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: #include <petscsys.h>
225: typedef struct _n_MatRootName *MatRootName;
226: struct _n_MatRootName {
227: char *rname, *sname, *mname;
228: MatRootName next;
229: };
231: PETSC_EXTERN MatRootName MatRootNameList;
233: /*
234: Utility private matrix routines used outside Mat
235: */
236: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatFindNonzeroRowsOrCols_Basic(Mat, PetscBool, PetscReal, IS *);
237: PETSC_EXTERN PetscErrorCode MatShellGetScalingShifts(Mat, PetscScalar *, PetscScalar *, Vec *, Vec *, Vec *, Mat *, IS *, IS *);
239: #define MAT_SHELL_NOT_ALLOWED (void *)-1
241: /*
242: Utility private matrix routines
243: */
244: PETSC_INTERN PetscErrorCode MatConvert_Basic(Mat, MatType, MatReuse, Mat *);
245: PETSC_INTERN PetscErrorCode MatConvert_Shell(Mat, MatType, MatReuse, Mat *);
246: PETSC_INTERN PetscErrorCode MatConvertFrom_Shell(Mat, MatType, MatReuse, Mat *);
247: PETSC_INTERN PetscErrorCode MatShellSetContext_Immutable(Mat, void *);
248: PETSC_INTERN PetscErrorCode MatShellSetContextDestroy_Immutable(Mat, PetscCtxDestroyFn *);
249: PETSC_INTERN PetscErrorCode MatShellSetManageScalingShifts_Immutable(Mat);
250: PETSC_INTERN PetscErrorCode MatCopy_Basic(Mat, Mat, MatStructure);
251: PETSC_INTERN PetscErrorCode MatDiagonalSet_Default(Mat, Vec, InsertMode);
252: #if PetscDefined(HAVE_SCALAPACK) && (PetscDefined(USE_REAL_SINGLE) || PetscDefined(USE_REAL_DOUBLE))
253: PETSC_INTERN PetscErrorCode MatConvert_Dense_ScaLAPACK(Mat, MatType, MatReuse, Mat *);
254: #endif
255: PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_Basic(Mat, PetscCount, PetscInt[], PetscInt[]);
256: PETSC_INTERN PetscErrorCode MatSetValuesCOO_Basic(Mat, const PetscScalar[], InsertMode);
258: /*
259: Index translation for the local submatrices of MATLOCALREF and MATIS, which both forward their local
260: insertions to another matrix. These need to be macros because they use sizeof
261: */
262: #define MatIndexSpaceGet_Private(buf, nrow, ncol, irowm, icolm) \
263: do { \
264: if ((nrow) + (ncol) > (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) { \
265: PetscCall(PetscMalloc2(nrow, &(irowm), ncol, &(icolm))); \
266: } else { \
267: irowm = &(buf)[0]; \
268: icolm = &(buf)[nrow]; \
269: } \
270: } while (0)
272: #define MatIndexSpaceRestore_Private(buf, nrow, ncol, irowm, icolm) \
273: do { \
274: if ((nrow) + (ncol) > (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) PetscCall(PetscFree2(irowm, icolm)); \
275: } while (0)
277: /* Turn n block indices into the n * bs scalar indices they stand for */
278: static inline void MatBlockIndicesExpand_Private(PetscInt n, const PetscInt idx[], PetscInt bs, PetscInt idxm[])
279: {
280: for (PetscInt i = 0; i < n; i++) {
281: for (PetscInt j = 0; j < bs; j++) idxm[i * bs + j] = idx[i] * bs + j;
282: }
283: }
285: /* Scattering of dense matrices with strided PetscSF */
286: PETSC_EXTERN PetscErrorCode MatDenseScatter_Private(PetscSF, Mat, Mat, InsertMode, ScatterMode);
288: /* This can be moved to the public header after implementing some missing MatProducts */
289: PETSC_INTERN PetscErrorCode MatCreateFromISLocalToGlobalMapping(ISLocalToGlobalMapping, Mat, PetscBool, PetscBool, MatType, Mat *);
291: /* these callbacks rely on the old matrix function pointers for
292: matmat operations. They are unsafe, and should be removed.
293: However, the amount of work needed to clean up all the
294: implementations is not negligible */
295: PETSC_INTERN PetscErrorCode MatProductSymbolic_AB(Mat);
296: PETSC_INTERN PetscErrorCode MatProductNumeric_AB(Mat);
297: PETSC_INTERN PetscErrorCode MatProductSymbolic_AtB(Mat);
298: PETSC_INTERN PetscErrorCode MatProductNumeric_AtB(Mat);
299: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABt(Mat);
300: PETSC_INTERN PetscErrorCode MatProductNumeric_ABt(Mat);
301: PETSC_INTERN PetscErrorCode MatProductNumeric_PtAP(Mat);
302: PETSC_INTERN PetscErrorCode MatProductNumeric_RARt(Mat);
303: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABC(Mat);
304: PETSC_INTERN PetscErrorCode MatProductNumeric_ABC(Mat);
306: PETSC_INTERN PetscErrorCode MatProductCreate_Private(Mat, Mat, Mat, Mat);
307: /* this callback handles all the different triple products and
308: does not rely on the function pointers; used by cuSPARSE/hipSPARSE and KOKKOS-KERNELS */
309: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABC_Basic(Mat);
311: /* CreateGraph is common to AIJ seq and mpi */
312: PETSC_INTERN PetscErrorCode MatCreateGraph_Simple_AIJ(Mat, PetscBool, PetscBool, PetscReal, PetscInt, PetscInt[], Mat *);
314: #if PetscDefined(CLANG_STATIC_ANALYZER)
315: template <typename Tm>
316: extern void MatCheckPreallocated(Tm, int);
317: template <typename Tm>
318: extern void MatCheckProduct(Tm, int);
319: #else /* PETSC_CLANG_STATIC_ANALYZER */
320: #define MatCheckPreallocated(A, arg) \
321: do { \
322: if (!(A)->preallocated) PetscCall(MatSetUp(A)); \
323: } while (0)
325: #if PetscDefined(USE_DEBUG)
326: #define MatCheckProduct(A, arg) \
327: do { \
328: PetscCheck((A)->product, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Argument %d \"%s\" is not a matrix obtained from MatProductCreate()", arg, #A); \
329: } while (0)
330: #else
331: #define MatCheckProduct(A, arg) \
332: do { \
333: } while (0)
334: #endif
335: #endif /* PETSC_CLANG_STATIC_ANALYZER */
337: /*
338: The stash is used to temporarily store inserted matrix values that
339: belong to another processor. During the assembly phase the stashed
340: values are moved to the correct processor and
341: */
343: typedef struct _MatStashSpace *PetscMatStashSpace;
345: struct _MatStashSpace {
346: PetscMatStashSpace next;
347: PetscScalar *space_head, *val;
348: PetscInt *idx, *idy;
349: PetscInt total_space_size;
350: PetscInt local_used;
351: PetscInt local_remaining;
352: };
354: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceGet(PetscInt, PetscInt, PetscMatStashSpace *);
355: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceContiguous(PetscInt, PetscMatStashSpace *, PetscScalar *, PetscInt *, PetscInt *);
356: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceDestroy(PetscMatStashSpace *);
358: typedef struct {
359: PetscInt count;
360: } MatStashHeader;
362: typedef struct {
363: void *buffer; /* Of type blocktype, dynamically constructed */
364: PetscInt count;
365: char pending;
366: } MatStashFrame;
368: typedef struct _MatStash MatStash;
369: struct _MatStash {
370: PetscInt nmax; /* maximum stash size */
371: PetscInt umax; /* user specified max-size */
372: PetscInt oldnmax; /* the nmax value used previously */
373: PetscInt n; /* stash size */
374: PetscInt bs; /* block size of the stash */
375: PetscInt reallocs; /* preserve the no of mallocs invoked */
376: PetscMatStashSpace space_head, space; /* linked list to hold stashed global row/column numbers and matrix values */
378: PetscErrorCode (*ScatterBegin)(Mat, MatStash *, PetscInt *);
379: PetscErrorCode (*ScatterGetMesg)(MatStash *, PetscMPIInt *, PetscInt **, PetscInt **, PetscScalar **, PetscInt *);
380: PetscErrorCode (*ScatterEnd)(MatStash *);
381: PetscErrorCode (*ScatterDestroy)(MatStash *);
383: /* The following variables are used for communication */
384: MPI_Comm comm;
385: PetscMPIInt size, rank;
386: PetscMPIInt tag1, tag2;
387: MPI_Request *send_waits; /* array of send requests */
388: MPI_Request *recv_waits; /* array of receive requests */
389: MPI_Status *send_status; /* array of send status */
390: PetscMPIInt nsends, nrecvs; /* numbers of sends and receives */
391: PetscScalar *svalues; /* sending data */
392: PetscInt *sindices;
393: PetscScalar **rvalues; /* receiving data (values) */
394: PetscInt **rindices; /* receiving data (indices) */
395: PetscMPIInt nprocessed; /* number of messages already processed */
396: PetscMPIInt *flg_v; /* indicates what messages have arrived so far and from whom */
397: PetscBool reproduce;
398: PetscMPIInt reproduce_count;
400: /* The following variables are used for BTS communication */
401: PetscBool first_assembly_done; /* Is the first time matrix assembly done? */
402: PetscBool use_status; /* Use MPI_Status to determine number of items in each message */
403: PetscMPIInt nsendranks;
404: PetscMPIInt nrecvranks;
405: PetscMPIInt *sendranks;
406: PetscMPIInt *recvranks;
407: MatStashHeader *sendhdr, *recvhdr;
408: MatStashFrame *sendframes; /* pointers to the main messages */
409: MatStashFrame *recvframes;
410: MatStashFrame *recvframe_active;
411: PetscInt recvframe_i; /* index of block within active frame */
412: PetscInt recvframe_count; /* Count actually sent for current frame */
413: PetscMPIInt recvcount; /* Number of receives processed so far */
414: PetscMPIInt *some_indices; /* From last call to MPI_Waitsome */
415: MPI_Status *some_statuses; /* Statuses from last call to MPI_Waitsome */
416: PetscMPIInt some_count; /* Number of requests completed in last call to MPI_Waitsome */
417: PetscMPIInt some_i; /* Index of request currently being processed */
418: MPI_Request *sendreqs;
419: MPI_Request *recvreqs;
420: PetscSegBuffer segsendblocks;
421: PetscSegBuffer segrecvframe;
422: PetscSegBuffer segrecvblocks;
423: MPI_Datatype blocktype;
424: size_t blocktype_size;
425: InsertMode *insertmode; /* Pointer to check mat->insertmode and set upon message arrival in case no local values have been set. */
426: };
428: #if !PetscDefined(HAVE_MPIUNI)
429: PETSC_INTERN PetscErrorCode MatStashScatterDestroy_BTS(MatStash *);
430: #endif
431: PETSC_INTERN PetscErrorCode MatStashCreate_Private(MPI_Comm, PetscInt, MatStash *);
432: PETSC_INTERN PetscErrorCode MatStashDestroy_Private(MatStash *);
433: PETSC_INTERN PetscErrorCode MatStashScatterEnd_Private(MatStash *);
434: PETSC_INTERN PetscErrorCode MatStashSetInitialSize_Private(MatStash *, PetscInt);
435: PETSC_INTERN PetscErrorCode MatStashGetInfo_Private(MatStash *, PetscInt *, PetscInt *);
436: PETSC_INTERN PetscErrorCode MatStashValuesRow_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscBool);
437: PETSC_INTERN PetscErrorCode MatStashValuesCol_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscBool);
438: PETSC_INTERN PetscErrorCode MatStashValuesRowBlocked_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscInt, PetscInt);
439: PETSC_INTERN PetscErrorCode MatStashValuesColBlocked_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscInt, PetscInt);
440: PETSC_INTERN PetscErrorCode MatStashScatterBegin_Private(Mat, MatStash *, PetscInt *);
441: PETSC_INTERN PetscErrorCode MatStashScatterGetMesg_Private(MatStash *, PetscMPIInt *, PetscInt **, PetscInt **, PetscScalar **, PetscInt *);
442: PETSC_INTERN PetscErrorCode MatGetInfo_External(Mat, MatInfoType, MatInfo *);
444: typedef struct {
445: PetscInt dim;
446: PetscInt dims[4];
447: PetscInt starts[4];
448: PetscBool noc; /* this is a single component problem, hence user will not set MatStencil.c */
449: } MatStencilInfo;
451: /* Info about using compressed row format */
452: typedef struct {
453: PetscBool use; /* indicates compressed rows have been checked and will be used */
454: PetscInt nrows; /* number of non-zero rows */
455: PetscInt *i; /* compressed row pointer */
456: PetscInt *rindex; /* compressed row index */
457: } Mat_CompressedRow;
458: PETSC_EXTERN PetscErrorCode MatCheckCompressedRow(Mat, PetscInt, Mat_CompressedRow *, PetscInt *, PetscInt, PetscReal);
460: typedef struct { /* used by MatCreateRedundantMatrix() for reusing matredundant */
461: PetscInt nzlocal, nsends, nrecvs;
462: PetscMPIInt *send_rank, *recv_rank;
463: PetscInt *sbuf_nz, *rbuf_nz, *sbuf_j, **rbuf_j;
464: PetscScalar *sbuf_a, **rbuf_a;
465: MPI_Comm subcomm; /* when user does not provide a subcomm */
466: IS isrow, iscol;
467: Mat *matseq;
468: } Mat_Redundant;
470: typedef struct { /* used by MatProduct() */
471: MatProductType type;
472: char *alg;
473: Mat A, B, C, Dwork;
474: 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, .. */
475: 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 .. */
476: PetscBool symbolic_used_the_fact_C_is_symmetric; /* MatMatMult(A,B,MAT_REUSE_MATRIX,..&C) is still legitimate), we need to redo symbolic! */
477: PetscObjectParameterDeclare(PetscReal, fill);
478: 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 */
479: PetscBool setfromoptionscalled;
481: /* Some products may display the information on the algorithm used */
482: PetscErrorCode (*view)(Mat, PetscViewer);
484: /* 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 */
485: PetscObjectState Anonzerostate;
487: /* many products have intermediate data structures, each specific to Mat types and product type */
488: PetscBool clear; /* whether or not to clear the data structures after MatProductNumeric has been called */
489: void *data; /* where to stash those structures */
490: PetscCtxDestroyFn *destroy; /* freeing data */
491: } Mat_Product;
493: struct _p_Mat {
494: PETSCHEADER(struct _MatOps);
495: PetscLayout rmap, cmap;
496: void *data; /* implementation-specific data */
497: MatFactorType factortype; /* MAT_FACTOR_LU, ILU, CHOLESKY or ICC */
498: PetscBool trivialsymbolic; /* indicates the symbolic factorization doesn't actually do a symbolic factorization, it is delayed to the numeric factorization */
499: PetscBool canuseordering; /* factorization can use ordering provide to routine (most PETSc implementations) */
500: MatOrderingType preferredordering[MAT_FACTOR_NUM_TYPES]; /* what is the preferred (or default) ordering for the matrix solver type */
501: PetscBool assembled; /* is the matrix assembled? */
502: PetscBool was_assembled; /* new values inserted into assembled mat */
503: PetscInt num_ass; /* number of times matrix has been assembled */
504: PetscObjectState nonzerostate; /* each time new nonzeros locations are introduced into the matrix this is updated */
505: PetscObjectState ass_nonzerostate; /* nonzero state at last assembly */
506: MatInfo info; /* matrix information */
507: InsertMode insertmode; /* have values been inserted in matrix or added? */
508: MatStash stash, bstash; /* used for assembling off-proc mat emements */
509: MatNullSpace nullsp; /* null space (operator is singular) */
510: MatNullSpace transnullsp; /* null space of transpose of operator */
511: MatNullSpace nearnullsp; /* near null space to be used by multigrid methods */
512: PetscInt congruentlayouts; /* are the rows and columns layouts congruent? */
513: PetscBool preallocated;
514: MatStencilInfo stencil; /* information for structured grid */
515: PetscBool3 symmetric, hermitian, structurally_symmetric, spd;
516: PetscBool symmetry_eternal, structural_symmetry_eternal, spd_eternal;
517: PetscBool nooffprocentries, nooffproczerorows;
518: PetscBool assembly_subset; /* set by MAT_SUBSET_OFF_PROC_ENTRIES */
519: PetscBool submat_singleis; /* for efficient PCSetUp_ASM() */
520: PetscBool structure_only;
521: PetscBool sortedfull; /* full, sorted rows are inserted */
522: PetscBool force_diagonals; /* set by MAT_FORCE_DIAGONAL_ENTRIES */
523: #if PetscDefined(HAVE_DEVICE)
524: PetscOffloadMask offloadmask; /* a mask which indicates where the valid matrix data is (GPU, CPU or both) */
525: PetscBool boundtocpu;
526: PetscBool bindingpropagates;
527: #endif
528: char *defaultrandtype;
529: void *spptr; /* pointer for special library like SuperLU */
530: char *solvertype;
531: PetscBool checksymmetryonassembly, checknullspaceonassembly;
532: PetscReal checksymmetrytol;
533: Mat schur; /* Schur complement matrix */
534: MatFactorSchurStatus schur_status; /* status of the Schur complement matrix */
535: Mat_Redundant *redundant; /* used by MatCreateRedundantMatrix() */
536: PetscBool erroriffailure; /* Generate an error if detected (for example a zero pivot) instead of returning */
537: MatFactorError factorerrortype; /* type of error in factorization */
538: PetscReal factorerror_zeropivot_value; /* If numerical zero pivot was detected this is the computed value */
539: PetscInt factorerror_zeropivot_row; /* Row where zero pivot was detected */
540: PetscInt nblocks, *bsizes; /* support for MatSetVariableBlockSizes() */
541: PetscInt p_cstart, p_rank, p_cend, n_rank; /* Information from parallel MatComputeVariableBlockEnvelope() */
542: PetscBool p_parallel;
543: char *defaultvectype;
544: Mat_Product *product;
545: PetscBool form_explicit_transpose; /* hint to generate an explicit mat tranpsose for operations like MatMultTranspose() */
546: PetscBool transupdated; /* whether or not the explicitly generated transpose is up-to-date */
547: char *factorprefix; /* the prefix to use with factored matrix that is created */
548: PetscBool hash_active; /* indicates MatSetValues() is being handled by hashing */
549: Vec dot_vec; /* work vector used by MatADot_Default() */
550: };
552: PETSC_INTERN PetscErrorCode MatAXPY_Basic(Mat, PetscScalar, Mat, MatStructure);
553: PETSC_INTERN PetscErrorCode MatAXPY_BasicWithPreallocation(Mat, Mat, PetscScalar, Mat, MatStructure);
554: PETSC_INTERN PetscErrorCode MatAXPY_Basic_Preallocate(Mat, Mat, Mat *);
555: PETSC_INTERN PetscErrorCode MatAXPY_Dense_Nest(Mat, PetscScalar, Mat);
557: /*
558: Utility for MatZeroRows
559: */
560: PETSC_INTERN PetscErrorCode MatZeroRowsMapLocal_Private(Mat, PetscInt, const PetscInt *, PetscInt *, PetscInt **);
562: /*
563: Utility for MatView/MatLoad
564: */
565: PETSC_INTERN PetscErrorCode MatView_Binary_BlockSizes(Mat, PetscViewer);
566: PETSC_INTERN PetscErrorCode MatLoad_Binary_BlockSizes(Mat, PetscViewer);
568: /*
569: Object for partitioning graphs
570: */
572: typedef struct _MatPartitioningOps *MatPartitioningOps;
573: struct _MatPartitioningOps {
574: PetscErrorCode (*apply)(MatPartitioning, IS *);
575: PetscErrorCode (*applynd)(MatPartitioning, IS *);
576: PetscErrorCode (*setfromoptions)(MatPartitioning, PetscOptionItems);
577: PetscErrorCode (*destroy)(MatPartitioning);
578: PetscErrorCode (*view)(MatPartitioning, PetscViewer);
579: PetscErrorCode (*improve)(MatPartitioning, IS *);
580: };
582: struct _p_MatPartitioning {
583: PETSCHEADER(struct _MatPartitioningOps);
584: Mat adj;
585: PetscInt *vertex_weights;
586: PetscReal *part_weights;
587: PetscInt n; /* number of partitions */
588: PetscInt ncon; /* number of vertex weights per vertex */
589: void *data;
590: PetscBool use_edge_weights; /* A flag indicates whether or not to use edge weights */
591: };
593: /* needed for parallel nested dissection by ParMETIS */
594: PETSC_INTERN PetscErrorCode MatPartitioningSizesToSep_Private(PetscInt, PetscInt[], PetscInt[], PetscInt[]);
596: /* shared by the MatPartitioning and PetscPartitioner bindings to PT-SCOTCH */
597: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode PetscPTSCOTCHPartitionGraph_Private(MPI_Comm, const PetscInt[], PetscInt[], PetscInt[], PetscInt[], PetscInt[], PetscBool, PetscInt, PetscInt[], PetscInt, double, PetscInt[]);
599: /*
600: Object for coarsen graphs
601: */
602: typedef struct _MatCoarsenOps *MatCoarsenOps;
603: struct _MatCoarsenOps {
604: PetscErrorCode (*apply)(MatCoarsen);
605: PetscErrorCode (*setfromoptions)(MatCoarsen, PetscOptionItems);
606: PetscErrorCode (*destroy)(MatCoarsen);
607: PetscErrorCode (*view)(MatCoarsen, PetscViewer);
608: };
610: #define MAT_COARSEN_STRENGTH_INDEX_SIZE 3
611: struct _p_MatCoarsen {
612: PETSCHEADER(struct _MatCoarsenOps);
613: Mat graph;
614: void *subctx;
615: /* */
616: PetscBool strict_aggs;
617: IS perm;
618: PetscCoarsenData *agg_lists;
619: PetscInt max_it; /* number of iterations in HEM */
620: PetscReal threshold; /* HEM can filter interim graphs */
621: PetscInt strength_index_size;
622: PetscInt strength_index[MAT_COARSEN_STRENGTH_INDEX_SIZE];
623: };
625: PETSC_EXTERN PetscErrorCode MatCoarsenMISKSetDistance(MatCoarsen, PetscInt);
626: PETSC_EXTERN PetscErrorCode MatCoarsenMISKGetDistance(MatCoarsen, PetscInt *);
628: /*
629: Used in aijdevice.h
630: */
631: typedef struct {
632: PetscInt *i;
633: PetscInt *j;
634: PetscScalar *a;
635: PetscInt n;
636: PetscInt ignorezeroentries;
637: } PetscCSRDataStructure;
639: /*
640: MatFDColoring is used to compute Jacobian matrices efficiently
641: via coloring. The data structure is explained below in an example.
643: Color = 0 1 0 2 | 2 3 0
644: ---------------------------------------------------
645: 00 01 | 05
646: 10 11 | 14 15 Processor 0
647: 22 23 | 25
648: 32 33 |
649: ===================================================
650: | 44 45 46
651: 50 | 55 Processor 1
652: | 64 66
653: ---------------------------------------------------
655: ncolors = 4;
657: ncolumns = {2,1,1,0}
658: columns = {{0,2},{1},{3},{}}
659: nrows = {4,2,3,3}
660: rows = {{0,1,2,3},{0,1},{1,2,3},{0,1,2}}
661: vwscale = {dx(0),dx(1),dx(2),dx(3)} MPI Vec
662: vscale = {dx(0),dx(1),dx(2),dx(3),dx(4),dx(5)} Seq Vec
664: ncolumns = {1,0,1,1}
665: columns = {{6},{},{4},{5}}
666: nrows = {3,0,2,2}
667: rows = {{0,1,2},{},{1,2},{1,2}}
668: vwscale = {dx(4),dx(5),dx(6)} MPI Vec
669: vscale = {dx(0),dx(4),dx(5),dx(6)} Seq Vec
671: See the routine MatFDColoringApply() for how this data is used
672: to compute the Jacobian.
674: */
675: typedef struct {
676: PetscInt row;
677: PetscInt col;
678: PetscScalar *valaddr; /* address of value */
679: } MatEntry;
681: typedef struct {
682: PetscInt row;
683: PetscScalar *valaddr; /* address of value */
684: } MatEntry2;
686: struct _p_MatFDColoring {
687: PETSCHEADER(int);
688: PetscInt M, N, m; /* total rows, columns; local rows */
689: PetscInt rstart; /* first row owned by local processor */
690: PetscInt ncolors; /* number of colors */
691: PetscInt *ncolumns; /* number of local columns for a color */
692: PetscInt **columns; /* lists the local columns of each color (using global column numbering) */
693: IS *isa; /* these are the IS that contain the column values given in columns */
694: PetscInt *nrows; /* number of local rows for each color */
695: MatEntry *matentry; /* holds (row, column, address of value) for Jacobian matrix entry */
696: MatEntry2 *matentry2; /* holds (row, address of value) for Jacobian matrix entry */
697: PetscScalar *dy; /* store a block of F(x+dx)-F(x) when J is in BAIJ format */
698: PetscReal error_rel; /* square root of relative error in computing function */
699: PetscReal umin; /* minimum allowable u'dx value */
700: Vec w1, w2, w3; /* work vectors used in computing Jacobian */
701: PetscBool fset; /* indicates that the initial function value F(X) is set */
702: MatFDColoringFn *f; /* function that defines Jacobian */
703: void *fctx; /* optional user-defined context for use by the function f */
704: Vec vscale; /* holds FD scaling, i.e. 1/dx for each perturbed column */
705: PetscInt currentcolor; /* color for which function evaluation is being done now */
706: const char *htype; /* "wp" or "ds" */
707: ISColoringType ctype; /* IS_COLORING_GLOBAL or IS_COLORING_LOCAL */
708: PetscInt brows, bcols; /* number of block rows or columns for speedup inserting the dense matrix into sparse Jacobian */
709: PetscBool setupcalled; /* true if setup has been called */
710: PetscBool viewed; /* true if the -mat_fd_coloring_view has been triggered already */
711: PetscFortranCallbackFn *ftn_func_pointer; /* serve the same purpose as *fortran_func_pointers in PETSc objects */
712: void *ftn_func_cntx;
713: PetscObjectId matid; /* matrix this object was created with, must always be the same */
714: };
716: typedef struct _MatColoringOps *MatColoringOps;
717: struct _MatColoringOps {
718: PetscErrorCode (*destroy)(MatColoring);
719: PetscErrorCode (*setfromoptions)(MatColoring, PetscOptionItems);
720: PetscErrorCode (*view)(MatColoring, PetscViewer);
721: PetscErrorCode (*apply)(MatColoring, ISColoring *);
722: PetscErrorCode (*weights)(MatColoring, PetscReal **, PetscInt **);
723: };
725: struct _p_MatColoring {
726: PETSCHEADER(struct _MatColoringOps);
727: Mat mat;
728: PetscInt dist; /* distance of the coloring */
729: PetscInt maxcolors; /* the maximum number of colors returned, maxcolors=1 for MIS */
730: void *data; /* inner context */
731: PetscBool valid; /* check to see if what is produced is a valid coloring */
732: MatColoringWeightType weight_type; /* type of weight computation to be performed */
733: PetscReal *user_weights; /* custom weights and permutation */
734: PetscInt *user_lperm;
735: PetscBool valid_iscoloring; /* check to see if matcoloring is produced a valid iscoloring */
736: };
738: struct _p_MatTransposeColoring {
739: PETSCHEADER(int);
740: PetscInt M, N, m; /* total rows, columns; local rows */
741: PetscInt rstart; /* first row owned by local processor */
742: PetscInt ncolors; /* number of colors */
743: PetscInt *ncolumns; /* number of local columns for a color */
744: PetscInt *nrows; /* number of local rows for each color */
745: PetscInt currentcolor; /* color for which function evaluation is being done now */
746: ISColoringType ctype; /* IS_COLORING_GLOBAL or IS_COLORING_LOCAL */
748: PetscInt *colorforrow, *colorforcol; /* pointer to rows and columns */
749: PetscInt *rows; /* lists the local rows for each color (using the local row numbering) */
750: PetscInt *den2sp; /* maps (row,color) in the dense matrix to index of sparse matrix array a->a */
751: PetscInt *columns; /* lists the local columns of each color (using global column numbering) */
752: PetscInt brows; /* number of rows for efficient implementation of MatTransColoringApplyDenToSp() */
753: PetscInt *lstart; /* array used for loop over row blocks of Csparse */
754: };
756: /*
757: Null space context for preconditioner/operators
758: */
759: struct _p_MatNullSpace {
760: PETSCHEADER(int);
761: PetscBool has_cnst;
762: PetscInt n;
763: Vec *vecs;
764: PetscScalar *alpha; /* for projections */
765: MatNullSpaceRemoveFn *remove; /* for user provided removal function */
766: void *rmctx; /* context for remove() function */
767: };
769: /*
770: Internal data structure for MATMPIDENSE
771: */
772: typedef struct {
773: Mat A; /* local submatrix */
775: /* The following variables are used for matrix assembly */
776: PetscBool donotstash; /* Flag indicating if values should be stashed */
777: MPI_Request *send_waits; /* array of send requests */
778: MPI_Request *recv_waits; /* array of receive requests */
779: PetscInt nsends, nrecvs; /* numbers of sends and receives */
780: PetscScalar *svalues, *rvalues; /* sending and receiving data */
781: PetscInt rmax; /* maximum message length */
783: /* The following variables are used for matrix-vector products */
784: Vec lvec; /* local vector */
785: PetscSF Mvctx; /* for mat-mult communications */
786: PetscBool roworiented; /* if true, row-oriented input (default) */
788: /* Support for MatDenseGetColumnVec and MatDenseGetSubMatrix */
789: Mat cmat; /* matrix representation of a given subset of columns */
790: Vec cvec; /* vector representation of a given column */
791: const PetscScalar *ptrinuse; /* holds array to be restored (just a placeholder) */
792: PetscInt vecinuse; /* if cvec is in use (col = vecinuse-1) */
793: PetscInt matinuse; /* if cmat is in use (cbegin = matinuse-1) */
794: /* if this is from MatDenseGetSubMatrix, which columns and rows does it correspond to? */
795: PetscInt sub_rbegin;
796: PetscInt sub_rend;
797: PetscInt sub_cbegin;
798: PetscInt sub_cend;
799: } Mat_MPIDense;
801: /*
802: Checking zero pivot for LU, ILU preconditioners.
803: */
804: typedef struct {
805: PetscInt nshift, nshift_max;
806: PetscReal shift_amount, shift_lo, shift_hi, shift_top, shift_fraction;
807: PetscBool newshift;
808: PetscReal rs; /* active row sum of abs(off-diagonals) */
809: PetscScalar pv; /* pivot of the active row */
810: } FactorShiftCtx;
812: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatTransposeCheckNonzeroState_Private(Mat, Mat);
814: PETSC_EXTERN PetscErrorCode MatFactorDumpMatrix(Mat);
815: PETSC_INTERN PetscErrorCode MatSetBlockSizes_Default(Mat, PetscInt, PetscInt);
817: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatShift_Basic(Mat, PetscScalar);
818: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatProductComputeWithMat_Private(Mat, Mat, MatProductType, Mat);
820: static inline PetscErrorCode MatPivotCheck_nz(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
821: {
822: PetscReal _rs = sctx->rs;
823: PetscReal _zero = info->zeropivot * _rs;
825: PetscFunctionBegin;
826: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
827: /* force |diag| > zeropivot*rs */
828: if (!sctx->nshift) sctx->shift_amount = info->shiftamount;
829: else sctx->shift_amount *= 2.0;
830: sctx->newshift = PETSC_TRUE;
831: sctx->nshift++;
832: } else {
833: sctx->newshift = PETSC_FALSE;
834: }
835: PetscFunctionReturn(PETSC_SUCCESS);
836: }
838: static inline PetscErrorCode MatPivotCheck_pd(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
839: {
840: PetscReal _rs = sctx->rs;
841: PetscReal _zero = info->zeropivot * _rs;
843: PetscFunctionBegin;
844: if (PetscRealPart(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
845: /* force matfactor to be diagonally dominant */
846: if (sctx->nshift == sctx->nshift_max) {
847: sctx->shift_fraction = sctx->shift_hi;
848: } else {
849: sctx->shift_lo = sctx->shift_fraction;
850: sctx->shift_fraction = (sctx->shift_hi + sctx->shift_lo) / (PetscReal)2.;
851: }
852: sctx->shift_amount = sctx->shift_fraction * sctx->shift_top;
853: sctx->nshift++;
854: sctx->newshift = PETSC_TRUE;
855: } else {
856: sctx->newshift = PETSC_FALSE;
857: }
858: PetscFunctionReturn(PETSC_SUCCESS);
859: }
861: static inline PetscErrorCode MatPivotCheck_inblocks(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
862: {
863: PetscReal _zero = info->zeropivot;
865: PetscFunctionBegin;
866: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
867: sctx->pv += info->shiftamount;
868: sctx->shift_amount = 0.0;
869: sctx->nshift++;
870: }
871: sctx->newshift = PETSC_FALSE;
872: PetscFunctionReturn(PETSC_SUCCESS);
873: }
875: static inline PetscErrorCode MatPivotCheck_none(Mat fact, Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PetscInt row)
876: {
877: PetscReal _zero = info->zeropivot;
879: PetscFunctionBegin;
880: sctx->newshift = PETSC_FALSE;
881: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
882: 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);
883: PetscCall(PetscInfo(mat, "Detected zero pivot in factorization in row %" PetscInt_FMT " value %g tolerance %g\n", row, (double)PetscAbsScalar(sctx->pv), (double)_zero));
884: fact->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
885: fact->factorerror_zeropivot_value = PetscAbsScalar(sctx->pv);
886: fact->factorerror_zeropivot_row = row;
887: }
888: PetscFunctionReturn(PETSC_SUCCESS);
889: }
891: static inline PetscErrorCode MatPivotCheck(Mat fact, Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PetscInt row)
892: {
893: PetscFunctionBegin;
894: if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) PetscCall(MatPivotCheck_nz(mat, info, sctx, row));
895: else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) PetscCall(MatPivotCheck_pd(mat, info, sctx, row));
896: else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) PetscCall(MatPivotCheck_inblocks(mat, info, sctx, row));
897: else PetscCall(MatPivotCheck_none(fact, mat, info, sctx, row));
898: PetscFunctionReturn(PETSC_SUCCESS);
899: }
901: #include <petscbt.h>
902: /*
903: Create and initialize a linked list
904: Input Parameters:
905: idx_start - starting index of the list
906: lnk_max - max value of lnk indicating the end of the list
907: nlnk - max length of the list
908: Output Parameters:
909: lnk - list initialized
910: bt - PetscBT (bitarray) with all bits set to false
911: lnk_empty - flg indicating the list is empty
912: */
913: #define PetscLLCreate(idx_start, lnk_max, nlnk, lnk, bt) ((PetscErrorCode)(PetscMalloc1(nlnk, &(lnk)) || PetscBTCreate(nlnk, &(bt)) || ((lnk)[idx_start] = lnk_max, PETSC_SUCCESS)))
915: #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)))
917: 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)
918: {
919: PetscInt location;
921: PetscFunctionBegin;
922: /* start from the beginning if entry < previous entry */
923: if (!assume_sorted && k && entry < *lnkdata) *lnkdata = idx_start;
924: /* search for insertion location */
925: do {
926: location = *lnkdata;
927: *lnkdata = lnk[location];
928: } while (entry > *lnkdata);
929: /* insertion location is found, add entry into lnk */
930: lnk[location] = entry;
931: lnk[entry] = *lnkdata;
932: ++(*nlnk);
933: *lnkdata = entry; /* next search starts from here if next_entry > entry */
934: PetscFunctionReturn(PETSC_SUCCESS);
935: }
937: 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)
938: {
939: PetscFunctionBegin;
940: *nlnk = 0;
941: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
942: const PetscInt entry = indices[k];
944: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, &lnkdata, lnk));
945: }
946: PetscFunctionReturn(PETSC_SUCCESS);
947: }
949: /*
950: Add an index set into a sorted linked list
951: Input Parameters:
952: nidx - number of input indices
953: indices - integer array
954: idx_start - starting index of the list
955: lnk - linked list(an integer array) that is created
956: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
957: output Parameters:
958: nlnk - number of newly added indices
959: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
960: bt - updated PetscBT (bitarray)
961: */
962: static inline PetscErrorCode PetscLLAdd(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt)
963: {
964: PetscFunctionBegin;
965: PetscCall(PetscLLAdd_Private(nidx, indices, idx_start, nlnk, lnk, bt, PETSC_FALSE));
966: PetscFunctionReturn(PETSC_SUCCESS);
967: }
969: /*
970: Add a SORTED ascending index set into a sorted linked list - same as PetscLLAdd() bus skip 'if (_k && _entry < _lnkdata) _lnkdata = idx_start;'
971: Input Parameters:
972: nidx - number of input indices
973: indices - sorted integer array
974: idx_start - starting index of the list
975: lnk - linked list(an integer array) that is created
976: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
977: output Parameters:
978: nlnk - number of newly added indices
979: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
980: bt - updated PetscBT (bitarray)
981: */
982: static inline PetscErrorCode PetscLLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt)
983: {
984: PetscFunctionBegin;
985: PetscCall(PetscLLAdd_Private(nidx, indices, idx_start, nlnk, lnk, bt, PETSC_TRUE));
986: PetscFunctionReturn(PETSC_SUCCESS);
987: }
989: /*
990: Add a permuted index set into a sorted linked list
991: Input Parameters:
992: nidx - number of input indices
993: indices - integer array
994: perm - permutation of indices
995: idx_start - starting index of the list
996: lnk - linked list(an integer array) that is created
997: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
998: output Parameters:
999: nlnk - number of newly added indices
1000: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
1001: bt - updated PetscBT (bitarray)
1002: */
1003: 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)
1004: {
1005: PetscFunctionBegin;
1006: *nlnk = 0;
1007: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1008: const PetscInt entry = perm[indices[k]];
1010: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_FALSE, k, idx_start, entry, nlnk, &lnkdata, lnk));
1011: }
1012: PetscFunctionReturn(PETSC_SUCCESS);
1013: }
1015: #if 0
1016: /* this appears to be unused? */
1017: static inline PetscErrorCode PetscLLAddSorted_new(PetscInt nidx, PetscInt *indices, PetscInt idx_start, PetscBool *lnk_empty, PetscInt *nlnk, PetscInt *lnk, PetscBT bt)
1018: {
1019: PetscInt lnkdata = idx_start;
1021: PetscFunctionBegin;
1022: if (*lnk_empty) {
1023: for (PetscInt k = 0; k < nidx; ++k) {
1024: const PetscInt entry = indices[k], location = lnkdata;
1026: PetscCall(PetscBTSet(bt,entry)); /* mark the new entry */
1027: lnkdata = lnk[location];
1028: /* insertion location is found, add entry into lnk */
1029: lnk[location] = entry;
1030: lnk[entry] = lnkdata;
1031: lnkdata = entry; /* next search starts from here */
1032: }
1033: /* lnk[indices[nidx-1]] = lnk[idx_start];
1034: lnk[idx_start] = indices[0];
1035: PetscCall(PetscBTSet(bt,indices[0]));
1036: for (_k=1; _k<nidx; _k++) {
1037: PetscCall(PetscBTSet(bt,indices[_k]));
1038: lnk[indices[_k-1]] = indices[_k];
1039: }
1040: */
1041: *nlnk = nidx;
1042: *lnk_empty = PETSC_FALSE;
1043: } else {
1044: *nlnk = 0;
1045: for (PetscInt k = 0; k < nidx; ++k) {
1046: const PetscInt entry = indices[k];
1048: if (!PetscBTLookupSet(bt,entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_TRUE,k,idx_start,entry,nlnk,&lnkdata,lnk));
1049: }
1050: }
1051: PetscFunctionReturn(PETSC_SUCCESS);
1052: }
1053: #endif
1055: /*
1056: Add a SORTED index set into a sorted linked list used for LUFactorSymbolic()
1057: Same as PetscLLAddSorted() with an additional operation:
1058: count the number of input indices that are no larger than 'diag'
1059: Input Parameters:
1060: indices - sorted integer array
1061: idx_start - starting index of the list, index of pivot row
1062: lnk - linked list(an integer array) that is created
1063: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1064: diag - index of the active row in LUFactorSymbolic
1065: nzbd - number of input indices with indices <= idx_start
1066: im - im[idx_start] is initialized as num of nonzero entries in row=idx_start
1067: output Parameters:
1068: nlnk - number of newly added indices
1069: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
1070: bt - updated PetscBT (bitarray)
1071: im - im[idx_start]: unchanged if diag is not an entry
1072: : num of entries with indices <= diag if diag is an entry
1073: */
1074: 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)
1075: {
1076: const PetscInt nidx = im[idx_start] - nzbd; /* num of entries with idx_start < index <= diag */
1078: PetscFunctionBegin;
1079: *nlnk = 0;
1080: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1081: const PetscInt entry = indices[k];
1083: ++nzbd;
1084: if (entry == diag) im[idx_start] = nzbd;
1085: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_TRUE, k, idx_start, entry, nlnk, &lnkdata, lnk));
1086: }
1087: PetscFunctionReturn(PETSC_SUCCESS);
1088: }
1090: /*
1091: Copy data on the list into an array, then initialize the list
1092: Input Parameters:
1093: idx_start - starting index of the list
1094: lnk_max - max value of lnk indicating the end of the list
1095: nlnk - number of data on the list to be copied
1096: lnk - linked list
1097: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1098: output Parameters:
1099: indices - array that contains the copied data
1100: lnk - linked list that is cleaned and initialize
1101: bt - PetscBT (bitarray) with all bits set to false
1102: */
1103: static inline PetscErrorCode PetscLLClean(PetscInt idx_start, PetscInt lnk_max, PetscInt nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT indices, PetscBT bt)
1104: {
1105: PetscFunctionBegin;
1106: for (PetscInt j = 0, idx = idx_start; j < nlnk; ++j) {
1107: idx = lnk[idx];
1108: indices[j] = idx;
1109: PetscCall(PetscBTClear(bt, idx));
1110: }
1111: lnk[idx_start] = lnk_max;
1112: PetscFunctionReturn(PETSC_SUCCESS);
1113: }
1115: /*
1116: Free memories used by the list
1117: */
1118: #define PetscLLDestroy(lnk, bt) ((PetscErrorCode)(PetscFree(lnk) || PetscBTDestroy(&(bt))))
1120: /* Routines below are used for incomplete matrix factorization */
1121: /*
1122: Create and initialize a linked list and its levels
1123: Input Parameters:
1124: idx_start - starting index of the list
1125: lnk_max - max value of lnk indicating the end of the list
1126: nlnk - max length of the list
1127: Output Parameters:
1128: lnk - list initialized
1129: lnk_lvl - array of size nlnk for storing levels of lnk
1130: bt - PetscBT (bitarray) with all bits set to false
1131: */
1132: #define PetscIncompleteLLCreate(idx_start, lnk_max, nlnk, lnk, lnk_lvl, bt) \
1133: ((PetscErrorCode)(PetscIntMultError(2, nlnk, NULL) || PetscMalloc1(2 * (nlnk), &(lnk)) || PetscBTCreate(nlnk, &(bt)) || ((lnk)[idx_start] = lnk_max, lnk_lvl = (lnk) + (nlnk), PETSC_SUCCESS)))
1135: 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)
1136: {
1137: PetscFunctionBegin;
1138: PetscCall(PetscLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, lnkdata, lnk));
1139: lnklvl[entry] = newval;
1140: PetscFunctionReturn(PETSC_SUCCESS);
1141: }
1143: /*
1144: Initialize a sorted linked list used for ILU and ICC
1145: Input Parameters:
1146: nidx - number of input idx
1147: idx - integer array used for storing column indices
1148: idx_start - starting index of the list
1149: perm - indices of an IS
1150: lnk - linked list(an integer array) that is created
1151: lnklvl - levels of lnk
1152: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1153: output Parameters:
1154: nlnk - number of newly added idx
1155: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1156: lnklvl - levels of lnk
1157: bt - updated PetscBT (bitarray)
1158: */
1159: 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)
1160: {
1161: PetscFunctionBegin;
1162: *nlnk = 0;
1163: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1164: const PetscInt entry = perm[idx[k]];
1166: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscIncompleteLLInsertLocation_Private(PETSC_FALSE, k, idx_start, entry, nlnk, &lnkdata, lnk, lnklvl, 0));
1167: }
1168: PetscFunctionReturn(PETSC_SUCCESS);
1169: }
1171: 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)
1172: {
1173: PetscFunctionBegin;
1174: *nlnk = 0;
1175: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1176: const PetscInt incrlev = idxlvl[k] + prow_offset + 1;
1178: if (incrlev <= level) {
1179: const PetscInt entry = idx[k];
1181: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscIncompleteLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, &lnkdata, lnk, lnklvl, incrlev));
1182: else if (lnklvl[entry] > incrlev) lnklvl[entry] = incrlev; /* existing entry */
1183: }
1184: }
1185: PetscFunctionReturn(PETSC_SUCCESS);
1186: }
1188: /*
1189: Add a SORTED index set into a sorted linked list for ICC
1190: Input Parameters:
1191: nidx - number of input indices
1192: idx - sorted integer array used for storing column indices
1193: level - level of fill, e.g., ICC(level)
1194: idxlvl - level of idx
1195: idx_start - starting index of the list
1196: lnk - linked list(an integer array) that is created
1197: lnklvl - levels of lnk
1198: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1199: idxlvl_prow - idxlvl[prow], where prow is the row number of the idx
1200: output Parameters:
1201: nlnk - number of newly added indices
1202: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1203: lnklvl - levels of lnk
1204: bt - updated PetscBT (bitarray)
1205: Note: the level of U(i,j) is set as lvl(i,j) = min{ lvl(i,j), lvl(prow,i)+lvl(prow,j)+1)
1206: where idx = non-zero columns of U(prow,prow+1:n-1), prow<i
1207: */
1208: 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)
1209: {
1210: PetscFunctionBegin;
1211: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, idxlvl_prow, PETSC_TRUE));
1212: PetscFunctionReturn(PETSC_SUCCESS);
1213: }
1215: /*
1216: Add a SORTED index set into a sorted linked list for ILU
1217: Input Parameters:
1218: nidx - number of input indices
1219: idx - sorted integer array used for storing column indices
1220: level - level of fill, e.g., ICC(level)
1221: idxlvl - level of idx
1222: idx_start - starting index of the list
1223: lnk - linked list(an integer array) that is created
1224: lnklvl - levels of lnk
1225: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1226: prow - the row number of idx
1227: output Parameters:
1228: nlnk - number of newly added idx
1229: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1230: lnklvl - levels of lnk
1231: bt - updated PetscBT (bitarray)
1233: Note: the level of factor(i,j) is set as lvl(i,j) = min{ lvl(i,j), lvl(i,prow)+lvl(prow,j)+1)
1234: where idx = non-zero columns of U(prow,prow+1:n-1), prow<i
1235: */
1236: 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)
1237: {
1238: PetscFunctionBegin;
1239: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, lnklvl[prow], PETSC_TRUE));
1240: PetscFunctionReturn(PETSC_SUCCESS);
1241: }
1243: /*
1244: Add a index set into a sorted linked list
1245: Input Parameters:
1246: nidx - number of input idx
1247: idx - integer array used for storing column indices
1248: level - level of fill, e.g., ICC(level)
1249: idxlvl - level of idx
1250: idx_start - starting index of the list
1251: lnk - linked list(an integer array) that is created
1252: lnklvl - levels of lnk
1253: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1254: output Parameters:
1255: nlnk - number of newly added idx
1256: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1257: lnklvl - levels of lnk
1258: bt - updated PetscBT (bitarray)
1259: */
1260: 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)
1261: {
1262: PetscFunctionBegin;
1263: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, 0, PETSC_FALSE));
1264: PetscFunctionReturn(PETSC_SUCCESS);
1265: }
1267: /*
1268: Add a SORTED index set into a sorted linked list
1269: Input Parameters:
1270: nidx - number of input indices
1271: idx - sorted integer array used for storing column indices
1272: level - level of fill, e.g., ICC(level)
1273: idxlvl - level of idx
1274: idx_start - starting index of the list
1275: lnk - linked list(an integer array) that is created
1276: lnklvl - levels of lnk
1277: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1278: output Parameters:
1279: nlnk - number of newly added idx
1280: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1281: lnklvl - levels of lnk
1282: bt - updated PetscBT (bitarray)
1283: */
1284: 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)
1285: {
1286: PetscFunctionBegin;
1287: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, 0, PETSC_TRUE));
1288: PetscFunctionReturn(PETSC_SUCCESS);
1289: }
1291: /*
1292: Copy data on the list into an array, then initialize the list
1293: Input Parameters:
1294: idx_start - starting index of the list
1295: lnk_max - max value of lnk indicating the end of the list
1296: nlnk - number of data on the list to be copied
1297: lnk - linked list
1298: lnklvl - level of lnk
1299: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1300: output Parameters:
1301: indices - array that contains the copied data
1302: lnk - linked list that is cleaned and initialize
1303: lnklvl - level of lnk that is reinitialized
1304: bt - PetscBT (bitarray) with all bits set to false
1305: */
1306: 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)
1307: {
1308: PetscFunctionBegin;
1309: for (PetscInt j = 0, idx = idx_start; j < nlnk; ++j) {
1310: idx = lnk[idx];
1311: indices[j] = idx;
1312: indiceslvl[j] = lnklvl[idx];
1313: lnklvl[idx] = -1;
1314: PetscCall(PetscBTClear(bt, idx));
1315: }
1316: lnk[idx_start] = lnk_max;
1317: PetscFunctionReturn(PETSC_SUCCESS);
1318: }
1320: /*
1321: Free memories used by the list
1322: */
1323: #define PetscIncompleteLLDestroy(lnk, bt) ((PetscErrorCode)(PetscFree(lnk) || PetscBTDestroy(&(bt))))
1325: #if !PetscDefined(CLANG_STATIC_ANALYZER)
1326: #define MatCheckSameLocalSize(A, ar1, B, ar2) \
1327: do { \
1328: PetscCheckSameComm(A, ar1, B, ar2); \
1329: 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, \
1330: (A)->rmap->n, (A)->cmap->n, ar2, (B)->rmap->n, (B)->cmap->n); \
1331: } while (0)
1332: #define MatCheckSameSize(A, ar1, B, ar2) \
1333: do { \
1334: 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, \
1335: (A)->rmap->N, (A)->cmap->N, ar2, (B)->rmap->N, (B)->cmap->N); \
1336: MatCheckSameLocalSize(A, ar1, B, ar2); \
1337: } while (0)
1338: #else
1339: template <typename Tm>
1340: extern void MatCheckSameLocalSize(Tm, int, Tm, int);
1341: template <typename Tm>
1342: extern void MatCheckSameSize(Tm, int, Tm, int);
1343: #endif
1345: #define VecCheckMatCompatible(M, x, ar1, b, ar2) \
1346: do { \
1347: 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, \
1348: (M)->cmap->N); \
1349: 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, \
1350: (M)->rmap->N); \
1351: } while (0)
1353: /*
1354: Create and initialize a condensed linked list -
1355: same as PetscLLCreate(), but uses a scalable array 'lnk' with size of max number of entries, not O(N).
1356: Barry suggested this approach (Dec. 6, 2011):
1357: I've thought of an alternative way of representing a linked list that is efficient but doesn't have the O(N) scaling issue
1358: (it may be faster than the O(N) even sequentially due to less crazy memory access).
1360: 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
1361: 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
1362: 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
1363: 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.
1364: 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
1365: to each other so memory access is much better than using the big array.
1367: Example:
1368: nlnk_max=5, lnk_max=36:
1369: Initial list: [0, 0 | 36, 2 | 0, 0 | 0, 0 | 0, 0 | 0, 0 | 0, 0]
1370: here, head_node has index 2 with value lnk[2]=lnk_max=36,
1371: 0-th entry is used to store the number of entries in the list,
1372: The initial lnk represents head -> tail(marked by 36) with number of entries = lnk[0]=0.
1374: Now adding a sorted set {2,4}, the list becomes
1375: [2, 0 | 36, 4 |2, 6 | 4, 2 | 0, 0 | 0, 0 | 0, 0 ]
1376: represents head -> 2 -> 4 -> tail with number of entries = lnk[0]=2.
1378: Then adding a sorted set {0,3,35}, the list
1379: [5, 0 | 36, 8 | 2, 10 | 4, 12 | 0, 4 | 3, 6 | 35, 2 ]
1380: represents head -> 0 -> 2 -> 3 -> 4 -> 35 -> tail with number of entries = lnk[0]=5.
1382: Input Parameters:
1383: nlnk_max - max length of the list
1384: lnk_max - max value of the entries
1385: Output Parameters:
1386: lnk - list created and initialized
1387: bt - PetscBT (bitarray) with all bits set to false. Note: bt has size lnk_max, not nln_max!
1388: */
1389: static inline PetscErrorCode PetscLLCondensedCreate(PetscInt nlnk_max, PetscInt lnk_max, PetscInt **lnk, PetscBT *bt)
1390: {
1391: PetscInt *llnk, lsize = 0;
1393: PetscFunctionBegin;
1394: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1395: PetscCall(PetscMalloc1(lsize, lnk));
1396: PetscCall(PetscBTCreate(lnk_max, bt));
1397: llnk = *lnk;
1398: llnk[0] = 0; /* number of entries on the list */
1399: llnk[2] = lnk_max; /* value in the head node */
1400: llnk[3] = 2; /* next for the head node */
1401: PetscFunctionReturn(PETSC_SUCCESS);
1402: }
1404: /*
1405: Add a SORTED ascending index set into a sorted linked list. See PetscLLCondensedCreate() for detailed description.
1406: Input Parameters:
1407: nidx - number of input indices
1408: indices - sorted integer array
1409: lnk - condensed linked list(an integer array) that is created
1410: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1411: output Parameters:
1412: lnk - the sorted(increasing order) linked list containing previous and newly added non-redundate indices
1413: bt - updated PetscBT (bitarray)
1414: */
1415: static inline PetscErrorCode PetscLLCondensedAddSorted(PetscInt nidx, const PetscInt indices[], PetscInt lnk[], PetscBT bt)
1416: {
1417: PetscInt location = 2; /* head */
1418: PetscInt nlnk = lnk[0]; /* num of entries on the input lnk */
1420: PetscFunctionBegin;
1421: for (PetscInt k = 0; k < nidx; k++) {
1422: const PetscInt entry = indices[k];
1423: if (!PetscBTLookupSet(bt, entry)) { /* new entry */
1424: PetscInt next, lnkdata;
1426: /* search for insertion location */
1427: do {
1428: next = location + 1; /* link from previous node to next node */
1429: location = lnk[next]; /* idx of next node */
1430: lnkdata = lnk[location]; /* value of next node */
1431: } while (entry > lnkdata);
1432: /* insertion location is found, add entry into lnk */
1433: const PetscInt newnode = 2 * (nlnk + 2); /* index for this new node */
1434: lnk[next] = newnode; /* connect previous node to the new node */
1435: lnk[newnode] = entry; /* set value of the new node */
1436: lnk[newnode + 1] = location; /* connect new node to next node */
1437: location = newnode; /* next search starts from the new node */
1438: nlnk++;
1439: }
1440: }
1441: lnk[0] = nlnk; /* number of entries in the list */
1442: PetscFunctionReturn(PETSC_SUCCESS);
1443: }
1445: static inline PetscErrorCode PetscLLCondensedClean(PetscInt lnk_max, PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt lnk[], PetscBT bt)
1446: {
1447: const PetscInt nlnk = lnk[0]; /* num of entries on the list */
1448: PetscInt next = lnk[3]; /* head node */
1450: PetscFunctionBegin;
1451: for (PetscInt k = 0; k < nlnk; k++) {
1452: indices[k] = lnk[next];
1453: next = lnk[next + 1];
1454: PetscCall(PetscBTClear(bt, indices[k]));
1455: }
1456: lnk[0] = 0; /* num of entries on the list */
1457: lnk[2] = lnk_max; /* initialize head node */
1458: lnk[3] = 2; /* head node */
1459: PetscFunctionReturn(PETSC_SUCCESS);
1460: }
1462: static inline PetscErrorCode PetscLLCondensedView(PetscInt *lnk)
1463: {
1464: PetscFunctionBegin;
1465: PetscCall(PetscPrintf(PETSC_COMM_SELF, "LLCondensed of size %" PetscInt_FMT ", (val, next)\n", lnk[0]));
1466: 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]));
1467: PetscFunctionReturn(PETSC_SUCCESS);
1468: }
1470: /*
1471: Free memories used by the list
1472: */
1473: static inline PetscErrorCode PetscLLCondensedDestroy(PetscInt *lnk, PetscBT bt)
1474: {
1475: PetscFunctionBegin;
1476: PetscCall(PetscFree(lnk));
1477: PetscCall(PetscBTDestroy(&bt));
1478: PetscFunctionReturn(PETSC_SUCCESS);
1479: }
1481: /*
1482: Same as PetscLLCondensedCreate(), but does not use non-scalable O(lnk_max) bitarray
1483: Input Parameters:
1484: nlnk_max - max length of the list
1485: Output Parameters:
1486: lnk - list created and initialized
1487: */
1488: static inline PetscErrorCode PetscLLCondensedCreate_Scalable(PetscInt nlnk_max, PetscInt **lnk)
1489: {
1490: PetscInt *llnk, lsize = 0;
1492: PetscFunctionBegin;
1493: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1494: PetscCall(PetscMalloc1(lsize, lnk));
1495: llnk = *lnk;
1496: llnk[0] = 0; /* number of entries on the list */
1497: llnk[2] = PETSC_INT_MAX; /* value in the head node */
1498: llnk[3] = 2; /* next for the head node */
1499: PetscFunctionReturn(PETSC_SUCCESS);
1500: }
1502: static inline PetscErrorCode PetscLLCondensedExpand_Scalable(PetscInt nlnk_max, PetscInt **lnk)
1503: {
1504: PetscInt lsize = 0;
1506: PetscFunctionBegin;
1507: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1508: PetscCall(PetscRealloc((size_t)lsize * sizeof(PetscInt), lnk));
1509: PetscFunctionReturn(PETSC_SUCCESS);
1510: }
1512: static inline PetscErrorCode PetscLLCondensedAddSorted_Scalable(PetscInt nidx, const PetscInt indices[], PetscInt lnk[])
1513: {
1514: PetscInt location = 2; /* head */
1515: PetscInt nlnk = lnk[0]; /* num of entries on the input lnk */
1517: for (PetscInt k = 0; k < nidx; k++) {
1518: const PetscInt entry = indices[k];
1519: PetscInt next, lnkdata;
1521: /* search for insertion location */
1522: do {
1523: next = location + 1; /* link from previous node to next node */
1524: location = lnk[next]; /* idx of next node */
1525: lnkdata = lnk[location]; /* value of next node */
1526: } while (entry > lnkdata);
1527: if (entry < lnkdata) {
1528: /* insertion location is found, add entry into lnk */
1529: const PetscInt newnode = 2 * (nlnk + 2); /* index for this new node */
1530: lnk[next] = newnode; /* connect previous node to the new node */
1531: lnk[newnode] = entry; /* set value of the new node */
1532: lnk[newnode + 1] = location; /* connect new node to next node */
1533: location = newnode; /* next search starts from the new node */
1534: nlnk++;
1535: }
1536: }
1537: lnk[0] = nlnk; /* number of entries in the list */
1538: return PETSC_SUCCESS;
1539: }
1541: static inline PetscErrorCode PetscLLCondensedClean_Scalable(PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt *lnk)
1542: {
1543: const PetscInt nlnk = lnk[0];
1544: PetscInt next = lnk[3]; /* head node */
1546: for (PetscInt k = 0; k < nlnk; k++) {
1547: indices[k] = lnk[next];
1548: next = lnk[next + 1];
1549: }
1550: lnk[0] = 0; /* num of entries on the list */
1551: lnk[3] = 2; /* head node */
1552: return PETSC_SUCCESS;
1553: }
1555: static inline PetscErrorCode PetscLLCondensedDestroy_Scalable(PetscInt *lnk)
1556: {
1557: return PetscFree(lnk);
1558: }
1560: /*
1561: lnk[0] number of links
1562: lnk[1] number of entries
1563: lnk[3n] value
1564: lnk[3n+1] len
1565: lnk[3n+2] link to next value
1567: The next three are always the first link
1569: lnk[3] PETSC_INT_MIN+1
1570: lnk[4] 1
1571: lnk[5] link to first real entry
1573: The next three are always the last link
1575: lnk[6] PETSC_INT_MAX - 1
1576: lnk[7] 1
1577: lnk[8] next valid link (this is the same as lnk[0] but without the decreases)
1578: */
1580: static inline PetscErrorCode PetscLLCondensedCreate_fast(PetscInt nlnk_max, PetscInt **lnk)
1581: {
1582: PetscInt *llnk;
1583: PetscInt lsize = 0;
1585: PetscFunctionBegin;
1586: PetscCall(PetscIntMultError(3, nlnk_max + 3, &lsize));
1587: PetscCall(PetscMalloc1(lsize, lnk));
1588: llnk = *lnk;
1589: llnk[0] = 0; /* nlnk: number of entries on the list */
1590: llnk[1] = 0; /* number of integer entries represented in list */
1591: llnk[3] = PETSC_INT_MIN + 1; /* value in the first node */
1592: llnk[4] = 1; /* count for the first node */
1593: llnk[5] = 6; /* next for the first node */
1594: llnk[6] = PETSC_INT_MAX - 1; /* value in the last node */
1595: llnk[7] = 1; /* count for the last node */
1596: llnk[8] = 0; /* next valid node to be used */
1597: PetscFunctionReturn(PETSC_SUCCESS);
1598: }
1600: static inline PetscErrorCode PetscLLCondensedAddSorted_fast(PetscInt nidx, const PetscInt indices[], PetscInt lnk[])
1601: {
1602: for (PetscInt k = 0, prev = 3 /* first value */; k < nidx; k++) {
1603: const PetscInt entry = indices[k];
1604: PetscInt next = lnk[prev + 2];
1606: /* search for insertion location */
1607: while (entry >= lnk[next]) {
1608: prev = next;
1609: next = lnk[next + 2];
1610: }
1611: /* entry is in range of previous list */
1612: if (entry < lnk[prev] + lnk[prev + 1]) continue;
1613: lnk[1]++;
1614: /* entry is right after previous list */
1615: if (entry == lnk[prev] + lnk[prev + 1]) {
1616: lnk[prev + 1]++;
1617: if (lnk[next] == entry + 1) { /* combine two contiguous strings */
1618: lnk[prev + 1] += lnk[next + 1];
1619: lnk[prev + 2] = lnk[next + 2];
1620: lnk[0]--;
1621: }
1622: continue;
1623: }
1624: /* entry is right before next list */
1625: if (entry == lnk[next] - 1) {
1626: lnk[next]--;
1627: lnk[next + 1]++;
1628: prev = next;
1629: continue;
1630: }
1631: /* add entry into lnk */
1632: lnk[prev + 2] = 3 * ((lnk[8]++) + 3); /* connect previous node to the new node */
1633: prev = lnk[prev + 2];
1634: lnk[prev] = entry; /* set value of the new node */
1635: lnk[prev + 1] = 1; /* number of values in contiguous string is one to start */
1636: lnk[prev + 2] = next; /* connect new node to next node */
1637: lnk[0]++;
1638: }
1639: return PETSC_SUCCESS;
1640: }
1642: static inline PetscErrorCode PetscLLCondensedClean_fast(PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt *lnk)
1643: {
1644: const PetscInt nlnk = lnk[0];
1645: PetscInt next = lnk[5]; /* first node */
1647: for (PetscInt k = 0, cnt = 0; k < nlnk; k++) {
1648: for (PetscInt j = 0; j < lnk[next + 1]; j++) indices[cnt++] = lnk[next] + j;
1649: next = lnk[next + 2];
1650: }
1651: lnk[0] = 0; /* nlnk: number of links */
1652: lnk[1] = 0; /* number of integer entries represented in list */
1653: lnk[3] = PETSC_INT_MIN + 1; /* value in the first node */
1654: lnk[4] = 1; /* count for the first node */
1655: lnk[5] = 6; /* next for the first node */
1656: lnk[6] = PETSC_INT_MAX - 1; /* value in the last node */
1657: lnk[7] = 1; /* count for the last node */
1658: lnk[8] = 0; /* next valid location to make link */
1659: return PETSC_SUCCESS;
1660: }
1662: static inline PetscErrorCode PetscLLCondensedView_fast(const PetscInt *lnk)
1663: {
1664: const PetscInt nlnk = lnk[0];
1665: PetscInt next = lnk[5]; /* first node */
1667: for (PetscInt k = 0; k < nlnk; k++) {
1668: #if 0 /* Debugging code */
1669: printf("%d value %d len %d next %d\n", next, lnk[next], lnk[next + 1], lnk[next + 2]);
1670: #endif
1671: next = lnk[next + 2];
1672: }
1673: return PETSC_SUCCESS;
1674: }
1676: static inline PetscErrorCode PetscLLCondensedDestroy_fast(PetscInt *lnk)
1677: {
1678: return PetscFree(lnk);
1679: }
1681: PETSC_EXTERN PetscErrorCode PetscCDCreate(PetscInt, PetscCoarsenData **);
1682: PETSC_EXTERN PetscErrorCode PetscCDDestroy(PetscCoarsenData *);
1683: PETSC_EXTERN PetscErrorCode PetscCDIntNdSetID(PetscCDIntNd *, PetscInt);
1684: PETSC_EXTERN PetscErrorCode PetscCDIntNdGetID(const PetscCDIntNd *, PetscInt *);
1685: PETSC_EXTERN PetscErrorCode PetscCDAppendID(PetscCoarsenData *, PetscInt, PetscInt);
1686: PETSC_EXTERN PetscErrorCode PetscCDMoveAppend(PetscCoarsenData *, PetscInt, PetscInt);
1687: PETSC_EXTERN PetscErrorCode PetscCDAppendNode(PetscCoarsenData *, PetscInt, PetscCDIntNd *);
1688: PETSC_EXTERN PetscErrorCode PetscCDRemoveNextNode(PetscCoarsenData *, PetscInt, PetscCDIntNd *);
1689: PETSC_EXTERN PetscErrorCode PetscCDCountAt(const PetscCoarsenData *, PetscInt, PetscInt *);
1690: PETSC_EXTERN PetscErrorCode PetscCDIsEmptyAt(const PetscCoarsenData *, PetscInt, PetscBool *);
1691: PETSC_EXTERN PetscErrorCode PetscCDSetChunkSize(PetscCoarsenData *, PetscInt);
1692: PETSC_EXTERN PetscErrorCode PetscCDPrint(const PetscCoarsenData *, PetscInt, MPI_Comm);
1693: PETSC_EXTERN PetscErrorCode PetscCDGetNonemptyIS(PetscCoarsenData *, IS *);
1694: PETSC_EXTERN PetscErrorCode PetscCDGetMat(PetscCoarsenData *, Mat *);
1695: PETSC_EXTERN PetscErrorCode PetscCDSetMat(PetscCoarsenData *, Mat);
1696: PETSC_EXTERN PetscErrorCode PetscCDClearMat(PetscCoarsenData *);
1697: PETSC_EXTERN PetscErrorCode PetscCDRemoveAllAt(PetscCoarsenData *, PetscInt);
1698: PETSC_EXTERN PetscErrorCode PetscCDCount(const PetscCoarsenData *, PetscInt *_sz);
1700: PETSC_EXTERN PetscErrorCode PetscCDGetHeadPos(const PetscCoarsenData *, PetscInt, PetscCDIntNd **);
1701: PETSC_EXTERN PetscErrorCode PetscCDGetNextPos(const PetscCoarsenData *, PetscInt, PetscCDIntNd **);
1702: PETSC_EXTERN PetscErrorCode PetscCDGetASMBlocks(const PetscCoarsenData *, const PetscInt, PetscInt *, IS **);
1704: PETSC_SINGLE_LIBRARY_VISIBILITY_INTERNAL PetscErrorCode MatFDColoringApply_AIJ(Mat, MatFDColoring, Vec, void *);
1706: typedef struct {
1707: Vec diag;
1708: PetscBool diag_valid;
1709: Vec inv_diag;
1710: PetscBool inv_diag_valid;
1711: PetscObjectState diag_state, inv_diag_state;
1712: PetscInt *col;
1713: PetscScalar *val;
1714: } Mat_Diagonal;
1716: #if PetscDefined(HAVE_CUDA)
1717: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqCUDA(Mat, Vec, Vec, PetscScalar *);
1718: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqCUDA(Mat, Vec, PetscReal *);
1719: #endif
1720: #if PetscDefined(HAVE_HIP)
1721: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqHIP(Mat, Vec, Vec, PetscScalar *);
1722: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqHIP(Mat, Vec, PetscReal *);
1723: #endif
1724: #if PetscDefined(HAVE_KOKKOS_KERNELS)
1725: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqKokkos(Mat, Vec, Vec, PetscScalar *);
1726: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqKokkos(Mat, Vec, PetscReal *);
1727: #endif
1729: PETSC_EXTERN PetscLogEvent MAT_Mult;
1730: PETSC_EXTERN PetscLogEvent MAT_MultAdd;
1731: PETSC_EXTERN PetscLogEvent MAT_MultTranspose;
1732: PETSC_EXTERN PetscLogEvent MAT_MultHermitianTranspose;
1733: PETSC_EXTERN PetscLogEvent MAT_MultTransposeAdd;
1734: PETSC_EXTERN PetscLogEvent MAT_MultHermitianTransposeAdd;
1735: PETSC_EXTERN PetscLogEvent MAT_ADot;
1736: PETSC_EXTERN PetscLogEvent MAT_ANorm;
1737: PETSC_EXTERN PetscLogEvent MAT_Solve;
1738: PETSC_EXTERN PetscLogEvent MAT_Solves;
1739: PETSC_EXTERN PetscLogEvent MAT_SolveAdd;
1740: PETSC_EXTERN PetscLogEvent MAT_SolveTranspose;
1741: PETSC_EXTERN PetscLogEvent MAT_SolveTransposeAdd;
1742: PETSC_EXTERN PetscLogEvent MAT_SOR;
1743: PETSC_EXTERN PetscLogEvent MAT_ForwardSolve;
1744: PETSC_EXTERN PetscLogEvent MAT_BackwardSolve;
1745: PETSC_EXTERN PetscLogEvent MAT_LUFactor;
1746: PETSC_EXTERN PetscLogEvent MAT_LUFactorSymbolic;
1747: PETSC_EXTERN PetscLogEvent MAT_LUFactorNumeric;
1748: PETSC_EXTERN PetscLogEvent MAT_QRFactor;
1749: PETSC_EXTERN PetscLogEvent MAT_QRFactorSymbolic;
1750: PETSC_EXTERN PetscLogEvent MAT_QRFactorNumeric;
1751: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactor;
1752: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactorSymbolic;
1753: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactorNumeric;
1754: PETSC_EXTERN PetscLogEvent MAT_ILUFactor;
1755: PETSC_EXTERN PetscLogEvent MAT_ILUFactorSymbolic;
1756: PETSC_EXTERN PetscLogEvent MAT_ICCFactorSymbolic;
1757: PETSC_EXTERN PetscLogEvent MAT_Copy;
1758: PETSC_EXTERN PetscLogEvent MAT_Convert;
1759: PETSC_EXTERN PetscLogEvent MAT_Scale;
1760: PETSC_EXTERN PetscLogEvent MAT_AssemblyBegin;
1761: PETSC_EXTERN PetscLogEvent MAT_AssemblyEnd;
1762: PETSC_EXTERN PetscLogEvent MAT_SetValues;
1763: PETSC_EXTERN PetscLogEvent MAT_GetValues;
1764: PETSC_EXTERN PetscLogEvent MAT_GetRow;
1765: PETSC_EXTERN PetscLogEvent MAT_GetRowIJ;
1766: PETSC_EXTERN PetscLogEvent MAT_CreateSubMats;
1767: PETSC_EXTERN PetscLogEvent MAT_GetOrdering;
1768: PETSC_EXTERN PetscLogEvent MAT_RedundantMat;
1769: PETSC_EXTERN PetscLogEvent MAT_IncreaseOverlap;
1770: PETSC_EXTERN PetscLogEvent MAT_Partitioning;
1771: PETSC_EXTERN PetscLogEvent MAT_PartitioningND;
1772: PETSC_EXTERN PetscLogEvent MAT_Coarsen;
1773: PETSC_EXTERN PetscLogEvent MAT_ZeroEntries;
1774: PETSC_EXTERN PetscLogEvent MAT_Load;
1775: PETSC_EXTERN PetscLogEvent MAT_View;
1776: PETSC_EXTERN PetscLogEvent MAT_AXPY;
1777: PETSC_EXTERN PetscLogEvent MAT_FDColoringCreate;
1778: PETSC_EXTERN PetscLogEvent MAT_TransposeColoringCreate;
1779: PETSC_EXTERN PetscLogEvent MAT_FDColoringSetUp;
1780: PETSC_EXTERN PetscLogEvent MAT_FDColoringApply;
1781: PETSC_EXTERN PetscLogEvent MAT_Transpose;
1782: PETSC_EXTERN PetscLogEvent MAT_FDColoringFunction;
1783: PETSC_EXTERN PetscLogEvent MAT_CreateSubMat;
1784: PETSC_EXTERN PetscLogEvent MAT_MatSolve;
1785: PETSC_EXTERN PetscLogEvent MAT_MatTrSolve;
1786: PETSC_EXTERN PetscLogEvent MAT_MatMultSymbolic;
1787: PETSC_EXTERN PetscLogEvent MAT_MatMultNumeric;
1788: PETSC_EXTERN PetscLogEvent MAT_Getlocalmatcondensed;
1789: PETSC_EXTERN PetscLogEvent MAT_GetBrowsOfAcols;
1790: PETSC_EXTERN PetscLogEvent MAT_GetBrowsOfAocols;
1791: PETSC_EXTERN PetscLogEvent MAT_PtAPSymbolic;
1792: PETSC_EXTERN PetscLogEvent MAT_PtAPNumeric;
1793: PETSC_EXTERN PetscLogEvent MAT_Seqstompinum;
1794: PETSC_EXTERN PetscLogEvent MAT_Seqstompisym;
1795: PETSC_EXTERN PetscLogEvent MAT_Seqstompi;
1796: PETSC_EXTERN PetscLogEvent MAT_Getlocalmat;
1797: PETSC_EXTERN PetscLogEvent MAT_RARtSymbolic;
1798: PETSC_EXTERN PetscLogEvent MAT_RARtNumeric;
1799: PETSC_EXTERN PetscLogEvent MAT_MatTransposeMultSymbolic;
1800: PETSC_EXTERN PetscLogEvent MAT_MatTransposeMultNumeric;
1801: PETSC_EXTERN PetscLogEvent MAT_TransposeMatMultSymbolic;
1802: PETSC_EXTERN PetscLogEvent MAT_TransposeMatMultNumeric;
1803: PETSC_EXTERN PetscLogEvent MAT_MatMatMultSymbolic;
1804: PETSC_EXTERN PetscLogEvent MAT_MatMatMultNumeric;
1805: PETSC_EXTERN PetscLogEvent MAT_Getsymtransreduced;
1806: PETSC_EXTERN PetscLogEvent MAT_GetSeqNonzeroStructure;
1807: PETSC_EXTERN PetscLogEvent MATMFFD_Mult;
1808: PETSC_EXTERN PetscLogEvent MAT_GetMultiProcBlock;
1809: PETSC_EXTERN PetscLogEvent MAT_CUSPARSECopyToGPU;
1810: PETSC_EXTERN PetscLogEvent MAT_CUSPARSECopyFromGPU;
1811: PETSC_EXTERN PetscLogEvent MAT_CUSPARSEGenerateTranspose;
1812: PETSC_EXTERN PetscLogEvent MAT_CUSPARSESolveAnalysis;
1813: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSECopyToGPU;
1814: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSECopyFromGPU;
1815: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSEGenerateTranspose;
1816: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSESolveAnalysis;
1817: PETSC_EXTERN PetscLogEvent MAT_SetValuesBatch;
1818: PETSC_EXTERN PetscLogEvent MAT_CreateGraph;
1819: PETSC_EXTERN PetscLogEvent MAT_ViennaCLCopyToGPU;
1820: PETSC_EXTERN PetscLogEvent MAT_DenseCopyToGPU;
1821: PETSC_EXTERN PetscLogEvent MAT_DenseCopyFromGPU;
1822: PETSC_EXTERN PetscLogEvent MAT_Merge;
1823: PETSC_EXTERN PetscLogEvent MAT_Residual;
1824: PETSC_EXTERN PetscLogEvent MAT_SetRandom;
1825: PETSC_EXTERN PetscLogEvent MAT_FactorFactS;
1826: PETSC_EXTERN PetscLogEvent MAT_FactorInvS;
1827: PETSC_EXTERN PetscLogEvent MAT_PreallCOO;
1828: PETSC_EXTERN PetscLogEvent MAT_SetVCOO;
1829: PETSC_EXTERN PetscLogEvent MATCOLORING_Apply;
1830: PETSC_EXTERN PetscLogEvent MATCOLORING_Comm;
1831: PETSC_EXTERN PetscLogEvent MATCOLORING_Local;
1832: PETSC_EXTERN PetscLogEvent MATCOLORING_ISCreate;
1833: PETSC_EXTERN PetscLogEvent MATCOLORING_SetUp;
1834: PETSC_EXTERN PetscLogEvent MATCOLORING_Weights;
1835: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Build;
1836: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Compress;
1837: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Orthog;
1838: PETSC_EXTERN PetscLogEvent MAT_H2Opus_LR;
1839: PETSC_EXTERN PetscLogEvent MAT_CUDACopyToGPU;
1840: PETSC_EXTERN PetscLogEvent MAT_HIPCopyToGPU;
1842: #if PetscDefined(CLANG_STATIC_ANALYZER)
1843: #define MatGetDiagonalMarkers(SeqXXX, bs)
1844: #else
1845: /*
1846: Adds diagonal pointers to sparse matrix nonzero structure and determines if all diagonal entries are present
1848: Rechecks the matrix data structure automatically if the nonzero structure of the matrix changed since the last call
1850: Potential optimization: since the a->j[j] are sorted this could use bisection to find the diagonal
1852: Developer Note:
1853: Uses the C preprocessor as a template mechanism to produce MatGetDiagonal_Seq[SB]AIJ() to avoid duplicate code
1854: */
1855: #define MatGetDiagonalMarkers(SeqXXX, bs) \
1856: PetscErrorCode MatGetDiagonalMarkers_##SeqXXX(Mat A, const PetscInt *diag[], PetscBool *diagDense) \
1857: { \
1858: Mat_##SeqXXX *a = (Mat_##SeqXXX *)A->data; \
1859: \
1860: PetscFunctionBegin; \
1861: if (A->factortype != MAT_FACTOR_NONE) { \
1862: if (diagDense) *diagDense = PETSC_TRUE; \
1863: if (diag) *diag = a->diag; \
1864: PetscFunctionReturn(PETSC_SUCCESS); \
1865: } \
1866: PetscCheck(diag != NULL || diagDense != NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "At least one of diag or diagDense must be requested"); \
1867: if (a->diagNonzeroState != A->nonzerostate || (diag != NULL && a->diag == NULL)) { \
1868: const PetscInt m = A->rmap->n / (bs); \
1869: \
1870: if (diag == NULL && a->diag == NULL) { \
1871: a->diagDense = PETSC_TRUE; \
1872: for (PetscInt i = 0; i < m; i++) { \
1873: PetscBool found = PETSC_FALSE; \
1874: \
1875: for (PetscInt j = a->i[i]; j < a->i[i + 1]; j++) { \
1876: if (a->j[j] == i) { \
1877: found = PETSC_TRUE; \
1878: break; \
1879: } \
1880: } \
1881: if (!found) { \
1882: a->diagDense = PETSC_FALSE; \
1883: *diagDense = a->diagDense; \
1884: a->diagNonzeroState = A->nonzerostate; \
1885: PetscFunctionReturn(PETSC_SUCCESS); \
1886: } \
1887: } \
1888: } else { \
1889: if (a->diag == NULL) PetscCall(PetscMalloc1(m, &a->diag)); \
1890: a->diagDense = PETSC_TRUE; \
1891: for (PetscInt i = 0; i < m; i++) { \
1892: PetscBool found = PETSC_FALSE; \
1893: \
1894: a->diag[i] = a->i[i + 1]; \
1895: for (PetscInt j = a->i[i]; j < a->i[i + 1]; j++) { \
1896: if (a->j[j] == i) { \
1897: a->diag[i] = j; \
1898: found = PETSC_TRUE; \
1899: break; \
1900: } \
1901: } \
1902: if (!found) a->diagDense = PETSC_FALSE; \
1903: } \
1904: } \
1905: a->diagNonzeroState = A->nonzerostate; \
1906: } \
1907: if (diag) *diag = a->diag; \
1908: if (diagDense) *diagDense = a->diagDense; \
1909: PetscFunctionReturn(PETSC_SUCCESS); \
1910: }
1911: #endif