Actual source code: aij.h
5: #include private/matimpl.h
7: /*
8: Struct header shared by SeqAIJ, SeqBAIJ and SeqSBAIJ matrix formats
9: */
10: #define SEQAIJHEADER(datatype) \
11: PetscTruth roworiented; /* if true, row-oriented input, default */\
12: PetscInt nonew; /* 1 don't add new nonzeros, -1 generate error on new */\
13: PetscInt nounused; /* -1 generate error on unused space */\
14: PetscTruth singlemalloc; /* if true a, i, and j have been obtained with one big malloc */\
15: PetscInt maxnz; /* allocated nonzeros */\
16: PetscInt *imax; /* maximum space allocated for each row */\
17: PetscInt *ilen; /* actual length of each row */\
18: PetscTruth free_imax_ilen; \
19: PetscInt reallocs; /* number of mallocs done during MatSetValues() \
20: as more values are set than were prealloced */\
21: PetscInt rmax; /* max nonzeros in any row */\
22: PetscTruth keepnonzeropattern; /* keeps matrix structure same in calls to MatZeroRows()*/\
23: PetscTruth ignorezeroentries; \
24: PetscInt *xtoy,*xtoyB; /* map nonzero pattern of X into Y's, used by MatAXPY() */\
25: Mat XtoY; /* used by MatAXPY() */\
26: PetscTruth free_ij; /* free the column indices j and row offsets i when the matrix is destroyed */ \
27: PetscTruth 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: PetscTruth free_diag; \
34: datatype *a; /* nonzero elements */ \
35: PetscScalar *solve_work; /* work space used in MatSolve */ \
36: IS row, col, icol; /* index sets, used for reorderings */ \
37: PetscTruth pivotinblocks; /* pivot inside factorization of each diagonal block */ \
38: Mat parent /* set if this matrix was formed with MatDuplicate(...,MAT_SHARE_NONZERO_PATTERN,....);
39: means that this shares some data structures with the parent including diag, ilen, imax, i, j */
41: /*
42: MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix
43: format) or compressed sparse row (CSR). The i[] and j[] arrays start at 0. For example,
44: j[i[k]+p] is the pth column in row k. Note that the diagonal
45: matrix elements are stored with the rest of the nonzeros (not separately).
46: */
48: /* Info about i-nodes (identical nodes) helper class for SeqAIJ */
49: typedef struct {
50: MatScalar *bdiag,*ibdiag,*ssor_work; /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */
51: PetscInt bdiagsize; /* length of bdiag and ibdiag */
52: PetscTruth ibdiagvalid; /* do ibdiag[] and bdiag[] contain the most recent values */
54: PetscTruth use;
55: PetscInt node_count; /* number of inodes */
56: PetscInt *size; /* size of each inode */
57: PetscInt limit; /* inode limit */
58: PetscInt max_limit; /* maximum supported inode limit */
59: PetscTruth checked; /* if inodes have been checked for */
60: } Mat_SeqAIJ_Inode;
62: EXTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat,PetscViewer);
63: EXTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat,MatAssemblyType);
64: EXTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat);
65: EXTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat);
66: EXTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat,MatOption,PetscTruth);
67: EXTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat,MatDuplicateOption,Mat*);
68: EXTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode_inplace(Mat,Mat,const MatFactorInfo*);
69: EXTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat,Mat,const MatFactorInfo*);
71: typedef struct {
72: SEQAIJHEADER(MatScalar);
73: Mat_SeqAIJ_Inode inode;
74: MatScalar *saved_values; /* location for stashing nonzero values of matrix */
76: PetscScalar *idiag,*mdiag,*ssor_work; /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */
77: PetscTruth idiagvalid; /* current idiag[] and mdiag[] are valid */
78: PetscScalar fshift,omega; /* last used omega and fshift */
80: ISColoring coloring; /* set with MatADSetColoring() used by MatADSetValues() */
81: } Mat_SeqAIJ;
83: /*
84: Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
85: */
88: PETSC_STATIC_INLINE PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA,MatScalar **a,PetscInt **j,PetscInt **i)
89: {
91: Mat_SeqAIJ *A = (Mat_SeqAIJ*) AA->data;
92: if (A->singlemalloc) {
93: PetscFree3(*a,*j,*i);
94: } else {
95: if (A->free_a && *a) {PetscFree(*a);}
96: if (A->free_ij && *j) {PetscFree(*j);}
97: if (A->free_ij && *i) {PetscFree(*i);}
98: }
99: *a = 0; *j = 0; *i = 0;
100: return 0;
101: }
103: /*
104: Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
105: */
106: #define MatSeqXAIJReallocateAIJ(Amat,AM,BS2,NROW,ROW,COL,RMAX,AA,AI,AJ,RP,AP,AIMAX,NONEW,datatype) \
107: if (NROW >= RMAX) {\
108: Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data;\
109: /* there is no extra room in row, therefore enlarge */ \
110: PetscInt new_nz = AI[AM] + CHUNKSIZE,len,*new_i=0,*new_j=0; \
111: datatype *new_a; \
112: \
113: if (NONEW == -2) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"New nonzero at (%D,%D) caused a malloc",ROW,COL); \
114: /* malloc new storage space */ \
115: PetscMalloc3(BS2*new_nz,datatype,&new_a,new_nz,PetscInt,&new_j,AM+1,PetscInt,&new_i);\
116: \
117: /* copy over old data into new slots */ \
118: for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \
119: for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \
120: PetscMemcpy(new_j,AJ,(AI[ROW]+NROW)*sizeof(PetscInt)); \
121: len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
122: PetscMemcpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len*sizeof(PetscInt)); \
123: PetscMemcpy(new_a,AA,BS2*(AI[ROW]+NROW)*sizeof(datatype)); \
124: PetscMemzero(new_a+BS2*(AI[ROW]+NROW),BS2*CHUNKSIZE*sizeof(datatype));\
125: PetscMemcpy(new_a+BS2*(AI[ROW]+NROW+CHUNKSIZE),AA+BS2*(AI[ROW]+NROW),BS2*len*sizeof(datatype)); \
126: /* free up old matrix storage */ \
127: MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i);\
128: AA = new_a; \
129: Ain->a = (MatScalar*) new_a; \
130: AI = Ain->i = new_i; AJ = Ain->j = new_j; \
131: Ain->singlemalloc = PETSC_TRUE; \
132: \
133: RP = AJ + AI[ROW]; AP = AA + BS2*AI[ROW]; \
134: RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
135: Ain->maxnz += CHUNKSIZE; \
136: Ain->reallocs++; \
137: } \
140: EXTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat,PetscInt,PetscInt*);
142: EXTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
143: EXTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
144: EXTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat,Mat,IS,IS,const MatFactorInfo*);
146: EXTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
147: EXTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
148: EXTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
149: EXTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
150: EXTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
151: EXTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
152: EXTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat,MatDuplicateOption,Mat*);
153: EXTERN PetscErrorCode MatCopy_SeqAIJ(Mat,Mat,MatStructure);
154: EXTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat,PetscTruth*,PetscInt*);
155: EXTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat);
157: EXTERN PetscErrorCode MatMult_SeqAIJ(Mat A,Vec,Vec);
158: EXTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec,Vec,Vec);
159: EXTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec,Vec);
160: EXTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec,Vec,Vec);
161: EXTERN PetscErrorCode MatSOR_SeqAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec);
163: EXTERN PetscErrorCode MatSetColoring_SeqAIJ(Mat,ISColoring);
164: EXTERN PetscErrorCode MatSetValuesAdic_SeqAIJ(Mat,void*);
165: EXTERN PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat,PetscInt,void*);
167: EXTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
168: EXTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat,PetscInt,PetscInt,PetscInt *[],PetscInt *[]);
169: EXTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
170: EXTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt,PetscInt*,PetscInt*,PetscInt,PetscInt,PetscInt**,PetscInt**);
171: EXTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
172: EXTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
173: EXTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
174: EXTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
175: EXTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat,Mat,const MatFactorInfo*);
176: EXTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat,IS,IS,const MatFactorInfo*);
177: EXTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat,Vec,Vec);
178: EXTERN PetscErrorCode MatSolve_SeqAIJ(Mat,Vec,Vec);
179: EXTERN PetscErrorCode MatSolve_SeqAIJ_Inode_inplace(Mat,Vec,Vec);
180: EXTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat,Vec,Vec);
181: EXTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering_inplace(Mat,Vec,Vec);
182: EXTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat,Vec,Vec);
183: EXTERN PetscErrorCode MatSolve_SeqAIJ_InplaceWithPerm(Mat,Vec,Vec);
184: EXTERN PetscErrorCode MatSolveAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
185: EXTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat,Vec,Vec,Vec);
186: EXTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat,Vec,Vec);
187: EXTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat,Vec,Vec);
188: EXTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
189: EXTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat,Vec,Vec,Vec);
190: EXTERN PetscErrorCode MatMatSolve_SeqAIJ_inplace(Mat,Mat,Mat);
191: EXTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat,Mat,Mat);
192: EXTERN PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscTruth* flg);
193: EXTERN PetscErrorCode MatFDColoringCreate_SeqAIJ(Mat,ISColoring,MatFDColoring);
194: EXTERN PetscErrorCode MatLoad_SeqAIJ(PetscViewer, const MatType,Mat*);
195: EXTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat);
196: EXTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
197: EXTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
198: EXTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ(Mat,Mat,PetscReal,Mat*);
199: EXTERN PetscErrorCode MatPtAPNumeric_SeqAIJ(Mat,Mat,Mat);
200: EXTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
201: EXTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
202: EXTERN PetscErrorCode MatMatMultTranspose_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
203: EXTERN PetscErrorCode MatMatMultTransposeSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
204: EXTERN PetscErrorCode MatMatMultTransposeNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
205: EXTERN PetscErrorCode MatSetValues_SeqAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode);
206: EXTERN PetscErrorCode MatGetRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
207: EXTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
208: EXTERN PetscErrorCode MatAXPY_SeqAIJ(Mat,PetscScalar,Mat,MatStructure);
209: EXTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat,PetscInt,PetscTruth,PetscTruth,PetscInt*,PetscInt *[],PetscInt *[],PetscTruth *);
210: EXTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat,PetscInt,PetscTruth,PetscTruth,PetscInt *,PetscInt *[],PetscInt *[],PetscTruth *);
211: EXTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat,PetscInt,PetscTruth,PetscTruth,PetscInt*,PetscInt *[],PetscInt *[],PetscTruth *);
212: EXTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat,PetscInt,PetscTruth,PetscTruth,PetscInt *,PetscInt *[],PetscInt *[],PetscTruth *);
213: EXTERN PetscErrorCode MatDestroy_SeqAIJ(Mat);
214: EXTERN PetscErrorCode MatView_SeqAIJ(Mat,PetscViewer);
216: EXTERN PetscErrorCode Mat_CheckInode(Mat,PetscTruth);
217: EXTERN PetscErrorCode Mat_CheckInode_FactorLU(Mat,PetscTruth);
220: EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat,const MatType,MatReuse,Mat*);
221: EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat,const MatType,MatReuse,Mat*);
222: EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqCSRPERM(Mat,const MatType,MatReuse,Mat*);
223: EXTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat,PetscReal,IS,IS);
224: EXTERN PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
227: /*
228: PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage
230: Input Parameters:
231: + nnz - the number of entries
232: . r - the array of vector values
233: . xv - the matrix values for the row
234: - xi - the column indices of the nonzeros in the row
236: Output Parameter:
237: . sum - negative the sum of results
239: PETSc compile flags:
240: + PETSC_KERNEL_USE_UNROLL_4 - don't use this; it changes nnz and hence is WRONG
241: - PETSC_KERNEL_USE_UNROLL_2 -
243: .seealso: PetscSparseDensePlusDot()
245: */
246: #ifdef PETSC_KERNEL_USE_UNROLL_4
247: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) {\
248: if (nnz > 0) {\
249: switch (nnz & 0x3) {\
250: case 3: sum -= *xv++ * r[*xi++];\
251: case 2: sum -= *xv++ * r[*xi++];\
252: case 1: sum -= *xv++ * r[*xi++];\
253: nnz -= 4;}\
254: while (nnz > 0) {\
255: sum -= xv[0] * r[xi[0]] - xv[1] * r[xi[1]] -\
256: xv[2] * r[xi[2]] - xv[3] * r[xi[3]];\
257: xv += 4; xi += 4; nnz -= 4; }}}
259: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
260: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) {\
261: PetscInt __i,__i1,__i2;\
262: for(__i=0;__i<nnz-1;__i+=2) {__i1 = xi[__i]; __i2=xi[__i+1];\
263: sum -= (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);}\
264: if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]];}
266: #else
267: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) {\
268: PetscInt __i;\
269: for(__i=0;__i<nnz;__i++) sum -= xv[__i] * r[xi[__i]];}
270: #endif
274: /*
275: PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage
277: Input Parameters:
278: + nnz - the number of entries
279: . r - the array of vector values
280: . xv - the matrix values for the row
281: - xi - the column indices of the nonzeros in the row
283: Output Parameter:
284: . sum - the sum of results
286: PETSc compile flags:
287: + PETSC_KERNEL_USE_UNROLL_4 - don't use this; it changes nnz and hence is WRONG
288: - PETSC_KERNEL_USE_UNROLL_2 -
290: .seealso: PetscSparseDenseMinusDot()
292: */
293: #ifdef PETSC_KERNEL_USE_UNROLL_4
294: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) {\
295: if (nnz > 0) {\
296: switch (nnz & 0x3) {\
297: case 3: sum += *xv++ * r[*xi++];\
298: case 2: sum += *xv++ * r[*xi++];\
299: case 1: sum += *xv++ * r[*xi++];\
300: nnz -= 4;}\
301: while (nnz > 0) {\
302: sum += xv[0] * r[xi[0]] + xv[1] * r[xi[1]] +\
303: xv[2] * r[xi[2]] + xv[3] * r[xi[3]];\
304: xv += 4; xi += 4; nnz -= 4; }}}
306: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
307: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) {\
308: PetscInt __i,__i1,__i2;\
309: for(__i=0;__i<nnz-1;__i+=2) {__i1 = xi[__i]; __i2=xi[__i+1];\
310: sum += (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);}\
311: if (nnz & 0x1) sum += xv[__i] * r[xi[__i]];}
313: #else
314: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) {\
315: PetscInt __i;\
316: for(__i=0;__i<nnz;__i++) sum += xv[__i] * r[xi[__i]];}
317: #endif
319: #endif