Actual source code: aij.h
petsc-3.12.5 2020-03-29
5: #include <petsc/private/matimpl.h>
6: #include <petscctable.h>
8: /*
9: Struct header shared by SeqAIJ, SeqBAIJ and SeqSBAIJ matrix formats
10: */
11: #define SEQAIJHEADER(datatype) \
12: PetscBool roworiented; /* if true, row-oriented input, default */ \
13: PetscInt nonew; /* 1 don't add new nonzeros, -1 generate error on new */ \
14: PetscInt nounused; /* -1 generate error on unused space */ \
15: PetscBool singlemalloc; /* if true a, i, and j have been obtained with one big malloc */ \
16: PetscInt maxnz; /* allocated nonzeros */ \
17: PetscInt *imax; /* maximum space allocated for each row */ \
18: PetscInt *ilen; /* actual length of each row */ \
19: PetscInt *ipre; /* space preallocated for each row by user */ \
20: PetscBool free_imax_ilen; \
21: PetscInt reallocs; /* number of mallocs done during MatSetValues() \
22: as more values are set than were prealloced */\
23: PetscInt rmax; /* max nonzeros in any row */ \
24: PetscBool keepnonzeropattern; /* keeps matrix structure same in calls to MatZeroRows()*/ \
25: PetscBool ignorezeroentries; \
26: PetscBool free_ij; /* free the column indices j and row offsets i when the matrix is destroyed */ \
27: PetscBool free_a; /* free the numerical values when matrix is destroy */ \
28: Mat_CompressedRow compressedrow; /* use compressed row format */ \
29: PetscInt nz; /* nonzeros */ \
30: PetscInt *i; /* pointer to beginning of each row */ \
31: PetscInt *j; /* column values: j + i[k] - 1 is start of row k */ \
32: PetscInt *diag; /* pointers to diagonal elements */ \
33: PetscInt nonzerorowcnt; /* how many rows have nonzero entries */ \
34: PetscBool free_diag; \
35: datatype *a; /* nonzero elements */ \
36: PetscScalar *solve_work; /* work space used in MatSolve */ \
37: IS row, col, icol; /* index sets, used for reorderings */ \
38: PetscBool pivotinblocks; /* pivot inside factorization of each diagonal block */ \
39: Mat parent; /* set if this matrix was formed with MatDuplicate(...,MAT_SHARE_NONZERO_PATTERN,....); \
40: means that this shares some data structures with the parent including diag, ilen, imax, i, j */\
41: Mat_SubSppt *submatis1 /* used by MatCreateSubMatrices_MPIXAIJ_Local */
43: typedef struct {
44: MatTransposeColoring matcoloring;
45: Mat Bt_den; /* dense matrix of B^T */
46: Mat ABt_den; /* dense matrix of A*B^T */
47: PetscBool usecoloring;
48: PetscErrorCode (*destroy)(Mat);
49: } Mat_MatMatTransMult;
51: typedef struct { /* used by MatTransposeMatMult() */
52: Mat At; /* transpose of the first matrix */
53: Mat mA; /* maij matrix of A */
54: Vec bt,ct; /* vectors to hold locally transposed arrays of B and C */
55: PetscErrorCode (*destroy)(Mat);
56: } Mat_MatTransMatMult;
58: typedef struct {
59: PetscInt *api,*apj; /* symbolic structure of A*P */
60: PetscScalar *apa; /* temporary array for storing one row of A*P */
61: PetscErrorCode (*destroy)(Mat);
62: } Mat_AP;
64: typedef struct {
65: MatTransposeColoring matcoloring;
66: Mat Rt; /* sparse or dense matrix of R^T */
67: Mat RARt; /* dense matrix of R*A*R^T */
68: Mat ARt; /* A*R^T used for the case -matrart_color_art */
69: MatScalar *work; /* work array to store columns of A*R^T used in MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqDense() */
70: PetscErrorCode (*destroy)(Mat);
71: } Mat_RARt;
73: typedef struct {
74: Mat BC; /* temp matrix for storing B*C */
75: PetscErrorCode (*destroy)(Mat);
76: } Mat_MatMatMatMult;
78: /*
79: MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix
80: format) or compressed sparse row (CSR). The i[] and j[] arrays start at 0. For example,
81: j[i[k]+p] is the pth column in row k. Note that the diagonal
82: matrix elements are stored with the rest of the nonzeros (not separately).
83: */
85: /* Info about i-nodes (identical nodes) helper class for SeqAIJ */
86: typedef struct {
87: MatScalar *bdiag,*ibdiag,*ssor_work; /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */
88: PetscInt bdiagsize; /* length of bdiag and ibdiag */
89: PetscBool ibdiagvalid; /* do ibdiag[] and bdiag[] contain the most recent values */
91: PetscBool use;
92: PetscInt node_count; /* number of inodes */
93: PetscInt *size; /* size of each inode */
94: PetscInt limit; /* inode limit */
95: PetscInt max_limit; /* maximum supported inode limit */
96: PetscBool checked; /* if inodes have been checked for */
97: PetscObjectState mat_nonzerostate; /* non-zero state when inodes were checked for */
98: } Mat_SeqAIJ_Inode;
100: PETSC_INTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat,PetscViewer);
101: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat,MatAssemblyType);
102: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat);
103: PETSC_INTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat);
104: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat,MatOption,PetscBool);
105: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat,MatDuplicateOption,Mat*);
106: PETSC_INTERN PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat,Mat,MatDuplicateOption,PetscBool);
107: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode_inplace(Mat,Mat,const MatFactorInfo*);
108: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat,Mat,const MatFactorInfo*);
109: PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat,PetscScalar**);
110: PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat,PetscScalar**);
112: typedef struct {
113: SEQAIJHEADER(MatScalar);
114: Mat_SeqAIJ_Inode inode;
115: MatScalar *saved_values; /* location for stashing nonzero values of matrix */
117: PetscScalar *idiag,*mdiag,*ssor_work; /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */
118: PetscBool idiagvalid; /* current idiag[] and mdiag[] are valid */
119: PetscScalar *ibdiag; /* inverses of block diagonals */
120: PetscBool ibdiagvalid; /* inverses of block diagonals are valid. */
121: PetscBool diagonaldense; /* all entries along the diagonal have been set; i.e. no missing diagonal terms */
122: PetscScalar fshift,omega; /* last used omega and fshift */
124: ISColoring coloring; /* set with MatADSetColoring() used by MatADSetValues() */
126: PetscScalar *matmult_abdense; /* used by MatMatMult() */
127: Mat_AP *ap; /* used by MatPtAP() */
128: Mat_MatMatMatMult *matmatmatmult; /* used by MatMatMatMult() */
129: Mat_RARt *rart; /* used by MatRARt() */
130: Mat_MatMatTransMult *abt; /* used by MatMatTransposeMult() */
131: Mat_MatTransMatMult *atb; /* used by MatTransposeMatMult() */
132: } Mat_SeqAIJ;
134: /*
135: Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
136: */
137: PETSC_STATIC_INLINE PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA,MatScalar **a,PetscInt **j,PetscInt **i)
138: {
140: Mat_SeqAIJ *A = (Mat_SeqAIJ*) AA->data;
141: if (A->singlemalloc) {
142: PetscFree3(*a,*j,*i);
143: } else {
144: if (A->free_a) {PetscFree(*a);}
145: if (A->free_ij) {PetscFree(*j);}
146: if (A->free_ij) {PetscFree(*i);}
147: }
148: return 0;
149: }
150: /*
151: Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
152: This is a macro because it takes the datatype as an argument which can be either a Mat or a MatScalar
153: */
154: #define MatSeqXAIJReallocateAIJ(Amat,AM,BS2,NROW,ROW,COL,RMAX,AA,AI,AJ,RP,AP,AIMAX,NONEW,datatype) \
155: if (NROW >= RMAX) { \
156: Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data; \
157: /* there is no extra room in row, therefore enlarge */ \
158: PetscInt CHUNKSIZE = 15,new_nz = AI[AM] + CHUNKSIZE,len,*new_i=0,*new_j=0; \
159: datatype *new_a; \
160: \
161: if (NONEW == -2) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"New nonzero at (%D,%D) caused a malloc\nUse MatSetOption(A, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE) to turn off this check",ROW,COL); \
162: /* malloc new storage space */ \
163: PetscMalloc3(BS2*new_nz,&new_a,new_nz,&new_j,AM+1,&new_i); \
164: \
165: /* copy over old data into new slots */ \
166: for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \
167: for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \
168: PetscArraycpy(new_j,AJ,AI[ROW]+NROW); \
169: len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
170: PetscArraycpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len); \
171: PetscArraycpy(new_a,AA,BS2*(AI[ROW]+NROW)); \
172: PetscArrayzero(new_a+BS2*(AI[ROW]+NROW),BS2*CHUNKSIZE); \
173: PetscArraycpy(new_a+BS2*(AI[ROW]+NROW+CHUNKSIZE),AA+BS2*(AI[ROW]+NROW),BS2*len); \
174: /* free up old matrix storage */ \
175: MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i); \
176: AA = new_a; \
177: Ain->a = (MatScalar*) new_a; \
178: AI = Ain->i = new_i; AJ = Ain->j = new_j; \
179: Ain->singlemalloc = PETSC_TRUE; \
180: \
181: RP = AJ + AI[ROW]; AP = AA + BS2*AI[ROW]; \
182: RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
183: Ain->maxnz += BS2*CHUNKSIZE; \
184: Ain->reallocs++; \
185: } \
187: #define MatSeqXAIJReallocateAIJ_structure_only(Amat,AM,BS2,NROW,ROW,COL,RMAX,AI,AJ,RP,AIMAX,NONEW,datatype) \
188: if (NROW >= RMAX) { \
189: Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data; \
190: /* there is no extra room in row, therefore enlarge */ \
191: PetscInt CHUNKSIZE = 15,new_nz = AI[AM] + CHUNKSIZE,len,*new_i=0,*new_j=0; \
192: \
193: if (NONEW == -2) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"New nonzero at (%D,%D) caused a malloc\nUse MatSetOption(A, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE) to turn off this check",ROW,COL); \
194: /* malloc new storage space */ \
195: PetscMalloc1(new_nz,&new_j); \
196: PetscMalloc1(AM+1,&new_i);\
197: \
198: /* copy over old data into new slots */ \
199: for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \
200: for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \
201: PetscArraycpy(new_j,AJ,AI[ROW]+NROW); \
202: len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
203: PetscArraycpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len); \
204: \
205: /* free up old matrix storage */ \
206: MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i); \
207: Ain->a = NULL; \
208: AI = Ain->i = new_i; AJ = Ain->j = new_j; \
209: Ain->singlemalloc = PETSC_FALSE; \
210: Ain->free_a = PETSC_FALSE; \
211: \
212: RP = AJ + AI[ROW]; \
213: RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
214: Ain->maxnz += BS2*CHUNKSIZE; \
215: Ain->reallocs++; \
216: } \
218: PETSC_INTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat,PetscInt,const PetscInt*);
219: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
220: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
221: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat,Mat,IS,IS,const MatFactorInfo*);
223: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
224: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
225: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
226: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
227: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
228: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
229: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat,MatDuplicateOption,Mat*);
230: PETSC_INTERN PetscErrorCode MatCopy_SeqAIJ(Mat,Mat,MatStructure);
231: PETSC_INTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat,PetscBool*,PetscInt*);
232: PETSC_INTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat);
233: PETSC_INTERN PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat,PetscInt*,PetscInt**);
235: PETSC_INTERN PetscErrorCode MatMult_SeqAIJ(Mat A,Vec,Vec);
236: PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec,Vec,Vec);
237: PETSC_INTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec,Vec);
238: PETSC_INTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec,Vec,Vec);
239: PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec);
241: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ(Mat,MatOption,PetscBool);
243: PETSC_INTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
244: PETSC_INTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat,PetscInt,PetscInt,PetscInt *[],PetscInt *[]);
245: PETSC_INTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
246: PETSC_INTERN PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat,Mat*);
247: PETSC_INTERN PetscErrorCode MatTranspose_SeqAIJ(Mat,MatReuse,Mat*);
248: PETSC_INTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt,PetscInt*,PetscInt*,PetscBool,PetscInt,PetscInt,PetscInt**,PetscInt**);
249: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
250: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
251: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
252: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
253: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat,Mat,const MatFactorInfo*);
254: PETSC_INTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat,IS,IS,const MatFactorInfo*);
255: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat,Vec,Vec);
256: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ(Mat,Vec,Vec);
257: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode_inplace(Mat,Vec,Vec);
258: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat,Vec,Vec);
259: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering_inplace(Mat,Vec,Vec);
260: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat,Vec,Vec);
261: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_InplaceWithPerm(Mat,Vec,Vec);
262: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
263: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat,Vec,Vec,Vec);
264: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat,Vec,Vec);
265: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat,Vec,Vec);
266: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
267: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat,Vec,Vec,Vec);
268: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ_inplace(Mat,Mat,Mat);
269: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat,Mat,Mat);
270: PETSC_INTERN PetscErrorCode MatEqual_SeqAIJ(Mat,Mat,PetscBool*);
271: PETSC_INTERN PetscErrorCode MatFDColoringCreate_SeqXAIJ(Mat,ISColoring,MatFDColoring);
272: PETSC_INTERN PetscErrorCode MatFDColoringSetUp_SeqXAIJ(Mat,ISColoring,MatFDColoring);
273: PETSC_INTERN PetscErrorCode MatFDColoringSetUpBlocked_AIJ_Private(Mat,MatFDColoring,PetscInt);
274: PETSC_INTERN PetscErrorCode MatLoad_AIJ_HDF5(Mat,PetscViewer);
275: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ_Binary(Mat,PetscViewer);
276: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ(Mat,PetscViewer);
277: PETSC_INTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat);
279: PETSC_INTERN PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
280: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
281: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(Mat,Mat,PetscReal,Mat*);
282: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat,Mat,PetscReal,Mat*);
283: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,PetscReal,Mat*);
284: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat,Mat,PetscReal,Mat*);
285: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat,Mat,PetscReal,Mat*);
286: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat,Mat,PetscReal,Mat*);
287: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(Mat,Mat,PetscReal,Mat*);
288: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat,Mat,PetscReal,Mat*);
289: #if defined(PETSC_HAVE_HYPRE)
290: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat,Mat,PetscReal,Mat*);
291: #endif
292: PETSC_INTERN PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ_Combined(Mat,Mat,PetscReal,Mat*);
293: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
294: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(Mat,Mat,Mat);
295: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat,Mat,Mat);
296: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,Mat);
297: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Combined(Mat,Mat,Mat);
299: PETSC_INTERN PetscErrorCode MatPtAP_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
300: PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,PetscReal,Mat*);
301: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
302: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,Mat);
304: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
305: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_matmattransposemult(Mat,Mat,PetscReal,Mat*);
306: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_colorrart(Mat,Mat,PetscReal,Mat*);
307: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
308: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_matmattransposemult(Mat,Mat,Mat);
309: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_colorrart(Mat,Mat,Mat);
311: PETSC_INTERN PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
312: PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
313: PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
314: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_MatTransMatMult(Mat);
316: PETSC_INTERN PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqDense(Mat,Mat,MatReuse,PetscReal,Mat*);
317: PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqDense(Mat,Mat,PetscReal,Mat*);
318: PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqDense(Mat,Mat,Mat);
320: PETSC_INTERN PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
321: PETSC_INTERN PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
322: PETSC_INTERN PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
323: PETSC_INTERN PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat,ISColoring,MatTransposeColoring);
324: PETSC_INTERN PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring,Mat,Mat);
325: PETSC_INTERN PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring,Mat,Mat);
327: PETSC_INTERN PetscErrorCode MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*);
328: PETSC_INTERN PetscErrorCode MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,PetscReal,Mat*);
329: PETSC_INTERN PetscErrorCode MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,Mat);
331: PETSC_INTERN PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat,PetscInt,PetscInt,PetscRandom);
332: PETSC_INTERN PetscErrorCode MatSetValues_SeqAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode);
333: PETSC_INTERN PetscErrorCode MatGetRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
334: PETSC_INTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
335: PETSC_INTERN PetscErrorCode MatScale_SeqAIJ(Mat,PetscScalar);
336: PETSC_INTERN PetscErrorCode MatDiagonalScale_SeqAIJ(Mat,Vec,Vec);
337: PETSC_INTERN PetscErrorCode MatDiagonalSet_SeqAIJ(Mat,Vec,InsertMode);
338: PETSC_INTERN PetscErrorCode MatAXPY_SeqAIJ(Mat,PetscScalar,Mat,MatStructure);
339: PETSC_INTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
340: PETSC_INTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
341: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
342: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
343: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscInt *[],PetscBool*);
344: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscInt *[],PetscBool*);
345: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat);
346: PETSC_INTERN PetscErrorCode MatSetUp_SeqAIJ(Mat);
347: PETSC_INTERN PetscErrorCode MatView_SeqAIJ(Mat,PetscViewer);
349: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat);
350: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal_Inode(Mat);
351: PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode(Mat);
352: PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode_FactorLU(Mat);
354: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat,Mat,PetscInt*);
356: PETSC_INTERN PetscErrorCode MatMatMatMult_Transpose_AIJ_AIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*);
357: #if defined(PETSC_HAVE_MATLAB_ENGINE)
358: PETSC_EXTERN PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*);
359: PETSC_EXTERN PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*);
360: #endif
361: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat,MatType,MatReuse,Mat*);
362: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat,MatType,MatReuse,Mat*);
363: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqDense(Mat,MatType,MatReuse,Mat*);
364: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*);
365: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat,MatType,MatReuse,Mat*);
366: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*);
367: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJPERM(Mat,MatType,MatReuse,Mat*);
368: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJSELL(Mat,MatType,MatReuse,Mat*);
369: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJMKL(Mat,MatType,MatReuse,Mat*);
370: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJViennaCL(Mat,MatType,MatReuse,Mat*);
371: PETSC_INTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat,PetscReal,IS,IS);
372: PETSC_INTERN PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
373: PETSC_INTERN PetscErrorCode MatRARt_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
374: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat);
375: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat,MatAssemblyType);
377: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt,const PetscInt*,const PetscInt*,const PetscInt*,const PetscInt*,PetscInt*);
378: PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm,Mat,PetscInt,MatReuse,Mat*);
379: PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIAIJ(MPI_Comm,Mat,PetscInt,MatReuse,Mat*);
381: PETSC_INTERN PetscErrorCode MatSetSeqMat_SeqAIJ(Mat,IS,IS,MatStructure,Mat);
382: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt*);
383: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat);
384: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Dummy(Mat);
385: PETSC_INTERN PetscErrorCode MatDestroySubMatrices_Dummy(PetscInt, Mat*[]);
386: PETSC_INTERN PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat,IS,IS,PetscInt,MatReuse,Mat*);
388: PETSC_INTERN PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat,ISLocalToGlobalMapping*);
390: /*
391: PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage
393: Input Parameters:
394: + nnz - the number of entries
395: . r - the array of vector values
396: . xv - the matrix values for the row
397: - xi - the column indices of the nonzeros in the row
399: Output Parameter:
400: . sum - negative the sum of results
402: PETSc compile flags:
403: + PETSC_KERNEL_USE_UNROLL_4
404: - PETSC_KERNEL_USE_UNROLL_2
406: Developer Notes:
407: The macro changes sum but not other parameters
409: .seealso: PetscSparseDensePlusDot()
411: */
412: #if defined(PETSC_KERNEL_USE_UNROLL_4)
413: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
414: if (nnz > 0) { \
415: PetscInt nnz2=nnz,rem=nnz&0x3; \
416: switch (rem) { \
417: case 3: sum -= *xv++ *r[*xi++]; \
418: case 2: sum -= *xv++ *r[*xi++]; \
419: case 1: sum -= *xv++ *r[*xi++]; \
420: nnz2 -= rem;} \
421: while (nnz2 > 0) { \
422: sum -= xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + \
423: xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
424: xv += 4; xi += 4; nnz2 -= 4; \
425: } \
426: xv -= nnz; xi -= nnz; \
427: } \
428: }
430: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
431: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
432: PetscInt __i,__i1,__i2; \
433: for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \
434: sum -= (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \
435: if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]];}
437: #else
438: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
439: PetscInt __i; \
440: for (__i=0; __i<nnz; __i++) sum -= xv[__i] * r[xi[__i]];}
441: #endif
445: /*
446: PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage
448: Input Parameters:
449: + nnz - the number of entries
450: . r - the array of vector values
451: . xv - the matrix values for the row
452: - xi - the column indices of the nonzeros in the row
454: Output Parameter:
455: . sum - the sum of results
457: PETSc compile flags:
458: + PETSC_KERNEL_USE_UNROLL_4
459: - PETSC_KERNEL_USE_UNROLL_2
461: Developer Notes:
462: The macro changes sum but not other parameters
464: .seealso: PetscSparseDenseMinusDot()
466: */
467: #if defined(PETSC_KERNEL_USE_UNROLL_4)
468: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
469: if (nnz > 0) { \
470: PetscInt nnz2=nnz,rem=nnz&0x3; \
471: switch (rem) { \
472: case 3: sum += *xv++ *r[*xi++]; \
473: case 2: sum += *xv++ *r[*xi++]; \
474: case 1: sum += *xv++ *r[*xi++]; \
475: nnz2 -= rem;} \
476: while (nnz2 > 0) { \
477: sum += xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + \
478: xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
479: xv += 4; xi += 4; nnz2 -= 4; \
480: } \
481: xv -= nnz; xi -= nnz; \
482: } \
483: }
485: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
486: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
487: PetscInt __i,__i1,__i2; \
488: for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \
489: sum += (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \
490: if (nnz & 0x1) sum += xv[__i] * r[xi[__i]];}
492: #elif 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)
493: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) PetscSparseDensePlusDot_AVX512_Private(&(sum),(r),(xv),(xi),(nnz))
495: #else
496: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
497: PetscInt __i; \
498: for (__i=0; __i<nnz; __i++) sum += xv[__i] * r[xi[__i]];}
499: #endif
501: #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)
502: #include <immintrin.h>
503: #if !defined(_MM_SCALE_8)
504: #define _MM_SCALE_8 8
505: #endif
507: PETSC_STATIC_INLINE void PetscSparseDensePlusDot_AVX512_Private(PetscScalar *sum,const PetscScalar *x,const MatScalar *aa,const PetscInt *aj,PetscInt n)
508: {
509: __m512d vec_x,vec_y,vec_vals;
510: __m256i vec_idx;
511: __mmask8 mask;
512: PetscInt j;
514: vec_y = _mm512_setzero_pd();
515: for (j=0; j<(n>>3); j++) {
516: vec_idx = _mm256_loadu_si256((__m256i const*)aj);
517: vec_vals = _mm512_loadu_pd(aa);
518: vec_x = _mm512_i32gather_pd(vec_idx,x,_MM_SCALE_8);
519: vec_y = _mm512_fmadd_pd(vec_x,vec_vals,vec_y);
520: aj += 8; aa += 8;
521: }
522: /* masked load does not work on KNL, it requires avx512vl */
523: if ((n&0x07)>2) {
524: mask = (__mmask8)(0xff >> (8-(n&0x07)));
525: vec_idx = _mm256_loadu_si256((__m256i const*)aj);
526: vec_vals = _mm512_loadu_pd(aa);
527: vec_x = _mm512_mask_i32gather_pd(vec_x,mask,vec_idx,x,_MM_SCALE_8);
528: vec_y = _mm512_mask3_fmadd_pd(vec_x,vec_vals,vec_y,mask);
529: } else if ((n&0x07)==2) {
530: *sum += aa[0]*x[aj[0]];
531: *sum += aa[1]*x[aj[1]];
532: } else if ((n&0x07)==1) {
533: *sum += aa[0]*x[aj[0]];
534: }
535: if (n>2) *sum += _mm512_reduce_add_pd(vec_y);
536: /*
537: for(j=0;j<(n&0x07);j++) *sum += aa[j]*x[aj[j]];
538: */
539: }
540: #endif
542: /*
543: PetscSparseDenseMaxDot - The inner kernel of a modified matrix-vector product \max_i xv[i] * r[xi[i]] for CSR storage
545: Input Parameters:
546: + nnz - the number of entries
547: . r - the array of vector values
548: . xv - the matrix values for the row
549: - xi - the column indices of the nonzeros in the row
551: Output Parameter:
552: . max - the max of results
554: .seealso: PetscSparseDensePlusDot(), PetscSparseDenseMinusDot()
556: */
557: #define PetscSparseDenseMaxDot(max,r,xv,xi,nnz) { \
558: PetscInt __i; \
559: for (__i=0; __i<nnz; __i++) max = PetscMax(PetscRealPart(max), PetscRealPart(xv[__i] * r[xi[__i]]));}
561: /*
562: Add column indices into table for counting the max nonzeros of merged rows
563: */
564: #define MatRowMergeMax_SeqAIJ(mat,nrows,ta) { \
565: PetscInt _j,_row,_nz,*_col; \
566: if (mat) { \
567: for (_row=0; _row<nrows; _row++) { \
568: _nz = mat->i[_row+1] - mat->i[_row]; \
569: for (_j=0; _j<_nz; _j++) { \
570: _col = _j + mat->j + mat->i[_row]; \
571: PetscTableAdd(ta,*_col+1,1,INSERT_VALUES); \
572: } \
573: } \
574: } \
575: }
577: /*
578: Add column indices into table for counting the nonzeros of merged rows
579: */
580: #define MatMergeRows_SeqAIJ(mat,nrows,rows,ta) { \
581: PetscInt _j,_row,_nz,*_col,_i; \
582: for (_i=0; _i<nrows; _i++) {\
583: _row = rows[_i]; \
584: _nz = mat->i[_row+1] - mat->i[_row]; \
585: for (_j=0; _j<_nz; _j++) { \
586: _col = _j + mat->j + mat->i[_row]; \
587: PetscTableAdd(ta,*_col+1,1,INSERT_VALUES); \
588: } \
589: } \
590: }
592: #endif