Actual source code: aij.h
petsc-3.4.2 2013-07-02
5: #include <petsc-private/matimpl.h>
7: /*
8: Struct header shared by SeqAIJ, SeqBAIJ and SeqSBAIJ matrix formats
9: */
10: #define SEQAIJHEADER(datatype) \
11: PetscBool 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: PetscBool 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: PetscBool 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: PetscBool keepnonzeropattern; /* keeps matrix structure same in calls to MatZeroRows()*/ \
23: PetscBool ignorezeroentries; \
24: PetscInt *xtoy,*xtoyB; /* map nonzero pattern of X into Y's, used by MatAXPY() */ \
25: Mat XtoY; /* used by MatAXPY() */ \
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: PetscBool 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: PetscBool 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: typedef struct {
42: MatTransposeColoring matcoloring;
43: Mat Bt_den; /* dense matrix of B^T */
44: Mat ABt_den; /* dense matrix of A*B^T */
45: PetscBool usecoloring;
46: PetscErrorCode (*destroy)(Mat);
47: } Mat_MatMatTransMult;
49: typedef struct {
50: PetscInt *api,*apj; /* symbolic structure of A*P */
51: PetscScalar *apa; /* temporary array for storing one row of A*P */
52: PetscErrorCode (*destroy)(Mat);
53: } Mat_PtAP;
55: typedef struct {
56: MatTransposeColoring matcoloring;
57: Mat Rt; /* dense matrix of R^T */
58: Mat RARt; /* dense matrix of R*A*R^T */
59: MatScalar *work; /* work array to store columns of A*R^T used in MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqDense() */
60: PetscErrorCode (*destroy)(Mat);
61: } Mat_RARt;
63: typedef struct {
64: Mat BC; /* temp matrix for storing B*C */
65: PetscErrorCode (*destroy)(Mat);
66: } Mat_MatMatMatMult;
68: /*
69: MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix
70: format) or compressed sparse row (CSR). The i[] and j[] arrays start at 0. For example,
71: j[i[k]+p] is the pth column in row k. Note that the diagonal
72: matrix elements are stored with the rest of the nonzeros (not separately).
73: */
75: /* Info about i-nodes (identical nodes) helper class for SeqAIJ */
76: typedef struct {
77: MatScalar *bdiag,*ibdiag,*ssor_work; /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */
78: PetscInt bdiagsize; /* length of bdiag and ibdiag */
79: PetscBool ibdiagvalid; /* do ibdiag[] and bdiag[] contain the most recent values */
81: PetscBool use;
82: PetscInt node_count; /* number of inodes */
83: PetscInt *size; /* size of each inode */
84: PetscInt limit; /* inode limit */
85: PetscInt max_limit; /* maximum supported inode limit */
86: PetscBool checked; /* if inodes have been checked for */
87: } Mat_SeqAIJ_Inode;
89: PETSC_INTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat,PetscViewer);
90: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat,MatAssemblyType);
91: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat);
92: PETSC_INTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat);
93: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat,MatOption,PetscBool);
94: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat,MatDuplicateOption,Mat*);
95: PETSC_INTERN PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat,Mat,MatDuplicateOption,PetscBool);
96: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode_inplace(Mat,Mat,const MatFactorInfo*);
97: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat,Mat,const MatFactorInfo*);
99: typedef struct {
100: SEQAIJHEADER(MatScalar);
101: Mat_SeqAIJ_Inode inode;
102: MatScalar *saved_values; /* location for stashing nonzero values of matrix */
104: PetscScalar *idiag,*mdiag,*ssor_work; /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */
105: PetscBool idiagvalid; /* current idiag[] and mdiag[] are valid */
106: PetscScalar *ibdiag; /* inverses of block diagonals */
107: PetscBool ibdiagvalid; /* inverses of block diagonals are valid. */
108: PetscScalar fshift,omega; /* last used omega and fshift */
110: ISColoring coloring; /* set with MatADSetColoring() used by MatADSetValues() */
112: PetscScalar *matmult_abdense; /* used by MatMatMult() */
113: Mat_PtAP *ptap; /* used by MatPtAP() */
114: Mat_MatMatMatMult *matmatmatmult; /* used by MatMatMatMult() */
115: } Mat_SeqAIJ;
117: /*
118: Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
119: */
122: PETSC_STATIC_INLINE PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA,MatScalar **a,PetscInt **j,PetscInt **i)
123: {
125: Mat_SeqAIJ *A = (Mat_SeqAIJ*) AA->data;
126: if (A->singlemalloc) {
127: PetscFree3(*a,*j,*i);
128: } else {
129: if (A->free_a) {PetscFree(*a);}
130: if (A->free_ij) {PetscFree(*j);}
131: if (A->free_ij) {PetscFree(*i);}
132: }
133: return 0;
134: }
135: /*
136: Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
137: This is a macro because it takes the datatype as an argument which can be either a Mat or a MatScalar
138: */
139: #define MatSeqXAIJReallocateAIJ(Amat,AM,BS2,NROW,ROW,COL,RMAX,AA,AI,AJ,RP,AP,AIMAX,NONEW,datatype) \
140: if (NROW >= RMAX) { \
141: Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data; \
142: /* there is no extra room in row, therefore enlarge */ \
143: PetscInt CHUNKSIZE = 15,new_nz = AI[AM] + CHUNKSIZE,len,*new_i=0,*new_j=0; \
144: datatype *new_a; \
145: \
146: if (NONEW == -2) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"New nonzero at (%D,%D) caused a malloc",ROW,COL); \
147: /* malloc new storage space */ \
148: PetscMalloc3(BS2*new_nz,datatype,&new_a,new_nz,PetscInt,&new_j,AM+1,PetscInt,&new_i); \
149: \
150: /* copy over old data into new slots */ \
151: for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \
152: for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \
153: PetscMemcpy(new_j,AJ,(AI[ROW]+NROW)*sizeof(PetscInt)); \
154: len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
155: PetscMemcpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len*sizeof(PetscInt)); \
156: PetscMemcpy(new_a,AA,BS2*(AI[ROW]+NROW)*sizeof(datatype)); \
157: PetscMemzero(new_a+BS2*(AI[ROW]+NROW),BS2*CHUNKSIZE*sizeof(datatype)); \
158: PetscMemcpy(new_a+BS2*(AI[ROW]+NROW+CHUNKSIZE),AA+BS2*(AI[ROW]+NROW),BS2*len*sizeof(datatype)); \
159: /* free up old matrix storage */ \
160: MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i); \
161: AA = new_a; \
162: Ain->a = (MatScalar*) new_a; \
163: AI = Ain->i = new_i; AJ = Ain->j = new_j; \
164: Ain->singlemalloc = PETSC_TRUE; \
165: \
166: RP = AJ + AI[ROW]; AP = AA + BS2*AI[ROW]; \
167: RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
168: Ain->maxnz += BS2*CHUNKSIZE; \
169: Ain->reallocs++; \
170: } \
173: PETSC_EXTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat,PetscInt,const PetscInt*);
174: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
175: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
176: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat,Mat,IS,IS,const MatFactorInfo*);
178: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
179: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
180: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
181: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
182: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
183: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
184: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat,MatDuplicateOption,Mat*);
185: PETSC_INTERN PetscErrorCode MatCopy_SeqAIJ(Mat,Mat,MatStructure);
186: PETSC_INTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat,PetscBool*,PetscInt*);
187: PETSC_INTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat);
188: PETSC_INTERN PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat,PetscInt*,PetscInt**);
190: PETSC_INTERN PetscErrorCode MatMult_SeqAIJ(Mat A,Vec,Vec);
191: PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec,Vec,Vec);
192: PETSC_INTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec,Vec);
193: PETSC_INTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec,Vec,Vec);
194: PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec);
196: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ(Mat,MatOption,PetscBool);
197: PETSC_INTERN PetscErrorCode MatSetColoring_SeqAIJ(Mat,ISColoring);
198: PETSC_INTERN PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat,PetscInt,void*);
200: PETSC_INTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
201: PETSC_INTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat,PetscInt,PetscInt,PetscInt *[],PetscInt *[]);
202: PETSC_INTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
203: PETSC_INTERN PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat,Mat*);
204: PETSC_INTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt,PetscInt*,PetscInt*,PetscInt,PetscInt,PetscInt**,PetscInt**);
205: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
206: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
207: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
208: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
209: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat,Mat,const MatFactorInfo*);
210: PETSC_INTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat,IS,IS,const MatFactorInfo*);
211: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat,Vec,Vec);
212: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ(Mat,Vec,Vec);
213: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode_inplace(Mat,Vec,Vec);
214: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat,Vec,Vec);
215: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering_inplace(Mat,Vec,Vec);
216: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat,Vec,Vec);
217: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_InplaceWithPerm(Mat,Vec,Vec);
218: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
219: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat,Vec,Vec,Vec);
220: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat,Vec,Vec);
221: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat,Vec,Vec);
222: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
223: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat,Vec,Vec,Vec);
224: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ_inplace(Mat,Mat,Mat);
225: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat,Mat,Mat);
226: PETSC_INTERN PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg);
227: PETSC_INTERN PetscErrorCode MatFDColoringCreate_SeqAIJ(Mat,ISColoring,MatFDColoring);
228: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ(Mat,PetscViewer);
229: PETSC_INTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat);
231: PETSC_INTERN PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
232: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
233: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,PetscReal,Mat*);
234: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat,Mat,PetscReal,Mat*);
235: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat,Mat,PetscReal,Mat*);
236: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat,Mat,PetscReal,Mat*);
237: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
238: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,Mat);
240: PETSC_INTERN PetscErrorCode MatPtAP_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
241: PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
242: PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,PetscReal,Mat*);
243: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
244: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,Mat);
246: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
247: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
249: PETSC_INTERN PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
250: PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
251: PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
252: PETSC_INTERN PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
253: PETSC_INTERN PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*);
254: PETSC_INTERN PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
255: PETSC_INTERN PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat,ISColoring,MatTransposeColoring);
256: PETSC_INTERN PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring,Mat,Mat);
257: PETSC_INTERN PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring,Mat,Mat);
259: PETSC_INTERN PetscErrorCode MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*);
260: PETSC_INTERN PetscErrorCode MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,PetscReal,Mat*);
261: PETSC_INTERN PetscErrorCode MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,Mat);
263: PETSC_INTERN PetscErrorCode MatSetValues_SeqAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode);
264: PETSC_INTERN PetscErrorCode MatGetRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
265: PETSC_INTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
266: PETSC_INTERN PetscErrorCode MatAXPY_SeqAIJ(Mat,PetscScalar,Mat,MatStructure);
267: PETSC_INTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
268: PETSC_INTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
269: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
270: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
271: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat);
272: PETSC_INTERN PetscErrorCode MatSetUp_SeqAIJ(Mat);
273: PETSC_INTERN PetscErrorCode MatView_SeqAIJ(Mat,PetscViewer);
275: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat);
276: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal_Inode(Mat);
277: PETSC_INTERN PetscErrorCode Mat_CheckInode(Mat,PetscBool);
278: PETSC_INTERN PetscErrorCode Mat_CheckInode_FactorLU(Mat,PetscBool);
280: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat,Mat,PetscInt*);
282: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat,MatType,MatReuse,Mat*);
283: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat,MatType,MatReuse,Mat*);
284: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJPERM(Mat,MatType,MatReuse,Mat*);
285: PETSC_INTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat,PetscReal,IS,IS);
286: PETSC_INTERN PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
287: PETSC_INTERN PetscErrorCode MatRARt_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
288: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat);
289: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat,MatAssemblyType);
290: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat);
293: /*
294: PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage
296: Input Parameters:
297: + nnz - the number of entries
298: . r - the array of vector values
299: . xv - the matrix values for the row
300: - xi - the column indices of the nonzeros in the row
302: Output Parameter:
303: . sum - negative the sum of results
305: PETSc compile flags:
306: + PETSC_KERNEL_USE_UNROLL_4 - don't use this; it changes nnz and hence is WRONG
307: - PETSC_KERNEL_USE_UNROLL_2 -
309: .seealso: PetscSparseDensePlusDot()
311: */
312: #if defined(PETSC_KERNEL_USE_UNROLL_4)
313: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
314: if (nnz > 0) { \
315: switch (nnz & 0x3) { \
316: case 3: sum -= *xv++ *r[*xi++]; \
317: case 2: sum -= *xv++ *r[*xi++]; \
318: case 1: sum -= *xv++ *r[*xi++]; \
319: nnz -= 4;} \
320: while (nnz > 0) { \
321: sum -= xv[0] * r[xi[0]] - xv[1] * r[xi[1]] - \
322: xv[2] * r[xi[2]] - xv[3] * r[xi[3]]; \
323: xv += 4; xi += 4; nnz -= 4; }}}
325: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
326: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
327: PetscInt __i,__i1,__i2; \
328: for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \
329: sum -= (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \
330: if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]];}
332: #else
333: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
334: PetscInt __i; \
335: for (__i=0; __i<nnz; __i++) sum -= xv[__i] * r[xi[__i]];}
336: #endif
340: /*
341: PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage
343: Input Parameters:
344: + nnz - the number of entries
345: . r - the array of vector values
346: . xv - the matrix values for the row
347: - xi - the column indices of the nonzeros in the row
349: Output Parameter:
350: . sum - the sum of results
352: PETSc compile flags:
353: + PETSC_KERNEL_USE_UNROLL_4 - don't use this; it changes nnz and hence is WRONG
354: - PETSC_KERNEL_USE_UNROLL_2 -
356: .seealso: PetscSparseDenseMinusDot()
358: */
359: #if defined(PETSC_KERNEL_USE_UNROLL_4)
360: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
361: if (nnz > 0) { \
362: switch (nnz & 0x3) { \
363: case 3: sum += *xv++ *r[*xi++]; \
364: case 2: sum += *xv++ *r[*xi++]; \
365: case 1: sum += *xv++ *r[*xi++]; \
366: nnz -= 4;} \
367: while (nnz > 0) { \
368: sum += xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + \
369: xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
370: xv += 4; xi += 4; nnz -= 4; }}}
372: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
373: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
374: PetscInt __i,__i1,__i2; \
375: for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \
376: sum += (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \
377: if (nnz & 0x1) sum += xv[__i] * r[xi[__i]];}
379: #else
380: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
381: PetscInt __i; \
382: for (__i=0; __i<nnz; __i++) sum += xv[__i] * r[xi[__i]];}
383: #endif
385: #endif