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 _p_MatRootName *MatRootName;
226: struct _p_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 defined(PETSC_HAVE_SCALAPACK) && (defined(PETSC_USE_REAL_SINGLE) || defined(PETSC_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: /* Scattering of dense matrices with strided PetscSF */
259: PETSC_EXTERN PetscErrorCode MatDenseScatter_Private(PetscSF, Mat, Mat, InsertMode, ScatterMode);
261: /* This can be moved to the public header after implementing some missing MatProducts */
262: PETSC_INTERN PetscErrorCode MatCreateFromISLocalToGlobalMapping(ISLocalToGlobalMapping, Mat, PetscBool, PetscBool, MatType, Mat *);
264: /* these callbacks rely on the old matrix function pointers for
265: matmat operations. They are unsafe, and should be removed.
266: However, the amount of work needed to clean up all the
267: implementations is not negligible */
268: PETSC_INTERN PetscErrorCode MatProductSymbolic_AB(Mat);
269: PETSC_INTERN PetscErrorCode MatProductNumeric_AB(Mat);
270: PETSC_INTERN PetscErrorCode MatProductSymbolic_AtB(Mat);
271: PETSC_INTERN PetscErrorCode MatProductNumeric_AtB(Mat);
272: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABt(Mat);
273: PETSC_INTERN PetscErrorCode MatProductNumeric_ABt(Mat);
274: PETSC_INTERN PetscErrorCode MatProductNumeric_PtAP(Mat);
275: PETSC_INTERN PetscErrorCode MatProductNumeric_RARt(Mat);
276: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABC(Mat);
277: PETSC_INTERN PetscErrorCode MatProductNumeric_ABC(Mat);
279: PETSC_INTERN PetscErrorCode MatProductCreate_Private(Mat, Mat, Mat, Mat);
280: /* this callback handles all the different triple products and
281: does not rely on the function pointers; used by cuSPARSE/hipSPARSE and KOKKOS-KERNELS */
282: PETSC_INTERN PetscErrorCode MatProductSymbolic_ABC_Basic(Mat);
284: /* CreateGraph is common to AIJ seq and mpi */
285: PETSC_INTERN PetscErrorCode MatCreateGraph_Simple_AIJ(Mat, PetscBool, PetscBool, PetscReal, PetscInt, PetscInt[], Mat *);
287: #if defined(PETSC_CLANG_STATIC_ANALYZER)
288: template <typename Tm>
289: extern void MatCheckPreallocated(Tm, int);
290: template <typename Tm>
291: extern void MatCheckProduct(Tm, int);
292: #else /* PETSC_CLANG_STATIC_ANALYZER */
293: #define MatCheckPreallocated(A, arg) \
294: do { \
295: if (!(A)->preallocated) PetscCall(MatSetUp(A)); \
296: } while (0)
298: #if defined(PETSC_USE_DEBUG)
299: #define MatCheckProduct(A, arg) \
300: do { \
301: PetscCheck((A)->product, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Argument %d \"%s\" is not a matrix obtained from MatProductCreate()", (arg), #A); \
302: } while (0)
303: #else
304: #define MatCheckProduct(A, arg) \
305: do { \
306: } while (0)
307: #endif
308: #endif /* PETSC_CLANG_STATIC_ANALYZER */
310: /*
311: The stash is used to temporarily store inserted matrix values that
312: belong to another processor. During the assembly phase the stashed
313: values are moved to the correct processor and
314: */
316: typedef struct _MatStashSpace *PetscMatStashSpace;
318: struct _MatStashSpace {
319: PetscMatStashSpace next;
320: PetscScalar *space_head, *val;
321: PetscInt *idx, *idy;
322: PetscInt total_space_size;
323: PetscInt local_used;
324: PetscInt local_remaining;
325: };
327: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceGet(PetscInt, PetscInt, PetscMatStashSpace *);
328: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceContiguous(PetscInt, PetscMatStashSpace *, PetscScalar *, PetscInt *, PetscInt *);
329: PETSC_EXTERN PetscErrorCode PetscMatStashSpaceDestroy(PetscMatStashSpace *);
331: typedef struct {
332: PetscInt count;
333: } MatStashHeader;
335: typedef struct {
336: void *buffer; /* Of type blocktype, dynamically constructed */
337: PetscInt count;
338: char pending;
339: } MatStashFrame;
341: typedef struct _MatStash MatStash;
342: struct _MatStash {
343: PetscInt nmax; /* maximum stash size */
344: PetscInt umax; /* user specified max-size */
345: PetscInt oldnmax; /* the nmax value used previously */
346: PetscInt n; /* stash size */
347: PetscInt bs; /* block size of the stash */
348: PetscInt reallocs; /* preserve the no of mallocs invoked */
349: PetscMatStashSpace space_head, space; /* linked list to hold stashed global row/column numbers and matrix values */
351: PetscErrorCode (*ScatterBegin)(Mat, MatStash *, PetscInt *);
352: PetscErrorCode (*ScatterGetMesg)(MatStash *, PetscMPIInt *, PetscInt **, PetscInt **, PetscScalar **, PetscInt *);
353: PetscErrorCode (*ScatterEnd)(MatStash *);
354: PetscErrorCode (*ScatterDestroy)(MatStash *);
356: /* The following variables are used for communication */
357: MPI_Comm comm;
358: PetscMPIInt size, rank;
359: PetscMPIInt tag1, tag2;
360: MPI_Request *send_waits; /* array of send requests */
361: MPI_Request *recv_waits; /* array of receive requests */
362: MPI_Status *send_status; /* array of send status */
363: PetscMPIInt nsends, nrecvs; /* numbers of sends and receives */
364: PetscScalar *svalues; /* sending data */
365: PetscInt *sindices;
366: PetscScalar **rvalues; /* receiving data (values) */
367: PetscInt **rindices; /* receiving data (indices) */
368: PetscMPIInt nprocessed; /* number of messages already processed */
369: PetscMPIInt *flg_v; /* indicates what messages have arrived so far and from whom */
370: PetscBool reproduce;
371: PetscMPIInt reproduce_count;
373: /* The following variables are used for BTS communication */
374: PetscBool first_assembly_done; /* Is the first time matrix assembly done? */
375: PetscBool use_status; /* Use MPI_Status to determine number of items in each message */
376: PetscMPIInt nsendranks;
377: PetscMPIInt nrecvranks;
378: PetscMPIInt *sendranks;
379: PetscMPIInt *recvranks;
380: MatStashHeader *sendhdr, *recvhdr;
381: MatStashFrame *sendframes; /* pointers to the main messages */
382: MatStashFrame *recvframes;
383: MatStashFrame *recvframe_active;
384: PetscInt recvframe_i; /* index of block within active frame */
385: PetscInt recvframe_count; /* Count actually sent for current frame */
386: PetscMPIInt recvcount; /* Number of receives processed so far */
387: PetscMPIInt *some_indices; /* From last call to MPI_Waitsome */
388: MPI_Status *some_statuses; /* Statuses from last call to MPI_Waitsome */
389: PetscMPIInt some_count; /* Number of requests completed in last call to MPI_Waitsome */
390: PetscMPIInt some_i; /* Index of request currently being processed */
391: MPI_Request *sendreqs;
392: MPI_Request *recvreqs;
393: PetscSegBuffer segsendblocks;
394: PetscSegBuffer segrecvframe;
395: PetscSegBuffer segrecvblocks;
396: MPI_Datatype blocktype;
397: size_t blocktype_size;
398: InsertMode *insertmode; /* Pointer to check mat->insertmode and set upon message arrival in case no local values have been set. */
399: };
401: #if !defined(PETSC_HAVE_MPIUNI)
402: PETSC_INTERN PetscErrorCode MatStashScatterDestroy_BTS(MatStash *);
403: #endif
404: PETSC_INTERN PetscErrorCode MatStashCreate_Private(MPI_Comm, PetscInt, MatStash *);
405: PETSC_INTERN PetscErrorCode MatStashDestroy_Private(MatStash *);
406: PETSC_INTERN PetscErrorCode MatStashScatterEnd_Private(MatStash *);
407: PETSC_INTERN PetscErrorCode MatStashSetInitialSize_Private(MatStash *, PetscInt);
408: PETSC_INTERN PetscErrorCode MatStashGetInfo_Private(MatStash *, PetscInt *, PetscInt *);
409: PETSC_INTERN PetscErrorCode MatStashValuesRow_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscBool);
410: PETSC_INTERN PetscErrorCode MatStashValuesCol_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscBool);
411: PETSC_INTERN PetscErrorCode MatStashValuesRowBlocked_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscInt, PetscInt);
412: PETSC_INTERN PetscErrorCode MatStashValuesColBlocked_Private(MatStash *, PetscInt, PetscInt, const PetscInt[], const PetscScalar[], PetscInt, PetscInt, PetscInt);
413: PETSC_INTERN PetscErrorCode MatStashScatterBegin_Private(Mat, MatStash *, PetscInt *);
414: PETSC_INTERN PetscErrorCode MatStashScatterGetMesg_Private(MatStash *, PetscMPIInt *, PetscInt **, PetscInt **, PetscScalar **, PetscInt *);
415: PETSC_INTERN PetscErrorCode MatGetInfo_External(Mat, MatInfoType, MatInfo *);
417: typedef struct {
418: PetscInt dim;
419: PetscInt dims[4];
420: PetscInt starts[4];
421: PetscBool noc; /* this is a single component problem, hence user will not set MatStencil.c */
422: } MatStencilInfo;
424: /* Info about using compressed row format */
425: typedef struct {
426: PetscBool use; /* indicates compressed rows have been checked and will be used */
427: PetscInt nrows; /* number of non-zero rows */
428: PetscInt *i; /* compressed row pointer */
429: PetscInt *rindex; /* compressed row index */
430: } Mat_CompressedRow;
431: PETSC_EXTERN PetscErrorCode MatCheckCompressedRow(Mat, PetscInt, Mat_CompressedRow *, PetscInt *, PetscInt, PetscReal);
433: typedef struct { /* used by MatCreateRedundantMatrix() for reusing matredundant */
434: PetscInt nzlocal, nsends, nrecvs;
435: PetscMPIInt *send_rank, *recv_rank;
436: PetscInt *sbuf_nz, *rbuf_nz, *sbuf_j, **rbuf_j;
437: PetscScalar *sbuf_a, **rbuf_a;
438: MPI_Comm subcomm; /* when user does not provide a subcomm */
439: IS isrow, iscol;
440: Mat *matseq;
441: } Mat_Redundant;
443: typedef struct { /* used by MatProduct() */
444: MatProductType type;
445: char *alg;
446: Mat A, B, C, Dwork;
447: 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, .. */
448: 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 .. */
449: PetscBool symbolic_used_the_fact_C_is_symmetric; /* MatMatMult(A,B,MAT_REUSE_MATRIX,..&C) is still legitimate), we need to redo symbolic! */
450: PetscObjectParameterDeclare(PetscReal, fill);
451: 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 */
452: PetscBool setfromoptionscalled;
454: /* Some products may display the information on the algorithm used */
455: PetscErrorCode (*view)(Mat, PetscViewer);
457: /* many products have intermediate data structures, each specific to Mat types and product type */
458: PetscBool clear; /* whether or not to clear the data structures after MatProductNumeric has been called */
459: void *data; /* where to stash those structures */
460: PetscCtxDestroyFn *destroy; /* freeing data */
461: } Mat_Product;
463: struct _p_Mat {
464: PETSCHEADER(struct _MatOps);
465: PetscLayout rmap, cmap;
466: void *data; /* implementation-specific data */
467: MatFactorType factortype; /* MAT_FACTOR_LU, ILU, CHOLESKY or ICC */
468: PetscBool trivialsymbolic; /* indicates the symbolic factorization doesn't actually do a symbolic factorization, it is delayed to the numeric factorization */
469: PetscBool canuseordering; /* factorization can use ordering provide to routine (most PETSc implementations) */
470: MatOrderingType preferredordering[MAT_FACTOR_NUM_TYPES]; /* what is the preferred (or default) ordering for the matrix solver type */
471: PetscBool assembled; /* is the matrix assembled? */
472: PetscBool was_assembled; /* new values inserted into assembled mat */
473: PetscInt num_ass; /* number of times matrix has been assembled */
474: PetscObjectState nonzerostate; /* each time new nonzeros locations are introduced into the matrix this is updated */
475: PetscObjectState ass_nonzerostate; /* nonzero state at last assembly */
476: MatInfo info; /* matrix information */
477: InsertMode insertmode; /* have values been inserted in matrix or added? */
478: MatStash stash, bstash; /* used for assembling off-proc mat emements */
479: MatNullSpace nullsp; /* null space (operator is singular) */
480: MatNullSpace transnullsp; /* null space of transpose of operator */
481: MatNullSpace nearnullsp; /* near null space to be used by multigrid methods */
482: PetscInt congruentlayouts; /* are the rows and columns layouts congruent? */
483: PetscBool preallocated;
484: MatStencilInfo stencil; /* information for structured grid */
485: PetscBool3 symmetric, hermitian, structurally_symmetric, spd;
486: PetscBool symmetry_eternal, structural_symmetry_eternal, spd_eternal;
487: PetscBool nooffprocentries, nooffproczerorows;
488: PetscBool assembly_subset; /* set by MAT_SUBSET_OFF_PROC_ENTRIES */
489: PetscBool submat_singleis; /* for efficient PCSetUp_ASM() */
490: PetscBool structure_only;
491: PetscBool sortedfull; /* full, sorted rows are inserted */
492: PetscBool force_diagonals; /* set by MAT_FORCE_DIAGONAL_ENTRIES */
493: #if defined(PETSC_HAVE_DEVICE)
494: PetscOffloadMask offloadmask; /* a mask which indicates where the valid matrix data is (GPU, CPU or both) */
495: PetscBool boundtocpu;
496: PetscBool bindingpropagates;
497: #endif
498: char *defaultrandtype;
499: void *spptr; /* pointer for special library like SuperLU */
500: char *solvertype;
501: PetscBool checksymmetryonassembly, checknullspaceonassembly;
502: PetscReal checksymmetrytol;
503: Mat schur; /* Schur complement matrix */
504: MatFactorSchurStatus schur_status; /* status of the Schur complement matrix */
505: Mat_Redundant *redundant; /* used by MatCreateRedundantMatrix() */
506: PetscBool erroriffailure; /* Generate an error if detected (for example a zero pivot) instead of returning */
507: MatFactorError factorerrortype; /* type of error in factorization */
508: PetscReal factorerror_zeropivot_value; /* If numerical zero pivot was detected this is the computed value */
509: PetscInt factorerror_zeropivot_row; /* Row where zero pivot was detected */
510: PetscInt nblocks, *bsizes; /* support for MatSetVariableBlockSizes() */
511: PetscInt p_cstart, p_rank, p_cend, n_rank; /* Information from parallel MatComputeVariableBlockEnvelope() */
512: PetscBool p_parallel;
513: char *defaultvectype;
514: Mat_Product *product;
515: PetscBool form_explicit_transpose; /* hint to generate an explicit mat tranpsose for operations like MatMultTranspose() */
516: PetscBool transupdated; /* whether or not the explicitly generated transpose is up-to-date */
517: char *factorprefix; /* the prefix to use with factored matrix that is created */
518: PetscBool hash_active; /* indicates MatSetValues() is being handled by hashing */
519: Vec dot_vec; /* work vector used by MatADot_Default() */
520: };
522: PETSC_INTERN PetscErrorCode MatAXPY_Basic(Mat, PetscScalar, Mat, MatStructure);
523: PETSC_INTERN PetscErrorCode MatAXPY_BasicWithPreallocation(Mat, Mat, PetscScalar, Mat, MatStructure);
524: PETSC_INTERN PetscErrorCode MatAXPY_Basic_Preallocate(Mat, Mat, Mat *);
525: PETSC_INTERN PetscErrorCode MatAXPY_Dense_Nest(Mat, PetscScalar, Mat);
527: PETSC_INTERN PetscErrorCode MatSetUp_Default(Mat);
529: /*
530: Utility for MatZeroRows
531: */
532: PETSC_INTERN PetscErrorCode MatZeroRowsMapLocal_Private(Mat, PetscInt, const PetscInt *, PetscInt *, PetscInt **);
534: /*
535: Utility for MatView/MatLoad
536: */
537: PETSC_INTERN PetscErrorCode MatView_Binary_BlockSizes(Mat, PetscViewer);
538: PETSC_INTERN PetscErrorCode MatLoad_Binary_BlockSizes(Mat, PetscViewer);
540: /*
541: Object for partitioning graphs
542: */
544: typedef struct _MatPartitioningOps *MatPartitioningOps;
545: struct _MatPartitioningOps {
546: PetscErrorCode (*apply)(MatPartitioning, IS *);
547: PetscErrorCode (*applynd)(MatPartitioning, IS *);
548: PetscErrorCode (*setfromoptions)(MatPartitioning, PetscOptionItems);
549: PetscErrorCode (*destroy)(MatPartitioning);
550: PetscErrorCode (*view)(MatPartitioning, PetscViewer);
551: PetscErrorCode (*improve)(MatPartitioning, IS *);
552: };
554: struct _p_MatPartitioning {
555: PETSCHEADER(struct _MatPartitioningOps);
556: Mat adj;
557: PetscInt *vertex_weights;
558: PetscReal *part_weights;
559: PetscInt n; /* number of partitions */
560: PetscInt ncon; /* number of vertex weights per vertex */
561: void *data;
562: PetscBool use_edge_weights; /* A flag indicates whether or not to use edge weights */
563: };
565: /* needed for parallel nested dissection by ParMetis and PTSCOTCH */
566: PETSC_INTERN PetscErrorCode MatPartitioningSizesToSep_Private(PetscInt, PetscInt[], PetscInt[], PetscInt[]);
568: /*
569: Object for coarsen graphs
570: */
571: typedef struct _MatCoarsenOps *MatCoarsenOps;
572: struct _MatCoarsenOps {
573: PetscErrorCode (*apply)(MatCoarsen);
574: PetscErrorCode (*setfromoptions)(MatCoarsen, PetscOptionItems);
575: PetscErrorCode (*destroy)(MatCoarsen);
576: PetscErrorCode (*view)(MatCoarsen, PetscViewer);
577: };
579: #define MAT_COARSEN_STRENGTH_INDEX_SIZE 3
580: struct _p_MatCoarsen {
581: PETSCHEADER(struct _MatCoarsenOps);
582: Mat graph;
583: void *subctx;
584: /* */
585: PetscBool strict_aggs;
586: IS perm;
587: PetscCoarsenData *agg_lists;
588: PetscInt max_it; /* number of iterations in HEM */
589: PetscReal threshold; /* HEM can filter interim graphs */
590: PetscInt strength_index_size;
591: PetscInt strength_index[MAT_COARSEN_STRENGTH_INDEX_SIZE];
592: };
594: PETSC_EXTERN PetscErrorCode MatCoarsenMISKSetDistance(MatCoarsen, PetscInt);
595: PETSC_EXTERN PetscErrorCode MatCoarsenMISKGetDistance(MatCoarsen, PetscInt *);
597: /*
598: Used in aijdevice.h
599: */
600: typedef struct {
601: PetscInt *i;
602: PetscInt *j;
603: PetscScalar *a;
604: PetscInt n;
605: PetscInt ignorezeroentries;
606: } PetscCSRDataStructure;
608: /*
609: MatFDColoring is used to compute Jacobian matrices efficiently
610: via coloring. The data structure is explained below in an example.
612: Color = 0 1 0 2 | 2 3 0
613: ---------------------------------------------------
614: 00 01 | 05
615: 10 11 | 14 15 Processor 0
616: 22 23 | 25
617: 32 33 |
618: ===================================================
619: | 44 45 46
620: 50 | 55 Processor 1
621: | 64 66
622: ---------------------------------------------------
624: ncolors = 4;
626: ncolumns = {2,1,1,0}
627: columns = {{0,2},{1},{3},{}}
628: nrows = {4,2,3,3}
629: rows = {{0,1,2,3},{0,1},{1,2,3},{0,1,2}}
630: vwscale = {dx(0),dx(1),dx(2),dx(3)} MPI Vec
631: vscale = {dx(0),dx(1),dx(2),dx(3),dx(4),dx(5)} Seq Vec
633: ncolumns = {1,0,1,1}
634: columns = {{6},{},{4},{5}}
635: nrows = {3,0,2,2}
636: rows = {{0,1,2},{},{1,2},{1,2}}
637: vwscale = {dx(4),dx(5),dx(6)} MPI Vec
638: vscale = {dx(0),dx(4),dx(5),dx(6)} Seq Vec
640: See the routine MatFDColoringApply() for how this data is used
641: to compute the Jacobian.
643: */
644: typedef struct {
645: PetscInt row;
646: PetscInt col;
647: PetscScalar *valaddr; /* address of value */
648: } MatEntry;
650: typedef struct {
651: PetscInt row;
652: PetscScalar *valaddr; /* address of value */
653: } MatEntry2;
655: struct _p_MatFDColoring {
656: PETSCHEADER(int);
657: PetscInt M, N, m; /* total rows, columns; local rows */
658: PetscInt rstart; /* first row owned by local processor */
659: PetscInt ncolors; /* number of colors */
660: PetscInt *ncolumns; /* number of local columns for a color */
661: PetscInt **columns; /* lists the local columns of each color (using global column numbering) */
662: IS *isa; /* these are the IS that contain the column values given in columns */
663: PetscInt *nrows; /* number of local rows for each color */
664: MatEntry *matentry; /* holds (row, column, address of value) for Jacobian matrix entry */
665: MatEntry2 *matentry2; /* holds (row, address of value) for Jacobian matrix entry */
666: PetscScalar *dy; /* store a block of F(x+dx)-F(x) when J is in BAIJ format */
667: PetscReal error_rel; /* square root of relative error in computing function */
668: PetscReal umin; /* minimum allowable u'dx value */
669: Vec w1, w2, w3; /* work vectors used in computing Jacobian */
670: PetscBool fset; /* indicates that the initial function value F(X) is set */
671: MatFDColoringFn *f; /* function that defines Jacobian */
672: void *fctx; /* optional user-defined context for use by the function f */
673: Vec vscale; /* holds FD scaling, i.e. 1/dx for each perturbed column */
674: PetscInt currentcolor; /* color for which function evaluation is being done now */
675: const char *htype; /* "wp" or "ds" */
676: ISColoringType ctype; /* IS_COLORING_GLOBAL or IS_COLORING_LOCAL */
677: PetscInt brows, bcols; /* number of block rows or columns for speedup inserting the dense matrix into sparse Jacobian */
678: PetscBool setupcalled; /* true if setup has been called */
679: PetscBool viewed; /* true if the -mat_fd_coloring_view has been triggered already */
680: PetscFortranCallbackFn *ftn_func_pointer; /* serve the same purpose as *fortran_func_pointers in PETSc objects */
681: void *ftn_func_cntx;
682: PetscObjectId matid; /* matrix this object was created with, must always be the same */
683: };
685: typedef struct _MatColoringOps *MatColoringOps;
686: struct _MatColoringOps {
687: PetscErrorCode (*destroy)(MatColoring);
688: PetscErrorCode (*setfromoptions)(MatColoring, PetscOptionItems);
689: PetscErrorCode (*view)(MatColoring, PetscViewer);
690: PetscErrorCode (*apply)(MatColoring, ISColoring *);
691: PetscErrorCode (*weights)(MatColoring, PetscReal **, PetscInt **);
692: };
694: struct _p_MatColoring {
695: PETSCHEADER(struct _MatColoringOps);
696: Mat mat;
697: PetscInt dist; /* distance of the coloring */
698: PetscInt maxcolors; /* the maximum number of colors returned, maxcolors=1 for MIS */
699: void *data; /* inner context */
700: PetscBool valid; /* check to see if what is produced is a valid coloring */
701: MatColoringWeightType weight_type; /* type of weight computation to be performed */
702: PetscReal *user_weights; /* custom weights and permutation */
703: PetscInt *user_lperm;
704: PetscBool valid_iscoloring; /* check to see if matcoloring is produced a valid iscoloring */
705: };
707: struct _p_MatTransposeColoring {
708: PETSCHEADER(int);
709: PetscInt M, N, m; /* total rows, columns; local rows */
710: PetscInt rstart; /* first row owned by local processor */
711: PetscInt ncolors; /* number of colors */
712: PetscInt *ncolumns; /* number of local columns for a color */
713: PetscInt *nrows; /* number of local rows for each color */
714: PetscInt currentcolor; /* color for which function evaluation is being done now */
715: ISColoringType ctype; /* IS_COLORING_GLOBAL or IS_COLORING_LOCAL */
717: PetscInt *colorforrow, *colorforcol; /* pointer to rows and columns */
718: PetscInt *rows; /* lists the local rows for each color (using the local row numbering) */
719: PetscInt *den2sp; /* maps (row,color) in the dense matrix to index of sparse matrix array a->a */
720: PetscInt *columns; /* lists the local columns of each color (using global column numbering) */
721: PetscInt brows; /* number of rows for efficient implementation of MatTransColoringApplyDenToSp() */
722: PetscInt *lstart; /* array used for loop over row blocks of Csparse */
723: };
725: /*
726: Null space context for preconditioner/operators
727: */
728: struct _p_MatNullSpace {
729: PETSCHEADER(int);
730: PetscBool has_cnst;
731: PetscInt n;
732: Vec *vecs;
733: PetscScalar *alpha; /* for projections */
734: MatNullSpaceRemoveFn *remove; /* for user provided removal function */
735: void *rmctx; /* context for remove() function */
736: };
738: /*
739: Internal data structure for MATMPIDENSE
740: */
741: typedef struct {
742: Mat A; /* local submatrix */
744: /* The following variables are used for matrix assembly */
745: PetscBool donotstash; /* Flag indicating if values should be stashed */
746: MPI_Request *send_waits; /* array of send requests */
747: MPI_Request *recv_waits; /* array of receive requests */
748: PetscInt nsends, nrecvs; /* numbers of sends and receives */
749: PetscScalar *svalues, *rvalues; /* sending and receiving data */
750: PetscInt rmax; /* maximum message length */
752: /* The following variables are used for matrix-vector products */
753: Vec lvec; /* local vector */
754: PetscSF Mvctx; /* for mat-mult communications */
755: PetscBool roworiented; /* if true, row-oriented input (default) */
757: /* Support for MatDenseGetColumnVec and MatDenseGetSubMatrix */
758: Mat cmat; /* matrix representation of a given subset of columns */
759: Vec cvec; /* vector representation of a given column */
760: const PetscScalar *ptrinuse; /* holds array to be restored (just a placeholder) */
761: PetscInt vecinuse; /* if cvec is in use (col = vecinuse-1) */
762: PetscInt matinuse; /* if cmat is in use (cbegin = matinuse-1) */
763: /* if this is from MatDenseGetSubMatrix, which columns and rows does it correspond to? */
764: PetscInt sub_rbegin;
765: PetscInt sub_rend;
766: PetscInt sub_cbegin;
767: PetscInt sub_cend;
768: } Mat_MPIDense;
770: /*
771: Checking zero pivot for LU, ILU preconditioners.
772: */
773: typedef struct {
774: PetscInt nshift, nshift_max;
775: PetscReal shift_amount, shift_lo, shift_hi, shift_top, shift_fraction;
776: PetscBool newshift;
777: PetscReal rs; /* active row sum of abs(off-diagonals) */
778: PetscScalar pv; /* pivot of the active row */
779: } FactorShiftCtx;
781: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatTransposeCheckNonzeroState_Private(Mat, Mat);
783: /*
784: Used by MatTranspose() and potentially other functions to track the matrix used in the generation of another matrix
785: */
786: typedef struct {
787: PetscObjectId id;
788: PetscObjectState state;
789: PetscObjectState nonzerostate;
790: } MatParentState;
792: PETSC_EXTERN PetscErrorCode MatFactorDumpMatrix(Mat);
793: PETSC_INTERN PetscErrorCode MatSetBlockSizes_Default(Mat, PetscInt, PetscInt);
795: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatShift_Basic(Mat, PetscScalar);
797: static inline PetscErrorCode MatPivotCheck_nz(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
798: {
799: PetscReal _rs = sctx->rs;
800: PetscReal _zero = info->zeropivot * _rs;
802: PetscFunctionBegin;
803: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
804: /* force |diag| > zeropivot*rs */
805: if (!sctx->nshift) sctx->shift_amount = info->shiftamount;
806: else sctx->shift_amount *= 2.0;
807: sctx->newshift = PETSC_TRUE;
808: (sctx->nshift)++;
809: } else {
810: sctx->newshift = PETSC_FALSE;
811: }
812: PetscFunctionReturn(PETSC_SUCCESS);
813: }
815: static inline PetscErrorCode MatPivotCheck_pd(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
816: {
817: PetscReal _rs = sctx->rs;
818: PetscReal _zero = info->zeropivot * _rs;
820: PetscFunctionBegin;
821: if (PetscRealPart(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
822: /* force matfactor to be diagonally dominant */
823: if (sctx->nshift == sctx->nshift_max) {
824: sctx->shift_fraction = sctx->shift_hi;
825: } else {
826: sctx->shift_lo = sctx->shift_fraction;
827: sctx->shift_fraction = (sctx->shift_hi + sctx->shift_lo) / (PetscReal)2.;
828: }
829: sctx->shift_amount = sctx->shift_fraction * sctx->shift_top;
830: sctx->nshift++;
831: sctx->newshift = PETSC_TRUE;
832: } else {
833: sctx->newshift = PETSC_FALSE;
834: }
835: PetscFunctionReturn(PETSC_SUCCESS);
836: }
838: static inline PetscErrorCode MatPivotCheck_inblocks(PETSC_UNUSED Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PETSC_UNUSED PetscInt row)
839: {
840: PetscReal _zero = info->zeropivot;
842: PetscFunctionBegin;
843: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
844: sctx->pv += info->shiftamount;
845: sctx->shift_amount = 0.0;
846: sctx->nshift++;
847: }
848: sctx->newshift = PETSC_FALSE;
849: PetscFunctionReturn(PETSC_SUCCESS);
850: }
852: static inline PetscErrorCode MatPivotCheck_none(Mat fact, Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PetscInt row)
853: {
854: PetscReal _zero = info->zeropivot;
856: PetscFunctionBegin;
857: sctx->newshift = PETSC_FALSE;
858: if (PetscAbsScalar(sctx->pv) <= _zero && !PetscIsNanScalar(sctx->pv)) {
859: 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);
860: PetscCall(PetscInfo(mat, "Detected zero pivot in factorization in row %" PetscInt_FMT " value %g tolerance %g\n", row, (double)PetscAbsScalar(sctx->pv), (double)_zero));
861: fact->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
862: fact->factorerror_zeropivot_value = PetscAbsScalar(sctx->pv);
863: fact->factorerror_zeropivot_row = row;
864: }
865: PetscFunctionReturn(PETSC_SUCCESS);
866: }
868: static inline PetscErrorCode MatPivotCheck(Mat fact, Mat mat, const MatFactorInfo *info, FactorShiftCtx *sctx, PetscInt row)
869: {
870: PetscFunctionBegin;
871: if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) PetscCall(MatPivotCheck_nz(mat, info, sctx, row));
872: else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) PetscCall(MatPivotCheck_pd(mat, info, sctx, row));
873: else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) PetscCall(MatPivotCheck_inblocks(mat, info, sctx, row));
874: else PetscCall(MatPivotCheck_none(fact, mat, info, sctx, row));
875: PetscFunctionReturn(PETSC_SUCCESS);
876: }
878: PETSC_INTERN PetscErrorCode MatADot_Default(Mat, Vec, Vec, PetscScalar *);
879: PETSC_INTERN PetscErrorCode MatANorm_Default(Mat, Vec, PetscReal *);
881: #include <petscbt.h>
882: /*
883: Create and initialize a linked list
884: Input Parameters:
885: idx_start - starting index of the list
886: lnk_max - max value of lnk indicating the end of the list
887: nlnk - max length of the list
888: Output Parameters:
889: lnk - list initialized
890: bt - PetscBT (bitarray) with all bits set to false
891: lnk_empty - flg indicating the list is empty
892: */
893: #define PetscLLCreate(idx_start, lnk_max, nlnk, lnk, bt) ((PetscErrorCode)(PetscMalloc1(nlnk, &lnk) || PetscBTCreate(nlnk, &(bt)) || (lnk[idx_start] = lnk_max, PETSC_SUCCESS)))
895: #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)))
897: 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)
898: {
899: PetscInt location;
901: PetscFunctionBegin;
902: /* start from the beginning if entry < previous entry */
903: if (!assume_sorted && k && entry < *lnkdata) *lnkdata = idx_start;
904: /* search for insertion location */
905: do {
906: location = *lnkdata;
907: *lnkdata = lnk[location];
908: } while (entry > *lnkdata);
909: /* insertion location is found, add entry into lnk */
910: lnk[location] = entry;
911: lnk[entry] = *lnkdata;
912: ++(*nlnk);
913: *lnkdata = entry; /* next search starts from here if next_entry > entry */
914: PetscFunctionReturn(PETSC_SUCCESS);
915: }
917: 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)
918: {
919: PetscFunctionBegin;
920: *nlnk = 0;
921: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
922: const PetscInt entry = indices[k];
924: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, &lnkdata, lnk));
925: }
926: PetscFunctionReturn(PETSC_SUCCESS);
927: }
929: /*
930: Add an index set into a sorted linked list
931: Input Parameters:
932: nidx - number of input indices
933: indices - integer array
934: idx_start - starting index of the list
935: lnk - linked list(an integer array) that is created
936: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
937: output Parameters:
938: nlnk - number of newly added indices
939: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
940: bt - updated PetscBT (bitarray)
941: */
942: static inline PetscErrorCode PetscLLAdd(PetscInt nidx, const PetscInt *PETSC_RESTRICT indices, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscBT bt)
943: {
944: PetscFunctionBegin;
945: PetscCall(PetscLLAdd_Private(nidx, indices, idx_start, nlnk, lnk, bt, PETSC_FALSE));
946: PetscFunctionReturn(PETSC_SUCCESS);
947: }
949: /*
950: Add a SORTED ascending index set into a sorted linked list - same as PetscLLAdd() bus skip 'if (_k && _entry < _lnkdata) _lnkdata = idx_start;'
951: Input Parameters:
952: nidx - number of input indices
953: indices - sorted 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 PetscLLAddSorted(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_TRUE));
966: PetscFunctionReturn(PETSC_SUCCESS);
967: }
969: /*
970: Add a permuted index set into a sorted linked list
971: Input Parameters:
972: nidx - number of input indices
973: indices - integer array
974: perm - permutation of indices
975: idx_start - starting index of the list
976: lnk - linked list(an integer array) that is created
977: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
978: output Parameters:
979: nlnk - number of newly added indices
980: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
981: bt - updated PetscBT (bitarray)
982: */
983: 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)
984: {
985: PetscFunctionBegin;
986: *nlnk = 0;
987: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
988: const PetscInt entry = perm[indices[k]];
990: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_FALSE, k, idx_start, entry, nlnk, &lnkdata, lnk));
991: }
992: PetscFunctionReturn(PETSC_SUCCESS);
993: }
995: #if 0
996: /* this appears to be unused? */
997: static inline PetscErrorCode PetscLLAddSorted_new(PetscInt nidx, PetscInt *indices, PetscInt idx_start, PetscBool *lnk_empty, PetscInt *nlnk, PetscInt *lnk, PetscBT bt)
998: {
999: PetscInt lnkdata = idx_start;
1001: PetscFunctionBegin;
1002: if (*lnk_empty) {
1003: for (PetscInt k = 0; k < nidx; ++k) {
1004: const PetscInt entry = indices[k], location = lnkdata;
1006: PetscCall(PetscBTSet(bt,entry)); /* mark the new entry */
1007: lnkdata = lnk[location];
1008: /* insertion location is found, add entry into lnk */
1009: lnk[location] = entry;
1010: lnk[entry] = lnkdata;
1011: lnkdata = entry; /* next search starts from here */
1012: }
1013: /* lnk[indices[nidx-1]] = lnk[idx_start];
1014: lnk[idx_start] = indices[0];
1015: PetscCall(PetscBTSet(bt,indices[0]));
1016: for (_k=1; _k<nidx; _k++) {
1017: PetscCall(PetscBTSet(bt,indices[_k]));
1018: lnk[indices[_k-1]] = indices[_k];
1019: }
1020: */
1021: *nlnk = nidx;
1022: *lnk_empty = PETSC_FALSE;
1023: } else {
1024: *nlnk = 0;
1025: for (PetscInt k = 0; k < nidx; ++k) {
1026: const PetscInt entry = indices[k];
1028: if (!PetscBTLookupSet(bt,entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_TRUE,k,idx_start,entry,nlnk,&lnkdata,lnk));
1029: }
1030: }
1031: PetscFunctionReturn(PETSC_SUCCESS);
1032: }
1033: #endif
1035: /*
1036: Add a SORTED index set into a sorted linked list used for LUFactorSymbolic()
1037: Same as PetscLLAddSorted() with an additional operation:
1038: count the number of input indices that are no larger than 'diag'
1039: Input Parameters:
1040: indices - sorted integer array
1041: idx_start - starting index of the list, index of pivot row
1042: lnk - linked list(an integer array) that is created
1043: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1044: diag - index of the active row in LUFactorSymbolic
1045: nzbd - number of input indices with indices <= idx_start
1046: im - im[idx_start] is initialized as num of nonzero entries in row=idx_start
1047: output Parameters:
1048: nlnk - number of newly added indices
1049: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from indices
1050: bt - updated PetscBT (bitarray)
1051: im - im[idx_start]: unchanged if diag is not an entry
1052: : num of entries with indices <= diag if diag is an entry
1053: */
1054: 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)
1055: {
1056: const PetscInt nidx = im[idx_start] - nzbd; /* num of entries with idx_start < index <= diag */
1058: PetscFunctionBegin;
1059: *nlnk = 0;
1060: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1061: const PetscInt entry = indices[k];
1063: ++nzbd;
1064: if (entry == diag) im[idx_start] = nzbd;
1065: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscLLInsertLocation_Private(PETSC_TRUE, k, idx_start, entry, nlnk, &lnkdata, lnk));
1066: }
1067: PetscFunctionReturn(PETSC_SUCCESS);
1068: }
1070: /*
1071: Copy data on the list into an array, then initialize the list
1072: Input Parameters:
1073: idx_start - starting index of the list
1074: lnk_max - max value of lnk indicating the end of the list
1075: nlnk - number of data on the list to be copied
1076: lnk - linked list
1077: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1078: output Parameters:
1079: indices - array that contains the copied data
1080: lnk - linked list that is cleaned and initialize
1081: bt - PetscBT (bitarray) with all bits set to false
1082: */
1083: static inline PetscErrorCode PetscLLClean(PetscInt idx_start, PetscInt lnk_max, PetscInt nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT indices, PetscBT bt)
1084: {
1085: PetscFunctionBegin;
1086: for (PetscInt j = 0, idx = idx_start; j < nlnk; ++j) {
1087: idx = lnk[idx];
1088: indices[j] = idx;
1089: PetscCall(PetscBTClear(bt, idx));
1090: }
1091: lnk[idx_start] = lnk_max;
1092: PetscFunctionReturn(PETSC_SUCCESS);
1093: }
1095: /*
1096: Free memories used by the list
1097: */
1098: #define PetscLLDestroy(lnk, bt) ((PetscErrorCode)(PetscFree(lnk) || PetscBTDestroy(&(bt))))
1100: /* Routines below are used for incomplete matrix factorization */
1101: /*
1102: Create and initialize a linked list and its levels
1103: Input Parameters:
1104: idx_start - starting index of the list
1105: lnk_max - max value of lnk indicating the end of the list
1106: nlnk - max length of the list
1107: Output Parameters:
1108: lnk - list initialized
1109: lnk_lvl - array of size nlnk for storing levels of lnk
1110: bt - PetscBT (bitarray) with all bits set to false
1111: */
1112: #define PetscIncompleteLLCreate(idx_start, lnk_max, nlnk, lnk, lnk_lvl, bt) \
1113: ((PetscErrorCode)(PetscIntMultError(2, nlnk, NULL) || PetscMalloc1(2 * nlnk, &lnk) || PetscBTCreate(nlnk, &(bt)) || (lnk[idx_start] = lnk_max, lnk_lvl = lnk + nlnk, PETSC_SUCCESS)))
1115: 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)
1116: {
1117: PetscFunctionBegin;
1118: PetscCall(PetscLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, lnkdata, lnk));
1119: lnklvl[entry] = newval;
1120: PetscFunctionReturn(PETSC_SUCCESS);
1121: }
1123: /*
1124: Initialize a sorted linked list used for ILU and ICC
1125: Input Parameters:
1126: nidx - number of input idx
1127: idx - integer array used for storing column indices
1128: idx_start - starting index of the list
1129: perm - indices of an IS
1130: lnk - linked list(an integer array) that is created
1131: lnklvl - levels of lnk
1132: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1133: output Parameters:
1134: nlnk - number of newly added idx
1135: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1136: lnklvl - levels of lnk
1137: bt - updated PetscBT (bitarray)
1138: */
1139: 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)
1140: {
1141: PetscFunctionBegin;
1142: *nlnk = 0;
1143: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1144: const PetscInt entry = perm[idx[k]];
1146: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscIncompleteLLInsertLocation_Private(PETSC_FALSE, k, idx_start, entry, nlnk, &lnkdata, lnk, lnklvl, 0));
1147: }
1148: PetscFunctionReturn(PETSC_SUCCESS);
1149: }
1151: static inline PetscErrorCode PetscIncompleteLLAdd_Private(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscReal 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)
1152: {
1153: PetscFunctionBegin;
1154: *nlnk = 0;
1155: for (PetscInt k = 0, lnkdata = idx_start; k < nidx; ++k) {
1156: const PetscInt incrlev = idxlvl[k] + prow_offset + 1;
1158: if (incrlev <= level) {
1159: const PetscInt entry = idx[k];
1161: if (!PetscBTLookupSet(bt, entry)) PetscCall(PetscIncompleteLLInsertLocation_Private(assume_sorted, k, idx_start, entry, nlnk, &lnkdata, lnk, lnklvl, incrlev));
1162: else if (lnklvl[entry] > incrlev) lnklvl[entry] = incrlev; /* existing entry */
1163: }
1164: }
1165: PetscFunctionReturn(PETSC_SUCCESS);
1166: }
1168: /*
1169: Add a SORTED index set into a sorted linked list for ICC
1170: Input Parameters:
1171: nidx - number of input indices
1172: idx - sorted integer array used for storing column indices
1173: level - level of fill, e.g., ICC(level)
1174: idxlvl - level of idx
1175: idx_start - starting index of the list
1176: lnk - linked list(an integer array) that is created
1177: lnklvl - levels of lnk
1178: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1179: idxlvl_prow - idxlvl[prow], where prow is the row number of the idx
1180: output Parameters:
1181: nlnk - number of newly added indices
1182: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1183: lnklvl - levels of lnk
1184: bt - updated PetscBT (bitarray)
1185: Note: the level of U(i,j) is set as lvl(i,j) = min{ lvl(i,j), lvl(prow,i)+lvl(prow,j)+1)
1186: where idx = non-zero columns of U(prow,prow+1:n-1), prow<i
1187: */
1188: static inline PetscErrorCode PetscICCLLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscReal 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)
1189: {
1190: PetscFunctionBegin;
1191: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, idxlvl_prow, PETSC_TRUE));
1192: PetscFunctionReturn(PETSC_SUCCESS);
1193: }
1195: /*
1196: Add a SORTED index set into a sorted linked list for ILU
1197: Input Parameters:
1198: nidx - number of input indices
1199: idx - sorted integer array used for storing column indices
1200: level - level of fill, e.g., ICC(level)
1201: idxlvl - level of idx
1202: idx_start - starting index of the list
1203: lnk - linked list(an integer array) that is created
1204: lnklvl - levels of lnk
1205: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1206: prow - the row number of idx
1207: output Parameters:
1208: nlnk - number of newly added idx
1209: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1210: lnklvl - levels of lnk
1211: bt - updated PetscBT (bitarray)
1213: Note: the level of factor(i,j) is set as lvl(i,j) = min{ lvl(i,j), lvl(i,prow)+lvl(prow,j)+1)
1214: where idx = non-zero columns of U(prow,prow+1:n-1), prow<i
1215: */
1216: 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)
1217: {
1218: PetscFunctionBegin;
1219: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, lnklvl[prow], PETSC_TRUE));
1220: PetscFunctionReturn(PETSC_SUCCESS);
1221: }
1223: /*
1224: Add a index set into a sorted linked list
1225: Input Parameters:
1226: nidx - number of input idx
1227: idx - integer array used for storing column indices
1228: level - level of fill, e.g., ICC(level)
1229: idxlvl - level of idx
1230: idx_start - starting index of the list
1231: lnk - linked list(an integer array) that is created
1232: lnklvl - levels of lnk
1233: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1234: output Parameters:
1235: nlnk - number of newly added idx
1236: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1237: lnklvl - levels of lnk
1238: bt - updated PetscBT (bitarray)
1239: */
1240: static inline PetscErrorCode PetscIncompleteLLAdd(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscReal level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt)
1241: {
1242: PetscFunctionBegin;
1243: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, 0, PETSC_FALSE));
1244: PetscFunctionReturn(PETSC_SUCCESS);
1245: }
1247: /*
1248: Add a SORTED index set into a sorted linked list
1249: Input Parameters:
1250: nidx - number of input indices
1251: idx - sorted integer array used for storing column indices
1252: level - level of fill, e.g., ICC(level)
1253: idxlvl - level of idx
1254: idx_start - starting index of the list
1255: lnk - linked list(an integer array) that is created
1256: lnklvl - levels of lnk
1257: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1258: output Parameters:
1259: nlnk - number of newly added idx
1260: lnk - the sorted(increasing order) linked list containing new and non-redundate entries from idx
1261: lnklvl - levels of lnk
1262: bt - updated PetscBT (bitarray)
1263: */
1264: static inline PetscErrorCode PetscIncompleteLLAddSorted(PetscInt nidx, const PetscInt *PETSC_RESTRICT idx, PetscReal level, const PetscInt *PETSC_RESTRICT idxlvl, PetscInt idx_start, PetscInt *PETSC_RESTRICT nlnk, PetscInt *PETSC_RESTRICT lnk, PetscInt *PETSC_RESTRICT lnklvl, PetscBT bt)
1265: {
1266: PetscFunctionBegin;
1267: PetscCall(PetscIncompleteLLAdd_Private(nidx, idx, level, idxlvl, idx_start, nlnk, lnk, lnklvl, bt, 0, PETSC_TRUE));
1268: PetscFunctionReturn(PETSC_SUCCESS);
1269: }
1271: /*
1272: Copy data on the list into an array, then initialize the list
1273: Input Parameters:
1274: idx_start - starting index of the list
1275: lnk_max - max value of lnk indicating the end of the list
1276: nlnk - number of data on the list to be copied
1277: lnk - linked list
1278: lnklvl - level of lnk
1279: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1280: output Parameters:
1281: indices - array that contains the copied data
1282: lnk - linked list that is cleaned and initialize
1283: lnklvl - level of lnk that is reinitialized
1284: bt - PetscBT (bitarray) with all bits set to false
1285: */
1286: 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)
1287: {
1288: PetscFunctionBegin;
1289: for (PetscInt j = 0, idx = idx_start; j < nlnk; ++j) {
1290: idx = lnk[idx];
1291: indices[j] = idx;
1292: indiceslvl[j] = lnklvl[idx];
1293: lnklvl[idx] = -1;
1294: PetscCall(PetscBTClear(bt, idx));
1295: }
1296: lnk[idx_start] = lnk_max;
1297: PetscFunctionReturn(PETSC_SUCCESS);
1298: }
1300: /*
1301: Free memories used by the list
1302: */
1303: #define PetscIncompleteLLDestroy(lnk, bt) ((PetscErrorCode)(PetscFree(lnk) || PetscBTDestroy(&(bt))))
1305: #if !defined(PETSC_CLANG_STATIC_ANALYZER)
1306: #define MatCheckSameLocalSize(A, ar1, B, ar2) \
1307: do { \
1308: PetscCheckSameComm(A, ar1, B, ar2); \
1309: 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, \
1310: (A)->rmap->n, (A)->cmap->n, ar2, (B)->rmap->n, (B)->cmap->n); \
1311: } while (0)
1312: #define MatCheckSameSize(A, ar1, B, ar2) \
1313: do { \
1314: 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, \
1315: (A)->rmap->N, (A)->cmap->N, ar2, (B)->rmap->N, (B)->cmap->N); \
1316: MatCheckSameLocalSize(A, ar1, B, ar2); \
1317: } while (0)
1318: #else
1319: template <typename Tm>
1320: extern void MatCheckSameLocalSize(Tm, int, Tm, int);
1321: template <typename Tm>
1322: extern void MatCheckSameSize(Tm, int, Tm, int);
1323: #endif
1325: #define VecCheckMatCompatible(M, x, ar1, b, ar2) \
1326: do { \
1327: 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, \
1328: (M)->cmap->N); \
1329: 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, \
1330: (M)->rmap->N); \
1331: } while (0)
1333: /*
1334: Create and initialize a condensed linked list -
1335: same as PetscLLCreate(), but uses a scalable array 'lnk' with size of max number of entries, not O(N).
1336: Barry suggested this approach (Dec. 6, 2011):
1337: I've thought of an alternative way of representing a linked list that is efficient but doesn't have the O(N) scaling issue
1338: (it may be faster than the O(N) even sequentially due to less crazy memory access).
1340: 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
1341: 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
1342: 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
1343: 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.
1344: 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
1345: to each other so memory access is much better than using the big array.
1347: Example:
1348: nlnk_max=5, lnk_max=36:
1349: Initial list: [0, 0 | 36, 2 | 0, 0 | 0, 0 | 0, 0 | 0, 0 | 0, 0]
1350: here, head_node has index 2 with value lnk[2]=lnk_max=36,
1351: 0-th entry is used to store the number of entries in the list,
1352: The initial lnk represents head -> tail(marked by 36) with number of entries = lnk[0]=0.
1354: Now adding a sorted set {2,4}, the list becomes
1355: [2, 0 | 36, 4 |2, 6 | 4, 2 | 0, 0 | 0, 0 | 0, 0 ]
1356: represents head -> 2 -> 4 -> tail with number of entries = lnk[0]=2.
1358: Then adding a sorted set {0,3,35}, the list
1359: [5, 0 | 36, 8 | 2, 10 | 4, 12 | 0, 4 | 3, 6 | 35, 2 ]
1360: represents head -> 0 -> 2 -> 3 -> 4 -> 35 -> tail with number of entries = lnk[0]=5.
1362: Input Parameters:
1363: nlnk_max - max length of the list
1364: lnk_max - max value of the entries
1365: Output Parameters:
1366: lnk - list created and initialized
1367: bt - PetscBT (bitarray) with all bits set to false. Note: bt has size lnk_max, not nln_max!
1368: */
1369: static inline PetscErrorCode PetscLLCondensedCreate(PetscInt nlnk_max, PetscInt lnk_max, PetscInt **lnk, PetscBT *bt)
1370: {
1371: PetscInt *llnk, lsize = 0;
1373: PetscFunctionBegin;
1374: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1375: PetscCall(PetscMalloc1(lsize, lnk));
1376: PetscCall(PetscBTCreate(lnk_max, bt));
1377: llnk = *lnk;
1378: llnk[0] = 0; /* number of entries on the list */
1379: llnk[2] = lnk_max; /* value in the head node */
1380: llnk[3] = 2; /* next for the head node */
1381: PetscFunctionReturn(PETSC_SUCCESS);
1382: }
1384: /*
1385: Add a SORTED ascending index set into a sorted linked list. See PetscLLCondensedCreate() for detailed description.
1386: Input Parameters:
1387: nidx - number of input indices
1388: indices - sorted integer array
1389: lnk - condensed linked list(an integer array) that is created
1390: bt - PetscBT (bitarray), bt[idx]=true marks idx is in lnk
1391: output Parameters:
1392: lnk - the sorted(increasing order) linked list containing previous and newly added non-redundate indices
1393: bt - updated PetscBT (bitarray)
1394: */
1395: static inline PetscErrorCode PetscLLCondensedAddSorted(PetscInt nidx, const PetscInt indices[], PetscInt lnk[], PetscBT bt)
1396: {
1397: PetscInt location = 2; /* head */
1398: PetscInt nlnk = lnk[0]; /* num of entries on the input lnk */
1400: PetscFunctionBegin;
1401: for (PetscInt k = 0; k < nidx; k++) {
1402: const PetscInt entry = indices[k];
1403: if (!PetscBTLookupSet(bt, entry)) { /* new entry */
1404: PetscInt next, lnkdata;
1406: /* search for insertion location */
1407: do {
1408: next = location + 1; /* link from previous node to next node */
1409: location = lnk[next]; /* idx of next node */
1410: lnkdata = lnk[location]; /* value of next node */
1411: } while (entry > lnkdata);
1412: /* insertion location is found, add entry into lnk */
1413: const PetscInt newnode = 2 * (nlnk + 2); /* index for this new node */
1414: lnk[next] = newnode; /* connect previous node to the new node */
1415: lnk[newnode] = entry; /* set value of the new node */
1416: lnk[newnode + 1] = location; /* connect new node to next node */
1417: location = newnode; /* next search starts from the new node */
1418: nlnk++;
1419: }
1420: }
1421: lnk[0] = nlnk; /* number of entries in the list */
1422: PetscFunctionReturn(PETSC_SUCCESS);
1423: }
1425: static inline PetscErrorCode PetscLLCondensedClean(PetscInt lnk_max, PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt lnk[], PetscBT bt)
1426: {
1427: const PetscInt nlnk = lnk[0]; /* num of entries on the list */
1428: PetscInt next = lnk[3]; /* head node */
1430: PetscFunctionBegin;
1431: for (PetscInt k = 0; k < nlnk; k++) {
1432: indices[k] = lnk[next];
1433: next = lnk[next + 1];
1434: PetscCall(PetscBTClear(bt, indices[k]));
1435: }
1436: lnk[0] = 0; /* num of entries on the list */
1437: lnk[2] = lnk_max; /* initialize head node */
1438: lnk[3] = 2; /* head node */
1439: PetscFunctionReturn(PETSC_SUCCESS);
1440: }
1442: static inline PetscErrorCode PetscLLCondensedView(PetscInt *lnk)
1443: {
1444: PetscFunctionBegin;
1445: PetscCall(PetscPrintf(PETSC_COMM_SELF, "LLCondensed of size %" PetscInt_FMT ", (val, next)\n", lnk[0]));
1446: 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]));
1447: PetscFunctionReturn(PETSC_SUCCESS);
1448: }
1450: /*
1451: Free memories used by the list
1452: */
1453: static inline PetscErrorCode PetscLLCondensedDestroy(PetscInt *lnk, PetscBT bt)
1454: {
1455: PetscFunctionBegin;
1456: PetscCall(PetscFree(lnk));
1457: PetscCall(PetscBTDestroy(&bt));
1458: PetscFunctionReturn(PETSC_SUCCESS);
1459: }
1461: /*
1462: Same as PetscLLCondensedCreate(), but does not use non-scalable O(lnk_max) bitarray
1463: Input Parameters:
1464: nlnk_max - max length of the list
1465: Output Parameters:
1466: lnk - list created and initialized
1467: */
1468: static inline PetscErrorCode PetscLLCondensedCreate_Scalable(PetscInt nlnk_max, PetscInt **lnk)
1469: {
1470: PetscInt *llnk, lsize = 0;
1472: PetscFunctionBegin;
1473: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1474: PetscCall(PetscMalloc1(lsize, lnk));
1475: llnk = *lnk;
1476: llnk[0] = 0; /* number of entries on the list */
1477: llnk[2] = PETSC_INT_MAX; /* value in the head node */
1478: llnk[3] = 2; /* next for the head node */
1479: PetscFunctionReturn(PETSC_SUCCESS);
1480: }
1482: static inline PetscErrorCode PetscLLCondensedExpand_Scalable(PetscInt nlnk_max, PetscInt **lnk)
1483: {
1484: PetscInt lsize = 0;
1486: PetscFunctionBegin;
1487: PetscCall(PetscIntMultError(2, nlnk_max + 2, &lsize));
1488: PetscCall(PetscRealloc(lsize * sizeof(PetscInt), lnk));
1489: PetscFunctionReturn(PETSC_SUCCESS);
1490: }
1492: static inline PetscErrorCode PetscLLCondensedAddSorted_Scalable(PetscInt nidx, const PetscInt indices[], PetscInt lnk[])
1493: {
1494: PetscInt location = 2; /* head */
1495: PetscInt nlnk = lnk[0]; /* num of entries on the input lnk */
1497: for (PetscInt k = 0; k < nidx; k++) {
1498: const PetscInt entry = indices[k];
1499: PetscInt next, lnkdata;
1501: /* search for insertion location */
1502: do {
1503: next = location + 1; /* link from previous node to next node */
1504: location = lnk[next]; /* idx of next node */
1505: lnkdata = lnk[location]; /* value of next node */
1506: } while (entry > lnkdata);
1507: if (entry < lnkdata) {
1508: /* insertion location is found, add entry into lnk */
1509: const PetscInt newnode = 2 * (nlnk + 2); /* index for this new node */
1510: lnk[next] = newnode; /* connect previous node to the new node */
1511: lnk[newnode] = entry; /* set value of the new node */
1512: lnk[newnode + 1] = location; /* connect new node to next node */
1513: location = newnode; /* next search starts from the new node */
1514: nlnk++;
1515: }
1516: }
1517: lnk[0] = nlnk; /* number of entries in the list */
1518: return PETSC_SUCCESS;
1519: }
1521: static inline PetscErrorCode PetscLLCondensedClean_Scalable(PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt *lnk)
1522: {
1523: const PetscInt nlnk = lnk[0];
1524: PetscInt next = lnk[3]; /* head node */
1526: for (PetscInt k = 0; k < nlnk; k++) {
1527: indices[k] = lnk[next];
1528: next = lnk[next + 1];
1529: }
1530: lnk[0] = 0; /* num of entries on the list */
1531: lnk[3] = 2; /* head node */
1532: return PETSC_SUCCESS;
1533: }
1535: static inline PetscErrorCode PetscLLCondensedDestroy_Scalable(PetscInt *lnk)
1536: {
1537: return PetscFree(lnk);
1538: }
1540: /*
1541: lnk[0] number of links
1542: lnk[1] number of entries
1543: lnk[3n] value
1544: lnk[3n+1] len
1545: lnk[3n+2] link to next value
1547: The next three are always the first link
1549: lnk[3] PETSC_INT_MIN+1
1550: lnk[4] 1
1551: lnk[5] link to first real entry
1553: The next three are always the last link
1555: lnk[6] PETSC_INT_MAX - 1
1556: lnk[7] 1
1557: lnk[8] next valid link (this is the same as lnk[0] but without the decreases)
1558: */
1560: static inline PetscErrorCode PetscLLCondensedCreate_fast(PetscInt nlnk_max, PetscInt **lnk)
1561: {
1562: PetscInt *llnk;
1563: PetscInt lsize = 0;
1565: PetscFunctionBegin;
1566: PetscCall(PetscIntMultError(3, nlnk_max + 3, &lsize));
1567: PetscCall(PetscMalloc1(lsize, lnk));
1568: llnk = *lnk;
1569: llnk[0] = 0; /* nlnk: number of entries on the list */
1570: llnk[1] = 0; /* number of integer entries represented in list */
1571: llnk[3] = PETSC_INT_MIN + 1; /* value in the first node */
1572: llnk[4] = 1; /* count for the first node */
1573: llnk[5] = 6; /* next for the first node */
1574: llnk[6] = PETSC_INT_MAX - 1; /* value in the last node */
1575: llnk[7] = 1; /* count for the last node */
1576: llnk[8] = 0; /* next valid node to be used */
1577: PetscFunctionReturn(PETSC_SUCCESS);
1578: }
1580: static inline PetscErrorCode PetscLLCondensedAddSorted_fast(PetscInt nidx, const PetscInt indices[], PetscInt lnk[])
1581: {
1582: for (PetscInt k = 0, prev = 3 /* first value */; k < nidx; k++) {
1583: const PetscInt entry = indices[k];
1584: PetscInt next = lnk[prev + 2];
1586: /* search for insertion location */
1587: while (entry >= lnk[next]) {
1588: prev = next;
1589: next = lnk[next + 2];
1590: }
1591: /* entry is in range of previous list */
1592: if (entry < lnk[prev] + lnk[prev + 1]) continue;
1593: lnk[1]++;
1594: /* entry is right after previous list */
1595: if (entry == lnk[prev] + lnk[prev + 1]) {
1596: lnk[prev + 1]++;
1597: if (lnk[next] == entry + 1) { /* combine two contiguous strings */
1598: lnk[prev + 1] += lnk[next + 1];
1599: lnk[prev + 2] = lnk[next + 2];
1600: next = lnk[next + 2];
1601: lnk[0]--;
1602: }
1603: continue;
1604: }
1605: /* entry is right before next list */
1606: if (entry == lnk[next] - 1) {
1607: lnk[next]--;
1608: lnk[next + 1]++;
1609: prev = next;
1610: next = lnk[prev + 2];
1611: continue;
1612: }
1613: /* add entry into lnk */
1614: lnk[prev + 2] = 3 * ((lnk[8]++) + 3); /* connect previous node to the new node */
1615: prev = lnk[prev + 2];
1616: lnk[prev] = entry; /* set value of the new node */
1617: lnk[prev + 1] = 1; /* number of values in contiguous string is one to start */
1618: lnk[prev + 2] = next; /* connect new node to next node */
1619: lnk[0]++;
1620: }
1621: return PETSC_SUCCESS;
1622: }
1624: static inline PetscErrorCode PetscLLCondensedClean_fast(PETSC_UNUSED PetscInt nidx, PetscInt *indices, PetscInt *lnk)
1625: {
1626: const PetscInt nlnk = lnk[0];
1627: PetscInt next = lnk[5]; /* first node */
1629: for (PetscInt k = 0, cnt = 0; k < nlnk; k++) {
1630: for (PetscInt j = 0; j < lnk[next + 1]; j++) indices[cnt++] = lnk[next] + j;
1631: next = lnk[next + 2];
1632: }
1633: lnk[0] = 0; /* nlnk: number of links */
1634: lnk[1] = 0; /* number of integer entries represented in list */
1635: lnk[3] = PETSC_INT_MIN + 1; /* value in the first node */
1636: lnk[4] = 1; /* count for the first node */
1637: lnk[5] = 6; /* next for the first node */
1638: lnk[6] = PETSC_INT_MAX - 1; /* value in the last node */
1639: lnk[7] = 1; /* count for the last node */
1640: lnk[8] = 0; /* next valid location to make link */
1641: return PETSC_SUCCESS;
1642: }
1644: static inline PetscErrorCode PetscLLCondensedView_fast(const PetscInt *lnk)
1645: {
1646: const PetscInt nlnk = lnk[0];
1647: PetscInt next = lnk[5]; /* first node */
1649: for (PetscInt k = 0; k < nlnk; k++) {
1650: #if 0 /* Debugging code */
1651: printf("%d value %d len %d next %d\n", next, lnk[next], lnk[next + 1], lnk[next + 2]);
1652: #endif
1653: next = lnk[next + 2];
1654: }
1655: return PETSC_SUCCESS;
1656: }
1658: static inline PetscErrorCode PetscLLCondensedDestroy_fast(PetscInt *lnk)
1659: {
1660: return PetscFree(lnk);
1661: }
1663: PETSC_EXTERN PetscErrorCode PetscCDCreate(PetscInt, PetscCoarsenData **);
1664: PETSC_EXTERN PetscErrorCode PetscCDDestroy(PetscCoarsenData *);
1665: PETSC_EXTERN PetscErrorCode PetscCDIntNdSetID(PetscCDIntNd *, PetscInt);
1666: PETSC_EXTERN PetscErrorCode PetscCDIntNdGetID(const PetscCDIntNd *, PetscInt *);
1667: PETSC_EXTERN PetscErrorCode PetscCDAppendID(PetscCoarsenData *, PetscInt, PetscInt);
1668: PETSC_EXTERN PetscErrorCode PetscCDMoveAppend(PetscCoarsenData *, PetscInt, PetscInt);
1669: PETSC_EXTERN PetscErrorCode PetscCDAppendNode(PetscCoarsenData *, PetscInt, PetscCDIntNd *);
1670: PETSC_EXTERN PetscErrorCode PetscCDRemoveNextNode(PetscCoarsenData *, PetscInt, PetscCDIntNd *);
1671: PETSC_EXTERN PetscErrorCode PetscCDCountAt(const PetscCoarsenData *, PetscInt, PetscInt *);
1672: PETSC_EXTERN PetscErrorCode PetscCDIsEmptyAt(const PetscCoarsenData *, PetscInt, PetscBool *);
1673: PETSC_EXTERN PetscErrorCode PetscCDSetChunkSize(PetscCoarsenData *, PetscInt);
1674: PETSC_EXTERN PetscErrorCode PetscCDPrint(const PetscCoarsenData *, PetscInt, MPI_Comm);
1675: PETSC_EXTERN PetscErrorCode PetscCDGetNonemptyIS(PetscCoarsenData *, IS *);
1676: PETSC_EXTERN PetscErrorCode PetscCDGetMat(PetscCoarsenData *, Mat *);
1677: PETSC_EXTERN PetscErrorCode PetscCDSetMat(PetscCoarsenData *, Mat);
1678: PETSC_EXTERN PetscErrorCode PetscCDClearMat(PetscCoarsenData *);
1679: PETSC_EXTERN PetscErrorCode PetscCDRemoveAllAt(PetscCoarsenData *, PetscInt);
1680: PETSC_EXTERN PetscErrorCode PetscCDCount(const PetscCoarsenData *, PetscInt *_sz);
1682: PETSC_EXTERN PetscErrorCode PetscCDGetHeadPos(const PetscCoarsenData *, PetscInt, PetscCDIntNd **);
1683: PETSC_EXTERN PetscErrorCode PetscCDGetNextPos(const PetscCoarsenData *, PetscInt, PetscCDIntNd **);
1684: PETSC_EXTERN PetscErrorCode PetscCDGetASMBlocks(const PetscCoarsenData *, const PetscInt, PetscInt *, IS **);
1686: PETSC_SINGLE_LIBRARY_VISIBILITY_INTERNAL PetscErrorCode MatFDColoringApply_AIJ(Mat, MatFDColoring, Vec, void *);
1688: typedef struct {
1689: Vec diag;
1690: PetscBool diag_valid;
1691: Vec inv_diag;
1692: PetscBool inv_diag_valid;
1693: PetscObjectState diag_state, inv_diag_state;
1694: PetscInt *col;
1695: PetscScalar *val;
1696: } Mat_Diagonal;
1698: #if PetscDefined(HAVE_CUDA)
1699: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqCUDA(Mat, Vec, Vec, PetscScalar *);
1700: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqCUDA(Mat, Vec, PetscReal *);
1701: #endif
1702: #if PetscDefined(HAVE_HIP)
1703: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqHIP(Mat, Vec, Vec, PetscScalar *);
1704: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqHIP(Mat, Vec, PetscReal *);
1705: #endif
1706: #if PetscDefined(HAVE_KOKKOS_KERNELS)
1707: PETSC_INTERN PetscErrorCode MatADot_Diagonal_SeqKokkos(Mat, Vec, Vec, PetscScalar *);
1708: PETSC_INTERN PetscErrorCode MatANormSq_Diagonal_SeqKokkos(Mat, Vec, PetscReal *);
1709: #endif
1711: PETSC_EXTERN PetscLogEvent MAT_Mult;
1712: PETSC_EXTERN PetscLogEvent MAT_MultAdd;
1713: PETSC_EXTERN PetscLogEvent MAT_MultTranspose;
1714: PETSC_EXTERN PetscLogEvent MAT_MultHermitianTranspose;
1715: PETSC_EXTERN PetscLogEvent MAT_MultTransposeAdd;
1716: PETSC_EXTERN PetscLogEvent MAT_MultHermitianTransposeAdd;
1717: PETSC_EXTERN PetscLogEvent MAT_ADot;
1718: PETSC_EXTERN PetscLogEvent MAT_ANorm;
1719: PETSC_EXTERN PetscLogEvent MAT_Solve;
1720: PETSC_EXTERN PetscLogEvent MAT_Solves;
1721: PETSC_EXTERN PetscLogEvent MAT_SolveAdd;
1722: PETSC_EXTERN PetscLogEvent MAT_SolveTranspose;
1723: PETSC_EXTERN PetscLogEvent MAT_SolveTransposeAdd;
1724: PETSC_EXTERN PetscLogEvent MAT_SOR;
1725: PETSC_EXTERN PetscLogEvent MAT_ForwardSolve;
1726: PETSC_EXTERN PetscLogEvent MAT_BackwardSolve;
1727: PETSC_EXTERN PetscLogEvent MAT_LUFactor;
1728: PETSC_EXTERN PetscLogEvent MAT_LUFactorSymbolic;
1729: PETSC_EXTERN PetscLogEvent MAT_LUFactorNumeric;
1730: PETSC_EXTERN PetscLogEvent MAT_QRFactor;
1731: PETSC_EXTERN PetscLogEvent MAT_QRFactorSymbolic;
1732: PETSC_EXTERN PetscLogEvent MAT_QRFactorNumeric;
1733: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactor;
1734: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactorSymbolic;
1735: PETSC_EXTERN PetscLogEvent MAT_CholeskyFactorNumeric;
1736: PETSC_EXTERN PetscLogEvent MAT_ILUFactor;
1737: PETSC_EXTERN PetscLogEvent MAT_ILUFactorSymbolic;
1738: PETSC_EXTERN PetscLogEvent MAT_ICCFactorSymbolic;
1739: PETSC_EXTERN PetscLogEvent MAT_Copy;
1740: PETSC_EXTERN PetscLogEvent MAT_Convert;
1741: PETSC_EXTERN PetscLogEvent MAT_Scale;
1742: PETSC_EXTERN PetscLogEvent MAT_AssemblyBegin;
1743: PETSC_EXTERN PetscLogEvent MAT_AssemblyEnd;
1744: PETSC_EXTERN PetscLogEvent MAT_SetValues;
1745: PETSC_EXTERN PetscLogEvent MAT_GetValues;
1746: PETSC_EXTERN PetscLogEvent MAT_GetRow;
1747: PETSC_EXTERN PetscLogEvent MAT_GetRowIJ;
1748: PETSC_EXTERN PetscLogEvent MAT_CreateSubMats;
1749: PETSC_EXTERN PetscLogEvent MAT_GetOrdering;
1750: PETSC_EXTERN PetscLogEvent MAT_RedundantMat;
1751: PETSC_EXTERN PetscLogEvent MAT_IncreaseOverlap;
1752: PETSC_EXTERN PetscLogEvent MAT_Partitioning;
1753: PETSC_EXTERN PetscLogEvent MAT_PartitioningND;
1754: PETSC_EXTERN PetscLogEvent MAT_Coarsen;
1755: PETSC_EXTERN PetscLogEvent MAT_ZeroEntries;
1756: PETSC_EXTERN PetscLogEvent MAT_Load;
1757: PETSC_EXTERN PetscLogEvent MAT_View;
1758: PETSC_EXTERN PetscLogEvent MAT_AXPY;
1759: PETSC_EXTERN PetscLogEvent MAT_FDColoringCreate;
1760: PETSC_EXTERN PetscLogEvent MAT_TransposeColoringCreate;
1761: PETSC_EXTERN PetscLogEvent MAT_FDColoringSetUp;
1762: PETSC_EXTERN PetscLogEvent MAT_FDColoringApply;
1763: PETSC_EXTERN PetscLogEvent MAT_Transpose;
1764: PETSC_EXTERN PetscLogEvent MAT_FDColoringFunction;
1765: PETSC_EXTERN PetscLogEvent MAT_CreateSubMat;
1766: PETSC_EXTERN PetscLogEvent MAT_MatSolve;
1767: PETSC_EXTERN PetscLogEvent MAT_MatTrSolve;
1768: PETSC_EXTERN PetscLogEvent MAT_MatMultSymbolic;
1769: PETSC_EXTERN PetscLogEvent MAT_MatMultNumeric;
1770: PETSC_EXTERN PetscLogEvent MAT_Getlocalmatcondensed;
1771: PETSC_EXTERN PetscLogEvent MAT_GetBrowsOfAcols;
1772: PETSC_EXTERN PetscLogEvent MAT_GetBrowsOfAocols;
1773: PETSC_EXTERN PetscLogEvent MAT_PtAPSymbolic;
1774: PETSC_EXTERN PetscLogEvent MAT_PtAPNumeric;
1775: PETSC_EXTERN PetscLogEvent MAT_Seqstompinum;
1776: PETSC_EXTERN PetscLogEvent MAT_Seqstompisym;
1777: PETSC_EXTERN PetscLogEvent MAT_Seqstompi;
1778: PETSC_EXTERN PetscLogEvent MAT_Getlocalmat;
1779: PETSC_EXTERN PetscLogEvent MAT_RARtSymbolic;
1780: PETSC_EXTERN PetscLogEvent MAT_RARtNumeric;
1781: PETSC_EXTERN PetscLogEvent MAT_MatTransposeMultSymbolic;
1782: PETSC_EXTERN PetscLogEvent MAT_MatTransposeMultNumeric;
1783: PETSC_EXTERN PetscLogEvent MAT_TransposeMatMultSymbolic;
1784: PETSC_EXTERN PetscLogEvent MAT_TransposeMatMultNumeric;
1785: PETSC_EXTERN PetscLogEvent MAT_MatMatMultSymbolic;
1786: PETSC_EXTERN PetscLogEvent MAT_MatMatMultNumeric;
1787: PETSC_EXTERN PetscLogEvent MAT_Getsymtransreduced;
1788: PETSC_EXTERN PetscLogEvent MAT_GetSeqNonzeroStructure;
1789: PETSC_EXTERN PetscLogEvent MATMFFD_Mult;
1790: PETSC_EXTERN PetscLogEvent MAT_GetMultiProcBlock;
1791: PETSC_EXTERN PetscLogEvent MAT_CUSPARSECopyToGPU;
1792: PETSC_EXTERN PetscLogEvent MAT_CUSPARSECopyFromGPU;
1793: PETSC_EXTERN PetscLogEvent MAT_CUSPARSEGenerateTranspose;
1794: PETSC_EXTERN PetscLogEvent MAT_CUSPARSESolveAnalysis;
1795: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSECopyToGPU;
1796: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSECopyFromGPU;
1797: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSEGenerateTranspose;
1798: PETSC_EXTERN PetscLogEvent MAT_HIPSPARSESolveAnalysis;
1799: PETSC_EXTERN PetscLogEvent MAT_SetValuesBatch;
1800: PETSC_EXTERN PetscLogEvent MAT_CreateGraph;
1801: PETSC_EXTERN PetscLogEvent MAT_ViennaCLCopyToGPU;
1802: PETSC_EXTERN PetscLogEvent MAT_DenseCopyToGPU;
1803: PETSC_EXTERN PetscLogEvent MAT_DenseCopyFromGPU;
1804: PETSC_EXTERN PetscLogEvent MAT_Merge;
1805: PETSC_EXTERN PetscLogEvent MAT_Residual;
1806: PETSC_EXTERN PetscLogEvent MAT_SetRandom;
1807: PETSC_EXTERN PetscLogEvent MAT_FactorFactS;
1808: PETSC_EXTERN PetscLogEvent MAT_FactorInvS;
1809: PETSC_EXTERN PetscLogEvent MAT_PreallCOO;
1810: PETSC_EXTERN PetscLogEvent MAT_SetVCOO;
1811: PETSC_EXTERN PetscLogEvent MATCOLORING_Apply;
1812: PETSC_EXTERN PetscLogEvent MATCOLORING_Comm;
1813: PETSC_EXTERN PetscLogEvent MATCOLORING_Local;
1814: PETSC_EXTERN PetscLogEvent MATCOLORING_ISCreate;
1815: PETSC_EXTERN PetscLogEvent MATCOLORING_SetUp;
1816: PETSC_EXTERN PetscLogEvent MATCOLORING_Weights;
1817: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Build;
1818: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Compress;
1819: PETSC_EXTERN PetscLogEvent MAT_H2Opus_Orthog;
1820: PETSC_EXTERN PetscLogEvent MAT_H2Opus_LR;
1821: PETSC_EXTERN PetscLogEvent MAT_CUDACopyToGPU;
1822: PETSC_EXTERN PetscLogEvent MAT_HIPCopyToGPU;
1824: #if defined(PETSC_CLANG_STATIC_ANALYZER)
1825: #define MatGetDiagonalMarkers(SeqXXX, bs)
1826: #else
1827: /*
1828: Adds diagonal pointers to sparse matrix nonzero structure and determines if all diagonal entries are present
1830: Rechecks the matrix data structure automatically if the nonzero structure of the matrix changed since the last call
1832: Potential optimization: since the a->j[j] are sorted this could use bisection to find the diagonal
1834: Developer Note:
1835: Uses the C preprocessor as a template mechanism to produce MatGetDiagonal_Seq[SB]AIJ() to avoid duplicate code
1836: */
1837: #define MatGetDiagonalMarkers(SeqXXX, bs) \
1838: PetscErrorCode MatGetDiagonalMarkers_##SeqXXX(Mat A, const PetscInt **diag, PetscBool *diagDense) \
1839: { \
1840: Mat_##SeqXXX *a = (Mat_##SeqXXX *)A->data; \
1841: \
1842: PetscFunctionBegin; \
1843: if (A->factortype != MAT_FACTOR_NONE) { \
1844: if (diagDense) *diagDense = PETSC_TRUE; \
1845: if (diag) *diag = a->diag; \
1846: PetscFunctionReturn(PETSC_SUCCESS); \
1847: } \
1848: PetscCheck(diag || diagDense, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "At least one of diag or diagDense must be requested"); \
1849: if (a->diagNonzeroState != A->nonzerostate || (diag && !a->diag)) { \
1850: const PetscInt m = A->rmap->n / bs; \
1851: \
1852: if (!diag && !a->diag) { \
1853: a->diagDense = PETSC_TRUE; \
1854: for (PetscInt i = 0; i < m; i++) { \
1855: PetscBool found = PETSC_FALSE; \
1856: \
1857: for (PetscInt j = a->i[i]; j < a->i[i + 1]; j++) { \
1858: if (a->j[j] == i) { \
1859: found = PETSC_TRUE; \
1860: break; \
1861: } \
1862: } \
1863: if (!found) { \
1864: a->diagDense = PETSC_FALSE; \
1865: *diagDense = a->diagDense; \
1866: a->diagNonzeroState = A->nonzerostate; \
1867: PetscFunctionReturn(PETSC_SUCCESS); \
1868: } \
1869: } \
1870: } else { \
1871: if (!a->diag) PetscCall(PetscMalloc1(m, &a->diag)); \
1872: a->diagDense = PETSC_TRUE; \
1873: for (PetscInt i = 0; i < m; i++) { \
1874: PetscBool found = PETSC_FALSE; \
1875: \
1876: a->diag[i] = a->i[i + 1]; \
1877: for (PetscInt j = a->i[i]; j < a->i[i + 1]; j++) { \
1878: if (a->j[j] == i) { \
1879: a->diag[i] = j; \
1880: found = PETSC_TRUE; \
1881: break; \
1882: } \
1883: } \
1884: if (!found) a->diagDense = PETSC_FALSE; \
1885: } \
1886: } \
1887: a->diagNonzeroState = A->nonzerostate; \
1888: } \
1889: if (diag) *diag = a->diag; \
1890: if (diagDense) *diagDense = a->diagDense; \
1891: PetscFunctionReturn(PETSC_SUCCESS); \
1892: }
1893: #endif