Actual source code: aij.h
1: #pragma once
3: #include <petsc/private/matimpl.h>
4: #include <petsc/private/hashmapi.h>
5: #include <petsc/private/hashmapijv.h>
7: /*
8: Used by MatCreateSubMatrices_MPIXAIJ_Local()
9: */
10: typedef struct { /* used by MatCreateSubMatrices_MPIAIJ_SingleIS_Local() and MatCreateSubMatrices_MPIAIJ_Local */
11: PetscInt id; /* index of submats, only submats[0] is responsible for deleting some arrays below */
12: PetscInt nrqs, nrqr;
13: PetscInt **rbuf1, **rbuf2, **rbuf3, **sbuf1, **sbuf2;
14: PetscInt **ptr;
15: PetscInt *tmp;
16: PetscInt *ctr;
17: PetscInt *pa; /* proc array */
18: PetscInt *req_size, *req_source1, *req_source2;
19: PetscBool allcolumns, allrows;
20: PetscBool singleis;
21: PetscInt *row2proc; /* row to proc map */
22: PetscInt nstages;
23: #if defined(PETSC_USE_CTABLE)
24: PetscHMapI cmap, rmap;
25: PetscInt *cmap_loc, *rmap_loc;
26: #else
27: PetscInt *cmap, *rmap;
28: #endif
29: PetscErrorCode (*destroy)(Mat);
30: } Mat_SubSppt;
32: /* Operations provided by MATSEQAIJ and its subclasses */
33: typedef struct {
34: PetscErrorCode (*getarray)(Mat, PetscScalar **);
35: PetscErrorCode (*restorearray)(Mat, PetscScalar **);
36: PetscErrorCode (*getarrayread)(Mat, const PetscScalar **);
37: PetscErrorCode (*restorearrayread)(Mat, const PetscScalar **);
38: PetscErrorCode (*getarraywrite)(Mat, PetscScalar **);
39: PetscErrorCode (*restorearraywrite)(Mat, PetscScalar **);
40: PetscErrorCode (*getcsrandmemtype)(Mat, const PetscInt **, const PetscInt **, PetscScalar **, PetscMemType *);
41: } Mat_SeqAIJOps;
43: /*
44: Struct header shared by SeqAIJ, SeqBAIJ and SeqSBAIJ matrix formats
45: */
46: #define SEQAIJHEADER(datatype) \
47: PetscBool roworiented; /* if true, row-oriented input, default */ \
48: PetscInt nonew; /* 1 don't add new nonzeros, -1 generate error on new */ \
49: PetscInt nounused; /* -1 generate error on unused space */ \
50: PetscInt maxnz; /* allocated nonzeros */ \
51: PetscInt *imax; /* maximum space allocated for each row */ \
52: PetscInt *ilen; /* actual length of each row */ \
53: PetscInt *ipre; /* space preallocated for each row by user */ \
54: PetscBool free_imax_ilen; \
55: PetscInt reallocs; /* number of mallocs done during MatSetValues() \
56: as more values are set than were prealloced */ \
57: PetscInt rmax; /* max nonzeros in any row */ \
58: PetscBool keepnonzeropattern; /* keeps matrix structure same in calls to MatZeroRows()*/ \
59: PetscBool ignorezeroentries; \
60: PetscBool free_ij; /* free the column indices j and row offsets i when the matrix is destroyed */ \
61: PetscBool free_a; /* free the numerical values when matrix is destroy */ \
62: Mat_CompressedRow compressedrow; /* use compressed row format */ \
63: PetscInt nz; /* nonzeros */ \
64: PetscInt *i; /* pointer to beginning of each row */ \
65: PetscInt *j; /* column values: j + i[k] - 1 is start of row k */ \
66: PetscInt *diag; /* pointers to diagonal elements */ \
67: PetscInt nonzerorowcnt; /* how many rows have nonzero entries */ \
68: PetscBool free_diag; \
69: datatype *a; /* nonzero elements */ \
70: PetscScalar *solve_work; /* work space used in MatSolve */ \
71: IS row, col, icol; /* index sets, used for reorderings */ \
72: PetscBool pivotinblocks; /* pivot inside factorization of each diagonal block */ \
73: Mat parent; /* set if this matrix was formed with MatDuplicate(...,MAT_SHARE_NONZERO_PATTERN,....); \
74: means that this shares some data structures with the parent including diag, ilen, imax, i, j */ \
75: Mat_SubSppt *submatis1; /* used by MatCreateSubMatrices_MPIXAIJ_Local */ \
76: Mat_SeqAIJOps ops[1] /* operations for SeqAIJ and its subclasses */
78: typedef struct {
79: MatTransposeColoring matcoloring;
80: Mat Bt_den; /* dense matrix of B^T */
81: Mat ABt_den; /* dense matrix of A*B^T */
82: PetscBool usecoloring;
83: } Mat_MatMatTransMult;
85: typedef struct { /* used by MatTransposeMatMult() */
86: Mat At; /* transpose of the first matrix */
87: Mat mA; /* maij matrix of A */
88: Vec bt, ct; /* vectors to hold locally transposed arrays of B and C */
89: /* used by PtAP */
90: void *data;
91: PetscErrorCode (*destroy)(void *);
92: } Mat_MatTransMatMult;
94: typedef struct {
95: PetscInt *api, *apj; /* symbolic structure of A*P */
96: PetscScalar *apa; /* temporary array for storing one row of A*P */
97: } Mat_AP;
99: typedef struct {
100: MatTransposeColoring matcoloring;
101: Mat Rt; /* sparse or dense matrix of R^T */
102: Mat RARt; /* dense matrix of R*A*R^T */
103: Mat ARt; /* A*R^T used for the case -matrart_color_art */
104: MatScalar *work; /* work array to store columns of A*R^T used in MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqDense() */
105: /* free intermediate products needed for PtAP */
106: void *data;
107: PetscErrorCode (*destroy)(void *);
108: } Mat_RARt;
110: typedef struct {
111: Mat BC; /* temp matrix for storing B*C */
112: } Mat_MatMatMatMult;
114: /*
115: MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix
116: format) or compressed sparse row (CSR). The i[] and j[] arrays start at 0. For example,
117: j[i[k]+p] is the pth column in row k. Note that the diagonal
118: matrix elements are stored with the rest of the nonzeros (not separately).
119: */
121: /* Info about i-nodes (identical nodes) helper class for SeqAIJ */
122: typedef struct {
123: MatScalar *bdiag, *ibdiag, *ssor_work; /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */
124: PetscInt bdiagsize; /* length of bdiag and ibdiag */
125: PetscBool ibdiagvalid; /* do ibdiag[] and bdiag[] contain the most recent values */
127: PetscBool use;
128: PetscInt node_count; /* number of inodes */
129: PetscInt *size; /* size of each inode */
130: PetscInt limit; /* inode limit */
131: PetscInt max_limit; /* maximum supported inode limit */
132: PetscBool checked; /* if inodes have been checked for */
133: PetscObjectState mat_nonzerostate; /* non-zero state when inodes were checked for */
134: } Mat_SeqAIJ_Inode;
136: PETSC_INTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat, PetscViewer);
137: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat, MatAssemblyType);
138: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat);
139: PETSC_INTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat);
140: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat, MatOption, PetscBool);
141: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat, MatDuplicateOption, Mat *);
142: PETSC_INTERN PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat, Mat, MatDuplicateOption, PetscBool);
143: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat, Mat, const MatFactorInfo *);
144: PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat, PetscScalar **);
145: PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat, PetscScalar **);
147: typedef struct {
148: SEQAIJHEADER(MatScalar);
149: Mat_SeqAIJ_Inode inode;
150: MatScalar *saved_values; /* location for stashing nonzero values of matrix */
152: PetscScalar *idiag, *mdiag, *ssor_work; /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */
153: PetscBool idiagvalid; /* current idiag[] and mdiag[] are valid */
154: PetscScalar *ibdiag; /* inverses of block diagonals */
155: PetscBool ibdiagvalid; /* inverses of block diagonals are valid. */
156: PetscBool diagonaldense; /* all entries along the diagonal have been set; i.e. no missing diagonal terms */
157: PetscScalar fshift, omega; /* last used omega and fshift */
159: /* MatSetValues() via hash related fields */
160: PetscHMapIJV ht;
161: PetscInt *dnz;
162: struct _MatOps cops;
163: } Mat_SeqAIJ;
165: typedef struct {
166: PetscInt nz; /* nz of the matrix after assembly */
167: PetscCount n; /* Number of entries in MatSetPreallocationCOO() */
168: PetscCount Atot; /* Total number of valid (i.e., w/ non-negative indices) entries in the COO array */
169: PetscCount *jmap; /* perm[jmap[i]..jmap[i+1]) give indices of entries in v[] associated with i-th nonzero of the matrix */
170: PetscCount *perm; /* The permutation array in sorting (i,j) by row and then by col */
171: } MatCOOStruct_SeqAIJ;
173: #define MatSeqXAIJGetOptions_Private(A) \
174: { \
175: const PetscBool oldvalues = (PetscBool)(A != PETSC_NULLPTR); \
176: PetscInt nonew = 0, nounused = 0; \
177: PetscBool roworiented = PETSC_FALSE; \
178: if (oldvalues) { \
179: nonew = ((Mat_SeqAIJ *)A->data)->nonew; \
180: nounused = ((Mat_SeqAIJ *)A->data)->nounused; \
181: roworiented = ((Mat_SeqAIJ *)A->data)->roworiented; \
182: } \
183: (void)0
185: #define MatSeqXAIJRestoreOptions_Private(A) \
186: if (oldvalues) { \
187: ((Mat_SeqAIJ *)A->data)->nonew = nonew; \
188: ((Mat_SeqAIJ *)A->data)->nounused = nounused; \
189: ((Mat_SeqAIJ *)A->data)->roworiented = roworiented; \
190: } \
191: } \
192: (void)0
194: static inline PetscErrorCode MatXAIJAllocatea(Mat A, PetscInt nz, PetscScalar **array)
195: {
196: Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data;
198: PetscFunctionBegin;
199: PetscCall(PetscShmgetAllocateArray(nz, sizeof(PetscScalar), (void **)array));
200: a->free_a = PETSC_TRUE;
201: PetscFunctionReturn(PETSC_SUCCESS);
202: }
204: static inline PetscErrorCode MatXAIJDeallocatea(Mat A, PetscScalar **array)
205: {
206: Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data;
208: PetscFunctionBegin;
209: if (a->free_a) PetscCall(PetscShmgetDeallocateArray((void **)array));
210: a->free_a = PETSC_FALSE;
211: PetscFunctionReturn(PETSC_SUCCESS);
212: }
214: /*
215: Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
216: */
217: static inline PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA, MatScalar **a, PetscInt **j, PetscInt **i)
218: {
219: Mat_SeqAIJ *A = (Mat_SeqAIJ *)AA->data;
221: PetscFunctionBegin;
222: if (A->free_a) PetscCall(PetscShmgetDeallocateArray((void **)a));
223: if (A->free_ij) PetscCall(PetscShmgetDeallocateArray((void **)j));
224: if (A->free_ij) PetscCall(PetscShmgetDeallocateArray((void **)i));
225: PetscFunctionReturn(PETSC_SUCCESS);
226: }
227: /*
228: Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
229: This is a macro because it takes the datatype as an argument which can be either a Mat or a MatScalar
230: */
231: #define MatSeqXAIJReallocateAIJ(Amat, AM, BS2, NROW, ROW, COL, RMAX, AA, AI, AJ, RP, AP, AIMAX, NONEW, datatype) \
232: do { \
233: if (NROW >= RMAX) { \
234: Mat_SeqAIJ *Ain = (Mat_SeqAIJ *)Amat->data; \
235: PetscInt CHUNKSIZE = 15, new_nz = AI[AM] + CHUNKSIZE, len, *new_i = NULL, *new_j = NULL; \
236: datatype *new_a; \
237: \
238: PetscCheck(NONEW != -2, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "New nonzero at (%" PetscInt_FMT ",%" PetscInt_FMT ") caused a malloc. Use MatSetOption(A, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE) to turn off this check", ROW, COL); \
239: /* malloc new storage space */ \
240: PetscCall(PetscShmgetAllocateArray(BS2 *new_nz, sizeof(PetscScalar), (void **)&new_a)); \
241: PetscCall(PetscShmgetAllocateArray(new_nz, sizeof(PetscInt), (void **)&new_j)); \
242: PetscCall(PetscShmgetAllocateArray(AM + 1, sizeof(PetscInt), (void **)&new_i)); \
243: Ain->free_a = PETSC_TRUE; \
244: Ain->free_ij = PETSC_TRUE; \
245: /* copy over old data into new slots */ \
246: for (ii = 0; ii < ROW + 1; ii++) new_i[ii] = AI[ii]; \
247: for (ii = ROW + 1; ii < AM + 1; ii++) new_i[ii] = AI[ii] + CHUNKSIZE; \
248: PetscCall(PetscArraycpy(new_j, AJ, AI[ROW] + NROW)); \
249: len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
250: PetscCall(PetscArraycpy(new_j + AI[ROW] + NROW + CHUNKSIZE, PetscSafePointerPlusOffset(AJ, AI[ROW] + NROW), len)); \
251: PetscCall(PetscArraycpy(new_a, AA, BS2 *(AI[ROW] + NROW))); \
252: PetscCall(PetscArrayzero(new_a + BS2 * (AI[ROW] + NROW), BS2 * CHUNKSIZE)); \
253: PetscCall(PetscArraycpy(new_a + BS2 * (AI[ROW] + NROW + CHUNKSIZE), PetscSafePointerPlusOffset(AA, BS2 * (AI[ROW] + NROW)), BS2 * len)); \
254: /* free up old matrix storage */ \
255: PetscCall(MatSeqXAIJFreeAIJ(A, &Ain->a, &Ain->j, &Ain->i)); \
256: AA = new_a; \
257: Ain->a = (MatScalar *)new_a; \
258: AI = Ain->i = new_i; \
259: AJ = Ain->j = new_j; \
260: \
261: RP = AJ + AI[ROW]; \
262: AP = AA + BS2 * AI[ROW]; \
263: RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
264: Ain->maxnz += BS2 * CHUNKSIZE; \
265: Ain->reallocs++; \
266: Amat->nonzerostate++; \
267: } \
268: } while (0)
270: #define MatSeqXAIJReallocateAIJ_structure_only(Amat, AM, BS2, NROW, ROW, COL, RMAX, AI, AJ, RP, AIMAX, NONEW, datatype) \
271: do { \
272: if (NROW >= RMAX) { \
273: Mat_SeqAIJ *Ain = (Mat_SeqAIJ *)Amat->data; \
274: /* there is no extra room in row, therefore enlarge */ \
275: PetscInt CHUNKSIZE = 15, new_nz = AI[AM] + CHUNKSIZE, len, *new_i = NULL, *new_j = NULL; \
276: \
277: PetscCheck(NONEW != -2, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "New nonzero at (%" PetscInt_FMT ",%" PetscInt_FMT ") caused a malloc. Use MatSetOption(A, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE) to turn off this check", ROW, COL); \
278: /* malloc new storage space */ \
279: PetscCall(PetscShmgetAllocateArray(new_nz, sizeof(PetscInt), (void **)&new_j)); \
280: PetscCall(PetscShmgetAllocateArray(AM + 1, sizeof(PetscInt), (void **)&new_i)); \
281: Ain->free_a = PETSC_FALSE; \
282: Ain->free_ij = PETSC_TRUE; \
283: \
284: /* copy over old data into new slots */ \
285: for (ii = 0; ii < ROW + 1; ii++) new_i[ii] = AI[ii]; \
286: for (ii = ROW + 1; ii < AM + 1; ii++) new_i[ii] = AI[ii] + CHUNKSIZE; \
287: PetscCall(PetscArraycpy(new_j, AJ, AI[ROW] + NROW)); \
288: len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
289: PetscCall(PetscArraycpy(new_j + AI[ROW] + NROW + CHUNKSIZE, AJ + AI[ROW] + NROW, len)); \
290: \
291: /* free up old matrix storage */ \
292: PetscCall(MatSeqXAIJFreeAIJ(A, &Ain->a, &Ain->j, &Ain->i)); \
293: Ain->a = NULL; \
294: AI = Ain->i = new_i; \
295: AJ = Ain->j = new_j; \
296: \
297: RP = AJ + AI[ROW]; \
298: RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
299: Ain->maxnz += BS2 * CHUNKSIZE; \
300: Ain->reallocs++; \
301: Amat->nonzerostate++; \
302: } \
303: } while (0)
305: PETSC_INTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat, PetscInt, const PetscInt *);
306: PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_SeqAIJ(Mat, PetscCount, PetscInt[], PetscInt[]);
308: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat, Mat, IS, IS, const MatFactorInfo *);
309: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat, Mat, IS, IS, const MatFactorInfo *);
311: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat, Mat, IS, const MatFactorInfo *);
312: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat, Mat, IS, const MatFactorInfo *);
313: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat, Mat, const MatFactorInfo *);
314: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat, Mat, const MatFactorInfo *);
315: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat, MatDuplicateOption, Mat *);
316: PETSC_INTERN PetscErrorCode MatCopy_SeqAIJ(Mat, Mat, MatStructure);
317: PETSC_INTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat, PetscBool *, PetscInt *);
318: PETSC_INTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat);
319: PETSC_INTERN PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat, PetscInt *, PetscInt **);
321: PETSC_INTERN PetscErrorCode MatMult_SeqAIJ(Mat, Vec, Vec);
322: PETSC_INTERN PetscErrorCode MatMult_SeqAIJ_Inode(Mat, Vec, Vec);
323: PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat, Vec, Vec, Vec);
324: PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ_Inode(Mat, Vec, Vec, Vec);
325: PETSC_INTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat, Vec, Vec);
326: PETSC_INTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat, Vec, Vec, Vec);
327: PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ(Mat, Vec, PetscReal, MatSORType, PetscReal, PetscInt, PetscInt, Vec);
328: PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ_Inode(Mat, Vec, PetscReal, MatSORType, PetscReal, PetscInt, PetscInt, Vec);
330: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ(Mat, MatOption, PetscBool);
332: PETSC_INTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat, PetscInt *[], PetscInt *[]);
333: PETSC_INTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat, PetscInt *[], PetscInt *[]);
334: PETSC_INTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat, PetscInt, PetscInt, PetscInt *[], PetscInt *[]);
335: PETSC_INTERN PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat, Mat *);
336: PETSC_INTERN PetscErrorCode MatTranspose_SeqAIJ(Mat, MatReuse, Mat *);
338: PETSC_INTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt, PetscInt *, PetscInt *, PetscBool, PetscInt, PetscInt, PetscInt **, PetscInt **);
339: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat, Mat, IS, IS, const MatFactorInfo *);
340: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat, Mat, const MatFactorInfo *);
341: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat, Mat, const MatFactorInfo *);
342: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat, Mat, const MatFactorInfo *);
343: PETSC_INTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat, IS, IS, const MatFactorInfo *);
344: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat, Vec, Vec);
345: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ(Mat, Vec, Vec);
346: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat, Vec, Vec);
347: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat, Vec, Vec);
348: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat, Vec, Vec, Vec);
349: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat, Vec, Vec);
350: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat, Vec, Vec);
351: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat, Vec, Vec, Vec);
352: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat, Vec, Vec, Vec);
353: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat, Mat, Mat);
354: PETSC_INTERN PetscErrorCode MatMatSolveTranspose_SeqAIJ(Mat, Mat, Mat);
355: PETSC_INTERN PetscErrorCode MatEqual_SeqAIJ(Mat, Mat, PetscBool *);
356: PETSC_INTERN PetscErrorCode MatFDColoringCreate_SeqXAIJ(Mat, ISColoring, MatFDColoring);
357: PETSC_INTERN PetscErrorCode MatFDColoringSetUp_SeqXAIJ(Mat, ISColoring, MatFDColoring);
358: PETSC_INTERN PetscErrorCode MatFDColoringSetUpBlocked_AIJ_Private(Mat, MatFDColoring, PetscInt);
359: PETSC_INTERN PetscErrorCode MatLoad_AIJ_HDF5(Mat, PetscViewer);
360: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ_Binary(Mat, PetscViewer);
361: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ(Mat, PetscViewer);
362: PETSC_INTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat);
364: #if defined(PETSC_HAVE_HYPRE)
365: PETSC_INTERN PetscErrorCode MatProductSetFromOptions_Transpose_AIJ_AIJ(Mat);
366: #endif
367: PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqAIJ(Mat);
369: PETSC_INTERN PetscErrorCode MatProductSymbolic_SeqAIJ_SeqAIJ(Mat);
370: PETSC_INTERN PetscErrorCode MatProductSymbolic_PtAP_SeqAIJ_SeqAIJ(Mat);
371: PETSC_INTERN PetscErrorCode MatProductSymbolic_RARt_SeqAIJ_SeqAIJ(Mat);
373: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat);
374: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(Mat, Mat, PetscReal, Mat);
375: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat, Mat, PetscReal, Mat);
376: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat, Mat, PetscReal, Mat);
377: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat, Mat, PetscReal, Mat);
378: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat, Mat, PetscReal, Mat);
379: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat, Mat, PetscReal, Mat);
380: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(Mat, Mat, PetscReal, Mat);
381: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat, Mat, PetscReal, Mat);
382: #if defined(PETSC_HAVE_HYPRE)
383: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat, Mat, PetscReal, Mat);
384: #endif
386: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat);
387: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(Mat, Mat, Mat);
389: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat, Mat, Mat);
390: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat, Mat, Mat);
392: PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy(Mat, Mat, PetscReal, Mat);
393: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat);
394: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy(Mat, Mat, Mat);
396: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat);
397: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_matmattransposemult(Mat, Mat, PetscReal, Mat);
398: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_colorrart(Mat, Mat, PetscReal, Mat);
399: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat);
400: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_matmattransposemult(Mat, Mat, Mat);
401: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_colorrart(Mat, Mat, Mat);
403: PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat);
404: PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat);
405: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_MatTransMatMult(void *);
407: PETSC_INTERN PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat);
408: PETSC_INTERN PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat);
409: PETSC_INTERN PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat, ISColoring, MatTransposeColoring);
410: PETSC_INTERN PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring, Mat, Mat);
411: PETSC_INTERN PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring, Mat, Mat);
413: PETSC_INTERN PetscErrorCode MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ(Mat, Mat, Mat, PetscReal, Mat);
414: PETSC_INTERN PetscErrorCode MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ(Mat, Mat, Mat, Mat);
416: PETSC_INTERN PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat, PetscInt, PetscInt, PetscRandom);
417: PETSC_INTERN PetscErrorCode MatSetValues_SeqAIJ(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], const PetscScalar[], InsertMode);
418: PETSC_INTERN PetscErrorCode MatGetRow_SeqAIJ(Mat, PetscInt, PetscInt *, PetscInt **, PetscScalar **);
419: PETSC_INTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat, PetscInt, PetscInt *, PetscInt **, PetscScalar **);
420: PETSC_INTERN PetscErrorCode MatScale_SeqAIJ(Mat, PetscScalar);
421: PETSC_INTERN PetscErrorCode MatDiagonalScale_SeqAIJ(Mat, Vec, Vec);
422: PETSC_INTERN PetscErrorCode MatDiagonalSet_SeqAIJ(Mat, Vec, InsertMode);
423: PETSC_INTERN PetscErrorCode MatAXPY_SeqAIJ(Mat, PetscScalar, Mat, MatStructure);
424: PETSC_INTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
425: PETSC_INTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
426: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
427: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *);
428: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscInt *[], PetscBool *);
429: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscInt *[], PetscBool *);
430: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat);
431: PETSC_INTERN PetscErrorCode MatView_SeqAIJ(Mat, PetscViewer);
433: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat);
434: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal_Inode(Mat);
435: PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode(Mat);
436: PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode_FactorLU(Mat);
438: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat, Mat, PetscInt *);
440: #if defined(PETSC_HAVE_MATLAB)
441: PETSC_EXTERN PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject, void *);
442: PETSC_EXTERN PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject, void *);
443: #endif
444: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat, MatType, MatReuse, Mat *);
445: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat, MatType, MatReuse, Mat *);
446: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqDense(Mat, MatType, MatReuse, Mat *);
447: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat, MatType, MatReuse, Mat *);
448: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat, MatType, MatReuse, Mat *);
449: #if defined(PETSC_HAVE_SCALAPACK)
450: PETSC_INTERN PetscErrorCode MatConvert_AIJ_ScaLAPACK(Mat, MatType, MatReuse, Mat *);
451: #endif
452: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat, MatType, MatReuse, Mat *);
453: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJPERM(Mat, MatType, MatReuse, Mat *);
454: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJSELL(Mat, MatType, MatReuse, Mat *);
455: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJMKL(Mat, MatType, MatReuse, Mat *);
456: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJViennaCL(Mat, MatType, MatReuse, Mat *);
457: PETSC_INTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat, PetscReal, IS, IS);
458: PETSC_INTERN PetscErrorCode MatRARt_SeqAIJ_SeqAIJ(Mat, Mat, MatReuse, PetscReal, Mat *);
459: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat);
460: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat, MatAssemblyType);
461: PETSC_EXTERN PetscErrorCode MatZeroEntries_SeqAIJ(Mat);
463: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt, const PetscInt *, const PetscInt *, const PetscInt *, const PetscInt *, PetscInt *);
464: PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm, Mat, PetscInt, MatReuse, Mat *);
465: PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIAIJ(MPI_Comm, Mat, PetscInt, MatReuse, Mat *);
467: PETSC_INTERN PetscErrorCode MatSetSeqMat_SeqAIJ(Mat, IS, IS, MatStructure, Mat);
468: PETSC_INTERN PetscErrorCode MatEliminateZeros_SeqAIJ(Mat, PetscBool);
469: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt *);
470: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat);
471: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Dummy(Mat);
472: PETSC_INTERN PetscErrorCode MatDestroySubMatrices_Dummy(PetscInt, Mat *[]);
473: PETSC_INTERN PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat, IS, IS, PetscInt, MatReuse, Mat *);
475: PETSC_INTERN PetscErrorCode MatSetSeqAIJWithArrays_private(MPI_Comm, PetscInt, PetscInt, PetscInt[], PetscInt[], PetscScalar[], MatType, Mat);
477: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat, ISLocalToGlobalMapping *);
479: /*
480: PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage
482: Input Parameters:
483: + nnz - the number of entries
484: . r - the array of vector values
485: . xv - the matrix values for the row
486: - xi - the column indices of the nonzeros in the row
488: Output Parameter:
489: . sum - negative the sum of results
491: PETSc compile flags:
492: + PETSC_KERNEL_USE_UNROLL_4
493: - PETSC_KERNEL_USE_UNROLL_2
495: Developer Note:
496: The macro changes sum but not other parameters
498: .seealso: `PetscSparseDensePlusDot()`
499: */
500: #if defined(PETSC_KERNEL_USE_UNROLL_4)
501: #define PetscSparseDenseMinusDot(sum, r, xv, xi, nnz) \
502: do { \
503: if (nnz > 0) { \
504: PetscInt nnz2 = nnz, rem = nnz & 0x3; \
505: switch (rem) { \
506: case 3: \
507: sum -= *xv++ * r[*xi++]; \
508: case 2: \
509: sum -= *xv++ * r[*xi++]; \
510: case 1: \
511: sum -= *xv++ * r[*xi++]; \
512: nnz2 -= rem; \
513: } \
514: while (nnz2 > 0) { \
515: sum -= xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
516: xv += 4; \
517: xi += 4; \
518: nnz2 -= 4; \
519: } \
520: xv -= nnz; \
521: xi -= nnz; \
522: } \
523: } while (0)
525: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
526: #define PetscSparseDenseMinusDot(sum, r, xv, xi, nnz) \
527: do { \
528: PetscInt __i, __i1, __i2; \
529: for (__i = 0; __i < nnz - 1; __i += 2) { \
530: __i1 = xi[__i]; \
531: __i2 = xi[__i + 1]; \
532: sum -= (xv[__i] * r[__i1] + xv[__i + 1] * r[__i2]); \
533: } \
534: if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]]; \
535: } while (0)
537: #else
538: #define PetscSparseDenseMinusDot(sum, r, xv, xi, nnz) \
539: do { \
540: PetscInt __i; \
541: for (__i = 0; __i < nnz; __i++) sum -= xv[__i] * r[xi[__i]]; \
542: } while (0)
543: #endif
545: /*
546: PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage
548: Input Parameters:
549: + nnz - the number of entries
550: . r - the array of vector values
551: . xv - the matrix values for the row
552: - xi - the column indices of the nonzeros in the row
554: Output Parameter:
555: . sum - the sum of results
557: PETSc compile flags:
558: + PETSC_KERNEL_USE_UNROLL_4
559: - PETSC_KERNEL_USE_UNROLL_2
561: Developer Note:
562: The macro changes sum but not other parameters
564: .seealso: `PetscSparseDenseMinusDot()`
565: */
566: #if defined(PETSC_KERNEL_USE_UNROLL_4)
567: #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) \
568: do { \
569: if (nnz > 0) { \
570: PetscInt nnz2 = nnz, rem = nnz & 0x3; \
571: switch (rem) { \
572: case 3: \
573: sum += *xv++ * r[*xi++]; \
574: case 2: \
575: sum += *xv++ * r[*xi++]; \
576: case 1: \
577: sum += *xv++ * r[*xi++]; \
578: nnz2 -= rem; \
579: } \
580: while (nnz2 > 0) { \
581: sum += xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
582: xv += 4; \
583: xi += 4; \
584: nnz2 -= 4; \
585: } \
586: xv -= nnz; \
587: xi -= nnz; \
588: } \
589: } while (0)
591: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
592: #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) \
593: do { \
594: PetscInt __i, __i1, __i2; \
595: for (__i = 0; __i < nnz - 1; __i += 2) { \
596: __i1 = xi[__i]; \
597: __i2 = xi[__i + 1]; \
598: sum += (xv[__i] * r[__i1] + xv[__i + 1] * r[__i2]); \
599: } \
600: if (nnz & 0x1) sum += xv[__i] * r[xi[__i]]; \
601: } while (0)
603: #elif !(defined(__GNUC__) && defined(_OPENMP)) && defined(PETSC_USE_AVX512_KERNELS) && defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX512F__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES) && !defined(PETSC_SKIP_IMMINTRIN_H_CUDAWORKAROUND)
604: #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) PetscSparseDensePlusDot_AVX512_Private(&(sum), (r), (xv), (xi), (nnz))
606: #else
607: #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) \
608: do { \
609: PetscInt __i; \
610: for (__i = 0; __i < nnz; __i++) sum += xv[__i] * r[xi[__i]]; \
611: } while (0)
612: #endif
614: #if defined(PETSC_USE_AVX512_KERNELS) && defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX512F__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES) && !defined(PETSC_SKIP_IMMINTRIN_H_CUDAWORKAROUND)
615: #include <immintrin.h>
616: #if !defined(_MM_SCALE_8)
617: #define _MM_SCALE_8 8
618: #endif
620: static inline void PetscSparseDensePlusDot_AVX512_Private(PetscScalar *sum, const PetscScalar *x, const MatScalar *aa, const PetscInt *aj, PetscInt n)
621: {
622: __m512d vec_x, vec_y, vec_vals;
623: __m256i vec_idx;
624: PetscInt j;
626: vec_y = _mm512_setzero_pd();
627: for (j = 0; j < (n >> 3); j++) {
628: vec_idx = _mm256_loadu_si256((__m256i const *)aj);
629: vec_vals = _mm512_loadu_pd(aa);
630: vec_x = _mm512_i32gather_pd(vec_idx, x, _MM_SCALE_8);
631: vec_y = _mm512_fmadd_pd(vec_x, vec_vals, vec_y);
632: aj += 8;
633: aa += 8;
634: }
635: #if defined(__AVX512VL__)
636: /* masked load requires avx512vl, which is not supported by KNL */
637: if (n & 0x07) {
638: __mmask8 mask;
639: mask = (__mmask8)(0xff >> (8 - (n & 0x07)));
640: vec_idx = _mm256_mask_loadu_epi32(vec_idx, mask, aj);
641: vec_vals = _mm512_mask_loadu_pd(vec_vals, mask, aa);
642: vec_x = _mm512_mask_i32gather_pd(vec_x, mask, vec_idx, x, _MM_SCALE_8);
643: vec_y = _mm512_mask3_fmadd_pd(vec_x, vec_vals, vec_y, mask);
644: }
645: *sum += _mm512_reduce_add_pd(vec_y);
646: #else
647: *sum += _mm512_reduce_add_pd(vec_y);
648: for (j = 0; j < (n & 0x07); j++) *sum += aa[j] * x[aj[j]];
649: #endif
650: }
651: #endif
653: /*
654: PetscSparseDenseMaxDot - The inner kernel of a modified matrix-vector product \max_i xv[i] * r[xi[i]] for CSR storage
656: Input Parameters:
657: + nnz - the number of entries
658: . r - the array of vector values
659: . xv - the matrix values for the row
660: - xi - the column indices of the nonzeros in the row
662: Output Parameter:
663: . max - the max of results
665: .seealso: `PetscSparseDensePlusDot()`, `PetscSparseDenseMinusDot()`
666: */
667: #define PetscSparseDenseMaxDot(max, r, xv, xi, nnz) \
668: do { \
669: for (PetscInt __i = 0; __i < (nnz); __i++) { max = PetscMax(PetscRealPart(max), PetscRealPart((xv)[__i] * (r)[(xi)[__i]])); } \
670: } while (0)
672: /*
673: Add column indices into table for counting the max nonzeros of merged rows
674: */
675: #define MatRowMergeMax_SeqAIJ(mat, nrows, ta) \
676: do { \
677: if (mat) { \
678: for (PetscInt _row = 0; _row < (nrows); _row++) { \
679: const PetscInt _nz = (mat)->i[_row + 1] - (mat)->i[_row]; \
680: for (PetscInt _j = 0; _j < _nz; _j++) { \
681: PetscInt *_col = _j + (mat)->j + (mat)->i[_row]; \
682: PetscCall(PetscHMapISet((ta), *_col + 1, 1)); \
683: } \
684: } \
685: } \
686: } while (0)
688: /*
689: Add column indices into table for counting the nonzeros of merged rows
690: */
691: #define MatMergeRows_SeqAIJ(mat, nrows, rows, ta) \
692: do { \
693: for (PetscInt _i = 0; _i < (nrows); _i++) { \
694: const PetscInt _row = (rows)[_i]; \
695: const PetscInt _nz = (mat)->i[_row + 1] - (mat)->i[_row]; \
696: for (PetscInt _j = 0; _j < _nz; _j++) { \
697: PetscInt *_col = _j + (mat)->j + (mat)->i[_row]; \
698: PetscCall(PetscHMapISetWithMode((ta), *_col + 1, 1, INSERT_VALUES)); \
699: } \
700: } \
701: } while (0)