Actual source code: matimpl.h
1: #pragma once
3: #include <petscmat.h>
4: #include <petscmatcoarsen.h>
5: #include <petsc/private/petscimpl.h>
7: PETSC_EXTERN PetscBool MatRegisterAllCalled;
8: PETSC_EXTERN PetscBool MatSeqAIJRegisterAllCalled;
9: PETSC_EXTERN PetscBool MatOrderingRegisterAllCalled;
10: PETSC_EXTERN PetscBool MatColoringRegisterAllCalled;
11: PETSC_EXTERN PetscBool MatPartitioningRegisterAllCalled;
12: PETSC_EXTERN PetscBool MatMeshToCellGraphRegisterAllCalled;
13: PETSC_EXTERN PetscBool MatCoarsenRegisterAllCalled;
14: PETSC_EXTERN PetscErrorCode MatRegisterAll(void);
15: PETSC_EXTERN PetscErrorCode MatOrderingRegisterAll(void);
16: PETSC_EXTERN PetscErrorCode MatColoringRegisterAll(void);
17: PETSC_EXTERN PetscErrorCode MatPartitioningRegisterAll(void);
18: PETSC_EXTERN PetscErrorCode MatMeshToCellGraphRegisterAll(void);
19: PETSC_EXTERN PetscErrorCode MatCoarsenRegisterAll(void);
20: PETSC_EXTERN PetscErrorCode MatSeqAIJRegisterAll(void);
22: /* Gets the root type of the input matrix's type (e.g., MATAIJ for MATSEQAIJ) */
23: PETSC_EXTERN PetscErrorCode MatGetRootType_Private(Mat, MatType *);
25: /* Gets the MPI type corresponding to the input matrix's type (e.g., MATMPIAIJ for MATSEQAIJ) */
26: PETSC_INTERN PetscErrorCode MatGetMPIMatType_Private(Mat, MatType *);
28: /*
29: This file defines the parts of the matrix data structure that are
30: shared by all matrix types.
31: */
33: /*
34: If you add entries here also add them to the MATOP enum
35: in include/petscmat.h
36: */
37: typedef struct _MatOps *MatOps;
38: struct _MatOps {
39: /* 0*/
40: PetscErrorCode (*setvalues)(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], const PetscScalar[], InsertMode);
41: PetscErrorCode (*getrow)(Mat, PetscInt, PetscInt *, PetscInt *[], PetscScalar *[]);
42: PetscErrorCode (*restorerow)(Mat, PetscInt, PetscInt *, PetscInt *[], PetscScalar *[]);
43: PetscErrorCode (*mult)(Mat, Vec, Vec);
44: PetscErrorCode (*multadd)(Mat, Vec, Vec, Vec);
45: /* 5*/
46: PetscErrorCode (*multtranspose)(Mat, Vec, Vec);
47: PetscErrorCode (*multtransposeadd)(Mat, Vec, Vec, Vec);
48: PetscErrorCode (*solve)(Mat, Vec, Vec);
49: PetscErrorCode (*solveadd)(Mat, Vec, Vec, Vec);
50: PetscErrorCode (*solvetranspose)(Mat, Vec, Vec);
51: /*10*/
52: PetscErrorCode (*solvetransposeadd)(Mat, Vec, Vec, Vec);
53: PetscErrorCode (*lufactor)(Mat, IS, IS, const MatFactorInfo *);
54: PetscErrorCode (*choleskyfactor)(Mat, IS, const MatFactorInfo *);
55: PetscErrorCode (*sor)(Mat, Vec, PetscReal, MatSORType, PetscReal, PetscInt, PetscInt, Vec);
56: PetscErrorCode (*transpose)(Mat, MatReuse, Mat *);
57: /*15*/
58: PetscErrorCode (*getinfo)(Mat, MatInfoType, MatInfo *);
59: PetscErrorCode (*equal)(Mat, Mat, PetscBool *);
60: PetscErrorCode (*getdiagonal)(Mat, Vec);
61: PetscErrorCode (*diagonalscale)(Mat, Vec, Vec);
62: PetscErrorCode (*norm)(Mat, NormType, PetscReal *);
63: /*20*/
64: PetscErrorCode (*assemblybegin)(Mat, MatAssemblyType);
65: PetscErrorCode (*assemblyend)(Mat, MatAssemblyType);
66: PetscErrorCode (*setoption)(Mat, MatOption, PetscBool);
67: PetscErrorCode (*zeroentries)(Mat);
68: /*24*/
69: PetscErrorCode (*zerorows)(Mat, PetscInt, const PetscInt[], PetscScalar, Vec, Vec);
70: PetscErrorCode (*lufactorsymbolic)(Mat, Mat, IS, IS, const MatFactorInfo *);
71: PetscErrorCode (*lufactornumeric)(Mat, Mat, const MatFactorInfo *);
72: PetscErrorCode (*choleskyfactorsymbolic)(Mat, Mat, IS, const MatFactorInfo *);
73: PetscErrorCode (*choleskyfactornumeric)(Mat, Mat, const MatFactorInfo *);
74: /*29*/
75: PetscErrorCode (*setup)(Mat);
76: PetscErrorCode (*ilufactorsymbolic)(Mat, Mat, IS, IS, const MatFactorInfo *);
77: PetscErrorCode (*iccfactorsymbolic)(Mat, Mat, IS, const MatFactorInfo *);
78: PetscErrorCode (*getdiagonalblock)(Mat, Mat *);
79: PetscErrorCode (*setinf)(Mat);
80: /*34*/
81: PetscErrorCode (*duplicate)(Mat, MatDuplicateOption, Mat *);
82: PetscErrorCode (*forwardsolve)(Mat, Vec, Vec);
83: PetscErrorCode (*backwardsolve)(Mat, Vec, Vec);
84: PetscErrorCode (*ilufactor)(Mat, IS, IS, const MatFactorInfo *);
85: PetscErrorCode (*iccfactor)(Mat, IS, const MatFactorInfo *);
86: /*39*/
87: PetscErrorCode (*axpy)(Mat, PetscScalar, Mat, MatStructure);
88: PetscErrorCode (*createsubmatrices)(Mat, PetscInt, const IS[], const IS[], MatReuse, Mat *[]);
89: PetscErrorCode (*increaseoverlap)(Mat, PetscInt, IS[], PetscInt);
90: PetscErrorCode (*getvalues)(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], PetscScalar[]);
91: PetscErrorCode (*copy)(Mat, Mat, MatStructure);
92: /*44*/
93: PetscErrorCode (*getrowmax)(Mat, Vec, PetscInt[]);
94: PetscErrorCode (*scale)(Mat, PetscScalar);
95: PetscErrorCode (*shift)(Mat, PetscScalar);
96: PetscErrorCode (*diagonalset)(Mat, Vec, InsertMode);
97: PetscErrorCode (*zerorowscolumns)(Mat, PetscInt, const PetscInt[], PetscScalar, Vec, Vec);
98: /*49*/
99: PetscErrorCode (*setrandom)(Mat, PetscRandom);
100: PetscErrorCode (*getrowij)(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
101: PetscErrorCode (*restorerowij)(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
102: PetscErrorCode (*getcolumnij)(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
103: PetscErrorCode (*restorecolumnij)(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
104: /*54*/
105: PetscErrorCode (*fdcoloringcreate)(Mat, ISColoring, MatFDColoring);
106: PetscErrorCode (*coloringpatch)(Mat, PetscInt, PetscInt, ISColoringValue[], ISColoring *);
107: PetscErrorCode (*setunfactored)(Mat);
108: PetscErrorCode (*permute)(Mat, IS, IS, Mat *);
109: PetscErrorCode (*setvaluesblocked)(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], const PetscScalar[], InsertMode);
110: /*59*/
111: PetscErrorCode (*createsubmatrix)(Mat, IS, IS, MatReuse, Mat *);
112: PetscErrorCode (*destroy)(Mat);
113: PetscErrorCode (*view)(Mat, PetscViewer);
114: PetscErrorCode (*convertfrom)(Mat, MatType, MatReuse, Mat *);
115: PetscErrorCode (*matmatmultsymbolic)(Mat, Mat, Mat, PetscReal, Mat);
116: /*64*/
117: PetscErrorCode (*matmatmultnumeric)(Mat, Mat, Mat, Mat);
118: PetscErrorCode (*setlocaltoglobalmapping)(Mat, ISLocalToGlobalMapping, ISLocalToGlobalMapping);
119: PetscErrorCode (*setvalueslocal)(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], const PetscScalar[], InsertMode);
120: PetscErrorCode (*zerorowslocal)(Mat, PetscInt, const PetscInt[], PetscScalar, Vec, Vec);
121: PetscErrorCode (*getrowmaxabs)(Mat, Vec, PetscInt[]);
122: /*69*/
123: PetscErrorCode (*getrowminabs)(Mat, Vec, PetscInt[]);
124: PetscErrorCode (*convert)(Mat, MatType, MatReuse, Mat *);
125: PetscErrorCode (*hasoperation)(Mat, MatOperation, PetscBool *);
126: PetscErrorCode (*fdcoloringapply)(Mat, MatFDColoring, Vec, void *);
127: PetscErrorCode (*setfromoptions)(Mat, PetscOptionItems);
128: /*74*/
129: PetscErrorCode (*findzerodiagonals)(Mat, IS *);
130: PetscErrorCode (*mults)(Mat, Vecs, Vecs);
131: PetscErrorCode (*solves)(Mat, Vecs, Vecs);
132: PetscErrorCode (*getinertia)(Mat, PetscInt *, PetscInt *, PetscInt *);
133: PetscErrorCode (*load)(Mat, PetscViewer);
134: /*79*/
135: PetscErrorCode (*issymmetric)(Mat, PetscReal, PetscBool *);
136: PetscErrorCode (*ishermitian)(Mat, PetscReal, PetscBool *);
137: PetscErrorCode (*isstructurallysymmetric)(Mat, PetscBool *);
138: PetscErrorCode (*setvaluesblockedlocal)(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], const PetscScalar[], InsertMode);
139: PetscErrorCode (*getvecs)(Mat, Vec *, Vec *);
140: /*84*/
141: PetscErrorCode (*matmultsymbolic)(Mat, Mat, PetscReal, Mat);
142: PetscErrorCode (*matmultnumeric)(Mat, Mat, Mat);
143: PetscErrorCode (*ptapnumeric)(Mat, Mat, Mat); /* double dispatch wrapper routine */
144: PetscErrorCode (*mattransposemultsymbolic)(Mat, Mat, PetscReal, Mat);
145: PetscErrorCode (*mattransposemultnumeric)(Mat, Mat, Mat);
146: /*89*/
147: PetscErrorCode (*bindtocpu)(Mat, PetscBool);
148: PetscErrorCode (*productsetfromoptions)(Mat);
149: PetscErrorCode (*productsymbolic)(Mat);
150: PetscErrorCode (*productnumeric)(Mat);
151: PetscErrorCode (*conjugate)(Mat); /* complex conjugate */
152: /*94*/
153: PetscErrorCode (*viewnative)(Mat, PetscViewer);
154: PetscErrorCode (*setvaluesrow)(Mat, PetscInt, const PetscScalar[]);
155: PetscErrorCode (*realpart)(Mat);
156: PetscErrorCode (*imaginarypart)(Mat);
157: PetscErrorCode (*getrowuppertriangular)(Mat);
158: /*99*/
159: PetscErrorCode (*restorerowuppertriangular)(Mat);
160: PetscErrorCode (*matsolve)(Mat, Mat, Mat);
161: PetscErrorCode (*matsolvetranspose)(Mat, Mat, Mat);
162: PetscErrorCode (*getrowmin)(Mat, Vec, PetscInt[]);
163: PetscErrorCode (*getcolumnvector)(Mat, Vec, PetscInt);
164: /*104*/
165: PetscErrorCode (*getseqnonzerostructure)(Mat, Mat *);
166: PetscErrorCode (*create)(Mat);
167: PetscErrorCode (*getghosts)(Mat, PetscInt *, const PetscInt *[]);
168: PetscErrorCode (*getlocalsubmatrix)(Mat, IS, IS, Mat *);
169: PetscErrorCode (*restorelocalsubmatrix)(Mat, IS, IS, Mat *);
170: /*109*/
171: PetscErrorCode (*multdiagonalblock)(Mat, Vec, Vec);
172: PetscErrorCode (*hermitiantranspose)(Mat, MatReuse, Mat *);
173: PetscErrorCode (*multhermitiantranspose)(Mat, Vec, Vec);
174: PetscErrorCode (*multhermitiantransposeadd)(Mat, Vec, Vec, Vec);
175: PetscErrorCode (*getmultiprocblock)(Mat, MPI_Comm, MatReuse, Mat *);
176: /*114*/
177: PetscErrorCode (*findnonzerorows)(Mat, IS *);
178: PetscErrorCode (*getcolumnreductions)(Mat, PetscInt, PetscReal *);
179: PetscErrorCode (*invertblockdiagonal)(Mat, const PetscScalar **);
180: PetscErrorCode (*invertvariableblockdiagonal)(Mat, PetscInt, const PetscInt *, PetscScalar *);
181: PetscErrorCode (*createsubmatricesmpi)(Mat, PetscInt, const IS[], const IS[], MatReuse, Mat **);
182: /*119*/
183: PetscErrorCode (*transposematmultsymbolic)(Mat, Mat, PetscReal, Mat);
184: PetscErrorCode (*transposematmultnumeric)(Mat, Mat, Mat);
185: PetscErrorCode (*transposecoloringcreate)(Mat, ISColoring, MatTransposeColoring);
186: PetscErrorCode (*transcoloringapplysptoden)(MatTransposeColoring, Mat, Mat);
187: PetscErrorCode (*transcoloringapplydentosp)(MatTransposeColoring, Mat, Mat);
188: /*124*/
189: PetscErrorCode (*rartnumeric)(Mat, Mat, Mat); /* double dispatch wrapper routine */
190: PetscErrorCode (*setblocksizes)(Mat, PetscInt, PetscInt);
191: PetscErrorCode (*residual)(Mat, Vec, Vec, Vec);
192: PetscErrorCode (*fdcoloringsetup)(Mat, ISColoring, MatFDColoring);
193: PetscErrorCode (*findoffblockdiagonalentries)(Mat, IS *);
194: /*129*/
195: PetscErrorCode (*creatempimatconcatenateseqmat)(MPI_Comm, Mat, PetscInt, MatReuse, Mat *);
196: PetscErrorCode (*destroysubmatrices)(PetscInt, Mat *[]);
197: PetscErrorCode (*mattransposesolve)(Mat, Mat, Mat);
198: PetscErrorCode (*getvalueslocal)(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], PetscScalar[]);
199: PetscErrorCode (*creategraph)(Mat, PetscBool, PetscBool, PetscReal, PetscInt, PetscInt[], Mat *);
200: /*134*/
201: PetscErrorCode (*transposesymbolic)(Mat, Mat *);
202: PetscErrorCode (*eliminatezeros)(Mat, PetscBool);
203: PetscErrorCode (*getrowsumabs)(Mat, Vec);
204: PetscErrorCode (*getfactor)(Mat, MatSolverType, MatFactorType, Mat *);
205: PetscErrorCode (*getblockdiagonal)(Mat, Mat *); // NOTE: the caller of get{block, vblock}diagonal owns the returned matrix;
206: /*139*/
207: PetscErrorCode (*getvblockdiagonal)(Mat, Mat *); // they must destroy it after use
208: PetscErrorCode (*copyhashtoxaij)(Mat, Mat);
209: PetscErrorCode (*getcurrentmemtype)(Mat, PetscMemType *);
210: PetscErrorCode (*zerorowscolumnslocal)(Mat, PetscInt, const PetscInt[], PetscScalar, Vec, Vec);
211: PetscErrorCode (*adot)(Mat, Vec, Vec, PetscScalar *); /* induced vector inner product */
212: /*144*/
213: PetscErrorCode (*anorm)(Mat, Vec, PetscReal *); /* induced vector norm */
214: PetscErrorCode (*adot_local)(Mat, Vec, Vec, PetscScalar *);
215: PetscErrorCode (*anorm_local)(Mat, Vec, PetscReal *);
216: PetscErrorCode (*getordering)(Mat, MatOrderingType, IS *, IS *);
217: };
218: /*
219: If you add MatOps entries above also add them to the MATOP enum
220: in include/petscmat.h
221: */
223: typedef struct _n_MatRootName *MatRootName;
224: struct _n_MatRootName {
225: char *rname, *sname, *mname;
226: MatRootName next;
227: };
229: PETSC_EXTERN MatRootName MatRootNameList;
231: /*
232: Utility private matrix routines used outside Mat
233: */
234: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatFindNonzeroRowsOrCols_Basic(Mat, PetscBool, PetscReal, IS *);
235: PETSC_EXTERN PetscErrorCode MatShellGetScalingShifts(Mat, PetscScalar *, PetscScalar *, Vec *, Vec *, Vec *, Mat *, IS *, IS *);
237: #define MAT_SHELL_NOT_ALLOWED (void *)-1
239: /*
240: Utility private matrix routines
241: */
242: PETSC_INTERN PetscErrorCode MatConvert_Basic(Mat, MatType, MatReuse, Mat *);
243: PETSC_INTERN PetscErrorCode MatConvert_Shell(Mat, MatType, MatReuse, Mat *);
244: PETSC_INTERN PetscErrorCode MatConvertFrom_Shell(Mat, MatType, MatReuse, Mat *);
245: PETSC_INTERN PetscErrorCode MatShellSetContext_Immutable(Mat, void *);
246: PETSC_INTERN PetscErrorCode MatShellSetContextDestroy_Immutable(Mat, PetscCtxDestroyFn *);
247: PETSC_INTERN PetscErrorCode MatShellSetManageScalingShifts_Immutable(Mat);
248: PETSC_INTERN PetscErrorCode MatCopy_Basic(Mat, Mat, MatStructure);
249: PETSC_INTERN PetscErrorCode MatDiagonalSet_Default(Mat, Vec, InsertMode);
250: #if PetscDefined(HAVE_SCALAPACK) && (PetscDefined(USE_REAL_SINGLE) || PetscDefined(USE_REAL_DOUBLE))
251: PETSC_INTERN PetscErrorCode MatConvert_Dense_ScaLAPACK(Mat, MatType, MatReuse, Mat *);
252: #endif
253: PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_Basic(Mat, PetscCount, PetscInt[], PetscInt[]);
254: PETSC_INTERN PetscErrorCode MatSetValuesCOO_Basic(Mat, const PetscScalar[], InsertMode);
256: /*
257: Index translation for the local submatrices of MATLOCALREF and MATIS, which both forward their local
258: insertions to another matrix. These need to be macros because they use sizeof
259: */
260: #define MatIndexSpaceGet_Private(buf, nrow, ncol, irowm, icolm) \
261: do { \
262: if ((nrow) + (ncol) > (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) { \
263: PetscCall(PetscMalloc2(nrow, &(irowm), ncol, &(icolm))); \
264: } else { \
265: irowm = &(buf)[0]; \
266: icolm = &(buf)[nrow]; \
267: } \
268: } while (0)
270: #define MatIndexSpaceRestore_Private(buf, nrow, ncol, irowm, icolm) \
271: do { \
272: if ((nrow) + (ncol) > (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) PetscCall(PetscFree2(irowm, icolm)); \
273: } while (0)
275: /* Turn n block indices into the n * bs scalar indices they stand for */
276: static inline void MatBlockIndicesExpand_Private(PetscInt n, const PetscInt idx[], PetscInt bs, PetscInt idxm[])
277: {
278: for (PetscInt i = 0; i < n; i++) {
279: for (PetscInt j = 0; j < bs; j++) idxm[i * bs + j] = idx[i] * bs + j;
280: }
281: }
283: /* Scattering of dense matrices with strided PetscSF */
284: PETSC_EXTERN PetscErrorCode MatDenseScatter_Private(PetscSF, Mat, Mat, InsertMode, ScatterMode);
285: PETSC_INTERN PetscErrorCode MatDenseScatterWithSF_Private(PetscSF, Mat, Mat, MPI_Op, ScatterMode, PetscBool);
287: /* This can be moved to the public header after implementing some missing MatProducts */
288: PETSC_INTERN PetscErrorCode MatCreateFromISLocalToGlobalMapping(ISLocalToGlobalMapping, Mat, PetscBool, PetscBool, MatType, Mat *);
290: /* these callbacks rely on the old matrix function pointers for
291: matmat operations. They are unsafe, and should be removed.
292: However, the amount of work needed to clean up all the
293: implementations is not negligible */
294: PETSC_INTERN PetscErrorCode MatProductSymbolic_AB(Mat);
295: PETSC_INTERN PetscErrorCode MatProductNumeric_AB(Mat);
296: PETSC_INTERN PetscErrorCode MatProductSymbolic_AtB(Mat);
297: PETSC_INTERN PetscErrorCode MatProductNumeric_AtB(Mat);
298: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABt(Mat);
299: PETSC_INTERN PetscErrorCode MatProductNumeric_ABt(Mat);
300: PETSC_INTERN PetscErrorCode MatProductNumeric_PtAP(Mat);
301: PETSC_INTERN PetscErrorCode MatProductNumeric_RARt(Mat);
302: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABC(Mat);
303: PETSC_INTERN PetscErrorCode MatProductNumeric_ABC(Mat);
305: PETSC_INTERN PetscErrorCode MatProductCreate_Private(Mat, Mat, Mat, Mat);
306: /* this callback handles all the different triple products and
307: does not rely on the function pointers; used by cuSPARSE/hipSPARSE and KOKKOS-KERNELS */
308: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABC_Basic(Mat);
310: /* CreateGraph is common to AIJ seq and mpi */
311: PETSC_INTERN PetscErrorCode MatCreateGraph_Simple_AIJ(Mat, PetscBool, PetscBool, PetscReal, PetscInt, PetscInt[], Mat *);
313: #if PetscDefined(CLANG_STATIC_ANALYZER)
314: template <typename Tm>
315: extern void MatCheckPreallocated(Tm, int);
316: template <typename Tm>
317: extern void MatCheckProduct(Tm, int);
318: #else /* PETSC_CLANG_STATIC_ANALYZER */
319: #define MatCheckPreallocated(A, arg) \
320: do { \
321: if (!(A)->preallocated) PetscCall(MatSetUp(A)); \
322: } while (0)
324: #if PetscDefined(USE_DEBUG)
325: #define MatCheckProduct(A, arg) \
326: do { \
327: PetscCheck((A)->product, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Argument %d \"%s\" is not a matrix obtained from MatProductCreate()", arg, #A); \
328: } while (0)
329: #else
330: #define MatCheckProduct(A, arg) \
331: do { \
332: } while (0)
333: #endif
334: #endif /* PETSC_CLANG_STATIC_ANALYZER */
336: /*
337: The stash is used to temporarily store inserted matrix values that
338: belong to another processor. During the assembly phase the stashed
339: values are moved to the correct processor and
340: */
342: typedef struct _MatStashSpace *PetscMatStashSpace;
344: struct _MatStashSpace {
345: PetscMatStashSpace next;
346: PetscScalar *space_head, *val;
347: PetscInt *idx, *idy;
348: PetscInt total_space_size;
349: PetscInt local_used;
350: PetscInt local_remaining;
351: };
353: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceGet(PetscInt, PetscInt, PetscMatStashSpace *);
354: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceContiguous(PetscInt, PetscMatStashSpace *, PetscScalar *, PetscInt *, PetscInt *);
355: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceDestroy(PetscMatStashSpace *);
357: typedef struct {
358: PetscInt count;
359: } MatStashHeader;
361: typedef struct {
362: void *buffer; /* Of type blocktype, dynamically constructed */
363: PetscInt count;
364: char pending;
365: } MatStashFrame;
367: typedef struct _MatStash MatStash;
368: struct _MatStash {
369: PetscInt nmax; /* maximum stash size */
370: PetscInt umax; /* user specified max-size */
371: PetscInt oldnmax; /* the nmax value used previously */
372: PetscInt n; /* stash size */
373: PetscInt bs; /* block size of the stash */
374: PetscInt reallocs; /* preserve the no of mallocs invoked */
375: PetscMatStashSpace space_head, space; /* linked list to hold stashed global row/column numbers and matrix values */
377: PetscErrorCode (*ScatterBegin)(Mat, MatStash *, PetscInt *);
378: PetscErrorCode (*ScatterGetMesg)(MatStash *, PetscMPIInt *, PetscInt **, PetscInt **, PetscScalar **, PetscInt *);
379: PetscErrorCode (*ScatterEnd)(MatStash *);
380: PetscErrorCode (*ScatterDestroy)(MatStash *);
382: /* The following variables are used for communication */
383: MPI_Comm comm;
384: PetscMPIInt size, rank;
385: PetscMPIInt tag1, tag2;
386: MPI_Request *send_waits; /* array of send requests */
387: MPI_Request *recv_waits; /* array of receive requests */
388: MPI_Status *send_status; /* array of send status */
389: PetscMPIInt nsends, nrecvs; /* numbers of sends and receives */
390: PetscScalar *svalues; /* sending data */
391: PetscInt *sindices;
392: PetscScalar **rvalues; /* receiving data (values) */
393: PetscInt **rindices; /* receiving data (indices) */
394: PetscMPIInt nprocessed; /* number of messages already processed */
395: PetscMPIInt *flg_v; /* indicates what messages have arrived so far and from whom */
396: PetscBool reproduce;
397: PetscMPIInt reproduce_count;
399: /* The following variables are used for BTS communication */
400: PetscBool first_assembly_done; /* Is the first time matrix assembly done? */
401: PetscBool use_status; /* Use MPI_Status to determine number of items in each message */
402: PetscMPIInt nsendranks;
403: PetscMPIInt nrecvranks;
404: PetscMPIInt *sendranks;
405: PetscMPIInt *recvranks;
406: MatStashHeader *sendhdr, *recvhdr;
407: MatStashFrame *sendframes; /* pointers to the main messages */
408: MatStashFrame *recvframes;
409: MatStashFrame *recvframe_active;
410: PetscInt recvframe_i; /* index of block within active frame */
411: PetscInt recvframe_count; /* Count actually sent for current frame */
412: PetscMPIInt recvcount; /* Number of receives processed so far */
413: PetscMPIInt *some_indices; /* From last call to MPI_Waitsome */
414: MPI_Status *some_statuses; /* Statuses from last call to MPI_Waitsome */
415: PetscMPIInt some_count; /* Number of requests completed in last call to MPI_Waitsome */
416: PetscMPIInt some_i; /* Index of request currently being processed */
417: MPI_Request *sendreqs;
418: MPI_Request *recvreqs;
419: PetscSegBuffer segsendblocks;
420: PetscSegBuffer segrecvframe;
421: PetscSegBuffer segrecvblocks;
422: MPI_Datatype blocktype;
423: size_t blocktype_size;
424: InsertMode *insertmode; /* Pointer to check mat->insertmode and set upon message arrival in case no local values have been set. */
425: };
427: #if !PetscDefined(HAVE_MPIUNI)
428: PETSC_INTERN PetscErrorCode MatStashScatterDestroy_BTS(MatStash *);
429: #endif
430: PETSC_INTERN PetscErrorCode MatStashCreate_Private(MPI_Comm, PetscInt, MatStash *);
431: PETSC_INTERN PetscErrorCode MatStashDestroy_Private(MatStash *);
432: PETSC_INTERN PetscErrorCode MatStashScatterEnd_Private(MatStash *);
433: PETSC_INTERN PetscErrorCode MatStashSetInitialSize_Private(MatStash *, PetscInt);
434: PETSC_INTERN PetscErrorCode MatStashGetInfo_Private(MatStash *, PetscInt *, PetscInt *);
435: PETSC_INTERN PetscErrorCode MatStashValuesRow_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscBool);
436: PETSC_INTERN PetscErrorCode MatStashValuesCol_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscBool);
437: PETSC_INTERN PetscErrorCode MatStashValuesRowBlocked_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscInt, PetscInt);
438: PETSC_INTERN PetscErrorCode MatStashValuesColBlocked_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscInt, PetscInt);
439: PETSC_INTERN PetscErrorCode MatStashScatterBegin_Private(Mat, MatStash *, PetscInt *);
440: PETSC_INTERN PetscErrorCode MatStashScatterGetMesg_Private(MatStash *, PetscMPIInt *, PetscInt **, PetscInt **, PetscScalar **, PetscInt *);
441: PETSC_INTERN PetscErrorCode MatGetInfo_External(Mat, MatInfoType, MatInfo *);
443: typedef struct {
444: PetscInt dim;
445: PetscInt dims[4];
446: PetscInt starts[4];
447: PetscBool noc; /* this is a single component problem, hence user will not set MatStencil.c */
448: } MatStencilInfo;
450: /* Info about using compressed row format */
451: typedef struct {
452: PetscBool use; /* indicates compressed rows have been checked and will be used */
453: PetscInt nrows; /* number of non-zero rows */
454: PetscInt *i; /* compressed row pointer */
455: PetscInt *rindex; /* compressed row index */
456: } Mat_CompressedRow;
457: PETSC_EXTERN PetscErrorCode MatCheckCompressedRow(Mat, PetscInt, Mat_CompressedRow *, PetscInt *, PetscInt, PetscReal);
459: typedef struct { /* used by MatCreateRedundantMatrix() for reusing matredundant */
460: PetscInt nzlocal, nsends, nrecvs;
461: PetscMPIInt *send_rank, *recv_rank;
462: PetscInt *sbuf_nz, *rbuf_nz, *sbuf_j, **rbuf_j;
463: PetscScalar *sbuf_a, **rbuf_a;
464: MPI_Comm subcomm; /* when user does not provide a subcomm */
465: IS isrow, iscol;
466: Mat *matseq;
467: } Mat_Redundant;
469: typedef struct { /* used by MatProduct() */
470: MatProductType type;
471: char *alg;
472: Mat A, B, C, Dwork;
473: PetscBool symbolic_used_the_fact_A_is_symmetric; /* Symbolic phase took advantage of the fact that A is symmetric, and optimized e.g. AtB as AB. Then, .. */
474: PetscBool symbolic_used_the_fact_B_is_symmetric; /* .. in the numeric phase, if a new A is not symmetric (but has the same sparsity as the old A therefore .. */
475: PetscBool symbolic_used_the_fact_C_is_symmetric; /* MatMatMult(A,B,MAT_REUSE_MATRIX,..&C) is still legitimate), we need to redo symbolic! */
476: PetscObjectParameterDeclare(PetscReal, fill);
477: PetscBool api_user; /* used to distinguish command line options and to indicate the matrix values are ready to be consumed at symbolic phase if needed */
478: PetscBool setfromoptionscalled;
480: /* Some products may display the information on the algorithm used */
481: PetscErrorCode (*view)(Mat, PetscViewer);
483: /* nonzero state of A at the last symbolic phase, used by MatProductComputeWithMat_Private() to run that phase again when the nonzero pattern of A changes */
484: PetscObjectState Anonzerostate;
486: /* many products have intermediate data structures, each specific to Mat types and product type */
487: PetscBool clear; /* whether or not to clear the data structures after MatProductNumeric has been called */
488: void *data; /* where to stash those structures */
489: PetscCtxDestroyFn *destroy; /* freeing data */
490: } Mat_Product;
492: struct _p_Mat {
493: PETSCHEADER(struct _MatOps);
494: PetscLayout rmap, cmap;
495: void *data; /* implementation-specific data */
496: MatFactorType factortype; /* MAT_FACTOR_LU, ILU, CHOLESKY or ICC */
497: PetscBool trivialsymbolic; /* indicates the symbolic factorization doesn't actually do a symbolic factorization, it is delayed to the numeric factorization */
498: PetscBool canuseordering; /* factorization can use ordering provide to routine (most PETSc implementations) */
499: MatOrderingType preferredordering[MAT_FACTOR_NUM_TYPES]; /* what is the preferred (or default) ordering for the matrix solver type */
500: PetscBool assembled; /* is the matrix assembled? */
501: PetscBool was_assembled; /* new values inserted into assembled mat */
502: PetscInt num_ass; /* number of times matrix has been assembled */
503: PetscObjectState nonzerostate; /* each time new nonzeros locations are introduced into the matrix this is updated */
504: PetscObjectState ass_nonzerostate; /* nonzero state at last assembly */
505: MatInfo info; /* matrix information */
506: InsertMode insertmode; /* have values been inserted in matrix or added? */
507: MatStash stash, bstash; /* used for assembling off-proc mat emements */
508: MatNullSpace nullsp; /* null space (operator is singular) */
509: MatNullSpace transnullsp; /* null space of transpose of operator */
510: MatNullSpace nearnullsp; /* near null space to be used by multigrid methods */
511: PetscInt congruentlayouts; /* are the rows and columns layouts congruent? */
512: PetscBool preallocated;
513: MatStencilInfo stencil; /* information for structured grid */
514: PetscBool3 symmetric, hermitian, structurally_symmetric, spd;
515: PetscBool symmetry_eternal, structural_symmetry_eternal, spd_eternal;
516: PetscBool nooffprocentries, nooffproczerorows;
517: PetscBool assembly_subset; /* set by MAT_SUBSET_OFF_PROC_ENTRIES */
518: PetscBool submat_singleis; /* for efficient PCSetUp_ASM() */
519: PetscBool structure_only;
520: PetscBool sortedfull; /* full, sorted rows are inserted */
521: PetscBool force_diagonals; /* set by MAT_FORCE_DIAGONAL_ENTRIES */
522: #if PetscDefined(HAVE_DEVICE)
523: PetscOffloadMask offloadmask; /* a mask which indicates where the valid matrix data is (GPU, CPU or both) */
524: PetscBool boundtocpu;
525: PetscBool bindingpropagates;
526: #endif
527: char *defaultrandtype;
528: void *spptr; /* pointer for special library like SuperLU */
529: char *solvertype;
530: PetscBool checksymmetryonassembly, checknullspaceonassembly;
531: PetscReal checksymmetrytol;
532: Mat schur; /* Schur complement matrix */
533: MatFactorSchurStatus schur_status; /* status of the Schur complement matrix */
534: Mat_Redundant *redundant; /* used by MatCreateRedundantMatrix() */
535: PetscBool erroriffailure; /* Generate an error if detected (for example a zero pivot) instead of returning */
536: MatFactorError factorerrortype; /* type of error in factorization */
537: PetscReal factorerror_zeropivot_value; /* If numerical zero pivot was detected this is the computed value */
538: PetscInt factorerror_zeropivot_row; /* Row where zero pivot was detected */
539: PetscInt nblocks, *bsizes; /* support for MatSetVariableBlockSizes() */
540: PetscInt p_cstart, p_rank, p_cend, n_rank; /* Information from parallel MatComputeVariableBlockEnvelope() */
541: PetscBool p_parallel;
542: char *defaultvectype;
543: Mat_Product *product;
544: PetscBool form_explicit_transpose; /* hint to generate an explicit mat tranpsose for operations like MatMultTranspose() */
545: PetscBool transupdated; /* whether or not the explicitly generated transpose is up-to-date */
546: char *factorprefix; /* the prefix to use with factored matrix that is created */
547: PetscBool hash_active; /* indicates MatSetValues() is being handled by hashing */
548: Vec dot_vec; /* work vector used by MatADot_Default() */
549: };
551: PETSC_INTERN PetscErrorCode MatAXPY_Basic(Mat, PetscScalar, Mat, MatStructure);
552: PETSC_INTERN PetscErrorCode MatAXPY_BasicWithPreallocation(Mat, Mat, PetscScalar, Mat, MatStructure);
553: PETSC_INTERN PetscErrorCode MatAXPY_Basic_Preallocate(Mat, Mat, Mat *);
554: PETSC_INTERN PetscErrorCode MatAXPY_Dense_Nest(Mat, PetscScalar, Mat);
556: /*
557: Utility for MatZeroRows
558: */
559: PETSC_INTERN PetscErrorCode MatZeroRowsMapLocal_Private(Mat, PetscInt, const PetscInt *, PetscInt *, PetscInt **);
561: /*
562: Utility for MatView/MatLoad
563: */
564: PETSC_INTERN PetscErrorCode MatView_Binary_BlockSizes(Mat, PetscViewer);
565: PETSC_INTERN PetscErrorCode MatLoad_Binary_BlockSizes(Mat, PetscViewer);
567: /*
568: Object for partitioning graphs
569: */
571: typedef struct _MatPartitioningOps *MatPartitioningOps;
572: struct _MatPartitioningOps {
573: PetscErrorCode (*apply)(MatPartitioning, IS *);
574: PetscErrorCode (*applynd)(MatPartitioning, IS *);
575: PetscErrorCode (*setfromoptions)(MatPartitioning, PetscOptionItems);
576: PetscErrorCode (*destroy)(MatPartitioning);
577: PetscErrorCode (*view)(MatPartitioning, PetscViewer);
578: PetscErrorCode (*improve)(MatPartitioning, IS *);
579: };
581: struct _p_MatPartitioning {
582: PETSCHEADER(struct _MatPartitioningOps);
583: Mat adj;
584: PetscInt *vertex_weights;
585: PetscReal *part_weights;
586: PetscInt n; /* number of partitions */
587: PetscInt ncon; /* number of vertex weights per vertex */
588: void *data;
589: PetscBool use_edge_weights; /* A flag indicates whether or not to use edge weights */
590: };
592: /* needed for parallel nested dissection by ParMETIS */
593: PETSC_INTERN PetscErrorCode MatPartitioningSizesToSep_Private(PetscInt, PetscInt[], PetscInt[], PetscInt[]);
595: /* shared by the MatPartitioning and PetscPartitioner bindings to PT-SCOTCH */
596: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode PetscPTSCOTCHPartitionGraph_Private(MPI_Comm, const PetscInt[], PetscInt[], PetscInt[], PetscInt[], PetscInt[], PetscBool, PetscInt, PetscInt[], PetscInt, double, PetscInt[]);
598: /*
599: Object for coarsen graphs
600: */
601: typedef struct _MatCoarsenOps *MatCoarsenOps;
602: struct _MatCoarsenOps {
603: PetscErrorCode (*apply)(MatCoarsen);
604: PetscErrorCode (*setfromoptions)(MatCoarsen, PetscOptionItems);
605: PetscErrorCode (*destroy)(MatCoarsen);
606: PetscErrorCode (*view)(MatCoarsen, PetscViewer);
607: };
609: #define MAT_COARSEN_STRENGTH_INDEX_SIZE 3
610: struct _p_MatCoarsen {
611: PETSCHEADER(struct _MatCoarsenOps);
612: Mat graph;
613: void *subctx;
614: /* */
615: PetscBool strict_aggs;
616: IS perm;
617: PetscCoarsenData *agg_lists;
618: PetscInt max_it; /* number of iterations in HEM */
619: PetscReal threshold; /* HEM can filter interim graphs */
620: PetscInt strength_index_size;
621: PetscInt strength_index[MAT_COARSEN_STRENGTH_INDEX_SIZE];
622: };
624: PETSC_EXTERN PetscErrorCode MatCoarsenMISKSetDistance(MatCoarsen, PetscInt);
625: PETSC_EXTERN PetscErrorCode MatCoarsenMISKGetDistance(MatCoarsen, PetscInt *);
627: /*
628: Used in aijdevice.h
629: */
630: typedef struct {
631: PetscInt *i;
632: PetscInt *j;
633: PetscScalar *a;
634: PetscInt n;
635: PetscInt ignorezeroentries;
636: } PetscCSRDataStructure;
638: /*
639: MatFDColoring is used to compute Jacobian matrices efficiently
640: via coloring. The data structure is explained below in an example.
642: Color = 0 1 0 2 | 2 3 0
643: ---------------------------------------------------
644: 00 01 | 05
645: 10 11 | 14 15 Processor 0
646: 22 23 | 25
647: 32 33 |
648: ===================================================
649: | 44 45 46
650: 50 | 55 Processor 1
651: | 64 66
652: ---------------------------------------------------
654: ncolors = 4;
656: ncolumns = {2,1,1,0}
657: columns = {{0,2},{1},{3},{}}
658: nrows = {4,2,3,3}
659: rows = {{0,1,2,3},{0,1},{1,2,3},{0,1,2}}
660: vwscale = {dx(0),dx(1),dx(2),dx(3)} MPI Vec
661: vscale = {dx(0),dx(1),dx(2),dx(3),dx(4),dx(5)} Seq Vec
663: ncolumns = {1,0,1,1}
664: columns = {{6},{},{4},{5}}
665: nrows = {3,0,2,2}
666: rows = {{0,1,2},{},{1,2},{1,2}}
667: vwscale = {dx(4),dx(5),dx(6)} MPI Vec
668: vscale = {dx(0),dx(4),dx(5),dx(6)} Seq Vec
670: See the routine MatFDColoringApply() for how this data is used
671: to compute the Jacobian.
673: */
674: typedef struct {
675: PetscInt row;
676: PetscInt col;
677: PetscScalar *valaddr; /* address of value */
678: } MatEntry;
680: typedef struct {
681: PetscInt row;
682: PetscScalar *valaddr; /* address of value */
683: } MatEntry2;
685: struct _p_MatFDColoring {
686: PETSCHEADER(int);
687: PetscInt M, N, m; /* total rows, columns; local rows */
688: PetscInt rstart; /* first row owned by local processor */
689: PetscInt ncolors; /* number of colors */
690: PetscInt *ncolumns; /* number of local columns for a color */
691: PetscInt **columns; /* lists the local columns of each color (using global column numbering) */
692: IS *isa; /* these are the IS that contain the column values given in columns */
693: PetscInt *nrows; /* number of local rows for each color */
694: MatEntry *matentry; /* holds (row, column, address of value) for Jacobian matrix entry */
695: MatEntry2 *matentry2; /* holds (row, address of value) for Jacobian matrix entry */
696: PetscScalar *dy; /* store a block of F(x+dx)-F(x) when J is in BAIJ format */
697: PetscReal error_rel; /* square root of relative error in computing function */
698: PetscReal umin; /* minimum allowable u'dx value */
699: Vec w1, w2, w3; /* work vectors used in computing Jacobian */
700: PetscBool fset; /* indicates that the initial function value F(X) is set */
701: MatFDColoringFn *f; /* function that defines Jacobian */
702: void *fctx; /* optional user-defined context for use by the function f */
703: Vec vscale; /* holds FD scaling, i.e. 1/dx for each perturbed column */
704: PetscInt currentcolor; /* color for which function evaluation is being done now */
705: const char *htype; /* "wp" or "ds" */
706: ISColoringType ctype; /* IS_COLORING_GLOBAL or IS_COLORING_LOCAL */
707: PetscInt brows, bcols; /* number of block rows or columns for speedup inserting the dense matrix into sparse Jacobian */
708: PetscBool setupcalled; /* true if setup has been called */
709: PetscBool viewed; /* true if the -mat_fd_coloring_view has been triggered already */
710: PetscFortranCallbackFn *ftn_func_pointer; /* serve the same purpose as *fortran_func_pointers in PETSc objects */
711: void *ftn_func_cntx;
712: PetscObjectId matid; /* matrix this object was created with, must always be the same */
713: };
715: typedef struct _MatColoringOps *MatColoringOps;
716: struct _MatColoringOps {
717: PetscErrorCode (*destroy)(MatColoring);
718: PetscErrorCode (*setfromoptions)(MatColoring, PetscOptionItems);
719: PetscErrorCode (*view)(MatColoring, PetscViewer);
720: PetscErrorCode (*apply)(MatColoring, ISColoring *);
721: PetscErrorCode (*weights)(MatColoring, PetscReal **, PetscInt **);
722: };
724: struct _p_MatColoring {
725: PETSCHEADER(struct _MatColoringOps);
726: Mat mat;
727: PetscInt dist; /* distance of the coloring */
728: PetscInt maxcolors; /* the maximum number of colors returned, maxcolors=1 for MIS */
729: void *data; /* inner context */
730: PetscBool valid; /* check to see if what is produced is a valid coloring */
731: MatColoringWeightType weight_type; /* type of weight computation to be performed */
732: PetscReal *user_weights; /* custom weights and permutation */
733: PetscInt *user_lperm;
734: PetscBool valid_iscoloring; /* check to see if matcoloring is produced a valid iscoloring */
735: };
737: struct _p_MatTransposeColoring {
738: PETSCHEADER(int);
739: PetscInt M, N, m; /* total rows, columns; local rows */
740: PetscInt rstart; /* first row owned by local processor */
741: PetscInt ncolors; /* number of colors */
742: PetscInt *ncolumns; /* number of local columns for a color */
743: PetscInt *nrows; /* number of local rows for each color */
744: PetscInt currentcolor; /* color for which function evaluation is being done now */
745: ISColoringType ctype; /* IS_COLORING_GLOBAL or IS_COLORING_LOCAL */
747: PetscInt *colorforrow, *colorforcol; /* pointer to rows and columns */
748: PetscInt *rows; /* lists the local rows for each color (using the local row numbering) */
749: PetscInt *den2sp; /* maps (row,color) in the dense matrix to index of sparse matrix array a->a */
750: PetscInt *columns; /* lists the local columns of each color (using global column numbering) */
751: PetscInt brows; /* number of rows for efficient implementation of MatTransColoringApplyDenToSp() */
752: PetscInt *lstart; /* array used for loop over row blocks of Csparse */
753: };
755: /*
756: Null space context for preconditioner/operators
757: */
758: struct _p_MatNullSpace {
759: PETSCHEADER(int);
760: PetscBool has_cnst;
761: PetscInt n;
762: Vec *vecs;
763: PetscScalar *alpha; /* for projections */
764: MatNullSpaceRemoveFn *remove; /* for user provided removal function */
765: void *rmctx; /* context for remove() function */
766: };
768: /*
769: Internal data structure for MATMPIDENSE
770: */
771: typedef struct {
772: Mat A; /* local submatrix */
774: /* The following variables are used for matrix assembly */
775: PetscBool donotstash; /* Flag indicating if values should be stashed */
776: MPI_Request *send_waits; /* array of send requests */
777: MPI_Request *recv_waits; /* array of receive requests */
778: PetscInt nsends, nrecvs; /* numbers of sends and receives */
779: PetscScalar *svalues, *rvalues; /* sending and receiving data */
780: PetscInt rmax; /* maximum message length */
782: /* The following variables are used for matrix-vector products */
783: Vec lvec; /* local vector */
784: PetscSF Mvctx; /* for mat-mult communications */
785: PetscBool roworiented; /* if true, row-oriented input (default) */
787: /* Support for MatDenseGetColumnVec and MatDenseGetSubMatrix */
788: Mat cmat; /* matrix representation of a given subset of columns */
789: Vec cvec; /* vector representation of a given column */
790: const PetscScalar *ptrinuse; /* holds array to be restored (just a placeholder) */
791: PetscInt vecinuse; /* if cvec is in use (col = vecinuse-1) */
792: PetscInt matinuse; /* if cmat is in use (cbegin = matinuse-1) */
793: /* if this is from MatDenseGetSubMatrix, which columns and rows does it correspond to? */
794: PetscInt sub_rbegin;
795: PetscInt sub_rend;
796: PetscInt sub_cbegin;
797: PetscInt sub_cend;
798: } Mat_MPIDense;
800: /*
801: Checking zero pivot for LU, ILU preconditioners.
802: */
803: typedef struct {
804: PetscInt nshift, nshift_max;
805: PetscReal shift_amount, shift_lo, shift_hi, shift_top, shift_fraction;
806: PetscBool newshift;
807: PetscReal rs; /* active row sum of abs(off-diagonals) */
808: PetscScalar pv; /* pivot of the active row */
809: } FactorShiftCtx;
811: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatTransposeCheckNonzeroState_Private(Mat, Mat);
813: PETSC_EXTERN PetscErrorCode MatFactorDumpMatrix(Mat);
814: PETSC_INTERN PetscErrorCode MatSetBlockSizes_Default(Mat, PetscInt, PetscInt);
816: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatShift_Basic(Mat, PetscScalar);
817: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatProductComputeWithMat_Private(Mat, Mat, MatProductType, Mat);
819: static inline PetscErrorCode MatPivotCheck_nz(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
820: {
821: PetscReal _rs = sctx->rs;
822: PetscReal _zero = info->zeropivot * _rs;
824: PetscFunctionBegin;
825: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
826: /* force |diag| > zeropivot*rs */
827: if (!sctx->nshift) sctx->shift_amount = info->shiftamount;
828: else sctx->shift_amount *= 2.0;
829: sctx->newshift = PETSC_TRUE;
830: sctx->nshift++;
831: } else {
832: sctx->newshift = PETSC_FALSE;
833: }
834: PetscFunctionReturn(PETSC_SUCCESS);
835: }
837: static inline PetscErrorCode MatPivotCheck_pd(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
838: {
839: PetscReal _rs = sctx->rs;
840: PetscReal _zero = info->zeropivot * _rs;
842: PetscFunctionBegin;
843: if (PetscRealPart(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
844: /* force matfactor to be diagonally dominant */
845: if (sctx->nshift == sctx->nshift_max) {
846: sctx->shift_fraction = sctx->shift_hi;
847: } else {
848: sctx->shift_lo = sctx->shift_fraction;
849: sctx->shift_fraction = (sctx->shift_hi + sctx->shift_lo) / (PetscReal)2.;
850: }
851: sctx->shift_amount = sctx->shift_fraction * sctx->shift_top;
852: sctx->nshift++;
853: sctx->newshift = PETSC_TRUE;
854: } else {
855: sctx->newshift = PETSC_FALSE;
856: }
857: PetscFunctionReturn(PETSC_SUCCESS);
858: }
860: static inline PetscErrorCode MatPivotCheck_inblocks(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
861: {
862: PetscReal _zero = info->zeropivot;
864: PetscFunctionBegin;
865: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
866: sctx->pv += info->shiftamount;
867: sctx->shift_amount = 0.0;
868: sctx->nshift++;
869: }
870: sctx->newshift = PETSC_FALSE;
871: PetscFunctionReturn(PETSC_SUCCESS);
872: }
874: static inline PetscErrorCode MatPivotCheck_none(Mat fact, Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PetscInt row)
875: {
876: PetscReal _zero = info->zeropivot;
878: PetscFunctionBegin;
879: sctx->newshift = PETSC_FALSE;
880: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
881: PetscCheck(!mat->erroriffailure, PETSC_COMM_SELF, PETSC_ERR_MAT_LU_ZRPVT, "Zero pivot row %" PetscInt_FMT " value %g tolerance %g", row, (double)PetscAbsScalar(sctx->pv), (double)_zero);
882: PetscCall(PetscInfo(mat, "Detected zero pivot in factorization in row %" PetscInt_FMT " value %g tolerance %g\n", row, (double)PetscAbsScalar(sctx->pv), (double)_zero));
883: fact->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
884: fact->factorerror_zeropivot_value = PetscAbsScalar(sctx->pv);
885: fact->factorerror_zeropivot_row = row;
886: }
887: PetscFunctionReturn(PETSC_SUCCESS);
888: }
890: static inline PetscErrorCode MatPivotCheck(Mat fact, Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PetscInt row)
891: {
892: PetscFunctionBegin;
893: if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) PetscCall(MatPivotCheck_nz(mat, info, sctx, row));
894: else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) PetscCall(MatPivotCheck_pd(mat, info, sctx, row));
895: else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) PetscCall(MatPivotCheck_inblocks(mat, info, sctx, row));
896: else PetscCall(MatPivotCheck_none(fact, mat, info, sctx, row));
897: PetscFunctionReturn(PETSC_SUCCESS);
898: }
900: /*
901: Create and initialize a linked list
902: Input Parameters:
903: idx_start - starting index of the list
904: lnk_max - max value of lnk indicating the end of the list
905: nlnk - max length of the list
906: Output Parameters:
907: lnk - list initialized
908: bt - PetscBT (bitarray) with all bits set to false
909: lnk_empty - flg indicating the list is empty
910: */
911: #define PetscLLCreate(idx_start, lnk_max, nlnk, lnk, bt) ((PetscErrorCode)(PetscMalloc1(nlnk, &(lnk)) || PetscBTCreate(nlnk, &(bt)) || ((lnk)[idx_start] = lnk_max, PETSC_SUCCESS)))
913: #define PetscLLCreate_new(idx_start, lnk_max, nlnk, lnk, bt, lnk_empty) ((PetscErrorCode)(PetscMalloc1(nlnk, &(lnk)) || PetscBTCreate(nlnk, &(bt)) || (lnk_empty = PETSC_TRUE, 0) || ((lnk)[idx_start] = lnk_max, PETSC_SUCCESS)))
915: static inline PetscErrorCode PetscLLInsertLocation_Private(PetscBool assume_sorted, PetscInt k, PetscInt idx_start, PetscInt entry, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnkdata, PetscInt *PETSC_RESTRICT lnk)
916: {
917: PetscInt location;
919: PetscFunctionBegin;
920: /* start from the beginning if entry < previous entry */
921: if (!assume_sorted && k && entry < *lnkdata) *lnkdata = idx_start;
922: /* search for insertion location */
923: do {
924: location = *lnkdata;
925: *lnkdata = lnk[location];
926: } while (entry > *lnkdata);
927: /* insertion location is found, add entry into lnk */
928: lnk[location] = entry;
929: lnk[entry] = *lnkdata;
930: ++(*nlnk);
931: *lnkdata = entry; /* next search starts from here if next_entry > entry */
932: PetscFunctionReturn(PETSC_SUCCESS);
933: }
935: static inline PetscErrorCode PetscLLAdd_Private(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt, PetscBool assume_sorted)
936: {
937: PetscFunctionBegin;
938: *nlnk = 0;
939: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
940: const PetscInt entry = indices[k];
942: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, &lnkdata, lnk));
943: }
944: PetscFunctionReturn(PETSC_SUCCESS);
945: }
947: /*
948: Add an index set into a sorted linked list
949: Input Parameters:
950: nidx - number of input indices
951: indices - integer array
952: idx_start - starting index of the list
953: lnk - linked list(an integer array) that is created
954: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
955: output Parameters:
956: nlnk - number of newly added indices
957: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
958: bt - updated PetscBT (bitarray)
959: */
960: static inline PetscErrorCode PetscLLAdd(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt)
961: {
962: PetscFunctionBegin;
963: PetscCall(PetscLLAdd_Private(nidx, indices, idx_start, nlnk, lnk, bt, PETSC_FALSE));
964: PetscFunctionReturn(PETSC_SUCCESS);
965: }
967: /*
968: Add a SORTED ascending index set into a sorted linked list - same as PetscLLAdd() bus skip 'if (_k && _entry < _lnkdata) _lnkdata = idx_start;'
969: Input Parameters:
970: nidx - number of input indices
971: indices - sorted integer array
972: idx_start - starting index of the list
973: lnk - linked list(an integer array) that is created
974: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
975: output Parameters:
976: nlnk - number of newly added indices
977: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
978: bt - updated PetscBT (bitarray)
979: */
980: static inline PetscErrorCode PetscLLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt)
981: {
982: PetscFunctionBegin;
983: PetscCall(PetscLLAdd_Private(nidx, indices, idx_start, nlnk, lnk, bt, PETSC_TRUE));
984: PetscFunctionReturn(PETSC_SUCCESS);
985: }
987: /*
988: Add a permuted index set into a sorted linked list
989: Input Parameters:
990: nidx - number of input indices
991: indices - integer array
992: perm - permutation of indices
993: idx_start - starting index of the list
994: lnk - linked list(an integer array) that is created
995: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
996: output Parameters:
997: nlnk - number of newly added indices
998: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
999: bt - updated PetscBT (bitarray)
1000: */
1001: static inline PetscErrorCode PetscLLAddPerm(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, const PetscInt *PETSC_RESTRICT perm, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt)
1002: {
1003: PetscFunctionBegin;
1004: *nlnk = 0;
1005: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1006: const PetscInt entry = perm[indices[k]];
1008: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_FALSE, k, idx_start, entry, nlnk, &lnkdata, lnk));
1009: }
1010: PetscFunctionReturn(PETSC_SUCCESS);
1011: }
1013: #if 0
1014: /* this appears to be unused? */
1015: static inline PetscErrorCode PetscLLAddSorted_new(PetscInt nidx, PetscInt *indices, PetscInt idx_start, PetscBool *lnk_empty, PetscInt *nlnk, PetscInt *lnk, PetscBT bt)
1016: {
1017: PetscInt lnkdata = idx_start;
1019: PetscFunctionBegin;
1020: if (*lnk_empty) {
1021: for (PetscInt k = 0; k < nidx; ++k) {
1022: const PetscInt entry = indices[k], location = lnkdata;
1024: PetscCall(PetscBTSet(bt,entry)); /* mark the new entry */
1025: lnkdata = lnk[location];
1026: /* insertion location is found, add entry into lnk */
1027: lnk[location] = entry;
1028: lnk[entry] = lnkdata;
1029: lnkdata = entry; /* next search starts from here */
1030: }
1031: /* lnk[indices[nidx-1]] = lnk[idx_start];
1032: lnk[idx_start] = indices[0];
1033: PetscCall(PetscBTSet(bt,indices[0]));
1034: for (_k=1; _k<nidx; _k++) {
1035: PetscCall(PetscBTSet(bt,indices[_k]));
1036: lnk[indices[_k-1]] = indices[_k];
1037: }
1038: */
1039: *nlnk = nidx;
1040: *lnk_empty = PETSC_FALSE;
1041: } else {
1042: *nlnk = 0;
1043: for (PetscInt k = 0; k < nidx; ++k) {
1044: const PetscInt entry = indices[k];
1046: if (!PetscBTLookupSet(bt,entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_TRUE,k,idx_start,entry,nlnk,&lnkdata,lnk));
1047: }
1048: }
1049: PetscFunctionReturn(PETSC_SUCCESS);
1050: }
1051: #endif
1053: /*
1054: Add a SORTED index set into a sorted linked list used for LUFactorSymbolic()
1055: Same as PetscLLAddSorted() with an additional operation:
1056: count the number of input indices that are no larger than 'diag'
1057: Input Parameters:
1058: indices - sorted integer array
1059: idx_start - starting index of the list, index of pivot row
1060: lnk - linked list(an integer array) that is created
1061: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1062: diag - index of the active row in LUFactorSymbolic
1063: nzbd - number of input indices with indices <= idx_start
1064: im - im[idx_start] is initialized as num of nonzero entries in row=idx_start
1065: output Parameters:
1066: nlnk - number of newly added indices
1067: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
1068: bt - updated PetscBT (bitarray)
1069: im - im[idx_start]: unchanged if diag is not an entry
1070: : num of entries with indices <= diag if diag is an entry
1071: */
1072: static inline PetscErrorCode PetscLLAddSortedLU(const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt, PetscInt diag, PetscInt nzbd, PetscInt *PETSC_RESTRICT im)
1073: {
1074: const PetscInt nidx = im[idx_start] - nzbd; /* num of entries with idx_start < index <= diag */
1076: PetscFunctionBegin;
1077: *nlnk = 0;
1078: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1079: const PetscInt entry = indices[k];
1081: ++nzbd;
1082: if (entry == diag) im[idx_start] = nzbd;
1083: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_TRUE, k, idx_start, entry, nlnk, &lnkdata, lnk));
1084: }
1085: PetscFunctionReturn(PETSC_SUCCESS);
1086: }
1088: /*
1089: Copy data on the list into an array, then initialize the list
1090: Input Parameters:
1091: idx_start - starting index of the list
1092: lnk_max - max value of lnk indicating the end of the list
1093: nlnk - number of data on the list to be copied
1094: lnk - linked list
1095: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1096: output Parameters:
1097: indices - array that contains the copied data
1098: lnk - linked list that is cleaned and initialize
1099: bt - PetscBT (bitarray) with all bits set to false
1100: */
1101: static inline PetscErrorCode PetscLLClean(PetscInt idx_start, PetscInt lnk_max, PetscInt nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT indices, PetscBT bt)
1102: {
1103: PetscFunctionBegin;
1104: for (PetscInt j = 0, idx = idx_start; j < nlnk; ++j) {
1105: idx = lnk[idx];
1106: indices[j] = idx;
1107: PetscCall(PetscBTClear(bt, idx));
1108: }
1109: lnk[idx_start] = lnk_max;
1110: PetscFunctionReturn(PETSC_SUCCESS);
1111: }
1113: /*
1114: Free memories used by the list
1115: */
1116: #define PetscLLDestroy(lnk, bt) ((PetscErrorCode)(PetscFree(lnk) || PetscBTDestroy(&(bt))))
1118: /* Routines below are used for incomplete matrix factorization */
1119: /*
1120: Create and initialize a linked list and its levels
1121: Input Parameters:
1122: idx_start - starting index of the list
1123: lnk_max - max value of lnk indicating the end of the list
1124: nlnk - max length of the list
1125: Output Parameters:
1126: lnk - list initialized
1127: lnk_lvl - array of size nlnk for storing levels of lnk
1128: bt - PetscBT (bitarray) with all bits set to false
1129: */
1130: #define PetscIncompleteLLCreate(idx_start, lnk_max, nlnk, lnk, lnk_lvl, bt) \
1131: ((PetscErrorCode)(PetscIntMultError(2, nlnk, NULL) || PetscMalloc1(2 * (nlnk), &(lnk)) || PetscBTCreate(nlnk, &(bt)) || ((lnk)[idx_start] = lnk_max, lnk_lvl = (lnk) + (nlnk), PETSC_SUCCESS)))
1133: static inline PetscErrorCode PetscIncompleteLLInsertLocation_Private(PetscBool assume_sorted, PetscInt k, PetscInt idx_start, PetscInt entry, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnkdata, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscInt newval)
1134: {
1135: PetscFunctionBegin;
1136: PetscCall(PetscLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, lnkdata, lnk));
1137: lnklvl[entry] = newval;
1138: PetscFunctionReturn(PETSC_SUCCESS);
1139: }
1141: /*
1142: Initialize a sorted linked list used for ILU and ICC
1143: Input Parameters:
1144: nidx - number of input idx
1145: idx - integer array used for storing column indices
1146: idx_start - starting index of the list
1147: perm - indices of an IS
1148: lnk - linked list(an integer array) that is created
1149: lnklvl - levels of lnk
1150: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1151: output Parameters:
1152: nlnk - number of newly added idx
1153: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1154: lnklvl - levels of lnk
1155: bt - updated PetscBT (bitarray)
1156: */
1157: static inline PetscErrorCode PetscIncompleteLLInit(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscInt idx_start, const PetscInt *PETSC_RESTRICT perm, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt)
1158: {
1159: PetscFunctionBegin;
1160: *nlnk = 0;
1161: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1162: const PetscInt entry = perm[idx[k]];
1164: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscIncompleteLLInsertLocation_Private(PETSC_FALSE, k, idx_start, entry, nlnk, &lnkdata, lnk, lnklvl, 0));
1165: }
1166: PetscFunctionReturn(PETSC_SUCCESS);
1167: }
1169: static inline PetscErrorCode PetscIncompleteLLAdd_Private(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscInt level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt, PetscInt prow_offset, PetscBool assume_sorted)
1170: {
1171: PetscFunctionBegin;
1172: *nlnk = 0;
1173: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1174: const PetscInt incrlev = idxlvl[k] + prow_offset + 1;
1176: if (incrlev <= level) {
1177: const PetscInt entry = idx[k];
1179: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscIncompleteLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, &lnkdata, lnk, lnklvl, incrlev));
1180: else if (lnklvl[entry] > incrlev) lnklvl[entry] = incrlev; /* existing entry */
1181: }
1182: }
1183: PetscFunctionReturn(PETSC_SUCCESS);
1184: }
1186: /*
1187: Add a SORTED index set into a sorted linked list for ICC
1188: Input Parameters:
1189: nidx - number of input indices
1190: idx - sorted integer array used for storing column indices
1191: level - level of fill, e.g., ICC(level)
1192: idxlvl - level of idx
1193: idx_start - starting index of the list
1194: lnk - linked list(an integer array) that is created
1195: lnklvl - levels of lnk
1196: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1197: idxlvl_prow - idxlvl[prow], where prow is the row number of the idx
1198: output Parameters:
1199: nlnk - number of newly added indices
1200: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1201: lnklvl - levels of lnk
1202: bt - updated PetscBT (bitarray)
1203: Note: the level of U(i,j) is set as lvl(i,j) = min{ lvl(i,j), lvl(prow,i)+lvl(prow,j)+1)
1204: where idx = non-zero columns of U(prow,prow+1:n-1), prow<i
1205: */
1206: static inline PetscErrorCode PetscICCLLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscInt level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt, PetscInt idxlvl_prow)
1207: {
1208: PetscFunctionBegin;
1209: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, idxlvl_prow, PETSC_TRUE));
1210: PetscFunctionReturn(PETSC_SUCCESS);
1211: }
1213: /*
1214: Add a SORTED index set into a sorted linked list for ILU
1215: Input Parameters:
1216: nidx - number of input indices
1217: idx - sorted integer array used for storing column indices
1218: level - level of fill, e.g., ICC(level)
1219: idxlvl - level of idx
1220: idx_start - starting index of the list
1221: lnk - linked list(an integer array) that is created
1222: lnklvl - levels of lnk
1223: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1224: prow - the row number of idx
1225: output Parameters:
1226: nlnk - number of newly added idx
1227: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1228: lnklvl - levels of lnk
1229: bt - updated PetscBT (bitarray)
1231: Note: the level of factor(i,j) is set as lvl(i,j) = min{ lvl(i,j), lvl(i,prow)+lvl(prow,j)+1)
1232: where idx = non-zero columns of U(prow,prow+1:n-1), prow<i
1233: */
1234: static inline PetscErrorCode PetscILULLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscInt level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt, PetscInt prow)
1235: {
1236: PetscFunctionBegin;
1237: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, lnklvl[prow], PETSC_TRUE));
1238: PetscFunctionReturn(PETSC_SUCCESS);
1239: }
1241: /*
1242: Add a index set into a sorted linked list
1243: Input Parameters:
1244: nidx - number of input idx
1245: idx - integer array used for storing column indices
1246: level - level of fill, e.g., ICC(level)
1247: idxlvl - level of idx
1248: idx_start - starting index of the list
1249: lnk - linked list(an integer array) that is created
1250: lnklvl - levels of lnk
1251: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1252: output Parameters:
1253: nlnk - number of newly added idx
1254: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1255: lnklvl - levels of lnk
1256: bt - updated PetscBT (bitarray)
1257: */
1258: static inline PetscErrorCode PetscIncompleteLLAdd(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscInt level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt)
1259: {
1260: PetscFunctionBegin;
1261: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, 0, PETSC_FALSE));
1262: PetscFunctionReturn(PETSC_SUCCESS);
1263: }
1265: /*
1266: Add a SORTED index set into a sorted linked list
1267: Input Parameters:
1268: nidx - number of input indices
1269: idx - sorted integer array used for storing column indices
1270: level - level of fill, e.g., ICC(level)
1271: idxlvl - level of idx
1272: idx_start - starting index of the list
1273: lnk - linked list(an integer array) that is created
1274: lnklvl - levels of lnk
1275: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1276: output Parameters:
1277: nlnk - number of newly added idx
1278: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1279: lnklvl - levels of lnk
1280: bt - updated PetscBT (bitarray)
1281: */
1282: static inline PetscErrorCode PetscIncompleteLLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscInt level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt)
1283: {
1284: PetscFunctionBegin;
1285: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, 0, PETSC_TRUE));
1286: PetscFunctionReturn(PETSC_SUCCESS);
1287: }
1289: /*
1290: Copy data on the list into an array, then initialize the list
1291: Input Parameters:
1292: idx_start - starting index of the list
1293: lnk_max - max value of lnk indicating the end of the list
1294: nlnk - number of data on the list to be copied
1295: lnk - linked list
1296: lnklvl - level of lnk
1297: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1298: output Parameters:
1299: indices - array that contains the copied data
1300: lnk - linked list that is cleaned and initialize
1301: lnklvl - level of lnk that is reinitialized
1302: bt - PetscBT (bitarray) with all bits set to false
1303: */
1304: static inline PetscErrorCode PetscIncompleteLLClean(PetscInt idx_start, PetscInt lnk_max, PetscInt nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscInt *PETSC_RESTRICT indices, PetscInt *PETSC_RESTRICT indiceslvl, PetscBT bt)
1305: {
1306: PetscFunctionBegin;
1307: for (PetscInt j = 0, idx = idx_start; j < nlnk; ++j) {
1308: idx = lnk[idx];
1309: indices[j] = idx;
1310: indiceslvl[j] = lnklvl[idx];
1311: lnklvl[idx] = -1;
1312: PetscCall(PetscBTClear(bt, idx));
1313: }
1314: lnk[idx_start] = lnk_max;
1315: PetscFunctionReturn(PETSC_SUCCESS);
1316: }
1318: /*
1319: Free memories used by the list
1320: */
1321: #define PetscIncompleteLLDestroy(lnk, bt) ((PetscErrorCode)(PetscFree(lnk) || PetscBTDestroy(&(bt))))
1323: #if !PetscDefined(CLANG_STATIC_ANALYZER)
1324: #define MatCheckSameLocalSize(A, ar1, B, ar2) \
1325: do { \
1326: PetscCheckSameComm(A, ar1, B, ar2); \
1327: PetscCheck(((A)->rmap->n == (B)->rmap->n) && ((A)->cmap->n == (B)->cmap->n), PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Incompatible matrix local sizes: parameter # %d (%" PetscInt_FMT " x %" PetscInt_FMT ") != parameter # %d (%" PetscInt_FMT " x %" PetscInt_FMT ")", ar1, \
1328: (A)->rmap->n, (A)->cmap->n, ar2, (B)->rmap->n, (B)->cmap->n); \
1329: } while (0)
1330: #define MatCheckSameSize(A, ar1, B, ar2) \
1331: do { \
1332: PetscCheck(((A)->rmap->N == (B)->rmap->N) && ((A)->cmap->N == (B)->cmap->N), PetscObjectComm((PetscObject)(A)), PETSC_ERR_ARG_INCOMP, "Incompatible matrix global sizes: parameter # %d (%" PetscInt_FMT " x %" PetscInt_FMT ") != parameter # %d (%" PetscInt_FMT " x %" PetscInt_FMT ")", ar1, \
1333: (A)->rmap->N, (A)->cmap->N, ar2, (B)->rmap->N, (B)->cmap->N); \
1334: MatCheckSameLocalSize(A, ar1, B, ar2); \
1335: } while (0)
1336: #else
1337: template <typename Tm>
1338: extern void MatCheckSameLocalSize(Tm, int, Tm, int);
1339: template <typename Tm>
1340: extern void MatCheckSameSize(Tm, int, Tm, int);
1341: #endif
1343: #define VecCheckMatCompatible(M, x, ar1, b, ar2) \
1344: do { \
1345: PetscCheck((M)->cmap->N == (x)->map->N, PetscObjectComm((PetscObject)(M)), PETSC_ERR_ARG_SIZ, "Vector global length incompatible with matrix: parameter # %d global size %" PetscInt_FMT " != matrix column global size %" PetscInt_FMT, ar1, (x)->map->N, \
1346: (M)->cmap->N); \
1347: PetscCheck((M)->rmap->N == (b)->map->N, PetscObjectComm((PetscObject)(M)), PETSC_ERR_ARG_SIZ, "Vector global length incompatible with matrix: parameter # %d global size %" PetscInt_FMT " != matrix row global size %" PetscInt_FMT, ar2, (b)->map->N, \
1348: (M)->rmap->N); \
1349: } while (0)
1351: /*
1352: Create and initialize a condensed linked list -
1353: same as PetscLLCreate(), but uses a scalable array 'lnk' with size of max number of entries, not O(N).
1354: Barry suggested this approach (Dec. 6, 2011):
1355: I've thought of an alternative way of representing a linked list that is efficient but doesn't have the O(N) scaling issue
1356: (it may be faster than the O(N) even sequentially due to less crazy memory access).
1358: Instead of having some like a 2 -> 4 -> 11 -> 22 list that uses slot 2 4 11 and 22 in a big array use a small array with two slots
1359: for each entry for example [ 2 1 | 4 3 | 22 -1 | 11 2] so the first number (of the pair) is the value while the second tells you where
1360: in the list the next entry is. Inserting a new link means just append another pair at the end. For example say we want to insert 13 into the
1361: list it would then become [2 1 | 4 3 | 22 -1 | 11 4 | 13 2 ] you just add a pair at the end and fix the point for the one that points to it.
1362: That is 11 use to point to the 2 slot, after the change 11 points to the 4th slot which has the value 13. Note that values are always next
1363: to each other so memory access is much better than using the big array.
1365: Example:
1366: nlnk_max=5, lnk_max=36:
1367: Initial list: [0, 0 | 36, 2 | 0, 0 | 0, 0 | 0, 0 | 0, 0 | 0, 0]
1368: here, head_node has index 2 with value lnk[2]=lnk_max=36,
1369: 0-th entry is used to store the number of entries in the list,
1370: The initial lnk represents head -> tail(marked by 36) with number of entries = lnk[0]=0.
1372: Now adding a sorted set {2,4}, the list becomes
1373: [2, 0 | 36, 4 |2, 6 | 4, 2 | 0, 0 | 0, 0 | 0, 0 ]
1374: represents head -> 2 -> 4 -> tail with number of entries = lnk[0]=2.
1376: Then adding a sorted set {0,3,35}, the list
1377: [5, 0 | 36, 8 | 2, 10 | 4, 12 | 0, 4 | 3, 6 | 35, 2 ]
1378: represents head -> 0 -> 2 -> 3 -> 4 -> 35 -> tail with number of entries = lnk[0]=5.
1380: Input Parameters:
1381: nlnk_max - max length of the list
1382: lnk_max - max value of the entries
1383: Output Parameters:
1384: lnk - list created and initialized
1385: bt - PetscBT (bitarray) with all bits set to false. Note: bt has size lnk_max, not nln_max!
1386: */
1387: static inline PetscErrorCode PetscLLCondensedCreate(PetscInt nlnk_max, PetscInt lnk_max, PetscInt **lnk, PetscBT *bt)
1388: {
1389: PetscInt *llnk, lsize = 0;
1391: PetscFunctionBegin;
1392: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1393: PetscCall(PetscMalloc1(lsize, lnk));
1394: PetscCall(PetscBTCreate(lnk_max, bt));
1395: llnk = *lnk;
1396: llnk[0] = 0; /* number of entries on the list */
1397: llnk[2] = lnk_max; /* value in the head node */
1398: llnk[3] = 2; /* next for the head node */
1399: PetscFunctionReturn(PETSC_SUCCESS);
1400: }
1402: /*
1403: Add a SORTED ascending index set into a sorted linked list. See PetscLLCondensedCreate() for detailed description.
1404: Input Parameters:
1405: nidx - number of input indices
1406: indices - sorted integer array
1407: lnk - condensed linked list(an integer array) that is created
1408: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1409: output Parameters:
1410: lnk - the sorted(increasing order) linked list containing previous and newly added non-redundate indices
1411: bt - updated PetscBT (bitarray)
1412: */
1413: static inline PetscErrorCode PetscLLCondensedAddSorted(PetscInt nidx, const PetscInt indices[], PetscInt lnk[], PetscBT bt)
1414: {
1415: PetscInt location = 2; /* head */
1416: PetscInt nlnk = lnk[0]; /* num of entries on the input lnk */
1418: PetscFunctionBegin;
1419: for (PetscInt k = 0; k < nidx; k++) {
1420: const PetscInt entry = indices[k];
1421: if (!PetscBTLookupSet(bt, entry)) { /* new entry */
1422: PetscInt next, lnkdata;
1424: /* search for insertion location */
1425: do {
1426: next = location + 1; /* link from previous node to next node */
1427: location = lnk[next]; /* idx of next node */
1428: lnkdata = lnk[location]; /* value of next node */
1429: } while (entry > lnkdata);
1430: /* insertion location is found, add entry into lnk */
1431: const PetscInt newnode = 2 * (nlnk + 2); /* index for this new node */
1432: lnk[next] = newnode; /* connect previous node to the new node */
1433: lnk[newnode] = entry; /* set value of the new node */
1434: lnk[newnode + 1] = location; /* connect new node to next node */
1435: location = newnode; /* next search starts from the new node */
1436: nlnk++;
1437: }
1438: }
1439: lnk[0] = nlnk; /* number of entries in the list */
1440: PetscFunctionReturn(PETSC_SUCCESS);
1441: }
1443: static inline PetscErrorCode PetscLLCondensedClean(PetscInt lnk_max, PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt lnk[], PetscBT bt)
1444: {
1445: const PetscInt nlnk = lnk[0]; /* num of entries on the list */
1446: PetscInt next = lnk[3]; /* head node */
1448: PetscFunctionBegin;
1449: for (PetscInt k = 0; k < nlnk; k++) {
1450: indices[k] = lnk[next];
1451: next = lnk[next + 1];
1452: PetscCall(PetscBTClear(bt, indices[k]));
1453: }
1454: lnk[0] = 0; /* num of entries on the list */
1455: lnk[2] = lnk_max; /* initialize head node */
1456: lnk[3] = 2; /* head node */
1457: PetscFunctionReturn(PETSC_SUCCESS);
1458: }
1460: static inline PetscErrorCode PetscLLCondensedView(PetscInt *lnk)
1461: {
1462: PetscFunctionBegin;
1463: PetscCall(PetscPrintf(PETSC_COMM_SELF, "LLCondensed of size %" PetscInt_FMT ", (val, next)\n", lnk[0]));
1464: for (PetscInt k = 2; k < lnk[0] + 2; ++k) PetscCall(PetscPrintf(PETSC_COMM_SELF, " %" PetscInt_FMT ": (%" PetscInt_FMT ", %" PetscInt_FMT ")\n", 2 * k, lnk[2 * k], lnk[2 * k + 1]));
1465: PetscFunctionReturn(PETSC_SUCCESS);
1466: }
1468: /*
1469: Free memories used by the list
1470: */
1471: static inline PetscErrorCode PetscLLCondensedDestroy(PetscInt *lnk, PetscBT bt)
1472: {
1473: PetscFunctionBegin;
1474: PetscCall(PetscFree(lnk));
1475: PetscCall(PetscBTDestroy(&bt));
1476: PetscFunctionReturn(PETSC_SUCCESS);
1477: }
1479: /*
1480: Same as PetscLLCondensedCreate(), but does not use non-scalable O(lnk_max) bitarray
1481: Input Parameters:
1482: nlnk_max - max length of the list
1483: Output Parameters:
1484: lnk - list created and initialized
1485: */
1486: static inline PetscErrorCode PetscLLCondensedCreate_Scalable(PetscInt nlnk_max, PetscInt **lnk)
1487: {
1488: PetscInt *llnk, lsize = 0;
1490: PetscFunctionBegin;
1491: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1492: PetscCall(PetscMalloc1(lsize, lnk));
1493: llnk = *lnk;
1494: llnk[0] = 0; /* number of entries on the list */
1495: llnk[2] = PETSC_INT_MAX; /* value in the head node */
1496: llnk[3] = 2; /* next for the head node */
1497: PetscFunctionReturn(PETSC_SUCCESS);
1498: }
1500: static inline PetscErrorCode PetscLLCondensedExpand_Scalable(PetscInt nlnk_max, PetscInt **lnk)
1501: {
1502: PetscInt lsize = 0;
1504: PetscFunctionBegin;
1505: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1506: PetscCall(PetscRealloc((size_t)lsize * sizeof(PetscInt), lnk));
1507: PetscFunctionReturn(PETSC_SUCCESS);
1508: }
1510: static inline PetscErrorCode PetscLLCondensedAddSorted_Scalable(PetscInt nidx, const PetscInt indices[], PetscInt lnk[])
1511: {
1512: PetscInt location = 2; /* head */
1513: PetscInt nlnk = lnk[0]; /* num of entries on the input lnk */
1515: for (PetscInt k = 0; k < nidx; k++) {
1516: const PetscInt entry = indices[k];
1517: PetscInt next, lnkdata;
1519: /* search for insertion location */
1520: do {
1521: next = location + 1; /* link from previous node to next node */
1522: location = lnk[next]; /* idx of next node */
1523: lnkdata = lnk[location]; /* value of next node */
1524: } while (entry > lnkdata);
1525: if (entry < lnkdata) {
1526: /* insertion location is found, add entry into lnk */
1527: const PetscInt newnode = 2 * (nlnk + 2); /* index for this new node */
1528: lnk[next] = newnode; /* connect previous node to the new node */
1529: lnk[newnode] = entry; /* set value of the new node */
1530: lnk[newnode + 1] = location; /* connect new node to next node */
1531: location = newnode; /* next search starts from the new node */
1532: nlnk++;
1533: }
1534: }
1535: lnk[0] = nlnk; /* number of entries in the list */
1536: return PETSC_SUCCESS;
1537: }
1539: static inline PetscErrorCode PetscLLCondensedClean_Scalable(PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt *lnk)
1540: {
1541: const PetscInt nlnk = lnk[0];
1542: PetscInt next = lnk[3]; /* head node */
1544: for (PetscInt k = 0; k < nlnk; k++) {
1545: indices[k] = lnk[next];
1546: next = lnk[next + 1];
1547: }
1548: lnk[0] = 0; /* num of entries on the list */
1549: lnk[3] = 2; /* head node */
1550: return PETSC_SUCCESS;
1551: }
1553: static inline PetscErrorCode PetscLLCondensedDestroy_Scalable(PetscInt *lnk)
1554: {
1555: return PetscFree(lnk);
1556: }
1558: /*
1559: lnk[0] number of links
1560: lnk[1] number of entries
1561: lnk[3n] value
1562: lnk[3n+1] len
1563: lnk[3n+2] link to next value
1565: The next three are always the first link
1567: lnk[3] PETSC_INT_MIN+1
1568: lnk[4] 1
1569: lnk[5] link to first real entry
1571: The next three are always the last link
1573: lnk[6] PETSC_INT_MAX - 1
1574: lnk[7] 1
1575: lnk[8] next valid link (this is the same as lnk[0] but without the decreases)
1576: */
1578: static inline PetscErrorCode PetscLLCondensedCreate_fast(PetscInt nlnk_max, PetscInt **lnk)
1579: {
1580: PetscInt *llnk;
1581: PetscInt lsize = 0;
1583: PetscFunctionBegin;
1584: PetscCall(PetscIntMultError(3, nlnk_max + 3, &lsize));
1585: PetscCall(PetscMalloc1(lsize, lnk));
1586: llnk = *lnk;
1587: llnk[0] = 0; /* nlnk: number of entries on the list */
1588: llnk[1] = 0; /* number of integer entries represented in list */
1589: llnk[3] = PETSC_INT_MIN + 1; /* value in the first node */
1590: llnk[4] = 1; /* count for the first node */
1591: llnk[5] = 6; /* next for the first node */
1592: llnk[6] = PETSC_INT_MAX - 1; /* value in the last node */
1593: llnk[7] = 1; /* count for the last node */
1594: llnk[8] = 0; /* next valid node to be used */
1595: PetscFunctionReturn(PETSC_SUCCESS);
1596: }
1598: static inline PetscErrorCode PetscLLCondensedAddSorted_fast(PetscInt nidx, const PetscInt indices[], PetscInt lnk[])
1599: {
1600: for (PetscInt k = 0, prev = 3 /* first value */; k < nidx; k++) {
1601: const PetscInt entry = indices[k];
1602: PetscInt next = lnk[prev + 2];
1604: /* search for insertion location */
1605: while (entry >= lnk[next]) {
1606: prev = next;
1607: next = lnk[next + 2];
1608: }
1609: /* entry is in range of previous list */
1610: if (entry < lnk[prev] + lnk[prev + 1]) continue;
1611: lnk[1]++;
1612: /* entry is right after previous list */
1613: if (entry == lnk[prev] + lnk[prev + 1]) {
1614: lnk[prev + 1]++;
1615: if (lnk[next] == entry + 1) { /* combine two contiguous strings */
1616: lnk[prev + 1] += lnk[next + 1];
1617: lnk[prev + 2] = lnk[next + 2];
1618: lnk[0]--;
1619: }
1620: continue;
1621: }
1622: /* entry is right before next list */
1623: if (entry == lnk[next] - 1) {
1624: lnk[next]--;
1625: lnk[next + 1]++;
1626: prev = next;
1627: continue;
1628: }
1629: /* add entry into lnk */
1630: lnk[prev + 2] = 3 * ((lnk[8]++) + 3); /* connect previous node to the new node */
1631: prev = lnk[prev + 2];
1632: lnk[prev] = entry; /* set value of the new node */
1633: lnk[prev + 1] = 1; /* number of values in contiguous string is one to start */
1634: lnk[prev + 2] = next; /* connect new node to next node */
1635: lnk[0]++;
1636: }
1637: return PETSC_SUCCESS;
1638: }
1640: static inline PetscErrorCode PetscLLCondensedClean_fast(PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt *lnk)
1641: {
1642: const PetscInt nlnk = lnk[0];
1643: PetscInt next = lnk[5]; /* first node */
1645: for (PetscInt k = 0, cnt = 0; k < nlnk; k++) {
1646: for (PetscInt j = 0; j < lnk[next + 1]; j++) indices[cnt++] = lnk[next] + j;
1647: next = lnk[next + 2];
1648: }
1649: lnk[0] = 0; /* nlnk: number of links */
1650: lnk[1] = 0; /* number of integer entries represented in list */
1651: lnk[3] = PETSC_INT_MIN + 1; /* value in the first node */
1652: lnk[4] = 1; /* count for the first node */
1653: lnk[5] = 6; /* next for the first node */
1654: lnk[6] = PETSC_INT_MAX - 1; /* value in the last node */
1655: lnk[7] = 1; /* count for the last node */
1656: lnk[8] = 0; /* next valid location to make link */
1657: return PETSC_SUCCESS;
1658: }
1660: static inline PetscErrorCode PetscLLCondensedView_fast(const PetscInt *lnk)
1661: {
1662: const PetscInt nlnk = lnk[0];
1663: PetscInt next = lnk[5]; /* first node */
1665: for (PetscInt k = 0; k < nlnk; k++) {
1666: #if 0 /* Debugging code */
1667: printf("%d value %d len %d next %d\n", next, lnk[next], lnk[next + 1], lnk[next + 2]);
1668: #endif
1669: next = lnk[next + 2];
1670: }
1671: return PETSC_SUCCESS;
1672: }
1674: static inline PetscErrorCode PetscLLCondensedDestroy_fast(PetscInt *lnk)
1675: {
1676: return PetscFree(lnk);
1677: }
1679: PETSC_EXTERN PetscErrorCode PetscCDCreate(PetscInt, PetscCoarsenData **);
1680: PETSC_EXTERN PetscErrorCode PetscCDDestroy(PetscCoarsenData *);
1681: PETSC_EXTERN PetscErrorCode PetscCDIntNdSetID(PetscCDIntNd *, PetscInt);
1682: PETSC_EXTERN PetscErrorCode PetscCDIntNdGetID(const PetscCDIntNd *, PetscInt *);
1683: PETSC_EXTERN PetscErrorCode PetscCDAppendID(PetscCoarsenData *, PetscInt, PetscInt);
1684: PETSC_EXTERN PetscErrorCode PetscCDMoveAppend(PetscCoarsenData *, PetscInt, PetscInt);
1685: PETSC_EXTERN PetscErrorCode PetscCDAppendNode(PetscCoarsenData *, PetscInt, PetscCDIntNd *);
1686: PETSC_EXTERN PetscErrorCode PetscCDRemoveNextNode(PetscCoarsenData *, PetscInt, PetscCDIntNd *);
1687: PETSC_EXTERN PetscErrorCode PetscCDCountAt(const PetscCoarsenData *, PetscInt, PetscInt *);
1688: PETSC_EXTERN PetscErrorCode PetscCDIsEmptyAt(const PetscCoarsenData *, PetscInt, PetscBool *);
1689: PETSC_EXTERN PetscErrorCode PetscCDSetChunkSize(PetscCoarsenData *, PetscInt);
1690: PETSC_EXTERN PetscErrorCode PetscCDPrint(const PetscCoarsenData *, PetscInt, MPI_Comm);
1691: PETSC_EXTERN PetscErrorCode PetscCDGetNonemptyIS(PetscCoarsenData *, IS *);
1692: PETSC_EXTERN PetscErrorCode PetscCDGetMat(PetscCoarsenData *, Mat *);
1693: PETSC_EXTERN PetscErrorCode PetscCDSetMat(PetscCoarsenData *, Mat);
1694: PETSC_EXTERN PetscErrorCode PetscCDClearMat(PetscCoarsenData *);
1695: PETSC_EXTERN PetscErrorCode PetscCDRemoveAllAt(PetscCoarsenData *, PetscInt);
1696: PETSC_EXTERN PetscErrorCode PetscCDCount(const PetscCoarsenData *, PetscInt *_sz);
1698: PETSC_EXTERN PetscErrorCode PetscCDGetHeadPos(const PetscCoarsenData *, PetscInt, PetscCDIntNd **);
1699: PETSC_EXTERN PetscErrorCode PetscCDGetNextPos(const PetscCoarsenData *, PetscInt, PetscCDIntNd **);
1700: PETSC_EXTERN PetscErrorCode PetscCDGetASMBlocks(const PetscCoarsenData *, const PetscInt, PetscInt *, IS **);
1702: PETSC_SINGLE_LIBRARY_VISIBILITY_INTERNAL PetscErrorCode MatFDColoringApply_AIJ(Mat, MatFDColoring, Vec, void *);
1704: typedef struct {
1705: Vec diag;
1706: PetscBool diag_valid;
1707: Vec inv_diag;
1708: PetscBool inv_diag_valid;
1709: PetscObjectState diag_state, inv_diag_state;
1710: PetscInt *col;
1711: PetscScalar *val;
1712: } Mat_Diagonal;
1714: #if PetscDefined(HAVE_CUDA)
1715: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqCUDA(Mat, Vec, Vec, PetscScalar *);
1716: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqCUDA(Mat, Vec, PetscReal *);
1717: #endif
1718: #if PetscDefined(HAVE_HIP)
1719: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqHIP(Mat, Vec, Vec, PetscScalar *);
1720: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqHIP(Mat, Vec, PetscReal *);
1721: #endif
1722: #if PetscDefined(HAVE_KOKKOS_KERNELS)
1723: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqKokkos(Mat, Vec, Vec, PetscScalar *);
1724: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqKokkos(Mat, Vec, PetscReal *);
1725: #endif
1727: PETSC_EXTERN PetscLogEvent MAT_Mult;
1728: PETSC_EXTERN PetscLogEvent MAT_MultAdd;
1729: PETSC_EXTERN PetscLogEvent MAT_MultTranspose;
1730: PETSC_EXTERN PetscLogEvent MAT_MultHermitianTranspose;
1731: PETSC_EXTERN PetscLogEvent MAT_MultTransposeAdd;
1732: PETSC_EXTERN PetscLogEvent MAT_MultHermitianTransposeAdd;
1733: PETSC_EXTERN PetscLogEvent MAT_ADot;
1734: PETSC_EXTERN PetscLogEvent MAT_ANorm;
1735: PETSC_EXTERN PetscLogEvent MAT_Solve;
1736: PETSC_EXTERN PetscLogEvent MAT_Solves;
1737: PETSC_EXTERN PetscLogEvent MAT_SolveAdd;
1738: PETSC_EXTERN PetscLogEvent MAT_SolveTranspose;
1739: PETSC_EXTERN PetscLogEvent MAT_SolveTransposeAdd;
1740: PETSC_EXTERN PetscLogEvent MAT_SOR;
1741: PETSC_EXTERN PetscLogEvent MAT_ForwardSolve;
1742: PETSC_EXTERN PetscLogEvent MAT_BackwardSolve;
1743: PETSC_EXTERN PetscLogEvent MAT_LUFactor;
1744: PETSC_EXTERN PetscLogEvent MAT_LUFactorSymbolic;
1745: PETSC_EXTERN PetscLogEvent MAT_LUFactorNumeric;
1746: PETSC_EXTERN PetscLogEvent MAT_QRFactor;
1747: PETSC_EXTERN PetscLogEvent MAT_QRFactorSymbolic;
1748: PETSC_EXTERN PetscLogEvent MAT_QRFactorNumeric;
1749: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactor;
1750: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactorSymbolic;
1751: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactorNumeric;
1752: PETSC_EXTERN PetscLogEvent MAT_ILUFactor;
1753: PETSC_EXTERN PetscLogEvent MAT_ILUFactorSymbolic;
1754: PETSC_EXTERN PetscLogEvent MAT_ICCFactorSymbolic;
1755: PETSC_EXTERN PetscLogEvent MAT_Copy;
1756: PETSC_EXTERN PetscLogEvent MAT_Convert;
1757: PETSC_EXTERN PetscLogEvent MAT_Scale;
1758: PETSC_EXTERN PetscLogEvent MAT_AssemblyBegin;
1759: PETSC_EXTERN PetscLogEvent MAT_AssemblyEnd;
1760: PETSC_EXTERN PetscLogEvent MAT_SetValues;
1761: PETSC_EXTERN PetscLogEvent MAT_GetValues;
1762: PETSC_EXTERN PetscLogEvent MAT_GetRow;
1763: PETSC_EXTERN PetscLogEvent MAT_GetRowIJ;
1764: PETSC_EXTERN PetscLogEvent MAT_CreateSubMats;
1765: PETSC_EXTERN PetscLogEvent MAT_GetOrdering;
1766: PETSC_EXTERN PetscLogEvent MAT_RedundantMat;
1767: PETSC_EXTERN PetscLogEvent MAT_IncreaseOverlap;
1768: PETSC_EXTERN PetscLogEvent MAT_Partitioning;
1769: PETSC_EXTERN PetscLogEvent MAT_PartitioningND;
1770: PETSC_EXTERN PetscLogEvent MAT_Coarsen;
1771: PETSC_EXTERN PetscLogEvent MAT_ZeroEntries;
1772: PETSC_EXTERN PetscLogEvent MAT_Load;
1773: PETSC_EXTERN PetscLogEvent MAT_View;
1774: PETSC_EXTERN PetscLogEvent MAT_AXPY;
1775: PETSC_EXTERN PetscLogEvent MAT_FDColoringCreate;
1776: PETSC_EXTERN PetscLogEvent MAT_TransposeColoringCreate;
1777: PETSC_EXTERN PetscLogEvent MAT_FDColoringSetUp;
1778: PETSC_EXTERN PetscLogEvent MAT_FDColoringApply;
1779: PETSC_EXTERN PetscLogEvent MAT_Transpose;
1780: PETSC_EXTERN PetscLogEvent MAT_FDColoringFunction;
1781: PETSC_EXTERN PetscLogEvent MAT_CreateSubMat;
1782: PETSC_EXTERN PetscLogEvent MAT_MatSolve;
1783: PETSC_EXTERN PetscLogEvent MAT_MatTrSolve;
1784: PETSC_EXTERN PetscLogEvent MAT_MatMultSymbolic;
1785: PETSC_EXTERN PetscLogEvent MAT_MatMultNumeric;
1786: PETSC_EXTERN PetscLogEvent MAT_Getlocalmatcondensed;
1787: PETSC_EXTERN PetscLogEvent MAT_GetBrowsOfAcols;
1788: PETSC_EXTERN PetscLogEvent MAT_GetBrowsOfAocols;
1789: PETSC_EXTERN PetscLogEvent MAT_PtAPSymbolic;
1790: PETSC_EXTERN PetscLogEvent MAT_PtAPNumeric;
1791: PETSC_EXTERN PetscLogEvent MAT_Seqstompinum;
1792: PETSC_EXTERN PetscLogEvent MAT_Seqstompisym;
1793: PETSC_EXTERN PetscLogEvent MAT_Seqstompi;
1794: PETSC_EXTERN PetscLogEvent MAT_Getlocalmat;
1795: PETSC_EXTERN PetscLogEvent MAT_RARtSymbolic;
1796: PETSC_EXTERN PetscLogEvent MAT_RARtNumeric;
1797: PETSC_EXTERN PetscLogEvent MAT_MatTransposeMultSymbolic;
1798: PETSC_EXTERN PetscLogEvent MAT_MatTransposeMultNumeric;
1799: PETSC_EXTERN PetscLogEvent MAT_TransposeMatMultSymbolic;
1800: PETSC_EXTERN PetscLogEvent MAT_TransposeMatMultNumeric;
1801: PETSC_EXTERN PetscLogEvent MAT_MatMatMultSymbolic;
1802: PETSC_EXTERN PetscLogEvent MAT_MatMatMultNumeric;
1803: PETSC_EXTERN PetscLogEvent MAT_Getsymtransreduced;
1804: PETSC_EXTERN PetscLogEvent MAT_GetSeqNonzeroStructure;
1805: PETSC_EXTERN PetscLogEvent MATMFFD_Mult;
1806: PETSC_EXTERN PetscLogEvent MAT_GetMultiProcBlock;
1807: PETSC_EXTERN PetscLogEvent MAT_CUSPARSECopyToGPU;
1808: PETSC_EXTERN PetscLogEvent MAT_CUSPARSECopyFromGPU;
1809: PETSC_EXTERN PetscLogEvent MAT_CUSPARSEGenerateTranspose;
1810: PETSC_EXTERN PetscLogEvent MAT_CUSPARSESolveAnalysis;
1811: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSECopyToGPU;
1812: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSECopyFromGPU;
1813: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSEGenerateTranspose;
1814: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSESolveAnalysis;
1815: PETSC_EXTERN PetscLogEvent MAT_SetValuesBatch;
1816: PETSC_EXTERN PetscLogEvent MAT_CreateGraph;
1817: PETSC_EXTERN PetscLogEvent MAT_ViennaCLCopyToGPU;
1818: PETSC_EXTERN PetscLogEvent MAT_DenseCopyToGPU;
1819: PETSC_EXTERN PetscLogEvent MAT_DenseCopyFromGPU;
1820: PETSC_EXTERN PetscLogEvent MAT_Merge;
1821: PETSC_EXTERN PetscLogEvent MAT_Residual;
1822: PETSC_EXTERN PetscLogEvent MAT_SetRandom;
1823: PETSC_EXTERN PetscLogEvent MAT_FactorFactS;
1824: PETSC_EXTERN PetscLogEvent MAT_FactorInvS;
1825: PETSC_EXTERN PetscLogEvent MAT_PreallCOO;
1826: PETSC_EXTERN PetscLogEvent MAT_SetVCOO;
1827: PETSC_EXTERN PetscLogEvent MATCOLORING_Apply;
1828: PETSC_EXTERN PetscLogEvent MATCOLORING_Comm;
1829: PETSC_EXTERN PetscLogEvent MATCOLORING_Local;
1830: PETSC_EXTERN PetscLogEvent MATCOLORING_ISCreate;
1831: PETSC_EXTERN PetscLogEvent MATCOLORING_SetUp;
1832: PETSC_EXTERN PetscLogEvent MATCOLORING_Weights;
1833: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Build;
1834: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Compress;
1835: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Orthog;
1836: PETSC_EXTERN PetscLogEvent MAT_H2Opus_LR;
1837: PETSC_EXTERN PetscLogEvent MAT_CUDACopyToGPU;
1838: PETSC_EXTERN PetscLogEvent MAT_HIPCopyToGPU;
1840: #if PetscDefined(CLANG_STATIC_ANALYZER)
1841: #define MatGetDiagonalMarkers(SeqXXX, bs)
1842: #else
1843: /*
1844: Adds diagonal pointers to sparse matrix nonzero structure and determines if all diagonal entries are present
1846: Rechecks the matrix data structure automatically if the nonzero structure of the matrix changed since the last call
1848: Potential optimization: since the a->j[j] are sorted this could use bisection to find the diagonal
1850: Developer Note:
1851: Uses the C preprocessor as a template mechanism to produce MatGetDiagonal_Seq[SB]AIJ() to avoid duplicate code
1852: */
1853: #define MatGetDiagonalMarkers(SeqXXX, bs) \
1854: PetscErrorCode MatGetDiagonalMarkers_##SeqXXX(Mat A, const PetscInt *diag[], PetscBool *diagDense) \
1855: { \
1856: Mat_##SeqXXX *a = (Mat_##SeqXXX *)A->data; \
1857: \
1858: PetscFunctionBegin; \
1859: if (A->factortype != MAT_FACTOR_NONE) { \
1860: if (diagDense) *diagDense = PETSC_TRUE; \
1861: if (diag) *diag = a->diag; \
1862: PetscFunctionReturn(PETSC_SUCCESS); \
1863: } \
1864: PetscCheck(diag != NULL || diagDense != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "At least one of diag or diagDense must be requested"); \
1865: if (a->diagNonzeroState != A->nonzerostate || (diag != NULL && a->diag == NULL)) { \
1866: const PetscInt m = A->rmap->n / (bs); \
1867: \
1868: if (diag == NULL && a->diag == NULL) { \
1869: a->diagDense = PETSC_TRUE; \
1870: for (PetscInt i = 0; i < m; i++) { \
1871: PetscBool found = PETSC_FALSE; \
1872: \
1873: for (PetscInt j = a->i[i]; j < a->i[i + 1]; j++) { \
1874: if (a->j[j] == i) { \
1875: found = PETSC_TRUE; \
1876: break; \
1877: } \
1878: } \
1879: if (!found) { \
1880: a->diagDense = PETSC_FALSE; \
1881: *diagDense = a->diagDense; \
1882: a->diagNonzeroState = A->nonzerostate; \
1883: PetscFunctionReturn(PETSC_SUCCESS); \
1884: } \
1885: } \
1886: } else { \
1887: if (a->diag == NULL) PetscCall(PetscMalloc1(m, &a->diag)); \
1888: a->diagDense = PETSC_TRUE; \
1889: for (PetscInt i = 0; i < m; i++) { \
1890: PetscBool found = PETSC_FALSE; \
1891: \
1892: a->diag[i] = a->i[i + 1]; \
1893: for (PetscInt j = a->i[i]; j < a->i[i + 1]; j++) { \
1894: if (a->j[j] == i) { \
1895: a->diag[i] = j; \
1896: found = PETSC_TRUE; \
1897: break; \
1898: } \
1899: } \
1900: if (!found) a->diagDense = PETSC_FALSE; \
1901: } \
1902: } \
1903: a->diagNonzeroState = A->nonzerostate; \
1904: } \
1905: if (diag) *diag = a->diag; \
1906: if (diagDense) *diagDense = a->diagDense; \
1907: PetscFunctionReturn(PETSC_SUCCESS); \
1908: }
1909: #endif