPWrite()`, `DMGetDMKSP()`, `DMSNES`, `DMTS`, `DMKSPSetComputeOperators()`, `DMKSPGetComputeOperators()`,
320: `DMKSPSetComputeRHS()`, `DMKSPSetComputeInitialGuess()`
321: S*/
322: struct _p_DMKSP {
323: PETSCHEADER(struct _DMKSPOps);
324: void *operatorsctx;
325: void *rhsctx;
326: void *initialguessctx;
327: void *data;
329: /* See developer note for `DMKSP` above */
330: DM originaldm;
332: void (*fortran_func_pointers[3])(void); /* Store our own function pointers so they are associated with the DMKSP instead of the DM */
333: };
334: PETSC_EXTERN PetscErrorCode DMGetDMKSP(DM, DMKSP *);
335: PETSC_EXTERN PetscErrorCode DMGetDMKSPWrite(DM, DMKSP *);
336: PETSC_EXTERN PetscErrorCode DMCopyDMKSP(DM, DM);
338: /*
339: These allow the various Krylov methods to apply to either the linear system or its transpose.
340: */
341: static inline PetscErrorCode KSP_RemoveNullSpace(KSP ksp, Vec y)
342: {
343: PetscFunctionBegin;
344: if (ksp->pc_side == PC_LEFT) {
345: Mat A;
346: MatNullSpace nullsp;
348: PetscCall(PCGetOperators(ksp->pc, &A, NULL));
349: PetscCall(MatGetNullSpace(A, &nullsp));
350: if (nullsp) PetscCall(MatNullSpaceRemove(nullsp, y));
351: }
352: PetscFunctionReturn(PETSC_SUCCESS);
353: }
355: static inline PetscErrorCode KSP_RemoveNullSpaceTranspose(KSP ksp, Vec y)
356: {
357: PetscFunctionBegin;
358: if (ksp->pc_side == PC_LEFT) {
359: Mat A;
360: MatNullSpace nullsp;
362: PetscCall(PCGetOperators(ksp->pc, &A, NULL));
363: PetscCall(MatGetTransposeNullSpace(A, &nullsp));
364: if (nullsp) PetscCall(MatNullSpaceRemove(nullsp, y));
365: }
366: PetscFunctionReturn(PETSC_SUCCESS);
367: }
369: static inline PetscErrorCode KSP_MatMult(KSP ksp, Mat A, Vec x, Vec y)
370: {
371: PetscFunctionBegin;
372: if (ksp->transpose_solve) PetscCall(MatMultTranspose(A, x, y));
373: else PetscCall(MatMult(A, x, y));
374: PetscFunctionReturn(PETSC_SUCCESS);
375: }
377: static inline PetscErrorCode KSP_MatMultTranspose(KSP ksp, Mat A, Vec x, Vec y)
378: {
379: PetscFunctionBegin;
380: if (ksp->transpose_solve) PetscCall(MatMult(A, x, y));
381: else PetscCall(MatMultTranspose(A, x, y));
382: PetscFunctionReturn(PETSC_SUCCESS);
383: }
385: static inline PetscErrorCode KSP_MatMultHermitianTranspose(KSP ksp, Mat A, Vec x, Vec y)
386: {
387: PetscFunctionBegin;
388: if (!ksp->transpose_solve) PetscCall(MatMultHermitianTranspose(A, x, y));
389: else {
390: Vec w;
392: PetscCall(VecDuplicate(x, &w));
393: PetscCall(VecCopy(x, w));
394: PetscCall(VecConjugate(w));
395: PetscCall(MatMult(A, w, y));
396: PetscCall(VecDestroy(&w));
397: PetscCall(VecConjugate(y));
398: }
399: PetscFunctionReturn(PETSC_SUCCESS);
400: }
402: static inline PetscErrorCode KSP_PCApply(KSP ksp, Vec x, Vec y)
403: {
404: PetscFunctionBegin;
405: if (ksp->transpose_solve) {
406: PetscCall(PCApplyTranspose(ksp->pc, x, y));
407: PetscCall(KSP_RemoveNullSpaceTranspose(ksp, y));
408: } else {
409: PetscCall(PCApply(ksp->pc, x, y));
410: PetscCall(KSP_RemoveNullSpace(ksp, y));
411: }
412: PetscFunctionReturn(PETSC_SUCCESS);
413: }
415: static inline PetscErrorCode KSP_PCApplyTranspose(KSP ksp, Vec x, Vec y)
416: {
417: PetscFunctionBegin;
418: if (ksp->transpose_solve) {
419: PetscCall(PCApply(ksp->pc, x, y));
420: PetscCall(KSP_RemoveNullSpace(ksp, y));
421: } else {
422: PetscCall(PCApplyTranspose(ksp->pc, x, y));
423: PetscCall(KSP_RemoveNullSpaceTranspose(ksp, y));
424: }
425: PetscFunctionReturn(PETSC_SUCCESS);
426: }
428: static inline PetscErrorCode KSP_PCApplyHermitianTranspose(KSP ksp, Vec x, Vec y)
429: {
430: PetscFunctionBegin;
431: PetscCall(VecConjugate(x));
432: PetscCall(KSP_PCApplyTranspose(ksp, x, y));
433: PetscCall(VecConjugate(x));
434: PetscCall(VecConjugate(y));
435: PetscFunctionReturn(PETSC_SUCCESS);
436: }
438: static inline PetscErrorCode KSP_PCMatApply(KSP ksp, Mat X, Mat Y)
439: {
440: PetscFunctionBegin;
441: if (ksp->transpose_solve) {
442: PetscBool flg;
443: PetscCall(PetscObjectTypeCompareAny((PetscObject)ksp->pc, &flg, PCNONE, PCICC, PCCHOLESKY, ""));
444: PetscCheck(flg, PetscObjectComm((PetscObject)ksp), PETSC_ERR_SUP, "PCMatApplyTranspose() not yet implemented for nonsymmetric PC");
445: }
446: PetscCall(PCMatApply(ksp->pc, X, Y));
447: PetscFunctionReturn(PETSC_SUCCESS);
448: }
450: static inline PetscErrorCode KSP_PCMatApplyTranspose(KSP ksp, Mat X, Mat Y)
451: {
452: PetscFunctionBegin;
453: if (!ksp->transpose_solve) {
454: PetscBool flg;
455: PetscCall(PetscObjectTypeCompareAny((PetscObject)ksp->pc, &flg, PCNONE, PCICC, PCCHOLESKY, ""));
456: PetscCheck(flg, PetscObjectComm((PetscObject)ksp), PETSC_ERR_SUP, "PCMatApplyTranspose() not yet implemented for nonsymmetric PC");
457: }
458: PetscCall(PCMatApply(ksp->pc, X, Y));
459: PetscFunctionReturn(PETSC_SUCCESS);
460: }
462: static inline PetscErrorCode KSP_PCApplyBAorAB(KSP ksp, Vec x, Vec y, Vec w)
463: {
464: PetscFunctionBegin;
465: if (ksp->transpose_solve) {
466: PetscCall(PCApplyBAorABTranspose(ksp->pc, ksp->pc_side, x, y, w));
467: PetscCall(KSP_RemoveNullSpaceTranspose(ksp, y));
468: } else {
469: PetscCall(PCApplyBAorAB(ksp->pc, ksp->pc_side, x, y, w));
470: PetscCall(KSP_RemoveNullSpace(ksp, y));
471: }
472: PetscFunctionReturn(PETSC_SUCCESS);
473: }
475: static inline PetscErrorCode KSP_PCApplyBAorABTranspose(KSP ksp, Vec x, Vec y, Vec w)
476: {
477: PetscFunctionBegin;
478: if (ksp->transpose_solve) PetscCall(PCApplyBAorAB(ksp->pc, ksp->pc_side, x, y, w));
479: else PetscCall(PCApplyBAorABTranspose(ksp->pc, ksp->pc_side, x, y, w));
480: PetscFunctionReturn(PETSC_SUCCESS);
481: }
483: PETSC_EXTERN PetscLogEvent KSP_GMRESOrthogonalization;
484: PETSC_EXTERN PetscLogEvent KSP_SetUp;
485: PETSC_EXTERN PetscLogEvent KSP_Solve;
486: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_0;
487: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_1;
488: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_2;
489: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_3;
490: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_4;
491: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_S;
492: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_L;
493: PETSC_EXTERN PetscLogEvent KSP_Solve_FS_U;
494: PETSC_EXTERN PetscLogEvent KSP_SolveTranspose;
495: PETSC_EXTERN PetscLogEvent KSP_MatSolve;
496: PETSC_EXTERN PetscLogEvent KSP_MatSolveTranspose;
498: PETSC_INTERN PetscErrorCode MatGetSchurComplement_Basic(Mat, IS, IS, IS, IS, MatReuse, Mat *, MatSchurComplementAinvType, MatReuse, Mat *);
499: PETSC_INTERN PetscErrorCode PCPreSolveChangeRHS(PC, PetscBool *);
501: /*MC
502: KSPCheckDot - Checks if the result of a dot product used by the corresponding `KSP` contains Inf or NaN. These indicate that the previous
503: application of the preconditioner generated an error. Sets a `KSPConvergedReason` and returns if the `PC` set a `PCFailedReason`.
505: Collective
507: Input Parameter:
508: . ksp - the linear solver `KSP` context.
510: Output Parameter:
511: . beta - the result of the inner product
513: Level: developer
515: Developer Notes:
516: Used to manage returning from `KSP` solvers collectively whose preconditioners have failed, possibly only a subset of MPI processes, in some way
518: It uses the fact that `KSP` piggy-backs the collectivity of certain error conditions on the results of norms and inner products.
520: .seealso: `PCFailedReason`, `KSPConvergedReason`, `PCGetFailedReasonRank()`, `KSP`, `KSPCreate()`, `KSPSetType()`, `KSP`, `KSPCheckNorm()`, `KSPCheckSolve()`,
521: `KSPSetErrorIfNotConverged()`
522: M*/
523: #define KSPCheckDot(ksp, beta) \
524: do { \
525: if (PetscIsInfOrNanScalar(beta)) { \
526: PetscCheck(!ksp->errorifnotconverged, PetscObjectComm((PetscObject)ksp), PETSC_ERR_NOT_CONVERGED, "KSPSolve has not converged due to Nan or Inf inner product"); \
527: { \
528: PCFailedReason pcreason; \
529: PetscCall(PCReduceFailedReason(ksp->pc)); \
530: PetscCall(PCGetFailedReasonRank(ksp->pc, &pcreason)); \
531: if (pcreason) { \
532: ksp->reason = KSP_DIVERGED_PC_FAILED; \
533: PetscCall(VecSetInf(ksp->vec_sol)); \
534: } else { \
535: ksp->reason = KSP_DIVERGED_NANORINF; \
536: } \
537: PetscFunctionReturn(PETSC_SUCCESS); \
538: } \
539: } \
540: } while (0)
542: /*MC
543: KSPCheckNorm - Checks if the result of a norm used by the corresponding `KSP` contains `inf` or `NaN`. These indicate that the previous
544: application of the preconditioner generated an error. Sets a `KSPConvergedReason` and returns if the `PC` set a `PCFailedReason`.
546: Collective
548: Input Parameter:
549: . ksp - the linear solver `KSP` context.
551: Output Parameter:
552: . beta - the result of the norm
554: Level: developer
556: Developer Notes:
557: Used to manage returning from `KSP` solvers collectively whose preconditioners have failed, possibly only a subset of MPI processes, in some way.
559: It uses the fact that `KSP` piggy-backs the collectivity of certain error conditions on the results of norms and inner products.
561: .seealso: `PCFailedReason`, `KSPConvergedReason`, `PCGetFailedReasonRank()`, `KSP`, `KSPCreate()`, `KSPSetType()`, `KSP`, `KSPCheckDot()`, `KSPCheckSolve()`,
562: `KSPSetErrorIfNotConverged()`
563: M*/
564: #define KSPCheckNorm(ksp, beta) \
565: do { \
566: if (PetscIsInfOrNanReal(beta)) { \
567: PetscCheck(!ksp->errorifnotconverged, PetscObjectComm((PetscObject)ksp), PETSC_ERR_NOT_CONVERGED, "KSPSolve has not converged due to Nan or Inf norm"); \
568: { \
569: PCFailedReason pcreason; \
570: PetscCall(PCReduceFailedReason(ksp->pc)); \
571: PetscCall(PCGetFailedReasonRank(ksp->pc, &pcreason)); \
572: if (pcreason) { \
573: ksp->reason = KSP_DIVERGED_PC_FAILED; \
574: PetscCall(VecSetInf(ksp->vec_sol)); \
575: ksp->rnorm = beta; \
576: } else { \
577: ksp->reason = KSP_DIVERGED_NANORINF; \
578: ksp->rnorm = beta; \
579: } \
580: PetscFunctionReturn(PETSC_SUCCESS); \
581: } \
582: } \
583: } while (0)
585: PETSC_INTERN PetscErrorCode KSPMonitorMakeKey_Internal(const char[], PetscViewerType, PetscViewerFormat, char[]);
586: PETSC_INTERN PetscErrorCode KSPMonitorRange_Private(KSP, PetscInt, PetscReal *);