Actual source code: aij.h



  5: #include <petsc/private/matimpl.h>
  6: #include <petscctable.h>

  8: /* Operations provided by MATSEQAIJ and its subclasses */
  9: typedef struct {
 10:   PetscErrorCode (*getarray)(Mat,PetscScalar **);
 11:   PetscErrorCode (*restorearray)(Mat,PetscScalar **);
 12:   PetscErrorCode (*getarrayread)(Mat,const PetscScalar **);
 13:   PetscErrorCode (*restorearrayread)(Mat,const PetscScalar **);
 14:   PetscErrorCode (*getarraywrite)(Mat,PetscScalar **);
 15:   PetscErrorCode (*restorearraywrite)(Mat,PetscScalar **);
 16:   PetscErrorCode (*getcsrandmemtype)(Mat,const PetscInt**,const PetscInt**,PetscScalar**,PetscMemType*);
 17: } Mat_SeqAIJOps;

 19: /*
 20:     Struct header shared by SeqAIJ, SeqBAIJ and SeqSBAIJ matrix formats
 21: */
 22: #define SEQAIJHEADER(datatype)        \
 23:   PetscBool roworiented;              /* if true, row-oriented input, default */ \
 24:   PetscInt  nonew;                    /* 1 don't add new nonzeros, -1 generate error on new */ \
 25:   PetscInt  nounused;                 /* -1 generate error on unused space */ \
 26:   PetscBool singlemalloc;             /* if true a, i, and j have been obtained with one big malloc */ \
 27:   PetscInt  maxnz;                    /* allocated nonzeros */ \
 28:   PetscInt  *imax;                    /* maximum space allocated for each row */ \
 29:   PetscInt  *ilen;                    /* actual length of each row */ \
 30:   PetscInt  *ipre;                    /* space preallocated for each row by user */ \
 31:   PetscBool free_imax_ilen;  \
 32:   PetscInt  reallocs;                 /* number of mallocs done during MatSetValues() \
 33:                                         as more values are set than were prealloced */\
 34:   PetscInt          rmax;             /* max nonzeros in any row */ \
 35:   PetscBool         keepnonzeropattern;   /* keeps matrix structure same in calls to MatZeroRows()*/ \
 36:   PetscBool         ignorezeroentries; \
 37:   PetscBool         free_ij;          /* free the column indices j and row offsets i when the matrix is destroyed */ \
 38:   PetscBool         free_a;           /* free the numerical values when matrix is destroy */ \
 39:   Mat_CompressedRow compressedrow;    /* use compressed row format */                      \
 40:   PetscInt          nz;               /* nonzeros */                                       \
 41:   PetscInt          *i;               /* pointer to beginning of each row */               \
 42:   PetscInt          *j;               /* column values: j + i[k] - 1 is start of row k */  \
 43:   PetscInt          *diag;            /* pointers to diagonal elements */                  \
 44:   PetscInt          nonzerorowcnt;    /* how many rows have nonzero entries */             \
 45:   PetscBool         free_diag;         \
 46:   datatype          *a;               /* nonzero elements */                               \
 47:   PetscScalar       *solve_work;      /* work space used in MatSolve */                    \
 48:   IS                row, col, icol;   /* index sets, used for reorderings */ \
 49:   PetscBool         pivotinblocks;    /* pivot inside factorization of each diagonal block */ \
 50:   Mat               parent;           /* set if this matrix was formed with MatDuplicate(...,MAT_SHARE_NONZERO_PATTERN,....); \
 51:                                          means that this shares some data structures with the parent including diag, ilen, imax, i, j */\
 52:   Mat_SubSppt       *submatis1;       /* used by MatCreateSubMatrices_MPIXAIJ_Local */    \
 53:   Mat_SeqAIJOps     ops[1]            /* operations for SeqAIJ and its subclasses */

 55: typedef struct {
 56:   MatTransposeColoring matcoloring;
 57:   Mat                  Bt_den;       /* dense matrix of B^T */
 58:   Mat                  ABt_den;      /* dense matrix of A*B^T */
 59:   PetscBool            usecoloring;
 60: } Mat_MatMatTransMult;

 62: typedef struct { /* used by MatTransposeMatMult() */
 63:   Mat          At;           /* transpose of the first matrix */
 64:   Mat          mA;           /* maij matrix of A */
 65:   Vec          bt,ct;        /* vectors to hold locally transposed arrays of B and C */
 66:   PetscBool    updateAt;     /* flg to avoid recomputing At in MatProductNumeric_AtB_SeqAIJ_SeqAIJ() */
 67:   /* used by PtAP */
 68:   void           *data;
 69:   PetscErrorCode (*destroy)(void*);
 70: } Mat_MatTransMatMult;

 72: typedef struct {
 73:   PetscInt    *api,*apj;       /* symbolic structure of A*P */
 74:   PetscScalar *apa;            /* temporary array for storing one row of A*P */
 75: } Mat_AP;

 77: typedef struct {
 78:   MatTransposeColoring matcoloring;
 79:   Mat                  Rt;    /* sparse or dense matrix of R^T */
 80:   Mat                  RARt;  /* dense matrix of R*A*R^T */
 81:   Mat                  ARt;   /* A*R^T used for the case -matrart_color_art */
 82:   MatScalar            *work; /* work array to store columns of A*R^T used in MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqDense() */
 83:   /* free intermediate products needed for PtAP */
 84:   void                 *data;
 85:   PetscErrorCode       (*destroy)(void*);
 86: } Mat_RARt;

 88: typedef struct {
 89:   Mat BC;               /* temp matrix for storing B*C */
 90: } Mat_MatMatMatMult;

 92: /*
 93:   MATSEQAIJ format - Compressed row storage (also called Yale sparse matrix
 94:   format) or compressed sparse row (CSR).  The i[] and j[] arrays start at 0. For example,
 95:   j[i[k]+p] is the pth column in row k.  Note that the diagonal
 96:   matrix elements are stored with the rest of the nonzeros (not separately).
 97: */

 99: /* Info about i-nodes (identical nodes) helper class for SeqAIJ */
100: typedef struct {
101:   MatScalar        *bdiag,*ibdiag,*ssor_work;        /* diagonal blocks of matrix used for MatSOR_SeqAIJ_Inode() */
102:   PetscInt         bdiagsize;                         /* length of bdiag and ibdiag */
103:   PetscBool        ibdiagvalid;                       /* do ibdiag[] and bdiag[] contain the most recent values */

105:   PetscBool        use;
106:   PetscInt         node_count;                     /* number of inodes */
107:   PetscInt         *size;                          /* size of each inode */
108:   PetscInt         limit;                          /* inode limit */
109:   PetscInt         max_limit;                      /* maximum supported inode limit */
110:   PetscBool        checked;                        /* if inodes have been checked for */
111:   PetscObjectState mat_nonzerostate;               /* non-zero state when inodes were checked for */
112: } Mat_SeqAIJ_Inode;

114: PETSC_INTERN PetscErrorCode MatView_SeqAIJ_Inode(Mat,PetscViewer);
115: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ_Inode(Mat,MatAssemblyType);
116: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_Inode(Mat);
117: PETSC_INTERN PetscErrorCode MatCreate_SeqAIJ_Inode(Mat);
118: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ_Inode(Mat,MatOption,PetscBool);
119: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ_Inode(Mat,MatDuplicateOption,Mat*);
120: PETSC_INTERN PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat,Mat,MatDuplicateOption,PetscBool);
121: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode_inplace(Mat,Mat,const MatFactorInfo*);
122: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_Inode(Mat,Mat,const MatFactorInfo*);
123: PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat,PetscScalar**);
124: PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat,PetscScalar**);

126: typedef struct {
127:   SEQAIJHEADER(MatScalar);
128:   Mat_SeqAIJ_Inode inode;
129:   MatScalar        *saved_values;             /* location for stashing nonzero values of matrix */

131:   PetscScalar *idiag,*mdiag,*ssor_work;       /* inverse of diagonal entries, diagonal values and workspace for Eisenstat trick */
132:   PetscBool   idiagvalid;                     /* current idiag[] and mdiag[] are valid */
133:   PetscScalar *ibdiag;                        /* inverses of block diagonals */
134:   PetscBool   ibdiagvalid;                    /* inverses of block diagonals are valid. */
135:   PetscBool   diagonaldense;                  /* all entries along the diagonal have been set; i.e. no missing diagonal terms */
136:   PetscScalar fshift,omega;                   /* last used omega and fshift */

138:   /* MatSetValuesCOO() related fields on host */
139:   PetscCount  coo_n;  /* Number of entries in MatSetPreallocationCOO() */
140:   PetscCount  Atot;   /* Total number of valid (i.e., w/ non-negative indices) entries in the COO array */
141:   PetscCount  *jmap;  /* perm[jmap[i]..jmap[i+1]) give indices of entries in v[] associated with i-th nonzero of the matrix */
142:   PetscCount  *perm;  /* The permutation array in sorting (i,j) by row and then by col */
143: } Mat_SeqAIJ;

145: /*
146:   Frees the a, i, and j arrays from the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
147: */
148: static inline PetscErrorCode MatSeqXAIJFreeAIJ(Mat AA,MatScalar **a,PetscInt **j,PetscInt **i)
149: {
150:   Mat_SeqAIJ     *A = (Mat_SeqAIJ*) AA->data;
151:   if (A->singlemalloc) {
152:     PetscFree3(*a,*j,*i);
153:   } else {
154:     if (A->free_a)  PetscFree(*a);
155:     if (A->free_ij) PetscFree(*j);
156:     if (A->free_ij) PetscFree(*i);
157:   }
158:   return 0;
159: }
160: /*
161:     Allocates larger a, i, and j arrays for the XAIJ (AIJ, BAIJ, and SBAIJ) matrix types
162:     This is a macro because it takes the datatype as an argument which can be either a Mat or a MatScalar
163: */
164: #define MatSeqXAIJReallocateAIJ(Amat,AM,BS2,NROW,ROW,COL,RMAX,AA,AI,AJ,RP,AP,AIMAX,NONEW,datatype) \
165:   if (NROW >= RMAX) { \
166:     Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data; \
167:     /* there is no extra room in row, therefore enlarge */ \
168:     PetscInt CHUNKSIZE = 15,new_nz = AI[AM] + CHUNKSIZE,len,*new_i=NULL,*new_j=NULL; \
169:     datatype *new_a; \
170:  \
172:     /* malloc new storage space */ \
173:     PetscMalloc3(BS2*new_nz,&new_a,new_nz,&new_j,AM+1,&new_i); \
174:  \
175:     /* copy over old data into new slots */ \
176:     for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \
177:     for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \
178:     PetscArraycpy(new_j,AJ,AI[ROW]+NROW); \
179:     len  = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
180:     PetscArraycpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len); \
181:     PetscArraycpy(new_a,AA,BS2*(AI[ROW]+NROW));    \
182:     PetscArrayzero(new_a+BS2*(AI[ROW]+NROW),BS2*CHUNKSIZE); \
183:     PetscArraycpy(new_a+BS2*(AI[ROW]+NROW+CHUNKSIZE),AA+BS2*(AI[ROW]+NROW),BS2*len);  \
184:     /* free up old matrix storage */ \
185:     MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i); \
186:     AA                = new_a; \
187:     Ain->a            = (MatScalar*) new_a;                   \
188:     AI                = Ain->i = new_i; AJ = Ain->j = new_j;  \
189:     Ain->singlemalloc = PETSC_TRUE; \
190:  \
191:     RP          = AJ + AI[ROW]; AP = AA + BS2*AI[ROW]; \
192:     RMAX        = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
193:     Ain->maxnz += BS2*CHUNKSIZE; \
194:     Ain->reallocs++; \
195:   } \

197: #define MatSeqXAIJReallocateAIJ_structure_only(Amat,AM,BS2,NROW,ROW,COL,RMAX,AI,AJ,RP,AIMAX,NONEW,datatype) \
198:   if (NROW >= RMAX) { \
199:     Mat_SeqAIJ *Ain = (Mat_SeqAIJ*)Amat->data; \
200:     /* there is no extra room in row, therefore enlarge */ \
201:     PetscInt CHUNKSIZE = 15,new_nz = AI[AM] + CHUNKSIZE,len,*new_i=NULL,*new_j=NULL; \
202:  \
204:     /* malloc new storage space */ \
205:     PetscMalloc1(new_nz,&new_j); \
206:     PetscMalloc1(AM+1,&new_i);\
207:  \
208:     /* copy over old data into new slots */ \
209:     for (ii=0; ii<ROW+1; ii++) {new_i[ii] = AI[ii];} \
210:     for (ii=ROW+1; ii<AM+1; ii++) {new_i[ii] = AI[ii]+CHUNKSIZE;} \
211:     PetscArraycpy(new_j,AJ,AI[ROW]+NROW); \
212:     len  = (new_nz - CHUNKSIZE - AI[ROW] - NROW); \
213:     PetscArraycpy(new_j+AI[ROW]+NROW+CHUNKSIZE,AJ+AI[ROW]+NROW,len); \
214:  \
215:     /* free up old matrix storage */ \
216:     MatSeqXAIJFreeAIJ(A,&Ain->a,&Ain->j,&Ain->i); \
217:     Ain->a            = NULL;                   \
218:     AI                = Ain->i = new_i; AJ = Ain->j = new_j;  \
219:     Ain->singlemalloc = PETSC_FALSE; \
220:     Ain->free_a       = PETSC_FALSE;             \
221:  \
222:     RP          = AJ + AI[ROW];    \
223:     RMAX        = AIMAX[ROW] = AIMAX[ROW] + CHUNKSIZE; \
224:     Ain->maxnz += BS2*CHUNKSIZE; \
225:     Ain->reallocs++; \
226:   } \

228: PETSC_INTERN PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat,PetscInt,const PetscInt*);
229: PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_SeqAIJ(Mat,PetscCount,const PetscInt[],const PetscInt[]);
230: PETSC_INTERN PetscErrorCode MatResetPreallocationCOO_SeqAIJ(Mat);

232: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
233: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
234: PETSC_INTERN PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat,Mat,IS,IS,const MatFactorInfo*);

236: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
237: PETSC_INTERN PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
238: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,const MatFactorInfo*);
239: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat,Mat,IS,const MatFactorInfo*);
240: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
241: PETSC_INTERN PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
242: PETSC_INTERN PetscErrorCode MatDuplicate_SeqAIJ(Mat,MatDuplicateOption,Mat*);
243: PETSC_INTERN PetscErrorCode MatCopy_SeqAIJ(Mat,Mat,MatStructure);
244: PETSC_INTERN PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat,PetscBool*,PetscInt*);
245: PETSC_INTERN PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat);
246: PETSC_INTERN PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat,PetscInt*,PetscInt**);

248: PETSC_INTERN PetscErrorCode MatMult_SeqAIJ(Mat,Vec,Vec);
249: PETSC_INTERN PetscErrorCode MatMult_SeqAIJ_Inode(Mat,Vec,Vec);
250: PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ(Mat,Vec,Vec,Vec);
251: PETSC_INTERN PetscErrorCode MatMultAdd_SeqAIJ_Inode(Mat,Vec,Vec,Vec);
252: PETSC_INTERN PetscErrorCode MatMultTranspose_SeqAIJ(Mat,Vec,Vec);
253: PETSC_INTERN PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat,Vec,Vec,Vec);
254: PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec);
255: PETSC_INTERN PetscErrorCode MatSOR_SeqAIJ_Inode(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec);

257: PETSC_INTERN PetscErrorCode MatSetOption_SeqAIJ(Mat,MatOption,PetscBool);

259: PETSC_INTERN PetscErrorCode MatGetSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
260: PETSC_INTERN PetscErrorCode MatGetSymbolicTransposeReduced_SeqAIJ(Mat,PetscInt,PetscInt,PetscInt *[],PetscInt *[]);
261: PETSC_INTERN PetscErrorCode MatRestoreSymbolicTranspose_SeqAIJ(Mat,PetscInt *[],PetscInt *[]);
262: PETSC_INTERN PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat,Mat*);
263: PETSC_INTERN PetscErrorCode MatTranspose_SeqAIJ(Mat,MatReuse,Mat*);
264: PETSC_INTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt,PetscInt*,PetscInt*,PetscBool,PetscInt,PetscInt,PetscInt**,PetscInt**);
265: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ_inplace(Mat,Mat,IS,IS,const MatFactorInfo*);
266: PETSC_INTERN PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat,Mat,IS,IS,const MatFactorInfo*);
267: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat,Mat,const MatFactorInfo*);
268: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat,Mat,const MatFactorInfo*);
269: PETSC_INTERN PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat,Mat,const MatFactorInfo*);
270: PETSC_INTERN PetscErrorCode MatLUFactor_SeqAIJ(Mat,IS,IS,const MatFactorInfo*);
271: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_inplace(Mat,Vec,Vec);
272: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ(Mat,Vec,Vec);
273: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode_inplace(Mat,Vec,Vec);
274: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_Inode(Mat,Vec,Vec);
275: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering_inplace(Mat,Vec,Vec);
276: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat,Vec,Vec);
277: PETSC_INTERN PetscErrorCode MatSolve_SeqAIJ_InplaceWithPerm(Mat,Vec,Vec);
278: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
279: PETSC_INTERN PetscErrorCode MatSolveAdd_SeqAIJ(Mat,Vec,Vec,Vec);
280: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat,Vec,Vec);
281: PETSC_INTERN PetscErrorCode MatSolveTranspose_SeqAIJ(Mat,Vec,Vec);
282: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat,Vec,Vec,Vec);
283: PETSC_INTERN PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat,Vec,Vec,Vec);
284: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ_inplace(Mat,Mat,Mat);
285: PETSC_INTERN PetscErrorCode MatMatSolve_SeqAIJ(Mat,Mat,Mat);
286: PETSC_INTERN PetscErrorCode MatEqual_SeqAIJ(Mat,Mat,PetscBool*);
287: PETSC_INTERN PetscErrorCode MatFDColoringCreate_SeqXAIJ(Mat,ISColoring,MatFDColoring);
288: PETSC_INTERN PetscErrorCode MatFDColoringSetUp_SeqXAIJ(Mat,ISColoring,MatFDColoring);
289: PETSC_INTERN PetscErrorCode MatFDColoringSetUpBlocked_AIJ_Private(Mat,MatFDColoring,PetscInt);
290: PETSC_INTERN PetscErrorCode MatLoad_AIJ_HDF5(Mat,PetscViewer);
291: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ_Binary(Mat,PetscViewer);
292: PETSC_INTERN PetscErrorCode MatLoad_SeqAIJ(Mat,PetscViewer);
293: PETSC_INTERN PetscErrorCode RegisterApplyPtAPRoutines_Private(Mat);

295: #if defined(PETSC_HAVE_HYPRE)
296: PETSC_INTERN PetscErrorCode MatProductSetFromOptions_Transpose_AIJ_AIJ(Mat);
297: #endif
298: PETSC_INTERN PetscErrorCode MatProductSetFromOptions_SeqAIJ(Mat);

300: PETSC_INTERN PetscErrorCode MatProductSymbolic_SeqAIJ_SeqAIJ(Mat);
301: PETSC_INTERN PetscErrorCode MatProductSymbolic_PtAP_SeqAIJ_SeqAIJ(Mat);
302: PETSC_INTERN PetscErrorCode MatProductSymbolic_RARt_SeqAIJ_SeqAIJ(Mat);

304: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat);
305: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Sorted(Mat,Mat,PetscReal,Mat);
306: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat,Mat,PetscReal,Mat);
307: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,PetscReal,Mat);
308: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Scalable_fast(Mat,Mat,PetscReal,Mat);
309: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_Heap(Mat,Mat,PetscReal,Mat);
310: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_BTHeap(Mat,Mat,PetscReal,Mat);
311: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_RowMerge(Mat,Mat,PetscReal,Mat);
312: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_SeqAIJ_SeqAIJ_LLCondensed(Mat,Mat,PetscReal,Mat);
313: #if defined(PETSC_HAVE_HYPRE)
314: PETSC_INTERN PetscErrorCode MatMatMultSymbolic_AIJ_AIJ_wHYPRE(Mat,Mat,PetscReal,Mat);
315: #endif

317: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
318: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted(Mat,Mat,Mat);

320: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat,Mat,Mat);
321: PETSC_INTERN PetscErrorCode MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable(Mat,Mat,Mat);

323: PETSC_INTERN PetscErrorCode MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,PetscReal,Mat);
324: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
325: PETSC_INTERN PetscErrorCode MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy(Mat,Mat,Mat);

327: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat);
328: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_matmattransposemult(Mat,Mat,PetscReal,Mat);
329: PETSC_INTERN PetscErrorCode MatRARtSymbolic_SeqAIJ_SeqAIJ_colorrart(Mat,Mat,PetscReal,Mat);
330: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
331: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_matmattransposemult(Mat,Mat,Mat);
332: PETSC_INTERN PetscErrorCode MatRARtNumeric_SeqAIJ_SeqAIJ_colorrart(Mat,Mat,Mat);

334: PETSC_INTERN PetscErrorCode MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat);
335: PETSC_INTERN PetscErrorCode MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
336: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ_MatTransMatMult(void*);

338: PETSC_INTERN PetscErrorCode MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ(Mat,Mat,PetscReal,Mat);
339: PETSC_INTERN PetscErrorCode MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ(Mat,Mat,Mat);
340: PETSC_INTERN PetscErrorCode MatTransposeColoringCreate_SeqAIJ(Mat,ISColoring,MatTransposeColoring);
341: PETSC_INTERN PetscErrorCode MatTransColoringApplySpToDen_SeqAIJ(MatTransposeColoring,Mat,Mat);
342: PETSC_INTERN PetscErrorCode MatTransColoringApplyDenToSp_SeqAIJ(MatTransposeColoring,Mat,Mat);

344: PETSC_INTERN PetscErrorCode MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,PetscReal,Mat);
345: PETSC_INTERN PetscErrorCode MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ(Mat,Mat,Mat,Mat);

347: PETSC_INTERN PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat,PetscInt,PetscInt,PetscRandom);
348: PETSC_INTERN PetscErrorCode MatSetValues_SeqAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode);
349: PETSC_INTERN PetscErrorCode MatGetRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
350: PETSC_INTERN PetscErrorCode MatRestoreRow_SeqAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
351: PETSC_INTERN PetscErrorCode MatScale_SeqAIJ(Mat,PetscScalar);
352: PETSC_INTERN PetscErrorCode MatDiagonalScale_SeqAIJ(Mat,Vec,Vec);
353: PETSC_INTERN PetscErrorCode MatDiagonalSet_SeqAIJ(Mat,Vec,InsertMode);
354: PETSC_INTERN PetscErrorCode MatAXPY_SeqAIJ(Mat,PetscScalar,Mat,MatStructure);
355: PETSC_INTERN PetscErrorCode MatGetRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
356: PETSC_INTERN PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
357: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
358: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscBool*);
359: PETSC_INTERN PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscInt *[],PetscBool*);
360: PETSC_INTERN PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat,PetscInt,PetscBool,PetscBool,PetscInt*,const PetscInt *[],const PetscInt *[],PetscInt *[],PetscBool*);
361: PETSC_INTERN PetscErrorCode MatDestroy_SeqAIJ(Mat);
362: PETSC_INTERN PetscErrorCode MatSetUp_SeqAIJ(Mat);
363: PETSC_INTERN PetscErrorCode MatView_SeqAIJ(Mat,PetscViewer);

365: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat);
366: PETSC_INTERN PetscErrorCode MatSeqAIJInvalidateDiagonal_Inode(Mat);
367: PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode(Mat);
368: PETSC_INTERN PetscErrorCode MatSeqAIJCheckInode_FactorLU(Mat);

370: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat,Mat,PetscInt*);

372: #if defined(PETSC_HAVE_MATLAB_ENGINE)
373: PETSC_EXTERN PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*);
374: PETSC_EXTERN PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*);
375: #endif
376: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqSBAIJ(Mat,MatType,MatReuse,Mat*);
377: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqBAIJ(Mat,MatType,MatReuse,Mat*);
378: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqDense(Mat,MatType,MatReuse,Mat*);
379: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*);
380: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat,MatType,MatReuse,Mat*);
381: #if defined(PETSC_HAVE_SCALAPACK)
382: PETSC_INTERN PetscErrorCode MatConvert_AIJ_ScaLAPACK(Mat,MatType,MatReuse,Mat*);
383: #endif
384: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*);
385: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJPERM(Mat,MatType,MatReuse,Mat*);
386: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJSELL(Mat,MatType,MatReuse,Mat*);
387: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJMKL(Mat,MatType,MatReuse,Mat*);
388: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJViennaCL(Mat,MatType,MatReuse,Mat*);
389: PETSC_INTERN PetscErrorCode MatReorderForNonzeroDiagonal_SeqAIJ(Mat,PetscReal,IS,IS);
390: PETSC_INTERN PetscErrorCode MatRARt_SeqAIJ_SeqAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
391: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat);
392: PETSC_INTERN PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat,MatAssemblyType);
393: PETSC_EXTERN PetscErrorCode MatZeroEntries_SeqAIJ(Mat);

395: PETSC_INTERN PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt,const PetscInt*,const PetscInt*,const PetscInt*,const PetscInt*,PetscInt*);
396: PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm,Mat,PetscInt,MatReuse,Mat*);
397: PETSC_INTERN PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIAIJ(MPI_Comm,Mat,PetscInt,MatReuse,Mat*);

399: PETSC_INTERN PetscErrorCode MatSetSeqMat_SeqAIJ(Mat,IS,IS,MatStructure,Mat);
400: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt*);
401: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat);
402: PETSC_INTERN PetscErrorCode MatDestroySubMatrix_Dummy(Mat);
403: PETSC_INTERN PetscErrorCode MatDestroySubMatrices_Dummy(PetscInt, Mat*[]);
404: PETSC_INTERN PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat,IS,IS,PetscInt,MatReuse,Mat*);

406: PETSC_INTERN PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat,ISLocalToGlobalMapping*);
407: PETSC_INTERN PetscErrorCode MatSetSeqAIJWithArrays_private(MPI_Comm,PetscInt,PetscInt,PetscInt[],PetscInt[],PetscScalar[],MatType,Mat);

409: /*
410:     PetscSparseDenseMinusDot - The inner kernel of triangular solves and Gauss-Siedel smoothing. \sum_i xv[i] * r[xi[i]] for CSR storage

412:   Input Parameters:
413: +  nnz - the number of entries
414: .  r - the array of vector values
415: .  xv - the matrix values for the row
416: -  xi - the column indices of the nonzeros in the row

418:   Output Parameter:
419: .  sum - negative the sum of results

421:   PETSc compile flags:
422: +   PETSC_KERNEL_USE_UNROLL_4
423: -   PETSC_KERNEL_USE_UNROLL_2

425:   Developer Notes:
426:     The macro changes sum but not other parameters

428: .seealso: PetscSparseDensePlusDot()

430: */
431: #if defined(PETSC_KERNEL_USE_UNROLL_4)
432: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
433:     if (nnz > 0) { \
434:       PetscInt nnz2=nnz,rem=nnz&0x3; \
435:       switch (rem) { \
436:       case 3: sum -= *xv++ *r[*xi++]; \
437:       case 2: sum -= *xv++ *r[*xi++]; \
438:       case 1: sum -= *xv++ *r[*xi++]; \
439:         nnz2      -= rem;} \
440:       while (nnz2 > 0) { \
441:         sum -=  xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + \
442:                 xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
443:         xv += 4; xi += 4; nnz2 -= 4; \
444:       } \
445:       xv -= nnz; xi -= nnz; \
446:     } \
447:   }

449: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
450: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
451:     PetscInt __i,__i1,__i2; \
452:     for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \
453:                                     sum -= (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \
454:     if (nnz & 0x1) sum -= xv[__i] * r[xi[__i]];}

456: #else
457: #define PetscSparseDenseMinusDot(sum,r,xv,xi,nnz) { \
458:     PetscInt __i; \
459:     for (__i=0; __i<nnz; __i++) sum -= xv[__i] * r[xi[__i]];}
460: #endif

462: /*
463:     PetscSparseDensePlusDot - The inner kernel of matrix-vector product \sum_i xv[i] * r[xi[i]] for CSR storage

465:   Input Parameters:
466: +  nnz - the number of entries
467: .  r - the array of vector values
468: .  xv - the matrix values for the row
469: -  xi - the column indices of the nonzeros in the row

471:   Output Parameter:
472: .  sum - the sum of results

474:   PETSc compile flags:
475: +   PETSC_KERNEL_USE_UNROLL_4
476: -   PETSC_KERNEL_USE_UNROLL_2

478:   Developer Notes:
479:     The macro changes sum but not other parameters

481: .seealso: PetscSparseDenseMinusDot()

483: */
484: #if defined(PETSC_KERNEL_USE_UNROLL_4)
485: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
486:     if (nnz > 0) { \
487:       PetscInt nnz2=nnz,rem=nnz&0x3; \
488:       switch (rem) { \
489:       case 3: sum += *xv++ *r[*xi++]; \
490:       case 2: sum += *xv++ *r[*xi++]; \
491:       case 1: sum += *xv++ *r[*xi++]; \
492:         nnz2      -= rem;} \
493:       while (nnz2 > 0) { \
494:         sum +=  xv[0] * r[xi[0]] + xv[1] * r[xi[1]] + \
495:                 xv[2] * r[xi[2]] + xv[3] * r[xi[3]]; \
496:         xv += 4; xi += 4; nnz2 -= 4; \
497:       } \
498:       xv -= nnz; xi -= nnz; \
499:     } \
500:   }

502: #elif defined(PETSC_KERNEL_USE_UNROLL_2)
503: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
504:     PetscInt __i,__i1,__i2; \
505:     for (__i=0; __i<nnz-1; __i+=2) {__i1 = xi[__i]; __i2=xi[__i+1]; \
506:                                     sum += (xv[__i]*r[__i1] + xv[__i+1]*r[__i2]);} \
507:     if (nnz & 0x1) sum += xv[__i] * r[xi[__i]];}

509: #elif defined(PETSC_USE_AVX512_KERNELS) && defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX512F__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES) && !defined(PETSC_SKIP_IMMINTRIN_H_CUDAWORKAROUND)
510: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) PetscSparseDensePlusDot_AVX512_Private(&(sum),(r),(xv),(xi),(nnz))

512: #else
513: #define PetscSparseDensePlusDot(sum,r,xv,xi,nnz) { \
514:     PetscInt __i; \
515:     for (__i=0; __i<nnz; __i++) sum += xv[__i] * r[xi[__i]];}
516: #endif

518: #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)
519:   #include <immintrin.h>
520:   #if !defined(_MM_SCALE_8)
521:   #define _MM_SCALE_8    8
522:   #endif

524: static inline void PetscSparseDensePlusDot_AVX512_Private(PetscScalar *sum,const PetscScalar *x,const MatScalar *aa,const PetscInt *aj,PetscInt n)
525: {
526:   __m512d  vec_x,vec_y,vec_vals;
527:   __m256i  vec_idx;
528:   __mmask8 mask;
529:   PetscInt j;

531:   vec_y = _mm512_setzero_pd();
532:   for (j=0; j<(n>>3); j++) {
533:     vec_idx  = _mm256_loadu_si256((__m256i const*)aj);
534:     vec_vals = _mm512_loadu_pd(aa);
535:     vec_x    = _mm512_i32gather_pd(vec_idx,x,_MM_SCALE_8);
536:     vec_y    = _mm512_fmadd_pd(vec_x,vec_vals,vec_y);
537:     aj += 8; aa += 8;
538:   }
539:   /* masked load does not work on KNL, it requires avx512vl */
540:   if ((n&0x07)>2) {
541:     mask     = (__mmask8)(0xff >> (8-(n&0x07)));
542:     vec_idx  = _mm256_loadu_si256((__m256i const*)aj);
543:     vec_vals = _mm512_loadu_pd(aa);
544:     vec_x    = _mm512_mask_i32gather_pd(vec_x,mask,vec_idx,x,_MM_SCALE_8);
545:     vec_y    = _mm512_mask3_fmadd_pd(vec_x,vec_vals,vec_y,mask);
546:   } else if ((n&0x07)==2) {
547:     *sum += aa[0]*x[aj[0]];
548:     *sum += aa[1]*x[aj[1]];
549:   } else if ((n&0x07)==1) {
550:     *sum += aa[0]*x[aj[0]];
551:   }
552:   if (n>2) *sum += _mm512_reduce_add_pd(vec_y);
553: /*
554:   for (j=0;j<(n&0x07);j++) *sum += aa[j]*x[aj[j]];
555: */
556: }
557: #endif

559: /*
560:     PetscSparseDenseMaxDot - The inner kernel of a modified matrix-vector product \max_i xv[i] * r[xi[i]] for CSR storage

562:   Input Parameters:
563: +  nnz - the number of entries
564: .  r - the array of vector values
565: .  xv - the matrix values for the row
566: -  xi - the column indices of the nonzeros in the row

568:   Output Parameter:
569: .  max - the max of results

571: .seealso: PetscSparseDensePlusDot(), PetscSparseDenseMinusDot()

573: */
574: #define PetscSparseDenseMaxDot(max,r,xv,xi,nnz) { \
575:     PetscInt __i; \
576:     for (__i=0; __i<nnz; __i++) max = PetscMax(PetscRealPart(max), PetscRealPart(xv[__i] * r[xi[__i]]));}

578: /*
579:  Add column indices into table for counting the max nonzeros of merged rows
580:  */
581: #define MatRowMergeMax_SeqAIJ(mat,nrows,ta) {       \
582:     PetscInt _j,_row,_nz,*_col;                     \
583:     if (mat) { \
584:       for (_row=0; _row<nrows; _row++) {   \
585:         _nz = mat->i[_row+1] - mat->i[_row];    \
586:         for (_j=0; _j<_nz; _j++) {               \
587:           _col = _j + mat->j + mat->i[_row];       \
588:           PetscTableAdd(ta,*_col+1,1,INSERT_VALUES);                    \
589:         }                                                               \
590:       }                                                                 \
591:     }    \
592: }

594: /*
595:  Add column indices into table for counting the nonzeros of merged rows
596:  */
597: #define MatMergeRows_SeqAIJ(mat,nrows,rows,ta) {    \
598:   PetscInt _j,_row,_nz,*_col,_i;                      \
599:     for (_i=0; _i<nrows; _i++) {\
600:       _row = rows[_i]; \
601:       _nz = mat->i[_row+1] - mat->i[_row]; \
602:       for (_j=0; _j<_nz; _j++) {                \
603:         _col = _j + mat->j + mat->i[_row];       \
604:         PetscTableAdd(ta,*_col+1,1,INSERT_VALUES); \
605:       }                                                                 \
606:     }                                                                   \
607: }

609: #endif