1 2 #if !defined(__AIJ_H) 3 #define __AIJ_H 4 5 #include <petsc-private/matimpl.h> 6 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 */ 40 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; 48 49 typedef struct { /* for MatTransposeMatMult_SeqAIJ_SeqDense() */ 50 Mat mA; /* maij matrix of A */ 51 Vec bt,ct; /* vectors to hold locally transposed arrays of B and C */ 52 PetscErrorCode (*destroy)(Mat); 53 } Mat_MatTransMatMult; 54 55 typedef struct { 56 PetscInt *api,*apj; /* symbolic structure of A*P */ 57 PetscScalar *apa; /* temporary array for storing one row of A*P */ 58 PetscErrorCode (*destroy)(Mat); 59 } Mat_PtAP; 60 61 typedef struct { 62 MatTransposeColoring matcoloring; 63 Mat Rt; /* sparse or dense matrix of R^T */ 64 Mat RARt; /* dense matrix of R*A*R^T */ 65 Mat ARt; /* A*R^T used for the case -matrart_color_art */ 66 MatScalar *work; /* work array to store columns of A*R^T used in MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqDense() */ 67 PetscErrorCode (*destroy)(Mat); 68 } Mat_RARt; 69 70 typedef struct { 71 Mat BC; /* temp matrix for storing B*C */ 72 PetscErrorCode (*destroy)(Mat); 73 } Mat_MatMatMatMult; 74 75 typedef struct { /* used by MatGetRedundantMatrix() for reusing matredundant */ 76 PetscInt nzlocal,nsends,nrecvs; 77 PetscMPIInt *send_rank,*recv_rank; 78 PetscInt *sbuf_nz,*rbuf_nz,*sbuf_j,**rbuf_j; 79 PetscScalar *sbuf_a,**rbuf_a; 80 PetscSubcomm psubcomm; 81 PetscErrorCode (*Destroy)(Mat); 82 } Mat_Redundant; 83 84 /* 85 MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix 86 format) or compressed sparse row (CSR). The i[] and j[] arrays start at 0. For example, 87 j[i[k]+p] is the pth column in row k. Note that the diagonal 88 matrix elements are stored with the rest of the nonzeros (not separately). 89 */ 90 91 /* Info about i-nodes (identical nodes) helper class for SeqAIJ */ 92 typedef struct { 93 MatScalar *bdiag,*ibdiag,*ssor_work; /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */ 94 PetscInt bdiagsize; /* length of bdiag and ibdiag */ 95 PetscBool ibdiagvalid; /* do ibdiag[] and bdiag[] contain the most recent values */ 96 97 PetscBool use; 98 PetscInt node_count; /* number of inodes */ 99 PetscInt *size; /* size of each inode */ 100 PetscInt limit; /* inode limit */ 101 PetscInt max_limit; /* maximum supported inode limit */ 102 PetscBool checked; /* if inodes have been checked for */ 103 } Mat_SeqAIJ_Inode; 104 105 PETSC_INTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat,PetscViewer); 106 PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat,MatAssemblyType); 107 PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat); 108 PETSC_INTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat); 109 PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat,MatOption,PetscBool); 110 PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat,MatDuplicateOption,Mat*); 111 PETSC_INTERN PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat,Mat,MatDuplicateOption,PetscBool); 112 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode_inplace(Mat,Mat,const MatFactorInfo*); 113 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat,Mat,const MatFactorInfo*); 114 115 typedef struct { 116 SEQAIJHEADER(MatScalar); 117 Mat_SeqAIJ_Inode inode; 118 MatScalar *saved_values; /* location for stashing nonzero values of matrix */ 119 120 PetscScalar *idiag,*mdiag,*ssor_work; /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */ 121 PetscBool idiagvalid; /* current idiag[] and mdiag[] are valid */ 122 PetscScalar *ibdiag; /* inverses of block diagonals */ 123 PetscBool ibdiagvalid; /* inverses of block diagonals are valid. */ 124 PetscScalar fshift,omega; /* last used omega and fshift */ 125 126 ISColoring coloring; /* set with MatADSetColoring() used by MatADSetValues() */ 127 128 PetscScalar *matmult_abdense; /* used by MatMatMult() */ 129 Mat_PtAP *ptap; /* used by MatPtAP() */ 130 Mat_MatMatMatMult *matmatmatmult; /* used by MatMatMatMult() */ 131 Mat_RARt *rart; /* used by MatRARt() */ 132 Mat_MatMatTransMult *abt; /* used by MatMatTransposeMult() */ 133 Mat_Redundant *redundant; /* used by MatGetRedundantMatrix() */ 134 } Mat_SeqAIJ; 135 136 /* 137 Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types 138 */ 139 #undef __FUNCT__ 140 #define __FUNCT__ "MatSeqXAIJFreeAIJ" 141 PETSC_STATIC_INLINE PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA,MatScalar **a,PetscInt **j,PetscInt **i) 142 { 143 PetscErrorCode ierr; 144 Mat_SeqAIJ *A = (Mat_SeqAIJ*) AA->data; 145 if (A->singlemalloc) { 146 ierr = PetscFree3(*a,*j,*i);CHKERRQ(ierr); 147 } else { 148 if (A->free_a) {ierr = PetscFree(*a);CHKERRQ(ierr);} 149 if (A->free_ij) {ierr = PetscFree(*j);CHKERRQ(ierr);} 150 if (A->free_ij) {ierr = PetscFree(*i);CHKERRQ(ierr);} 151 } 152 return 0; 153 } 154 /* 155 Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types 156 This is a macro because it takes the datatype as an argument which can be either a Mat or a MatScalar 157 */ 158 #define MatSeqXAIJReallocateAIJ(Amat,AM,BS2,NROW,ROW,COL,RMAX,AA,AI,AJ,RP,AP,AIMAX,NONEW,datatype) \ 159 if (NROW >= RMAX) { \ 160 Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data; \ 161 /* there is no extra room in row, therefore enlarge */ \ 162 PetscInt CHUNKSIZE = 15,new_nz = AI[AM] + CHUNKSIZE,len,*new_i=0,*new_j=0; \ 163 datatype *new_a; \ 164 \ 165 if (NONEW == -2) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"New nonzero at (%D,%D) caused a malloc",ROW,COL); \ 166 /* malloc new storage space */ \ 167 ierr = PetscMalloc3(BS2*new_nz,datatype,&new_a,new_nz,PetscInt,&new_j,AM+1,PetscInt,&new_i);CHKERRQ(ierr); \ 168 \ 169 /* copy over old data into new slots */ \ 170 for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \ 171 for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \ 172 ierr = PetscMemcpy(new_j,AJ,(AI[ROW]+NROW)*sizeof(PetscInt));CHKERRQ(ierr); \ 173 len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \ 174 ierr = PetscMemcpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len*sizeof(PetscInt));CHKERRQ(ierr); \ 175 ierr = PetscMemcpy(new_a,AA,BS2*(AI[ROW]+NROW)*sizeof(datatype));CHKERRQ(ierr); \ 176 ierr = PetscMemzero(new_a+BS2*(AI[ROW]+NROW),BS2*CHUNKSIZE*sizeof(datatype));CHKERRQ(ierr); \ 177 ierr = PetscMemcpy(new_a+BS2*(AI[ROW]+NROW+CHUNKSIZE),AA+BS2*(AI[ROW]+NROW),BS2*len*sizeof(datatype));CHKERRQ(ierr); \ 178 /* free up old matrix storage */ \ 179 ierr = MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i);CHKERRQ(ierr); \ 180 AA = new_a; \ 181 Ain->a = (MatScalar*) new_a; \ 182 AI = Ain->i = new_i; AJ = Ain->j = new_j; \ 183 Ain->singlemalloc = PETSC_TRUE; \ 184 \ 185 RP = AJ + AI[ROW]; AP = AA + BS2*AI[ROW]; \ 186 RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \ 187 Ain->maxnz += BS2*CHUNKSIZE; \ 188 Ain->reallocs++; \ 189 } \ 190 191 192 PETSC_EXTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat,PetscInt,const PetscInt*); 193 PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*); 194 PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*); 195 PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat,Mat,IS,IS,const MatFactorInfo*); 196 197 PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*); 198 PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*); 199 PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*); 200 PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*); 201 PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*); 202 PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*); 203 PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat,MatDuplicateOption,Mat*); 204 PETSC_INTERN PetscErrorCode MatCopy_SeqAIJ(Mat,Mat,MatStructure); 205 PETSC_INTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat,PetscBool*,PetscInt*); 206 PETSC_INTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat); 207 PETSC_INTERN PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat,PetscInt*,PetscInt**); 208 209 PETSC_INTERN PetscErrorCode MatMult_SeqAIJ(Mat A,Vec,Vec); 210 PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec,Vec,Vec); 211 PETSC_INTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec,Vec); 212 PETSC_INTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec,Vec,Vec); 213 PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec); 214 215 PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ(Mat,MatOption,PetscBool); 216 PETSC_INTERN PetscErrorCode MatSetColoring_SeqAIJ(Mat,ISColoring); 217 PETSC_INTERN PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat,PetscInt,void*); 218 219 PETSC_INTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]); 220 PETSC_INTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat,PetscInt,PetscInt,PetscInt *[],PetscInt *[]); 221 PETSC_INTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]); 222 PETSC_INTERN PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat,Mat*); 223 PETSC_INTERN PetscErrorCode MatTranspose_SeqAIJ(Mat,MatReuse,Mat*); 224 PETSC_INTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt,PetscInt*,PetscInt*,PetscInt,PetscInt,PetscInt**,PetscInt**); 225 PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*); 226 PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*); 227 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*); 228 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*); 229 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat,Mat,const MatFactorInfo*); 230 PETSC_INTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat,IS,IS,const MatFactorInfo*); 231 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat,Vec,Vec); 232 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ(Mat,Vec,Vec); 233 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode_inplace(Mat,Vec,Vec); 234 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat,Vec,Vec); 235 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering_inplace(Mat,Vec,Vec); 236 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat,Vec,Vec); 237 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_InplaceWithPerm(Mat,Vec,Vec); 238 PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec); 239 PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat,Vec,Vec,Vec); 240 PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat,Vec,Vec); 241 PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat,Vec,Vec); 242 PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec); 243 PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat,Vec,Vec,Vec); 244 PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ_inplace(Mat,Mat,Mat); 245 PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat,Mat,Mat); 246 PETSC_INTERN PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg); 247 PETSC_INTERN PetscErrorCode MatFDColoringCreate_SeqAIJ(Mat,ISColoring,MatFDColoring); 248 PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ(Mat,PetscViewer); 249 PETSC_INTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat); 250 251 PETSC_INTERN PetscErrorCode MatMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*); 252 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*); 253 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,PetscReal,Mat*); 254 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat,Mat,PetscReal,Mat*); 255 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat,Mat,PetscReal,Mat*); 256 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat,Mat,PetscReal,Mat*); 257 PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat); 258 PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,Mat); 259 260 PETSC_INTERN PetscErrorCode MatPtAP_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*); 261 PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_DenseAxpy(Mat,Mat,PetscReal,Mat*); 262 PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,PetscReal,Mat*); 263 PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat); 264 PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,Mat); 265 266 PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*); 267 PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_matmattransposemult(Mat,Mat,PetscReal,Mat*); 268 PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_colorrart(Mat,Mat,PetscReal,Mat*); 269 PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat); 270 PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_matmattransposemult(Mat,Mat,Mat); 271 PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_colorrart(Mat,Mat,Mat); 272 273 PETSC_INTERN PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*); 274 PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*); 275 PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat); 276 277 PETSC_INTERN PetscErrorCode MatTransposeMatMult_SeqAIJ_SeqDense(Mat,Mat,MatReuse,PetscReal,Mat*); 278 PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqDense(Mat,Mat,PetscReal,Mat*); 279 PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqDense(Mat,Mat,Mat); 280 281 PETSC_INTERN PetscErrorCode MatMatTransposeMult_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*); 282 PETSC_INTERN PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat*); 283 PETSC_INTERN PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat); 284 PETSC_INTERN PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat,ISColoring,MatTransposeColoring); 285 PETSC_INTERN PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring,Mat,Mat); 286 PETSC_INTERN PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring,Mat,Mat); 287 288 PETSC_INTERN PetscErrorCode MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*); 289 PETSC_INTERN PetscErrorCode MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,PetscReal,Mat*); 290 PETSC_INTERN PetscErrorCode MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,Mat); 291 292 PETSC_INTERN PetscErrorCode MatSetValues_SeqAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode); 293 PETSC_INTERN PetscErrorCode MatGetRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**); 294 PETSC_INTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**); 295 PETSC_INTERN PetscErrorCode MatAXPY_SeqAIJ(Mat,PetscScalar,Mat,MatStructure); 296 PETSC_INTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*); 297 PETSC_INTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*); 298 PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*); 299 PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*); 300 PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat); 301 PETSC_INTERN PetscErrorCode MatSetUp_SeqAIJ(Mat); 302 PETSC_INTERN PetscErrorCode MatView_SeqAIJ(Mat,PetscViewer); 303 304 PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat); 305 PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal_Inode(Mat); 306 PETSC_INTERN PetscErrorCode Mat_CheckInode(Mat,PetscBool); 307 PETSC_INTERN PetscErrorCode Mat_CheckInode_FactorLU(Mat,PetscBool); 308 309 PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat,Mat,PetscInt*); 310 311 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat,MatType,MatReuse,Mat*); 312 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat,MatType,MatReuse,Mat*); 313 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJPERM(Mat,MatType,MatReuse,Mat*); 314 PETSC_INTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat,PetscReal,IS,IS); 315 PETSC_INTERN PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*); 316 PETSC_INTERN PetscErrorCode MatRARt_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*); 317 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat); 318 PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat,MatAssemblyType); 319 PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat); 320 321 322 /* 323 PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage 324 325 Input Parameters: 326 + nnz - the number of entries 327 . r - the array of vector values 328 . xv - the matrix values for the row 329 - xi - the column indices of the nonzeros in the row 330 331 Output Parameter: 332 . sum - negative the sum of results 333 334 PETSc compile flags: 335 + PETSC_KERNEL_USE_UNROLL_4 - don't use this; it changes nnz and hence is WRONG 336 - PETSC_KERNEL_USE_UNROLL_2 - 337 338 .seealso: PetscSparseDensePlusDot() 339 340 */ 341 #if defined(PETSC_KERNEL_USE_UNROLL_4) 342 #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \ 343 if (nnz > 0) { \ 344 switch (nnz & 0x3) { \ 345 case 3: sum -= *xv++ *r[*xi++]; \ 346 case 2: sum -= *xv++ *r[*xi++]; \ 347 case 1: sum -= *xv++ *r[*xi++]; \ 348 nnz -= 4;} \ 349 while (nnz > 0) { \ 350 sum -= xv[0] * r[xi[0]] - xv[1] * r[xi[1]] - \ 351 xv[2] * r[xi[2]] - xv[3] * r[xi[3]]; \ 352 xv += 4; xi += 4; nnz -= 4; }}} 353 354 #elif defined(PETSC_KERNEL_USE_UNROLL_2) 355 #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \ 356 PetscInt __i,__i1,__i2; \ 357 for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \ 358 sum -= (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \ 359 if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]];} 360 361 #else 362 #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \ 363 PetscInt __i; \ 364 for (__i=0; __i<nnz; __i++) sum -= xv[__i] * r[xi[__i]];} 365 #endif 366 367 368 369 /* 370 PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage 371 372 Input Parameters: 373 + nnz - the number of entries 374 . r - the array of vector values 375 . xv - the matrix values for the row 376 - xi - the column indices of the nonzeros in the row 377 378 Output Parameter: 379 . sum - the sum of results 380 381 PETSc compile flags: 382 + PETSC_KERNEL_USE_UNROLL_4 - don't use this; it changes nnz and hence is WRONG 383 - PETSC_KERNEL_USE_UNROLL_2 - 384 385 .seealso: PetscSparseDenseMinusDot() 386 387 */ 388 #if defined(PETSC_KERNEL_USE_UNROLL_4) 389 #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \ 390 if (nnz > 0) { \ 391 switch (nnz & 0x3) { \ 392 case 3: sum += *xv++ *r[*xi++]; \ 393 case 2: sum += *xv++ *r[*xi++]; \ 394 case 1: sum += *xv++ *r[*xi++]; \ 395 nnz -= 4;} \ 396 while (nnz > 0) { \ 397 sum += xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + \ 398 xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \ 399 xv += 4; xi += 4; nnz -= 4; }}} 400 401 #elif defined(PETSC_KERNEL_USE_UNROLL_2) 402 #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \ 403 PetscInt __i,__i1,__i2; \ 404 for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \ 405 sum += (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \ 406 if (nnz & 0x1) sum += xv[__i] * r[xi[__i]];} 407 408 #else 409 #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \ 410 PetscInt __i; \ 411 for (__i=0; __i<nnz; __i++) sum += xv[__i] * r[xi[__i]];} 412 #endif 413 414 #endif 415