1 /* 2 Defines the basic matrix operations for the AIJ (compressed row) 3 matrix storage format. 4 */ 5 6 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 7 #include <petscblaslapack.h> 8 #include <petscbt.h> 9 #include <petsc/private/kernels/blocktranspose.h> 10 11 PetscErrorCode MatSeqAIJSetTypeFromOptions(Mat A) 12 { 13 PetscBool flg; 14 char type[256]; 15 16 PetscFunctionBegin; 17 PetscObjectOptionsBegin((PetscObject)A); 18 PetscCall(PetscOptionsFList("-mat_seqaij_type","Matrix SeqAIJ type","MatSeqAIJSetType",MatSeqAIJList,"seqaij",type,256,&flg)); 19 if (flg) PetscCall(MatSeqAIJSetType(A,type)); 20 PetscOptionsEnd(); 21 PetscFunctionReturn(0); 22 } 23 24 PetscErrorCode MatGetColumnReductions_SeqAIJ(Mat A,PetscInt type,PetscReal *reductions) 25 { 26 PetscInt i,m,n; 27 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 28 29 PetscFunctionBegin; 30 PetscCall(MatGetSize(A,&m,&n)); 31 PetscCall(PetscArrayzero(reductions,n)); 32 if (type == NORM_2) { 33 for (i=0; i<aij->i[m]; i++) { 34 reductions[aij->j[i]] += PetscAbsScalar(aij->a[i]*aij->a[i]); 35 } 36 } else if (type == NORM_1) { 37 for (i=0; i<aij->i[m]; i++) { 38 reductions[aij->j[i]] += PetscAbsScalar(aij->a[i]); 39 } 40 } else if (type == NORM_INFINITY) { 41 for (i=0; i<aij->i[m]; i++) { 42 reductions[aij->j[i]] = PetscMax(PetscAbsScalar(aij->a[i]),reductions[aij->j[i]]); 43 } 44 } else if (type == REDUCTION_SUM_REALPART || type == REDUCTION_MEAN_REALPART) { 45 for (i=0; i<aij->i[m]; i++) { 46 reductions[aij->j[i]] += PetscRealPart(aij->a[i]); 47 } 48 } else if (type == REDUCTION_SUM_IMAGINARYPART || type == REDUCTION_MEAN_IMAGINARYPART) { 49 for (i=0; i<aij->i[m]; i++) { 50 reductions[aij->j[i]] += PetscImaginaryPart(aij->a[i]); 51 } 52 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown reduction type"); 53 54 if (type == NORM_2) { 55 for (i=0; i<n; i++) reductions[i] = PetscSqrtReal(reductions[i]); 56 } else if (type == REDUCTION_MEAN_REALPART || type == REDUCTION_MEAN_IMAGINARYPART) { 57 for (i=0; i<n; i++) reductions[i] /= m; 58 } 59 PetscFunctionReturn(0); 60 } 61 62 PetscErrorCode MatFindOffBlockDiagonalEntries_SeqAIJ(Mat A,IS *is) 63 { 64 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 65 PetscInt i,m=A->rmap->n,cnt = 0, bs = A->rmap->bs; 66 const PetscInt *jj = a->j,*ii = a->i; 67 PetscInt *rows; 68 69 PetscFunctionBegin; 70 for (i=0; i<m; i++) { 71 if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) { 72 cnt++; 73 } 74 } 75 PetscCall(PetscMalloc1(cnt,&rows)); 76 cnt = 0; 77 for (i=0; i<m; i++) { 78 if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) { 79 rows[cnt] = i; 80 cnt++; 81 } 82 } 83 PetscCall(ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,is)); 84 PetscFunctionReturn(0); 85 } 86 87 PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat A,PetscInt *nrows,PetscInt **zrows) 88 { 89 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 90 const MatScalar *aa; 91 PetscInt i,m=A->rmap->n,cnt = 0; 92 const PetscInt *ii = a->i,*jj = a->j,*diag; 93 PetscInt *rows; 94 95 PetscFunctionBegin; 96 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 97 PetscCall(MatMarkDiagonal_SeqAIJ(A)); 98 diag = a->diag; 99 for (i=0; i<m; i++) { 100 if ((diag[i] >= ii[i+1]) || (jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) { 101 cnt++; 102 } 103 } 104 PetscCall(PetscMalloc1(cnt,&rows)); 105 cnt = 0; 106 for (i=0; i<m; i++) { 107 if ((diag[i] >= ii[i+1]) || (jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) { 108 rows[cnt++] = i; 109 } 110 } 111 *nrows = cnt; 112 *zrows = rows; 113 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 114 PetscFunctionReturn(0); 115 } 116 117 PetscErrorCode MatFindZeroDiagonals_SeqAIJ(Mat A,IS *zrows) 118 { 119 PetscInt nrows,*rows; 120 121 PetscFunctionBegin; 122 *zrows = NULL; 123 PetscCall(MatFindZeroDiagonals_SeqAIJ_Private(A,&nrows,&rows)); 124 PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)A),nrows,rows,PETSC_OWN_POINTER,zrows)); 125 PetscFunctionReturn(0); 126 } 127 128 PetscErrorCode MatFindNonzeroRows_SeqAIJ(Mat A,IS *keptrows) 129 { 130 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 131 const MatScalar *aa; 132 PetscInt m=A->rmap->n,cnt = 0; 133 const PetscInt *ii; 134 PetscInt n,i,j,*rows; 135 136 PetscFunctionBegin; 137 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 138 *keptrows = NULL; 139 ii = a->i; 140 for (i=0; i<m; i++) { 141 n = ii[i+1] - ii[i]; 142 if (!n) { 143 cnt++; 144 goto ok1; 145 } 146 for (j=ii[i]; j<ii[i+1]; j++) { 147 if (aa[j] != 0.0) goto ok1; 148 } 149 cnt++; 150 ok1:; 151 } 152 if (!cnt) { 153 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 154 PetscFunctionReturn(0); 155 } 156 PetscCall(PetscMalloc1(A->rmap->n-cnt,&rows)); 157 cnt = 0; 158 for (i=0; i<m; i++) { 159 n = ii[i+1] - ii[i]; 160 if (!n) continue; 161 for (j=ii[i]; j<ii[i+1]; j++) { 162 if (aa[j] != 0.0) { 163 rows[cnt++] = i; 164 break; 165 } 166 } 167 } 168 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 169 PetscCall(ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,keptrows)); 170 PetscFunctionReturn(0); 171 } 172 173 PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is) 174 { 175 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data; 176 PetscInt i,m = Y->rmap->n; 177 const PetscInt *diag; 178 MatScalar *aa; 179 const PetscScalar *v; 180 PetscBool missing; 181 182 PetscFunctionBegin; 183 if (Y->assembled) { 184 PetscCall(MatMissingDiagonal_SeqAIJ(Y,&missing,NULL)); 185 if (!missing) { 186 diag = aij->diag; 187 PetscCall(VecGetArrayRead(D,&v)); 188 PetscCall(MatSeqAIJGetArray(Y,&aa)); 189 if (is == INSERT_VALUES) { 190 for (i=0; i<m; i++) { 191 aa[diag[i]] = v[i]; 192 } 193 } else { 194 for (i=0; i<m; i++) { 195 aa[diag[i]] += v[i]; 196 } 197 } 198 PetscCall(MatSeqAIJRestoreArray(Y,&aa)); 199 PetscCall(VecRestoreArrayRead(D,&v)); 200 PetscFunctionReturn(0); 201 } 202 PetscCall(MatSeqAIJInvalidateDiagonal(Y)); 203 } 204 PetscCall(MatDiagonalSet_Default(Y,D,is)); 205 PetscFunctionReturn(0); 206 } 207 208 PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *m,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 209 { 210 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 211 PetscInt i,ishift; 212 213 PetscFunctionBegin; 214 *m = A->rmap->n; 215 if (!ia) PetscFunctionReturn(0); 216 ishift = 0; 217 if (symmetric && !A->structurally_symmetric) { 218 PetscCall(MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,PETSC_TRUE,ishift,oshift,(PetscInt**)ia,(PetscInt**)ja)); 219 } else if (oshift == 1) { 220 PetscInt *tia; 221 PetscInt nz = a->i[A->rmap->n]; 222 /* malloc space and add 1 to i and j indices */ 223 PetscCall(PetscMalloc1(A->rmap->n+1,&tia)); 224 for (i=0; i<A->rmap->n+1; i++) tia[i] = a->i[i] + 1; 225 *ia = tia; 226 if (ja) { 227 PetscInt *tja; 228 PetscCall(PetscMalloc1(nz+1,&tja)); 229 for (i=0; i<nz; i++) tja[i] = a->j[i] + 1; 230 *ja = tja; 231 } 232 } else { 233 *ia = a->i; 234 if (ja) *ja = a->j; 235 } 236 PetscFunctionReturn(0); 237 } 238 239 PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 240 { 241 PetscFunctionBegin; 242 if (!ia) PetscFunctionReturn(0); 243 if ((symmetric && !A->structurally_symmetric) || oshift == 1) { 244 PetscCall(PetscFree(*ia)); 245 if (ja) PetscCall(PetscFree(*ja)); 246 } 247 PetscFunctionReturn(0); 248 } 249 250 PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 251 { 252 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 253 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 254 PetscInt nz = a->i[m],row,*jj,mr,col; 255 256 PetscFunctionBegin; 257 *nn = n; 258 if (!ia) PetscFunctionReturn(0); 259 if (symmetric) { 260 PetscCall(MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,PETSC_TRUE,0,oshift,(PetscInt**)ia,(PetscInt**)ja)); 261 } else { 262 PetscCall(PetscCalloc1(n,&collengths)); 263 PetscCall(PetscMalloc1(n+1,&cia)); 264 PetscCall(PetscMalloc1(nz,&cja)); 265 jj = a->j; 266 for (i=0; i<nz; i++) { 267 collengths[jj[i]]++; 268 } 269 cia[0] = oshift; 270 for (i=0; i<n; i++) { 271 cia[i+1] = cia[i] + collengths[i]; 272 } 273 PetscCall(PetscArrayzero(collengths,n)); 274 jj = a->j; 275 for (row=0; row<m; row++) { 276 mr = a->i[row+1] - a->i[row]; 277 for (i=0; i<mr; i++) { 278 col = *jj++; 279 280 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 281 } 282 } 283 PetscCall(PetscFree(collengths)); 284 *ia = cia; *ja = cja; 285 } 286 PetscFunctionReturn(0); 287 } 288 289 PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 290 { 291 PetscFunctionBegin; 292 if (!ia) PetscFunctionReturn(0); 293 294 PetscCall(PetscFree(*ia)); 295 PetscCall(PetscFree(*ja)); 296 PetscFunctionReturn(0); 297 } 298 299 /* 300 MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from 301 MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output 302 spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ() and MatFDColoringCreate_SeqXAIJ() 303 */ 304 PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 305 { 306 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 307 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 308 PetscInt nz = a->i[m],row,mr,col,tmp; 309 PetscInt *cspidx; 310 const PetscInt *jj; 311 312 PetscFunctionBegin; 313 *nn = n; 314 if (!ia) PetscFunctionReturn(0); 315 316 PetscCall(PetscCalloc1(n,&collengths)); 317 PetscCall(PetscMalloc1(n+1,&cia)); 318 PetscCall(PetscMalloc1(nz,&cja)); 319 PetscCall(PetscMalloc1(nz,&cspidx)); 320 jj = a->j; 321 for (i=0; i<nz; i++) { 322 collengths[jj[i]]++; 323 } 324 cia[0] = oshift; 325 for (i=0; i<n; i++) { 326 cia[i+1] = cia[i] + collengths[i]; 327 } 328 PetscCall(PetscArrayzero(collengths,n)); 329 jj = a->j; 330 for (row=0; row<m; row++) { 331 mr = a->i[row+1] - a->i[row]; 332 for (i=0; i<mr; i++) { 333 col = *jj++; 334 tmp = cia[col] + collengths[col]++ - oshift; 335 cspidx[tmp] = a->i[row] + i; /* index of a->j */ 336 cja[tmp] = row + oshift; 337 } 338 } 339 PetscCall(PetscFree(collengths)); 340 *ia = cia; 341 *ja = cja; 342 *spidx = cspidx; 343 PetscFunctionReturn(0); 344 } 345 346 PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 347 { 348 PetscFunctionBegin; 349 PetscCall(MatRestoreColumnIJ_SeqAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done)); 350 PetscCall(PetscFree(*spidx)); 351 PetscFunctionReturn(0); 352 } 353 354 PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[]) 355 { 356 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 357 PetscInt *ai = a->i; 358 PetscScalar *aa; 359 360 PetscFunctionBegin; 361 PetscCall(MatSeqAIJGetArray(A,&aa)); 362 PetscCall(PetscArraycpy(aa+ai[row],v,ai[row+1]-ai[row])); 363 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 364 PetscFunctionReturn(0); 365 } 366 367 /* 368 MatSeqAIJSetValuesLocalFast - An optimized version of MatSetValuesLocal() for SeqAIJ matrices with several assumptions 369 370 - a single row of values is set with each call 371 - no row or column indices are negative or (in error) larger than the number of rows or columns 372 - the values are always added to the matrix, not set 373 - no new locations are introduced in the nonzero structure of the matrix 374 375 This does NOT assume the global column indices are sorted 376 377 */ 378 379 #include <petsc/private/isimpl.h> 380 PetscErrorCode MatSeqAIJSetValuesLocalFast(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 381 { 382 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 383 PetscInt low,high,t,row,nrow,i,col,l; 384 const PetscInt *rp,*ai = a->i,*ailen = a->ilen,*aj = a->j; 385 PetscInt lastcol = -1; 386 MatScalar *ap,value,*aa; 387 const PetscInt *ridx = A->rmap->mapping->indices,*cidx = A->cmap->mapping->indices; 388 389 PetscFunctionBegin; 390 PetscCall(MatSeqAIJGetArray(A,&aa)); 391 row = ridx[im[0]]; 392 rp = aj + ai[row]; 393 ap = aa + ai[row]; 394 nrow = ailen[row]; 395 low = 0; 396 high = nrow; 397 for (l=0; l<n; l++) { /* loop over added columns */ 398 col = cidx[in[l]]; 399 value = v[l]; 400 401 if (col <= lastcol) low = 0; 402 else high = nrow; 403 lastcol = col; 404 while (high-low > 5) { 405 t = (low+high)/2; 406 if (rp[t] > col) high = t; 407 else low = t; 408 } 409 for (i=low; i<high; i++) { 410 if (rp[i] == col) { 411 ap[i] += value; 412 low = i + 1; 413 break; 414 } 415 } 416 } 417 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 418 return 0; 419 } 420 421 PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 422 { 423 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 424 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 425 PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen; 426 PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1; 427 MatScalar *ap=NULL,value=0.0,*aa; 428 PetscBool ignorezeroentries = a->ignorezeroentries; 429 PetscBool roworiented = a->roworiented; 430 431 PetscFunctionBegin; 432 PetscCall(MatSeqAIJGetArray(A,&aa)); 433 for (k=0; k<m; k++) { /* loop over added rows */ 434 row = im[k]; 435 if (row < 0) continue; 436 PetscCheck(row < A->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT,row,A->rmap->n-1); 437 rp = aj + ai[row]; 438 if (!A->structure_only) ap = aa + ai[row]; 439 rmax = imax[row]; nrow = ailen[row]; 440 low = 0; 441 high = nrow; 442 for (l=0; l<n; l++) { /* loop over added columns */ 443 if (in[l] < 0) continue; 444 PetscCheck(in[l] < A->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT,in[l],A->cmap->n-1); 445 col = in[l]; 446 if (v && !A->structure_only) value = roworiented ? v[l + k*n] : v[k + l*m]; 447 if (!A->structure_only && value == 0.0 && ignorezeroentries && is == ADD_VALUES && row != col) continue; 448 449 if (col <= lastcol) low = 0; 450 else high = nrow; 451 lastcol = col; 452 while (high-low > 5) { 453 t = (low+high)/2; 454 if (rp[t] > col) high = t; 455 else low = t; 456 } 457 for (i=low; i<high; i++) { 458 if (rp[i] > col) break; 459 if (rp[i] == col) { 460 if (!A->structure_only) { 461 if (is == ADD_VALUES) { 462 ap[i] += value; 463 (void)PetscLogFlops(1.0); 464 } 465 else ap[i] = value; 466 } 467 low = i + 1; 468 goto noinsert; 469 } 470 } 471 if (value == 0.0 && ignorezeroentries && row != col) goto noinsert; 472 if (nonew == 1) goto noinsert; 473 PetscCheck(nonew != -1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%" PetscInt_FMT ",%" PetscInt_FMT ") in the matrix",row,col); 474 if (A->structure_only) { 475 MatSeqXAIJReallocateAIJ_structure_only(A,A->rmap->n,1,nrow,row,col,rmax,ai,aj,rp,imax,nonew,MatScalar); 476 } else { 477 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 478 } 479 N = nrow++ - 1; a->nz++; high++; 480 /* shift up all the later entries in this row */ 481 PetscCall(PetscArraymove(rp+i+1,rp+i,N-i+1)); 482 rp[i] = col; 483 if (!A->structure_only) { 484 PetscCall(PetscArraymove(ap+i+1,ap+i,N-i+1)); 485 ap[i] = value; 486 } 487 low = i + 1; 488 A->nonzerostate++; 489 noinsert:; 490 } 491 ailen[row] = nrow; 492 } 493 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 494 PetscFunctionReturn(0); 495 } 496 497 PetscErrorCode MatSetValues_SeqAIJ_SortedFullNoPreallocation(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 498 { 499 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 500 PetscInt *rp,k,row; 501 PetscInt *ai = a->i; 502 PetscInt *aj = a->j; 503 MatScalar *aa,*ap; 504 505 PetscFunctionBegin; 506 PetscCheck(!A->was_assembled,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot call on assembled matrix."); 507 PetscCheck(m*n+a->nz <= a->maxnz,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Number of entries in matrix will be larger than maximum nonzeros allocated for %" PetscInt_FMT " in MatSeqAIJSetTotalPreallocation()",a->maxnz); 508 509 PetscCall(MatSeqAIJGetArray(A,&aa)); 510 for (k=0; k<m; k++) { /* loop over added rows */ 511 row = im[k]; 512 rp = aj + ai[row]; 513 ap = aa + ai[row]; 514 515 PetscCall(PetscMemcpy(rp,in,n*sizeof(PetscInt))); 516 if (!A->structure_only) { 517 if (v) { 518 PetscCall(PetscMemcpy(ap,v,n*sizeof(PetscScalar))); 519 v += n; 520 } else { 521 PetscCall(PetscMemzero(ap,n*sizeof(PetscScalar))); 522 } 523 } 524 a->ilen[row] = n; 525 a->imax[row] = n; 526 a->i[row+1] = a->i[row]+n; 527 a->nz += n; 528 } 529 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 530 PetscFunctionReturn(0); 531 } 532 533 /*@ 534 MatSeqAIJSetTotalPreallocation - Sets an upper bound on the total number of expected nonzeros in the matrix. 535 536 Input Parameters: 537 + A - the SeqAIJ matrix 538 - nztotal - bound on the number of nonzeros 539 540 Level: advanced 541 542 Notes: 543 This can be called if you will be provided the matrix row by row (from row zero) with sorted column indices for each row. 544 Simply call MatSetValues() after this call to provide the matrix entries in the usual manner. This matrix may be used 545 as always with multiple matrix assemblies. 546 547 .seealso: `MatSetOption()`, `MAT_SORTED_FULL`, `MatSetValues()`, `MatSeqAIJSetPreallocation()` 548 @*/ 549 550 PetscErrorCode MatSeqAIJSetTotalPreallocation(Mat A,PetscInt nztotal) 551 { 552 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 553 554 PetscFunctionBegin; 555 PetscCall(PetscLayoutSetUp(A->rmap)); 556 PetscCall(PetscLayoutSetUp(A->cmap)); 557 a->maxnz = nztotal; 558 if (!a->imax) { 559 PetscCall(PetscMalloc1(A->rmap->n,&a->imax)); 560 PetscCall(PetscLogObjectMemory((PetscObject)A,A->rmap->n*sizeof(PetscInt))); 561 } 562 if (!a->ilen) { 563 PetscCall(PetscMalloc1(A->rmap->n,&a->ilen)); 564 PetscCall(PetscLogObjectMemory((PetscObject)A,A->rmap->n*sizeof(PetscInt))); 565 } else { 566 PetscCall(PetscMemzero(a->ilen,A->rmap->n*sizeof(PetscInt))); 567 } 568 569 /* allocate the matrix space */ 570 if (A->structure_only) { 571 PetscCall(PetscMalloc1(nztotal,&a->j)); 572 PetscCall(PetscMalloc1(A->rmap->n+1,&a->i)); 573 PetscCall(PetscLogObjectMemory((PetscObject)A,(A->rmap->n+1)*sizeof(PetscInt)+nztotal*sizeof(PetscInt))); 574 } else { 575 PetscCall(PetscMalloc3(nztotal,&a->a,nztotal,&a->j,A->rmap->n+1,&a->i)); 576 PetscCall(PetscLogObjectMemory((PetscObject)A,(A->rmap->n+1)*sizeof(PetscInt)+nztotal*(sizeof(PetscScalar)+sizeof(PetscInt)))); 577 } 578 a->i[0] = 0; 579 if (A->structure_only) { 580 a->singlemalloc = PETSC_FALSE; 581 a->free_a = PETSC_FALSE; 582 } else { 583 a->singlemalloc = PETSC_TRUE; 584 a->free_a = PETSC_TRUE; 585 } 586 a->free_ij = PETSC_TRUE; 587 A->ops->setvalues = MatSetValues_SeqAIJ_SortedFullNoPreallocation; 588 A->preallocated = PETSC_TRUE; 589 PetscFunctionReturn(0); 590 } 591 592 PetscErrorCode MatSetValues_SeqAIJ_SortedFull(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 593 { 594 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 595 PetscInt *rp,k,row; 596 PetscInt *ai = a->i,*ailen = a->ilen; 597 PetscInt *aj = a->j; 598 MatScalar *aa,*ap; 599 600 PetscFunctionBegin; 601 PetscCall(MatSeqAIJGetArray(A,&aa)); 602 for (k=0; k<m; k++) { /* loop over added rows */ 603 row = im[k]; 604 PetscCheck(n <= a->imax[row],PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Preallocation for row %" PetscInt_FMT " does not match number of columns provided",n); 605 rp = aj + ai[row]; 606 ap = aa + ai[row]; 607 if (!A->was_assembled) { 608 PetscCall(PetscMemcpy(rp,in,n*sizeof(PetscInt))); 609 } 610 if (!A->structure_only) { 611 if (v) { 612 PetscCall(PetscMemcpy(ap,v,n*sizeof(PetscScalar))); 613 v += n; 614 } else { 615 PetscCall(PetscMemzero(ap,n*sizeof(PetscScalar))); 616 } 617 } 618 ailen[row] = n; 619 a->nz += n; 620 } 621 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 622 PetscFunctionReturn(0); 623 } 624 625 PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[]) 626 { 627 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 628 PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j; 629 PetscInt *ai = a->i,*ailen = a->ilen; 630 MatScalar *ap,*aa; 631 632 PetscFunctionBegin; 633 PetscCall(MatSeqAIJGetArray(A,&aa)); 634 for (k=0; k<m; k++) { /* loop over rows */ 635 row = im[k]; 636 if (row < 0) {v += n; continue;} /* negative row */ 637 PetscCheck(row < A->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT,row,A->rmap->n-1); 638 rp = aj + ai[row]; ap = aa + ai[row]; 639 nrow = ailen[row]; 640 for (l=0; l<n; l++) { /* loop over columns */ 641 if (in[l] < 0) {v++; continue;} /* negative column */ 642 PetscCheck(in[l] < A->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT,in[l],A->cmap->n-1); 643 col = in[l]; 644 high = nrow; low = 0; /* assume unsorted */ 645 while (high-low > 5) { 646 t = (low+high)/2; 647 if (rp[t] > col) high = t; 648 else low = t; 649 } 650 for (i=low; i<high; i++) { 651 if (rp[i] > col) break; 652 if (rp[i] == col) { 653 *v++ = ap[i]; 654 goto finished; 655 } 656 } 657 *v++ = 0.0; 658 finished:; 659 } 660 } 661 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 662 PetscFunctionReturn(0); 663 } 664 665 PetscErrorCode MatView_SeqAIJ_Binary(Mat mat,PetscViewer viewer) 666 { 667 Mat_SeqAIJ *A = (Mat_SeqAIJ*)mat->data; 668 const PetscScalar *av; 669 PetscInt header[4],M,N,m,nz,i; 670 PetscInt *rowlens; 671 672 PetscFunctionBegin; 673 PetscCall(PetscViewerSetUp(viewer)); 674 675 M = mat->rmap->N; 676 N = mat->cmap->N; 677 m = mat->rmap->n; 678 nz = A->nz; 679 680 /* write matrix header */ 681 header[0] = MAT_FILE_CLASSID; 682 header[1] = M; header[2] = N; header[3] = nz; 683 PetscCall(PetscViewerBinaryWrite(viewer,header,4,PETSC_INT)); 684 685 /* fill in and store row lengths */ 686 PetscCall(PetscMalloc1(m,&rowlens)); 687 for (i=0; i<m; i++) rowlens[i] = A->i[i+1] - A->i[i]; 688 PetscCall(PetscViewerBinaryWrite(viewer,rowlens,m,PETSC_INT)); 689 PetscCall(PetscFree(rowlens)); 690 /* store column indices */ 691 PetscCall(PetscViewerBinaryWrite(viewer,A->j,nz,PETSC_INT)); 692 /* store nonzero values */ 693 PetscCall(MatSeqAIJGetArrayRead(mat,&av)); 694 PetscCall(PetscViewerBinaryWrite(viewer,av,nz,PETSC_SCALAR)); 695 PetscCall(MatSeqAIJRestoreArrayRead(mat,&av)); 696 697 /* write block size option to the viewer's .info file */ 698 PetscCall(MatView_Binary_BlockSizes(mat,viewer)); 699 PetscFunctionReturn(0); 700 } 701 702 static PetscErrorCode MatView_SeqAIJ_ASCII_structonly(Mat A,PetscViewer viewer) 703 { 704 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 705 PetscInt i,k,m=A->rmap->N; 706 707 PetscFunctionBegin; 708 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 709 for (i=0; i<m; i++) { 710 PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT ":",i)); 711 for (k=a->i[i]; k<a->i[i+1]; k++) { 712 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ") ",a->j[k])); 713 } 714 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 715 } 716 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 717 PetscFunctionReturn(0); 718 } 719 720 extern PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer); 721 722 PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer) 723 { 724 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 725 const PetscScalar *av; 726 PetscInt i,j,m = A->rmap->n; 727 const char *name; 728 PetscViewerFormat format; 729 730 PetscFunctionBegin; 731 if (A->structure_only) { 732 PetscCall(MatView_SeqAIJ_ASCII_structonly(A,viewer)); 733 PetscFunctionReturn(0); 734 } 735 736 PetscCall(PetscViewerGetFormat(viewer,&format)); 737 if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) PetscFunctionReturn(0); 738 739 /* trigger copy to CPU if needed */ 740 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 741 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 742 if (format == PETSC_VIEWER_ASCII_MATLAB) { 743 PetscInt nofinalvalue = 0; 744 if (m && ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap->n-1))) { 745 /* Need a dummy value to ensure the dimension of the matrix. */ 746 nofinalvalue = 1; 747 } 748 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 749 PetscCall(PetscViewerASCIIPrintf(viewer,"%% Size = %" PetscInt_FMT " %" PetscInt_FMT " \n",m,A->cmap->n)); 750 PetscCall(PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %" PetscInt_FMT " \n",a->nz)); 751 #if defined(PETSC_USE_COMPLEX) 752 PetscCall(PetscViewerASCIIPrintf(viewer,"zzz = zeros(%" PetscInt_FMT ",4);\n",a->nz+nofinalvalue)); 753 #else 754 PetscCall(PetscViewerASCIIPrintf(viewer,"zzz = zeros(%" PetscInt_FMT ",3);\n",a->nz+nofinalvalue)); 755 #endif 756 PetscCall(PetscViewerASCIIPrintf(viewer,"zzz = [\n")); 757 758 for (i=0; i<m; i++) { 759 for (j=a->i[i]; j<a->i[i+1]; j++) { 760 #if defined(PETSC_USE_COMPLEX) 761 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %18.16e %18.16e\n",i+1,a->j[j]+1,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 762 #else 763 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %18.16e\n",i+1,a->j[j]+1,(double)a->a[j])); 764 #endif 765 } 766 } 767 if (nofinalvalue) { 768 #if defined(PETSC_USE_COMPLEX) 769 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %18.16e %18.16e\n",m,A->cmap->n,0.,0.)); 770 #else 771 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %18.16e\n",m,A->cmap->n,0.0)); 772 #endif 773 } 774 PetscCall(PetscObjectGetName((PetscObject)A,&name)); 775 PetscCall(PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name)); 776 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 777 } else if (format == PETSC_VIEWER_ASCII_COMMON) { 778 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 779 for (i=0; i<m; i++) { 780 PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT ":",i)); 781 for (j=a->i[i]; j<a->i[i+1]; j++) { 782 #if defined(PETSC_USE_COMPLEX) 783 if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) { 784 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 785 } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) { 786 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]))); 787 } else if (PetscRealPart(a->a[j]) != 0.0) { 788 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)PetscRealPart(a->a[j]))); 789 } 790 #else 791 if (a->a[j] != 0.0) PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)a->a[j])); 792 #endif 793 } 794 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 795 } 796 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 797 } else if (format == PETSC_VIEWER_ASCII_SYMMODU) { 798 PetscInt nzd=0,fshift=1,*sptr; 799 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 800 PetscCall(PetscMalloc1(m+1,&sptr)); 801 for (i=0; i<m; i++) { 802 sptr[i] = nzd+1; 803 for (j=a->i[i]; j<a->i[i+1]; j++) { 804 if (a->j[j] >= i) { 805 #if defined(PETSC_USE_COMPLEX) 806 if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++; 807 #else 808 if (a->a[j] != 0.0) nzd++; 809 #endif 810 } 811 } 812 } 813 sptr[m] = nzd+1; 814 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " %" PetscInt_FMT "\n\n",m,nzd)); 815 for (i=0; i<m+1; i+=6) { 816 if (i+4<m) { 817 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT "\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4],sptr[i+5])); 818 } else if (i+3<m) { 819 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT "\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4])); 820 } else if (i+2<m) { 821 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT "\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3])); 822 } else if (i+1<m) { 823 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT "\n",sptr[i],sptr[i+1],sptr[i+2])); 824 } else if (i<m) { 825 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " %" PetscInt_FMT "\n",sptr[i],sptr[i+1])); 826 } else { 827 PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT "\n",sptr[i])); 828 } 829 } 830 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 831 PetscCall(PetscFree(sptr)); 832 for (i=0; i<m; i++) { 833 for (j=a->i[i]; j<a->i[i+1]; j++) { 834 if (a->j[j] >= i) PetscCall(PetscViewerASCIIPrintf(viewer," %" PetscInt_FMT " ",a->j[j]+fshift)); 835 } 836 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 837 } 838 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 839 for (i=0; i<m; i++) { 840 for (j=a->i[i]; j<a->i[i+1]; j++) { 841 if (a->j[j] >= i) { 842 #if defined(PETSC_USE_COMPLEX) 843 if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) { 844 PetscCall(PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 845 } 846 #else 847 if (a->a[j] != 0.0) PetscCall(PetscViewerASCIIPrintf(viewer," %18.16e ",(double)a->a[j])); 848 #endif 849 } 850 } 851 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 852 } 853 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 854 } else if (format == PETSC_VIEWER_ASCII_DENSE) { 855 PetscInt cnt = 0,jcnt; 856 PetscScalar value; 857 #if defined(PETSC_USE_COMPLEX) 858 PetscBool realonly = PETSC_TRUE; 859 860 for (i=0; i<a->i[m]; i++) { 861 if (PetscImaginaryPart(a->a[i]) != 0.0) { 862 realonly = PETSC_FALSE; 863 break; 864 } 865 } 866 #endif 867 868 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 869 for (i=0; i<m; i++) { 870 jcnt = 0; 871 for (j=0; j<A->cmap->n; j++) { 872 if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) { 873 value = a->a[cnt++]; 874 jcnt++; 875 } else { 876 value = 0.0; 877 } 878 #if defined(PETSC_USE_COMPLEX) 879 if (realonly) { 880 PetscCall(PetscViewerASCIIPrintf(viewer," %7.5e ",(double)PetscRealPart(value))); 881 } else { 882 PetscCall(PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",(double)PetscRealPart(value),(double)PetscImaginaryPart(value))); 883 } 884 #else 885 PetscCall(PetscViewerASCIIPrintf(viewer," %7.5e ",(double)value)); 886 #endif 887 } 888 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 889 } 890 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 891 } else if (format == PETSC_VIEWER_ASCII_MATRIXMARKET) { 892 PetscInt fshift=1; 893 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 894 #if defined(PETSC_USE_COMPLEX) 895 PetscCall(PetscViewerASCIIPrintf(viewer,"%%%%MatrixMarket matrix coordinate complex general\n")); 896 #else 897 PetscCall(PetscViewerASCIIPrintf(viewer,"%%%%MatrixMarket matrix coordinate real general\n")); 898 #endif 899 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT "\n", m, A->cmap->n, a->nz)); 900 for (i=0; i<m; i++) { 901 for (j=a->i[i]; j<a->i[i+1]; j++) { 902 #if defined(PETSC_USE_COMPLEX) 903 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %g %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 904 #else 905 PetscCall(PetscViewerASCIIPrintf(viewer,"%" PetscInt_FMT " %" PetscInt_FMT " %g\n", i+fshift, a->j[j]+fshift, (double)a->a[j])); 906 #endif 907 } 908 } 909 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 910 } else { 911 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE)); 912 if (A->factortype) { 913 for (i=0; i<m; i++) { 914 PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT ":",i)); 915 /* L part */ 916 for (j=a->i[i]; j<a->i[i+1]; j++) { 917 #if defined(PETSC_USE_COMPLEX) 918 if (PetscImaginaryPart(a->a[j]) > 0.0) { 919 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 920 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 921 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])))); 922 } else { 923 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)PetscRealPart(a->a[j]))); 924 } 925 #else 926 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)a->a[j])); 927 #endif 928 } 929 /* diagonal */ 930 j = a->diag[i]; 931 #if defined(PETSC_USE_COMPLEX) 932 if (PetscImaginaryPart(a->a[j]) > 0.0) { 933 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)PetscImaginaryPart(1.0/a->a[j]))); 934 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 935 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)(-PetscImaginaryPart(1.0/a->a[j])))); 936 } else { 937 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)PetscRealPart(1.0/a->a[j]))); 938 } 939 #else 940 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)(1.0/a->a[j]))); 941 #endif 942 943 /* U part */ 944 for (j=a->diag[i+1]+1; j<a->diag[i]; j++) { 945 #if defined(PETSC_USE_COMPLEX) 946 if (PetscImaginaryPart(a->a[j]) > 0.0) { 947 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 948 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 949 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])))); 950 } else { 951 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)PetscRealPart(a->a[j]))); 952 } 953 #else 954 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)a->a[j])); 955 #endif 956 } 957 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 958 } 959 } else { 960 for (i=0; i<m; i++) { 961 PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT ":",i)); 962 for (j=a->i[i]; j<a->i[i+1]; j++) { 963 #if defined(PETSC_USE_COMPLEX) 964 if (PetscImaginaryPart(a->a[j]) > 0.0) { 965 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]))); 966 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 967 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]))); 968 } else { 969 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)PetscRealPart(a->a[j]))); 970 } 971 #else 972 PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",a->j[j],(double)a->a[j])); 973 #endif 974 } 975 PetscCall(PetscViewerASCIIPrintf(viewer,"\n")); 976 } 977 } 978 PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE)); 979 } 980 PetscCall(PetscViewerFlush(viewer)); 981 PetscFunctionReturn(0); 982 } 983 984 #include <petscdraw.h> 985 PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa) 986 { 987 Mat A = (Mat) Aa; 988 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 989 PetscInt i,j,m = A->rmap->n; 990 int color; 991 PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r; 992 PetscViewer viewer; 993 PetscViewerFormat format; 994 const PetscScalar *aa; 995 996 PetscFunctionBegin; 997 PetscCall(PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer)); 998 PetscCall(PetscViewerGetFormat(viewer,&format)); 999 PetscCall(PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr)); 1000 1001 /* loop over matrix elements drawing boxes */ 1002 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 1003 if (format != PETSC_VIEWER_DRAW_CONTOUR) { 1004 PetscDrawCollectiveBegin(draw); 1005 /* Blue for negative, Cyan for zero and Red for positive */ 1006 color = PETSC_DRAW_BLUE; 1007 for (i=0; i<m; i++) { 1008 y_l = m - i - 1.0; y_r = y_l + 1.0; 1009 for (j=a->i[i]; j<a->i[i+1]; j++) { 1010 x_l = a->j[j]; x_r = x_l + 1.0; 1011 if (PetscRealPart(aa[j]) >= 0.) continue; 1012 PetscCall(PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color)); 1013 } 1014 } 1015 color = PETSC_DRAW_CYAN; 1016 for (i=0; i<m; i++) { 1017 y_l = m - i - 1.0; y_r = y_l + 1.0; 1018 for (j=a->i[i]; j<a->i[i+1]; j++) { 1019 x_l = a->j[j]; x_r = x_l + 1.0; 1020 if (aa[j] != 0.) continue; 1021 PetscCall(PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color)); 1022 } 1023 } 1024 color = PETSC_DRAW_RED; 1025 for (i=0; i<m; i++) { 1026 y_l = m - i - 1.0; y_r = y_l + 1.0; 1027 for (j=a->i[i]; j<a->i[i+1]; j++) { 1028 x_l = a->j[j]; x_r = x_l + 1.0; 1029 if (PetscRealPart(aa[j]) <= 0.) continue; 1030 PetscCall(PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color)); 1031 } 1032 } 1033 PetscDrawCollectiveEnd(draw); 1034 } else { 1035 /* use contour shading to indicate magnitude of values */ 1036 /* first determine max of all nonzero values */ 1037 PetscReal minv = 0.0, maxv = 0.0; 1038 PetscInt nz = a->nz, count = 0; 1039 PetscDraw popup; 1040 1041 for (i=0; i<nz; i++) { 1042 if (PetscAbsScalar(aa[i]) > maxv) maxv = PetscAbsScalar(aa[i]); 1043 } 1044 if (minv >= maxv) maxv = minv + PETSC_SMALL; 1045 PetscCall(PetscDrawGetPopup(draw,&popup)); 1046 PetscCall(PetscDrawScalePopup(popup,minv,maxv)); 1047 1048 PetscDrawCollectiveBegin(draw); 1049 for (i=0; i<m; i++) { 1050 y_l = m - i - 1.0; 1051 y_r = y_l + 1.0; 1052 for (j=a->i[i]; j<a->i[i+1]; j++) { 1053 x_l = a->j[j]; 1054 x_r = x_l + 1.0; 1055 color = PetscDrawRealToColor(PetscAbsScalar(aa[count]),minv,maxv); 1056 PetscCall(PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color)); 1057 count++; 1058 } 1059 } 1060 PetscDrawCollectiveEnd(draw); 1061 } 1062 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 1063 PetscFunctionReturn(0); 1064 } 1065 1066 #include <petscdraw.h> 1067 PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer) 1068 { 1069 PetscDraw draw; 1070 PetscReal xr,yr,xl,yl,h,w; 1071 PetscBool isnull; 1072 1073 PetscFunctionBegin; 1074 PetscCall(PetscViewerDrawGetDraw(viewer,0,&draw)); 1075 PetscCall(PetscDrawIsNull(draw,&isnull)); 1076 if (isnull) PetscFunctionReturn(0); 1077 1078 xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0; 1079 xr += w; yr += h; xl = -w; yl = -h; 1080 PetscCall(PetscDrawSetCoordinates(draw,xl,yl,xr,yr)); 1081 PetscCall(PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer)); 1082 PetscCall(PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A)); 1083 PetscCall(PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL)); 1084 PetscCall(PetscDrawSave(draw)); 1085 PetscFunctionReturn(0); 1086 } 1087 1088 PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer) 1089 { 1090 PetscBool iascii,isbinary,isdraw; 1091 1092 PetscFunctionBegin; 1093 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii)); 1094 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary)); 1095 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw)); 1096 if (iascii) { 1097 PetscCall(MatView_SeqAIJ_ASCII(A,viewer)); 1098 } else if (isbinary) { 1099 PetscCall(MatView_SeqAIJ_Binary(A,viewer)); 1100 } else if (isdraw) { 1101 PetscCall(MatView_SeqAIJ_Draw(A,viewer)); 1102 } 1103 PetscCall(MatView_SeqAIJ_Inode(A,viewer)); 1104 PetscFunctionReturn(0); 1105 } 1106 1107 PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode) 1108 { 1109 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1110 PetscInt fshift = 0,i,*ai = a->i,*aj = a->j,*imax = a->imax; 1111 PetscInt m = A->rmap->n,*ip,N,*ailen = a->ilen,rmax = 0; 1112 MatScalar *aa = a->a,*ap; 1113 PetscReal ratio = 0.6; 1114 1115 PetscFunctionBegin; 1116 if (mode == MAT_FLUSH_ASSEMBLY) PetscFunctionReturn(0); 1117 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 1118 if (A->was_assembled && A->ass_nonzerostate == A->nonzerostate) { 1119 /* we need to respect users asking to use or not the inodes routine in between matrix assemblies */ 1120 PetscCall(MatAssemblyEnd_SeqAIJ_Inode(A,mode)); 1121 PetscFunctionReturn(0); 1122 } 1123 1124 if (m) rmax = ailen[0]; /* determine row with most nonzeros */ 1125 for (i=1; i<m; i++) { 1126 /* move each row back by the amount of empty slots (fshift) before it*/ 1127 fshift += imax[i-1] - ailen[i-1]; 1128 rmax = PetscMax(rmax,ailen[i]); 1129 if (fshift) { 1130 ip = aj + ai[i]; 1131 ap = aa + ai[i]; 1132 N = ailen[i]; 1133 PetscCall(PetscArraymove(ip-fshift,ip,N)); 1134 if (!A->structure_only) { 1135 PetscCall(PetscArraymove(ap-fshift,ap,N)); 1136 } 1137 } 1138 ai[i] = ai[i-1] + ailen[i-1]; 1139 } 1140 if (m) { 1141 fshift += imax[m-1] - ailen[m-1]; 1142 ai[m] = ai[m-1] + ailen[m-1]; 1143 } 1144 1145 /* reset ilen and imax for each row */ 1146 a->nonzerorowcnt = 0; 1147 if (A->structure_only) { 1148 PetscCall(PetscFree(a->imax)); 1149 PetscCall(PetscFree(a->ilen)); 1150 } else { /* !A->structure_only */ 1151 for (i=0; i<m; i++) { 1152 ailen[i] = imax[i] = ai[i+1] - ai[i]; 1153 a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0); 1154 } 1155 } 1156 a->nz = ai[m]; 1157 PetscCheck(!fshift || a->nounused != -1,PETSC_COMM_SELF,PETSC_ERR_PLIB, "Unused space detected in matrix: %" PetscInt_FMT " X %" PetscInt_FMT ", %" PetscInt_FMT " unneeded", m, A->cmap->n, fshift); 1158 1159 PetscCall(MatMarkDiagonal_SeqAIJ(A)); 1160 PetscCall(PetscInfo(A,"Matrix size: %" PetscInt_FMT " X %" PetscInt_FMT "; storage space: %" PetscInt_FMT " unneeded,%" PetscInt_FMT " used\n",m,A->cmap->n,fshift,a->nz)); 1161 PetscCall(PetscInfo(A,"Number of mallocs during MatSetValues() is %" PetscInt_FMT "\n",a->reallocs)); 1162 PetscCall(PetscInfo(A,"Maximum nonzeros in any row is %" PetscInt_FMT "\n",rmax)); 1163 1164 A->info.mallocs += a->reallocs; 1165 a->reallocs = 0; 1166 A->info.nz_unneeded = (PetscReal)fshift; 1167 a->rmax = rmax; 1168 1169 if (!A->structure_only) { 1170 PetscCall(MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,m,ratio)); 1171 } 1172 PetscCall(MatAssemblyEnd_SeqAIJ_Inode(A,mode)); 1173 PetscFunctionReturn(0); 1174 } 1175 1176 PetscErrorCode MatRealPart_SeqAIJ(Mat A) 1177 { 1178 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1179 PetscInt i,nz = a->nz; 1180 MatScalar *aa; 1181 1182 PetscFunctionBegin; 1183 PetscCall(MatSeqAIJGetArray(A,&aa)); 1184 for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]); 1185 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 1186 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 1187 PetscFunctionReturn(0); 1188 } 1189 1190 PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A) 1191 { 1192 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1193 PetscInt i,nz = a->nz; 1194 MatScalar *aa; 1195 1196 PetscFunctionBegin; 1197 PetscCall(MatSeqAIJGetArray(A,&aa)); 1198 for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]); 1199 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 1200 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 1201 PetscFunctionReturn(0); 1202 } 1203 1204 PetscErrorCode MatZeroEntries_SeqAIJ(Mat A) 1205 { 1206 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1207 MatScalar *aa; 1208 1209 PetscFunctionBegin; 1210 PetscCall(MatSeqAIJGetArrayWrite(A,&aa)); 1211 PetscCall(PetscArrayzero(aa,a->i[A->rmap->n])); 1212 PetscCall(MatSeqAIJRestoreArrayWrite(A,&aa)); 1213 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 1214 PetscFunctionReturn(0); 1215 } 1216 1217 PETSC_INTERN PetscErrorCode MatResetPreallocationCOO_SeqAIJ(Mat A) 1218 { 1219 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1220 1221 PetscFunctionBegin; 1222 PetscCall(PetscFree(a->perm)); 1223 PetscCall(PetscFree(a->jmap)); 1224 PetscFunctionReturn(0); 1225 } 1226 1227 PetscErrorCode MatDestroy_SeqAIJ(Mat A) 1228 { 1229 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1230 1231 PetscFunctionBegin; 1232 #if defined(PETSC_USE_LOG) 1233 PetscLogObjectState((PetscObject)A,"Rows=%" PetscInt_FMT ", Cols=%" PetscInt_FMT ", NZ=%" PetscInt_FMT,A->rmap->n,A->cmap->n,a->nz); 1234 #endif 1235 PetscCall(MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i)); 1236 PetscCall(MatResetPreallocationCOO_SeqAIJ(A)); 1237 PetscCall(ISDestroy(&a->row)); 1238 PetscCall(ISDestroy(&a->col)); 1239 PetscCall(PetscFree(a->diag)); 1240 PetscCall(PetscFree(a->ibdiag)); 1241 PetscCall(PetscFree(a->imax)); 1242 PetscCall(PetscFree(a->ilen)); 1243 PetscCall(PetscFree(a->ipre)); 1244 PetscCall(PetscFree3(a->idiag,a->mdiag,a->ssor_work)); 1245 PetscCall(PetscFree(a->solve_work)); 1246 PetscCall(ISDestroy(&a->icol)); 1247 PetscCall(PetscFree(a->saved_values)); 1248 PetscCall(PetscFree2(a->compressedrow.i,a->compressedrow.rindex)); 1249 PetscCall(MatDestroy_SeqAIJ_Inode(A)); 1250 PetscCall(PetscFree(A->data)); 1251 1252 /* MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted may allocate this. 1253 That function is so heavily used (sometimes in an hidden way through multnumeric function pointers) 1254 that is hard to properly add this data to the MatProduct data. We free it here to avoid 1255 users reusing the matrix object with different data to incur in obscure segmentation faults 1256 due to different matrix sizes */ 1257 PetscCall(PetscObjectCompose((PetscObject)A,"__PETSc__ab_dense",NULL)); 1258 1259 PetscCall(PetscObjectChangeTypeName((PetscObject)A,NULL)); 1260 PetscCall(PetscObjectComposeFunction((PetscObject)A,"PetscMatlabEnginePut_C",NULL)); 1261 PetscCall(PetscObjectComposeFunction((PetscObject)A,"PetscMatlabEngineGet_C",NULL)); 1262 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetColumnIndices_C",NULL)); 1263 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL)); 1264 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL)); 1265 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsbaij_C",NULL)); 1266 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqbaij_C",NULL)); 1267 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijperm_C",NULL)); 1268 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijsell_C",NULL)); 1269 #if defined(PETSC_HAVE_MKL_SPARSE) 1270 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijmkl_C",NULL)); 1271 #endif 1272 #if defined(PETSC_HAVE_CUDA) 1273 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijcusparse_C",NULL)); 1274 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaijcusparse_seqaij_C",NULL)); 1275 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaij_seqaijcusparse_C",NULL)); 1276 #endif 1277 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 1278 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijkokkos_C",NULL)); 1279 #endif 1280 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijcrl_C",NULL)); 1281 #if defined(PETSC_HAVE_ELEMENTAL) 1282 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_elemental_C",NULL)); 1283 #endif 1284 #if defined(PETSC_HAVE_SCALAPACK) 1285 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_scalapack_C",NULL)); 1286 #endif 1287 #if defined(PETSC_HAVE_HYPRE) 1288 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_hypre_C",NULL)); 1289 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_transpose_seqaij_seqaij_C",NULL)); 1290 #endif 1291 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqdense_C",NULL)); 1292 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsell_C",NULL)); 1293 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_is_C",NULL)); 1294 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL)); 1295 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatIsHermitianTranspose_C",NULL)); 1296 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocation_C",NULL)); 1297 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatResetPreallocation_C",NULL)); 1298 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C",NULL)); 1299 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatReorderForNonzeroDiagonal_C",NULL)); 1300 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_is_seqaij_C",NULL)); 1301 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqdense_seqaij_C",NULL)); 1302 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaij_seqaij_C",NULL)); 1303 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJKron_C",NULL)); 1304 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSetPreallocationCOO_C",NULL)); 1305 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSetValuesCOO_C",NULL)); 1306 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverType_C",NULL)); 1307 /* these calls do not belong here: the subclasses Duplicate/Destroy are wrong */ 1308 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaijsell_seqaij_C",NULL)); 1309 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaijperm_seqaij_C",NULL)); 1310 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijviennacl_C",NULL)); 1311 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaijviennacl_seqdense_C",NULL)); 1312 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaijviennacl_seqaij_C",NULL)); 1313 PetscFunctionReturn(0); 1314 } 1315 1316 PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op,PetscBool flg) 1317 { 1318 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1319 1320 PetscFunctionBegin; 1321 switch (op) { 1322 case MAT_ROW_ORIENTED: 1323 a->roworiented = flg; 1324 break; 1325 case MAT_KEEP_NONZERO_PATTERN: 1326 a->keepnonzeropattern = flg; 1327 break; 1328 case MAT_NEW_NONZERO_LOCATIONS: 1329 a->nonew = (flg ? 0 : 1); 1330 break; 1331 case MAT_NEW_NONZERO_LOCATION_ERR: 1332 a->nonew = (flg ? -1 : 0); 1333 break; 1334 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1335 a->nonew = (flg ? -2 : 0); 1336 break; 1337 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1338 a->nounused = (flg ? -1 : 0); 1339 break; 1340 case MAT_IGNORE_ZERO_ENTRIES: 1341 a->ignorezeroentries = flg; 1342 break; 1343 case MAT_SPD: 1344 case MAT_SYMMETRIC: 1345 case MAT_STRUCTURALLY_SYMMETRIC: 1346 case MAT_HERMITIAN: 1347 case MAT_SYMMETRY_ETERNAL: 1348 case MAT_STRUCTURE_ONLY: 1349 /* These options are handled directly by MatSetOption() */ 1350 break; 1351 case MAT_FORCE_DIAGONAL_ENTRIES: 1352 case MAT_IGNORE_OFF_PROC_ENTRIES: 1353 case MAT_USE_HASH_TABLE: 1354 PetscCall(PetscInfo(A,"Option %s ignored\n",MatOptions[op])); 1355 break; 1356 case MAT_USE_INODES: 1357 PetscCall(MatSetOption_SeqAIJ_Inode(A,MAT_USE_INODES,flg)); 1358 break; 1359 case MAT_SUBMAT_SINGLEIS: 1360 A->submat_singleis = flg; 1361 break; 1362 case MAT_SORTED_FULL: 1363 if (flg) A->ops->setvalues = MatSetValues_SeqAIJ_SortedFull; 1364 else A->ops->setvalues = MatSetValues_SeqAIJ; 1365 break; 1366 case MAT_FORM_EXPLICIT_TRANSPOSE: 1367 A->form_explicit_transpose = flg; 1368 break; 1369 default: 1370 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op); 1371 } 1372 PetscFunctionReturn(0); 1373 } 1374 1375 PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v) 1376 { 1377 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1378 PetscInt i,j,n,*ai=a->i,*aj=a->j; 1379 PetscScalar *x; 1380 const PetscScalar *aa; 1381 1382 PetscFunctionBegin; 1383 PetscCall(VecGetLocalSize(v,&n)); 1384 PetscCheck(n == A->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 1385 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 1386 if (A->factortype == MAT_FACTOR_ILU || A->factortype == MAT_FACTOR_LU) { 1387 PetscInt *diag=a->diag; 1388 PetscCall(VecGetArrayWrite(v,&x)); 1389 for (i=0; i<n; i++) x[i] = 1.0/aa[diag[i]]; 1390 PetscCall(VecRestoreArrayWrite(v,&x)); 1391 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 1392 PetscFunctionReturn(0); 1393 } 1394 1395 PetscCall(VecGetArrayWrite(v,&x)); 1396 for (i=0; i<n; i++) { 1397 x[i] = 0.0; 1398 for (j=ai[i]; j<ai[i+1]; j++) { 1399 if (aj[j] == i) { 1400 x[i] = aa[j]; 1401 break; 1402 } 1403 } 1404 } 1405 PetscCall(VecRestoreArrayWrite(v,&x)); 1406 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 1407 PetscFunctionReturn(0); 1408 } 1409 1410 #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h> 1411 PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy) 1412 { 1413 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1414 const MatScalar *aa; 1415 PetscScalar *y; 1416 const PetscScalar *x; 1417 PetscInt m = A->rmap->n; 1418 #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ) 1419 const MatScalar *v; 1420 PetscScalar alpha; 1421 PetscInt n,i,j; 1422 const PetscInt *idx,*ii,*ridx=NULL; 1423 Mat_CompressedRow cprow = a->compressedrow; 1424 PetscBool usecprow = cprow.use; 1425 #endif 1426 1427 PetscFunctionBegin; 1428 if (zz != yy) PetscCall(VecCopy(zz,yy)); 1429 PetscCall(VecGetArrayRead(xx,&x)); 1430 PetscCall(VecGetArray(yy,&y)); 1431 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 1432 1433 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ) 1434 fortranmulttransposeaddaij_(&m,x,a->i,a->j,aa,y); 1435 #else 1436 if (usecprow) { 1437 m = cprow.nrows; 1438 ii = cprow.i; 1439 ridx = cprow.rindex; 1440 } else { 1441 ii = a->i; 1442 } 1443 for (i=0; i<m; i++) { 1444 idx = a->j + ii[i]; 1445 v = aa + ii[i]; 1446 n = ii[i+1] - ii[i]; 1447 if (usecprow) { 1448 alpha = x[ridx[i]]; 1449 } else { 1450 alpha = x[i]; 1451 } 1452 for (j=0; j<n; j++) y[idx[j]] += alpha*v[j]; 1453 } 1454 #endif 1455 PetscCall(PetscLogFlops(2.0*a->nz)); 1456 PetscCall(VecRestoreArrayRead(xx,&x)); 1457 PetscCall(VecRestoreArray(yy,&y)); 1458 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 1459 PetscFunctionReturn(0); 1460 } 1461 1462 PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy) 1463 { 1464 PetscFunctionBegin; 1465 PetscCall(VecSet(yy,0.0)); 1466 PetscCall(MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy)); 1467 PetscFunctionReturn(0); 1468 } 1469 1470 #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h> 1471 1472 PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy) 1473 { 1474 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1475 PetscScalar *y; 1476 const PetscScalar *x; 1477 const MatScalar *aa,*a_a; 1478 PetscInt m=A->rmap->n; 1479 const PetscInt *aj,*ii,*ridx=NULL; 1480 PetscInt n,i; 1481 PetscScalar sum; 1482 PetscBool usecprow=a->compressedrow.use; 1483 1484 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1485 #pragma disjoint(*x,*y,*aa) 1486 #endif 1487 1488 PetscFunctionBegin; 1489 if (a->inode.use && a->inode.checked) { 1490 PetscCall(MatMult_SeqAIJ_Inode(A,xx,yy)); 1491 PetscFunctionReturn(0); 1492 } 1493 PetscCall(MatSeqAIJGetArrayRead(A,&a_a)); 1494 PetscCall(VecGetArrayRead(xx,&x)); 1495 PetscCall(VecGetArray(yy,&y)); 1496 ii = a->i; 1497 if (usecprow) { /* use compressed row format */ 1498 PetscCall(PetscArrayzero(y,m)); 1499 m = a->compressedrow.nrows; 1500 ii = a->compressedrow.i; 1501 ridx = a->compressedrow.rindex; 1502 for (i=0; i<m; i++) { 1503 n = ii[i+1] - ii[i]; 1504 aj = a->j + ii[i]; 1505 aa = a_a + ii[i]; 1506 sum = 0.0; 1507 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1508 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1509 y[*ridx++] = sum; 1510 } 1511 } else { /* do not use compressed row format */ 1512 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ) 1513 aj = a->j; 1514 aa = a_a; 1515 fortranmultaij_(&m,x,ii,aj,aa,y); 1516 #else 1517 for (i=0; i<m; i++) { 1518 n = ii[i+1] - ii[i]; 1519 aj = a->j + ii[i]; 1520 aa = a_a + ii[i]; 1521 sum = 0.0; 1522 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1523 y[i] = sum; 1524 } 1525 #endif 1526 } 1527 PetscCall(PetscLogFlops(2.0*a->nz - a->nonzerorowcnt)); 1528 PetscCall(VecRestoreArrayRead(xx,&x)); 1529 PetscCall(VecRestoreArray(yy,&y)); 1530 PetscCall(MatSeqAIJRestoreArrayRead(A,&a_a)); 1531 PetscFunctionReturn(0); 1532 } 1533 1534 PetscErrorCode MatMultMax_SeqAIJ(Mat A,Vec xx,Vec yy) 1535 { 1536 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1537 PetscScalar *y; 1538 const PetscScalar *x; 1539 const MatScalar *aa,*a_a; 1540 PetscInt m=A->rmap->n; 1541 const PetscInt *aj,*ii,*ridx=NULL; 1542 PetscInt n,i,nonzerorow=0; 1543 PetscScalar sum; 1544 PetscBool usecprow=a->compressedrow.use; 1545 1546 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1547 #pragma disjoint(*x,*y,*aa) 1548 #endif 1549 1550 PetscFunctionBegin; 1551 PetscCall(MatSeqAIJGetArrayRead(A,&a_a)); 1552 PetscCall(VecGetArrayRead(xx,&x)); 1553 PetscCall(VecGetArray(yy,&y)); 1554 if (usecprow) { /* use compressed row format */ 1555 m = a->compressedrow.nrows; 1556 ii = a->compressedrow.i; 1557 ridx = a->compressedrow.rindex; 1558 for (i=0; i<m; i++) { 1559 n = ii[i+1] - ii[i]; 1560 aj = a->j + ii[i]; 1561 aa = a_a + ii[i]; 1562 sum = 0.0; 1563 nonzerorow += (n>0); 1564 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1565 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1566 y[*ridx++] = sum; 1567 } 1568 } else { /* do not use compressed row format */ 1569 ii = a->i; 1570 for (i=0; i<m; i++) { 1571 n = ii[i+1] - ii[i]; 1572 aj = a->j + ii[i]; 1573 aa = a_a + ii[i]; 1574 sum = 0.0; 1575 nonzerorow += (n>0); 1576 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1577 y[i] = sum; 1578 } 1579 } 1580 PetscCall(PetscLogFlops(2.0*a->nz - nonzerorow)); 1581 PetscCall(VecRestoreArrayRead(xx,&x)); 1582 PetscCall(VecRestoreArray(yy,&y)); 1583 PetscCall(MatSeqAIJRestoreArrayRead(A,&a_a)); 1584 PetscFunctionReturn(0); 1585 } 1586 1587 PetscErrorCode MatMultAddMax_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1588 { 1589 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1590 PetscScalar *y,*z; 1591 const PetscScalar *x; 1592 const MatScalar *aa,*a_a; 1593 PetscInt m = A->rmap->n,*aj,*ii; 1594 PetscInt n,i,*ridx=NULL; 1595 PetscScalar sum; 1596 PetscBool usecprow=a->compressedrow.use; 1597 1598 PetscFunctionBegin; 1599 PetscCall(MatSeqAIJGetArrayRead(A,&a_a)); 1600 PetscCall(VecGetArrayRead(xx,&x)); 1601 PetscCall(VecGetArrayPair(yy,zz,&y,&z)); 1602 if (usecprow) { /* use compressed row format */ 1603 if (zz != yy) { 1604 PetscCall(PetscArraycpy(z,y,m)); 1605 } 1606 m = a->compressedrow.nrows; 1607 ii = a->compressedrow.i; 1608 ridx = a->compressedrow.rindex; 1609 for (i=0; i<m; i++) { 1610 n = ii[i+1] - ii[i]; 1611 aj = a->j + ii[i]; 1612 aa = a_a + ii[i]; 1613 sum = y[*ridx]; 1614 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1615 z[*ridx++] = sum; 1616 } 1617 } else { /* do not use compressed row format */ 1618 ii = a->i; 1619 for (i=0; i<m; i++) { 1620 n = ii[i+1] - ii[i]; 1621 aj = a->j + ii[i]; 1622 aa = a_a + ii[i]; 1623 sum = y[i]; 1624 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1625 z[i] = sum; 1626 } 1627 } 1628 PetscCall(PetscLogFlops(2.0*a->nz)); 1629 PetscCall(VecRestoreArrayRead(xx,&x)); 1630 PetscCall(VecRestoreArrayPair(yy,zz,&y,&z)); 1631 PetscCall(MatSeqAIJRestoreArrayRead(A,&a_a)); 1632 PetscFunctionReturn(0); 1633 } 1634 1635 #include <../src/mat/impls/aij/seq/ftn-kernels/fmultadd.h> 1636 PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1637 { 1638 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1639 PetscScalar *y,*z; 1640 const PetscScalar *x; 1641 const MatScalar *aa,*a_a; 1642 const PetscInt *aj,*ii,*ridx=NULL; 1643 PetscInt m = A->rmap->n,n,i; 1644 PetscScalar sum; 1645 PetscBool usecprow=a->compressedrow.use; 1646 1647 PetscFunctionBegin; 1648 if (a->inode.use && a->inode.checked) { 1649 PetscCall(MatMultAdd_SeqAIJ_Inode(A,xx,yy,zz)); 1650 PetscFunctionReturn(0); 1651 } 1652 PetscCall(MatSeqAIJGetArrayRead(A,&a_a)); 1653 PetscCall(VecGetArrayRead(xx,&x)); 1654 PetscCall(VecGetArrayPair(yy,zz,&y,&z)); 1655 if (usecprow) { /* use compressed row format */ 1656 if (zz != yy) { 1657 PetscCall(PetscArraycpy(z,y,m)); 1658 } 1659 m = a->compressedrow.nrows; 1660 ii = a->compressedrow.i; 1661 ridx = a->compressedrow.rindex; 1662 for (i=0; i<m; i++) { 1663 n = ii[i+1] - ii[i]; 1664 aj = a->j + ii[i]; 1665 aa = a_a + ii[i]; 1666 sum = y[*ridx]; 1667 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1668 z[*ridx++] = sum; 1669 } 1670 } else { /* do not use compressed row format */ 1671 ii = a->i; 1672 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ) 1673 aj = a->j; 1674 aa = a_a; 1675 fortranmultaddaij_(&m,x,ii,aj,aa,y,z); 1676 #else 1677 for (i=0; i<m; i++) { 1678 n = ii[i+1] - ii[i]; 1679 aj = a->j + ii[i]; 1680 aa = a_a + ii[i]; 1681 sum = y[i]; 1682 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1683 z[i] = sum; 1684 } 1685 #endif 1686 } 1687 PetscCall(PetscLogFlops(2.0*a->nz)); 1688 PetscCall(VecRestoreArrayRead(xx,&x)); 1689 PetscCall(VecRestoreArrayPair(yy,zz,&y,&z)); 1690 PetscCall(MatSeqAIJRestoreArrayRead(A,&a_a)); 1691 PetscFunctionReturn(0); 1692 } 1693 1694 /* 1695 Adds diagonal pointers to sparse matrix structure. 1696 */ 1697 PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A) 1698 { 1699 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1700 PetscInt i,j,m = A->rmap->n; 1701 PetscBool alreadySet = PETSC_TRUE; 1702 1703 PetscFunctionBegin; 1704 if (!a->diag) { 1705 PetscCall(PetscMalloc1(m,&a->diag)); 1706 PetscCall(PetscLogObjectMemory((PetscObject)A, m*sizeof(PetscInt))); 1707 alreadySet = PETSC_FALSE; 1708 } 1709 for (i=0; i<A->rmap->n; i++) { 1710 /* If A's diagonal is already correctly set, this fast track enables cheap and repeated MatMarkDiagonal_SeqAIJ() calls */ 1711 if (alreadySet) { 1712 PetscInt pos = a->diag[i]; 1713 if (pos >= a->i[i] && pos < a->i[i+1] && a->j[pos] == i) continue; 1714 } 1715 1716 a->diag[i] = a->i[i+1]; 1717 for (j=a->i[i]; j<a->i[i+1]; j++) { 1718 if (a->j[j] == i) { 1719 a->diag[i] = j; 1720 break; 1721 } 1722 } 1723 } 1724 PetscFunctionReturn(0); 1725 } 1726 1727 PetscErrorCode MatShift_SeqAIJ(Mat A,PetscScalar v) 1728 { 1729 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1730 const PetscInt *diag = (const PetscInt*)a->diag; 1731 const PetscInt *ii = (const PetscInt*) a->i; 1732 PetscInt i,*mdiag = NULL; 1733 PetscInt cnt = 0; /* how many diagonals are missing */ 1734 1735 PetscFunctionBegin; 1736 if (!A->preallocated || !a->nz) { 1737 PetscCall(MatSeqAIJSetPreallocation(A,1,NULL)); 1738 PetscCall(MatShift_Basic(A,v)); 1739 PetscFunctionReturn(0); 1740 } 1741 1742 if (a->diagonaldense) { 1743 cnt = 0; 1744 } else { 1745 PetscCall(PetscCalloc1(A->rmap->n,&mdiag)); 1746 for (i=0; i<A->rmap->n; i++) { 1747 if (i < A->cmap->n && diag[i] >= ii[i+1]) { /* 'out of range' rows never have diagonals */ 1748 cnt++; 1749 mdiag[i] = 1; 1750 } 1751 } 1752 } 1753 if (!cnt) { 1754 PetscCall(MatShift_Basic(A,v)); 1755 } else { 1756 PetscScalar *olda = a->a; /* preserve pointers to current matrix nonzeros structure and values */ 1757 PetscInt *oldj = a->j, *oldi = a->i; 1758 PetscBool singlemalloc = a->singlemalloc,free_a = a->free_a,free_ij = a->free_ij; 1759 1760 a->a = NULL; 1761 a->j = NULL; 1762 a->i = NULL; 1763 /* increase the values in imax for each row where a diagonal is being inserted then reallocate the matrix data structures */ 1764 for (i=0; i<PetscMin(A->rmap->n,A->cmap->n); i++) { 1765 a->imax[i] += mdiag[i]; 1766 } 1767 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(A,0,a->imax)); 1768 1769 /* copy old values into new matrix data structure */ 1770 for (i=0; i<A->rmap->n; i++) { 1771 PetscCall(MatSetValues(A,1,&i,a->imax[i] - mdiag[i],&oldj[oldi[i]],&olda[oldi[i]],ADD_VALUES)); 1772 if (i < A->cmap->n) { 1773 PetscCall(MatSetValue(A,i,i,v,ADD_VALUES)); 1774 } 1775 } 1776 PetscCall(MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY)); 1777 PetscCall(MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY)); 1778 if (singlemalloc) { 1779 PetscCall(PetscFree3(olda,oldj,oldi)); 1780 } else { 1781 if (free_a) PetscCall(PetscFree(olda)); 1782 if (free_ij) PetscCall(PetscFree(oldj)); 1783 if (free_ij) PetscCall(PetscFree(oldi)); 1784 } 1785 } 1786 PetscCall(PetscFree(mdiag)); 1787 a->diagonaldense = PETSC_TRUE; 1788 PetscFunctionReturn(0); 1789 } 1790 1791 /* 1792 Checks for missing diagonals 1793 */ 1794 PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscBool *missing,PetscInt *d) 1795 { 1796 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1797 PetscInt *diag,*ii = a->i,i; 1798 1799 PetscFunctionBegin; 1800 *missing = PETSC_FALSE; 1801 if (A->rmap->n > 0 && !ii) { 1802 *missing = PETSC_TRUE; 1803 if (d) *d = 0; 1804 PetscCall(PetscInfo(A,"Matrix has no entries therefore is missing diagonal\n")); 1805 } else { 1806 PetscInt n; 1807 n = PetscMin(A->rmap->n, A->cmap->n); 1808 diag = a->diag; 1809 for (i=0; i<n; i++) { 1810 if (diag[i] >= ii[i+1]) { 1811 *missing = PETSC_TRUE; 1812 if (d) *d = i; 1813 PetscCall(PetscInfo(A,"Matrix is missing diagonal number %" PetscInt_FMT "\n",i)); 1814 break; 1815 } 1816 } 1817 } 1818 PetscFunctionReturn(0); 1819 } 1820 1821 #include <petscblaslapack.h> 1822 #include <petsc/private/kernels/blockinvert.h> 1823 1824 /* 1825 Note that values is allocated externally by the PC and then passed into this routine 1826 */ 1827 PetscErrorCode MatInvertVariableBlockDiagonal_SeqAIJ(Mat A,PetscInt nblocks,const PetscInt *bsizes,PetscScalar *diag) 1828 { 1829 PetscInt n = A->rmap->n, i, ncnt = 0, *indx,j,bsizemax = 0,*v_pivots; 1830 PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE; 1831 const PetscReal shift = 0.0; 1832 PetscInt ipvt[5]; 1833 PetscScalar work[25],*v_work; 1834 1835 PetscFunctionBegin; 1836 allowzeropivot = PetscNot(A->erroriffailure); 1837 for (i=0; i<nblocks; i++) ncnt += bsizes[i]; 1838 PetscCheck(ncnt == n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Total blocksizes %" PetscInt_FMT " doesn't match number matrix rows %" PetscInt_FMT,ncnt,n); 1839 for (i=0; i<nblocks; i++) { 1840 bsizemax = PetscMax(bsizemax,bsizes[i]); 1841 } 1842 PetscCall(PetscMalloc1(bsizemax,&indx)); 1843 if (bsizemax > 7) { 1844 PetscCall(PetscMalloc2(bsizemax,&v_work,bsizemax,&v_pivots)); 1845 } 1846 ncnt = 0; 1847 for (i=0; i<nblocks; i++) { 1848 for (j=0; j<bsizes[i]; j++) indx[j] = ncnt+j; 1849 PetscCall(MatGetValues(A,bsizes[i],indx,bsizes[i],indx,diag)); 1850 switch (bsizes[i]) { 1851 case 1: 1852 *diag = 1.0/(*diag); 1853 break; 1854 case 2: 1855 PetscCall(PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected)); 1856 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1857 PetscCall(PetscKernel_A_gets_transpose_A_2(diag)); 1858 break; 1859 case 3: 1860 PetscCall(PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected)); 1861 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1862 PetscCall(PetscKernel_A_gets_transpose_A_3(diag)); 1863 break; 1864 case 4: 1865 PetscCall(PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected)); 1866 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1867 PetscCall(PetscKernel_A_gets_transpose_A_4(diag)); 1868 break; 1869 case 5: 1870 PetscCall(PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected)); 1871 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1872 PetscCall(PetscKernel_A_gets_transpose_A_5(diag)); 1873 break; 1874 case 6: 1875 PetscCall(PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected)); 1876 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1877 PetscCall(PetscKernel_A_gets_transpose_A_6(diag)); 1878 break; 1879 case 7: 1880 PetscCall(PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected)); 1881 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1882 PetscCall(PetscKernel_A_gets_transpose_A_7(diag)); 1883 break; 1884 default: 1885 PetscCall(PetscKernel_A_gets_inverse_A(bsizes[i],diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected)); 1886 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1887 PetscCall(PetscKernel_A_gets_transpose_A_N(diag,bsizes[i])); 1888 } 1889 ncnt += bsizes[i]; 1890 diag += bsizes[i]*bsizes[i]; 1891 } 1892 if (bsizemax > 7) { 1893 PetscCall(PetscFree2(v_work,v_pivots)); 1894 } 1895 PetscCall(PetscFree(indx)); 1896 PetscFunctionReturn(0); 1897 } 1898 1899 /* 1900 Negative shift indicates do not generate an error if there is a zero diagonal, just invert it anyways 1901 */ 1902 PetscErrorCode MatInvertDiagonal_SeqAIJ(Mat A,PetscScalar omega,PetscScalar fshift) 1903 { 1904 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 1905 PetscInt i,*diag,m = A->rmap->n; 1906 const MatScalar *v; 1907 PetscScalar *idiag,*mdiag; 1908 1909 PetscFunctionBegin; 1910 if (a->idiagvalid) PetscFunctionReturn(0); 1911 PetscCall(MatMarkDiagonal_SeqAIJ(A)); 1912 diag = a->diag; 1913 if (!a->idiag) { 1914 PetscCall(PetscMalloc3(m,&a->idiag,m,&a->mdiag,m,&a->ssor_work)); 1915 PetscCall(PetscLogObjectMemory((PetscObject)A,3*m*sizeof(PetscScalar))); 1916 } 1917 1918 mdiag = a->mdiag; 1919 idiag = a->idiag; 1920 PetscCall(MatSeqAIJGetArrayRead(A,&v)); 1921 if (omega == 1.0 && PetscRealPart(fshift) <= 0.0) { 1922 for (i=0; i<m; i++) { 1923 mdiag[i] = v[diag[i]]; 1924 if (!PetscAbsScalar(mdiag[i])) { /* zero diagonal */ 1925 if (PetscRealPart(fshift)) { 1926 PetscCall(PetscInfo(A,"Zero diagonal on row %" PetscInt_FMT "\n",i)); 1927 A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1928 A->factorerror_zeropivot_value = 0.0; 1929 A->factorerror_zeropivot_row = i; 1930 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %" PetscInt_FMT,i); 1931 } 1932 idiag[i] = 1.0/v[diag[i]]; 1933 } 1934 PetscCall(PetscLogFlops(m)); 1935 } else { 1936 for (i=0; i<m; i++) { 1937 mdiag[i] = v[diag[i]]; 1938 idiag[i] = omega/(fshift + v[diag[i]]); 1939 } 1940 PetscCall(PetscLogFlops(2.0*m)); 1941 } 1942 a->idiagvalid = PETSC_TRUE; 1943 PetscCall(MatSeqAIJRestoreArrayRead(A,&v)); 1944 PetscFunctionReturn(0); 1945 } 1946 1947 #include <../src/mat/impls/aij/seq/ftn-kernels/frelax.h> 1948 PetscErrorCode MatSOR_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1949 { 1950 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1951 PetscScalar *x,d,sum,*t,scale; 1952 const MatScalar *v,*idiag=NULL,*mdiag,*aa; 1953 const PetscScalar *b, *bs,*xb, *ts; 1954 PetscInt n,m = A->rmap->n,i; 1955 const PetscInt *idx,*diag; 1956 1957 PetscFunctionBegin; 1958 if (a->inode.use && a->inode.checked && omega == 1.0 && fshift == 0.0) { 1959 PetscCall(MatSOR_SeqAIJ_Inode(A,bb,omega,flag,fshift,its,lits,xx)); 1960 PetscFunctionReturn(0); 1961 } 1962 its = its*lits; 1963 1964 if (fshift != a->fshift || omega != a->omega) a->idiagvalid = PETSC_FALSE; /* must recompute idiag[] */ 1965 if (!a->idiagvalid) PetscCall(MatInvertDiagonal_SeqAIJ(A,omega,fshift)); 1966 a->fshift = fshift; 1967 a->omega = omega; 1968 1969 diag = a->diag; 1970 t = a->ssor_work; 1971 idiag = a->idiag; 1972 mdiag = a->mdiag; 1973 1974 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 1975 PetscCall(VecGetArray(xx,&x)); 1976 PetscCall(VecGetArrayRead(bb,&b)); 1977 /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */ 1978 if (flag == SOR_APPLY_UPPER) { 1979 /* apply (U + D/omega) to the vector */ 1980 bs = b; 1981 for (i=0; i<m; i++) { 1982 d = fshift + mdiag[i]; 1983 n = a->i[i+1] - diag[i] - 1; 1984 idx = a->j + diag[i] + 1; 1985 v = aa + diag[i] + 1; 1986 sum = b[i]*d/omega; 1987 PetscSparseDensePlusDot(sum,bs,v,idx,n); 1988 x[i] = sum; 1989 } 1990 PetscCall(VecRestoreArray(xx,&x)); 1991 PetscCall(VecRestoreArrayRead(bb,&b)); 1992 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 1993 PetscCall(PetscLogFlops(a->nz)); 1994 PetscFunctionReturn(0); 1995 } 1996 1997 PetscCheck(flag != SOR_APPLY_LOWER,PETSC_COMM_SELF,PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented"); 1998 else if (flag & SOR_EISENSTAT) { 1999 /* Let A = L + U + D; where L is lower triangular, 2000 U is upper triangular, E = D/omega; This routine applies 2001 2002 (L + E)^{-1} A (U + E)^{-1} 2003 2004 to a vector efficiently using Eisenstat's trick. 2005 */ 2006 scale = (2.0/omega) - 1.0; 2007 2008 /* x = (E + U)^{-1} b */ 2009 for (i=m-1; i>=0; i--) { 2010 n = a->i[i+1] - diag[i] - 1; 2011 idx = a->j + diag[i] + 1; 2012 v = aa + diag[i] + 1; 2013 sum = b[i]; 2014 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2015 x[i] = sum*idiag[i]; 2016 } 2017 2018 /* t = b - (2*E - D)x */ 2019 v = aa; 2020 for (i=0; i<m; i++) t[i] = b[i] - scale*(v[*diag++])*x[i]; 2021 2022 /* t = (E + L)^{-1}t */ 2023 ts = t; 2024 diag = a->diag; 2025 for (i=0; i<m; i++) { 2026 n = diag[i] - a->i[i]; 2027 idx = a->j + a->i[i]; 2028 v = aa + a->i[i]; 2029 sum = t[i]; 2030 PetscSparseDenseMinusDot(sum,ts,v,idx,n); 2031 t[i] = sum*idiag[i]; 2032 /* x = x + t */ 2033 x[i] += t[i]; 2034 } 2035 2036 PetscCall(PetscLogFlops(6.0*m-1 + 2.0*a->nz)); 2037 PetscCall(VecRestoreArray(xx,&x)); 2038 PetscCall(VecRestoreArrayRead(bb,&b)); 2039 PetscFunctionReturn(0); 2040 } 2041 if (flag & SOR_ZERO_INITIAL_GUESS) { 2042 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 2043 for (i=0; i<m; i++) { 2044 n = diag[i] - a->i[i]; 2045 idx = a->j + a->i[i]; 2046 v = aa + a->i[i]; 2047 sum = b[i]; 2048 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2049 t[i] = sum; 2050 x[i] = sum*idiag[i]; 2051 } 2052 xb = t; 2053 PetscCall(PetscLogFlops(a->nz)); 2054 } else xb = b; 2055 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 2056 for (i=m-1; i>=0; i--) { 2057 n = a->i[i+1] - diag[i] - 1; 2058 idx = a->j + diag[i] + 1; 2059 v = aa + diag[i] + 1; 2060 sum = xb[i]; 2061 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2062 if (xb == b) { 2063 x[i] = sum*idiag[i]; 2064 } else { 2065 x[i] = (1-omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 2066 } 2067 } 2068 PetscCall(PetscLogFlops(a->nz)); /* assumes 1/2 in upper */ 2069 } 2070 its--; 2071 } 2072 while (its--) { 2073 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 2074 for (i=0; i<m; i++) { 2075 /* lower */ 2076 n = diag[i] - a->i[i]; 2077 idx = a->j + a->i[i]; 2078 v = aa + a->i[i]; 2079 sum = b[i]; 2080 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2081 t[i] = sum; /* save application of the lower-triangular part */ 2082 /* upper */ 2083 n = a->i[i+1] - diag[i] - 1; 2084 idx = a->j + diag[i] + 1; 2085 v = aa + diag[i] + 1; 2086 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2087 x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 2088 } 2089 xb = t; 2090 PetscCall(PetscLogFlops(2.0*a->nz)); 2091 } else xb = b; 2092 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 2093 for (i=m-1; i>=0; i--) { 2094 sum = xb[i]; 2095 if (xb == b) { 2096 /* whole matrix (no checkpointing available) */ 2097 n = a->i[i+1] - a->i[i]; 2098 idx = a->j + a->i[i]; 2099 v = aa + a->i[i]; 2100 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2101 x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i]; 2102 } else { /* lower-triangular part has been saved, so only apply upper-triangular */ 2103 n = a->i[i+1] - diag[i] - 1; 2104 idx = a->j + diag[i] + 1; 2105 v = aa + diag[i] + 1; 2106 PetscSparseDenseMinusDot(sum,x,v,idx,n); 2107 x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 2108 } 2109 } 2110 if (xb == b) { 2111 PetscCall(PetscLogFlops(2.0*a->nz)); 2112 } else { 2113 PetscCall(PetscLogFlops(a->nz)); /* assumes 1/2 in upper */ 2114 } 2115 } 2116 } 2117 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 2118 PetscCall(VecRestoreArray(xx,&x)); 2119 PetscCall(VecRestoreArrayRead(bb,&b)); 2120 PetscFunctionReturn(0); 2121 } 2122 2123 PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info) 2124 { 2125 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2126 2127 PetscFunctionBegin; 2128 info->block_size = 1.0; 2129 info->nz_allocated = a->maxnz; 2130 info->nz_used = a->nz; 2131 info->nz_unneeded = (a->maxnz - a->nz); 2132 info->assemblies = A->num_ass; 2133 info->mallocs = A->info.mallocs; 2134 info->memory = ((PetscObject)A)->mem; 2135 if (A->factortype) { 2136 info->fill_ratio_given = A->info.fill_ratio_given; 2137 info->fill_ratio_needed = A->info.fill_ratio_needed; 2138 info->factor_mallocs = A->info.factor_mallocs; 2139 } else { 2140 info->fill_ratio_given = 0; 2141 info->fill_ratio_needed = 0; 2142 info->factor_mallocs = 0; 2143 } 2144 PetscFunctionReturn(0); 2145 } 2146 2147 PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 2148 { 2149 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2150 PetscInt i,m = A->rmap->n - 1; 2151 const PetscScalar *xx; 2152 PetscScalar *bb,*aa; 2153 PetscInt d = 0; 2154 2155 PetscFunctionBegin; 2156 if (x && b) { 2157 PetscCall(VecGetArrayRead(x,&xx)); 2158 PetscCall(VecGetArray(b,&bb)); 2159 for (i=0; i<N; i++) { 2160 PetscCheck(rows[i] >= 0 && rows[i] <= m,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range", rows[i]); 2161 if (rows[i] >= A->cmap->n) continue; 2162 bb[rows[i]] = diag*xx[rows[i]]; 2163 } 2164 PetscCall(VecRestoreArrayRead(x,&xx)); 2165 PetscCall(VecRestoreArray(b,&bb)); 2166 } 2167 2168 PetscCall(MatSeqAIJGetArray(A,&aa)); 2169 if (a->keepnonzeropattern) { 2170 for (i=0; i<N; i++) { 2171 PetscCheck(rows[i] >= 0 && rows[i] <= m,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range", rows[i]); 2172 PetscCall(PetscArrayzero(&aa[a->i[rows[i]]],a->ilen[rows[i]])); 2173 } 2174 if (diag != 0.0) { 2175 for (i=0; i<N; i++) { 2176 d = rows[i]; 2177 if (rows[i] >= A->cmap->n) continue; 2178 PetscCheck(a->diag[d] < a->i[d+1],PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in the zeroed row %" PetscInt_FMT,d); 2179 } 2180 for (i=0; i<N; i++) { 2181 if (rows[i] >= A->cmap->n) continue; 2182 aa[a->diag[rows[i]]] = diag; 2183 } 2184 } 2185 } else { 2186 if (diag != 0.0) { 2187 for (i=0; i<N; i++) { 2188 PetscCheck(rows[i] >= 0 && rows[i] <= m,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range", rows[i]); 2189 if (a->ilen[rows[i]] > 0) { 2190 if (rows[i] >= A->cmap->n) { 2191 a->ilen[rows[i]] = 0; 2192 } else { 2193 a->ilen[rows[i]] = 1; 2194 aa[a->i[rows[i]]] = diag; 2195 a->j[a->i[rows[i]]] = rows[i]; 2196 } 2197 } else if (rows[i] < A->cmap->n) { /* in case row was completely empty */ 2198 PetscCall(MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES)); 2199 } 2200 } 2201 } else { 2202 for (i=0; i<N; i++) { 2203 PetscCheck(rows[i] >= 0 && rows[i] <= m,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range", rows[i]); 2204 a->ilen[rows[i]] = 0; 2205 } 2206 } 2207 A->nonzerostate++; 2208 } 2209 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 2210 PetscCall((*A->ops->assemblyend)(A,MAT_FINAL_ASSEMBLY)); 2211 PetscFunctionReturn(0); 2212 } 2213 2214 PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 2215 { 2216 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2217 PetscInt i,j,m = A->rmap->n - 1,d = 0; 2218 PetscBool missing,*zeroed,vecs = PETSC_FALSE; 2219 const PetscScalar *xx; 2220 PetscScalar *bb,*aa; 2221 2222 PetscFunctionBegin; 2223 if (!N) PetscFunctionReturn(0); 2224 PetscCall(MatSeqAIJGetArray(A,&aa)); 2225 if (x && b) { 2226 PetscCall(VecGetArrayRead(x,&xx)); 2227 PetscCall(VecGetArray(b,&bb)); 2228 vecs = PETSC_TRUE; 2229 } 2230 PetscCall(PetscCalloc1(A->rmap->n,&zeroed)); 2231 for (i=0; i<N; i++) { 2232 PetscCheck(rows[i] >= 0 && rows[i] <= m,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range", rows[i]); 2233 PetscCall(PetscArrayzero(&aa[a->i[rows[i]]],a->ilen[rows[i]])); 2234 2235 zeroed[rows[i]] = PETSC_TRUE; 2236 } 2237 for (i=0; i<A->rmap->n; i++) { 2238 if (!zeroed[i]) { 2239 for (j=a->i[i]; j<a->i[i+1]; j++) { 2240 if (a->j[j] < A->rmap->n && zeroed[a->j[j]]) { 2241 if (vecs) bb[i] -= aa[j]*xx[a->j[j]]; 2242 aa[j] = 0.0; 2243 } 2244 } 2245 } else if (vecs && i < A->cmap->N) bb[i] = diag*xx[i]; 2246 } 2247 if (x && b) { 2248 PetscCall(VecRestoreArrayRead(x,&xx)); 2249 PetscCall(VecRestoreArray(b,&bb)); 2250 } 2251 PetscCall(PetscFree(zeroed)); 2252 if (diag != 0.0) { 2253 PetscCall(MatMissingDiagonal_SeqAIJ(A,&missing,&d)); 2254 if (missing) { 2255 for (i=0; i<N; i++) { 2256 if (rows[i] >= A->cmap->N) continue; 2257 PetscCheck(!a->nonew || rows[i] < d,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %" PetscInt_FMT " (%" PetscInt_FMT ")",d,rows[i]); 2258 PetscCall(MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES)); 2259 } 2260 } else { 2261 for (i=0; i<N; i++) { 2262 aa[a->diag[rows[i]]] = diag; 2263 } 2264 } 2265 } 2266 PetscCall(MatSeqAIJRestoreArray(A,&aa)); 2267 PetscCall((*A->ops->assemblyend)(A,MAT_FINAL_ASSEMBLY)); 2268 PetscFunctionReturn(0); 2269 } 2270 2271 PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 2272 { 2273 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2274 const PetscScalar *aa; 2275 PetscInt *itmp; 2276 2277 PetscFunctionBegin; 2278 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 2279 *nz = a->i[row+1] - a->i[row]; 2280 if (v) *v = (PetscScalar*)(aa + a->i[row]); 2281 if (idx) { 2282 itmp = a->j + a->i[row]; 2283 if (*nz) *idx = itmp; 2284 else *idx = NULL; 2285 } 2286 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 2287 PetscFunctionReturn(0); 2288 } 2289 2290 PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 2291 { 2292 PetscFunctionBegin; 2293 if (nz) *nz = 0; 2294 if (idx) *idx = NULL; 2295 if (v) *v = NULL; 2296 PetscFunctionReturn(0); 2297 } 2298 2299 PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm) 2300 { 2301 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2302 const MatScalar *v; 2303 PetscReal sum = 0.0; 2304 PetscInt i,j; 2305 2306 PetscFunctionBegin; 2307 PetscCall(MatSeqAIJGetArrayRead(A,&v)); 2308 if (type == NORM_FROBENIUS) { 2309 #if defined(PETSC_USE_REAL___FP16) 2310 PetscBLASInt one = 1,nz = a->nz; 2311 PetscStackCallBLAS("BLASnrm2",*nrm = BLASnrm2_(&nz,v,&one)); 2312 #else 2313 for (i=0; i<a->nz; i++) { 2314 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 2315 } 2316 *nrm = PetscSqrtReal(sum); 2317 #endif 2318 PetscCall(PetscLogFlops(2.0*a->nz)); 2319 } else if (type == NORM_1) { 2320 PetscReal *tmp; 2321 PetscInt *jj = a->j; 2322 PetscCall(PetscCalloc1(A->cmap->n+1,&tmp)); 2323 *nrm = 0.0; 2324 for (j=0; j<a->nz; j++) { 2325 tmp[*jj++] += PetscAbsScalar(*v); v++; 2326 } 2327 for (j=0; j<A->cmap->n; j++) { 2328 if (tmp[j] > *nrm) *nrm = tmp[j]; 2329 } 2330 PetscCall(PetscFree(tmp)); 2331 PetscCall(PetscLogFlops(PetscMax(a->nz-1,0))); 2332 } else if (type == NORM_INFINITY) { 2333 *nrm = 0.0; 2334 for (j=0; j<A->rmap->n; j++) { 2335 const PetscScalar *v2 = v + a->i[j]; 2336 sum = 0.0; 2337 for (i=0; i<a->i[j+1]-a->i[j]; i++) { 2338 sum += PetscAbsScalar(*v2); v2++; 2339 } 2340 if (sum > *nrm) *nrm = sum; 2341 } 2342 PetscCall(PetscLogFlops(PetscMax(a->nz-1,0))); 2343 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm"); 2344 PetscCall(MatSeqAIJRestoreArrayRead(A,&v)); 2345 PetscFunctionReturn(0); 2346 } 2347 2348 /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */ 2349 PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B) 2350 { 2351 PetscInt i,j,anzj; 2352 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b; 2353 PetscInt an=A->cmap->N,am=A->rmap->N; 2354 PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j; 2355 2356 PetscFunctionBegin; 2357 /* Allocate space for symbolic transpose info and work array */ 2358 PetscCall(PetscCalloc1(an+1,&ati)); 2359 PetscCall(PetscMalloc1(ai[am],&atj)); 2360 PetscCall(PetscMalloc1(an,&atfill)); 2361 2362 /* Walk through aj and count ## of non-zeros in each row of A^T. */ 2363 /* Note: offset by 1 for fast conversion into csr format. */ 2364 for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1; 2365 /* Form ati for csr format of A^T. */ 2366 for (i=0;i<an;i++) ati[i+1] += ati[i]; 2367 2368 /* Copy ati into atfill so we have locations of the next free space in atj */ 2369 PetscCall(PetscArraycpy(atfill,ati,an)); 2370 2371 /* Walk through A row-wise and mark nonzero entries of A^T. */ 2372 for (i=0;i<am;i++) { 2373 anzj = ai[i+1] - ai[i]; 2374 for (j=0;j<anzj;j++) { 2375 atj[atfill[*aj]] = i; 2376 atfill[*aj++] += 1; 2377 } 2378 } 2379 2380 /* Clean up temporary space and complete requests. */ 2381 PetscCall(PetscFree(atfill)); 2382 PetscCall(MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B)); 2383 PetscCall(MatSetBlockSizes(*B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs))); 2384 PetscCall(MatSetType(*B,((PetscObject)A)->type_name)); 2385 2386 b = (Mat_SeqAIJ*)((*B)->data); 2387 b->free_a = PETSC_FALSE; 2388 b->free_ij = PETSC_TRUE; 2389 b->nonew = 0; 2390 PetscFunctionReturn(0); 2391 } 2392 2393 PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2394 { 2395 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) B->data; 2396 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2397 const MatScalar *va,*vb; 2398 PetscInt ma,na,mb,nb, i; 2399 2400 PetscFunctionBegin; 2401 PetscCall(MatGetSize(A,&ma,&na)); 2402 PetscCall(MatGetSize(B,&mb,&nb)); 2403 if (ma!=nb || na!=mb) { 2404 *f = PETSC_FALSE; 2405 PetscFunctionReturn(0); 2406 } 2407 PetscCall(MatSeqAIJGetArrayRead(A,&va)); 2408 PetscCall(MatSeqAIJGetArrayRead(B,&vb)); 2409 aii = aij->i; bii = bij->i; 2410 adx = aij->j; bdx = bij->j; 2411 PetscCall(PetscMalloc1(ma,&aptr)); 2412 PetscCall(PetscMalloc1(mb,&bptr)); 2413 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2414 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2415 2416 *f = PETSC_TRUE; 2417 for (i=0; i<ma; i++) { 2418 while (aptr[i]<aii[i+1]) { 2419 PetscInt idc,idr; 2420 PetscScalar vc,vr; 2421 /* column/row index/value */ 2422 idc = adx[aptr[i]]; 2423 idr = bdx[bptr[idc]]; 2424 vc = va[aptr[i]]; 2425 vr = vb[bptr[idc]]; 2426 if (i!=idr || PetscAbsScalar(vc-vr) > tol) { 2427 *f = PETSC_FALSE; 2428 goto done; 2429 } else { 2430 aptr[i]++; 2431 if (B || i!=idc) bptr[idc]++; 2432 } 2433 } 2434 } 2435 done: 2436 PetscCall(PetscFree(aptr)); 2437 PetscCall(PetscFree(bptr)); 2438 PetscCall(MatSeqAIJRestoreArrayRead(A,&va)); 2439 PetscCall(MatSeqAIJRestoreArrayRead(B,&vb)); 2440 PetscFunctionReturn(0); 2441 } 2442 2443 PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2444 { 2445 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) B->data; 2446 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2447 MatScalar *va,*vb; 2448 PetscInt ma,na,mb,nb, i; 2449 2450 PetscFunctionBegin; 2451 PetscCall(MatGetSize(A,&ma,&na)); 2452 PetscCall(MatGetSize(B,&mb,&nb)); 2453 if (ma!=nb || na!=mb) { 2454 *f = PETSC_FALSE; 2455 PetscFunctionReturn(0); 2456 } 2457 aii = aij->i; bii = bij->i; 2458 adx = aij->j; bdx = bij->j; 2459 va = aij->a; vb = bij->a; 2460 PetscCall(PetscMalloc1(ma,&aptr)); 2461 PetscCall(PetscMalloc1(mb,&bptr)); 2462 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2463 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2464 2465 *f = PETSC_TRUE; 2466 for (i=0; i<ma; i++) { 2467 while (aptr[i]<aii[i+1]) { 2468 PetscInt idc,idr; 2469 PetscScalar vc,vr; 2470 /* column/row index/value */ 2471 idc = adx[aptr[i]]; 2472 idr = bdx[bptr[idc]]; 2473 vc = va[aptr[i]]; 2474 vr = vb[bptr[idc]]; 2475 if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) { 2476 *f = PETSC_FALSE; 2477 goto done; 2478 } else { 2479 aptr[i]++; 2480 if (B || i!=idc) bptr[idc]++; 2481 } 2482 } 2483 } 2484 done: 2485 PetscCall(PetscFree(aptr)); 2486 PetscCall(PetscFree(bptr)); 2487 PetscFunctionReturn(0); 2488 } 2489 2490 PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2491 { 2492 PetscFunctionBegin; 2493 PetscCall(MatIsTranspose_SeqAIJ(A,A,tol,f)); 2494 PetscFunctionReturn(0); 2495 } 2496 2497 PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2498 { 2499 PetscFunctionBegin; 2500 PetscCall(MatIsHermitianTranspose_SeqAIJ(A,A,tol,f)); 2501 PetscFunctionReturn(0); 2502 } 2503 2504 PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr) 2505 { 2506 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2507 const PetscScalar *l,*r; 2508 PetscScalar x; 2509 MatScalar *v; 2510 PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz; 2511 const PetscInt *jj; 2512 2513 PetscFunctionBegin; 2514 if (ll) { 2515 /* The local size is used so that VecMPI can be passed to this routine 2516 by MatDiagonalScale_MPIAIJ */ 2517 PetscCall(VecGetLocalSize(ll,&m)); 2518 PetscCheck(m == A->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length"); 2519 PetscCall(VecGetArrayRead(ll,&l)); 2520 PetscCall(MatSeqAIJGetArray(A,&v)); 2521 for (i=0; i<m; i++) { 2522 x = l[i]; 2523 M = a->i[i+1] - a->i[i]; 2524 for (j=0; j<M; j++) (*v++) *= x; 2525 } 2526 PetscCall(VecRestoreArrayRead(ll,&l)); 2527 PetscCall(PetscLogFlops(nz)); 2528 PetscCall(MatSeqAIJRestoreArray(A,&v)); 2529 } 2530 if (rr) { 2531 PetscCall(VecGetLocalSize(rr,&n)); 2532 PetscCheck(n == A->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length"); 2533 PetscCall(VecGetArrayRead(rr,&r)); 2534 PetscCall(MatSeqAIJGetArray(A,&v)); 2535 jj = a->j; 2536 for (i=0; i<nz; i++) (*v++) *= r[*jj++]; 2537 PetscCall(MatSeqAIJRestoreArray(A,&v)); 2538 PetscCall(VecRestoreArrayRead(rr,&r)); 2539 PetscCall(PetscLogFlops(nz)); 2540 } 2541 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 2542 PetscFunctionReturn(0); 2543 } 2544 2545 PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B) 2546 { 2547 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c; 2548 PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens; 2549 PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi; 2550 const PetscInt *irow,*icol; 2551 const PetscScalar *aa; 2552 PetscInt nrows,ncols; 2553 PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen; 2554 MatScalar *a_new,*mat_a; 2555 Mat C; 2556 PetscBool stride; 2557 2558 PetscFunctionBegin; 2559 PetscCall(ISGetIndices(isrow,&irow)); 2560 PetscCall(ISGetLocalSize(isrow,&nrows)); 2561 PetscCall(ISGetLocalSize(iscol,&ncols)); 2562 2563 PetscCall(PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride)); 2564 if (stride) { 2565 PetscCall(ISStrideGetInfo(iscol,&first,&step)); 2566 } else { 2567 first = 0; 2568 step = 0; 2569 } 2570 if (stride && step == 1) { 2571 /* special case of contiguous rows */ 2572 PetscCall(PetscMalloc2(nrows,&lens,nrows,&starts)); 2573 /* loop over new rows determining lens and starting points */ 2574 for (i=0; i<nrows; i++) { 2575 kstart = ai[irow[i]]; 2576 kend = kstart + ailen[irow[i]]; 2577 starts[i] = kstart; 2578 for (k=kstart; k<kend; k++) { 2579 if (aj[k] >= first) { 2580 starts[i] = k; 2581 break; 2582 } 2583 } 2584 sum = 0; 2585 while (k < kend) { 2586 if (aj[k++] >= first+ncols) break; 2587 sum++; 2588 } 2589 lens[i] = sum; 2590 } 2591 /* create submatrix */ 2592 if (scall == MAT_REUSE_MATRIX) { 2593 PetscInt n_cols,n_rows; 2594 PetscCall(MatGetSize(*B,&n_rows,&n_cols)); 2595 PetscCheck(n_rows == nrows && n_cols == ncols,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size"); 2596 PetscCall(MatZeroEntries(*B)); 2597 C = *B; 2598 } else { 2599 PetscInt rbs,cbs; 2600 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&C)); 2601 PetscCall(MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE)); 2602 PetscCall(ISGetBlockSize(isrow,&rbs)); 2603 PetscCall(ISGetBlockSize(iscol,&cbs)); 2604 PetscCall(MatSetBlockSizes(C,rbs,cbs)); 2605 PetscCall(MatSetType(C,((PetscObject)A)->type_name)); 2606 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens)); 2607 } 2608 c = (Mat_SeqAIJ*)C->data; 2609 2610 /* loop over rows inserting into submatrix */ 2611 a_new = c->a; 2612 j_new = c->j; 2613 i_new = c->i; 2614 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 2615 for (i=0; i<nrows; i++) { 2616 ii = starts[i]; 2617 lensi = lens[i]; 2618 for (k=0; k<lensi; k++) { 2619 *j_new++ = aj[ii+k] - first; 2620 } 2621 PetscCall(PetscArraycpy(a_new,aa + starts[i],lensi)); 2622 a_new += lensi; 2623 i_new[i+1] = i_new[i] + lensi; 2624 c->ilen[i] = lensi; 2625 } 2626 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 2627 PetscCall(PetscFree2(lens,starts)); 2628 } else { 2629 PetscCall(ISGetIndices(iscol,&icol)); 2630 PetscCall(PetscCalloc1(oldcols,&smap)); 2631 PetscCall(PetscMalloc1(1+nrows,&lens)); 2632 for (i=0; i<ncols; i++) { 2633 PetscCheck(icol[i] < oldcols,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Requesting column beyond largest column icol[%" PetscInt_FMT "] %" PetscInt_FMT " >= A->cmap->n %" PetscInt_FMT,i,icol[i],oldcols); 2634 smap[icol[i]] = i+1; 2635 } 2636 2637 /* determine lens of each row */ 2638 for (i=0; i<nrows; i++) { 2639 kstart = ai[irow[i]]; 2640 kend = kstart + a->ilen[irow[i]]; 2641 lens[i] = 0; 2642 for (k=kstart; k<kend; k++) { 2643 if (smap[aj[k]]) { 2644 lens[i]++; 2645 } 2646 } 2647 } 2648 /* Create and fill new matrix */ 2649 if (scall == MAT_REUSE_MATRIX) { 2650 PetscBool equal; 2651 2652 c = (Mat_SeqAIJ*)((*B)->data); 2653 PetscCheck((*B)->rmap->n == nrows && (*B)->cmap->n == ncols,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size"); 2654 PetscCall(PetscArraycmp(c->ilen,lens,(*B)->rmap->n,&equal)); 2655 PetscCheck(equal,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros"); 2656 PetscCall(PetscArrayzero(c->ilen,(*B)->rmap->n)); 2657 C = *B; 2658 } else { 2659 PetscInt rbs,cbs; 2660 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&C)); 2661 PetscCall(MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE)); 2662 PetscCall(ISGetBlockSize(isrow,&rbs)); 2663 PetscCall(ISGetBlockSize(iscol,&cbs)); 2664 PetscCall(MatSetBlockSizes(C,rbs,cbs)); 2665 PetscCall(MatSetType(C,((PetscObject)A)->type_name)); 2666 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens)); 2667 } 2668 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 2669 c = (Mat_SeqAIJ*)(C->data); 2670 for (i=0; i<nrows; i++) { 2671 row = irow[i]; 2672 kstart = ai[row]; 2673 kend = kstart + a->ilen[row]; 2674 mat_i = c->i[i]; 2675 mat_j = c->j + mat_i; 2676 mat_a = c->a + mat_i; 2677 mat_ilen = c->ilen + i; 2678 for (k=kstart; k<kend; k++) { 2679 if ((tcol=smap[a->j[k]])) { 2680 *mat_j++ = tcol - 1; 2681 *mat_a++ = aa[k]; 2682 (*mat_ilen)++; 2683 2684 } 2685 } 2686 } 2687 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 2688 /* Free work space */ 2689 PetscCall(ISRestoreIndices(iscol,&icol)); 2690 PetscCall(PetscFree(smap)); 2691 PetscCall(PetscFree(lens)); 2692 /* sort */ 2693 for (i = 0; i < nrows; i++) { 2694 PetscInt ilen; 2695 2696 mat_i = c->i[i]; 2697 mat_j = c->j + mat_i; 2698 mat_a = c->a + mat_i; 2699 ilen = c->ilen[i]; 2700 PetscCall(PetscSortIntWithScalarArray(ilen,mat_j,mat_a)); 2701 } 2702 } 2703 #if defined(PETSC_HAVE_DEVICE) 2704 PetscCall(MatBindToCPU(C,A->boundtocpu)); 2705 #endif 2706 PetscCall(MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY)); 2707 PetscCall(MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY)); 2708 2709 PetscCall(ISRestoreIndices(isrow,&irow)); 2710 *B = C; 2711 PetscFunctionReturn(0); 2712 } 2713 2714 PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat) 2715 { 2716 Mat B; 2717 2718 PetscFunctionBegin; 2719 if (scall == MAT_INITIAL_MATRIX) { 2720 PetscCall(MatCreate(subComm,&B)); 2721 PetscCall(MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n)); 2722 PetscCall(MatSetBlockSizesFromMats(B,mat,mat)); 2723 PetscCall(MatSetType(B,MATSEQAIJ)); 2724 PetscCall(MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE)); 2725 *subMat = B; 2726 } else { 2727 PetscCall(MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN)); 2728 } 2729 PetscFunctionReturn(0); 2730 } 2731 2732 PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info) 2733 { 2734 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2735 Mat outA; 2736 PetscBool row_identity,col_identity; 2737 2738 PetscFunctionBegin; 2739 PetscCheck(info->levels == 0,PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu"); 2740 2741 PetscCall(ISIdentity(row,&row_identity)); 2742 PetscCall(ISIdentity(col,&col_identity)); 2743 2744 outA = inA; 2745 outA->factortype = MAT_FACTOR_LU; 2746 PetscCall(PetscFree(inA->solvertype)); 2747 PetscCall(PetscStrallocpy(MATSOLVERPETSC,&inA->solvertype)); 2748 2749 PetscCall(PetscObjectReference((PetscObject)row)); 2750 PetscCall(ISDestroy(&a->row)); 2751 2752 a->row = row; 2753 2754 PetscCall(PetscObjectReference((PetscObject)col)); 2755 PetscCall(ISDestroy(&a->col)); 2756 2757 a->col = col; 2758 2759 /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */ 2760 PetscCall(ISDestroy(&a->icol)); 2761 PetscCall(ISInvertPermutation(col,PETSC_DECIDE,&a->icol)); 2762 PetscCall(PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol)); 2763 2764 if (!a->solve_work) { /* this matrix may have been factored before */ 2765 PetscCall(PetscMalloc1(inA->rmap->n+1,&a->solve_work)); 2766 PetscCall(PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar))); 2767 } 2768 2769 PetscCall(MatMarkDiagonal_SeqAIJ(inA)); 2770 if (row_identity && col_identity) { 2771 PetscCall(MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info)); 2772 } else { 2773 PetscCall(MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info)); 2774 } 2775 PetscFunctionReturn(0); 2776 } 2777 2778 PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha) 2779 { 2780 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2781 PetscScalar *v; 2782 PetscBLASInt one = 1,bnz; 2783 2784 PetscFunctionBegin; 2785 PetscCall(MatSeqAIJGetArray(inA,&v)); 2786 PetscCall(PetscBLASIntCast(a->nz,&bnz)); 2787 PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&alpha,v,&one)); 2788 PetscCall(PetscLogFlops(a->nz)); 2789 PetscCall(MatSeqAIJRestoreArray(inA,&v)); 2790 PetscCall(MatSeqAIJInvalidateDiagonal(inA)); 2791 PetscFunctionReturn(0); 2792 } 2793 2794 PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt *submatj) 2795 { 2796 PetscInt i; 2797 2798 PetscFunctionBegin; 2799 if (!submatj->id) { /* delete data that are linked only to submats[id=0] */ 2800 PetscCall(PetscFree4(submatj->sbuf1,submatj->ptr,submatj->tmp,submatj->ctr)); 2801 2802 for (i=0; i<submatj->nrqr; ++i) { 2803 PetscCall(PetscFree(submatj->sbuf2[i])); 2804 } 2805 PetscCall(PetscFree3(submatj->sbuf2,submatj->req_size,submatj->req_source1)); 2806 2807 if (submatj->rbuf1) { 2808 PetscCall(PetscFree(submatj->rbuf1[0])); 2809 PetscCall(PetscFree(submatj->rbuf1)); 2810 } 2811 2812 for (i=0; i<submatj->nrqs; ++i) { 2813 PetscCall(PetscFree(submatj->rbuf3[i])); 2814 } 2815 PetscCall(PetscFree3(submatj->req_source2,submatj->rbuf2,submatj->rbuf3)); 2816 PetscCall(PetscFree(submatj->pa)); 2817 } 2818 2819 #if defined(PETSC_USE_CTABLE) 2820 PetscCall(PetscTableDestroy((PetscTable*)&submatj->rmap)); 2821 if (submatj->cmap_loc) PetscCall(PetscFree(submatj->cmap_loc)); 2822 PetscCall(PetscFree(submatj->rmap_loc)); 2823 #else 2824 PetscCall(PetscFree(submatj->rmap)); 2825 #endif 2826 2827 if (!submatj->allcolumns) { 2828 #if defined(PETSC_USE_CTABLE) 2829 PetscCall(PetscTableDestroy((PetscTable*)&submatj->cmap)); 2830 #else 2831 PetscCall(PetscFree(submatj->cmap)); 2832 #endif 2833 } 2834 PetscCall(PetscFree(submatj->row2proc)); 2835 2836 PetscCall(PetscFree(submatj)); 2837 PetscFunctionReturn(0); 2838 } 2839 2840 PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat C) 2841 { 2842 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data; 2843 Mat_SubSppt *submatj = c->submatis1; 2844 2845 PetscFunctionBegin; 2846 PetscCall((*submatj->destroy)(C)); 2847 PetscCall(MatDestroySubMatrix_Private(submatj)); 2848 PetscFunctionReturn(0); 2849 } 2850 2851 PetscErrorCode MatDestroySubMatrices_SeqAIJ(PetscInt n,Mat *mat[]) 2852 { 2853 PetscInt i; 2854 Mat C; 2855 Mat_SeqAIJ *c; 2856 Mat_SubSppt *submatj; 2857 2858 PetscFunctionBegin; 2859 for (i=0; i<n; i++) { 2860 C = (*mat)[i]; 2861 c = (Mat_SeqAIJ*)C->data; 2862 submatj = c->submatis1; 2863 if (submatj) { 2864 if (--((PetscObject)C)->refct <= 0) { 2865 PetscCall((*submatj->destroy)(C)); 2866 PetscCall(MatDestroySubMatrix_Private(submatj)); 2867 PetscCall(PetscFree(C->defaultvectype)); 2868 PetscCall(PetscLayoutDestroy(&C->rmap)); 2869 PetscCall(PetscLayoutDestroy(&C->cmap)); 2870 PetscCall(PetscHeaderDestroy(&C)); 2871 } 2872 } else { 2873 PetscCall(MatDestroy(&C)); 2874 } 2875 } 2876 2877 /* Destroy Dummy submatrices created for reuse */ 2878 PetscCall(MatDestroySubMatrices_Dummy(n,mat)); 2879 2880 PetscCall(PetscFree(*mat)); 2881 PetscFunctionReturn(0); 2882 } 2883 2884 PetscErrorCode MatCreateSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[]) 2885 { 2886 PetscInt i; 2887 2888 PetscFunctionBegin; 2889 if (scall == MAT_INITIAL_MATRIX) { 2890 PetscCall(PetscCalloc1(n+1,B)); 2891 } 2892 2893 for (i=0; i<n; i++) { 2894 PetscCall(MatCreateSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i])); 2895 } 2896 PetscFunctionReturn(0); 2897 } 2898 2899 PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov) 2900 { 2901 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2902 PetscInt row,i,j,k,l,m,n,*nidx,isz,val; 2903 const PetscInt *idx; 2904 PetscInt start,end,*ai,*aj; 2905 PetscBT table; 2906 2907 PetscFunctionBegin; 2908 m = A->rmap->n; 2909 ai = a->i; 2910 aj = a->j; 2911 2912 PetscCheck(ov >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used"); 2913 2914 PetscCall(PetscMalloc1(m+1,&nidx)); 2915 PetscCall(PetscBTCreate(m,&table)); 2916 2917 for (i=0; i<is_max; i++) { 2918 /* Initialize the two local arrays */ 2919 isz = 0; 2920 PetscCall(PetscBTMemzero(m,table)); 2921 2922 /* Extract the indices, assume there can be duplicate entries */ 2923 PetscCall(ISGetIndices(is[i],&idx)); 2924 PetscCall(ISGetLocalSize(is[i],&n)); 2925 2926 /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */ 2927 for (j=0; j<n; ++j) { 2928 if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j]; 2929 } 2930 PetscCall(ISRestoreIndices(is[i],&idx)); 2931 PetscCall(ISDestroy(&is[i])); 2932 2933 k = 0; 2934 for (j=0; j<ov; j++) { /* for each overlap */ 2935 n = isz; 2936 for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */ 2937 row = nidx[k]; 2938 start = ai[row]; 2939 end = ai[row+1]; 2940 for (l = start; l<end; l++) { 2941 val = aj[l]; 2942 if (!PetscBTLookupSet(table,val)) nidx[isz++] = val; 2943 } 2944 } 2945 } 2946 PetscCall(ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i))); 2947 } 2948 PetscCall(PetscBTDestroy(&table)); 2949 PetscCall(PetscFree(nidx)); 2950 PetscFunctionReturn(0); 2951 } 2952 2953 /* -------------------------------------------------------------- */ 2954 PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B) 2955 { 2956 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2957 PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n; 2958 const PetscInt *row,*col; 2959 PetscInt *cnew,j,*lens; 2960 IS icolp,irowp; 2961 PetscInt *cwork = NULL; 2962 PetscScalar *vwork = NULL; 2963 2964 PetscFunctionBegin; 2965 PetscCall(ISInvertPermutation(rowp,PETSC_DECIDE,&irowp)); 2966 PetscCall(ISGetIndices(irowp,&row)); 2967 PetscCall(ISInvertPermutation(colp,PETSC_DECIDE,&icolp)); 2968 PetscCall(ISGetIndices(icolp,&col)); 2969 2970 /* determine lengths of permuted rows */ 2971 PetscCall(PetscMalloc1(m+1,&lens)); 2972 for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i]; 2973 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),B)); 2974 PetscCall(MatSetSizes(*B,m,n,m,n)); 2975 PetscCall(MatSetBlockSizesFromMats(*B,A,A)); 2976 PetscCall(MatSetType(*B,((PetscObject)A)->type_name)); 2977 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens)); 2978 PetscCall(PetscFree(lens)); 2979 2980 PetscCall(PetscMalloc1(n,&cnew)); 2981 for (i=0; i<m; i++) { 2982 PetscCall(MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork)); 2983 for (j=0; j<nz; j++) cnew[j] = col[cwork[j]]; 2984 PetscCall(MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES)); 2985 PetscCall(MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork)); 2986 } 2987 PetscCall(PetscFree(cnew)); 2988 2989 (*B)->assembled = PETSC_FALSE; 2990 2991 #if defined(PETSC_HAVE_DEVICE) 2992 PetscCall(MatBindToCPU(*B,A->boundtocpu)); 2993 #endif 2994 PetscCall(MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY)); 2995 PetscCall(MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY)); 2996 PetscCall(ISRestoreIndices(irowp,&row)); 2997 PetscCall(ISRestoreIndices(icolp,&col)); 2998 PetscCall(ISDestroy(&irowp)); 2999 PetscCall(ISDestroy(&icolp)); 3000 if (rowp == colp) { 3001 PetscCall(MatPropagateSymmetryOptions(A,*B)); 3002 } 3003 PetscFunctionReturn(0); 3004 } 3005 3006 PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str) 3007 { 3008 PetscFunctionBegin; 3009 /* If the two matrices have the same copy implementation, use fast copy. */ 3010 if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) { 3011 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3012 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 3013 const PetscScalar *aa; 3014 3015 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 3016 PetscCheck(a->i[A->rmap->n] == b->i[B->rmap->n],PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzeros in two matrices are different %" PetscInt_FMT " != %" PetscInt_FMT,a->i[A->rmap->n],b->i[B->rmap->n]); 3017 PetscCall(PetscArraycpy(b->a,aa,a->i[A->rmap->n])); 3018 PetscCall(PetscObjectStateIncrease((PetscObject)B)); 3019 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 3020 } else { 3021 PetscCall(MatCopy_Basic(A,B,str)); 3022 } 3023 PetscFunctionReturn(0); 3024 } 3025 3026 PetscErrorCode MatSetUp_SeqAIJ(Mat A) 3027 { 3028 PetscFunctionBegin; 3029 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,NULL)); 3030 PetscFunctionReturn(0); 3031 } 3032 3033 PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[]) 3034 { 3035 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3036 3037 PetscFunctionBegin; 3038 *array = a->a; 3039 PetscFunctionReturn(0); 3040 } 3041 3042 PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[]) 3043 { 3044 PetscFunctionBegin; 3045 *array = NULL; 3046 PetscFunctionReturn(0); 3047 } 3048 3049 /* 3050 Computes the number of nonzeros per row needed for preallocation when X and Y 3051 have different nonzero structure. 3052 */ 3053 PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt m,const PetscInt *xi,const PetscInt *xj,const PetscInt *yi,const PetscInt *yj,PetscInt *nnz) 3054 { 3055 PetscInt i,j,k,nzx,nzy; 3056 3057 PetscFunctionBegin; 3058 /* Set the number of nonzeros in the new matrix */ 3059 for (i=0; i<m; i++) { 3060 const PetscInt *xjj = xj+xi[i],*yjj = yj+yi[i]; 3061 nzx = xi[i+1] - xi[i]; 3062 nzy = yi[i+1] - yi[i]; 3063 nnz[i] = 0; 3064 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 3065 for (; k<nzy && yjj[k]<xjj[j]; k++) nnz[i]++; /* Catch up to X */ 3066 if (k<nzy && yjj[k]==xjj[j]) k++; /* Skip duplicate */ 3067 nnz[i]++; 3068 } 3069 for (; k<nzy; k++) nnz[i]++; 3070 } 3071 PetscFunctionReturn(0); 3072 } 3073 3074 PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz) 3075 { 3076 PetscInt m = Y->rmap->N; 3077 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 3078 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 3079 3080 PetscFunctionBegin; 3081 /* Set the number of nonzeros in the new matrix */ 3082 PetscCall(MatAXPYGetPreallocation_SeqX_private(m,x->i,x->j,y->i,y->j,nnz)); 3083 PetscFunctionReturn(0); 3084 } 3085 3086 PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 3087 { 3088 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data; 3089 3090 PetscFunctionBegin; 3091 if (str == UNKNOWN_NONZERO_PATTERN || (PetscDefined(USE_DEBUG) && str == SAME_NONZERO_PATTERN)) { 3092 PetscBool e = x->nz == y->nz ? PETSC_TRUE : PETSC_FALSE; 3093 if (e) { 3094 PetscCall(PetscArraycmp(x->i,y->i,Y->rmap->n+1,&e)); 3095 if (e) { 3096 PetscCall(PetscArraycmp(x->j,y->j,y->nz,&e)); 3097 if (e) str = SAME_NONZERO_PATTERN; 3098 } 3099 } 3100 if (!e) PetscCheck(str != SAME_NONZERO_PATTERN,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"MatStructure is not SAME_NONZERO_PATTERN"); 3101 } 3102 if (str == SAME_NONZERO_PATTERN) { 3103 const PetscScalar *xa; 3104 PetscScalar *ya,alpha = a; 3105 PetscBLASInt one = 1,bnz; 3106 3107 PetscCall(PetscBLASIntCast(x->nz,&bnz)); 3108 PetscCall(MatSeqAIJGetArray(Y,&ya)); 3109 PetscCall(MatSeqAIJGetArrayRead(X,&xa)); 3110 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,xa,&one,ya,&one)); 3111 PetscCall(MatSeqAIJRestoreArrayRead(X,&xa)); 3112 PetscCall(MatSeqAIJRestoreArray(Y,&ya)); 3113 PetscCall(PetscLogFlops(2.0*bnz)); 3114 PetscCall(MatSeqAIJInvalidateDiagonal(Y)); 3115 PetscCall(PetscObjectStateIncrease((PetscObject)Y)); 3116 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 3117 PetscCall(MatAXPY_Basic(Y,a,X,str)); 3118 } else { 3119 Mat B; 3120 PetscInt *nnz; 3121 PetscCall(PetscMalloc1(Y->rmap->N,&nnz)); 3122 PetscCall(MatCreate(PetscObjectComm((PetscObject)Y),&B)); 3123 PetscCall(PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name)); 3124 PetscCall(MatSetLayouts(B,Y->rmap,Y->cmap)); 3125 PetscCall(MatSetType(B,((PetscObject)Y)->type_name)); 3126 PetscCall(MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz)); 3127 PetscCall(MatSeqAIJSetPreallocation(B,0,nnz)); 3128 PetscCall(MatAXPY_BasicWithPreallocation(B,Y,a,X,str)); 3129 PetscCall(MatHeaderMerge(Y,&B)); 3130 PetscCall(PetscFree(nnz)); 3131 } 3132 PetscFunctionReturn(0); 3133 } 3134 3135 PETSC_INTERN PetscErrorCode MatConjugate_SeqAIJ(Mat mat) 3136 { 3137 #if defined(PETSC_USE_COMPLEX) 3138 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3139 PetscInt i,nz; 3140 PetscScalar *a; 3141 3142 PetscFunctionBegin; 3143 nz = aij->nz; 3144 PetscCall(MatSeqAIJGetArray(mat,&a)); 3145 for (i=0; i<nz; i++) a[i] = PetscConj(a[i]); 3146 PetscCall(MatSeqAIJRestoreArray(mat,&a)); 3147 #else 3148 PetscFunctionBegin; 3149 #endif 3150 PetscFunctionReturn(0); 3151 } 3152 3153 PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3154 { 3155 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3156 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3157 PetscReal atmp; 3158 PetscScalar *x; 3159 const MatScalar *aa,*av; 3160 3161 PetscFunctionBegin; 3162 PetscCheck(!A->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3163 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 3164 aa = av; 3165 ai = a->i; 3166 aj = a->j; 3167 3168 PetscCall(VecSet(v,0.0)); 3169 PetscCall(VecGetArrayWrite(v,&x)); 3170 PetscCall(VecGetLocalSize(v,&n)); 3171 PetscCheck(n == A->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3172 for (i=0; i<m; i++) { 3173 ncols = ai[1] - ai[0]; ai++; 3174 for (j=0; j<ncols; j++) { 3175 atmp = PetscAbsScalar(*aa); 3176 if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3177 aa++; aj++; 3178 } 3179 } 3180 PetscCall(VecRestoreArrayWrite(v,&x)); 3181 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 3182 PetscFunctionReturn(0); 3183 } 3184 3185 PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3186 { 3187 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3188 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3189 PetscScalar *x; 3190 const MatScalar *aa,*av; 3191 3192 PetscFunctionBegin; 3193 PetscCheck(!A->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3194 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 3195 aa = av; 3196 ai = a->i; 3197 aj = a->j; 3198 3199 PetscCall(VecSet(v,0.0)); 3200 PetscCall(VecGetArrayWrite(v,&x)); 3201 PetscCall(VecGetLocalSize(v,&n)); 3202 PetscCheck(n == A->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3203 for (i=0; i<m; i++) { 3204 ncols = ai[1] - ai[0]; ai++; 3205 if (ncols == A->cmap->n) { /* row is dense */ 3206 x[i] = *aa; if (idx) idx[i] = 0; 3207 } else { /* row is sparse so already KNOW maximum is 0.0 or higher */ 3208 x[i] = 0.0; 3209 if (idx) { 3210 for (j=0; j<ncols; j++) { /* find first implicit 0.0 in the row */ 3211 if (aj[j] > j) { 3212 idx[i] = j; 3213 break; 3214 } 3215 } 3216 /* in case first implicit 0.0 in the row occurs at ncols-th column */ 3217 if (j==ncols && j < A->cmap->n) idx[i] = j; 3218 } 3219 } 3220 for (j=0; j<ncols; j++) { 3221 if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3222 aa++; aj++; 3223 } 3224 } 3225 PetscCall(VecRestoreArrayWrite(v,&x)); 3226 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 3227 PetscFunctionReturn(0); 3228 } 3229 3230 PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3231 { 3232 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3233 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3234 PetscScalar *x; 3235 const MatScalar *aa,*av; 3236 3237 PetscFunctionBegin; 3238 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 3239 aa = av; 3240 ai = a->i; 3241 aj = a->j; 3242 3243 PetscCall(VecSet(v,0.0)); 3244 PetscCall(VecGetArrayWrite(v,&x)); 3245 PetscCall(VecGetLocalSize(v,&n)); 3246 PetscCheck(n == m,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector, %" PetscInt_FMT " vs. %" PetscInt_FMT " rows", m, n); 3247 for (i=0; i<m; i++) { 3248 ncols = ai[1] - ai[0]; ai++; 3249 if (ncols == A->cmap->n) { /* row is dense */ 3250 x[i] = *aa; if (idx) idx[i] = 0; 3251 } else { /* row is sparse so already KNOW minimum is 0.0 or higher */ 3252 x[i] = 0.0; 3253 if (idx) { /* find first implicit 0.0 in the row */ 3254 for (j=0; j<ncols; j++) { 3255 if (aj[j] > j) { 3256 idx[i] = j; 3257 break; 3258 } 3259 } 3260 /* in case first implicit 0.0 in the row occurs at ncols-th column */ 3261 if (j==ncols && j < A->cmap->n) idx[i] = j; 3262 } 3263 } 3264 for (j=0; j<ncols; j++) { 3265 if (PetscAbsScalar(x[i]) > PetscAbsScalar(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3266 aa++; aj++; 3267 } 3268 } 3269 PetscCall(VecRestoreArrayWrite(v,&x)); 3270 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 3271 PetscFunctionReturn(0); 3272 } 3273 3274 PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3275 { 3276 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3277 PetscInt i,j,m = A->rmap->n,ncols,n; 3278 const PetscInt *ai,*aj; 3279 PetscScalar *x; 3280 const MatScalar *aa,*av; 3281 3282 PetscFunctionBegin; 3283 PetscCheck(!A->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3284 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 3285 aa = av; 3286 ai = a->i; 3287 aj = a->j; 3288 3289 PetscCall(VecSet(v,0.0)); 3290 PetscCall(VecGetArrayWrite(v,&x)); 3291 PetscCall(VecGetLocalSize(v,&n)); 3292 PetscCheck(n == m,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3293 for (i=0; i<m; i++) { 3294 ncols = ai[1] - ai[0]; ai++; 3295 if (ncols == A->cmap->n) { /* row is dense */ 3296 x[i] = *aa; if (idx) idx[i] = 0; 3297 } else { /* row is sparse so already KNOW minimum is 0.0 or lower */ 3298 x[i] = 0.0; 3299 if (idx) { /* find first implicit 0.0 in the row */ 3300 for (j=0; j<ncols; j++) { 3301 if (aj[j] > j) { 3302 idx[i] = j; 3303 break; 3304 } 3305 } 3306 /* in case first implicit 0.0 in the row occurs at ncols-th column */ 3307 if (j==ncols && j < A->cmap->n) idx[i] = j; 3308 } 3309 } 3310 for (j=0; j<ncols; j++) { 3311 if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3312 aa++; aj++; 3313 } 3314 } 3315 PetscCall(VecRestoreArrayWrite(v,&x)); 3316 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 3317 PetscFunctionReturn(0); 3318 } 3319 3320 PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values) 3321 { 3322 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 3323 PetscInt i,bs = PetscAbs(A->rmap->bs),mbs = A->rmap->n/bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j; 3324 MatScalar *diag,work[25],*v_work; 3325 const PetscReal shift = 0.0; 3326 PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE; 3327 3328 PetscFunctionBegin; 3329 allowzeropivot = PetscNot(A->erroriffailure); 3330 if (a->ibdiagvalid) { 3331 if (values) *values = a->ibdiag; 3332 PetscFunctionReturn(0); 3333 } 3334 PetscCall(MatMarkDiagonal_SeqAIJ(A)); 3335 if (!a->ibdiag) { 3336 PetscCall(PetscMalloc1(bs2*mbs,&a->ibdiag)); 3337 PetscCall(PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar))); 3338 } 3339 diag = a->ibdiag; 3340 if (values) *values = a->ibdiag; 3341 /* factor and invert each block */ 3342 switch (bs) { 3343 case 1: 3344 for (i=0; i<mbs; i++) { 3345 PetscCall(MatGetValues(A,1,&i,1,&i,diag+i)); 3346 if (PetscAbsScalar(diag[i] + shift) < PETSC_MACHINE_EPSILON) { 3347 if (allowzeropivot) { 3348 A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3349 A->factorerror_zeropivot_value = PetscAbsScalar(diag[i]); 3350 A->factorerror_zeropivot_row = i; 3351 PetscCall(PetscInfo(A,"Zero pivot, row %" PetscInt_FMT " pivot %g tolerance %g\n",i,(double)PetscAbsScalar(diag[i]),(double)PETSC_MACHINE_EPSILON)); 3352 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot, row %" PetscInt_FMT " pivot %g tolerance %g",i,(double)PetscAbsScalar(diag[i]),(double)PETSC_MACHINE_EPSILON); 3353 } 3354 diag[i] = (PetscScalar)1.0 / (diag[i] + shift); 3355 } 3356 break; 3357 case 2: 3358 for (i=0; i<mbs; i++) { 3359 ij[0] = 2*i; ij[1] = 2*i + 1; 3360 PetscCall(MatGetValues(A,2,ij,2,ij,diag)); 3361 PetscCall(PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected)); 3362 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3363 PetscCall(PetscKernel_A_gets_transpose_A_2(diag)); 3364 diag += 4; 3365 } 3366 break; 3367 case 3: 3368 for (i=0; i<mbs; i++) { 3369 ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2; 3370 PetscCall(MatGetValues(A,3,ij,3,ij,diag)); 3371 PetscCall(PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected)); 3372 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3373 PetscCall(PetscKernel_A_gets_transpose_A_3(diag)); 3374 diag += 9; 3375 } 3376 break; 3377 case 4: 3378 for (i=0; i<mbs; i++) { 3379 ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3; 3380 PetscCall(MatGetValues(A,4,ij,4,ij,diag)); 3381 PetscCall(PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected)); 3382 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3383 PetscCall(PetscKernel_A_gets_transpose_A_4(diag)); 3384 diag += 16; 3385 } 3386 break; 3387 case 5: 3388 for (i=0; i<mbs; i++) { 3389 ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4; 3390 PetscCall(MatGetValues(A,5,ij,5,ij,diag)); 3391 PetscCall(PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected)); 3392 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3393 PetscCall(PetscKernel_A_gets_transpose_A_5(diag)); 3394 diag += 25; 3395 } 3396 break; 3397 case 6: 3398 for (i=0; i<mbs; i++) { 3399 ij[0] = 6*i; ij[1] = 6*i + 1; ij[2] = 6*i + 2; ij[3] = 6*i + 3; ij[4] = 6*i + 4; ij[5] = 6*i + 5; 3400 PetscCall(MatGetValues(A,6,ij,6,ij,diag)); 3401 PetscCall(PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected)); 3402 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3403 PetscCall(PetscKernel_A_gets_transpose_A_6(diag)); 3404 diag += 36; 3405 } 3406 break; 3407 case 7: 3408 for (i=0; i<mbs; i++) { 3409 ij[0] = 7*i; ij[1] = 7*i + 1; ij[2] = 7*i + 2; ij[3] = 7*i + 3; ij[4] = 7*i + 4; ij[5] = 7*i + 5; ij[5] = 7*i + 6; 3410 PetscCall(MatGetValues(A,7,ij,7,ij,diag)); 3411 PetscCall(PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected)); 3412 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3413 PetscCall(PetscKernel_A_gets_transpose_A_7(diag)); 3414 diag += 49; 3415 } 3416 break; 3417 default: 3418 PetscCall(PetscMalloc3(bs,&v_work,bs,&v_pivots,bs,&IJ)); 3419 for (i=0; i<mbs; i++) { 3420 for (j=0; j<bs; j++) { 3421 IJ[j] = bs*i + j; 3422 } 3423 PetscCall(MatGetValues(A,bs,IJ,bs,IJ,diag)); 3424 PetscCall(PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected)); 3425 if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 3426 PetscCall(PetscKernel_A_gets_transpose_A_N(diag,bs)); 3427 diag += bs2; 3428 } 3429 PetscCall(PetscFree3(v_work,v_pivots,IJ)); 3430 } 3431 a->ibdiagvalid = PETSC_TRUE; 3432 PetscFunctionReturn(0); 3433 } 3434 3435 static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx) 3436 { 3437 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data; 3438 PetscScalar a,*aa; 3439 PetscInt m,n,i,j,col; 3440 3441 PetscFunctionBegin; 3442 if (!x->assembled) { 3443 PetscCall(MatGetSize(x,&m,&n)); 3444 for (i=0; i<m; i++) { 3445 for (j=0; j<aij->imax[i]; j++) { 3446 PetscCall(PetscRandomGetValue(rctx,&a)); 3447 col = (PetscInt)(n*PetscRealPart(a)); 3448 PetscCall(MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES)); 3449 } 3450 } 3451 } else { 3452 PetscCall(MatSeqAIJGetArrayWrite(x,&aa)); 3453 for (i=0; i<aij->nz; i++) PetscCall(PetscRandomGetValue(rctx,aa+i)); 3454 PetscCall(MatSeqAIJRestoreArrayWrite(x,&aa)); 3455 } 3456 PetscCall(MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY)); 3457 PetscCall(MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY)); 3458 PetscFunctionReturn(0); 3459 } 3460 3461 /* Like MatSetRandom_SeqAIJ, but do not set values on columns in range of [low, high) */ 3462 PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat x,PetscInt low,PetscInt high,PetscRandom rctx) 3463 { 3464 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data; 3465 PetscScalar a; 3466 PetscInt m,n,i,j,col,nskip; 3467 3468 PetscFunctionBegin; 3469 nskip = high - low; 3470 PetscCall(MatGetSize(x,&m,&n)); 3471 n -= nskip; /* shrink number of columns where nonzeros can be set */ 3472 for (i=0; i<m; i++) { 3473 for (j=0; j<aij->imax[i]; j++) { 3474 PetscCall(PetscRandomGetValue(rctx,&a)); 3475 col = (PetscInt)(n*PetscRealPart(a)); 3476 if (col >= low) col += nskip; /* shift col rightward to skip the hole */ 3477 PetscCall(MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES)); 3478 } 3479 } 3480 PetscCall(MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY)); 3481 PetscCall(MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY)); 3482 PetscFunctionReturn(0); 3483 } 3484 3485 /* -------------------------------------------------------------------*/ 3486 static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ, 3487 MatGetRow_SeqAIJ, 3488 MatRestoreRow_SeqAIJ, 3489 MatMult_SeqAIJ, 3490 /* 4*/ MatMultAdd_SeqAIJ, 3491 MatMultTranspose_SeqAIJ, 3492 MatMultTransposeAdd_SeqAIJ, 3493 NULL, 3494 NULL, 3495 NULL, 3496 /* 10*/ NULL, 3497 MatLUFactor_SeqAIJ, 3498 NULL, 3499 MatSOR_SeqAIJ, 3500 MatTranspose_SeqAIJ, 3501 /*1 5*/ MatGetInfo_SeqAIJ, 3502 MatEqual_SeqAIJ, 3503 MatGetDiagonal_SeqAIJ, 3504 MatDiagonalScale_SeqAIJ, 3505 MatNorm_SeqAIJ, 3506 /* 20*/ NULL, 3507 MatAssemblyEnd_SeqAIJ, 3508 MatSetOption_SeqAIJ, 3509 MatZeroEntries_SeqAIJ, 3510 /* 24*/ MatZeroRows_SeqAIJ, 3511 NULL, 3512 NULL, 3513 NULL, 3514 NULL, 3515 /* 29*/ MatSetUp_SeqAIJ, 3516 NULL, 3517 NULL, 3518 NULL, 3519 NULL, 3520 /* 34*/ MatDuplicate_SeqAIJ, 3521 NULL, 3522 NULL, 3523 MatILUFactor_SeqAIJ, 3524 NULL, 3525 /* 39*/ MatAXPY_SeqAIJ, 3526 MatCreateSubMatrices_SeqAIJ, 3527 MatIncreaseOverlap_SeqAIJ, 3528 MatGetValues_SeqAIJ, 3529 MatCopy_SeqAIJ, 3530 /* 44*/ MatGetRowMax_SeqAIJ, 3531 MatScale_SeqAIJ, 3532 MatShift_SeqAIJ, 3533 MatDiagonalSet_SeqAIJ, 3534 MatZeroRowsColumns_SeqAIJ, 3535 /* 49*/ MatSetRandom_SeqAIJ, 3536 MatGetRowIJ_SeqAIJ, 3537 MatRestoreRowIJ_SeqAIJ, 3538 MatGetColumnIJ_SeqAIJ, 3539 MatRestoreColumnIJ_SeqAIJ, 3540 /* 54*/ MatFDColoringCreate_SeqXAIJ, 3541 NULL, 3542 NULL, 3543 MatPermute_SeqAIJ, 3544 NULL, 3545 /* 59*/ NULL, 3546 MatDestroy_SeqAIJ, 3547 MatView_SeqAIJ, 3548 NULL, 3549 NULL, 3550 /* 64*/ NULL, 3551 MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ, 3552 NULL, 3553 NULL, 3554 NULL, 3555 /* 69*/ MatGetRowMaxAbs_SeqAIJ, 3556 MatGetRowMinAbs_SeqAIJ, 3557 NULL, 3558 NULL, 3559 NULL, 3560 /* 74*/ NULL, 3561 MatFDColoringApply_AIJ, 3562 NULL, 3563 NULL, 3564 NULL, 3565 /* 79*/ MatFindZeroDiagonals_SeqAIJ, 3566 NULL, 3567 NULL, 3568 NULL, 3569 MatLoad_SeqAIJ, 3570 /* 84*/ MatIsSymmetric_SeqAIJ, 3571 MatIsHermitian_SeqAIJ, 3572 NULL, 3573 NULL, 3574 NULL, 3575 /* 89*/ NULL, 3576 NULL, 3577 MatMatMultNumeric_SeqAIJ_SeqAIJ, 3578 NULL, 3579 NULL, 3580 /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy, 3581 NULL, 3582 NULL, 3583 MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ, 3584 NULL, 3585 /* 99*/ MatProductSetFromOptions_SeqAIJ, 3586 NULL, 3587 NULL, 3588 MatConjugate_SeqAIJ, 3589 NULL, 3590 /*104*/ MatSetValuesRow_SeqAIJ, 3591 MatRealPart_SeqAIJ, 3592 MatImaginaryPart_SeqAIJ, 3593 NULL, 3594 NULL, 3595 /*109*/ MatMatSolve_SeqAIJ, 3596 NULL, 3597 MatGetRowMin_SeqAIJ, 3598 NULL, 3599 MatMissingDiagonal_SeqAIJ, 3600 /*114*/ NULL, 3601 NULL, 3602 NULL, 3603 NULL, 3604 NULL, 3605 /*119*/ NULL, 3606 NULL, 3607 NULL, 3608 NULL, 3609 MatGetMultiProcBlock_SeqAIJ, 3610 /*124*/ MatFindNonzeroRows_SeqAIJ, 3611 MatGetColumnReductions_SeqAIJ, 3612 MatInvertBlockDiagonal_SeqAIJ, 3613 MatInvertVariableBlockDiagonal_SeqAIJ, 3614 NULL, 3615 /*129*/ NULL, 3616 NULL, 3617 NULL, 3618 MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ, 3619 MatTransposeColoringCreate_SeqAIJ, 3620 /*134*/ MatTransColoringApplySpToDen_SeqAIJ, 3621 MatTransColoringApplyDenToSp_SeqAIJ, 3622 NULL, 3623 NULL, 3624 MatRARtNumeric_SeqAIJ_SeqAIJ, 3625 /*139*/NULL, 3626 NULL, 3627 NULL, 3628 MatFDColoringSetUp_SeqXAIJ, 3629 MatFindOffBlockDiagonalEntries_SeqAIJ, 3630 MatCreateMPIMatConcatenateSeqMat_SeqAIJ, 3631 /*145*/MatDestroySubMatrices_SeqAIJ, 3632 NULL, 3633 NULL 3634 }; 3635 3636 PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices) 3637 { 3638 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3639 PetscInt i,nz,n; 3640 3641 PetscFunctionBegin; 3642 nz = aij->maxnz; 3643 n = mat->rmap->n; 3644 for (i=0; i<nz; i++) { 3645 aij->j[i] = indices[i]; 3646 } 3647 aij->nz = nz; 3648 for (i=0; i<n; i++) { 3649 aij->ilen[i] = aij->imax[i]; 3650 } 3651 PetscFunctionReturn(0); 3652 } 3653 3654 /* 3655 * Given a sparse matrix with global column indices, compact it by using a local column space. 3656 * The result matrix helps saving memory in other algorithms, such as MatPtAPSymbolic_MPIAIJ_MPIAIJ_scalable() 3657 */ 3658 PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat mat, ISLocalToGlobalMapping *mapping) 3659 { 3660 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3661 PetscTable gid1_lid1; 3662 PetscTablePosition tpos; 3663 PetscInt gid,lid,i,ec,nz = aij->nz; 3664 PetscInt *garray,*jj = aij->j; 3665 3666 PetscFunctionBegin; 3667 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3668 PetscValidPointer(mapping,2); 3669 /* use a table */ 3670 PetscCall(PetscTableCreate(mat->rmap->n,mat->cmap->N+1,&gid1_lid1)); 3671 ec = 0; 3672 for (i=0; i<nz; i++) { 3673 PetscInt data,gid1 = jj[i] + 1; 3674 PetscCall(PetscTableFind(gid1_lid1,gid1,&data)); 3675 if (!data) { 3676 /* one based table */ 3677 PetscCall(PetscTableAdd(gid1_lid1,gid1,++ec,INSERT_VALUES)); 3678 } 3679 } 3680 /* form array of columns we need */ 3681 PetscCall(PetscMalloc1(ec,&garray)); 3682 PetscCall(PetscTableGetHeadPosition(gid1_lid1,&tpos)); 3683 while (tpos) { 3684 PetscCall(PetscTableGetNext(gid1_lid1,&tpos,&gid,&lid)); 3685 gid--; 3686 lid--; 3687 garray[lid] = gid; 3688 } 3689 PetscCall(PetscSortInt(ec,garray)); /* sort, and rebuild */ 3690 PetscCall(PetscTableRemoveAll(gid1_lid1)); 3691 for (i=0; i<ec; i++) { 3692 PetscCall(PetscTableAdd(gid1_lid1,garray[i]+1,i+1,INSERT_VALUES)); 3693 } 3694 /* compact out the extra columns in B */ 3695 for (i=0; i<nz; i++) { 3696 PetscInt gid1 = jj[i] + 1; 3697 PetscCall(PetscTableFind(gid1_lid1,gid1,&lid)); 3698 lid--; 3699 jj[i] = lid; 3700 } 3701 PetscCall(PetscLayoutDestroy(&mat->cmap)); 3702 PetscCall(PetscTableDestroy(&gid1_lid1)); 3703 PetscCall(PetscLayoutCreateFromSizes(PetscObjectComm((PetscObject)mat),ec,ec,1,&mat->cmap)); 3704 PetscCall(ISLocalToGlobalMappingCreate(PETSC_COMM_SELF,mat->cmap->bs,mat->cmap->n,garray,PETSC_OWN_POINTER,mapping)); 3705 PetscCall(ISLocalToGlobalMappingSetType(*mapping,ISLOCALTOGLOBALMAPPINGHASH)); 3706 PetscFunctionReturn(0); 3707 } 3708 3709 /*@ 3710 MatSeqAIJSetColumnIndices - Set the column indices for all the rows 3711 in the matrix. 3712 3713 Input Parameters: 3714 + mat - the SeqAIJ matrix 3715 - indices - the column indices 3716 3717 Level: advanced 3718 3719 Notes: 3720 This can be called if you have precomputed the nonzero structure of the 3721 matrix and want to provide it to the matrix object to improve the performance 3722 of the MatSetValues() operation. 3723 3724 You MUST have set the correct numbers of nonzeros per row in the call to 3725 MatCreateSeqAIJ(), and the columns indices MUST be sorted. 3726 3727 MUST be called before any calls to MatSetValues(); 3728 3729 The indices should start with zero, not one. 3730 3731 @*/ 3732 PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices) 3733 { 3734 PetscFunctionBegin; 3735 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3736 PetscValidIntPointer(indices,2); 3737 PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices)); 3738 PetscFunctionReturn(0); 3739 } 3740 3741 /* ----------------------------------------------------------------------------------------*/ 3742 3743 PetscErrorCode MatStoreValues_SeqAIJ(Mat mat) 3744 { 3745 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3746 size_t nz = aij->i[mat->rmap->n]; 3747 3748 PetscFunctionBegin; 3749 PetscCheck(aij->nonew,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3750 3751 /* allocate space for values if not already there */ 3752 if (!aij->saved_values) { 3753 PetscCall(PetscMalloc1(nz+1,&aij->saved_values)); 3754 PetscCall(PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar))); 3755 } 3756 3757 /* copy values over */ 3758 PetscCall(PetscArraycpy(aij->saved_values,aij->a,nz)); 3759 PetscFunctionReturn(0); 3760 } 3761 3762 /*@ 3763 MatStoreValues - Stashes a copy of the matrix values; this allows, for 3764 example, reuse of the linear part of a Jacobian, while recomputing the 3765 nonlinear portion. 3766 3767 Collect on Mat 3768 3769 Input Parameters: 3770 . mat - the matrix (currently only AIJ matrices support this option) 3771 3772 Level: advanced 3773 3774 Common Usage, with SNESSolve(): 3775 $ Create Jacobian matrix 3776 $ Set linear terms into matrix 3777 $ Apply boundary conditions to matrix, at this time matrix must have 3778 $ final nonzero structure (i.e. setting the nonlinear terms and applying 3779 $ boundary conditions again will not change the nonzero structure 3780 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3781 $ ierr = MatStoreValues(mat); 3782 $ Call SNESSetJacobian() with matrix 3783 $ In your Jacobian routine 3784 $ ierr = MatRetrieveValues(mat); 3785 $ Set nonlinear terms in matrix 3786 3787 Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself: 3788 $ // build linear portion of Jacobian 3789 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3790 $ ierr = MatStoreValues(mat); 3791 $ loop over nonlinear iterations 3792 $ ierr = MatRetrieveValues(mat); 3793 $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian 3794 $ // call MatAssemblyBegin/End() on matrix 3795 $ Solve linear system with Jacobian 3796 $ endloop 3797 3798 Notes: 3799 Matrix must already be assemblied before calling this routine 3800 Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before 3801 calling this routine. 3802 3803 When this is called multiple times it overwrites the previous set of stored values 3804 and does not allocated additional space. 3805 3806 .seealso: `MatRetrieveValues()` 3807 3808 @*/ 3809 PetscErrorCode MatStoreValues(Mat mat) 3810 { 3811 PetscFunctionBegin; 3812 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3813 PetscCheck(mat->assembled,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3814 PetscCheck(!mat->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3815 PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat)); 3816 PetscFunctionReturn(0); 3817 } 3818 3819 PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat) 3820 { 3821 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3822 PetscInt nz = aij->i[mat->rmap->n]; 3823 3824 PetscFunctionBegin; 3825 PetscCheck(aij->nonew,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3826 PetscCheck(aij->saved_values,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first"); 3827 /* copy values over */ 3828 PetscCall(PetscArraycpy(aij->a,aij->saved_values,nz)); 3829 PetscFunctionReturn(0); 3830 } 3831 3832 /*@ 3833 MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for 3834 example, reuse of the linear part of a Jacobian, while recomputing the 3835 nonlinear portion. 3836 3837 Collect on Mat 3838 3839 Input Parameters: 3840 . mat - the matrix (currently only AIJ matrices support this option) 3841 3842 Level: advanced 3843 3844 .seealso: `MatStoreValues()` 3845 3846 @*/ 3847 PetscErrorCode MatRetrieveValues(Mat mat) 3848 { 3849 PetscFunctionBegin; 3850 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3851 PetscCheck(mat->assembled,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3852 PetscCheck(!mat->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3853 PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat)); 3854 PetscFunctionReturn(0); 3855 } 3856 3857 /* --------------------------------------------------------------------------------*/ 3858 /*@C 3859 MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format 3860 (the default parallel PETSc format). For good matrix assembly performance 3861 the user should preallocate the matrix storage by setting the parameter nz 3862 (or the array nnz). By setting these parameters accurately, performance 3863 during matrix assembly can be increased by more than a factor of 50. 3864 3865 Collective 3866 3867 Input Parameters: 3868 + comm - MPI communicator, set to PETSC_COMM_SELF 3869 . m - number of rows 3870 . n - number of columns 3871 . nz - number of nonzeros per row (same for all rows) 3872 - nnz - array containing the number of nonzeros in the various rows 3873 (possibly different for each row) or NULL 3874 3875 Output Parameter: 3876 . A - the matrix 3877 3878 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3879 MatXXXXSetPreallocation() paradigm instead of this routine directly. 3880 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3881 3882 Notes: 3883 If nnz is given then nz is ignored 3884 3885 The AIJ format (also called the Yale sparse matrix format or 3886 compressed row storage), is fully compatible with standard Fortran 77 3887 storage. That is, the stored row and column indices can begin at 3888 either one (as in Fortran) or zero. See the users' manual for details. 3889 3890 Specify the preallocated storage with either nz or nnz (not both). 3891 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3892 allocation. For large problems you MUST preallocate memory or you 3893 will get TERRIBLE performance, see the users' manual chapter on matrices. 3894 3895 By default, this format uses inodes (identical nodes) when possible, to 3896 improve numerical efficiency of matrix-vector products and solves. We 3897 search for consecutive rows with the same nonzero structure, thereby 3898 reusing matrix information to achieve increased efficiency. 3899 3900 Options Database Keys: 3901 + -mat_no_inode - Do not use inodes 3902 - -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3903 3904 Level: intermediate 3905 3906 .seealso: `MatCreate()`, `MatCreateAIJ()`, `MatSetValues()`, `MatSeqAIJSetColumnIndices()`, `MatCreateSeqAIJWithArrays()` 3907 3908 @*/ 3909 PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A) 3910 { 3911 PetscFunctionBegin; 3912 PetscCall(MatCreate(comm,A)); 3913 PetscCall(MatSetSizes(*A,m,n,m,n)); 3914 PetscCall(MatSetType(*A,MATSEQAIJ)); 3915 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz)); 3916 PetscFunctionReturn(0); 3917 } 3918 3919 /*@C 3920 MatSeqAIJSetPreallocation - For good matrix assembly performance 3921 the user should preallocate the matrix storage by setting the parameter nz 3922 (or the array nnz). By setting these parameters accurately, performance 3923 during matrix assembly can be increased by more than a factor of 50. 3924 3925 Collective 3926 3927 Input Parameters: 3928 + B - The matrix 3929 . nz - number of nonzeros per row (same for all rows) 3930 - nnz - array containing the number of nonzeros in the various rows 3931 (possibly different for each row) or NULL 3932 3933 Notes: 3934 If nnz is given then nz is ignored 3935 3936 The AIJ format (also called the Yale sparse matrix format or 3937 compressed row storage), is fully compatible with standard Fortran 77 3938 storage. That is, the stored row and column indices can begin at 3939 either one (as in Fortran) or zero. See the users' manual for details. 3940 3941 Specify the preallocated storage with either nz or nnz (not both). 3942 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3943 allocation. For large problems you MUST preallocate memory or you 3944 will get TERRIBLE performance, see the users' manual chapter on matrices. 3945 3946 You can call MatGetInfo() to get information on how effective the preallocation was; 3947 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3948 You can also run with the option -info and look for messages with the string 3949 malloc in them to see if additional memory allocation was needed. 3950 3951 Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix 3952 entries or columns indices 3953 3954 By default, this format uses inodes (identical nodes) when possible, to 3955 improve numerical efficiency of matrix-vector products and solves. We 3956 search for consecutive rows with the same nonzero structure, thereby 3957 reusing matrix information to achieve increased efficiency. 3958 3959 Options Database Keys: 3960 + -mat_no_inode - Do not use inodes 3961 - -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3962 3963 Level: intermediate 3964 3965 .seealso: `MatCreate()`, `MatCreateAIJ()`, `MatSetValues()`, `MatSeqAIJSetColumnIndices()`, `MatCreateSeqAIJWithArrays()`, `MatGetInfo()`, 3966 `MatSeqAIJSetTotalPreallocation()` 3967 3968 @*/ 3969 PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[]) 3970 { 3971 PetscFunctionBegin; 3972 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3973 PetscValidType(B,1); 3974 PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz)); 3975 PetscFunctionReturn(0); 3976 } 3977 3978 PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz) 3979 { 3980 Mat_SeqAIJ *b; 3981 PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE; 3982 PetscInt i; 3983 3984 PetscFunctionBegin; 3985 if (nz >= 0 || nnz) realalloc = PETSC_TRUE; 3986 if (nz == MAT_SKIP_ALLOCATION) { 3987 skipallocation = PETSC_TRUE; 3988 nz = 0; 3989 } 3990 PetscCall(PetscLayoutSetUp(B->rmap)); 3991 PetscCall(PetscLayoutSetUp(B->cmap)); 3992 3993 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3994 PetscCheck(nz >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %" PetscInt_FMT,nz); 3995 if (PetscUnlikelyDebug(nnz)) { 3996 for (i=0; i<B->rmap->n; i++) { 3997 PetscCheck(nnz[i] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %" PetscInt_FMT " value %" PetscInt_FMT,i,nnz[i]); 3998 PetscCheck(nnz[i] <= B->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than row length: local row %" PetscInt_FMT " value %" PetscInt_FMT " rowlength %" PetscInt_FMT,i,nnz[i],B->cmap->n); 3999 } 4000 } 4001 4002 B->preallocated = PETSC_TRUE; 4003 4004 b = (Mat_SeqAIJ*)B->data; 4005 4006 if (!skipallocation) { 4007 if (!b->imax) { 4008 PetscCall(PetscMalloc1(B->rmap->n,&b->imax)); 4009 PetscCall(PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt))); 4010 } 4011 if (!b->ilen) { 4012 /* b->ilen will count nonzeros in each row so far. */ 4013 PetscCall(PetscCalloc1(B->rmap->n,&b->ilen)); 4014 PetscCall(PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt))); 4015 } else { 4016 PetscCall(PetscMemzero(b->ilen,B->rmap->n*sizeof(PetscInt))); 4017 } 4018 if (!b->ipre) { 4019 PetscCall(PetscMalloc1(B->rmap->n,&b->ipre)); 4020 PetscCall(PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt))); 4021 } 4022 if (!nnz) { 4023 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10; 4024 else if (nz < 0) nz = 1; 4025 nz = PetscMin(nz,B->cmap->n); 4026 for (i=0; i<B->rmap->n; i++) b->imax[i] = nz; 4027 nz = nz*B->rmap->n; 4028 } else { 4029 PetscInt64 nz64 = 0; 4030 for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz64 += nnz[i];} 4031 PetscCall(PetscIntCast(nz64,&nz)); 4032 } 4033 4034 /* allocate the matrix space */ 4035 /* FIXME: should B's old memory be unlogged? */ 4036 PetscCall(MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i)); 4037 if (B->structure_only) { 4038 PetscCall(PetscMalloc1(nz,&b->j)); 4039 PetscCall(PetscMalloc1(B->rmap->n+1,&b->i)); 4040 PetscCall(PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*sizeof(PetscInt))); 4041 } else { 4042 PetscCall(PetscMalloc3(nz,&b->a,nz,&b->j,B->rmap->n+1,&b->i)); 4043 PetscCall(PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)))); 4044 } 4045 b->i[0] = 0; 4046 for (i=1; i<B->rmap->n+1; i++) { 4047 b->i[i] = b->i[i-1] + b->imax[i-1]; 4048 } 4049 if (B->structure_only) { 4050 b->singlemalloc = PETSC_FALSE; 4051 b->free_a = PETSC_FALSE; 4052 } else { 4053 b->singlemalloc = PETSC_TRUE; 4054 b->free_a = PETSC_TRUE; 4055 } 4056 b->free_ij = PETSC_TRUE; 4057 } else { 4058 b->free_a = PETSC_FALSE; 4059 b->free_ij = PETSC_FALSE; 4060 } 4061 4062 if (b->ipre && nnz != b->ipre && b->imax) { 4063 /* reserve user-requested sparsity */ 4064 PetscCall(PetscArraycpy(b->ipre,b->imax,B->rmap->n)); 4065 } 4066 4067 b->nz = 0; 4068 b->maxnz = nz; 4069 B->info.nz_unneeded = (double)b->maxnz; 4070 if (realalloc) { 4071 PetscCall(MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE)); 4072 } 4073 B->was_assembled = PETSC_FALSE; 4074 B->assembled = PETSC_FALSE; 4075 /* We simply deem preallocation has changed nonzero state. Updating the state 4076 will give clients (like AIJKokkos) a chance to know something has happened. 4077 */ 4078 B->nonzerostate++; 4079 PetscFunctionReturn(0); 4080 } 4081 4082 PetscErrorCode MatResetPreallocation_SeqAIJ(Mat A) 4083 { 4084 Mat_SeqAIJ *a; 4085 PetscInt i; 4086 4087 PetscFunctionBegin; 4088 PetscValidHeaderSpecific(A,MAT_CLASSID,1); 4089 4090 /* Check local size. If zero, then return */ 4091 if (!A->rmap->n) PetscFunctionReturn(0); 4092 4093 a = (Mat_SeqAIJ*)A->data; 4094 /* if no saved info, we error out */ 4095 PetscCheck(a->ipre,PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"No saved preallocation info "); 4096 4097 PetscCheck(a->i && a->j && a->a && a->imax && a->ilen,PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"Memory info is incomplete, and can not reset preallocation "); 4098 4099 PetscCall(PetscArraycpy(a->imax,a->ipre,A->rmap->n)); 4100 PetscCall(PetscArrayzero(a->ilen,A->rmap->n)); 4101 a->i[0] = 0; 4102 for (i=1; i<A->rmap->n+1; i++) { 4103 a->i[i] = a->i[i-1] + a->imax[i-1]; 4104 } 4105 A->preallocated = PETSC_TRUE; 4106 a->nz = 0; 4107 a->maxnz = a->i[A->rmap->n]; 4108 A->info.nz_unneeded = (double)a->maxnz; 4109 A->was_assembled = PETSC_FALSE; 4110 A->assembled = PETSC_FALSE; 4111 PetscFunctionReturn(0); 4112 } 4113 4114 /*@ 4115 MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format. 4116 4117 Input Parameters: 4118 + B - the matrix 4119 . i - the indices into j for the start of each row (starts with zero) 4120 . j - the column indices for each row (starts with zero) these must be sorted for each row 4121 - v - optional values in the matrix 4122 4123 Level: developer 4124 4125 Notes: 4126 The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays() 4127 4128 This routine may be called multiple times with different nonzero patterns (or the same nonzero pattern). The nonzero 4129 structure will be the union of all the previous nonzero structures. 4130 4131 Developer Notes: 4132 An optimization could be added to the implementation where it checks if the i, and j are identical to the current i and j and 4133 then just copies the v values directly with PetscMemcpy(). 4134 4135 This routine could also take a PetscCopyMode argument to allow sharing the values instead of always copying them. 4136 4137 .seealso: `MatCreate()`, `MatCreateSeqAIJ()`, `MatSetValues()`, `MatSeqAIJSetPreallocation()`, `MatCreateSeqAIJ()`, `MATSEQAIJ`, `MatResetPreallocation()` 4138 @*/ 4139 PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[]) 4140 { 4141 PetscFunctionBegin; 4142 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 4143 PetscValidType(B,1); 4144 PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v)); 4145 PetscFunctionReturn(0); 4146 } 4147 4148 PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 4149 { 4150 PetscInt i; 4151 PetscInt m,n; 4152 PetscInt nz; 4153 PetscInt *nnz; 4154 4155 PetscFunctionBegin; 4156 PetscCheck(Ii[0] == 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %" PetscInt_FMT, Ii[0]); 4157 4158 PetscCall(PetscLayoutSetUp(B->rmap)); 4159 PetscCall(PetscLayoutSetUp(B->cmap)); 4160 4161 PetscCall(MatGetSize(B, &m, &n)); 4162 PetscCall(PetscMalloc1(m+1, &nnz)); 4163 for (i = 0; i < m; i++) { 4164 nz = Ii[i+1]- Ii[i]; 4165 PetscCheck(nz >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %" PetscInt_FMT " has a negative number of columns %" PetscInt_FMT, i, nz); 4166 nnz[i] = nz; 4167 } 4168 PetscCall(MatSeqAIJSetPreallocation(B, 0, nnz)); 4169 PetscCall(PetscFree(nnz)); 4170 4171 for (i = 0; i < m; i++) { 4172 PetscCall(MatSetValues_SeqAIJ(B, 1, &i, Ii[i+1] - Ii[i], J+Ii[i], v ? v + Ii[i] : NULL, INSERT_VALUES)); 4173 } 4174 4175 PetscCall(MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY)); 4176 PetscCall(MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY)); 4177 4178 PetscCall(MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE)); 4179 PetscFunctionReturn(0); 4180 } 4181 4182 /*@ 4183 MatSeqAIJKron - Computes C, the Kronecker product of A and B. 4184 4185 Input Parameters: 4186 + A - left-hand side matrix 4187 . B - right-hand side matrix 4188 - reuse - either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX 4189 4190 Output Parameter: 4191 . C - Kronecker product of A and B 4192 4193 Level: intermediate 4194 4195 Notes: 4196 MAT_REUSE_MATRIX can only be used when the nonzero structure of the product matrix has not changed from that last call to MatSeqAIJKron(). 4197 4198 .seealso: `MatCreateSeqAIJ()`, `MATSEQAIJ`, `MATKAIJ`, `MatReuse` 4199 @*/ 4200 PetscErrorCode MatSeqAIJKron(Mat A,Mat B,MatReuse reuse,Mat *C) 4201 { 4202 PetscFunctionBegin; 4203 PetscValidHeaderSpecific(A,MAT_CLASSID,1); 4204 PetscValidType(A,1); 4205 PetscValidHeaderSpecific(B,MAT_CLASSID,2); 4206 PetscValidType(B,2); 4207 PetscValidPointer(C,4); 4208 if (reuse == MAT_REUSE_MATRIX) { 4209 PetscValidHeaderSpecific(*C,MAT_CLASSID,4); 4210 PetscValidType(*C,4); 4211 } 4212 PetscTryMethod(A,"MatSeqAIJKron_C",(Mat,Mat,MatReuse,Mat*),(A,B,reuse,C)); 4213 PetscFunctionReturn(0); 4214 } 4215 4216 PetscErrorCode MatSeqAIJKron_SeqAIJ(Mat A,Mat B,MatReuse reuse,Mat *C) 4217 { 4218 Mat newmat; 4219 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4220 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 4221 PetscScalar *v; 4222 const PetscScalar *aa,*ba; 4223 PetscInt *i,*j,m,n,p,q,nnz = 0,am = A->rmap->n,bm = B->rmap->n,an = A->cmap->n, bn = B->cmap->n; 4224 PetscBool flg; 4225 4226 PetscFunctionBegin; 4227 PetscCheck(!A->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 4228 PetscCheck(A->assembled,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 4229 PetscCheck(!B->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 4230 PetscCheck(B->assembled,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 4231 PetscCall(PetscObjectTypeCompare((PetscObject)B,MATSEQAIJ,&flg)); 4232 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_SUP,"MatType %s",((PetscObject)B)->type_name); 4233 PetscCheck(reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX,PETSC_COMM_SELF,PETSC_ERR_SUP,"MatReuse %d",(int)reuse); 4234 if (reuse == MAT_INITIAL_MATRIX) { 4235 PetscCall(PetscMalloc2(am*bm+1,&i,a->i[am]*b->i[bm],&j)); 4236 PetscCall(MatCreate(PETSC_COMM_SELF,&newmat)); 4237 PetscCall(MatSetSizes(newmat,am*bm,an*bn,am*bm,an*bn)); 4238 PetscCall(MatSetType(newmat,MATAIJ)); 4239 i[0] = 0; 4240 for (m = 0; m < am; ++m) { 4241 for (p = 0; p < bm; ++p) { 4242 i[m*bm + p + 1] = i[m*bm + p] + (a->i[m+1] - a->i[m]) * (b->i[p+1] - b->i[p]); 4243 for (n = a->i[m]; n < a->i[m+1]; ++n) { 4244 for (q = b->i[p]; q < b->i[p+1]; ++q) { 4245 j[nnz++] = a->j[n]*bn + b->j[q]; 4246 } 4247 } 4248 } 4249 } 4250 PetscCall(MatSeqAIJSetPreallocationCSR(newmat,i,j,NULL)); 4251 *C = newmat; 4252 PetscCall(PetscFree2(i,j)); 4253 nnz = 0; 4254 } 4255 PetscCall(MatSeqAIJGetArray(*C,&v)); 4256 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 4257 PetscCall(MatSeqAIJGetArrayRead(B,&ba)); 4258 for (m = 0; m < am; ++m) { 4259 for (p = 0; p < bm; ++p) { 4260 for (n = a->i[m]; n < a->i[m+1]; ++n) { 4261 for (q = b->i[p]; q < b->i[p+1]; ++q) { 4262 v[nnz++] = aa[n] * ba[q]; 4263 } 4264 } 4265 } 4266 } 4267 PetscCall(MatSeqAIJRestoreArray(*C,&v)); 4268 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 4269 PetscCall(MatSeqAIJRestoreArrayRead(B,&ba)); 4270 PetscFunctionReturn(0); 4271 } 4272 4273 #include <../src/mat/impls/dense/seq/dense.h> 4274 #include <petsc/private/kernels/petscaxpy.h> 4275 4276 /* 4277 Computes (B'*A')' since computing B*A directly is untenable 4278 4279 n p p 4280 [ ] [ ] [ ] 4281 m [ A ] * n [ B ] = m [ C ] 4282 [ ] [ ] [ ] 4283 4284 */ 4285 PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C) 4286 { 4287 Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data; 4288 Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data; 4289 Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data; 4290 PetscInt i,j,n,m,q,p; 4291 const PetscInt *ii,*idx; 4292 const PetscScalar *b,*a,*a_q; 4293 PetscScalar *c,*c_q; 4294 PetscInt clda = sub_c->lda; 4295 PetscInt alda = sub_a->lda; 4296 4297 PetscFunctionBegin; 4298 m = A->rmap->n; 4299 n = A->cmap->n; 4300 p = B->cmap->n; 4301 a = sub_a->v; 4302 b = sub_b->a; 4303 c = sub_c->v; 4304 if (clda == m) { 4305 PetscCall(PetscArrayzero(c,m*p)); 4306 } else { 4307 for (j=0;j<p;j++) 4308 for (i=0;i<m;i++) 4309 c[j*clda + i] = 0.0; 4310 } 4311 ii = sub_b->i; 4312 idx = sub_b->j; 4313 for (i=0; i<n; i++) { 4314 q = ii[i+1] - ii[i]; 4315 while (q-->0) { 4316 c_q = c + clda*(*idx); 4317 a_q = a + alda*i; 4318 PetscKernelAXPY(c_q,*b,a_q,m); 4319 idx++; 4320 b++; 4321 } 4322 } 4323 PetscFunctionReturn(0); 4324 } 4325 4326 PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C) 4327 { 4328 PetscInt m=A->rmap->n,n=B->cmap->n; 4329 PetscBool cisdense; 4330 4331 PetscFunctionBegin; 4332 PetscCheck(A->cmap->n == B->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"A->cmap->n %" PetscInt_FMT " != B->rmap->n %" PetscInt_FMT,A->cmap->n,B->rmap->n); 4333 PetscCall(MatSetSizes(C,m,n,m,n)); 4334 PetscCall(MatSetBlockSizesFromMats(C,A,B)); 4335 PetscCall(PetscObjectTypeCompareAny((PetscObject)C,&cisdense,MATSEQDENSE,MATSEQDENSECUDA,"")); 4336 if (!cisdense) { 4337 PetscCall(MatSetType(C,MATDENSE)); 4338 } 4339 PetscCall(MatSetUp(C)); 4340 4341 C->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ; 4342 PetscFunctionReturn(0); 4343 } 4344 4345 /* ----------------------------------------------------------------*/ 4346 /*MC 4347 MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices, 4348 based on compressed sparse row format. 4349 4350 Options Database Keys: 4351 . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions() 4352 4353 Level: beginner 4354 4355 Notes: 4356 MatSetValues() may be called for this matrix type with a NULL argument for the numerical values, 4357 in this case the values associated with the rows and columns one passes in are set to zero 4358 in the matrix 4359 4360 MatSetOptions(,MAT_STRUCTURE_ONLY,PETSC_TRUE) may be called for this matrix type. In this no 4361 space is allocated for the nonzero entries and any entries passed with MatSetValues() are ignored 4362 4363 Developer Notes: 4364 It would be nice if all matrix formats supported passing NULL in for the numerical values 4365 4366 .seealso: `MatCreateSeqAIJ()`, `MatSetFromOptions()`, `MatSetType()`, `MatCreate()`, `MatType`, `MATSELL`, `MATSEQSELL`, `MATMPISELL` 4367 M*/ 4368 4369 /*MC 4370 MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices. 4371 4372 This matrix type is identical to MATSEQAIJ when constructed with a single process communicator, 4373 and MATMPIAIJ otherwise. As a result, for single process communicators, 4374 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 4375 for communicators controlling multiple processes. It is recommended that you call both of 4376 the above preallocation routines for simplicity. 4377 4378 Options Database Keys: 4379 . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions() 4380 4381 Developer Notes: 4382 Subclasses include MATAIJCUSPARSE, MATAIJPERM, MATAIJSELL, MATAIJMKL, MATAIJCRL, and also automatically switches over to use inodes when 4383 enough exist. 4384 4385 Level: beginner 4386 4387 .seealso: `MatCreateAIJ()`, `MatCreateSeqAIJ()`, `MATSEQAIJ`, `MATMPIAIJ`, `MATSELL`, `MATSEQSELL`, `MATMPISELL` 4388 M*/ 4389 4390 /*MC 4391 MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices. 4392 4393 This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator, 4394 and MATMPIAIJCRL otherwise. As a result, for single process communicators, 4395 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 4396 for communicators controlling multiple processes. It is recommended that you call both of 4397 the above preallocation routines for simplicity. 4398 4399 Options Database Keys: 4400 . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions() 4401 4402 Level: beginner 4403 4404 .seealso: `MatCreateMPIAIJCRL`, `MATSEQAIJCRL`, `MATMPIAIJCRL`, `MATSEQAIJCRL`, `MATMPIAIJCRL` 4405 M*/ 4406 4407 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*); 4408 #if defined(PETSC_HAVE_ELEMENTAL) 4409 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat,MatType,MatReuse,Mat*); 4410 #endif 4411 #if defined(PETSC_HAVE_SCALAPACK) 4412 PETSC_INTERN PetscErrorCode MatConvert_AIJ_ScaLAPACK(Mat,MatType,MatReuse,Mat*); 4413 #endif 4414 #if defined(PETSC_HAVE_HYPRE) 4415 PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat A,MatType,MatReuse,Mat*); 4416 #endif 4417 4418 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqSELL(Mat,MatType,MatReuse,Mat*); 4419 PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat,MatType,MatReuse,Mat*); 4420 PETSC_INTERN PetscErrorCode MatProductSetFromOptions_IS_XAIJ(Mat); 4421 4422 /*@C 4423 MatSeqAIJGetArray - gives read/write access to the array where the data for a MATSEQAIJ matrix is stored 4424 4425 Not Collective 4426 4427 Input Parameter: 4428 . mat - a MATSEQAIJ matrix 4429 4430 Output Parameter: 4431 . array - pointer to the data 4432 4433 Level: intermediate 4434 4435 .seealso: `MatSeqAIJRestoreArray()`, `MatSeqAIJGetArrayF90()` 4436 @*/ 4437 PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array) 4438 { 4439 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4440 4441 PetscFunctionBegin; 4442 if (aij->ops->getarray) { 4443 PetscCall((*aij->ops->getarray)(A,array)); 4444 } else { 4445 *array = aij->a; 4446 } 4447 PetscFunctionReturn(0); 4448 } 4449 4450 /*@C 4451 MatSeqAIJRestoreArray - returns access to the array where the data for a MATSEQAIJ matrix is stored obtained by MatSeqAIJGetArray() 4452 4453 Not Collective 4454 4455 Input Parameters: 4456 + mat - a MATSEQAIJ matrix 4457 - array - pointer to the data 4458 4459 Level: intermediate 4460 4461 .seealso: `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArrayF90()` 4462 @*/ 4463 PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array) 4464 { 4465 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4466 4467 PetscFunctionBegin; 4468 if (aij->ops->restorearray) { 4469 PetscCall((*aij->ops->restorearray)(A,array)); 4470 } else { 4471 *array = NULL; 4472 } 4473 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 4474 PetscCall(PetscObjectStateIncrease((PetscObject)A)); 4475 PetscFunctionReturn(0); 4476 } 4477 4478 /*@C 4479 MatSeqAIJGetArrayRead - gives read-only access to the array where the data for a MATSEQAIJ matrix is stored 4480 4481 Not Collective 4482 4483 Input Parameter: 4484 . mat - a MATSEQAIJ matrix 4485 4486 Output Parameter: 4487 . array - pointer to the data 4488 4489 Level: intermediate 4490 4491 .seealso: `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArrayRead()` 4492 @*/ 4493 PetscErrorCode MatSeqAIJGetArrayRead(Mat A,const PetscScalar **array) 4494 { 4495 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4496 4497 PetscFunctionBegin; 4498 if (aij->ops->getarrayread) { 4499 PetscCall((*aij->ops->getarrayread)(A,array)); 4500 } else { 4501 *array = aij->a; 4502 } 4503 PetscFunctionReturn(0); 4504 } 4505 4506 /*@C 4507 MatSeqAIJRestoreArrayRead - restore the read-only access array obtained from MatSeqAIJGetArrayRead 4508 4509 Not Collective 4510 4511 Input Parameter: 4512 . mat - a MATSEQAIJ matrix 4513 4514 Output Parameter: 4515 . array - pointer to the data 4516 4517 Level: intermediate 4518 4519 .seealso: `MatSeqAIJGetArray()`, `MatSeqAIJGetArrayRead()` 4520 @*/ 4521 PetscErrorCode MatSeqAIJRestoreArrayRead(Mat A,const PetscScalar **array) 4522 { 4523 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4524 4525 PetscFunctionBegin; 4526 if (aij->ops->restorearrayread) { 4527 PetscCall((*aij->ops->restorearrayread)(A,array)); 4528 } else { 4529 *array = NULL; 4530 } 4531 PetscFunctionReturn(0); 4532 } 4533 4534 /*@C 4535 MatSeqAIJGetArrayWrite - gives write-only access to the array where the data for a MATSEQAIJ matrix is stored 4536 4537 Not Collective 4538 4539 Input Parameter: 4540 . mat - a MATSEQAIJ matrix 4541 4542 Output Parameter: 4543 . array - pointer to the data 4544 4545 Level: intermediate 4546 4547 .seealso: `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArrayRead()` 4548 @*/ 4549 PetscErrorCode MatSeqAIJGetArrayWrite(Mat A,PetscScalar **array) 4550 { 4551 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4552 4553 PetscFunctionBegin; 4554 if (aij->ops->getarraywrite) { 4555 PetscCall((*aij->ops->getarraywrite)(A,array)); 4556 } else { 4557 *array = aij->a; 4558 } 4559 PetscCall(MatSeqAIJInvalidateDiagonal(A)); 4560 PetscCall(PetscObjectStateIncrease((PetscObject)A)); 4561 PetscFunctionReturn(0); 4562 } 4563 4564 /*@C 4565 MatSeqAIJRestoreArrayWrite - restore the read-only access array obtained from MatSeqAIJGetArrayRead 4566 4567 Not Collective 4568 4569 Input Parameter: 4570 . mat - a MATSEQAIJ matrix 4571 4572 Output Parameter: 4573 . array - pointer to the data 4574 4575 Level: intermediate 4576 4577 .seealso: `MatSeqAIJGetArray()`, `MatSeqAIJGetArrayRead()` 4578 @*/ 4579 PetscErrorCode MatSeqAIJRestoreArrayWrite(Mat A,PetscScalar **array) 4580 { 4581 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4582 4583 PetscFunctionBegin; 4584 if (aij->ops->restorearraywrite) { 4585 PetscCall((*aij->ops->restorearraywrite)(A,array)); 4586 } else { 4587 *array = NULL; 4588 } 4589 PetscFunctionReturn(0); 4590 } 4591 4592 /*@C 4593 MatSeqAIJGetCSRAndMemType - Get the CSR arrays and the memory type of the SEQAIJ matrix 4594 4595 Not Collective 4596 4597 Input Parameter: 4598 . mat - a matrix of type MATSEQAIJ or its subclasses 4599 4600 Output Parameters: 4601 + i - row map array of the matrix 4602 . j - column index array of the matrix 4603 . a - data array of the matrix 4604 - memtype - memory type of the arrays 4605 4606 Notes: 4607 Any of the output parameters can be NULL, in which case the corresponding value is not returned. 4608 If mat is a device matrix, the arrays are on the device. Otherwise, they are on the host. 4609 4610 One can call this routine on a preallocated but not assembled matrix to just get the memory of the CSR underneath the matrix. 4611 If the matrix is assembled, the data array 'a' is guaranteed to have the latest values of the matrix. 4612 4613 Level: Developer 4614 4615 .seealso: `MatSeqAIJGetArray()`, `MatSeqAIJGetArrayRead()` 4616 @*/ 4617 PetscErrorCode MatSeqAIJGetCSRAndMemType(Mat mat,const PetscInt **i,const PetscInt **j,PetscScalar **a,PetscMemType *mtype) 4618 { 4619 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 4620 4621 PetscFunctionBegin; 4622 PetscCheck(mat->preallocated,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"matrix is not preallocated"); 4623 if (aij->ops->getcsrandmemtype) { 4624 PetscCall((*aij->ops->getcsrandmemtype)(mat,i,j,a,mtype)); 4625 } else { 4626 if (i) *i = aij->i; 4627 if (j) *j = aij->j; 4628 if (a) *a = aij->a; 4629 if (mtype) *mtype = PETSC_MEMTYPE_HOST; 4630 } 4631 PetscFunctionReturn(0); 4632 } 4633 4634 /*@C 4635 MatSeqAIJGetMaxRowNonzeros - returns the maximum number of nonzeros in any row 4636 4637 Not Collective 4638 4639 Input Parameter: 4640 . mat - a MATSEQAIJ matrix 4641 4642 Output Parameter: 4643 . nz - the maximum number of nonzeros in any row 4644 4645 Level: intermediate 4646 4647 .seealso: `MatSeqAIJRestoreArray()`, `MatSeqAIJGetArrayF90()` 4648 @*/ 4649 PetscErrorCode MatSeqAIJGetMaxRowNonzeros(Mat A,PetscInt *nz) 4650 { 4651 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4652 4653 PetscFunctionBegin; 4654 *nz = aij->rmax; 4655 PetscFunctionReturn(0); 4656 } 4657 4658 PetscErrorCode MatSetPreallocationCOO_SeqAIJ(Mat mat, PetscCount coo_n, const PetscInt coo_i[], const PetscInt coo_j[]) 4659 { 4660 MPI_Comm comm; 4661 PetscInt *i,*j; 4662 PetscInt M,N,row; 4663 PetscCount k,p,q,nneg,nnz,start,end; /* Index the coo array, so use PetscCount as their type */ 4664 PetscInt *Ai; /* Change to PetscCount once we use it for row pointers */ 4665 PetscInt *Aj; 4666 PetscScalar *Aa; 4667 Mat_SeqAIJ *seqaij = (Mat_SeqAIJ*)(mat->data); 4668 MatType rtype; 4669 PetscCount *perm,*jmap; 4670 4671 PetscFunctionBegin; 4672 PetscCall(MatResetPreallocationCOO_SeqAIJ(mat)); 4673 PetscCall(PetscObjectGetComm((PetscObject)mat,&comm)); 4674 PetscCall(MatGetSize(mat,&M,&N)); 4675 PetscCall(PetscMalloc2(coo_n,&i,coo_n,&j)); 4676 PetscCall(PetscArraycpy(i,coo_i,coo_n)); /* Make a copy since we'll modify it */ 4677 PetscCall(PetscArraycpy(j,coo_j,coo_n)); 4678 PetscCall(PetscMalloc1(coo_n,&perm)); 4679 for (k=0; k<coo_n; k++) { /* Ignore entries with negative row or col indices */ 4680 if (j[k] < 0) i[k] = -1; 4681 perm[k] = k; 4682 } 4683 4684 /* Sort by row */ 4685 PetscCall(PetscSortIntWithIntCountArrayPair(coo_n,i,j,perm)); 4686 for (k=0; k<coo_n; k++) {if (i[k] >= 0) break;} /* Advance k to the first row with a non-negative index */ 4687 nneg = k; 4688 PetscCall(PetscMalloc1(coo_n-nneg+1,&jmap)); /* +1 to make a CSR-like data structure. jmap[i] originally is the number of repeats for i-th nonzero */ 4689 nnz = 0; /* Total number of unique nonzeros to be counted */ 4690 jmap++; /* Inc jmap by 1 for convinience */ 4691 4692 PetscCall(PetscCalloc1(M+1,&Ai)); /* CSR of A */ 4693 PetscCall(PetscMalloc1(coo_n-nneg,&Aj)); /* We have at most coo_n-nneg unique nonzeros */ 4694 4695 /* In each row, sort by column, then unique column indices to get row length */ 4696 Ai++; /* Inc by 1 for convinience */ 4697 q = 0; /* q-th unique nonzero, with q starting from 0 */ 4698 while (k<coo_n) { 4699 row = i[k]; 4700 start = k; /* [start,end) indices for this row */ 4701 while (k<coo_n && i[k] == row) k++; 4702 end = k; 4703 PetscCall(PetscSortIntWithCountArray(end-start,j+start,perm+start)); 4704 /* Find number of unique col entries in this row */ 4705 Aj[q] = j[start]; /* Log the first nonzero in this row */ 4706 jmap[q] = 1; /* Number of repeats of this nozero entry */ 4707 Ai[row] = 1; 4708 nnz++; 4709 4710 for (p=start+1; p<end; p++) { /* Scan remaining nonzero in this row */ 4711 if (j[p] != j[p-1]) { /* Meet a new nonzero */ 4712 q++; 4713 jmap[q] = 1; 4714 Aj[q] = j[p]; 4715 Ai[row]++; 4716 nnz++; 4717 } else { 4718 jmap[q]++; 4719 } 4720 } 4721 q++; /* Move to next row and thus next unique nonzero */ 4722 } 4723 PetscCall(PetscFree2(i,j)); 4724 4725 Ai--; /* Back to the beginning of Ai[] */ 4726 for (k=0; k<M; k++) Ai[k+1] += Ai[k]; 4727 jmap--; /* Back to the beginning of jmap[] */ 4728 jmap[0] = 0; 4729 for (k=0; k<nnz; k++) jmap[k+1] += jmap[k]; 4730 if (nnz < coo_n-nneg) { /* Realloc with actual number of unique nonzeros */ 4731 PetscCount *jmap_new; 4732 PetscInt *Aj_new; 4733 4734 PetscCall(PetscMalloc1(nnz+1,&jmap_new)); 4735 PetscCall(PetscArraycpy(jmap_new,jmap,nnz+1)); 4736 PetscCall(PetscFree(jmap)); 4737 jmap = jmap_new; 4738 4739 PetscCall(PetscMalloc1(nnz,&Aj_new)); 4740 PetscCall(PetscArraycpy(Aj_new,Aj,nnz)); 4741 PetscCall(PetscFree(Aj)); 4742 Aj = Aj_new; 4743 } 4744 4745 if (nneg) { /* Discard heading entries with negative indices in perm[], as we'll access it from index 0 in MatSetValuesCOO */ 4746 PetscCount *perm_new; 4747 4748 PetscCall(PetscMalloc1(coo_n-nneg,&perm_new)); 4749 PetscCall(PetscArraycpy(perm_new,perm+nneg,coo_n-nneg)); 4750 PetscCall(PetscFree(perm)); 4751 perm = perm_new; 4752 } 4753 4754 PetscCall(MatGetRootType_Private(mat,&rtype)); 4755 PetscCall(PetscCalloc1(nnz,&Aa)); /* Zero the matrix */ 4756 PetscCall(MatSetSeqAIJWithArrays_private(PETSC_COMM_SELF,M,N,Ai,Aj,Aa,rtype,mat)); 4757 4758 seqaij->singlemalloc = PETSC_FALSE; /* Ai, Aj and Aa are not allocated in one big malloc */ 4759 seqaij->free_a = seqaij->free_ij = PETSC_TRUE; /* Let newmat own Ai, Aj and Aa */ 4760 /* Record COO fields */ 4761 seqaij->coo_n = coo_n; 4762 seqaij->Atot = coo_n-nneg; /* Annz is seqaij->nz, so no need to record that again */ 4763 seqaij->jmap = jmap; /* of length nnz+1 */ 4764 seqaij->perm = perm; 4765 PetscFunctionReturn(0); 4766 } 4767 4768 static PetscErrorCode MatSetValuesCOO_SeqAIJ(Mat A,const PetscScalar v[],InsertMode imode) 4769 { 4770 Mat_SeqAIJ *aseq = (Mat_SeqAIJ*)A->data; 4771 PetscCount i,j,Annz = aseq->nz; 4772 PetscCount *perm = aseq->perm,*jmap = aseq->jmap; 4773 PetscScalar *Aa; 4774 4775 PetscFunctionBegin; 4776 PetscCall(MatSeqAIJGetArray(A,&Aa)); 4777 for (i=0; i<Annz; i++) { 4778 PetscScalar sum = 0.0; 4779 for (j=jmap[i]; j<jmap[i+1]; j++) sum += v[perm[j]]; 4780 Aa[i] = (imode == INSERT_VALUES? 0.0 : Aa[i]) + sum; 4781 } 4782 PetscCall(MatSeqAIJRestoreArray(A,&Aa)); 4783 PetscFunctionReturn(0); 4784 } 4785 4786 #if defined(PETSC_HAVE_CUDA) 4787 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCUSPARSE(Mat,MatType,MatReuse,Mat*); 4788 #endif 4789 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 4790 PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJKokkos(Mat,MatType,MatReuse,Mat*); 4791 #endif 4792 4793 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B) 4794 { 4795 Mat_SeqAIJ *b; 4796 PetscMPIInt size; 4797 4798 PetscFunctionBegin; 4799 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)B),&size)); 4800 PetscCheck(size <= 1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1"); 4801 4802 PetscCall(PetscNewLog(B,&b)); 4803 4804 B->data = (void*)b; 4805 4806 PetscCall(PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps))); 4807 if (B->sortedfull) B->ops->setvalues = MatSetValues_SeqAIJ_SortedFull; 4808 4809 b->row = NULL; 4810 b->col = NULL; 4811 b->icol = NULL; 4812 b->reallocs = 0; 4813 b->ignorezeroentries = PETSC_FALSE; 4814 b->roworiented = PETSC_TRUE; 4815 b->nonew = 0; 4816 b->diag = NULL; 4817 b->solve_work = NULL; 4818 B->spptr = NULL; 4819 b->saved_values = NULL; 4820 b->idiag = NULL; 4821 b->mdiag = NULL; 4822 b->ssor_work = NULL; 4823 b->omega = 1.0; 4824 b->fshift = 0.0; 4825 b->idiagvalid = PETSC_FALSE; 4826 b->ibdiagvalid = PETSC_FALSE; 4827 b->keepnonzeropattern = PETSC_FALSE; 4828 4829 PetscCall(PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ)); 4830 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4831 PetscCall(PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ)); 4832 PetscCall(PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ)); 4833 #endif 4834 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ)); 4835 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ)); 4836 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ)); 4837 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ)); 4838 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ)); 4839 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM)); 4840 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijsell_C",MatConvert_SeqAIJ_SeqAIJSELL)); 4841 #if defined(PETSC_HAVE_MKL_SPARSE) 4842 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijmkl_C",MatConvert_SeqAIJ_SeqAIJMKL)); 4843 #endif 4844 #if defined(PETSC_HAVE_CUDA) 4845 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcusparse_C",MatConvert_SeqAIJ_SeqAIJCUSPARSE)); 4846 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqaijcusparse_seqaij_C",MatProductSetFromOptions_SeqAIJ)); 4847 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqaij_seqaijcusparse_C",MatProductSetFromOptions_SeqAIJ)); 4848 #endif 4849 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 4850 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijkokkos_C",MatConvert_SeqAIJ_SeqAIJKokkos)); 4851 #endif 4852 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL)); 4853 #if defined(PETSC_HAVE_ELEMENTAL) 4854 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_elemental_C",MatConvert_SeqAIJ_Elemental)); 4855 #endif 4856 #if defined(PETSC_HAVE_SCALAPACK) 4857 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_scalapack_C",MatConvert_AIJ_ScaLAPACK)); 4858 #endif 4859 #if defined(PETSC_HAVE_HYPRE) 4860 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_hypre_C",MatConvert_AIJ_HYPRE)); 4861 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_transpose_seqaij_seqaij_C",MatProductSetFromOptions_Transpose_AIJ_AIJ)); 4862 #endif 4863 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqdense_C",MatConvert_SeqAIJ_SeqDense)); 4864 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsell_C",MatConvert_SeqAIJ_SeqSELL)); 4865 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_is_C",MatConvert_XAIJ_IS)); 4866 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ)); 4867 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ)); 4868 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ)); 4869 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatResetPreallocation_C",MatResetPreallocation_SeqAIJ)); 4870 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ)); 4871 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ)); 4872 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_is_seqaij_C",MatProductSetFromOptions_IS_XAIJ)); 4873 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqdense_seqaij_C",MatProductSetFromOptions_SeqDense_SeqAIJ)); 4874 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqaij_seqaij_C",MatProductSetFromOptions_SeqAIJ)); 4875 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJKron_C",MatSeqAIJKron_SeqAIJ)); 4876 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSetPreallocationCOO_C",MatSetPreallocationCOO_SeqAIJ)); 4877 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSetValuesCOO_C",MatSetValuesCOO_SeqAIJ)); 4878 PetscCall(MatCreate_SeqAIJ_Inode(B)); 4879 PetscCall(PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ)); 4880 PetscCall(MatSeqAIJSetTypeFromOptions(B)); /* this allows changing the matrix subtype to say MATSEQAIJPERM */ 4881 PetscFunctionReturn(0); 4882 } 4883 4884 /* 4885 Given a matrix generated with MatGetFactor() duplicates all the information in A into B 4886 */ 4887 PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace) 4888 { 4889 Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data,*a = (Mat_SeqAIJ*)A->data; 4890 PetscInt m = A->rmap->n,i; 4891 4892 PetscFunctionBegin; 4893 PetscCheck(A->assembled || cpvalues == MAT_DO_NOT_COPY_VALUES,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot duplicate unassembled matrix"); 4894 4895 C->factortype = A->factortype; 4896 c->row = NULL; 4897 c->col = NULL; 4898 c->icol = NULL; 4899 c->reallocs = 0; 4900 4901 C->assembled = A->assembled; 4902 C->preallocated = A->preallocated; 4903 4904 if (A->preallocated) { 4905 PetscCall(PetscLayoutReference(A->rmap,&C->rmap)); 4906 PetscCall(PetscLayoutReference(A->cmap,&C->cmap)); 4907 4908 PetscCall(PetscMalloc1(m,&c->imax)); 4909 PetscCall(PetscMemcpy(c->imax,a->imax,m*sizeof(PetscInt))); 4910 PetscCall(PetscMalloc1(m,&c->ilen)); 4911 PetscCall(PetscMemcpy(c->ilen,a->ilen,m*sizeof(PetscInt))); 4912 PetscCall(PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt))); 4913 4914 /* allocate the matrix space */ 4915 if (mallocmatspace) { 4916 PetscCall(PetscMalloc3(a->i[m],&c->a,a->i[m],&c->j,m+1,&c->i)); 4917 PetscCall(PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt))); 4918 4919 c->singlemalloc = PETSC_TRUE; 4920 4921 PetscCall(PetscArraycpy(c->i,a->i,m+1)); 4922 if (m > 0) { 4923 PetscCall(PetscArraycpy(c->j,a->j,a->i[m])); 4924 if (cpvalues == MAT_COPY_VALUES) { 4925 const PetscScalar *aa; 4926 4927 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 4928 PetscCall(PetscArraycpy(c->a,aa,a->i[m])); 4929 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 4930 } else { 4931 PetscCall(PetscArrayzero(c->a,a->i[m])); 4932 } 4933 } 4934 } 4935 4936 c->ignorezeroentries = a->ignorezeroentries; 4937 c->roworiented = a->roworiented; 4938 c->nonew = a->nonew; 4939 if (a->diag) { 4940 PetscCall(PetscMalloc1(m+1,&c->diag)); 4941 PetscCall(PetscMemcpy(c->diag,a->diag,m*sizeof(PetscInt))); 4942 PetscCall(PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt))); 4943 } else c->diag = NULL; 4944 4945 c->solve_work = NULL; 4946 c->saved_values = NULL; 4947 c->idiag = NULL; 4948 c->ssor_work = NULL; 4949 c->keepnonzeropattern = a->keepnonzeropattern; 4950 c->free_a = PETSC_TRUE; 4951 c->free_ij = PETSC_TRUE; 4952 4953 c->rmax = a->rmax; 4954 c->nz = a->nz; 4955 c->maxnz = a->nz; /* Since we allocate exactly the right amount */ 4956 4957 c->compressedrow.use = a->compressedrow.use; 4958 c->compressedrow.nrows = a->compressedrow.nrows; 4959 if (a->compressedrow.use) { 4960 i = a->compressedrow.nrows; 4961 PetscCall(PetscMalloc2(i+1,&c->compressedrow.i,i,&c->compressedrow.rindex)); 4962 PetscCall(PetscArraycpy(c->compressedrow.i,a->compressedrow.i,i+1)); 4963 PetscCall(PetscArraycpy(c->compressedrow.rindex,a->compressedrow.rindex,i)); 4964 } else { 4965 c->compressedrow.use = PETSC_FALSE; 4966 c->compressedrow.i = NULL; 4967 c->compressedrow.rindex = NULL; 4968 } 4969 c->nonzerorowcnt = a->nonzerorowcnt; 4970 C->nonzerostate = A->nonzerostate; 4971 4972 PetscCall(MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C)); 4973 } 4974 PetscCall(PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist)); 4975 PetscFunctionReturn(0); 4976 } 4977 4978 PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B) 4979 { 4980 PetscFunctionBegin; 4981 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),B)); 4982 PetscCall(MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n)); 4983 if (!(A->rmap->n % A->rmap->bs) && !(A->cmap->n % A->cmap->bs)) { 4984 PetscCall(MatSetBlockSizesFromMats(*B,A,A)); 4985 } 4986 PetscCall(MatSetType(*B,((PetscObject)A)->type_name)); 4987 PetscCall(MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE)); 4988 PetscFunctionReturn(0); 4989 } 4990 4991 PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer) 4992 { 4993 PetscBool isbinary, ishdf5; 4994 4995 PetscFunctionBegin; 4996 PetscValidHeaderSpecific(newMat,MAT_CLASSID,1); 4997 PetscValidHeaderSpecific(viewer,PETSC_VIEWER_CLASSID,2); 4998 /* force binary viewer to load .info file if it has not yet done so */ 4999 PetscCall(PetscViewerSetUp(viewer)); 5000 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary)); 5001 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERHDF5, &ishdf5)); 5002 if (isbinary) { 5003 PetscCall(MatLoad_SeqAIJ_Binary(newMat,viewer)); 5004 } else if (ishdf5) { 5005 #if defined(PETSC_HAVE_HDF5) 5006 PetscCall(MatLoad_AIJ_HDF5(newMat,viewer)); 5007 #else 5008 SETERRQ(PetscObjectComm((PetscObject)newMat),PETSC_ERR_SUP,"HDF5 not supported in this build.\nPlease reconfigure using --download-hdf5"); 5009 #endif 5010 } else { 5011 SETERRQ(PetscObjectComm((PetscObject)newMat),PETSC_ERR_SUP,"Viewer type %s not yet supported for reading %s matrices",((PetscObject)viewer)->type_name,((PetscObject)newMat)->type_name); 5012 } 5013 PetscFunctionReturn(0); 5014 } 5015 5016 PetscErrorCode MatLoad_SeqAIJ_Binary(Mat mat, PetscViewer viewer) 5017 { 5018 Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->data; 5019 PetscInt header[4],*rowlens,M,N,nz,sum,rows,cols,i; 5020 5021 PetscFunctionBegin; 5022 PetscCall(PetscViewerSetUp(viewer)); 5023 5024 /* read in matrix header */ 5025 PetscCall(PetscViewerBinaryRead(viewer,header,4,NULL,PETSC_INT)); 5026 PetscCheck(header[0] == MAT_FILE_CLASSID,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Not a matrix object in file"); 5027 M = header[1]; N = header[2]; nz = header[3]; 5028 PetscCheck(M >= 0,PetscObjectComm((PetscObject)viewer),PETSC_ERR_FILE_UNEXPECTED,"Matrix row size (%" PetscInt_FMT ") in file is negative",M); 5029 PetscCheck(N >= 0,PetscObjectComm((PetscObject)viewer),PETSC_ERR_FILE_UNEXPECTED,"Matrix column size (%" PetscInt_FMT ") in file is negative",N); 5030 PetscCheck(nz >= 0,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk, cannot load as SeqAIJ"); 5031 5032 /* set block sizes from the viewer's .info file */ 5033 PetscCall(MatLoad_Binary_BlockSizes(mat,viewer)); 5034 /* set local and global sizes if not set already */ 5035 if (mat->rmap->n < 0) mat->rmap->n = M; 5036 if (mat->cmap->n < 0) mat->cmap->n = N; 5037 if (mat->rmap->N < 0) mat->rmap->N = M; 5038 if (mat->cmap->N < 0) mat->cmap->N = N; 5039 PetscCall(PetscLayoutSetUp(mat->rmap)); 5040 PetscCall(PetscLayoutSetUp(mat->cmap)); 5041 5042 /* check if the matrix sizes are correct */ 5043 PetscCall(MatGetSize(mat,&rows,&cols)); 5044 PetscCheck(M == rows && N == cols,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different sizes (%" PetscInt_FMT ", %" PetscInt_FMT ") than the input matrix (%" PetscInt_FMT ", %" PetscInt_FMT ")",M,N,rows,cols); 5045 5046 /* read in row lengths */ 5047 PetscCall(PetscMalloc1(M,&rowlens)); 5048 PetscCall(PetscViewerBinaryRead(viewer,rowlens,M,NULL,PETSC_INT)); 5049 /* check if sum(rowlens) is same as nz */ 5050 sum = 0; for (i=0; i<M; i++) sum += rowlens[i]; 5051 PetscCheck(sum == nz,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent matrix data in file: nonzeros = %" PetscInt_FMT ", sum-row-lengths = %" PetscInt_FMT,nz,sum); 5052 /* preallocate and check sizes */ 5053 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(mat,0,rowlens)); 5054 PetscCall(MatGetSize(mat,&rows,&cols)); 5055 PetscCheck(M == rows && N == cols,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different length (%" PetscInt_FMT ", %" PetscInt_FMT ") than the input matrix (%" PetscInt_FMT ", %" PetscInt_FMT ")",M,N,rows,cols); 5056 /* store row lengths */ 5057 PetscCall(PetscArraycpy(a->ilen,rowlens,M)); 5058 PetscCall(PetscFree(rowlens)); 5059 5060 /* fill in "i" row pointers */ 5061 a->i[0] = 0; for (i=0; i<M; i++) a->i[i+1] = a->i[i] + a->ilen[i]; 5062 /* read in "j" column indices */ 5063 PetscCall(PetscViewerBinaryRead(viewer,a->j,nz,NULL,PETSC_INT)); 5064 /* read in "a" nonzero values */ 5065 PetscCall(PetscViewerBinaryRead(viewer,a->a,nz,NULL,PETSC_SCALAR)); 5066 5067 PetscCall(MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY)); 5068 PetscCall(MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY)); 5069 PetscFunctionReturn(0); 5070 } 5071 5072 PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg) 5073 { 5074 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data; 5075 const PetscScalar *aa,*ba; 5076 #if defined(PETSC_USE_COMPLEX) 5077 PetscInt k; 5078 #endif 5079 5080 PetscFunctionBegin; 5081 /* If the matrix dimensions are not equal,or no of nonzeros */ 5082 if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) { 5083 *flg = PETSC_FALSE; 5084 PetscFunctionReturn(0); 5085 } 5086 5087 /* if the a->i are the same */ 5088 PetscCall(PetscArraycmp(a->i,b->i,A->rmap->n+1,flg)); 5089 if (!*flg) PetscFunctionReturn(0); 5090 5091 /* if a->j are the same */ 5092 PetscCall(PetscArraycmp(a->j,b->j,a->nz,flg)); 5093 if (!*flg) PetscFunctionReturn(0); 5094 5095 PetscCall(MatSeqAIJGetArrayRead(A,&aa)); 5096 PetscCall(MatSeqAIJGetArrayRead(B,&ba)); 5097 /* if a->a are the same */ 5098 #if defined(PETSC_USE_COMPLEX) 5099 for (k=0; k<a->nz; k++) { 5100 if (PetscRealPart(aa[k]) != PetscRealPart(ba[k]) || PetscImaginaryPart(aa[k]) != PetscImaginaryPart(ba[k])) { 5101 *flg = PETSC_FALSE; 5102 PetscFunctionReturn(0); 5103 } 5104 } 5105 #else 5106 PetscCall(PetscArraycmp(aa,ba,a->nz,flg)); 5107 #endif 5108 PetscCall(MatSeqAIJRestoreArrayRead(A,&aa)); 5109 PetscCall(MatSeqAIJRestoreArrayRead(B,&ba)); 5110 PetscFunctionReturn(0); 5111 } 5112 5113 /*@ 5114 MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format) 5115 provided by the user. 5116 5117 Collective 5118 5119 Input Parameters: 5120 + comm - must be an MPI communicator of size 1 5121 . m - number of rows 5122 . n - number of columns 5123 . i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of elements in that row of the matrix 5124 . j - column indices 5125 - a - matrix values 5126 5127 Output Parameter: 5128 . mat - the matrix 5129 5130 Level: intermediate 5131 5132 Notes: 5133 The i, j, and a arrays are not copied by this routine, the user must free these arrays 5134 once the matrix is destroyed and not before 5135 5136 You cannot set new nonzero locations into this matrix, that will generate an error. 5137 5138 The i and j indices are 0 based 5139 5140 The format which is used for the sparse matrix input, is equivalent to a 5141 row-major ordering.. i.e for the following matrix, the input data expected is 5142 as shown 5143 5144 $ 1 0 0 5145 $ 2 0 3 5146 $ 4 5 6 5147 $ 5148 $ i = {0,1,3,6} [size = nrow+1 = 3+1] 5149 $ j = {0,0,2,0,1,2} [size = 6]; values must be sorted for each row 5150 $ v = {1,2,3,4,5,6} [size = 6] 5151 5152 .seealso: `MatCreate()`, `MatCreateAIJ()`, `MatCreateSeqAIJ()`, `MatCreateMPIAIJWithArrays()`, `MatMPIAIJSetPreallocationCSR()` 5153 5154 @*/ 5155 PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat) 5156 { 5157 PetscInt ii; 5158 Mat_SeqAIJ *aij; 5159 PetscInt jj; 5160 5161 PetscFunctionBegin; 5162 PetscCheck(m <= 0 || i[0] == 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 5163 PetscCall(MatCreate(comm,mat)); 5164 PetscCall(MatSetSizes(*mat,m,n,m,n)); 5165 /* PetscCall(MatSetBlockSizes(*mat,,)); */ 5166 PetscCall(MatSetType(*mat,MATSEQAIJ)); 5167 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,NULL)); 5168 aij = (Mat_SeqAIJ*)(*mat)->data; 5169 PetscCall(PetscMalloc1(m,&aij->imax)); 5170 PetscCall(PetscMalloc1(m,&aij->ilen)); 5171 5172 aij->i = i; 5173 aij->j = j; 5174 aij->a = a; 5175 aij->singlemalloc = PETSC_FALSE; 5176 aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 5177 aij->free_a = PETSC_FALSE; 5178 aij->free_ij = PETSC_FALSE; 5179 5180 for (ii=0,aij->nonzerorowcnt=0,aij->rmax=0; ii<m; ii++) { 5181 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 5182 if (PetscDefined(USE_DEBUG)) { 5183 PetscCheck(i[ii+1] - i[ii] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row length in i (row indices) row = %" PetscInt_FMT " length = %" PetscInt_FMT,ii,i[ii+1] - i[ii]); 5184 for (jj=i[ii]+1; jj<i[ii+1]; jj++) { 5185 PetscCheck(j[jj] >= j[jj-1],PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %" PetscInt_FMT " (actual column %" PetscInt_FMT ") in row %" PetscInt_FMT " is not sorted",jj-i[ii],j[jj],ii); 5186 PetscCheck(j[jj] != j[jj-1],PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %" PetscInt_FMT " (actual column %" PetscInt_FMT ") in row %" PetscInt_FMT " is identical to previous entry",jj-i[ii],j[jj],ii); 5187 } 5188 } 5189 } 5190 if (PetscDefined(USE_DEBUG)) { 5191 for (ii=0; ii<aij->i[m]; ii++) { 5192 PetscCheck(j[ii] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %" PetscInt_FMT " index = %" PetscInt_FMT,ii,j[ii]); 5193 PetscCheck(j[ii] <= n - 1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column index to large at location = %" PetscInt_FMT " index = %" PetscInt_FMT,ii,j[ii]); 5194 } 5195 } 5196 5197 PetscCall(MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY)); 5198 PetscCall(MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY)); 5199 PetscFunctionReturn(0); 5200 } 5201 5202 /*@ 5203 MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format) 5204 provided by the user. 5205 5206 Collective 5207 5208 Input Parameters: 5209 + comm - must be an MPI communicator of size 1 5210 . m - number of rows 5211 . n - number of columns 5212 . i - row indices 5213 . j - column indices 5214 . a - matrix values 5215 . nz - number of nonzeros 5216 - idx - if the i and j indices start with 1 use PETSC_TRUE otherwise use PETSC_FALSE 5217 5218 Output Parameter: 5219 . mat - the matrix 5220 5221 Level: intermediate 5222 5223 Example: 5224 For the following matrix, the input data expected is as shown (using 0 based indexing) 5225 .vb 5226 1 0 0 5227 2 0 3 5228 4 5 6 5229 5230 i = {0,1,1,2,2,2} 5231 j = {0,0,2,0,1,2} 5232 v = {1,2,3,4,5,6} 5233 .ve 5234 5235 .seealso: `MatCreate()`, `MatCreateAIJ()`, `MatCreateSeqAIJ()`, `MatCreateSeqAIJWithArrays()`, `MatMPIAIJSetPreallocationCSR()`, `MatSetValuesCOO()` 5236 5237 @*/ 5238 PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat,PetscInt nz,PetscBool idx) 5239 { 5240 PetscInt ii, *nnz, one = 1,row,col; 5241 5242 PetscFunctionBegin; 5243 PetscCall(PetscCalloc1(m,&nnz)); 5244 for (ii = 0; ii < nz; ii++) { 5245 nnz[i[ii] - !!idx] += 1; 5246 } 5247 PetscCall(MatCreate(comm,mat)); 5248 PetscCall(MatSetSizes(*mat,m,n,m,n)); 5249 PetscCall(MatSetType(*mat,MATSEQAIJ)); 5250 PetscCall(MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz)); 5251 for (ii = 0; ii < nz; ii++) { 5252 if (idx) { 5253 row = i[ii] - 1; 5254 col = j[ii] - 1; 5255 } else { 5256 row = i[ii]; 5257 col = j[ii]; 5258 } 5259 PetscCall(MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES)); 5260 } 5261 PetscCall(MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY)); 5262 PetscCall(MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY)); 5263 PetscCall(PetscFree(nnz)); 5264 PetscFunctionReturn(0); 5265 } 5266 5267 PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A) 5268 { 5269 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 5270 5271 PetscFunctionBegin; 5272 a->idiagvalid = PETSC_FALSE; 5273 a->ibdiagvalid = PETSC_FALSE; 5274 5275 PetscCall(MatSeqAIJInvalidateDiagonal_Inode(A)); 5276 PetscFunctionReturn(0); 5277 } 5278 5279 PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat) 5280 { 5281 PetscMPIInt size; 5282 5283 PetscFunctionBegin; 5284 PetscCallMPI(MPI_Comm_size(comm,&size)); 5285 if (size == 1) { 5286 if (scall == MAT_INITIAL_MATRIX) { 5287 PetscCall(MatDuplicate(inmat,MAT_COPY_VALUES,outmat)); 5288 } else { 5289 PetscCall(MatCopy(inmat,*outmat,SAME_NONZERO_PATTERN)); 5290 } 5291 } else { 5292 PetscCall(MatCreateMPIMatConcatenateSeqMat_MPIAIJ(comm,inmat,n,scall,outmat)); 5293 } 5294 PetscFunctionReturn(0); 5295 } 5296 5297 /* 5298 Permute A into C's *local* index space using rowemb,colemb. 5299 The embedding are supposed to be injections and the above implies that the range of rowemb is a subset 5300 of [0,m), colemb is in [0,n). 5301 If pattern == DIFFERENT_NONZERO_PATTERN, C is preallocated according to A. 5302 */ 5303 PetscErrorCode MatSetSeqMat_SeqAIJ(Mat C,IS rowemb,IS colemb,MatStructure pattern,Mat B) 5304 { 5305 /* If making this function public, change the error returned in this function away from _PLIB. */ 5306 Mat_SeqAIJ *Baij; 5307 PetscBool seqaij; 5308 PetscInt m,n,*nz,i,j,count; 5309 PetscScalar v; 5310 const PetscInt *rowindices,*colindices; 5311 5312 PetscFunctionBegin; 5313 if (!B) PetscFunctionReturn(0); 5314 /* Check to make sure the target matrix (and embeddings) are compatible with C and each other. */ 5315 PetscCall(PetscObjectBaseTypeCompare((PetscObject)B,MATSEQAIJ,&seqaij)); 5316 PetscCheck(seqaij,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is of wrong type"); 5317 if (rowemb) { 5318 PetscCall(ISGetLocalSize(rowemb,&m)); 5319 PetscCheck(m == B->rmap->n,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Row IS of size %" PetscInt_FMT " is incompatible with matrix row size %" PetscInt_FMT,m,B->rmap->n); 5320 } else { 5321 PetscCheck(C->rmap->n == B->rmap->n,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is row-incompatible with the target matrix"); 5322 } 5323 if (colemb) { 5324 PetscCall(ISGetLocalSize(colemb,&n)); 5325 PetscCheck(n == B->cmap->n,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Diag col IS of size %" PetscInt_FMT " is incompatible with input matrix col size %" PetscInt_FMT,n,B->cmap->n); 5326 } else { 5327 PetscCheck(C->cmap->n == B->cmap->n,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is col-incompatible with the target matrix"); 5328 } 5329 5330 Baij = (Mat_SeqAIJ*)(B->data); 5331 if (pattern == DIFFERENT_NONZERO_PATTERN) { 5332 PetscCall(PetscMalloc1(B->rmap->n,&nz)); 5333 for (i=0; i<B->rmap->n; i++) { 5334 nz[i] = Baij->i[i+1] - Baij->i[i]; 5335 } 5336 PetscCall(MatSeqAIJSetPreallocation(C,0,nz)); 5337 PetscCall(PetscFree(nz)); 5338 } 5339 if (pattern == SUBSET_NONZERO_PATTERN) { 5340 PetscCall(MatZeroEntries(C)); 5341 } 5342 count = 0; 5343 rowindices = NULL; 5344 colindices = NULL; 5345 if (rowemb) { 5346 PetscCall(ISGetIndices(rowemb,&rowindices)); 5347 } 5348 if (colemb) { 5349 PetscCall(ISGetIndices(colemb,&colindices)); 5350 } 5351 for (i=0; i<B->rmap->n; i++) { 5352 PetscInt row; 5353 row = i; 5354 if (rowindices) row = rowindices[i]; 5355 for (j=Baij->i[i]; j<Baij->i[i+1]; j++) { 5356 PetscInt col; 5357 col = Baij->j[count]; 5358 if (colindices) col = colindices[col]; 5359 v = Baij->a[count]; 5360 PetscCall(MatSetValues(C,1,&row,1,&col,&v,INSERT_VALUES)); 5361 ++count; 5362 } 5363 } 5364 /* FIXME: set C's nonzerostate correctly. */ 5365 /* Assembly for C is necessary. */ 5366 C->preallocated = PETSC_TRUE; 5367 C->assembled = PETSC_TRUE; 5368 C->was_assembled = PETSC_FALSE; 5369 PetscFunctionReturn(0); 5370 } 5371 5372 PetscFunctionList MatSeqAIJList = NULL; 5373 5374 /*@C 5375 MatSeqAIJSetType - Converts a MATSEQAIJ matrix to a subtype 5376 5377 Collective on Mat 5378 5379 Input Parameters: 5380 + mat - the matrix object 5381 - matype - matrix type 5382 5383 Options Database Key: 5384 . -mat_seqai_type <method> - for example seqaijcrl 5385 5386 Level: intermediate 5387 5388 .seealso: `PCSetType()`, `VecSetType()`, `MatCreate()`, `MatType`, `Mat` 5389 @*/ 5390 PetscErrorCode MatSeqAIJSetType(Mat mat, MatType matype) 5391 { 5392 PetscBool sametype; 5393 PetscErrorCode (*r)(Mat,MatType,MatReuse,Mat*); 5394 5395 PetscFunctionBegin; 5396 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 5397 PetscCall(PetscObjectTypeCompare((PetscObject)mat,matype,&sametype)); 5398 if (sametype) PetscFunctionReturn(0); 5399 5400 PetscCall(PetscFunctionListFind(MatSeqAIJList,matype,&r)); 5401 PetscCheck(r,PETSC_COMM_SELF,PETSC_ERR_ARG_UNKNOWN_TYPE,"Unknown Mat type given: %s",matype); 5402 PetscCall((*r)(mat,matype,MAT_INPLACE_MATRIX,&mat)); 5403 PetscFunctionReturn(0); 5404 } 5405 5406 /*@C 5407 MatSeqAIJRegister - - Adds a new sub-matrix type for sequential AIJ matrices 5408 5409 Not Collective 5410 5411 Input Parameters: 5412 + name - name of a new user-defined matrix type, for example MATSEQAIJCRL 5413 - function - routine to convert to subtype 5414 5415 Notes: 5416 MatSeqAIJRegister() may be called multiple times to add several user-defined solvers. 5417 5418 Then, your matrix can be chosen with the procedural interface at runtime via the option 5419 $ -mat_seqaij_type my_mat 5420 5421 Level: advanced 5422 5423 .seealso: `MatSeqAIJRegisterAll()` 5424 5425 Level: advanced 5426 @*/ 5427 PetscErrorCode MatSeqAIJRegister(const char sname[],PetscErrorCode (*function)(Mat,MatType,MatReuse,Mat *)) 5428 { 5429 PetscFunctionBegin; 5430 PetscCall(MatInitializePackage()); 5431 PetscCall(PetscFunctionListAdd(&MatSeqAIJList,sname,function)); 5432 PetscFunctionReturn(0); 5433 } 5434 5435 PetscBool MatSeqAIJRegisterAllCalled = PETSC_FALSE; 5436 5437 /*@C 5438 MatSeqAIJRegisterAll - Registers all of the matrix subtypes of SeqAIJ 5439 5440 Not Collective 5441 5442 Level: advanced 5443 5444 .seealso: `MatRegisterAll()`, `MatSeqAIJRegister()` 5445 @*/ 5446 PetscErrorCode MatSeqAIJRegisterAll(void) 5447 { 5448 PetscFunctionBegin; 5449 if (MatSeqAIJRegisterAllCalled) PetscFunctionReturn(0); 5450 MatSeqAIJRegisterAllCalled = PETSC_TRUE; 5451 5452 PetscCall(MatSeqAIJRegister(MATSEQAIJCRL, MatConvert_SeqAIJ_SeqAIJCRL)); 5453 PetscCall(MatSeqAIJRegister(MATSEQAIJPERM, MatConvert_SeqAIJ_SeqAIJPERM)); 5454 PetscCall(MatSeqAIJRegister(MATSEQAIJSELL, MatConvert_SeqAIJ_SeqAIJSELL)); 5455 #if defined(PETSC_HAVE_MKL_SPARSE) 5456 PetscCall(MatSeqAIJRegister(MATSEQAIJMKL, MatConvert_SeqAIJ_SeqAIJMKL)); 5457 #endif 5458 #if defined(PETSC_HAVE_CUDA) 5459 PetscCall(MatSeqAIJRegister(MATSEQAIJCUSPARSE, MatConvert_SeqAIJ_SeqAIJCUSPARSE)); 5460 #endif 5461 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 5462 PetscCall(MatSeqAIJRegister(MATSEQAIJKOKKOS, MatConvert_SeqAIJ_SeqAIJKokkos)); 5463 #endif 5464 #if defined(PETSC_HAVE_VIENNACL) && defined(PETSC_HAVE_VIENNACL_NO_CUDA) 5465 PetscCall(MatSeqAIJRegister(MATMPIAIJVIENNACL, MatConvert_SeqAIJ_SeqAIJViennaCL)); 5466 #endif 5467 PetscFunctionReturn(0); 5468 } 5469 5470 /* 5471 Special version for direct calls from Fortran 5472 */ 5473 #include <petsc/private/fortranimpl.h> 5474 #if defined(PETSC_HAVE_FORTRAN_CAPS) 5475 #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ 5476 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 5477 #define matsetvaluesseqaij_ matsetvaluesseqaij 5478 #endif 5479 5480 /* Change these macros so can be used in void function */ 5481 5482 /* Change these macros so can be used in void function */ 5483 /* Identical to PetscCallVoid, except it assigns to *_ierr */ 5484 #undef PetscCall 5485 #define PetscCall(...) do { \ 5486 PetscErrorCode ierr_msv_mpiaij = __VA_ARGS__; \ 5487 if (PetscUnlikely(ierr_msv_mpiaij)) { \ 5488 *_ierr = PetscError(PETSC_COMM_SELF,__LINE__,PETSC_FUNCTION_NAME,__FILE__,ierr_msv_mpiaij,PETSC_ERROR_REPEAT," "); \ 5489 return; \ 5490 } \ 5491 } while (0) 5492 5493 #undef SETERRQ 5494 #define SETERRQ(comm,ierr,...) do { \ 5495 *_ierr = PetscError(comm,__LINE__,PETSC_FUNCTION_NAME,__FILE__,ierr,PETSC_ERROR_INITIAL,__VA_ARGS__); \ 5496 return; \ 5497 } while (0) 5498 5499 PETSC_EXTERN void matsetvaluesseqaij_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[],InsertMode *isis, PetscErrorCode *_ierr) 5500 { 5501 Mat A = *AA; 5502 PetscInt m = *mm, n = *nn; 5503 InsertMode is = *isis; 5504 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 5505 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 5506 PetscInt *imax,*ai,*ailen; 5507 PetscInt *aj,nonew = a->nonew,lastcol = -1; 5508 MatScalar *ap,value,*aa; 5509 PetscBool ignorezeroentries = a->ignorezeroentries; 5510 PetscBool roworiented = a->roworiented; 5511 5512 PetscFunctionBegin; 5513 MatCheckPreallocated(A,1); 5514 imax = a->imax; 5515 ai = a->i; 5516 ailen = a->ilen; 5517 aj = a->j; 5518 aa = a->a; 5519 5520 for (k=0; k<m; k++) { /* loop over added rows */ 5521 row = im[k]; 5522 if (row < 0) continue; 5523 PetscCheck(row < A->rmap->n,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large"); 5524 rp = aj + ai[row]; ap = aa + ai[row]; 5525 rmax = imax[row]; nrow = ailen[row]; 5526 low = 0; 5527 high = nrow; 5528 for (l=0; l<n; l++) { /* loop over added columns */ 5529 if (in[l] < 0) continue; 5530 PetscCheck(in[l] < A->cmap->n,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large"); 5531 col = in[l]; 5532 if (roworiented) value = v[l + k*n]; 5533 else value = v[k + l*m]; 5534 5535 if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue; 5536 5537 if (col <= lastcol) low = 0; 5538 else high = nrow; 5539 lastcol = col; 5540 while (high-low > 5) { 5541 t = (low+high)/2; 5542 if (rp[t] > col) high = t; 5543 else low = t; 5544 } 5545 for (i=low; i<high; i++) { 5546 if (rp[i] > col) break; 5547 if (rp[i] == col) { 5548 if (is == ADD_VALUES) ap[i] += value; 5549 else ap[i] = value; 5550 goto noinsert; 5551 } 5552 } 5553 if (value == 0.0 && ignorezeroentries) goto noinsert; 5554 if (nonew == 1) goto noinsert; 5555 PetscCheck(nonew != -1,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); 5556 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 5557 N = nrow++ - 1; a->nz++; high++; 5558 /* shift up all the later entries in this row */ 5559 for (ii=N; ii>=i; ii--) { 5560 rp[ii+1] = rp[ii]; 5561 ap[ii+1] = ap[ii]; 5562 } 5563 rp[i] = col; 5564 ap[i] = value; 5565 A->nonzerostate++; 5566 noinsert:; 5567 low = i + 1; 5568 } 5569 ailen[row] = nrow; 5570 } 5571 PetscFunctionReturnVoid(); 5572 } 5573 /* Undefining these here since they were redefined from their original definition above! No 5574 * other PETSc functions should be defined past this point, as it is impossible to recover the 5575 * original definitions */ 5576 #undef PetscCall 5577 #undef SETERRQ 5578