1 #pragma once 2 3 #include <petsc/private/matimpl.h> 4 #include <petsc/private/hashmapi.h> 5 #include <petsc/private/hashmapijv.h> 6 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 PetscMPIInt nrqs, nrqr; 13 PetscInt **rbuf1, **rbuf2, **rbuf3, **sbuf1, **sbuf2; 14 PetscInt **ptr; 15 PetscInt *tmp; 16 PetscInt *ctr; 17 PetscMPIInt *pa; /* process array */ 18 PetscInt *req_size; 19 PetscMPIInt *req_source1, *req_source2; 20 PetscBool allcolumns, allrows; 21 PetscBool singleis; 22 PetscMPIInt *row2proc; /* row to process (MPI rank) map */ 23 PetscInt nstages; 24 #if defined(PETSC_USE_CTABLE) 25 PetscHMapI cmap, rmap; 26 PetscInt *cmap_loc, *rmap_loc; 27 #else 28 PetscInt *cmap, *rmap; 29 #endif 30 PetscErrorCode (*destroy)(Mat); 31 } Mat_SubSppt; 32 33 /* Operations provided by MATSEQAIJ and its subclasses */ 34 typedef struct { 35 PetscErrorCode (*getarray)(Mat, PetscScalar **); 36 PetscErrorCode (*restorearray)(Mat, PetscScalar **); 37 PetscErrorCode (*getarrayread)(Mat, const PetscScalar **); 38 PetscErrorCode (*restorearrayread)(Mat, const PetscScalar **); 39 PetscErrorCode (*getarraywrite)(Mat, PetscScalar **); 40 PetscErrorCode (*restorearraywrite)(Mat, PetscScalar **); 41 PetscErrorCode (*getcsrandmemtype)(Mat, const PetscInt **, const PetscInt **, PetscScalar **, PetscMemType *); 42 } Mat_SeqAIJOps; 43 44 /* 45 Struct header shared by SeqAIJ, SeqBAIJ and SeqSBAIJ matrix formats 46 */ 47 #define SEQAIJHEADER(datatype) \ 48 PetscBool roworiented; /* if true, row-oriented input, default */ \ 49 PetscInt nonew; /* 1 don't add new nonzeros, -1 generate error on new */ \ 50 PetscInt nounused; /* -1 generate error on unused space */ \ 51 PetscInt maxnz; /* allocated nonzeros */ \ 52 PetscInt *imax; /* maximum space allocated for each row */ \ 53 PetscInt *ilen; /* actual length of each row */ \ 54 PetscInt *ipre; /* space preallocated for each row by user */ \ 55 PetscBool free_imax_ilen; \ 56 PetscInt reallocs; /* number of mallocs done during MatSetValues() \ 57 as more values are set than were prealloced */ \ 58 PetscInt rmax; /* max nonzeros in any row */ \ 59 PetscBool keepnonzeropattern; /* keeps matrix nonzero structure same in calls to MatZeroRows()*/ \ 60 PetscBool ignorezeroentries; \ 61 PetscBool free_ij; /* free the column indices j and row offsets i when the matrix is destroyed */ \ 62 PetscBool free_a; /* free the numerical values when matrix is destroy */ \ 63 Mat_CompressedRow compressedrow; /* use compressed row format */ \ 64 PetscInt nz; /* nonzeros */ \ 65 PetscInt *i; /* pointer to beginning of each row */ \ 66 PetscInt *j; /* column values: j + i[k] - 1 is start of row k */ \ 67 PetscInt *diag; /* pointers to diagonal elements */ \ 68 PetscInt nonzerorowcnt; /* how many rows have nonzero entries */ \ 69 PetscBool free_diag; \ 70 datatype *a; /* nonzero elements */ \ 71 PetscScalar *solve_work; /* work space used in MatSolve */ \ 72 IS row, col, icol; /* index sets, used for reorderings */ \ 73 PetscBool pivotinblocks; /* pivot inside factorization of each diagonal block */ \ 74 Mat parent; /* set if this matrix was formed with MatDuplicate(...,MAT_SHARE_NONZERO_PATTERN,....); \ 75 means that this shares some data structures with the parent including diag, ilen, imax, i, j */ \ 76 Mat_SubSppt *submatis1; /* used by MatCreateSubMatrices_MPIXAIJ_Local */ \ 77 Mat_SeqAIJOps ops[1] /* operations for SeqAIJ and its subclasses */ 78 79 typedef struct { 80 MatTransposeColoring matcoloring; 81 Mat Bt_den; /* dense matrix of B^T */ 82 Mat ABt_den; /* dense matrix of A*B^T */ 83 PetscBool usecoloring; 84 } Mat_MatMatTransMult; 85 86 typedef struct { /* used by MatTransposeMatMult() */ 87 Mat At; /* transpose of the first matrix */ 88 Mat mA; /* maij matrix of A */ 89 Vec bt, ct; /* vectors to hold locally transposed arrays of B and C */ 90 /* used by PtAP */ 91 void *data; 92 PetscErrorCode (*destroy)(void *); 93 } Mat_MatTransMatMult; 94 95 typedef struct { 96 PetscInt *api, *apj; /* symbolic structure of A*P */ 97 PetscScalar *apa; /* temporary array for storing one row of A*P */ 98 } Mat_AP; 99 100 typedef struct { 101 MatTransposeColoring matcoloring; 102 Mat Rt; /* sparse or dense matrix of R^T */ 103 Mat RARt; /* dense matrix of R*A*R^T */ 104 Mat ARt; /* A*R^T used for the case -matrart_color_art */ 105 MatScalar *work; /* work array to store columns of A*R^T used in MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqDense() */ 106 /* free intermediate products needed for PtAP */ 107 void *data; 108 PetscErrorCode (*destroy)(void *); 109 } Mat_RARt; 110 111 typedef struct { 112 Mat BC; /* temp matrix for storing B*C */ 113 } Mat_MatMatMatMult; 114 115 /* 116 MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix 117 format) or compressed sparse row (CSR). The i[] and j[] arrays start at 0. For example, 118 j[i[k]+p] is the pth column in row k. Note that the diagonal 119 matrix elements are stored with the rest of the nonzeros (not separately). 120 */ 121 122 /* Info about i-nodes (identical nodes) helper class for SeqAIJ */ 123 typedef struct { 124 MatScalar *bdiag, *ibdiag, *ssor_work; /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */ 125 PetscInt bdiagsize; /* length of bdiag and ibdiag */ 126 PetscBool ibdiagvalid; /* do ibdiag[] and bdiag[] contain the most recent values */ 127 128 PetscBool use; 129 PetscInt node_count; /* number of inodes */ 130 PetscInt *size; /* size of each inode */ 131 PetscInt limit; /* inode limit */ 132 PetscInt max_limit; /* maximum supported inode limit */ 133 PetscBool checked; /* if inodes have been checked for */ 134 PetscObjectState mat_nonzerostate; /* non-zero state when inodes were checked for */ 135 } Mat_SeqAIJ_Inode; 136 137 PETSC_INTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat, PetscViewer); 138 PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat, MatAssemblyType); 139 PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat); 140 PETSC_INTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat); 141 PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat, MatOption, PetscBool); 142 PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat, MatDuplicateOption, Mat *); 143 PETSC_INTERN PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat, Mat, MatDuplicateOption, PetscBool); 144 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat, Mat, const MatFactorInfo *); 145 PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat, PetscScalar **); 146 PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat, PetscScalar **); 147 148 typedef struct { 149 SEQAIJHEADER(MatScalar); 150 Mat_SeqAIJ_Inode inode; 151 MatScalar *saved_values; /* location for stashing nonzero values of matrix */ 152 153 PetscScalar *idiag, *mdiag, *ssor_work; /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */ 154 PetscBool idiagvalid; /* current idiag[] and mdiag[] are valid */ 155 PetscScalar *ibdiag; /* inverses of block diagonals */ 156 PetscBool ibdiagvalid; /* inverses of block diagonals are valid. */ 157 PetscBool diagonaldense; /* all entries along the diagonal have been set; i.e. no missing diagonal terms */ 158 PetscScalar fshift, omega; /* last used omega and fshift */ 159 160 /* MatSetValues() via hash related fields */ 161 PetscHMapIJV ht; 162 PetscInt *dnz; 163 struct _MatOps cops; 164 } Mat_SeqAIJ; 165 166 typedef struct { 167 PetscInt nz; /* nz of the matrix after assembly */ 168 PetscCount n; /* Number of entries in MatSetPreallocationCOO() */ 169 PetscCount Atot; /* Total number of valid (i.e., w/ non-negative indices) entries in the COO array */ 170 PetscCount *jmap; /* perm[jmap[i]..jmap[i+1]) give indices of entries in v[] associated with i-th nonzero of the matrix */ 171 PetscCount *perm; /* The permutation array in sorting (i,j) by row and then by col */ 172 } MatCOOStruct_SeqAIJ; 173 174 #define MatSeqXAIJGetOptions_Private(A) \ 175 { \ 176 const PetscBool oldvalues = (PetscBool)(A != PETSC_NULLPTR); \ 177 PetscInt nonew = 0, nounused = 0; \ 178 PetscBool roworiented = PETSC_FALSE; \ 179 if (oldvalues) { \ 180 nonew = ((Mat_SeqAIJ *)A->data)->nonew; \ 181 nounused = ((Mat_SeqAIJ *)A->data)->nounused; \ 182 roworiented = ((Mat_SeqAIJ *)A->data)->roworiented; \ 183 } \ 184 (void)0 185 186 #define MatSeqXAIJRestoreOptions_Private(A) \ 187 if (oldvalues) { \ 188 ((Mat_SeqAIJ *)A->data)->nonew = nonew; \ 189 ((Mat_SeqAIJ *)A->data)->nounused = nounused; \ 190 ((Mat_SeqAIJ *)A->data)->roworiented = roworiented; \ 191 } \ 192 } \ 193 (void)0 194 195 static inline PetscErrorCode MatXAIJAllocatea(Mat A, PetscInt nz, PetscScalar **array) 196 { 197 Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data; 198 199 PetscFunctionBegin; 200 PetscCall(PetscShmgetAllocateArray(nz, sizeof(PetscScalar), (void **)array)); 201 a->free_a = PETSC_TRUE; 202 PetscFunctionReturn(PETSC_SUCCESS); 203 } 204 205 static inline PetscErrorCode MatXAIJDeallocatea(Mat A, PetscScalar **array) 206 { 207 Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data; 208 209 PetscFunctionBegin; 210 if (a->free_a) PetscCall(PetscShmgetDeallocateArray((void **)array)); 211 a->free_a = PETSC_FALSE; 212 PetscFunctionReturn(PETSC_SUCCESS); 213 } 214 215 /* 216 Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types 217 */ 218 static inline PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA, MatScalar **a, PetscInt **j, PetscInt **i) 219 { 220 Mat_SeqAIJ *A = (Mat_SeqAIJ *)AA->data; 221 222 PetscFunctionBegin; 223 if (A->free_a) PetscCall(PetscShmgetDeallocateArray((void **)a)); 224 if (A->free_ij) PetscCall(PetscShmgetDeallocateArray((void **)j)); 225 if (A->free_ij) PetscCall(PetscShmgetDeallocateArray((void **)i)); 226 PetscFunctionReturn(PETSC_SUCCESS); 227 } 228 /* 229 Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types 230 This is a macro because it takes the datatype as an argument which can be either a Mat or a MatScalar 231 */ 232 #define MatSeqXAIJReallocateAIJ(Amat, AM, BS2, NROW, ROW, COL, RMAX, AA, AI, AJ, RP, AP, AIMAX, NONEW, datatype) \ 233 do { \ 234 if (NROW >= RMAX) { \ 235 Mat_SeqAIJ *Ain = (Mat_SeqAIJ *)Amat->data; \ 236 PetscInt CHUNKSIZE = 15, new_nz = AI[AM] + CHUNKSIZE, len, *new_i = NULL, *new_j = NULL; \ 237 datatype *new_a; \ 238 \ 239 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); \ 240 /* malloc new storage space */ \ 241 PetscCall(PetscShmgetAllocateArray(BS2 *new_nz, sizeof(PetscScalar), (void **)&new_a)); \ 242 PetscCall(PetscShmgetAllocateArray(new_nz, sizeof(PetscInt), (void **)&new_j)); \ 243 PetscCall(PetscShmgetAllocateArray(AM + 1, sizeof(PetscInt), (void **)&new_i)); \ 244 Ain->free_a = PETSC_TRUE; \ 245 Ain->free_ij = PETSC_TRUE; \ 246 /* copy over old data into new slots */ \ 247 for (ii = 0; ii < ROW + 1; ii++) new_i[ii] = AI[ii]; \ 248 for (ii = ROW + 1; ii < AM + 1; ii++) new_i[ii] = AI[ii] + CHUNKSIZE; \ 249 PetscCall(PetscArraycpy(new_j, AJ, AI[ROW] + NROW)); \ 250 len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \ 251 PetscCall(PetscArraycpy(new_j + AI[ROW] + NROW + CHUNKSIZE, PetscSafePointerPlusOffset(AJ, AI[ROW] + NROW), len)); \ 252 PetscCall(PetscArraycpy(new_a, AA, BS2 *(AI[ROW] + NROW))); \ 253 PetscCall(PetscArrayzero(new_a + BS2 * (AI[ROW] + NROW), BS2 * CHUNKSIZE)); \ 254 PetscCall(PetscArraycpy(new_a + BS2 * (AI[ROW] + NROW + CHUNKSIZE), PetscSafePointerPlusOffset(AA, BS2 * (AI[ROW] + NROW)), BS2 * len)); \ 255 /* free up old matrix storage */ \ 256 PetscCall(MatSeqXAIJFreeAIJ(A, &Ain->a, &Ain->j, &Ain->i)); \ 257 AA = new_a; \ 258 Ain->a = new_a; \ 259 AI = Ain->i = new_i; \ 260 AJ = Ain->j = new_j; \ 261 \ 262 RP = AJ + AI[ROW]; \ 263 AP = AA + BS2 * AI[ROW]; \ 264 RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \ 265 Ain->maxnz += BS2 * CHUNKSIZE; \ 266 Ain->reallocs++; \ 267 Amat->nonzerostate++; \ 268 } \ 269 } while (0) 270 271 #define MatSeqXAIJReallocateAIJ_structure_only(Amat, AM, BS2, NROW, ROW, COL, RMAX, AI, AJ, RP, AIMAX, NONEW, datatype) \ 272 do { \ 273 if (NROW >= RMAX) { \ 274 Mat_SeqAIJ *Ain = (Mat_SeqAIJ *)Amat->data; \ 275 /* there is no extra room in row, therefore enlarge */ \ 276 PetscInt CHUNKSIZE = 15, new_nz = AI[AM] + CHUNKSIZE, len, *new_i = NULL, *new_j = NULL; \ 277 \ 278 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); \ 279 /* malloc new storage space */ \ 280 PetscCall(PetscShmgetAllocateArray(new_nz, sizeof(PetscInt), (void **)&new_j)); \ 281 PetscCall(PetscShmgetAllocateArray(AM + 1, sizeof(PetscInt), (void **)&new_i)); \ 282 Ain->free_a = PETSC_FALSE; \ 283 Ain->free_ij = PETSC_TRUE; \ 284 \ 285 /* copy over old data into new slots */ \ 286 for (ii = 0; ii < ROW + 1; ii++) new_i[ii] = AI[ii]; \ 287 for (ii = ROW + 1; ii < AM + 1; ii++) new_i[ii] = AI[ii] + CHUNKSIZE; \ 288 PetscCall(PetscArraycpy(new_j, AJ, AI[ROW] + NROW)); \ 289 len = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \ 290 PetscCall(PetscArraycpy(new_j + AI[ROW] + NROW + CHUNKSIZE, AJ + AI[ROW] + NROW, len)); \ 291 \ 292 /* free up old matrix storage */ \ 293 PetscCall(MatSeqXAIJFreeAIJ(A, &Ain->a, &Ain->j, &Ain->i)); \ 294 Ain->a = NULL; \ 295 AI = Ain->i = new_i; \ 296 AJ = Ain->j = new_j; \ 297 \ 298 RP = AJ + AI[ROW]; \ 299 RMAX = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \ 300 Ain->maxnz += BS2 * CHUNKSIZE; \ 301 Ain->reallocs++; \ 302 Amat->nonzerostate++; \ 303 } \ 304 } while (0) 305 306 PETSC_INTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat, PetscInt, const PetscInt *); 307 PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_SeqAIJ(Mat, PetscCount, PetscInt[], PetscInt[]); 308 309 PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat, Mat, IS, IS, const MatFactorInfo *); 310 PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat, Mat, IS, IS, const MatFactorInfo *); 311 312 PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat, Mat, IS, const MatFactorInfo *); 313 PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat, Mat, IS, const MatFactorInfo *); 314 PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat, Mat, const MatFactorInfo *); 315 PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat, Mat, const MatFactorInfo *); 316 PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat, MatDuplicateOption, Mat *); 317 PETSC_INTERN PetscErrorCode MatCopy_SeqAIJ(Mat, Mat, MatStructure); 318 PETSC_INTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat, PetscBool *, PetscInt *); 319 PETSC_INTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat); 320 PETSC_INTERN PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat, PetscInt *, PetscInt **); 321 322 PETSC_INTERN PetscErrorCode MatMult_SeqAIJ(Mat, Vec, Vec); 323 PETSC_INTERN PetscErrorCode MatMult_SeqAIJ_Inode(Mat, Vec, Vec); 324 PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat, Vec, Vec, Vec); 325 PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ_Inode(Mat, Vec, Vec, Vec); 326 PETSC_INTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat, Vec, Vec); 327 PETSC_INTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat, Vec, Vec, Vec); 328 PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ(Mat, Vec, PetscReal, MatSORType, PetscReal, PetscInt, PetscInt, Vec); 329 PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ_Inode(Mat, Vec, PetscReal, MatSORType, PetscReal, PetscInt, PetscInt, Vec); 330 331 PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ(Mat, MatOption, PetscBool); 332 333 PETSC_INTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat, PetscInt *[], PetscInt *[]); 334 PETSC_INTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat, PetscInt *[], PetscInt *[]); 335 PETSC_INTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat, PetscInt, PetscInt, PetscInt *[], PetscInt *[]); 336 PETSC_INTERN PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat, Mat *); 337 PETSC_INTERN PetscErrorCode MatTranspose_SeqAIJ(Mat, MatReuse, Mat *); 338 339 PETSC_INTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt, PetscInt *, PetscInt *, PetscBool, PetscInt, PetscInt, PetscInt **, PetscInt **); 340 PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat, Mat, IS, IS, const MatFactorInfo *); 341 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat, Mat, const MatFactorInfo *); 342 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat, Mat, const MatFactorInfo *); 343 PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat, Mat, const MatFactorInfo *); 344 PETSC_INTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat, IS, IS, const MatFactorInfo *); 345 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat, Vec, Vec); 346 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ(Mat, Vec, Vec); 347 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat, Vec, Vec); 348 PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat, Vec, Vec); 349 PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat, Vec, Vec, Vec); 350 PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat, Vec, Vec); 351 PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat, Vec, Vec); 352 PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat, Vec, Vec, Vec); 353 PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat, Vec, Vec, Vec); 354 PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat, Mat, Mat); 355 PETSC_INTERN PetscErrorCode MatMatSolveTranspose_SeqAIJ(Mat, Mat, Mat); 356 PETSC_INTERN PetscErrorCode MatEqual_SeqAIJ(Mat, Mat, PetscBool *); 357 PETSC_INTERN PetscErrorCode MatFDColoringCreate_SeqXAIJ(Mat, ISColoring, MatFDColoring); 358 PETSC_INTERN PetscErrorCode MatFDColoringSetUp_SeqXAIJ(Mat, ISColoring, MatFDColoring); 359 PETSC_INTERN PetscErrorCode MatFDColoringSetUpBlocked_AIJ_Private(Mat, MatFDColoring, PetscInt); 360 PETSC_INTERN PetscErrorCode MatLoad_AIJ_HDF5(Mat, PetscViewer); 361 PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ_Binary(Mat, PetscViewer); 362 PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ(Mat, PetscViewer); 363 PETSC_INTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat); 364 365 #if defined(PETSC_HAVE_HYPRE) 366 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_Transpose_AIJ_AIJ(Mat); 367 #endif 368 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqAIJ(Mat); 369 370 PETSC_INTERN PetscErrorCode MatProductSymbolic_SeqAIJ_SeqAIJ(Mat); 371 PETSC_INTERN PetscErrorCode MatProductSymbolic_PtAP_SeqAIJ_SeqAIJ(Mat); 372 PETSC_INTERN PetscErrorCode MatProductSymbolic_RARt_SeqAIJ_SeqAIJ(Mat); 373 374 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat); 375 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(Mat, Mat, PetscReal, Mat); 376 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat, Mat, PetscReal, Mat); 377 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat, Mat, PetscReal, Mat); 378 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat, Mat, PetscReal, Mat); 379 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat, Mat, PetscReal, Mat); 380 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat, Mat, PetscReal, Mat); 381 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(Mat, Mat, PetscReal, Mat); 382 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat, Mat, PetscReal, Mat); 383 #if defined(PETSC_HAVE_HYPRE) 384 PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat, Mat, PetscReal, Mat); 385 #endif 386 387 PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat); 388 PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(Mat, Mat, Mat); 389 390 PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat, Mat, Mat); 391 PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat, Mat, Mat); 392 393 PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy(Mat, Mat, PetscReal, Mat); 394 PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat); 395 PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy(Mat, Mat, Mat); 396 397 PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat); 398 PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_matmattransposemult(Mat, Mat, PetscReal, Mat); 399 PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_colorrart(Mat, Mat, PetscReal, Mat); 400 PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat); 401 PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_matmattransposemult(Mat, Mat, Mat); 402 PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_colorrart(Mat, Mat, Mat); 403 404 PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat); 405 PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat); 406 PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_MatTransMatMult(void *); 407 408 PETSC_INTERN PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat, Mat, PetscReal, Mat); 409 PETSC_INTERN PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat, Mat, Mat); 410 PETSC_INTERN PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat, ISColoring, MatTransposeColoring); 411 PETSC_INTERN PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring, Mat, Mat); 412 PETSC_INTERN PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring, Mat, Mat); 413 414 PETSC_INTERN PetscErrorCode MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ(Mat, Mat, Mat, PetscReal, Mat); 415 PETSC_INTERN PetscErrorCode MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ(Mat, Mat, Mat, Mat); 416 417 PETSC_INTERN PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat, PetscInt, PetscInt, PetscRandom); 418 PETSC_INTERN PetscErrorCode MatSetValues_SeqAIJ(Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[], const PetscScalar[], InsertMode); 419 PETSC_INTERN PetscErrorCode MatGetRow_SeqAIJ(Mat, PetscInt, PetscInt *, PetscInt **, PetscScalar **); 420 PETSC_INTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat, PetscInt, PetscInt *, PetscInt **, PetscScalar **); 421 PETSC_INTERN PetscErrorCode MatScale_SeqAIJ(Mat, PetscScalar); 422 PETSC_INTERN PetscErrorCode MatDiagonalScale_SeqAIJ(Mat, Vec, Vec); 423 PETSC_INTERN PetscErrorCode MatDiagonalSet_SeqAIJ(Mat, Vec, InsertMode); 424 PETSC_INTERN PetscErrorCode MatAXPY_SeqAIJ(Mat, PetscScalar, Mat, MatStructure); 425 PETSC_INTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *); 426 PETSC_INTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *); 427 PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *); 428 PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscBool *); 429 PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscInt *[], PetscBool *); 430 PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat, PetscInt, PetscBool, PetscBool, PetscInt *, const PetscInt *[], const PetscInt *[], PetscInt *[], PetscBool *); 431 PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat); 432 PETSC_INTERN PetscErrorCode MatView_SeqAIJ(Mat, PetscViewer); 433 434 PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat); 435 PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal_Inode(Mat); 436 PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode(Mat); 437 PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode_FactorLU(Mat); 438 439 PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat, Mat, PetscInt *); 440 441 #if defined(PETSC_HAVE_MATLAB) 442 PETSC_EXTERN PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject, void *); 443 PETSC_EXTERN PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject, void *); 444 #endif 445 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat, MatType, MatReuse, Mat *); 446 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat, MatType, MatReuse, Mat *); 447 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqDense(Mat, MatType, MatReuse, Mat *); 448 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat, MatType, MatReuse, Mat *); 449 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat, MatType, MatReuse, Mat *); 450 #if defined(PETSC_HAVE_SCALAPACK) 451 PETSC_INTERN PetscErrorCode MatConvert_AIJ_ScaLAPACK(Mat, MatType, MatReuse, Mat *); 452 #endif 453 PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat, MatType, MatReuse, Mat *); 454 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJPERM(Mat, MatType, MatReuse, Mat *); 455 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJSELL(Mat, MatType, MatReuse, Mat *); 456 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJMKL(Mat, MatType, MatReuse, Mat *); 457 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJViennaCL(Mat, MatType, MatReuse, Mat *); 458 PETSC_INTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat, PetscReal, IS, IS); 459 PETSC_INTERN PetscErrorCode MatRARt_SeqAIJ_SeqAIJ(Mat, Mat, MatReuse, PetscReal, Mat *); 460 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat); 461 PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat, MatAssemblyType); 462 PETSC_EXTERN PetscErrorCode MatZeroEntries_SeqAIJ(Mat); 463 464 PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt, const PetscInt *, const PetscInt *, const PetscInt *, const PetscInt *, PetscInt *); 465 PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm, Mat, PetscInt, MatReuse, Mat *); 466 PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIAIJ(MPI_Comm, Mat, PetscInt, MatReuse, Mat *); 467 468 PETSC_INTERN PetscErrorCode MatSetSeqMat_SeqAIJ(Mat, IS, IS, MatStructure, Mat); 469 PETSC_INTERN PetscErrorCode MatEliminateZeros_SeqAIJ(Mat, PetscBool); 470 PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt *); 471 PETSC_INTERN PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat); 472 PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Dummy(Mat); 473 PETSC_INTERN PetscErrorCode MatDestroySubMatrices_Dummy(PetscInt, Mat *[]); 474 PETSC_INTERN PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat, IS, IS, PetscInt, MatReuse, Mat *); 475 476 PETSC_INTERN PetscErrorCode MatSetSeqAIJWithArrays_private(MPI_Comm, PetscInt, PetscInt, PetscInt[], PetscInt[], PetscScalar[], MatType, Mat); 477 478 PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat, ISLocalToGlobalMapping *); 479 480 /* 481 PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage 482 483 Input Parameters: 484 + nnz - the number of entries 485 . r - the array of vector values 486 . xv - the matrix values for the row 487 - xi - the column indices of the nonzeros in the row 488 489 Output Parameter: 490 . sum - negative the sum of results 491 492 PETSc compile flags: 493 + PETSC_KERNEL_USE_UNROLL_4 494 - PETSC_KERNEL_USE_UNROLL_2 495 496 Developer Note: 497 The macro changes sum but not other parameters 498 499 .seealso: `PetscSparseDensePlusDot()` 500 */ 501 #if defined(PETSC_KERNEL_USE_UNROLL_4) 502 #define PetscSparseDenseMinusDot(sum, r, xv, xi, nnz) \ 503 do { \ 504 if (nnz > 0) { \ 505 PetscInt nnz2 = nnz, rem = nnz & 0x3; \ 506 switch (rem) { \ 507 case 3: \ 508 sum -= *xv++ * r[*xi++]; \ 509 case 2: \ 510 sum -= *xv++ * r[*xi++]; \ 511 case 1: \ 512 sum -= *xv++ * r[*xi++]; \ 513 nnz2 -= rem; \ 514 } \ 515 while (nnz2 > 0) { \ 516 sum -= xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \ 517 xv += 4; \ 518 xi += 4; \ 519 nnz2 -= 4; \ 520 } \ 521 xv -= nnz; \ 522 xi -= nnz; \ 523 } \ 524 } while (0) 525 526 #elif defined(PETSC_KERNEL_USE_UNROLL_2) 527 #define PetscSparseDenseMinusDot(sum, r, xv, xi, nnz) \ 528 do { \ 529 PetscInt __i, __i1, __i2; \ 530 for (__i = 0; __i < nnz - 1; __i += 2) { \ 531 __i1 = xi[__i]; \ 532 __i2 = xi[__i + 1]; \ 533 sum -= (xv[__i] * r[__i1] + xv[__i + 1] * r[__i2]); \ 534 } \ 535 if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]]; \ 536 } while (0) 537 538 #else 539 #define PetscSparseDenseMinusDot(sum, r, xv, xi, nnz) \ 540 do { \ 541 PetscInt __i; \ 542 for (__i = 0; __i < nnz; __i++) sum -= xv[__i] * r[xi[__i]]; \ 543 } while (0) 544 #endif 545 546 /* 547 PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage 548 549 Input Parameters: 550 + nnz - the number of entries 551 . r - the array of vector values 552 . xv - the matrix values for the row 553 - xi - the column indices of the nonzeros in the row 554 555 Output Parameter: 556 . sum - the sum of results 557 558 PETSc compile flags: 559 + PETSC_KERNEL_USE_UNROLL_4 560 - PETSC_KERNEL_USE_UNROLL_2 561 562 Developer Note: 563 The macro changes sum but not other parameters 564 565 .seealso: `PetscSparseDenseMinusDot()` 566 */ 567 #if defined(PETSC_KERNEL_USE_UNROLL_4) 568 #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) \ 569 do { \ 570 if (nnz > 0) { \ 571 PetscInt nnz2 = nnz, rem = nnz & 0x3; \ 572 switch (rem) { \ 573 case 3: \ 574 sum += *xv++ * r[*xi++]; \ 575 case 2: \ 576 sum += *xv++ * r[*xi++]; \ 577 case 1: \ 578 sum += *xv++ * r[*xi++]; \ 579 nnz2 -= rem; \ 580 } \ 581 while (nnz2 > 0) { \ 582 sum += xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \ 583 xv += 4; \ 584 xi += 4; \ 585 nnz2 -= 4; \ 586 } \ 587 xv -= nnz; \ 588 xi -= nnz; \ 589 } \ 590 } while (0) 591 592 #elif defined(PETSC_KERNEL_USE_UNROLL_2) 593 #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) \ 594 do { \ 595 PetscInt __i, __i1, __i2; \ 596 for (__i = 0; __i < nnz - 1; __i += 2) { \ 597 __i1 = xi[__i]; \ 598 __i2 = xi[__i + 1]; \ 599 sum += (xv[__i] * r[__i1] + xv[__i + 1] * r[__i2]); \ 600 } \ 601 if (nnz & 0x1) sum += xv[__i] * r[xi[__i]]; \ 602 } while (0) 603 604 #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) 605 #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) PetscSparseDensePlusDot_AVX512_Private(&(sum), (r), (xv), (xi), (nnz)) 606 607 #else 608 #define PetscSparseDensePlusDot(sum, r, xv, xi, nnz) \ 609 do { \ 610 PetscInt __i; \ 611 for (__i = 0; __i < nnz; __i++) sum += xv[__i] * r[xi[__i]]; \ 612 } while (0) 613 #endif 614 615 #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) 616 #include <immintrin.h> 617 #if !defined(_MM_SCALE_8) 618 #define _MM_SCALE_8 8 619 #endif 620 621 static inline void PetscSparseDensePlusDot_AVX512_Private(PetscScalar *sum, const PetscScalar *x, const MatScalar *aa, const PetscInt *aj, PetscInt n) 622 { 623 __m512d vec_x, vec_y, vec_vals; 624 __m256i vec_idx; 625 PetscInt j; 626 627 vec_y = _mm512_setzero_pd(); 628 for (j = 0; j < (n >> 3); j++) { 629 vec_idx = _mm256_loadu_si256((__m256i const *)aj); 630 vec_vals = _mm512_loadu_pd(aa); 631 vec_x = _mm512_i32gather_pd(vec_idx, x, _MM_SCALE_8); 632 vec_y = _mm512_fmadd_pd(vec_x, vec_vals, vec_y); 633 aj += 8; 634 aa += 8; 635 } 636 #if defined(__AVX512VL__) 637 /* masked load requires avx512vl, which is not supported by KNL */ 638 if (n & 0x07) { 639 __mmask8 mask; 640 mask = (__mmask8)(0xff >> (8 - (n & 0x07))); 641 vec_idx = _mm256_mask_loadu_epi32(vec_idx, mask, aj); 642 vec_vals = _mm512_mask_loadu_pd(vec_vals, mask, aa); 643 vec_x = _mm512_mask_i32gather_pd(vec_x, mask, vec_idx, x, _MM_SCALE_8); 644 vec_y = _mm512_mask3_fmadd_pd(vec_x, vec_vals, vec_y, mask); 645 } 646 *sum += _mm512_reduce_add_pd(vec_y); 647 #else 648 *sum += _mm512_reduce_add_pd(vec_y); 649 for (j = 0; j < (n & 0x07); j++) *sum += aa[j] * x[aj[j]]; 650 #endif 651 } 652 #endif 653 654 /* 655 PetscSparseDenseMaxDot - The inner kernel of a modified matrix-vector product \max_i xv[i] * r[xi[i]] for CSR storage 656 657 Input Parameters: 658 + nnz - the number of entries 659 . r - the array of vector values 660 . xv - the matrix values for the row 661 - xi - the column indices of the nonzeros in the row 662 663 Output Parameter: 664 . max - the max of results 665 666 .seealso: `PetscSparseDensePlusDot()`, `PetscSparseDenseMinusDot()` 667 */ 668 #define PetscSparseDenseMaxDot(max, r, xv, xi, nnz) \ 669 do { \ 670 for (PetscInt __i = 0; __i < (nnz); __i++) { max = PetscMax(PetscRealPart(max), PetscRealPart((xv)[__i] * (r)[(xi)[__i]])); } \ 671 } while (0) 672 673 /* 674 Add column indices into table for counting the max nonzeros of merged rows 675 */ 676 #define MatRowMergeMax_SeqAIJ(mat, nrows, ta) \ 677 do { \ 678 if (mat) { \ 679 for (PetscInt _row = 0; _row < (nrows); _row++) { \ 680 const PetscInt _nz = (mat)->i[_row + 1] - (mat)->i[_row]; \ 681 for (PetscInt _j = 0; _j < _nz; _j++) { \ 682 PetscInt *_col = _j + (mat)->j + (mat)->i[_row]; \ 683 PetscCall(PetscHMapISet((ta), *_col + 1, 1)); \ 684 } \ 685 } \ 686 } \ 687 } while (0) 688 689 /* 690 Add column indices into table for counting the nonzeros of merged rows 691 */ 692 #define MatMergeRows_SeqAIJ(mat, nrows, rows, ta) \ 693 do { \ 694 for (PetscInt _i = 0; _i < (nrows); _i++) { \ 695 const PetscInt _row = (rows)[_i]; \ 696 const PetscInt _nz = (mat)->i[_row + 1] - (mat)->i[_row]; \ 697 for (PetscInt _j = 0; _j < _nz; _j++) { \ 698 PetscInt *_col = _j + (mat)->j + (mat)->i[_row]; \ 699 PetscCall(PetscHMapISetWithMode((ta), *_col + 1, 1, INSERT_VALUES)); \ 700 } \ 701 } \ 702 } while (0) 703