1 2 /* 3 Defines the basic matrix operations for the AIJ (compressed row) 4 matrix storage format. 5 */ 6 7 8 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 9 #include <petscblaslapack.h> 10 #include <petscbt.h> 11 #include <petsc-private/kernels/blocktranspose.h> 12 #if defined(PETSC_THREADCOMM_ACTIVE) 13 #include <petscthreadcomm.h> 14 #endif 15 16 #undef __FUNCT__ 17 #define __FUNCT__ "MatGetColumnNorms_SeqAIJ" 18 PetscErrorCode MatGetColumnNorms_SeqAIJ(Mat A,NormType type,PetscReal *norms) 19 { 20 PetscErrorCode ierr; 21 PetscInt i,m,n; 22 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 23 24 PetscFunctionBegin; 25 ierr = MatGetSize(A,&m,&n);CHKERRQ(ierr); 26 ierr = PetscMemzero(norms,n*sizeof(PetscReal));CHKERRQ(ierr); 27 if (type == NORM_2) { 28 for (i=0; i<aij->i[m]; i++) { 29 norms[aij->j[i]] += PetscAbsScalar(aij->a[i]*aij->a[i]); 30 } 31 } else if (type == NORM_1) { 32 for (i=0; i<aij->i[m]; i++) { 33 norms[aij->j[i]] += PetscAbsScalar(aij->a[i]); 34 } 35 } else if (type == NORM_INFINITY) { 36 for (i=0; i<aij->i[m]; i++) { 37 norms[aij->j[i]] = PetscMax(PetscAbsScalar(aij->a[i]),norms[aij->j[i]]); 38 } 39 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown NormType"); 40 41 if (type == NORM_2) { 42 for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]); 43 } 44 PetscFunctionReturn(0); 45 } 46 47 #undef __FUNCT__ 48 #define __FUNCT__ "MatFindOffBlockDiagonalEntries_SeqAIJ" 49 PetscErrorCode MatFindOffBlockDiagonalEntries_SeqAIJ(Mat A,IS *is) 50 { 51 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 52 PetscInt i,m=A->rmap->n,cnt = 0, bs = A->rmap->bs; 53 const PetscInt *jj = a->j,*ii = a->i; 54 PetscInt *rows; 55 PetscErrorCode ierr; 56 57 PetscFunctionBegin; 58 for (i=0; i<m; i++) { 59 if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) { 60 cnt++; 61 } 62 } 63 ierr = PetscMalloc1(cnt,&rows);CHKERRQ(ierr); 64 cnt = 0; 65 for (i=0; i<m; i++) { 66 if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) { 67 rows[cnt] = i; 68 cnt++; 69 } 70 } 71 ierr = ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,is);CHKERRQ(ierr); 72 PetscFunctionReturn(0); 73 } 74 75 #undef __FUNCT__ 76 #define __FUNCT__ "MatFindZeroDiagonals_SeqAIJ_Private" 77 PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat A,PetscInt *nrows,PetscInt **zrows) 78 { 79 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 80 const MatScalar *aa = a->a; 81 PetscInt i,m=A->rmap->n,cnt = 0; 82 const PetscInt *jj = a->j,*diag; 83 PetscInt *rows; 84 PetscErrorCode ierr; 85 86 PetscFunctionBegin; 87 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 88 diag = a->diag; 89 for (i=0; i<m; i++) { 90 if ((jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) { 91 cnt++; 92 } 93 } 94 ierr = PetscMalloc1(cnt,&rows);CHKERRQ(ierr); 95 cnt = 0; 96 for (i=0; i<m; i++) { 97 if ((jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) { 98 rows[cnt++] = i; 99 } 100 } 101 *nrows = cnt; 102 *zrows = rows; 103 PetscFunctionReturn(0); 104 } 105 106 #undef __FUNCT__ 107 #define __FUNCT__ "MatFindZeroDiagonals_SeqAIJ" 108 PetscErrorCode MatFindZeroDiagonals_SeqAIJ(Mat A,IS *zrows) 109 { 110 PetscInt nrows,*rows; 111 PetscErrorCode ierr; 112 113 PetscFunctionBegin; 114 *zrows = NULL; 115 ierr = MatFindZeroDiagonals_SeqAIJ_Private(A,&nrows,&rows);CHKERRQ(ierr); 116 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)A),nrows,rows,PETSC_OWN_POINTER,zrows);CHKERRQ(ierr); 117 PetscFunctionReturn(0); 118 } 119 120 #undef __FUNCT__ 121 #define __FUNCT__ "MatFindNonzeroRows_SeqAIJ" 122 PetscErrorCode MatFindNonzeroRows_SeqAIJ(Mat A,IS *keptrows) 123 { 124 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 125 const MatScalar *aa; 126 PetscInt m=A->rmap->n,cnt = 0; 127 const PetscInt *ii; 128 PetscInt n,i,j,*rows; 129 PetscErrorCode ierr; 130 131 PetscFunctionBegin; 132 *keptrows = 0; 133 ii = a->i; 134 for (i=0; i<m; i++) { 135 n = ii[i+1] - ii[i]; 136 if (!n) { 137 cnt++; 138 goto ok1; 139 } 140 aa = a->a + ii[i]; 141 for (j=0; j<n; j++) { 142 if (aa[j] != 0.0) goto ok1; 143 } 144 cnt++; 145 ok1:; 146 } 147 if (!cnt) PetscFunctionReturn(0); 148 ierr = PetscMalloc1((A->rmap->n-cnt),&rows);CHKERRQ(ierr); 149 cnt = 0; 150 for (i=0; i<m; i++) { 151 n = ii[i+1] - ii[i]; 152 if (!n) continue; 153 aa = a->a + ii[i]; 154 for (j=0; j<n; j++) { 155 if (aa[j] != 0.0) { 156 rows[cnt++] = i; 157 break; 158 } 159 } 160 } 161 ierr = ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,keptrows);CHKERRQ(ierr); 162 PetscFunctionReturn(0); 163 } 164 165 #undef __FUNCT__ 166 #define __FUNCT__ "MatDiagonalSet_SeqAIJ" 167 PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is) 168 { 169 PetscErrorCode ierr; 170 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data; 171 PetscInt i,*diag, m = Y->rmap->n; 172 MatScalar *aa = aij->a; 173 PetscScalar *v; 174 PetscBool missing; 175 176 PetscFunctionBegin; 177 if (Y->assembled) { 178 ierr = MatMissingDiagonal_SeqAIJ(Y,&missing,NULL);CHKERRQ(ierr); 179 if (!missing) { 180 diag = aij->diag; 181 ierr = VecGetArray(D,&v);CHKERRQ(ierr); 182 if (is == INSERT_VALUES) { 183 for (i=0; i<m; i++) { 184 aa[diag[i]] = v[i]; 185 } 186 } else { 187 for (i=0; i<m; i++) { 188 aa[diag[i]] += v[i]; 189 } 190 } 191 ierr = VecRestoreArray(D,&v);CHKERRQ(ierr); 192 PetscFunctionReturn(0); 193 } 194 ierr = MatSeqAIJInvalidateDiagonal(Y);CHKERRQ(ierr); 195 } 196 ierr = MatDiagonalSet_Default(Y,D,is);CHKERRQ(ierr); 197 PetscFunctionReturn(0); 198 } 199 200 #undef __FUNCT__ 201 #define __FUNCT__ "MatGetRowIJ_SeqAIJ" 202 PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *m,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 203 { 204 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 205 PetscErrorCode ierr; 206 PetscInt i,ishift; 207 208 PetscFunctionBegin; 209 *m = A->rmap->n; 210 if (!ia) PetscFunctionReturn(0); 211 ishift = 0; 212 if (symmetric && !A->structurally_symmetric) { 213 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,ishift,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 214 } else if (oshift == 1) { 215 PetscInt *tia; 216 PetscInt nz = a->i[A->rmap->n]; 217 /* malloc space and add 1 to i and j indices */ 218 ierr = PetscMalloc1((A->rmap->n+1),&tia);CHKERRQ(ierr); 219 for (i=0; i<A->rmap->n+1; i++) tia[i] = a->i[i] + 1; 220 *ia = tia; 221 if (ja) { 222 PetscInt *tja; 223 ierr = PetscMalloc1((nz+1),&tja);CHKERRQ(ierr); 224 for (i=0; i<nz; i++) tja[i] = a->j[i] + 1; 225 *ja = tja; 226 } 227 } else { 228 *ia = a->i; 229 if (ja) *ja = a->j; 230 } 231 PetscFunctionReturn(0); 232 } 233 234 #undef __FUNCT__ 235 #define __FUNCT__ "MatRestoreRowIJ_SeqAIJ" 236 PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 237 { 238 PetscErrorCode ierr; 239 240 PetscFunctionBegin; 241 if (!ia) PetscFunctionReturn(0); 242 if ((symmetric && !A->structurally_symmetric) || oshift == 1) { 243 ierr = PetscFree(*ia);CHKERRQ(ierr); 244 if (ja) {ierr = PetscFree(*ja);CHKERRQ(ierr);} 245 } 246 PetscFunctionReturn(0); 247 } 248 249 #undef __FUNCT__ 250 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ" 251 PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 252 { 253 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 254 PetscErrorCode ierr; 255 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 256 PetscInt nz = a->i[m],row,*jj,mr,col; 257 258 PetscFunctionBegin; 259 *nn = n; 260 if (!ia) PetscFunctionReturn(0); 261 if (symmetric) { 262 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 263 } else { 264 ierr = PetscCalloc1(n+1,&collengths);CHKERRQ(ierr); 265 ierr = PetscMalloc1((n+1),&cia);CHKERRQ(ierr); 266 ierr = PetscMalloc1((nz+1),&cja);CHKERRQ(ierr); 267 jj = a->j; 268 for (i=0; i<nz; i++) { 269 collengths[jj[i]]++; 270 } 271 cia[0] = oshift; 272 for (i=0; i<n; i++) { 273 cia[i+1] = cia[i] + collengths[i]; 274 } 275 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 276 jj = a->j; 277 for (row=0; row<m; row++) { 278 mr = a->i[row+1] - a->i[row]; 279 for (i=0; i<mr; i++) { 280 col = *jj++; 281 282 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 283 } 284 } 285 ierr = PetscFree(collengths);CHKERRQ(ierr); 286 *ia = cia; *ja = cja; 287 } 288 PetscFunctionReturn(0); 289 } 290 291 #undef __FUNCT__ 292 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ" 293 PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 294 { 295 PetscErrorCode ierr; 296 297 PetscFunctionBegin; 298 if (!ia) PetscFunctionReturn(0); 299 300 ierr = PetscFree(*ia);CHKERRQ(ierr); 301 ierr = PetscFree(*ja);CHKERRQ(ierr); 302 PetscFunctionReturn(0); 303 } 304 305 /* 306 MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from 307 MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output 308 spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ() and MatFDColoringCreate_SeqXAIJ() 309 */ 310 #undef __FUNCT__ 311 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ_Color" 312 PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 313 { 314 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 315 PetscErrorCode ierr; 316 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 317 PetscInt nz = a->i[m],row,*jj,mr,col; 318 PetscInt *cspidx; 319 320 PetscFunctionBegin; 321 *nn = n; 322 if (!ia) PetscFunctionReturn(0); 323 324 ierr = PetscCalloc1(n+1,&collengths);CHKERRQ(ierr); 325 ierr = PetscMalloc1((n+1),&cia);CHKERRQ(ierr); 326 ierr = PetscMalloc1((nz+1),&cja);CHKERRQ(ierr); 327 ierr = PetscMalloc1((nz+1),&cspidx);CHKERRQ(ierr); 328 jj = a->j; 329 for (i=0; i<nz; i++) { 330 collengths[jj[i]]++; 331 } 332 cia[0] = oshift; 333 for (i=0; i<n; i++) { 334 cia[i+1] = cia[i] + collengths[i]; 335 } 336 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 337 jj = a->j; 338 for (row=0; row<m; row++) { 339 mr = a->i[row+1] - a->i[row]; 340 for (i=0; i<mr; i++) { 341 col = *jj++; 342 cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */ 343 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 344 } 345 } 346 ierr = PetscFree(collengths);CHKERRQ(ierr); 347 *ia = cia; *ja = cja; 348 *spidx = cspidx; 349 PetscFunctionReturn(0); 350 } 351 352 #undef __FUNCT__ 353 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ_Color" 354 PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 355 { 356 PetscErrorCode ierr; 357 358 PetscFunctionBegin; 359 ierr = MatRestoreColumnIJ_SeqAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done);CHKERRQ(ierr); 360 ierr = PetscFree(*spidx);CHKERRQ(ierr); 361 PetscFunctionReturn(0); 362 } 363 364 #undef __FUNCT__ 365 #define __FUNCT__ "MatSetValuesRow_SeqAIJ" 366 PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[]) 367 { 368 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 369 PetscInt *ai = a->i; 370 PetscErrorCode ierr; 371 372 PetscFunctionBegin; 373 ierr = PetscMemcpy(a->a+ai[row],v,(ai[row+1]-ai[row])*sizeof(PetscScalar));CHKERRQ(ierr); 374 PetscFunctionReturn(0); 375 } 376 377 /* 378 MatSeqAIJSetValuesLocalFast - An optimized version of MatSetValuesLocal() for SeqAIJ matrices with several assumptions 379 380 - a single row of values is set with each call 381 - no row or column indices are negative or (in error) larger than the number of rows or columns 382 - the values are always added to the matrix, not set 383 - no new locations are introduced in the nonzero structure of the matrix 384 385 This does NOT assume the global column indices are sorted 386 387 */ 388 389 #include <petsc-private/isimpl.h> 390 #undef __FUNCT__ 391 #define __FUNCT__ "MatSeqAIJSetValuesLocalFast" 392 PetscErrorCode MatSeqAIJSetValuesLocalFast(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 393 { 394 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 395 PetscInt low,high,t,row,nrow,i,col,l; 396 const PetscInt *rp,*ai = a->i,*ailen = a->ilen,*aj = a->j; 397 PetscInt lastcol = -1; 398 MatScalar *ap,value,*aa = a->a; 399 const PetscInt *ridx = A->rmap->mapping->indices,*cidx = A->cmap->mapping->indices; 400 401 row = ridx[im[0]]; 402 rp = aj + ai[row]; 403 ap = aa + ai[row]; 404 nrow = ailen[row]; 405 low = 0; 406 high = nrow; 407 for (l=0; l<n; l++) { /* loop over added columns */ 408 col = cidx[in[l]]; 409 value = v[l]; 410 411 if (col <= lastcol) low = 0; 412 else high = nrow; 413 lastcol = col; 414 while (high-low > 5) { 415 t = (low+high)/2; 416 if (rp[t] > col) high = t; 417 else low = t; 418 } 419 for (i=low; i<high; i++) { 420 if (rp[i] == col) { 421 ap[i] += value; 422 low = i + 1; 423 break; 424 } 425 } 426 } 427 return 0; 428 } 429 430 #undef __FUNCT__ 431 #define __FUNCT__ "MatSetValues_SeqAIJ" 432 PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 433 { 434 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 435 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 436 PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen; 437 PetscErrorCode ierr; 438 PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1; 439 MatScalar *ap,value,*aa = a->a; 440 PetscBool ignorezeroentries = a->ignorezeroentries; 441 PetscBool roworiented = a->roworiented; 442 443 PetscFunctionBegin; 444 for (k=0; k<m; k++) { /* loop over added rows */ 445 row = im[k]; 446 if (row < 0) continue; 447 #if defined(PETSC_USE_DEBUG) 448 if (row >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap->n-1); 449 #endif 450 rp = aj + ai[row]; ap = aa + ai[row]; 451 rmax = imax[row]; nrow = ailen[row]; 452 low = 0; 453 high = nrow; 454 for (l=0; l<n; l++) { /* loop over added columns */ 455 if (in[l] < 0) continue; 456 #if defined(PETSC_USE_DEBUG) 457 if (in[l] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap->n-1); 458 #endif 459 col = in[l]; 460 if (roworiented) { 461 value = v[l + k*n]; 462 } else { 463 value = v[k + l*m]; 464 } 465 if ((value == 0.0 && ignorezeroentries) && (is == ADD_VALUES)) continue; 466 467 if (col <= lastcol) low = 0; 468 else high = nrow; 469 lastcol = col; 470 while (high-low > 5) { 471 t = (low+high)/2; 472 if (rp[t] > col) high = t; 473 else low = t; 474 } 475 for (i=low; i<high; i++) { 476 if (rp[i] > col) break; 477 if (rp[i] == col) { 478 if (is == ADD_VALUES) ap[i] += value; 479 else ap[i] = value; 480 low = i + 1; 481 goto noinsert; 482 } 483 } 484 if (value == 0.0 && ignorezeroentries) goto noinsert; 485 if (nonew == 1) goto noinsert; 486 if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%D,%D) in the matrix",row,col); 487 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 488 N = nrow++ - 1; a->nz++; high++; 489 /* shift up all the later entries in this row */ 490 for (ii=N; ii>=i; ii--) { 491 rp[ii+1] = rp[ii]; 492 ap[ii+1] = ap[ii]; 493 } 494 rp[i] = col; 495 ap[i] = value; 496 low = i + 1; 497 A->nonzerostate++; 498 noinsert:; 499 } 500 ailen[row] = nrow; 501 } 502 PetscFunctionReturn(0); 503 } 504 505 506 #undef __FUNCT__ 507 #define __FUNCT__ "MatGetValues_SeqAIJ" 508 PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[]) 509 { 510 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 511 PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j; 512 PetscInt *ai = a->i,*ailen = a->ilen; 513 MatScalar *ap,*aa = a->a; 514 515 PetscFunctionBegin; 516 for (k=0; k<m; k++) { /* loop over rows */ 517 row = im[k]; 518 if (row < 0) {v += n; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",row); */ 519 if (row >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap->n-1); 520 rp = aj + ai[row]; ap = aa + ai[row]; 521 nrow = ailen[row]; 522 for (l=0; l<n; l++) { /* loop over columns */ 523 if (in[l] < 0) {v++; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",in[l]); */ 524 if (in[l] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap->n-1); 525 col = in[l]; 526 high = nrow; low = 0; /* assume unsorted */ 527 while (high-low > 5) { 528 t = (low+high)/2; 529 if (rp[t] > col) high = t; 530 else low = t; 531 } 532 for (i=low; i<high; i++) { 533 if (rp[i] > col) break; 534 if (rp[i] == col) { 535 *v++ = ap[i]; 536 goto finished; 537 } 538 } 539 *v++ = 0.0; 540 finished:; 541 } 542 } 543 PetscFunctionReturn(0); 544 } 545 546 547 #undef __FUNCT__ 548 #define __FUNCT__ "MatView_SeqAIJ_Binary" 549 PetscErrorCode MatView_SeqAIJ_Binary(Mat A,PetscViewer viewer) 550 { 551 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 552 PetscErrorCode ierr; 553 PetscInt i,*col_lens; 554 int fd; 555 FILE *file; 556 557 PetscFunctionBegin; 558 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 559 ierr = PetscMalloc1((4+A->rmap->n),&col_lens);CHKERRQ(ierr); 560 561 col_lens[0] = MAT_FILE_CLASSID; 562 col_lens[1] = A->rmap->n; 563 col_lens[2] = A->cmap->n; 564 col_lens[3] = a->nz; 565 566 /* store lengths of each row and write (including header) to file */ 567 for (i=0; i<A->rmap->n; i++) { 568 col_lens[4+i] = a->i[i+1] - a->i[i]; 569 } 570 ierr = PetscBinaryWrite(fd,col_lens,4+A->rmap->n,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 571 ierr = PetscFree(col_lens);CHKERRQ(ierr); 572 573 /* store column indices (zero start index) */ 574 ierr = PetscBinaryWrite(fd,a->j,a->nz,PETSC_INT,PETSC_FALSE);CHKERRQ(ierr); 575 576 /* store nonzero values */ 577 ierr = PetscBinaryWrite(fd,a->a,a->nz,PETSC_SCALAR,PETSC_FALSE);CHKERRQ(ierr); 578 579 ierr = PetscViewerBinaryGetInfoPointer(viewer,&file);CHKERRQ(ierr); 580 if (file) { 581 fprintf(file,"-matload_block_size %d\n",(int)PetscAbs(A->rmap->bs)); 582 } 583 PetscFunctionReturn(0); 584 } 585 586 extern PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer); 587 588 #undef __FUNCT__ 589 #define __FUNCT__ "MatView_SeqAIJ_ASCII" 590 PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer) 591 { 592 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 593 PetscErrorCode ierr; 594 PetscInt i,j,m = A->rmap->n; 595 const char *name; 596 PetscViewerFormat format; 597 598 PetscFunctionBegin; 599 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 600 if (format == PETSC_VIEWER_ASCII_MATLAB) { 601 PetscInt nofinalvalue = 0; 602 if (m && ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap->n-1))) { 603 /* Need a dummy value to ensure the dimension of the matrix. */ 604 nofinalvalue = 1; 605 } 606 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 607 ierr = PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",m,A->cmap->n);CHKERRQ(ierr); 608 ierr = PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %D \n",a->nz);CHKERRQ(ierr); 609 #if defined(PETSC_USE_COMPLEX) 610 ierr = PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,4);\n",a->nz+nofinalvalue);CHKERRQ(ierr); 611 #else 612 ierr = PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,3);\n",a->nz+nofinalvalue);CHKERRQ(ierr); 613 #endif 614 ierr = PetscViewerASCIIPrintf(viewer,"zzz = [\n");CHKERRQ(ierr); 615 616 for (i=0; i<m; i++) { 617 for (j=a->i[i]; j<a->i[i+1]; j++) { 618 #if defined(PETSC_USE_COMPLEX) 619 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e %18.16e\n",i+1,a->j[j]+1,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 620 #else 621 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",i+1,a->j[j]+1,(double)a->a[j]);CHKERRQ(ierr); 622 #endif 623 } 624 } 625 if (nofinalvalue) { 626 #if defined(PETSC_USE_COMPLEX) 627 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e %18.16e\n",m,A->cmap->n,0.,0.);CHKERRQ(ierr); 628 #else 629 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",m,A->cmap->n,0.0);CHKERRQ(ierr); 630 #endif 631 } 632 ierr = PetscObjectGetName((PetscObject)A,&name);CHKERRQ(ierr); 633 ierr = PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);CHKERRQ(ierr); 634 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 635 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO) { 636 PetscFunctionReturn(0); 637 } else if (format == PETSC_VIEWER_ASCII_COMMON) { 638 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 639 for (i=0; i<m; i++) { 640 ierr = PetscViewerASCIIPrintf(viewer,"row %D:",i);CHKERRQ(ierr); 641 for (j=a->i[i]; j<a->i[i+1]; j++) { 642 #if defined(PETSC_USE_COMPLEX) 643 if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) { 644 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 645 } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) { 646 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 647 } else if (PetscRealPart(a->a[j]) != 0.0) { 648 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));CHKERRQ(ierr); 649 } 650 #else 651 if (a->a[j] != 0.0) {ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);CHKERRQ(ierr);} 652 #endif 653 } 654 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 655 } 656 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 657 } else if (format == PETSC_VIEWER_ASCII_SYMMODU) { 658 PetscInt nzd=0,fshift=1,*sptr; 659 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 660 ierr = PetscMalloc1((m+1),&sptr);CHKERRQ(ierr); 661 for (i=0; i<m; i++) { 662 sptr[i] = nzd+1; 663 for (j=a->i[i]; j<a->i[i+1]; j++) { 664 if (a->j[j] >= i) { 665 #if defined(PETSC_USE_COMPLEX) 666 if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++; 667 #else 668 if (a->a[j] != 0.0) nzd++; 669 #endif 670 } 671 } 672 } 673 sptr[m] = nzd+1; 674 ierr = PetscViewerASCIIPrintf(viewer," %D %D\n\n",m,nzd);CHKERRQ(ierr); 675 for (i=0; i<m+1; i+=6) { 676 if (i+4<m) { 677 ierr = PetscViewerASCIIPrintf(viewer," %D %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4],sptr[i+5]);CHKERRQ(ierr); 678 } else if (i+3<m) { 679 ierr = PetscViewerASCIIPrintf(viewer," %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4]);CHKERRQ(ierr); 680 } else if (i+2<m) { 681 ierr = PetscViewerASCIIPrintf(viewer," %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);CHKERRQ(ierr); 682 } else if (i+1<m) { 683 ierr = PetscViewerASCIIPrintf(viewer," %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2]);CHKERRQ(ierr); 684 } else if (i<m) { 685 ierr = PetscViewerASCIIPrintf(viewer," %D %D\n",sptr[i],sptr[i+1]);CHKERRQ(ierr); 686 } else { 687 ierr = PetscViewerASCIIPrintf(viewer," %D\n",sptr[i]);CHKERRQ(ierr); 688 } 689 } 690 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 691 ierr = PetscFree(sptr);CHKERRQ(ierr); 692 for (i=0; i<m; i++) { 693 for (j=a->i[i]; j<a->i[i+1]; j++) { 694 if (a->j[j] >= i) {ierr = PetscViewerASCIIPrintf(viewer," %D ",a->j[j]+fshift);CHKERRQ(ierr);} 695 } 696 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 697 } 698 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 699 for (i=0; i<m; i++) { 700 for (j=a->i[i]; j<a->i[i+1]; j++) { 701 if (a->j[j] >= i) { 702 #if defined(PETSC_USE_COMPLEX) 703 if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) { 704 ierr = PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 705 } 706 #else 707 if (a->a[j] != 0.0) {ierr = PetscViewerASCIIPrintf(viewer," %18.16e ",(double)a->a[j]);CHKERRQ(ierr);} 708 #endif 709 } 710 } 711 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 712 } 713 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 714 } else if (format == PETSC_VIEWER_ASCII_DENSE) { 715 PetscInt cnt = 0,jcnt; 716 PetscScalar value; 717 #if defined(PETSC_USE_COMPLEX) 718 PetscBool realonly = PETSC_TRUE; 719 720 for (i=0; i<a->i[m]; i++) { 721 if (PetscImaginaryPart(a->a[i]) != 0.0) { 722 realonly = PETSC_FALSE; 723 break; 724 } 725 } 726 #endif 727 728 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 729 for (i=0; i<m; i++) { 730 jcnt = 0; 731 for (j=0; j<A->cmap->n; j++) { 732 if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) { 733 value = a->a[cnt++]; 734 jcnt++; 735 } else { 736 value = 0.0; 737 } 738 #if defined(PETSC_USE_COMPLEX) 739 if (realonly) { 740 ierr = PetscViewerASCIIPrintf(viewer," %7.5e ",(double)PetscRealPart(value));CHKERRQ(ierr); 741 } else { 742 ierr = PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",(double)PetscRealPart(value),(double)PetscImaginaryPart(value));CHKERRQ(ierr); 743 } 744 #else 745 ierr = PetscViewerASCIIPrintf(viewer," %7.5e ",(double)value);CHKERRQ(ierr); 746 #endif 747 } 748 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 749 } 750 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 751 } else if (format == PETSC_VIEWER_ASCII_MATRIXMARKET) { 752 PetscInt fshift=1; 753 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 754 #if defined(PETSC_USE_COMPLEX) 755 ierr = PetscViewerASCIIPrintf(viewer,"%%matrix complex general\n");CHKERRQ(ierr); 756 #else 757 ierr = PetscViewerASCIIPrintf(viewer,"%%matrix real general\n");CHKERRQ(ierr); 758 #endif 759 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %D\n", m, A->cmap->n, a->nz);CHKERRQ(ierr); 760 for (i=0; i<m; i++) { 761 for (j=a->i[i]; j<a->i[i+1]; j++) { 762 #if defined(PETSC_USE_COMPLEX) 763 if (PetscImaginaryPart(a->a[j]) > 0.0) { 764 ierr = PetscViewerASCIIPrintf(viewer,"%D %D, %g %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 765 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 766 ierr = PetscViewerASCIIPrintf(viewer,"%D %D, %g -%g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 767 } else { 768 ierr = PetscViewerASCIIPrintf(viewer,"%D %D, %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]));CHKERRQ(ierr); 769 } 770 #else 771 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %g\n", i+fshift, a->j[j]+fshift, (double)a->a[j]);CHKERRQ(ierr); 772 #endif 773 } 774 } 775 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 776 } else { 777 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 778 if (A->factortype) { 779 for (i=0; i<m; i++) { 780 ierr = PetscViewerASCIIPrintf(viewer,"row %D:",i);CHKERRQ(ierr); 781 /* L part */ 782 for (j=a->i[i]; j<a->i[i+1]; j++) { 783 #if defined(PETSC_USE_COMPLEX) 784 if (PetscImaginaryPart(a->a[j]) > 0.0) { 785 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 786 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 787 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));CHKERRQ(ierr); 788 } else { 789 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));CHKERRQ(ierr); 790 } 791 #else 792 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);CHKERRQ(ierr); 793 #endif 794 } 795 /* diagonal */ 796 j = a->diag[i]; 797 #if defined(PETSC_USE_COMPLEX) 798 if (PetscImaginaryPart(a->a[j]) > 0.0) { 799 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)PetscImaginaryPart(1.0/a->a[j]));CHKERRQ(ierr); 800 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 801 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)(-PetscImaginaryPart(1.0/a->a[j])));CHKERRQ(ierr); 802 } else { 803 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(1.0/a->a[j]));CHKERRQ(ierr); 804 } 805 #else 806 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)(1.0/a->a[j]));CHKERRQ(ierr); 807 #endif 808 809 /* U part */ 810 for (j=a->diag[i+1]+1; j<a->diag[i]; j++) { 811 #if defined(PETSC_USE_COMPLEX) 812 if (PetscImaginaryPart(a->a[j]) > 0.0) { 813 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 814 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 815 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));CHKERRQ(ierr); 816 } else { 817 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));CHKERRQ(ierr); 818 } 819 #else 820 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);CHKERRQ(ierr); 821 #endif 822 } 823 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 824 } 825 } else { 826 for (i=0; i<m; i++) { 827 ierr = PetscViewerASCIIPrintf(viewer,"row %D:",i);CHKERRQ(ierr); 828 for (j=a->i[i]; j<a->i[i+1]; j++) { 829 #if defined(PETSC_USE_COMPLEX) 830 if (PetscImaginaryPart(a->a[j]) > 0.0) { 831 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 832 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 833 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 834 } else { 835 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));CHKERRQ(ierr); 836 } 837 #else 838 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);CHKERRQ(ierr); 839 #endif 840 } 841 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 842 } 843 } 844 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 845 } 846 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 847 PetscFunctionReturn(0); 848 } 849 850 #include <petscdraw.h> 851 #undef __FUNCT__ 852 #define __FUNCT__ "MatView_SeqAIJ_Draw_Zoom" 853 PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa) 854 { 855 Mat A = (Mat) Aa; 856 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 857 PetscErrorCode ierr; 858 PetscInt i,j,m = A->rmap->n,color; 859 PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r,maxv = 0.0; 860 PetscViewer viewer; 861 PetscViewerFormat format; 862 863 PetscFunctionBegin; 864 ierr = PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);CHKERRQ(ierr); 865 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 866 867 ierr = PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);CHKERRQ(ierr); 868 /* loop over matrix elements drawing boxes */ 869 870 if (format != PETSC_VIEWER_DRAW_CONTOUR) { 871 /* Blue for negative, Cyan for zero and Red for positive */ 872 color = PETSC_DRAW_BLUE; 873 for (i=0; i<m; i++) { 874 y_l = m - i - 1.0; y_r = y_l + 1.0; 875 for (j=a->i[i]; j<a->i[i+1]; j++) { 876 x_l = a->j[j]; x_r = x_l + 1.0; 877 if (PetscRealPart(a->a[j]) >= 0.) continue; 878 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 879 } 880 } 881 color = PETSC_DRAW_CYAN; 882 for (i=0; i<m; i++) { 883 y_l = m - i - 1.0; y_r = y_l + 1.0; 884 for (j=a->i[i]; j<a->i[i+1]; j++) { 885 x_l = a->j[j]; x_r = x_l + 1.0; 886 if (a->a[j] != 0.) continue; 887 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 888 } 889 } 890 color = PETSC_DRAW_RED; 891 for (i=0; i<m; i++) { 892 y_l = m - i - 1.0; y_r = y_l + 1.0; 893 for (j=a->i[i]; j<a->i[i+1]; j++) { 894 x_l = a->j[j]; x_r = x_l + 1.0; 895 if (PetscRealPart(a->a[j]) <= 0.) continue; 896 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 897 } 898 } 899 } else { 900 /* use contour shading to indicate magnitude of values */ 901 /* first determine max of all nonzero values */ 902 PetscInt nz = a->nz,count; 903 PetscDraw popup; 904 PetscReal scale; 905 906 for (i=0; i<nz; i++) { 907 if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]); 908 } 909 scale = (245.0 - PETSC_DRAW_BASIC_COLORS)/maxv; 910 ierr = PetscDrawGetPopup(draw,&popup);CHKERRQ(ierr); 911 if (popup) { 912 ierr = PetscDrawScalePopup(popup,0.0,maxv);CHKERRQ(ierr); 913 } 914 count = 0; 915 for (i=0; i<m; i++) { 916 y_l = m - i - 1.0; y_r = y_l + 1.0; 917 for (j=a->i[i]; j<a->i[i+1]; j++) { 918 x_l = a->j[j]; x_r = x_l + 1.0; 919 color = PETSC_DRAW_BASIC_COLORS + (PetscInt)(scale*PetscAbsScalar(a->a[count])); 920 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 921 count++; 922 } 923 } 924 } 925 PetscFunctionReturn(0); 926 } 927 928 #include <petscdraw.h> 929 #undef __FUNCT__ 930 #define __FUNCT__ "MatView_SeqAIJ_Draw" 931 PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer) 932 { 933 PetscErrorCode ierr; 934 PetscDraw draw; 935 PetscReal xr,yr,xl,yl,h,w; 936 PetscBool isnull; 937 938 PetscFunctionBegin; 939 ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr); 940 ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); 941 if (isnull) PetscFunctionReturn(0); 942 943 ierr = PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);CHKERRQ(ierr); 944 xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0; 945 xr += w; yr += h; xl = -w; yl = -h; 946 ierr = PetscDrawSetCoordinates(draw,xl,yl,xr,yr);CHKERRQ(ierr); 947 ierr = PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);CHKERRQ(ierr); 948 ierr = PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL);CHKERRQ(ierr); 949 PetscFunctionReturn(0); 950 } 951 952 #undef __FUNCT__ 953 #define __FUNCT__ "MatView_SeqAIJ" 954 PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer) 955 { 956 PetscErrorCode ierr; 957 PetscBool iascii,isbinary,isdraw; 958 959 PetscFunctionBegin; 960 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 961 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 962 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 963 if (iascii) { 964 ierr = MatView_SeqAIJ_ASCII(A,viewer);CHKERRQ(ierr); 965 } else if (isbinary) { 966 ierr = MatView_SeqAIJ_Binary(A,viewer);CHKERRQ(ierr); 967 } else if (isdraw) { 968 ierr = MatView_SeqAIJ_Draw(A,viewer);CHKERRQ(ierr); 969 } 970 ierr = MatView_SeqAIJ_Inode(A,viewer);CHKERRQ(ierr); 971 PetscFunctionReturn(0); 972 } 973 974 #undef __FUNCT__ 975 #define __FUNCT__ "MatAssemblyEnd_SeqAIJ" 976 PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode) 977 { 978 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 979 PetscErrorCode ierr; 980 PetscInt fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax; 981 PetscInt m = A->rmap->n,*ip,N,*ailen = a->ilen,rmax = 0; 982 MatScalar *aa = a->a,*ap; 983 PetscReal ratio = 0.6; 984 985 PetscFunctionBegin; 986 if (mode == MAT_FLUSH_ASSEMBLY) PetscFunctionReturn(0); 987 988 if (m) rmax = ailen[0]; /* determine row with most nonzeros */ 989 for (i=1; i<m; i++) { 990 /* move each row back by the amount of empty slots (fshift) before it*/ 991 fshift += imax[i-1] - ailen[i-1]; 992 rmax = PetscMax(rmax,ailen[i]); 993 if (fshift) { 994 ip = aj + ai[i]; 995 ap = aa + ai[i]; 996 N = ailen[i]; 997 for (j=0; j<N; j++) { 998 ip[j-fshift] = ip[j]; 999 ap[j-fshift] = ap[j]; 1000 } 1001 } 1002 ai[i] = ai[i-1] + ailen[i-1]; 1003 } 1004 if (m) { 1005 fshift += imax[m-1] - ailen[m-1]; 1006 ai[m] = ai[m-1] + ailen[m-1]; 1007 } 1008 1009 /* reset ilen and imax for each row */ 1010 a->nonzerorowcnt = 0; 1011 for (i=0; i<m; i++) { 1012 ailen[i] = imax[i] = ai[i+1] - ai[i]; 1013 a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0); 1014 } 1015 a->nz = ai[m]; 1016 if (fshift && a->nounused == -1) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_PLIB, "Unused space detected in matrix: %D X %D, %D unneeded", m, A->cmap->n, fshift); 1017 1018 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 1019 ierr = PetscInfo4(A,"Matrix size: %D X %D; storage space: %D unneeded,%D used\n",m,A->cmap->n,fshift,a->nz);CHKERRQ(ierr); 1020 ierr = PetscInfo1(A,"Number of mallocs during MatSetValues() is %D\n",a->reallocs);CHKERRQ(ierr); 1021 ierr = PetscInfo1(A,"Maximum nonzeros in any row is %D\n",rmax);CHKERRQ(ierr); 1022 1023 A->info.mallocs += a->reallocs; 1024 a->reallocs = 0; 1025 A->info.nz_unneeded = (PetscReal)fshift; 1026 a->rmax = rmax; 1027 1028 ierr = MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,m,ratio);CHKERRQ(ierr); 1029 ierr = MatAssemblyEnd_SeqAIJ_Inode(A,mode);CHKERRQ(ierr); 1030 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1031 PetscFunctionReturn(0); 1032 } 1033 1034 #undef __FUNCT__ 1035 #define __FUNCT__ "MatRealPart_SeqAIJ" 1036 PetscErrorCode MatRealPart_SeqAIJ(Mat A) 1037 { 1038 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1039 PetscInt i,nz = a->nz; 1040 MatScalar *aa = a->a; 1041 PetscErrorCode ierr; 1042 1043 PetscFunctionBegin; 1044 for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]); 1045 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1046 PetscFunctionReturn(0); 1047 } 1048 1049 #undef __FUNCT__ 1050 #define __FUNCT__ "MatImaginaryPart_SeqAIJ" 1051 PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A) 1052 { 1053 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1054 PetscInt i,nz = a->nz; 1055 MatScalar *aa = a->a; 1056 PetscErrorCode ierr; 1057 1058 PetscFunctionBegin; 1059 for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]); 1060 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1061 PetscFunctionReturn(0); 1062 } 1063 1064 #if defined(PETSC_THREADCOMM_ACTIVE) 1065 PetscErrorCode MatZeroEntries_SeqAIJ_Kernel(PetscInt thread_id,Mat A) 1066 { 1067 PetscErrorCode ierr; 1068 PetscInt *trstarts=A->rmap->trstarts; 1069 PetscInt n,start,end; 1070 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1071 1072 start = trstarts[thread_id]; 1073 end = trstarts[thread_id+1]; 1074 n = a->i[end] - a->i[start]; 1075 ierr = PetscMemzero(a->a+a->i[start],n*sizeof(PetscScalar));CHKERRQ(ierr); 1076 return 0; 1077 } 1078 1079 #undef __FUNCT__ 1080 #define __FUNCT__ "MatZeroEntries_SeqAIJ" 1081 PetscErrorCode MatZeroEntries_SeqAIJ(Mat A) 1082 { 1083 PetscErrorCode ierr; 1084 1085 PetscFunctionBegin; 1086 ierr = PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatZeroEntries_SeqAIJ_Kernel,1,A);CHKERRQ(ierr); 1087 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1088 PetscFunctionReturn(0); 1089 } 1090 #else 1091 #undef __FUNCT__ 1092 #define __FUNCT__ "MatZeroEntries_SeqAIJ" 1093 PetscErrorCode MatZeroEntries_SeqAIJ(Mat A) 1094 { 1095 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1096 PetscErrorCode ierr; 1097 1098 PetscFunctionBegin; 1099 ierr = PetscMemzero(a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));CHKERRQ(ierr); 1100 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1101 PetscFunctionReturn(0); 1102 } 1103 #endif 1104 1105 extern PetscErrorCode MatDestroy_Redundant(Mat_Redundant **); 1106 1107 #undef __FUNCT__ 1108 #define __FUNCT__ "MatDestroy_SeqAIJ" 1109 PetscErrorCode MatDestroy_SeqAIJ(Mat A) 1110 { 1111 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1112 PetscErrorCode ierr; 1113 1114 PetscFunctionBegin; 1115 #if defined(PETSC_USE_LOG) 1116 PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap->n,A->cmap->n,a->nz); 1117 #endif 1118 ierr = MatDestroy_Redundant(&a->redundant);CHKERRQ(ierr); 1119 ierr = MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);CHKERRQ(ierr); 1120 ierr = ISDestroy(&a->row);CHKERRQ(ierr); 1121 ierr = ISDestroy(&a->col);CHKERRQ(ierr); 1122 ierr = PetscFree(a->diag);CHKERRQ(ierr); 1123 ierr = PetscFree(a->ibdiag);CHKERRQ(ierr); 1124 ierr = PetscFree2(a->imax,a->ilen);CHKERRQ(ierr); 1125 ierr = PetscFree3(a->idiag,a->mdiag,a->ssor_work);CHKERRQ(ierr); 1126 ierr = PetscFree(a->solve_work);CHKERRQ(ierr); 1127 ierr = ISDestroy(&a->icol);CHKERRQ(ierr); 1128 ierr = PetscFree(a->saved_values);CHKERRQ(ierr); 1129 ierr = ISColoringDestroy(&a->coloring);CHKERRQ(ierr); 1130 ierr = PetscFree(a->xtoy);CHKERRQ(ierr); 1131 ierr = MatDestroy(&a->XtoY);CHKERRQ(ierr); 1132 ierr = PetscFree2(a->compressedrow.i,a->compressedrow.rindex);CHKERRQ(ierr); 1133 ierr = PetscFree(a->matmult_abdense);CHKERRQ(ierr); 1134 1135 ierr = MatDestroy_SeqAIJ_Inode(A);CHKERRQ(ierr); 1136 ierr = PetscFree(A->data);CHKERRQ(ierr); 1137 1138 ierr = PetscObjectChangeTypeName((PetscObject)A,0);CHKERRQ(ierr); 1139 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetColumnIndices_C",NULL);CHKERRQ(ierr); 1140 ierr = PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL);CHKERRQ(ierr); 1141 ierr = PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL);CHKERRQ(ierr); 1142 ierr = PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsbaij_C",NULL);CHKERRQ(ierr); 1143 ierr = PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqbaij_C",NULL);CHKERRQ(ierr); 1144 ierr = PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijperm_C",NULL);CHKERRQ(ierr); 1145 ierr = PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL);CHKERRQ(ierr); 1146 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocation_C",NULL);CHKERRQ(ierr); 1147 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C",NULL);CHKERRQ(ierr); 1148 ierr = PetscObjectComposeFunction((PetscObject)A,"MatReorderForNonzeroDiagonal_C",NULL);CHKERRQ(ierr); 1149 PetscFunctionReturn(0); 1150 } 1151 1152 #undef __FUNCT__ 1153 #define __FUNCT__ "MatSetOption_SeqAIJ" 1154 PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op,PetscBool flg) 1155 { 1156 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1157 PetscErrorCode ierr; 1158 1159 PetscFunctionBegin; 1160 switch (op) { 1161 case MAT_ROW_ORIENTED: 1162 a->roworiented = flg; 1163 break; 1164 case MAT_KEEP_NONZERO_PATTERN: 1165 a->keepnonzeropattern = flg; 1166 break; 1167 case MAT_NEW_NONZERO_LOCATIONS: 1168 a->nonew = (flg ? 0 : 1); 1169 break; 1170 case MAT_NEW_NONZERO_LOCATION_ERR: 1171 a->nonew = (flg ? -1 : 0); 1172 break; 1173 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1174 a->nonew = (flg ? -2 : 0); 1175 break; 1176 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1177 a->nounused = (flg ? -1 : 0); 1178 break; 1179 case MAT_IGNORE_ZERO_ENTRIES: 1180 a->ignorezeroentries = flg; 1181 break; 1182 case MAT_SPD: 1183 case MAT_SYMMETRIC: 1184 case MAT_STRUCTURALLY_SYMMETRIC: 1185 case MAT_HERMITIAN: 1186 case MAT_SYMMETRY_ETERNAL: 1187 /* These options are handled directly by MatSetOption() */ 1188 break; 1189 case MAT_NEW_DIAGONALS: 1190 case MAT_IGNORE_OFF_PROC_ENTRIES: 1191 case MAT_USE_HASH_TABLE: 1192 ierr = PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);CHKERRQ(ierr); 1193 break; 1194 case MAT_USE_INODES: 1195 /* Not an error because MatSetOption_SeqAIJ_Inode handles this one */ 1196 break; 1197 default: 1198 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op); 1199 } 1200 ierr = MatSetOption_SeqAIJ_Inode(A,op,flg);CHKERRQ(ierr); 1201 PetscFunctionReturn(0); 1202 } 1203 1204 #undef __FUNCT__ 1205 #define __FUNCT__ "MatGetDiagonal_SeqAIJ" 1206 PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v) 1207 { 1208 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1209 PetscErrorCode ierr; 1210 PetscInt i,j,n,*ai=a->i,*aj=a->j,nz; 1211 PetscScalar *aa=a->a,*x,zero=0.0; 1212 1213 PetscFunctionBegin; 1214 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 1215 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 1216 1217 if (A->factortype == MAT_FACTOR_ILU || A->factortype == MAT_FACTOR_LU) { 1218 PetscInt *diag=a->diag; 1219 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 1220 for (i=0; i<n; i++) x[i] = 1.0/aa[diag[i]]; 1221 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 1222 PetscFunctionReturn(0); 1223 } 1224 1225 ierr = VecSet(v,zero);CHKERRQ(ierr); 1226 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 1227 for (i=0; i<n; i++) { 1228 nz = ai[i+1] - ai[i]; 1229 if (!nz) x[i] = 0.0; 1230 for (j=ai[i]; j<ai[i+1]; j++) { 1231 if (aj[j] == i) { 1232 x[i] = aa[j]; 1233 break; 1234 } 1235 } 1236 } 1237 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 1238 PetscFunctionReturn(0); 1239 } 1240 1241 #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h> 1242 #undef __FUNCT__ 1243 #define __FUNCT__ "MatMultTransposeAdd_SeqAIJ" 1244 PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy) 1245 { 1246 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1247 PetscScalar *x,*y; 1248 PetscErrorCode ierr; 1249 PetscInt m = A->rmap->n; 1250 #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ) 1251 MatScalar *v; 1252 PetscScalar alpha; 1253 PetscInt n,i,j,*idx,*ii,*ridx=NULL; 1254 Mat_CompressedRow cprow = a->compressedrow; 1255 PetscBool usecprow = cprow.use; 1256 #endif 1257 1258 PetscFunctionBegin; 1259 if (zz != yy) {ierr = VecCopy(zz,yy);CHKERRQ(ierr);} 1260 ierr = VecGetArray(xx,&x);CHKERRQ(ierr); 1261 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1262 1263 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ) 1264 fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y); 1265 #else 1266 if (usecprow) { 1267 m = cprow.nrows; 1268 ii = cprow.i; 1269 ridx = cprow.rindex; 1270 } else { 1271 ii = a->i; 1272 } 1273 for (i=0; i<m; i++) { 1274 idx = a->j + ii[i]; 1275 v = a->a + ii[i]; 1276 n = ii[i+1] - ii[i]; 1277 if (usecprow) { 1278 alpha = x[ridx[i]]; 1279 } else { 1280 alpha = x[i]; 1281 } 1282 for (j=0; j<n; j++) y[idx[j]] += alpha*v[j]; 1283 } 1284 #endif 1285 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1286 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1287 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1288 PetscFunctionReturn(0); 1289 } 1290 1291 #undef __FUNCT__ 1292 #define __FUNCT__ "MatMultTranspose_SeqAIJ" 1293 PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy) 1294 { 1295 PetscErrorCode ierr; 1296 1297 PetscFunctionBegin; 1298 ierr = VecSet(yy,0.0);CHKERRQ(ierr); 1299 ierr = MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);CHKERRQ(ierr); 1300 PetscFunctionReturn(0); 1301 } 1302 1303 #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h> 1304 #if defined(PETSC_THREADCOMM_ACTIVE) 1305 PetscErrorCode MatMult_SeqAIJ_Kernel(PetscInt thread_id,Mat A,Vec xx,Vec yy) 1306 { 1307 PetscErrorCode ierr; 1308 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1309 PetscScalar *y; 1310 const PetscScalar *x; 1311 const MatScalar *aa; 1312 PetscInt *trstarts=A->rmap->trstarts; 1313 PetscInt n,start,end,i; 1314 const PetscInt *aj,*ai; 1315 PetscScalar sum; 1316 1317 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1318 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1319 start = trstarts[thread_id]; 1320 end = trstarts[thread_id+1]; 1321 aj = a->j; 1322 aa = a->a; 1323 ai = a->i; 1324 for (i=start; i<end; i++) { 1325 n = ai[i+1] - ai[i]; 1326 aj = a->j + ai[i]; 1327 aa = a->a + ai[i]; 1328 sum = 0.0; 1329 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1330 y[i] = sum; 1331 } 1332 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1333 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1334 return 0; 1335 } 1336 1337 #undef __FUNCT__ 1338 #define __FUNCT__ "MatMult_SeqAIJ" 1339 PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy) 1340 { 1341 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1342 PetscScalar *y; 1343 const PetscScalar *x; 1344 const MatScalar *aa; 1345 PetscErrorCode ierr; 1346 PetscInt m=A->rmap->n; 1347 const PetscInt *aj,*ii,*ridx=NULL; 1348 PetscInt n,i; 1349 PetscScalar sum; 1350 PetscBool usecprow=a->compressedrow.use; 1351 1352 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1353 #pragma disjoint(*x,*y,*aa) 1354 #endif 1355 1356 PetscFunctionBegin; 1357 aj = a->j; 1358 aa = a->a; 1359 ii = a->i; 1360 if (usecprow) { /* use compressed row format */ 1361 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1362 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1363 ierr = PetscMemzero(y,m*sizeof(PetscScalar));CHKERRQ(ierr); 1364 m = a->compressedrow.nrows; 1365 ii = a->compressedrow.i; 1366 ridx = a->compressedrow.rindex; 1367 for (i=0; i<m; i++) { 1368 n = ii[i+1] - ii[i]; 1369 aj = a->j + ii[i]; 1370 aa = a->a + ii[i]; 1371 sum = 0.0; 1372 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1373 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1374 y[*ridx++] = sum; 1375 } 1376 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1377 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1378 } else { /* do not use compressed row format */ 1379 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ) 1380 fortranmultaij_(&m,x,ii,aj,aa,y); 1381 #else 1382 ierr = PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatMult_SeqAIJ_Kernel,3,A,xx,yy);CHKERRQ(ierr); 1383 #endif 1384 } 1385 ierr = PetscLogFlops(2.0*a->nz - a->nonzerorowcnt);CHKERRQ(ierr); 1386 PetscFunctionReturn(0); 1387 } 1388 #else 1389 #undef __FUNCT__ 1390 #define __FUNCT__ "MatMult_SeqAIJ" 1391 PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy) 1392 { 1393 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1394 PetscScalar *y; 1395 const PetscScalar *x; 1396 const MatScalar *aa; 1397 PetscErrorCode ierr; 1398 PetscInt m=A->rmap->n; 1399 const PetscInt *aj,*ii,*ridx=NULL; 1400 PetscInt n,i; 1401 PetscScalar sum; 1402 PetscBool usecprow=a->compressedrow.use; 1403 1404 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1405 #pragma disjoint(*x,*y,*aa) 1406 #endif 1407 1408 PetscFunctionBegin; 1409 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1410 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1411 aj = a->j; 1412 aa = a->a; 1413 ii = a->i; 1414 if (usecprow) { /* use compressed row format */ 1415 ierr = PetscMemzero(y,m*sizeof(PetscScalar));CHKERRQ(ierr); 1416 m = a->compressedrow.nrows; 1417 ii = a->compressedrow.i; 1418 ridx = a->compressedrow.rindex; 1419 for (i=0; i<m; i++) { 1420 n = ii[i+1] - ii[i]; 1421 aj = a->j + ii[i]; 1422 aa = a->a + ii[i]; 1423 sum = 0.0; 1424 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1425 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1426 y[*ridx++] = sum; 1427 } 1428 } else { /* do not use compressed row format */ 1429 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ) 1430 fortranmultaij_(&m,x,ii,aj,aa,y); 1431 #else 1432 #if defined(PETSC_THREADCOMM_ACTIVE) 1433 ierr = PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatMult_SeqAIJ_Kernel,3,A,xx,yy);CHKERRQ(ierr); 1434 #else 1435 for (i=0; i<m; i++) { 1436 n = ii[i+1] - ii[i]; 1437 aj = a->j + ii[i]; 1438 aa = a->a + ii[i]; 1439 sum = 0.0; 1440 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1441 y[i] = sum; 1442 } 1443 #endif 1444 #endif 1445 } 1446 ierr = PetscLogFlops(2.0*a->nz - a->nonzerorowcnt);CHKERRQ(ierr); 1447 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1448 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1449 PetscFunctionReturn(0); 1450 } 1451 #endif 1452 1453 #undef __FUNCT__ 1454 #define __FUNCT__ "MatMultMax_SeqAIJ" 1455 PetscErrorCode MatMultMax_SeqAIJ(Mat A,Vec xx,Vec yy) 1456 { 1457 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1458 PetscScalar *y; 1459 const PetscScalar *x; 1460 const MatScalar *aa; 1461 PetscErrorCode ierr; 1462 PetscInt m=A->rmap->n; 1463 const PetscInt *aj,*ii,*ridx=NULL; 1464 PetscInt n,i,nonzerorow=0; 1465 PetscScalar sum; 1466 PetscBool usecprow=a->compressedrow.use; 1467 1468 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1469 #pragma disjoint(*x,*y,*aa) 1470 #endif 1471 1472 PetscFunctionBegin; 1473 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1474 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1475 aj = a->j; 1476 aa = a->a; 1477 ii = a->i; 1478 if (usecprow) { /* use compressed row format */ 1479 m = a->compressedrow.nrows; 1480 ii = a->compressedrow.i; 1481 ridx = a->compressedrow.rindex; 1482 for (i=0; i<m; i++) { 1483 n = ii[i+1] - ii[i]; 1484 aj = a->j + ii[i]; 1485 aa = a->a + ii[i]; 1486 sum = 0.0; 1487 nonzerorow += (n>0); 1488 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1489 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1490 y[*ridx++] = sum; 1491 } 1492 } else { /* do not use compressed row format */ 1493 for (i=0; i<m; i++) { 1494 n = ii[i+1] - ii[i]; 1495 aj = a->j + ii[i]; 1496 aa = a->a + ii[i]; 1497 sum = 0.0; 1498 nonzerorow += (n>0); 1499 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1500 y[i] = sum; 1501 } 1502 } 1503 ierr = PetscLogFlops(2.0*a->nz - nonzerorow);CHKERRQ(ierr); 1504 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1505 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1506 PetscFunctionReturn(0); 1507 } 1508 1509 #undef __FUNCT__ 1510 #define __FUNCT__ "MatMultAddMax_SeqAIJ" 1511 PetscErrorCode MatMultAddMax_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1512 { 1513 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1514 PetscScalar *y,*z; 1515 const PetscScalar *x; 1516 const MatScalar *aa; 1517 PetscErrorCode ierr; 1518 PetscInt m = A->rmap->n,*aj,*ii; 1519 PetscInt n,i,*ridx=NULL; 1520 PetscScalar sum; 1521 PetscBool usecprow=a->compressedrow.use; 1522 1523 PetscFunctionBegin; 1524 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1525 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1526 if (zz != yy) { 1527 ierr = VecGetArray(zz,&z);CHKERRQ(ierr); 1528 } else { 1529 z = y; 1530 } 1531 1532 aj = a->j; 1533 aa = a->a; 1534 ii = a->i; 1535 if (usecprow) { /* use compressed row format */ 1536 if (zz != yy) { 1537 ierr = PetscMemcpy(z,y,m*sizeof(PetscScalar));CHKERRQ(ierr); 1538 } 1539 m = a->compressedrow.nrows; 1540 ii = a->compressedrow.i; 1541 ridx = a->compressedrow.rindex; 1542 for (i=0; i<m; i++) { 1543 n = ii[i+1] - ii[i]; 1544 aj = a->j + ii[i]; 1545 aa = a->a + ii[i]; 1546 sum = y[*ridx]; 1547 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1548 z[*ridx++] = sum; 1549 } 1550 } else { /* do not use compressed row format */ 1551 for (i=0; i<m; i++) { 1552 n = ii[i+1] - ii[i]; 1553 aj = a->j + ii[i]; 1554 aa = a->a + ii[i]; 1555 sum = y[i]; 1556 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1557 z[i] = sum; 1558 } 1559 } 1560 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1561 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1562 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1563 if (zz != yy) { 1564 ierr = VecRestoreArray(zz,&z);CHKERRQ(ierr); 1565 } 1566 PetscFunctionReturn(0); 1567 } 1568 1569 #include <../src/mat/impls/aij/seq/ftn-kernels/fmultadd.h> 1570 #undef __FUNCT__ 1571 #define __FUNCT__ "MatMultAdd_SeqAIJ" 1572 PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1573 { 1574 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1575 PetscScalar *y,*z; 1576 const PetscScalar *x; 1577 const MatScalar *aa; 1578 PetscErrorCode ierr; 1579 PetscInt m = A->rmap->n,*aj,*ii; 1580 PetscInt n,i,*ridx=NULL; 1581 PetscScalar sum; 1582 PetscBool usecprow=a->compressedrow.use; 1583 1584 PetscFunctionBegin; 1585 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1586 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1587 if (zz != yy) { 1588 ierr = VecGetArray(zz,&z);CHKERRQ(ierr); 1589 } else { 1590 z = y; 1591 } 1592 1593 aj = a->j; 1594 aa = a->a; 1595 ii = a->i; 1596 if (usecprow) { /* use compressed row format */ 1597 if (zz != yy) { 1598 ierr = PetscMemcpy(z,y,m*sizeof(PetscScalar));CHKERRQ(ierr); 1599 } 1600 m = a->compressedrow.nrows; 1601 ii = a->compressedrow.i; 1602 ridx = a->compressedrow.rindex; 1603 for (i=0; i<m; i++) { 1604 n = ii[i+1] - ii[i]; 1605 aj = a->j + ii[i]; 1606 aa = a->a + ii[i]; 1607 sum = y[*ridx]; 1608 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1609 z[*ridx++] = sum; 1610 } 1611 } else { /* do not use compressed row format */ 1612 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ) 1613 fortranmultaddaij_(&m,x,ii,aj,aa,y,z); 1614 #else 1615 for (i=0; i<m; i++) { 1616 n = ii[i+1] - ii[i]; 1617 aj = a->j + ii[i]; 1618 aa = a->a + ii[i]; 1619 sum = y[i]; 1620 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1621 z[i] = sum; 1622 } 1623 #endif 1624 } 1625 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1626 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1627 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1628 if (zz != yy) { 1629 ierr = VecRestoreArray(zz,&z);CHKERRQ(ierr); 1630 } 1631 #if defined(PETSC_HAVE_CUSP) 1632 /* 1633 ierr = VecView(xx,0);CHKERRQ(ierr); 1634 ierr = VecView(zz,0);CHKERRQ(ierr); 1635 ierr = MatView(A,0);CHKERRQ(ierr); 1636 */ 1637 #endif 1638 PetscFunctionReturn(0); 1639 } 1640 1641 /* 1642 Adds diagonal pointers to sparse matrix structure. 1643 */ 1644 #undef __FUNCT__ 1645 #define __FUNCT__ "MatMarkDiagonal_SeqAIJ" 1646 PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A) 1647 { 1648 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1649 PetscErrorCode ierr; 1650 PetscInt i,j,m = A->rmap->n; 1651 1652 PetscFunctionBegin; 1653 if (!a->diag) { 1654 ierr = PetscMalloc1(m,&a->diag);CHKERRQ(ierr); 1655 ierr = PetscLogObjectMemory((PetscObject)A, m*sizeof(PetscInt));CHKERRQ(ierr); 1656 } 1657 for (i=0; i<A->rmap->n; i++) { 1658 a->diag[i] = a->i[i+1]; 1659 for (j=a->i[i]; j<a->i[i+1]; j++) { 1660 if (a->j[j] == i) { 1661 a->diag[i] = j; 1662 break; 1663 } 1664 } 1665 } 1666 PetscFunctionReturn(0); 1667 } 1668 1669 /* 1670 Checks for missing diagonals 1671 */ 1672 #undef __FUNCT__ 1673 #define __FUNCT__ "MatMissingDiagonal_SeqAIJ" 1674 PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscBool *missing,PetscInt *d) 1675 { 1676 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1677 PetscInt *diag,*ii = a->i,i; 1678 1679 PetscFunctionBegin; 1680 *missing = PETSC_FALSE; 1681 if (A->rmap->n > 0 && !ii) { 1682 *missing = PETSC_TRUE; 1683 if (d) *d = 0; 1684 PetscInfo(A,"Matrix has no entries therefore is missing diagonal"); 1685 } else { 1686 diag = a->diag; 1687 for (i=0; i<A->rmap->n; i++) { 1688 if (diag[i] >= ii[i+1]) { 1689 *missing = PETSC_TRUE; 1690 if (d) *d = i; 1691 PetscInfo1(A,"Matrix is missing diagonal number %D",i); 1692 break; 1693 } 1694 } 1695 } 1696 PetscFunctionReturn(0); 1697 } 1698 1699 #undef __FUNCT__ 1700 #define __FUNCT__ "MatInvertDiagonal_SeqAIJ" 1701 PetscErrorCode MatInvertDiagonal_SeqAIJ(Mat A,PetscScalar omega,PetscScalar fshift) 1702 { 1703 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 1704 PetscErrorCode ierr; 1705 PetscInt i,*diag,m = A->rmap->n; 1706 MatScalar *v = a->a; 1707 PetscScalar *idiag,*mdiag; 1708 1709 PetscFunctionBegin; 1710 if (a->idiagvalid) PetscFunctionReturn(0); 1711 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 1712 diag = a->diag; 1713 if (!a->idiag) { 1714 ierr = PetscMalloc3(m,&a->idiag,m,&a->mdiag,m,&a->ssor_work);CHKERRQ(ierr); 1715 ierr = PetscLogObjectMemory((PetscObject)A, 3*m*sizeof(PetscScalar));CHKERRQ(ierr); 1716 v = a->a; 1717 } 1718 mdiag = a->mdiag; 1719 idiag = a->idiag; 1720 1721 if (omega == 1.0 && !PetscAbsScalar(fshift)) { 1722 for (i=0; i<m; i++) { 1723 mdiag[i] = v[diag[i]]; 1724 if (!PetscAbsScalar(mdiag[i])) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %D",i); 1725 idiag[i] = 1.0/v[diag[i]]; 1726 } 1727 ierr = PetscLogFlops(m);CHKERRQ(ierr); 1728 } else { 1729 for (i=0; i<m; i++) { 1730 mdiag[i] = v[diag[i]]; 1731 idiag[i] = omega/(fshift + v[diag[i]]); 1732 } 1733 ierr = PetscLogFlops(2.0*m);CHKERRQ(ierr); 1734 } 1735 a->idiagvalid = PETSC_TRUE; 1736 PetscFunctionReturn(0); 1737 } 1738 1739 #include <../src/mat/impls/aij/seq/ftn-kernels/frelax.h> 1740 #undef __FUNCT__ 1741 #define __FUNCT__ "MatSOR_SeqAIJ" 1742 PetscErrorCode MatSOR_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1743 { 1744 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1745 PetscScalar *x,d,sum,*t,scale; 1746 const MatScalar *v = a->a,*idiag=0,*mdiag; 1747 const PetscScalar *b, *bs,*xb, *ts; 1748 PetscErrorCode ierr; 1749 PetscInt n = A->cmap->n,m = A->rmap->n,i; 1750 const PetscInt *idx,*diag; 1751 1752 PetscFunctionBegin; 1753 its = its*lits; 1754 1755 if (fshift != a->fshift || omega != a->omega) a->idiagvalid = PETSC_FALSE; /* must recompute idiag[] */ 1756 if (!a->idiagvalid) {ierr = MatInvertDiagonal_SeqAIJ(A,omega,fshift);CHKERRQ(ierr);} 1757 a->fshift = fshift; 1758 a->omega = omega; 1759 1760 diag = a->diag; 1761 t = a->ssor_work; 1762 idiag = a->idiag; 1763 mdiag = a->mdiag; 1764 1765 ierr = VecGetArray(xx,&x);CHKERRQ(ierr); 1766 ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr); 1767 /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */ 1768 if (flag == SOR_APPLY_UPPER) { 1769 /* apply (U + D/omega) to the vector */ 1770 bs = b; 1771 for (i=0; i<m; i++) { 1772 d = fshift + mdiag[i]; 1773 n = a->i[i+1] - diag[i] - 1; 1774 idx = a->j + diag[i] + 1; 1775 v = a->a + diag[i] + 1; 1776 sum = b[i]*d/omega; 1777 PetscSparseDensePlusDot(sum,bs,v,idx,n); 1778 x[i] = sum; 1779 } 1780 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1781 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1782 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 1783 PetscFunctionReturn(0); 1784 } 1785 1786 if (flag == SOR_APPLY_LOWER) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented"); 1787 else if (flag & SOR_EISENSTAT) { 1788 /* Let A = L + U + D; where L is lower trianglar, 1789 U is upper triangular, E = D/omega; This routine applies 1790 1791 (L + E)^{-1} A (U + E)^{-1} 1792 1793 to a vector efficiently using Eisenstat's trick. 1794 */ 1795 scale = (2.0/omega) - 1.0; 1796 1797 /* x = (E + U)^{-1} b */ 1798 for (i=m-1; i>=0; i--) { 1799 n = a->i[i+1] - diag[i] - 1; 1800 idx = a->j + diag[i] + 1; 1801 v = a->a + diag[i] + 1; 1802 sum = b[i]; 1803 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1804 x[i] = sum*idiag[i]; 1805 } 1806 1807 /* t = b - (2*E - D)x */ 1808 v = a->a; 1809 for (i=0; i<m; i++) t[i] = b[i] - scale*(v[*diag++])*x[i]; 1810 1811 /* t = (E + L)^{-1}t */ 1812 ts = t; 1813 diag = a->diag; 1814 for (i=0; i<m; i++) { 1815 n = diag[i] - a->i[i]; 1816 idx = a->j + a->i[i]; 1817 v = a->a + a->i[i]; 1818 sum = t[i]; 1819 PetscSparseDenseMinusDot(sum,ts,v,idx,n); 1820 t[i] = sum*idiag[i]; 1821 /* x = x + t */ 1822 x[i] += t[i]; 1823 } 1824 1825 ierr = PetscLogFlops(6.0*m-1 + 2.0*a->nz);CHKERRQ(ierr); 1826 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1827 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1828 PetscFunctionReturn(0); 1829 } 1830 if (flag & SOR_ZERO_INITIAL_GUESS) { 1831 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 1832 for (i=0; i<m; i++) { 1833 n = diag[i] - a->i[i]; 1834 idx = a->j + a->i[i]; 1835 v = a->a + a->i[i]; 1836 sum = b[i]; 1837 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1838 t[i] = sum; 1839 x[i] = sum*idiag[i]; 1840 } 1841 xb = t; 1842 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 1843 } else xb = b; 1844 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 1845 for (i=m-1; i>=0; i--) { 1846 n = a->i[i+1] - diag[i] - 1; 1847 idx = a->j + diag[i] + 1; 1848 v = a->a + diag[i] + 1; 1849 sum = xb[i]; 1850 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1851 if (xb == b) { 1852 x[i] = sum*idiag[i]; 1853 } else { 1854 x[i] = (1-omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 1855 } 1856 } 1857 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); /* assumes 1/2 in upper */ 1858 } 1859 its--; 1860 } 1861 while (its--) { 1862 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 1863 for (i=0; i<m; i++) { 1864 /* lower */ 1865 n = diag[i] - a->i[i]; 1866 idx = a->j + a->i[i]; 1867 v = a->a + a->i[i]; 1868 sum = b[i]; 1869 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1870 t[i] = sum; /* save application of the lower-triangular part */ 1871 /* upper */ 1872 n = a->i[i+1] - diag[i] - 1; 1873 idx = a->j + diag[i] + 1; 1874 v = a->a + diag[i] + 1; 1875 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1876 x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 1877 } 1878 xb = t; 1879 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1880 } else xb = b; 1881 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 1882 for (i=m-1; i>=0; i--) { 1883 sum = xb[i]; 1884 if (xb == b) { 1885 /* whole matrix (no checkpointing available) */ 1886 n = a->i[i+1] - a->i[i]; 1887 idx = a->j + a->i[i]; 1888 v = a->a + a->i[i]; 1889 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1890 x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i]; 1891 } else { /* lower-triangular part has been saved, so only apply upper-triangular */ 1892 n = a->i[i+1] - diag[i] - 1; 1893 idx = a->j + diag[i] + 1; 1894 v = a->a + diag[i] + 1; 1895 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1896 x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 1897 } 1898 } 1899 if (xb == b) { 1900 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1901 } else { 1902 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); /* assumes 1/2 in upper */ 1903 } 1904 } 1905 } 1906 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1907 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1908 PetscFunctionReturn(0); 1909 } 1910 1911 1912 #undef __FUNCT__ 1913 #define __FUNCT__ "MatGetInfo_SeqAIJ" 1914 PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info) 1915 { 1916 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1917 1918 PetscFunctionBegin; 1919 info->block_size = 1.0; 1920 info->nz_allocated = (double)a->maxnz; 1921 info->nz_used = (double)a->nz; 1922 info->nz_unneeded = (double)(a->maxnz - a->nz); 1923 info->assemblies = (double)A->num_ass; 1924 info->mallocs = (double)A->info.mallocs; 1925 info->memory = ((PetscObject)A)->mem; 1926 if (A->factortype) { 1927 info->fill_ratio_given = A->info.fill_ratio_given; 1928 info->fill_ratio_needed = A->info.fill_ratio_needed; 1929 info->factor_mallocs = A->info.factor_mallocs; 1930 } else { 1931 info->fill_ratio_given = 0; 1932 info->fill_ratio_needed = 0; 1933 info->factor_mallocs = 0; 1934 } 1935 PetscFunctionReturn(0); 1936 } 1937 1938 #undef __FUNCT__ 1939 #define __FUNCT__ "MatZeroRows_SeqAIJ" 1940 PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 1941 { 1942 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1943 PetscInt i,m = A->rmap->n - 1,d = 0; 1944 PetscErrorCode ierr; 1945 const PetscScalar *xx; 1946 PetscScalar *bb; 1947 PetscBool missing; 1948 1949 PetscFunctionBegin; 1950 if (x && b) { 1951 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 1952 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 1953 for (i=0; i<N; i++) { 1954 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1955 bb[rows[i]] = diag*xx[rows[i]]; 1956 } 1957 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 1958 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 1959 } 1960 1961 if (a->keepnonzeropattern) { 1962 for (i=0; i<N; i++) { 1963 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1964 ierr = PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));CHKERRQ(ierr); 1965 } 1966 if (diag != 0.0) { 1967 ierr = MatMissingDiagonal_SeqAIJ(A,&missing,&d);CHKERRQ(ierr); 1968 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d); 1969 for (i=0; i<N; i++) { 1970 a->a[a->diag[rows[i]]] = diag; 1971 } 1972 } 1973 } else { 1974 if (diag != 0.0) { 1975 for (i=0; i<N; i++) { 1976 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1977 if (a->ilen[rows[i]] > 0) { 1978 a->ilen[rows[i]] = 1; 1979 a->a[a->i[rows[i]]] = diag; 1980 a->j[a->i[rows[i]]] = rows[i]; 1981 } else { /* in case row was completely empty */ 1982 ierr = MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);CHKERRQ(ierr); 1983 } 1984 } 1985 } else { 1986 for (i=0; i<N; i++) { 1987 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1988 a->ilen[rows[i]] = 0; 1989 } 1990 } 1991 A->nonzerostate++; 1992 } 1993 ierr = MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1994 PetscFunctionReturn(0); 1995 } 1996 1997 #undef __FUNCT__ 1998 #define __FUNCT__ "MatZeroRowsColumns_SeqAIJ" 1999 PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 2000 { 2001 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2002 PetscInt i,j,m = A->rmap->n - 1,d = 0; 2003 PetscErrorCode ierr; 2004 PetscBool missing,*zeroed,vecs = PETSC_FALSE; 2005 const PetscScalar *xx; 2006 PetscScalar *bb; 2007 2008 PetscFunctionBegin; 2009 if (x && b) { 2010 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 2011 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 2012 vecs = PETSC_TRUE; 2013 } 2014 ierr = PetscCalloc1(A->rmap->n,&zeroed);CHKERRQ(ierr); 2015 for (i=0; i<N; i++) { 2016 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 2017 ierr = PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));CHKERRQ(ierr); 2018 2019 zeroed[rows[i]] = PETSC_TRUE; 2020 } 2021 for (i=0; i<A->rmap->n; i++) { 2022 if (!zeroed[i]) { 2023 for (j=a->i[i]; j<a->i[i+1]; j++) { 2024 if (zeroed[a->j[j]]) { 2025 if (vecs) bb[i] -= a->a[j]*xx[a->j[j]]; 2026 a->a[j] = 0.0; 2027 } 2028 } 2029 } else if (vecs) bb[i] = diag*xx[i]; 2030 } 2031 if (x && b) { 2032 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 2033 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 2034 } 2035 ierr = PetscFree(zeroed);CHKERRQ(ierr); 2036 if (diag != 0.0) { 2037 ierr = MatMissingDiagonal_SeqAIJ(A,&missing,&d);CHKERRQ(ierr); 2038 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d); 2039 for (i=0; i<N; i++) { 2040 a->a[a->diag[rows[i]]] = diag; 2041 } 2042 } 2043 ierr = MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2044 PetscFunctionReturn(0); 2045 } 2046 2047 #undef __FUNCT__ 2048 #define __FUNCT__ "MatGetRow_SeqAIJ" 2049 PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 2050 { 2051 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2052 PetscInt *itmp; 2053 2054 PetscFunctionBegin; 2055 if (row < 0 || row >= A->rmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row); 2056 2057 *nz = a->i[row+1] - a->i[row]; 2058 if (v) *v = a->a + a->i[row]; 2059 if (idx) { 2060 itmp = a->j + a->i[row]; 2061 if (*nz) *idx = itmp; 2062 else *idx = 0; 2063 } 2064 PetscFunctionReturn(0); 2065 } 2066 2067 /* remove this function? */ 2068 #undef __FUNCT__ 2069 #define __FUNCT__ "MatRestoreRow_SeqAIJ" 2070 PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 2071 { 2072 PetscFunctionBegin; 2073 PetscFunctionReturn(0); 2074 } 2075 2076 #undef __FUNCT__ 2077 #define __FUNCT__ "MatNorm_SeqAIJ" 2078 PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm) 2079 { 2080 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2081 MatScalar *v = a->a; 2082 PetscReal sum = 0.0; 2083 PetscErrorCode ierr; 2084 PetscInt i,j; 2085 2086 PetscFunctionBegin; 2087 if (type == NORM_FROBENIUS) { 2088 for (i=0; i<a->nz; i++) { 2089 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 2090 } 2091 *nrm = PetscSqrtReal(sum); 2092 } else if (type == NORM_1) { 2093 PetscReal *tmp; 2094 PetscInt *jj = a->j; 2095 ierr = PetscCalloc1(A->cmap->n+1,&tmp);CHKERRQ(ierr); 2096 *nrm = 0.0; 2097 for (j=0; j<a->nz; j++) { 2098 tmp[*jj++] += PetscAbsScalar(*v); v++; 2099 } 2100 for (j=0; j<A->cmap->n; j++) { 2101 if (tmp[j] > *nrm) *nrm = tmp[j]; 2102 } 2103 ierr = PetscFree(tmp);CHKERRQ(ierr); 2104 } else if (type == NORM_INFINITY) { 2105 *nrm = 0.0; 2106 for (j=0; j<A->rmap->n; j++) { 2107 v = a->a + a->i[j]; 2108 sum = 0.0; 2109 for (i=0; i<a->i[j+1]-a->i[j]; i++) { 2110 sum += PetscAbsScalar(*v); v++; 2111 } 2112 if (sum > *nrm) *nrm = sum; 2113 } 2114 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm"); 2115 PetscFunctionReturn(0); 2116 } 2117 2118 /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */ 2119 #undef __FUNCT__ 2120 #define __FUNCT__ "MatTransposeSymbolic_SeqAIJ" 2121 PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B) 2122 { 2123 PetscErrorCode ierr; 2124 PetscInt i,j,anzj; 2125 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b; 2126 PetscInt an=A->cmap->N,am=A->rmap->N; 2127 PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j; 2128 2129 PetscFunctionBegin; 2130 /* Allocate space for symbolic transpose info and work array */ 2131 ierr = PetscCalloc1((an+1),&ati);CHKERRQ(ierr); 2132 ierr = PetscMalloc1(ai[am],&atj);CHKERRQ(ierr); 2133 ierr = PetscMalloc1(an,&atfill);CHKERRQ(ierr); 2134 2135 /* Walk through aj and count ## of non-zeros in each row of A^T. */ 2136 /* Note: offset by 1 for fast conversion into csr format. */ 2137 for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1; 2138 /* Form ati for csr format of A^T. */ 2139 for (i=0;i<an;i++) ati[i+1] += ati[i]; 2140 2141 /* Copy ati into atfill so we have locations of the next free space in atj */ 2142 ierr = PetscMemcpy(atfill,ati,an*sizeof(PetscInt));CHKERRQ(ierr); 2143 2144 /* Walk through A row-wise and mark nonzero entries of A^T. */ 2145 for (i=0;i<am;i++) { 2146 anzj = ai[i+1] - ai[i]; 2147 for (j=0;j<anzj;j++) { 2148 atj[atfill[*aj]] = i; 2149 atfill[*aj++] += 1; 2150 } 2151 } 2152 2153 /* Clean up temporary space and complete requests. */ 2154 ierr = PetscFree(atfill);CHKERRQ(ierr); 2155 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B);CHKERRQ(ierr); 2156 ierr = MatSetBlockSizes(*B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));CHKERRQ(ierr); 2157 2158 b = (Mat_SeqAIJ*)((*B)->data); 2159 b->free_a = PETSC_FALSE; 2160 b->free_ij = PETSC_TRUE; 2161 b->nonew = 0; 2162 PetscFunctionReturn(0); 2163 } 2164 2165 #undef __FUNCT__ 2166 #define __FUNCT__ "MatTranspose_SeqAIJ" 2167 PetscErrorCode MatTranspose_SeqAIJ(Mat A,MatReuse reuse,Mat *B) 2168 { 2169 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2170 Mat C; 2171 PetscErrorCode ierr; 2172 PetscInt i,*aj = a->j,*ai = a->i,m = A->rmap->n,len,*col; 2173 MatScalar *array = a->a; 2174 2175 PetscFunctionBegin; 2176 if (reuse == MAT_REUSE_MATRIX && A == *B && m != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Square matrix only for in-place"); 2177 2178 if (reuse == MAT_INITIAL_MATRIX || *B == A) { 2179 ierr = PetscCalloc1((1+A->cmap->n),&col);CHKERRQ(ierr); 2180 2181 for (i=0; i<ai[m]; i++) col[aj[i]] += 1; 2182 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2183 ierr = MatSetSizes(C,A->cmap->n,m,A->cmap->n,m);CHKERRQ(ierr); 2184 ierr = MatSetBlockSizes(C,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));CHKERRQ(ierr); 2185 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2186 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,col);CHKERRQ(ierr); 2187 ierr = PetscFree(col);CHKERRQ(ierr); 2188 } else { 2189 C = *B; 2190 } 2191 2192 for (i=0; i<m; i++) { 2193 len = ai[i+1]-ai[i]; 2194 ierr = MatSetValues_SeqAIJ(C,len,aj,1,&i,array,INSERT_VALUES);CHKERRQ(ierr); 2195 array += len; 2196 aj += len; 2197 } 2198 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2199 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2200 2201 if (reuse == MAT_INITIAL_MATRIX || *B != A) { 2202 *B = C; 2203 } else { 2204 ierr = MatHeaderMerge(A,C);CHKERRQ(ierr); 2205 } 2206 PetscFunctionReturn(0); 2207 } 2208 2209 #undef __FUNCT__ 2210 #define __FUNCT__ "MatIsTranspose_SeqAIJ" 2211 PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2212 { 2213 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data; 2214 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2215 MatScalar *va,*vb; 2216 PetscErrorCode ierr; 2217 PetscInt ma,na,mb,nb, i; 2218 2219 PetscFunctionBegin; 2220 bij = (Mat_SeqAIJ*) B->data; 2221 2222 ierr = MatGetSize(A,&ma,&na);CHKERRQ(ierr); 2223 ierr = MatGetSize(B,&mb,&nb);CHKERRQ(ierr); 2224 if (ma!=nb || na!=mb) { 2225 *f = PETSC_FALSE; 2226 PetscFunctionReturn(0); 2227 } 2228 aii = aij->i; bii = bij->i; 2229 adx = aij->j; bdx = bij->j; 2230 va = aij->a; vb = bij->a; 2231 ierr = PetscMalloc1(ma,&aptr);CHKERRQ(ierr); 2232 ierr = PetscMalloc1(mb,&bptr);CHKERRQ(ierr); 2233 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2234 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2235 2236 *f = PETSC_TRUE; 2237 for (i=0; i<ma; i++) { 2238 while (aptr[i]<aii[i+1]) { 2239 PetscInt idc,idr; 2240 PetscScalar vc,vr; 2241 /* column/row index/value */ 2242 idc = adx[aptr[i]]; 2243 idr = bdx[bptr[idc]]; 2244 vc = va[aptr[i]]; 2245 vr = vb[bptr[idc]]; 2246 if (i!=idr || PetscAbsScalar(vc-vr) > tol) { 2247 *f = PETSC_FALSE; 2248 goto done; 2249 } else { 2250 aptr[i]++; 2251 if (B || i!=idc) bptr[idc]++; 2252 } 2253 } 2254 } 2255 done: 2256 ierr = PetscFree(aptr);CHKERRQ(ierr); 2257 ierr = PetscFree(bptr);CHKERRQ(ierr); 2258 PetscFunctionReturn(0); 2259 } 2260 2261 #undef __FUNCT__ 2262 #define __FUNCT__ "MatIsHermitianTranspose_SeqAIJ" 2263 PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2264 { 2265 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data; 2266 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2267 MatScalar *va,*vb; 2268 PetscErrorCode ierr; 2269 PetscInt ma,na,mb,nb, i; 2270 2271 PetscFunctionBegin; 2272 bij = (Mat_SeqAIJ*) B->data; 2273 2274 ierr = MatGetSize(A,&ma,&na);CHKERRQ(ierr); 2275 ierr = MatGetSize(B,&mb,&nb);CHKERRQ(ierr); 2276 if (ma!=nb || na!=mb) { 2277 *f = PETSC_FALSE; 2278 PetscFunctionReturn(0); 2279 } 2280 aii = aij->i; bii = bij->i; 2281 adx = aij->j; bdx = bij->j; 2282 va = aij->a; vb = bij->a; 2283 ierr = PetscMalloc1(ma,&aptr);CHKERRQ(ierr); 2284 ierr = PetscMalloc1(mb,&bptr);CHKERRQ(ierr); 2285 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2286 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2287 2288 *f = PETSC_TRUE; 2289 for (i=0; i<ma; i++) { 2290 while (aptr[i]<aii[i+1]) { 2291 PetscInt idc,idr; 2292 PetscScalar vc,vr; 2293 /* column/row index/value */ 2294 idc = adx[aptr[i]]; 2295 idr = bdx[bptr[idc]]; 2296 vc = va[aptr[i]]; 2297 vr = vb[bptr[idc]]; 2298 if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) { 2299 *f = PETSC_FALSE; 2300 goto done; 2301 } else { 2302 aptr[i]++; 2303 if (B || i!=idc) bptr[idc]++; 2304 } 2305 } 2306 } 2307 done: 2308 ierr = PetscFree(aptr);CHKERRQ(ierr); 2309 ierr = PetscFree(bptr);CHKERRQ(ierr); 2310 PetscFunctionReturn(0); 2311 } 2312 2313 #undef __FUNCT__ 2314 #define __FUNCT__ "MatIsSymmetric_SeqAIJ" 2315 PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2316 { 2317 PetscErrorCode ierr; 2318 2319 PetscFunctionBegin; 2320 ierr = MatIsTranspose_SeqAIJ(A,A,tol,f);CHKERRQ(ierr); 2321 PetscFunctionReturn(0); 2322 } 2323 2324 #undef __FUNCT__ 2325 #define __FUNCT__ "MatIsHermitian_SeqAIJ" 2326 PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2327 { 2328 PetscErrorCode ierr; 2329 2330 PetscFunctionBegin; 2331 ierr = MatIsHermitianTranspose_SeqAIJ(A,A,tol,f);CHKERRQ(ierr); 2332 PetscFunctionReturn(0); 2333 } 2334 2335 #undef __FUNCT__ 2336 #define __FUNCT__ "MatDiagonalScale_SeqAIJ" 2337 PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr) 2338 { 2339 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2340 PetscScalar *l,*r,x; 2341 MatScalar *v; 2342 PetscErrorCode ierr; 2343 PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz,*jj; 2344 2345 PetscFunctionBegin; 2346 if (ll) { 2347 /* The local size is used so that VecMPI can be passed to this routine 2348 by MatDiagonalScale_MPIAIJ */ 2349 ierr = VecGetLocalSize(ll,&m);CHKERRQ(ierr); 2350 if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length"); 2351 ierr = VecGetArray(ll,&l);CHKERRQ(ierr); 2352 v = a->a; 2353 for (i=0; i<m; i++) { 2354 x = l[i]; 2355 M = a->i[i+1] - a->i[i]; 2356 for (j=0; j<M; j++) (*v++) *= x; 2357 } 2358 ierr = VecRestoreArray(ll,&l);CHKERRQ(ierr); 2359 ierr = PetscLogFlops(nz);CHKERRQ(ierr); 2360 } 2361 if (rr) { 2362 ierr = VecGetLocalSize(rr,&n);CHKERRQ(ierr); 2363 if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length"); 2364 ierr = VecGetArray(rr,&r);CHKERRQ(ierr); 2365 v = a->a; jj = a->j; 2366 for (i=0; i<nz; i++) (*v++) *= r[*jj++]; 2367 ierr = VecRestoreArray(rr,&r);CHKERRQ(ierr); 2368 ierr = PetscLogFlops(nz);CHKERRQ(ierr); 2369 } 2370 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 2371 PetscFunctionReturn(0); 2372 } 2373 2374 #undef __FUNCT__ 2375 #define __FUNCT__ "MatGetSubMatrix_SeqAIJ" 2376 PetscErrorCode MatGetSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B) 2377 { 2378 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c; 2379 PetscErrorCode ierr; 2380 PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens; 2381 PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi; 2382 const PetscInt *irow,*icol; 2383 PetscInt nrows,ncols; 2384 PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen; 2385 MatScalar *a_new,*mat_a; 2386 Mat C; 2387 PetscBool stride,sorted; 2388 2389 PetscFunctionBegin; 2390 ierr = ISSorted(isrow,&sorted);CHKERRQ(ierr); 2391 if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"ISrow is not sorted"); 2392 ierr = ISSorted(iscol,&sorted);CHKERRQ(ierr); 2393 if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"IScol is not sorted"); 2394 2395 ierr = ISGetIndices(isrow,&irow);CHKERRQ(ierr); 2396 ierr = ISGetLocalSize(isrow,&nrows);CHKERRQ(ierr); 2397 ierr = ISGetLocalSize(iscol,&ncols);CHKERRQ(ierr); 2398 2399 ierr = PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride);CHKERRQ(ierr); 2400 if (stride) { 2401 ierr = ISStrideGetInfo(iscol,&first,&step);CHKERRQ(ierr); 2402 } else { 2403 first = 0; 2404 step = 0; 2405 } 2406 if (stride && step == 1) { 2407 /* special case of contiguous rows */ 2408 ierr = PetscMalloc2(nrows,&lens,nrows,&starts);CHKERRQ(ierr); 2409 /* loop over new rows determining lens and starting points */ 2410 for (i=0; i<nrows; i++) { 2411 kstart = ai[irow[i]]; 2412 kend = kstart + ailen[irow[i]]; 2413 for (k=kstart; k<kend; k++) { 2414 if (aj[k] >= first) { 2415 starts[i] = k; 2416 break; 2417 } 2418 } 2419 sum = 0; 2420 while (k < kend) { 2421 if (aj[k++] >= first+ncols) break; 2422 sum++; 2423 } 2424 lens[i] = sum; 2425 } 2426 /* create submatrix */ 2427 if (scall == MAT_REUSE_MATRIX) { 2428 PetscInt n_cols,n_rows; 2429 ierr = MatGetSize(*B,&n_rows,&n_cols);CHKERRQ(ierr); 2430 if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size"); 2431 ierr = MatZeroEntries(*B);CHKERRQ(ierr); 2432 C = *B; 2433 } else { 2434 PetscInt rbs,cbs; 2435 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2436 ierr = MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2437 ierr = ISGetBlockSize(isrow,&rbs);CHKERRQ(ierr); 2438 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2439 ierr = MatSetBlockSizes(C,rbs,cbs);CHKERRQ(ierr); 2440 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2441 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);CHKERRQ(ierr); 2442 } 2443 c = (Mat_SeqAIJ*)C->data; 2444 2445 /* loop over rows inserting into submatrix */ 2446 a_new = c->a; 2447 j_new = c->j; 2448 i_new = c->i; 2449 2450 for (i=0; i<nrows; i++) { 2451 ii = starts[i]; 2452 lensi = lens[i]; 2453 for (k=0; k<lensi; k++) { 2454 *j_new++ = aj[ii+k] - first; 2455 } 2456 ierr = PetscMemcpy(a_new,a->a + starts[i],lensi*sizeof(PetscScalar));CHKERRQ(ierr); 2457 a_new += lensi; 2458 i_new[i+1] = i_new[i] + lensi; 2459 c->ilen[i] = lensi; 2460 } 2461 ierr = PetscFree2(lens,starts);CHKERRQ(ierr); 2462 } else { 2463 ierr = ISGetIndices(iscol,&icol);CHKERRQ(ierr); 2464 ierr = PetscCalloc1(oldcols,&smap);CHKERRQ(ierr); 2465 ierr = PetscMalloc1((1+nrows),&lens);CHKERRQ(ierr); 2466 for (i=0; i<ncols; i++) { 2467 #if defined(PETSC_USE_DEBUG) 2468 if (icol[i] >= oldcols) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Requesting column beyond largest column icol[%D] %D <= A->cmap->n %D",i,icol[i],oldcols); 2469 #endif 2470 smap[icol[i]] = i+1; 2471 } 2472 2473 /* determine lens of each row */ 2474 for (i=0; i<nrows; i++) { 2475 kstart = ai[irow[i]]; 2476 kend = kstart + a->ilen[irow[i]]; 2477 lens[i] = 0; 2478 for (k=kstart; k<kend; k++) { 2479 if (smap[aj[k]]) { 2480 lens[i]++; 2481 } 2482 } 2483 } 2484 /* Create and fill new matrix */ 2485 if (scall == MAT_REUSE_MATRIX) { 2486 PetscBool equal; 2487 2488 c = (Mat_SeqAIJ*)((*B)->data); 2489 if ((*B)->rmap->n != nrows || (*B)->cmap->n != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size"); 2490 ierr = PetscMemcmp(c->ilen,lens,(*B)->rmap->n*sizeof(PetscInt),&equal);CHKERRQ(ierr); 2491 if (!equal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros"); 2492 ierr = PetscMemzero(c->ilen,(*B)->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 2493 C = *B; 2494 } else { 2495 PetscInt rbs,cbs; 2496 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2497 ierr = MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2498 ierr = ISGetBlockSize(isrow,&rbs);CHKERRQ(ierr); 2499 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2500 ierr = MatSetBlockSizes(C,rbs,cbs);CHKERRQ(ierr); 2501 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2502 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);CHKERRQ(ierr); 2503 } 2504 c = (Mat_SeqAIJ*)(C->data); 2505 for (i=0; i<nrows; i++) { 2506 row = irow[i]; 2507 kstart = ai[row]; 2508 kend = kstart + a->ilen[row]; 2509 mat_i = c->i[i]; 2510 mat_j = c->j + mat_i; 2511 mat_a = c->a + mat_i; 2512 mat_ilen = c->ilen + i; 2513 for (k=kstart; k<kend; k++) { 2514 if ((tcol=smap[a->j[k]])) { 2515 *mat_j++ = tcol - 1; 2516 *mat_a++ = a->a[k]; 2517 (*mat_ilen)++; 2518 2519 } 2520 } 2521 } 2522 /* Free work space */ 2523 ierr = ISRestoreIndices(iscol,&icol);CHKERRQ(ierr); 2524 ierr = PetscFree(smap);CHKERRQ(ierr); 2525 ierr = PetscFree(lens);CHKERRQ(ierr); 2526 } 2527 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2528 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2529 2530 ierr = ISRestoreIndices(isrow,&irow);CHKERRQ(ierr); 2531 *B = C; 2532 PetscFunctionReturn(0); 2533 } 2534 2535 #undef __FUNCT__ 2536 #define __FUNCT__ "MatGetMultiProcBlock_SeqAIJ" 2537 PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat) 2538 { 2539 PetscErrorCode ierr; 2540 Mat B; 2541 2542 PetscFunctionBegin; 2543 if (scall == MAT_INITIAL_MATRIX) { 2544 ierr = MatCreate(subComm,&B);CHKERRQ(ierr); 2545 ierr = MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n);CHKERRQ(ierr); 2546 ierr = MatSetBlockSizesFromMats(B,mat,mat);CHKERRQ(ierr); 2547 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 2548 ierr = MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE);CHKERRQ(ierr); 2549 *subMat = B; 2550 } else { 2551 ierr = MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 2552 } 2553 PetscFunctionReturn(0); 2554 } 2555 2556 #undef __FUNCT__ 2557 #define __FUNCT__ "MatILUFactor_SeqAIJ" 2558 PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info) 2559 { 2560 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2561 PetscErrorCode ierr; 2562 Mat outA; 2563 PetscBool row_identity,col_identity; 2564 2565 PetscFunctionBegin; 2566 if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu"); 2567 2568 ierr = ISIdentity(row,&row_identity);CHKERRQ(ierr); 2569 ierr = ISIdentity(col,&col_identity);CHKERRQ(ierr); 2570 2571 outA = inA; 2572 outA->factortype = MAT_FACTOR_LU; 2573 2574 ierr = PetscObjectReference((PetscObject)row);CHKERRQ(ierr); 2575 ierr = ISDestroy(&a->row);CHKERRQ(ierr); 2576 2577 a->row = row; 2578 2579 ierr = PetscObjectReference((PetscObject)col);CHKERRQ(ierr); 2580 ierr = ISDestroy(&a->col);CHKERRQ(ierr); 2581 2582 a->col = col; 2583 2584 /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */ 2585 ierr = ISDestroy(&a->icol);CHKERRQ(ierr); 2586 ierr = ISInvertPermutation(col,PETSC_DECIDE,&a->icol);CHKERRQ(ierr); 2587 ierr = PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);CHKERRQ(ierr); 2588 2589 if (!a->solve_work) { /* this matrix may have been factored before */ 2590 ierr = PetscMalloc1((inA->rmap->n+1),&a->solve_work);CHKERRQ(ierr); 2591 ierr = PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar));CHKERRQ(ierr); 2592 } 2593 2594 ierr = MatMarkDiagonal_SeqAIJ(inA);CHKERRQ(ierr); 2595 if (row_identity && col_identity) { 2596 ierr = MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info);CHKERRQ(ierr); 2597 } else { 2598 ierr = MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info);CHKERRQ(ierr); 2599 } 2600 PetscFunctionReturn(0); 2601 } 2602 2603 #undef __FUNCT__ 2604 #define __FUNCT__ "MatScale_SeqAIJ" 2605 PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha) 2606 { 2607 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2608 PetscScalar oalpha = alpha; 2609 PetscErrorCode ierr; 2610 PetscBLASInt one = 1,bnz; 2611 2612 PetscFunctionBegin; 2613 ierr = PetscBLASIntCast(a->nz,&bnz);CHKERRQ(ierr); 2614 PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&oalpha,a->a,&one)); 2615 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 2616 ierr = MatSeqAIJInvalidateDiagonal(inA);CHKERRQ(ierr); 2617 PetscFunctionReturn(0); 2618 } 2619 2620 #undef __FUNCT__ 2621 #define __FUNCT__ "MatGetSubMatrices_SeqAIJ" 2622 PetscErrorCode MatGetSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[]) 2623 { 2624 PetscErrorCode ierr; 2625 PetscInt i; 2626 2627 PetscFunctionBegin; 2628 if (scall == MAT_INITIAL_MATRIX) { 2629 ierr = PetscMalloc1((n+1),B);CHKERRQ(ierr); 2630 } 2631 2632 for (i=0; i<n; i++) { 2633 ierr = MatGetSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);CHKERRQ(ierr); 2634 } 2635 PetscFunctionReturn(0); 2636 } 2637 2638 #undef __FUNCT__ 2639 #define __FUNCT__ "MatIncreaseOverlap_SeqAIJ" 2640 PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov) 2641 { 2642 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2643 PetscErrorCode ierr; 2644 PetscInt row,i,j,k,l,m,n,*nidx,isz,val; 2645 const PetscInt *idx; 2646 PetscInt start,end,*ai,*aj; 2647 PetscBT table; 2648 2649 PetscFunctionBegin; 2650 m = A->rmap->n; 2651 ai = a->i; 2652 aj = a->j; 2653 2654 if (ov < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used"); 2655 2656 ierr = PetscMalloc1((m+1),&nidx);CHKERRQ(ierr); 2657 ierr = PetscBTCreate(m,&table);CHKERRQ(ierr); 2658 2659 for (i=0; i<is_max; i++) { 2660 /* Initialize the two local arrays */ 2661 isz = 0; 2662 ierr = PetscBTMemzero(m,table);CHKERRQ(ierr); 2663 2664 /* Extract the indices, assume there can be duplicate entries */ 2665 ierr = ISGetIndices(is[i],&idx);CHKERRQ(ierr); 2666 ierr = ISGetLocalSize(is[i],&n);CHKERRQ(ierr); 2667 2668 /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */ 2669 for (j=0; j<n; ++j) { 2670 if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j]; 2671 } 2672 ierr = ISRestoreIndices(is[i],&idx);CHKERRQ(ierr); 2673 ierr = ISDestroy(&is[i]);CHKERRQ(ierr); 2674 2675 k = 0; 2676 for (j=0; j<ov; j++) { /* for each overlap */ 2677 n = isz; 2678 for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */ 2679 row = nidx[k]; 2680 start = ai[row]; 2681 end = ai[row+1]; 2682 for (l = start; l<end; l++) { 2683 val = aj[l]; 2684 if (!PetscBTLookupSet(table,val)) nidx[isz++] = val; 2685 } 2686 } 2687 } 2688 ierr = ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i));CHKERRQ(ierr); 2689 } 2690 ierr = PetscBTDestroy(&table);CHKERRQ(ierr); 2691 ierr = PetscFree(nidx);CHKERRQ(ierr); 2692 PetscFunctionReturn(0); 2693 } 2694 2695 /* -------------------------------------------------------------- */ 2696 #undef __FUNCT__ 2697 #define __FUNCT__ "MatPermute_SeqAIJ" 2698 PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B) 2699 { 2700 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2701 PetscErrorCode ierr; 2702 PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n; 2703 const PetscInt *row,*col; 2704 PetscInt *cnew,j,*lens; 2705 IS icolp,irowp; 2706 PetscInt *cwork = NULL; 2707 PetscScalar *vwork = NULL; 2708 2709 PetscFunctionBegin; 2710 ierr = ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);CHKERRQ(ierr); 2711 ierr = ISGetIndices(irowp,&row);CHKERRQ(ierr); 2712 ierr = ISInvertPermutation(colp,PETSC_DECIDE,&icolp);CHKERRQ(ierr); 2713 ierr = ISGetIndices(icolp,&col);CHKERRQ(ierr); 2714 2715 /* determine lengths of permuted rows */ 2716 ierr = PetscMalloc1((m+1),&lens);CHKERRQ(ierr); 2717 for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i]; 2718 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 2719 ierr = MatSetSizes(*B,m,n,m,n);CHKERRQ(ierr); 2720 ierr = MatSetBlockSizesFromMats(*B,A,A);CHKERRQ(ierr); 2721 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 2722 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);CHKERRQ(ierr); 2723 ierr = PetscFree(lens);CHKERRQ(ierr); 2724 2725 ierr = PetscMalloc1(n,&cnew);CHKERRQ(ierr); 2726 for (i=0; i<m; i++) { 2727 ierr = MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);CHKERRQ(ierr); 2728 for (j=0; j<nz; j++) cnew[j] = col[cwork[j]]; 2729 ierr = MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);CHKERRQ(ierr); 2730 ierr = MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);CHKERRQ(ierr); 2731 } 2732 ierr = PetscFree(cnew);CHKERRQ(ierr); 2733 2734 (*B)->assembled = PETSC_FALSE; 2735 2736 ierr = MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2737 ierr = MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2738 ierr = ISRestoreIndices(irowp,&row);CHKERRQ(ierr); 2739 ierr = ISRestoreIndices(icolp,&col);CHKERRQ(ierr); 2740 ierr = ISDestroy(&irowp);CHKERRQ(ierr); 2741 ierr = ISDestroy(&icolp);CHKERRQ(ierr); 2742 PetscFunctionReturn(0); 2743 } 2744 2745 #undef __FUNCT__ 2746 #define __FUNCT__ "MatCopy_SeqAIJ" 2747 PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str) 2748 { 2749 PetscErrorCode ierr; 2750 2751 PetscFunctionBegin; 2752 /* If the two matrices have the same copy implementation, use fast copy. */ 2753 if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) { 2754 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2755 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 2756 2757 if (a->i[A->rmap->n] != b->i[B->rmap->n]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzeros in two matrices are different"); 2758 ierr = PetscMemcpy(b->a,a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));CHKERRQ(ierr); 2759 } else { 2760 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 2761 } 2762 PetscFunctionReturn(0); 2763 } 2764 2765 #undef __FUNCT__ 2766 #define __FUNCT__ "MatSetUp_SeqAIJ" 2767 PetscErrorCode MatSetUp_SeqAIJ(Mat A) 2768 { 2769 PetscErrorCode ierr; 2770 2771 PetscFunctionBegin; 2772 ierr = MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,0);CHKERRQ(ierr); 2773 PetscFunctionReturn(0); 2774 } 2775 2776 #undef __FUNCT__ 2777 #define __FUNCT__ "MatSeqAIJGetArray_SeqAIJ" 2778 PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[]) 2779 { 2780 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2781 2782 PetscFunctionBegin; 2783 *array = a->a; 2784 PetscFunctionReturn(0); 2785 } 2786 2787 #undef __FUNCT__ 2788 #define __FUNCT__ "MatSeqAIJRestoreArray_SeqAIJ" 2789 PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[]) 2790 { 2791 PetscFunctionBegin; 2792 PetscFunctionReturn(0); 2793 } 2794 2795 /* 2796 Computes the number of nonzeros per row needed for preallocation when X and Y 2797 have different nonzero structure. 2798 */ 2799 #undef __FUNCT__ 2800 #define __FUNCT__ "MatAXPYGetPreallocation_SeqX_private" 2801 PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt m,const PetscInt *xi,const PetscInt *xj,const PetscInt *yi,const PetscInt *yj,PetscInt *nnz) 2802 { 2803 PetscInt i,j,k,nzx,nzy; 2804 2805 PetscFunctionBegin; 2806 /* Set the number of nonzeros in the new matrix */ 2807 for (i=0; i<m; i++) { 2808 const PetscInt *xjj = xj+xi[i],*yjj = yj+yi[i]; 2809 nzx = xi[i+1] - xi[i]; 2810 nzy = yi[i+1] - yi[i]; 2811 nnz[i] = 0; 2812 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 2813 for (; k<nzy && yjj[k]<xjj[j]; k++) nnz[i]++; /* Catch up to X */ 2814 if (k<nzy && yjj[k]==xjj[j]) k++; /* Skip duplicate */ 2815 nnz[i]++; 2816 } 2817 for (; k<nzy; k++) nnz[i]++; 2818 } 2819 PetscFunctionReturn(0); 2820 } 2821 2822 #undef __FUNCT__ 2823 #define __FUNCT__ "MatAXPYGetPreallocation_SeqAIJ" 2824 PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz) 2825 { 2826 PetscInt m = Y->rmap->N; 2827 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 2828 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 2829 PetscErrorCode ierr; 2830 2831 PetscFunctionBegin; 2832 /* Set the number of nonzeros in the new matrix */ 2833 ierr = MatAXPYGetPreallocation_SeqX_private(m,x->i,x->j,y->i,y->j,nnz);CHKERRQ(ierr); 2834 PetscFunctionReturn(0); 2835 } 2836 2837 #undef __FUNCT__ 2838 #define __FUNCT__ "MatAXPY_SeqAIJ" 2839 PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 2840 { 2841 PetscErrorCode ierr; 2842 PetscInt i; 2843 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data; 2844 PetscBLASInt one=1,bnz; 2845 2846 PetscFunctionBegin; 2847 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 2848 if (str == SAME_NONZERO_PATTERN) { 2849 PetscScalar alpha = a; 2850 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 2851 ierr = MatSeqAIJInvalidateDiagonal(Y);CHKERRQ(ierr); 2852 ierr = PetscObjectStateIncrease((PetscObject)Y);CHKERRQ(ierr); 2853 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 2854 if (y->xtoy && y->XtoY != X) { 2855 ierr = PetscFree(y->xtoy);CHKERRQ(ierr); 2856 ierr = MatDestroy(&y->XtoY);CHKERRQ(ierr); 2857 } 2858 if (!y->xtoy) { /* get xtoy */ 2859 ierr = MatAXPYGetxtoy_Private(X->rmap->n,x->i,x->j,NULL, y->i,y->j,NULL, &y->xtoy);CHKERRQ(ierr); 2860 y->XtoY = X; 2861 ierr = PetscObjectReference((PetscObject)X);CHKERRQ(ierr); 2862 } 2863 for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]); 2864 ierr = PetscObjectStateIncrease((PetscObject)Y);CHKERRQ(ierr); 2865 ierr = PetscInfo3(Y,"ratio of nnz(X)/nnz(Y): %D/%D = %g\n",x->nz,y->nz,(double)((PetscReal)(x->nz)/(y->nz+1)));CHKERRQ(ierr); 2866 } else { 2867 Mat B; 2868 PetscInt *nnz; 2869 ierr = PetscMalloc1(Y->rmap->N,&nnz);CHKERRQ(ierr); 2870 ierr = MatCreate(PetscObjectComm((PetscObject)Y),&B);CHKERRQ(ierr); 2871 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 2872 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 2873 ierr = MatSetBlockSizesFromMats(B,Y,Y);CHKERRQ(ierr); 2874 ierr = MatSetType(B,(MatType) ((PetscObject)Y)->type_name);CHKERRQ(ierr); 2875 ierr = MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);CHKERRQ(ierr); 2876 ierr = MatSeqAIJSetPreallocation(B,0,nnz);CHKERRQ(ierr); 2877 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 2878 ierr = MatHeaderReplace(Y,B);CHKERRQ(ierr); 2879 ierr = PetscFree(nnz);CHKERRQ(ierr); 2880 } 2881 PetscFunctionReturn(0); 2882 } 2883 2884 #undef __FUNCT__ 2885 #define __FUNCT__ "MatConjugate_SeqAIJ" 2886 PetscErrorCode MatConjugate_SeqAIJ(Mat mat) 2887 { 2888 #if defined(PETSC_USE_COMPLEX) 2889 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 2890 PetscInt i,nz; 2891 PetscScalar *a; 2892 2893 PetscFunctionBegin; 2894 nz = aij->nz; 2895 a = aij->a; 2896 for (i=0; i<nz; i++) a[i] = PetscConj(a[i]); 2897 #else 2898 PetscFunctionBegin; 2899 #endif 2900 PetscFunctionReturn(0); 2901 } 2902 2903 #undef __FUNCT__ 2904 #define __FUNCT__ "MatGetRowMaxAbs_SeqAIJ" 2905 PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 2906 { 2907 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2908 PetscErrorCode ierr; 2909 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 2910 PetscReal atmp; 2911 PetscScalar *x; 2912 MatScalar *aa; 2913 2914 PetscFunctionBegin; 2915 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 2916 aa = a->a; 2917 ai = a->i; 2918 aj = a->j; 2919 2920 ierr = VecSet(v,0.0);CHKERRQ(ierr); 2921 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 2922 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 2923 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 2924 for (i=0; i<m; i++) { 2925 ncols = ai[1] - ai[0]; ai++; 2926 x[i] = 0.0; 2927 for (j=0; j<ncols; j++) { 2928 atmp = PetscAbsScalar(*aa); 2929 if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 2930 aa++; aj++; 2931 } 2932 } 2933 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 2934 PetscFunctionReturn(0); 2935 } 2936 2937 #undef __FUNCT__ 2938 #define __FUNCT__ "MatGetRowMax_SeqAIJ" 2939 PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 2940 { 2941 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2942 PetscErrorCode ierr; 2943 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 2944 PetscScalar *x; 2945 MatScalar *aa; 2946 2947 PetscFunctionBegin; 2948 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 2949 aa = a->a; 2950 ai = a->i; 2951 aj = a->j; 2952 2953 ierr = VecSet(v,0.0);CHKERRQ(ierr); 2954 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 2955 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 2956 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 2957 for (i=0; i<m; i++) { 2958 ncols = ai[1] - ai[0]; ai++; 2959 if (ncols == A->cmap->n) { /* row is dense */ 2960 x[i] = *aa; if (idx) idx[i] = 0; 2961 } else { /* row is sparse so already KNOW maximum is 0.0 or higher */ 2962 x[i] = 0.0; 2963 if (idx) { 2964 idx[i] = 0; /* in case ncols is zero */ 2965 for (j=0;j<ncols;j++) { /* find first implicit 0.0 in the row */ 2966 if (aj[j] > j) { 2967 idx[i] = j; 2968 break; 2969 } 2970 } 2971 } 2972 } 2973 for (j=0; j<ncols; j++) { 2974 if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 2975 aa++; aj++; 2976 } 2977 } 2978 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 2979 PetscFunctionReturn(0); 2980 } 2981 2982 #undef __FUNCT__ 2983 #define __FUNCT__ "MatGetRowMinAbs_SeqAIJ" 2984 PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 2985 { 2986 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2987 PetscErrorCode ierr; 2988 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 2989 PetscReal atmp; 2990 PetscScalar *x; 2991 MatScalar *aa; 2992 2993 PetscFunctionBegin; 2994 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 2995 aa = a->a; 2996 ai = a->i; 2997 aj = a->j; 2998 2999 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3000 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3001 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3002 if (n != A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector, %D vs. %D rows", A->rmap->n, n); 3003 for (i=0; i<m; i++) { 3004 ncols = ai[1] - ai[0]; ai++; 3005 if (ncols) { 3006 /* Get first nonzero */ 3007 for (j = 0; j < ncols; j++) { 3008 atmp = PetscAbsScalar(aa[j]); 3009 if (atmp > 1.0e-12) { 3010 x[i] = atmp; 3011 if (idx) idx[i] = aj[j]; 3012 break; 3013 } 3014 } 3015 if (j == ncols) {x[i] = PetscAbsScalar(*aa); if (idx) idx[i] = *aj;} 3016 } else { 3017 x[i] = 0.0; if (idx) idx[i] = 0; 3018 } 3019 for (j = 0; j < ncols; j++) { 3020 atmp = PetscAbsScalar(*aa); 3021 if (atmp > 1.0e-12 && PetscAbsScalar(x[i]) > atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3022 aa++; aj++; 3023 } 3024 } 3025 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3026 PetscFunctionReturn(0); 3027 } 3028 3029 #undef __FUNCT__ 3030 #define __FUNCT__ "MatGetRowMin_SeqAIJ" 3031 PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3032 { 3033 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3034 PetscErrorCode ierr; 3035 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3036 PetscScalar *x; 3037 MatScalar *aa; 3038 3039 PetscFunctionBegin; 3040 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3041 aa = a->a; 3042 ai = a->i; 3043 aj = a->j; 3044 3045 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3046 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3047 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3048 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3049 for (i=0; i<m; i++) { 3050 ncols = ai[1] - ai[0]; ai++; 3051 if (ncols == A->cmap->n) { /* row is dense */ 3052 x[i] = *aa; if (idx) idx[i] = 0; 3053 } else { /* row is sparse so already KNOW minimum is 0.0 or lower */ 3054 x[i] = 0.0; 3055 if (idx) { /* find first implicit 0.0 in the row */ 3056 idx[i] = 0; /* in case ncols is zero */ 3057 for (j=0; j<ncols; j++) { 3058 if (aj[j] > j) { 3059 idx[i] = j; 3060 break; 3061 } 3062 } 3063 } 3064 } 3065 for (j=0; j<ncols; j++) { 3066 if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3067 aa++; aj++; 3068 } 3069 } 3070 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3071 PetscFunctionReturn(0); 3072 } 3073 3074 #include <petscblaslapack.h> 3075 #include <petsc-private/kernels/blockinvert.h> 3076 3077 #undef __FUNCT__ 3078 #define __FUNCT__ "MatInvertBlockDiagonal_SeqAIJ" 3079 PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values) 3080 { 3081 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 3082 PetscErrorCode ierr; 3083 PetscInt i,bs = PetscAbs(A->rmap->bs),mbs = A->rmap->n/bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j; 3084 MatScalar *diag,work[25],*v_work; 3085 PetscReal shift = 0.0; 3086 3087 PetscFunctionBegin; 3088 if (a->ibdiagvalid) { 3089 if (values) *values = a->ibdiag; 3090 PetscFunctionReturn(0); 3091 } 3092 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 3093 if (!a->ibdiag) { 3094 ierr = PetscMalloc1(bs2*mbs,&a->ibdiag);CHKERRQ(ierr); 3095 ierr = PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));CHKERRQ(ierr); 3096 } 3097 diag = a->ibdiag; 3098 if (values) *values = a->ibdiag; 3099 /* factor and invert each block */ 3100 switch (bs) { 3101 case 1: 3102 for (i=0; i<mbs; i++) { 3103 ierr = MatGetValues(A,1,&i,1,&i,diag+i);CHKERRQ(ierr); 3104 diag[i] = (PetscScalar)1.0 / (diag[i] + shift); 3105 } 3106 break; 3107 case 2: 3108 for (i=0; i<mbs; i++) { 3109 ij[0] = 2*i; ij[1] = 2*i + 1; 3110 ierr = MatGetValues(A,2,ij,2,ij,diag);CHKERRQ(ierr); 3111 ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 3112 ierr = PetscKernel_A_gets_transpose_A_2(diag);CHKERRQ(ierr); 3113 diag += 4; 3114 } 3115 break; 3116 case 3: 3117 for (i=0; i<mbs; i++) { 3118 ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2; 3119 ierr = MatGetValues(A,3,ij,3,ij,diag);CHKERRQ(ierr); 3120 ierr = PetscKernel_A_gets_inverse_A_3(diag,shift);CHKERRQ(ierr); 3121 ierr = PetscKernel_A_gets_transpose_A_3(diag);CHKERRQ(ierr); 3122 diag += 9; 3123 } 3124 break; 3125 case 4: 3126 for (i=0; i<mbs; i++) { 3127 ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3; 3128 ierr = MatGetValues(A,4,ij,4,ij,diag);CHKERRQ(ierr); 3129 ierr = PetscKernel_A_gets_inverse_A_4(diag,shift);CHKERRQ(ierr); 3130 ierr = PetscKernel_A_gets_transpose_A_4(diag);CHKERRQ(ierr); 3131 diag += 16; 3132 } 3133 break; 3134 case 5: 3135 for (i=0; i<mbs; i++) { 3136 ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4; 3137 ierr = MatGetValues(A,5,ij,5,ij,diag);CHKERRQ(ierr); 3138 ierr = PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift);CHKERRQ(ierr); 3139 ierr = PetscKernel_A_gets_transpose_A_5(diag);CHKERRQ(ierr); 3140 diag += 25; 3141 } 3142 break; 3143 case 6: 3144 for (i=0; i<mbs; i++) { 3145 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; 3146 ierr = MatGetValues(A,6,ij,6,ij,diag);CHKERRQ(ierr); 3147 ierr = PetscKernel_A_gets_inverse_A_6(diag,shift);CHKERRQ(ierr); 3148 ierr = PetscKernel_A_gets_transpose_A_6(diag);CHKERRQ(ierr); 3149 diag += 36; 3150 } 3151 break; 3152 case 7: 3153 for (i=0; i<mbs; i++) { 3154 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; 3155 ierr = MatGetValues(A,7,ij,7,ij,diag);CHKERRQ(ierr); 3156 ierr = PetscKernel_A_gets_inverse_A_7(diag,shift);CHKERRQ(ierr); 3157 ierr = PetscKernel_A_gets_transpose_A_7(diag);CHKERRQ(ierr); 3158 diag += 49; 3159 } 3160 break; 3161 default: 3162 ierr = PetscMalloc3(bs,&v_work,bs,&v_pivots,bs,&IJ);CHKERRQ(ierr); 3163 for (i=0; i<mbs; i++) { 3164 for (j=0; j<bs; j++) { 3165 IJ[j] = bs*i + j; 3166 } 3167 ierr = MatGetValues(A,bs,IJ,bs,IJ,diag);CHKERRQ(ierr); 3168 ierr = PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work);CHKERRQ(ierr); 3169 ierr = PetscKernel_A_gets_transpose_A_N(diag,bs);CHKERRQ(ierr); 3170 diag += bs2; 3171 } 3172 ierr = PetscFree3(v_work,v_pivots,IJ);CHKERRQ(ierr); 3173 } 3174 a->ibdiagvalid = PETSC_TRUE; 3175 PetscFunctionReturn(0); 3176 } 3177 3178 #undef __FUNCT__ 3179 #define __FUNCT__ "MatSetRandom_SeqAIJ" 3180 static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx) 3181 { 3182 PetscErrorCode ierr; 3183 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data; 3184 PetscScalar a; 3185 PetscInt m,n,i,j,col; 3186 3187 PetscFunctionBegin; 3188 if (!x->assembled) { 3189 ierr = MatGetSize(x,&m,&n);CHKERRQ(ierr); 3190 for (i=0; i<m; i++) { 3191 for (j=0; j<aij->imax[i]; j++) { 3192 ierr = PetscRandomGetValue(rctx,&a);CHKERRQ(ierr); 3193 col = (PetscInt)(n*PetscRealPart(a)); 3194 ierr = MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);CHKERRQ(ierr); 3195 } 3196 } 3197 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not yet coded"); 3198 ierr = MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3199 ierr = MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3200 PetscFunctionReturn(0); 3201 } 3202 3203 /* -------------------------------------------------------------------*/ 3204 static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ, 3205 MatGetRow_SeqAIJ, 3206 MatRestoreRow_SeqAIJ, 3207 MatMult_SeqAIJ, 3208 /* 4*/ MatMultAdd_SeqAIJ, 3209 MatMultTranspose_SeqAIJ, 3210 MatMultTransposeAdd_SeqAIJ, 3211 0, 3212 0, 3213 0, 3214 /* 10*/ 0, 3215 MatLUFactor_SeqAIJ, 3216 0, 3217 MatSOR_SeqAIJ, 3218 MatTranspose_SeqAIJ, 3219 /*1 5*/ MatGetInfo_SeqAIJ, 3220 MatEqual_SeqAIJ, 3221 MatGetDiagonal_SeqAIJ, 3222 MatDiagonalScale_SeqAIJ, 3223 MatNorm_SeqAIJ, 3224 /* 20*/ 0, 3225 MatAssemblyEnd_SeqAIJ, 3226 MatSetOption_SeqAIJ, 3227 MatZeroEntries_SeqAIJ, 3228 /* 24*/ MatZeroRows_SeqAIJ, 3229 0, 3230 0, 3231 0, 3232 0, 3233 /* 29*/ MatSetUp_SeqAIJ, 3234 0, 3235 0, 3236 0, 3237 0, 3238 /* 34*/ MatDuplicate_SeqAIJ, 3239 0, 3240 0, 3241 MatILUFactor_SeqAIJ, 3242 0, 3243 /* 39*/ MatAXPY_SeqAIJ, 3244 MatGetSubMatrices_SeqAIJ, 3245 MatIncreaseOverlap_SeqAIJ, 3246 MatGetValues_SeqAIJ, 3247 MatCopy_SeqAIJ, 3248 /* 44*/ MatGetRowMax_SeqAIJ, 3249 MatScale_SeqAIJ, 3250 0, 3251 MatDiagonalSet_SeqAIJ, 3252 MatZeroRowsColumns_SeqAIJ, 3253 /* 49*/ MatSetRandom_SeqAIJ, 3254 MatGetRowIJ_SeqAIJ, 3255 MatRestoreRowIJ_SeqAIJ, 3256 MatGetColumnIJ_SeqAIJ, 3257 MatRestoreColumnIJ_SeqAIJ, 3258 /* 54*/ MatFDColoringCreate_SeqXAIJ, 3259 0, 3260 0, 3261 MatPermute_SeqAIJ, 3262 0, 3263 /* 59*/ 0, 3264 MatDestroy_SeqAIJ, 3265 MatView_SeqAIJ, 3266 0, 3267 MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ, 3268 /* 64*/ MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ, 3269 MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ, 3270 0, 3271 0, 3272 0, 3273 /* 69*/ MatGetRowMaxAbs_SeqAIJ, 3274 MatGetRowMinAbs_SeqAIJ, 3275 0, 3276 MatSetColoring_SeqAIJ, 3277 0, 3278 /* 74*/ MatSetValuesAdifor_SeqAIJ, 3279 MatFDColoringApply_AIJ, 3280 0, 3281 0, 3282 0, 3283 /* 79*/ MatFindZeroDiagonals_SeqAIJ, 3284 0, 3285 0, 3286 0, 3287 MatLoad_SeqAIJ, 3288 /* 84*/ MatIsSymmetric_SeqAIJ, 3289 MatIsHermitian_SeqAIJ, 3290 0, 3291 0, 3292 0, 3293 /* 89*/ MatMatMult_SeqAIJ_SeqAIJ, 3294 MatMatMultSymbolic_SeqAIJ_SeqAIJ, 3295 MatMatMultNumeric_SeqAIJ_SeqAIJ, 3296 MatPtAP_SeqAIJ_SeqAIJ, 3297 MatPtAPSymbolic_SeqAIJ_SeqAIJ_DenseAxpy, 3298 /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ, 3299 MatMatTransposeMult_SeqAIJ_SeqAIJ, 3300 MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ, 3301 MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ, 3302 0, 3303 /* 99*/ 0, 3304 0, 3305 0, 3306 MatConjugate_SeqAIJ, 3307 0, 3308 /*104*/ MatSetValuesRow_SeqAIJ, 3309 MatRealPart_SeqAIJ, 3310 MatImaginaryPart_SeqAIJ, 3311 0, 3312 0, 3313 /*109*/ MatMatSolve_SeqAIJ, 3314 0, 3315 MatGetRowMin_SeqAIJ, 3316 0, 3317 MatMissingDiagonal_SeqAIJ, 3318 /*114*/ 0, 3319 0, 3320 0, 3321 0, 3322 0, 3323 /*119*/ 0, 3324 0, 3325 0, 3326 0, 3327 MatGetMultiProcBlock_SeqAIJ, 3328 /*124*/ MatFindNonzeroRows_SeqAIJ, 3329 MatGetColumnNorms_SeqAIJ, 3330 MatInvertBlockDiagonal_SeqAIJ, 3331 0, 3332 0, 3333 /*129*/ 0, 3334 MatTransposeMatMult_SeqAIJ_SeqAIJ, 3335 MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ, 3336 MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ, 3337 MatTransposeColoringCreate_SeqAIJ, 3338 /*134*/ MatTransColoringApplySpToDen_SeqAIJ, 3339 MatTransColoringApplyDenToSp_SeqAIJ, 3340 MatRARt_SeqAIJ_SeqAIJ, 3341 MatRARtSymbolic_SeqAIJ_SeqAIJ, 3342 MatRARtNumeric_SeqAIJ_SeqAIJ, 3343 /*139*/0, 3344 0, 3345 0, 3346 MatFDColoringSetUp_SeqXAIJ, 3347 MatFindOffBlockDiagonalEntries_SeqAIJ 3348 }; 3349 3350 #undef __FUNCT__ 3351 #define __FUNCT__ "MatSeqAIJSetColumnIndices_SeqAIJ" 3352 PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices) 3353 { 3354 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3355 PetscInt i,nz,n; 3356 3357 PetscFunctionBegin; 3358 nz = aij->maxnz; 3359 n = mat->rmap->n; 3360 for (i=0; i<nz; i++) { 3361 aij->j[i] = indices[i]; 3362 } 3363 aij->nz = nz; 3364 for (i=0; i<n; i++) { 3365 aij->ilen[i] = aij->imax[i]; 3366 } 3367 PetscFunctionReturn(0); 3368 } 3369 3370 #undef __FUNCT__ 3371 #define __FUNCT__ "MatSeqAIJSetColumnIndices" 3372 /*@ 3373 MatSeqAIJSetColumnIndices - Set the column indices for all the rows 3374 in the matrix. 3375 3376 Input Parameters: 3377 + mat - the SeqAIJ matrix 3378 - indices - the column indices 3379 3380 Level: advanced 3381 3382 Notes: 3383 This can be called if you have precomputed the nonzero structure of the 3384 matrix and want to provide it to the matrix object to improve the performance 3385 of the MatSetValues() operation. 3386 3387 You MUST have set the correct numbers of nonzeros per row in the call to 3388 MatCreateSeqAIJ(), and the columns indices MUST be sorted. 3389 3390 MUST be called before any calls to MatSetValues(); 3391 3392 The indices should start with zero, not one. 3393 3394 @*/ 3395 PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices) 3396 { 3397 PetscErrorCode ierr; 3398 3399 PetscFunctionBegin; 3400 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3401 PetscValidPointer(indices,2); 3402 ierr = PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));CHKERRQ(ierr); 3403 PetscFunctionReturn(0); 3404 } 3405 3406 /* ----------------------------------------------------------------------------------------*/ 3407 3408 #undef __FUNCT__ 3409 #define __FUNCT__ "MatStoreValues_SeqAIJ" 3410 PetscErrorCode MatStoreValues_SeqAIJ(Mat mat) 3411 { 3412 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3413 PetscErrorCode ierr; 3414 size_t nz = aij->i[mat->rmap->n]; 3415 3416 PetscFunctionBegin; 3417 if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3418 3419 /* allocate space for values if not already there */ 3420 if (!aij->saved_values) { 3421 ierr = PetscMalloc1((nz+1),&aij->saved_values);CHKERRQ(ierr); 3422 ierr = PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));CHKERRQ(ierr); 3423 } 3424 3425 /* copy values over */ 3426 ierr = PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3427 PetscFunctionReturn(0); 3428 } 3429 3430 #undef __FUNCT__ 3431 #define __FUNCT__ "MatStoreValues" 3432 /*@ 3433 MatStoreValues - Stashes a copy of the matrix values; this allows, for 3434 example, reuse of the linear part of a Jacobian, while recomputing the 3435 nonlinear portion. 3436 3437 Collect on Mat 3438 3439 Input Parameters: 3440 . mat - the matrix (currently only AIJ matrices support this option) 3441 3442 Level: advanced 3443 3444 Common Usage, with SNESSolve(): 3445 $ Create Jacobian matrix 3446 $ Set linear terms into matrix 3447 $ Apply boundary conditions to matrix, at this time matrix must have 3448 $ final nonzero structure (i.e. setting the nonlinear terms and applying 3449 $ boundary conditions again will not change the nonzero structure 3450 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3451 $ ierr = MatStoreValues(mat); 3452 $ Call SNESSetJacobian() with matrix 3453 $ In your Jacobian routine 3454 $ ierr = MatRetrieveValues(mat); 3455 $ Set nonlinear terms in matrix 3456 3457 Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself: 3458 $ // build linear portion of Jacobian 3459 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3460 $ ierr = MatStoreValues(mat); 3461 $ loop over nonlinear iterations 3462 $ ierr = MatRetrieveValues(mat); 3463 $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian 3464 $ // call MatAssemblyBegin/End() on matrix 3465 $ Solve linear system with Jacobian 3466 $ endloop 3467 3468 Notes: 3469 Matrix must already be assemblied before calling this routine 3470 Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before 3471 calling this routine. 3472 3473 When this is called multiple times it overwrites the previous set of stored values 3474 and does not allocated additional space. 3475 3476 .seealso: MatRetrieveValues() 3477 3478 @*/ 3479 PetscErrorCode MatStoreValues(Mat mat) 3480 { 3481 PetscErrorCode ierr; 3482 3483 PetscFunctionBegin; 3484 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3485 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3486 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3487 ierr = PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));CHKERRQ(ierr); 3488 PetscFunctionReturn(0); 3489 } 3490 3491 #undef __FUNCT__ 3492 #define __FUNCT__ "MatRetrieveValues_SeqAIJ" 3493 PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat) 3494 { 3495 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3496 PetscErrorCode ierr; 3497 PetscInt nz = aij->i[mat->rmap->n]; 3498 3499 PetscFunctionBegin; 3500 if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3501 if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first"); 3502 /* copy values over */ 3503 ierr = PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3504 PetscFunctionReturn(0); 3505 } 3506 3507 #undef __FUNCT__ 3508 #define __FUNCT__ "MatRetrieveValues" 3509 /*@ 3510 MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for 3511 example, reuse of the linear part of a Jacobian, while recomputing the 3512 nonlinear portion. 3513 3514 Collect on Mat 3515 3516 Input Parameters: 3517 . mat - the matrix (currently on AIJ matrices support this option) 3518 3519 Level: advanced 3520 3521 .seealso: MatStoreValues() 3522 3523 @*/ 3524 PetscErrorCode MatRetrieveValues(Mat mat) 3525 { 3526 PetscErrorCode ierr; 3527 3528 PetscFunctionBegin; 3529 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3530 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3531 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3532 ierr = PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));CHKERRQ(ierr); 3533 PetscFunctionReturn(0); 3534 } 3535 3536 3537 /* --------------------------------------------------------------------------------*/ 3538 #undef __FUNCT__ 3539 #define __FUNCT__ "MatCreateSeqAIJ" 3540 /*@C 3541 MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format 3542 (the default parallel PETSc format). For good matrix assembly performance 3543 the user should preallocate the matrix storage by setting the parameter nz 3544 (or the array nnz). By setting these parameters accurately, performance 3545 during matrix assembly can be increased by more than a factor of 50. 3546 3547 Collective on MPI_Comm 3548 3549 Input Parameters: 3550 + comm - MPI communicator, set to PETSC_COMM_SELF 3551 . m - number of rows 3552 . n - number of columns 3553 . nz - number of nonzeros per row (same for all rows) 3554 - nnz - array containing the number of nonzeros in the various rows 3555 (possibly different for each row) or NULL 3556 3557 Output Parameter: 3558 . A - the matrix 3559 3560 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3561 MatXXXXSetPreallocation() paradgm instead of this routine directly. 3562 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3563 3564 Notes: 3565 If nnz is given then nz is ignored 3566 3567 The AIJ format (also called the Yale sparse matrix format or 3568 compressed row storage), is fully compatible with standard Fortran 77 3569 storage. That is, the stored row and column indices can begin at 3570 either one (as in Fortran) or zero. See the users' manual for details. 3571 3572 Specify the preallocated storage with either nz or nnz (not both). 3573 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3574 allocation. For large problems you MUST preallocate memory or you 3575 will get TERRIBLE performance, see the users' manual chapter on matrices. 3576 3577 By default, this format uses inodes (identical nodes) when possible, to 3578 improve numerical efficiency of matrix-vector products and solves. We 3579 search for consecutive rows with the same nonzero structure, thereby 3580 reusing matrix information to achieve increased efficiency. 3581 3582 Options Database Keys: 3583 + -mat_no_inode - Do not use inodes 3584 - -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3585 3586 Level: intermediate 3587 3588 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays() 3589 3590 @*/ 3591 PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A) 3592 { 3593 PetscErrorCode ierr; 3594 3595 PetscFunctionBegin; 3596 ierr = MatCreate(comm,A);CHKERRQ(ierr); 3597 ierr = MatSetSizes(*A,m,n,m,n);CHKERRQ(ierr); 3598 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 3599 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);CHKERRQ(ierr); 3600 PetscFunctionReturn(0); 3601 } 3602 3603 #undef __FUNCT__ 3604 #define __FUNCT__ "MatSeqAIJSetPreallocation" 3605 /*@C 3606 MatSeqAIJSetPreallocation - For good matrix assembly performance 3607 the user should preallocate the matrix storage by setting the parameter nz 3608 (or the array nnz). By setting these parameters accurately, performance 3609 during matrix assembly can be increased by more than a factor of 50. 3610 3611 Collective on MPI_Comm 3612 3613 Input Parameters: 3614 + B - The matrix 3615 . nz - number of nonzeros per row (same for all rows) 3616 - nnz - array containing the number of nonzeros in the various rows 3617 (possibly different for each row) or NULL 3618 3619 Notes: 3620 If nnz is given then nz is ignored 3621 3622 The AIJ format (also called the Yale sparse matrix format or 3623 compressed row storage), is fully compatible with standard Fortran 77 3624 storage. That is, the stored row and column indices can begin at 3625 either one (as in Fortran) or zero. See the users' manual for details. 3626 3627 Specify the preallocated storage with either nz or nnz (not both). 3628 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3629 allocation. For large problems you MUST preallocate memory or you 3630 will get TERRIBLE performance, see the users' manual chapter on matrices. 3631 3632 You can call MatGetInfo() to get information on how effective the preallocation was; 3633 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3634 You can also run with the option -info and look for messages with the string 3635 malloc in them to see if additional memory allocation was needed. 3636 3637 Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix 3638 entries or columns indices 3639 3640 By default, this format uses inodes (identical nodes) when possible, to 3641 improve numerical efficiency of matrix-vector products and solves. We 3642 search for consecutive rows with the same nonzero structure, thereby 3643 reusing matrix information to achieve increased efficiency. 3644 3645 Options Database Keys: 3646 + -mat_no_inode - Do not use inodes 3647 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3648 - -mat_aij_oneindex - Internally use indexing starting at 1 3649 rather than 0. Note that when calling MatSetValues(), 3650 the user still MUST index entries starting at 0! 3651 3652 Level: intermediate 3653 3654 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo() 3655 3656 @*/ 3657 PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[]) 3658 { 3659 PetscErrorCode ierr; 3660 3661 PetscFunctionBegin; 3662 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3663 PetscValidType(B,1); 3664 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));CHKERRQ(ierr); 3665 PetscFunctionReturn(0); 3666 } 3667 3668 #undef __FUNCT__ 3669 #define __FUNCT__ "MatSeqAIJSetPreallocation_SeqAIJ" 3670 PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz) 3671 { 3672 Mat_SeqAIJ *b; 3673 PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE; 3674 PetscErrorCode ierr; 3675 PetscInt i; 3676 3677 PetscFunctionBegin; 3678 if (nz >= 0 || nnz) realalloc = PETSC_TRUE; 3679 if (nz == MAT_SKIP_ALLOCATION) { 3680 skipallocation = PETSC_TRUE; 3681 nz = 0; 3682 } 3683 3684 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3685 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3686 3687 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3688 if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %D",nz); 3689 if (nnz) { 3690 for (i=0; i<B->rmap->n; i++) { 3691 if (nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %D value %D",i,nnz[i]); 3692 if (nnz[i] > B->cmap->n) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than row length: local row %D value %d rowlength %D",i,nnz[i],B->cmap->n); 3693 } 3694 } 3695 3696 B->preallocated = PETSC_TRUE; 3697 3698 b = (Mat_SeqAIJ*)B->data; 3699 3700 if (!skipallocation) { 3701 if (!b->imax) { 3702 ierr = PetscMalloc2(B->rmap->n,&b->imax,B->rmap->n,&b->ilen);CHKERRQ(ierr); 3703 ierr = PetscLogObjectMemory((PetscObject)B,2*B->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 3704 } 3705 if (!nnz) { 3706 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10; 3707 else if (nz < 0) nz = 1; 3708 for (i=0; i<B->rmap->n; i++) b->imax[i] = nz; 3709 nz = nz*B->rmap->n; 3710 } else { 3711 nz = 0; 3712 for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz += nnz[i];} 3713 } 3714 /* b->ilen will count nonzeros in each row so far. */ 3715 for (i=0; i<B->rmap->n; i++) b->ilen[i] = 0; 3716 3717 /* allocate the matrix space */ 3718 ierr = MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);CHKERRQ(ierr); 3719 ierr = PetscMalloc3(nz,&b->a,nz,&b->j,B->rmap->n+1,&b->i);CHKERRQ(ierr); 3720 ierr = PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));CHKERRQ(ierr); 3721 b->i[0] = 0; 3722 for (i=1; i<B->rmap->n+1; i++) { 3723 b->i[i] = b->i[i-1] + b->imax[i-1]; 3724 } 3725 b->singlemalloc = PETSC_TRUE; 3726 b->free_a = PETSC_TRUE; 3727 b->free_ij = PETSC_TRUE; 3728 #if defined(PETSC_THREADCOMM_ACTIVE) 3729 ierr = MatZeroEntries_SeqAIJ(B);CHKERRQ(ierr); 3730 #endif 3731 } else { 3732 b->free_a = PETSC_FALSE; 3733 b->free_ij = PETSC_FALSE; 3734 } 3735 3736 b->nz = 0; 3737 b->maxnz = nz; 3738 B->info.nz_unneeded = (double)b->maxnz; 3739 if (realalloc) { 3740 ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3741 } 3742 PetscFunctionReturn(0); 3743 } 3744 3745 #undef __FUNCT__ 3746 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR" 3747 /*@ 3748 MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format. 3749 3750 Input Parameters: 3751 + B - the matrix 3752 . i - the indices into j for the start of each row (starts with zero) 3753 . j - the column indices for each row (starts with zero) these must be sorted for each row 3754 - v - optional values in the matrix 3755 3756 Level: developer 3757 3758 The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays() 3759 3760 .keywords: matrix, aij, compressed row, sparse, sequential 3761 3762 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), SeqAIJ 3763 @*/ 3764 PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[]) 3765 { 3766 PetscErrorCode ierr; 3767 3768 PetscFunctionBegin; 3769 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3770 PetscValidType(B,1); 3771 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3772 PetscFunctionReturn(0); 3773 } 3774 3775 #undef __FUNCT__ 3776 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR_SeqAIJ" 3777 PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3778 { 3779 PetscInt i; 3780 PetscInt m,n; 3781 PetscInt nz; 3782 PetscInt *nnz, nz_max = 0; 3783 PetscScalar *values; 3784 PetscErrorCode ierr; 3785 3786 PetscFunctionBegin; 3787 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]); 3788 3789 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3790 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3791 3792 ierr = MatGetSize(B, &m, &n);CHKERRQ(ierr); 3793 ierr = PetscMalloc1((m+1), &nnz);CHKERRQ(ierr); 3794 for (i = 0; i < m; i++) { 3795 nz = Ii[i+1]- Ii[i]; 3796 nz_max = PetscMax(nz_max, nz); 3797 if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz); 3798 nnz[i] = nz; 3799 } 3800 ierr = MatSeqAIJSetPreallocation(B, 0, nnz);CHKERRQ(ierr); 3801 ierr = PetscFree(nnz);CHKERRQ(ierr); 3802 3803 if (v) { 3804 values = (PetscScalar*) v; 3805 } else { 3806 ierr = PetscCalloc1(nz_max, &values);CHKERRQ(ierr); 3807 } 3808 3809 for (i = 0; i < m; i++) { 3810 nz = Ii[i+1] - Ii[i]; 3811 ierr = MatSetValues_SeqAIJ(B, 1, &i, nz, J+Ii[i], values + (v ? Ii[i] : 0), INSERT_VALUES);CHKERRQ(ierr); 3812 } 3813 3814 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3815 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3816 3817 if (!v) { 3818 ierr = PetscFree(values);CHKERRQ(ierr); 3819 } 3820 ierr = MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3821 PetscFunctionReturn(0); 3822 } 3823 3824 #include <../src/mat/impls/dense/seq/dense.h> 3825 #include <petsc-private/kernels/petscaxpy.h> 3826 3827 #undef __FUNCT__ 3828 #define __FUNCT__ "MatMatMultNumeric_SeqDense_SeqAIJ" 3829 /* 3830 Computes (B'*A')' since computing B*A directly is untenable 3831 3832 n p p 3833 ( ) ( ) ( ) 3834 m ( A ) * n ( B ) = m ( C ) 3835 ( ) ( ) ( ) 3836 3837 */ 3838 PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C) 3839 { 3840 PetscErrorCode ierr; 3841 Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data; 3842 Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data; 3843 Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data; 3844 PetscInt i,n,m,q,p; 3845 const PetscInt *ii,*idx; 3846 const PetscScalar *b,*a,*a_q; 3847 PetscScalar *c,*c_q; 3848 3849 PetscFunctionBegin; 3850 m = A->rmap->n; 3851 n = A->cmap->n; 3852 p = B->cmap->n; 3853 a = sub_a->v; 3854 b = sub_b->a; 3855 c = sub_c->v; 3856 ierr = PetscMemzero(c,m*p*sizeof(PetscScalar));CHKERRQ(ierr); 3857 3858 ii = sub_b->i; 3859 idx = sub_b->j; 3860 for (i=0; i<n; i++) { 3861 q = ii[i+1] - ii[i]; 3862 while (q-->0) { 3863 c_q = c + m*(*idx); 3864 a_q = a + m*i; 3865 PetscKernelAXPY(c_q,*b,a_q,m); 3866 idx++; 3867 b++; 3868 } 3869 } 3870 PetscFunctionReturn(0); 3871 } 3872 3873 #undef __FUNCT__ 3874 #define __FUNCT__ "MatMatMultSymbolic_SeqDense_SeqAIJ" 3875 PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 3876 { 3877 PetscErrorCode ierr; 3878 PetscInt m=A->rmap->n,n=B->cmap->n; 3879 Mat Cmat; 3880 3881 PetscFunctionBegin; 3882 if (A->cmap->n != B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"A->cmap->n %D != B->rmap->n %D\n",A->cmap->n,B->rmap->n); 3883 ierr = MatCreate(PetscObjectComm((PetscObject)A),&Cmat);CHKERRQ(ierr); 3884 ierr = MatSetSizes(Cmat,m,n,m,n);CHKERRQ(ierr); 3885 ierr = MatSetBlockSizesFromMats(Cmat,A,B);CHKERRQ(ierr); 3886 ierr = MatSetType(Cmat,MATSEQDENSE);CHKERRQ(ierr); 3887 ierr = MatSeqDenseSetPreallocation(Cmat,NULL);CHKERRQ(ierr); 3888 3889 Cmat->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ; 3890 3891 *C = Cmat; 3892 PetscFunctionReturn(0); 3893 } 3894 3895 /* ----------------------------------------------------------------*/ 3896 #undef __FUNCT__ 3897 #define __FUNCT__ "MatMatMult_SeqDense_SeqAIJ" 3898 PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 3899 { 3900 PetscErrorCode ierr; 3901 3902 PetscFunctionBegin; 3903 if (scall == MAT_INITIAL_MATRIX) { 3904 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 3905 ierr = MatMatMultSymbolic_SeqDense_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 3906 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 3907 } 3908 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 3909 ierr = MatMatMultNumeric_SeqDense_SeqAIJ(A,B,*C);CHKERRQ(ierr); 3910 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 3911 PetscFunctionReturn(0); 3912 } 3913 3914 3915 /*MC 3916 MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices, 3917 based on compressed sparse row format. 3918 3919 Options Database Keys: 3920 . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions() 3921 3922 Level: beginner 3923 3924 .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType 3925 M*/ 3926 3927 /*MC 3928 MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices. 3929 3930 This matrix type is identical to MATSEQAIJ when constructed with a single process communicator, 3931 and MATMPIAIJ otherwise. As a result, for single process communicators, 3932 MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation is supported 3933 for communicators controlling multiple processes. It is recommended that you call both of 3934 the above preallocation routines for simplicity. 3935 3936 Options Database Keys: 3937 . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions() 3938 3939 Developer Notes: Subclasses include MATAIJCUSP, MATAIJPERM, MATAIJCRL, and also automatically switches over to use inodes when 3940 enough exist. 3941 3942 Level: beginner 3943 3944 .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ 3945 M*/ 3946 3947 /*MC 3948 MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices. 3949 3950 This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator, 3951 and MATMPIAIJCRL otherwise. As a result, for single process communicators, 3952 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 3953 for communicators controlling multiple processes. It is recommended that you call both of 3954 the above preallocation routines for simplicity. 3955 3956 Options Database Keys: 3957 . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions() 3958 3959 Level: beginner 3960 3961 .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL 3962 M*/ 3963 3964 #if defined(PETSC_HAVE_PASTIX) 3965 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_pastix(Mat,MatFactorType,Mat*); 3966 #endif 3967 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 3968 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_essl(Mat,MatFactorType,Mat*); 3969 #endif 3970 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*); 3971 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_petsc(Mat,MatFactorType,Mat*); 3972 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_bas(Mat,MatFactorType,Mat*); 3973 extern PetscErrorCode MatGetFactorAvailable_seqaij_petsc(Mat,MatFactorType,PetscBool*); 3974 #if defined(PETSC_HAVE_MUMPS) 3975 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mumps(Mat,MatFactorType,Mat*); 3976 #endif 3977 #if defined(PETSC_HAVE_SUPERLU) 3978 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat,MatFactorType,Mat*); 3979 #endif 3980 #if defined(PETSC_HAVE_MKL_PARDISO) 3981 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mkl_pardiso(Mat,MatFactorType,Mat*); 3982 #endif 3983 #if defined(PETSC_HAVE_SUPERLU_DIST) 3984 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu_dist(Mat,MatFactorType,Mat*); 3985 #endif 3986 #if defined(PETSC_HAVE_SUITESPARSE) 3987 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_umfpack(Mat,MatFactorType,Mat*); 3988 #endif 3989 #if defined(PETSC_HAVE_SUITESPARSE) 3990 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat,MatFactorType,Mat*); 3991 #endif 3992 #if defined(PETSC_HAVE_SUITESPARSE) 3993 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_klu(Mat,MatFactorType,Mat*); 3994 #endif 3995 #if defined(PETSC_HAVE_LUSOL) 3996 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_lusol(Mat,MatFactorType,Mat*); 3997 #endif 3998 #if defined(PETSC_HAVE_MATLAB_ENGINE) 3999 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_matlab(Mat,MatFactorType,Mat*); 4000 extern PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*); 4001 extern PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*); 4002 #endif 4003 #if defined(PETSC_HAVE_CLIQUE) 4004 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_clique(Mat,MatFactorType,Mat*); 4005 #endif 4006 4007 4008 #undef __FUNCT__ 4009 #define __FUNCT__ "MatSeqAIJGetArray" 4010 /*@C 4011 MatSeqAIJGetArray - gives access to the array where the data for a SeqSeqAIJ matrix is stored 4012 4013 Not Collective 4014 4015 Input Parameter: 4016 . mat - a MATSEQDENSE matrix 4017 4018 Output Parameter: 4019 . array - pointer to the data 4020 4021 Level: intermediate 4022 4023 .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90() 4024 @*/ 4025 PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array) 4026 { 4027 PetscErrorCode ierr; 4028 4029 PetscFunctionBegin; 4030 ierr = PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4031 PetscFunctionReturn(0); 4032 } 4033 4034 #undef __FUNCT__ 4035 #define __FUNCT__ "MatSeqAIJGetMaxRowNonzeros" 4036 /*@C 4037 MatSeqAIJGetMaxRowNonzeros - returns the maximum number of nonzeros in any row 4038 4039 Not Collective 4040 4041 Input Parameter: 4042 . mat - a MATSEQDENSE matrix 4043 4044 Output Parameter: 4045 . nz - the maximum number of nonzeros in any row 4046 4047 Level: intermediate 4048 4049 .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90() 4050 @*/ 4051 PetscErrorCode MatSeqAIJGetMaxRowNonzeros(Mat A,PetscInt *nz) 4052 { 4053 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data; 4054 4055 PetscFunctionBegin; 4056 *nz = aij->rmax; 4057 PetscFunctionReturn(0); 4058 } 4059 4060 #undef __FUNCT__ 4061 #define __FUNCT__ "MatSeqAIJRestoreArray" 4062 /*@C 4063 MatSeqAIJRestoreArray - returns access to the array where the data for a SeqSeqAIJ matrix is stored obtained by MatSeqAIJGetArray() 4064 4065 Not Collective 4066 4067 Input Parameters: 4068 . mat - a MATSEQDENSE matrix 4069 . array - pointer to the data 4070 4071 Level: intermediate 4072 4073 .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90() 4074 @*/ 4075 PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array) 4076 { 4077 PetscErrorCode ierr; 4078 4079 PetscFunctionBegin; 4080 ierr = PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4081 PetscFunctionReturn(0); 4082 } 4083 4084 #undef __FUNCT__ 4085 #define __FUNCT__ "MatCreate_SeqAIJ" 4086 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B) 4087 { 4088 Mat_SeqAIJ *b; 4089 PetscErrorCode ierr; 4090 PetscMPIInt size; 4091 4092 PetscFunctionBegin; 4093 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);CHKERRQ(ierr); 4094 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1"); 4095 4096 ierr = PetscNewLog(B,&b);CHKERRQ(ierr); 4097 4098 B->data = (void*)b; 4099 4100 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 4101 4102 b->row = 0; 4103 b->col = 0; 4104 b->icol = 0; 4105 b->reallocs = 0; 4106 b->ignorezeroentries = PETSC_FALSE; 4107 b->roworiented = PETSC_TRUE; 4108 b->nonew = 0; 4109 b->diag = 0; 4110 b->solve_work = 0; 4111 B->spptr = 0; 4112 b->saved_values = 0; 4113 b->idiag = 0; 4114 b->mdiag = 0; 4115 b->ssor_work = 0; 4116 b->omega = 1.0; 4117 b->fshift = 0.0; 4118 b->idiagvalid = PETSC_FALSE; 4119 b->ibdiagvalid = PETSC_FALSE; 4120 b->keepnonzeropattern = PETSC_FALSE; 4121 b->xtoy = 0; 4122 b->XtoY = 0; 4123 4124 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4125 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);CHKERRQ(ierr); 4126 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);CHKERRQ(ierr); 4127 4128 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4129 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_matlab_C",MatGetFactor_seqaij_matlab);CHKERRQ(ierr); 4130 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);CHKERRQ(ierr); 4131 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);CHKERRQ(ierr); 4132 #endif 4133 #if defined(PETSC_HAVE_PASTIX) 4134 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_pastix_C",MatGetFactor_seqaij_pastix);CHKERRQ(ierr); 4135 #endif 4136 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4137 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_essl_C",MatGetFactor_seqaij_essl);CHKERRQ(ierr); 4138 #endif 4139 #if defined(PETSC_HAVE_SUPERLU) 4140 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_C",MatGetFactor_seqaij_superlu);CHKERRQ(ierr); 4141 #endif 4142 #if defined(PETSC_HAVE_MKL_PARDISO) 4143 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mkl_pardiso_C",MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 4144 #endif 4145 #if defined(PETSC_HAVE_SUPERLU_DIST) 4146 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_dist_C",MatGetFactor_seqaij_superlu_dist);CHKERRQ(ierr); 4147 #endif 4148 #if defined(PETSC_HAVE_MUMPS) 4149 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mumps_C",MatGetFactor_aij_mumps);CHKERRQ(ierr); 4150 #endif 4151 #if defined(PETSC_HAVE_SUITESPARSE) 4152 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_umfpack_C",MatGetFactor_seqaij_umfpack);CHKERRQ(ierr); 4153 #endif 4154 #if defined(PETSC_HAVE_SUITESPARSE) 4155 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_cholmod_C",MatGetFactor_seqaij_cholmod);CHKERRQ(ierr); 4156 #endif 4157 #if defined(PETSC_HAVE_SUITESPARSE) 4158 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_klu_C",MatGetFactor_seqaij_klu);CHKERRQ(ierr); 4159 #endif 4160 #if defined(PETSC_HAVE_LUSOL) 4161 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_lusol_C",MatGetFactor_seqaij_lusol);CHKERRQ(ierr); 4162 #endif 4163 #if defined(PETSC_HAVE_CLIQUE) 4164 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_clique_C",MatGetFactor_aij_clique);CHKERRQ(ierr); 4165 #endif 4166 4167 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_petsc_C",MatGetFactor_seqaij_petsc);CHKERRQ(ierr); 4168 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactorAvailable_petsc_C",MatGetFactorAvailable_seqaij_petsc);CHKERRQ(ierr); 4169 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_bas_C",MatGetFactor_seqaij_bas);CHKERRQ(ierr); 4170 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);CHKERRQ(ierr); 4171 ierr = PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);CHKERRQ(ierr); 4172 ierr = PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);CHKERRQ(ierr); 4173 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);CHKERRQ(ierr); 4174 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);CHKERRQ(ierr); 4175 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);CHKERRQ(ierr); 4176 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);CHKERRQ(ierr); 4177 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4178 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4179 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);CHKERRQ(ierr); 4180 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);CHKERRQ(ierr); 4181 ierr = PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);CHKERRQ(ierr); 4182 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMult_seqdense_seqaij_C",MatMatMult_SeqDense_SeqAIJ);CHKERRQ(ierr); 4183 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_seqdense_seqaij_C",MatMatMultSymbolic_SeqDense_SeqAIJ);CHKERRQ(ierr); 4184 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_seqdense_seqaij_C",MatMatMultNumeric_SeqDense_SeqAIJ);CHKERRQ(ierr); 4185 ierr = MatCreate_SeqAIJ_Inode(B);CHKERRQ(ierr); 4186 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4187 PetscFunctionReturn(0); 4188 } 4189 4190 #undef __FUNCT__ 4191 #define __FUNCT__ "MatDuplicateNoCreate_SeqAIJ" 4192 /* 4193 Given a matrix generated with MatGetFactor() duplicates all the information in A into B 4194 */ 4195 PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace) 4196 { 4197 Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data; 4198 PetscErrorCode ierr; 4199 PetscInt i,m = A->rmap->n; 4200 4201 PetscFunctionBegin; 4202 c = (Mat_SeqAIJ*)C->data; 4203 4204 C->factortype = A->factortype; 4205 c->row = 0; 4206 c->col = 0; 4207 c->icol = 0; 4208 c->reallocs = 0; 4209 4210 C->assembled = PETSC_TRUE; 4211 4212 ierr = PetscLayoutReference(A->rmap,&C->rmap);CHKERRQ(ierr); 4213 ierr = PetscLayoutReference(A->cmap,&C->cmap);CHKERRQ(ierr); 4214 4215 ierr = PetscMalloc2(m,&c->imax,m,&c->ilen);CHKERRQ(ierr); 4216 ierr = PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));CHKERRQ(ierr); 4217 for (i=0; i<m; i++) { 4218 c->imax[i] = a->imax[i]; 4219 c->ilen[i] = a->ilen[i]; 4220 } 4221 4222 /* allocate the matrix space */ 4223 if (mallocmatspace) { 4224 ierr = PetscMalloc3(a->i[m],&c->a,a->i[m],&c->j,m+1,&c->i);CHKERRQ(ierr); 4225 ierr = PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4226 4227 c->singlemalloc = PETSC_TRUE; 4228 4229 ierr = PetscMemcpy(c->i,a->i,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4230 if (m > 0) { 4231 ierr = PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(PetscInt));CHKERRQ(ierr); 4232 if (cpvalues == MAT_COPY_VALUES) { 4233 ierr = PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4234 } else { 4235 ierr = PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4236 } 4237 } 4238 } 4239 4240 c->ignorezeroentries = a->ignorezeroentries; 4241 c->roworiented = a->roworiented; 4242 c->nonew = a->nonew; 4243 if (a->diag) { 4244 ierr = PetscMalloc1((m+1),&c->diag);CHKERRQ(ierr); 4245 ierr = PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4246 for (i=0; i<m; i++) { 4247 c->diag[i] = a->diag[i]; 4248 } 4249 } else c->diag = 0; 4250 4251 c->solve_work = 0; 4252 c->saved_values = 0; 4253 c->idiag = 0; 4254 c->ssor_work = 0; 4255 c->keepnonzeropattern = a->keepnonzeropattern; 4256 c->free_a = PETSC_TRUE; 4257 c->free_ij = PETSC_TRUE; 4258 c->xtoy = 0; 4259 c->XtoY = 0; 4260 4261 c->rmax = a->rmax; 4262 c->nz = a->nz; 4263 c->maxnz = a->nz; /* Since we allocate exactly the right amount */ 4264 C->preallocated = PETSC_TRUE; 4265 4266 c->compressedrow.use = a->compressedrow.use; 4267 c->compressedrow.nrows = a->compressedrow.nrows; 4268 if (a->compressedrow.use) { 4269 i = a->compressedrow.nrows; 4270 ierr = PetscMalloc2(i+1,&c->compressedrow.i,i,&c->compressedrow.rindex);CHKERRQ(ierr); 4271 ierr = PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));CHKERRQ(ierr); 4272 ierr = PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));CHKERRQ(ierr); 4273 } else { 4274 c->compressedrow.use = PETSC_FALSE; 4275 c->compressedrow.i = NULL; 4276 c->compressedrow.rindex = NULL; 4277 } 4278 C->nonzerostate = A->nonzerostate; 4279 4280 ierr = MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);CHKERRQ(ierr); 4281 ierr = PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);CHKERRQ(ierr); 4282 PetscFunctionReturn(0); 4283 } 4284 4285 #undef __FUNCT__ 4286 #define __FUNCT__ "MatDuplicate_SeqAIJ" 4287 PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B) 4288 { 4289 PetscErrorCode ierr; 4290 4291 PetscFunctionBegin; 4292 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 4293 ierr = MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);CHKERRQ(ierr); 4294 if (!(A->rmap->n % A->rmap->bs) && !(A->cmap->n % A->cmap->bs)) { 4295 ierr = MatSetBlockSizesFromMats(*B,A,A);CHKERRQ(ierr); 4296 } 4297 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 4298 ierr = MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);CHKERRQ(ierr); 4299 PetscFunctionReturn(0); 4300 } 4301 4302 #undef __FUNCT__ 4303 #define __FUNCT__ "MatLoad_SeqAIJ" 4304 PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer) 4305 { 4306 Mat_SeqAIJ *a; 4307 PetscErrorCode ierr; 4308 PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N,rows,cols; 4309 int fd; 4310 PetscMPIInt size; 4311 MPI_Comm comm; 4312 PetscInt bs = 1; 4313 4314 PetscFunctionBegin; 4315 ierr = PetscObjectGetComm((PetscObject)viewer,&comm);CHKERRQ(ierr); 4316 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4317 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"view must have one processor"); 4318 4319 ierr = PetscOptionsBegin(comm,NULL,"Options for loading SEQAIJ matrix","Mat");CHKERRQ(ierr); 4320 ierr = PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);CHKERRQ(ierr); 4321 ierr = PetscOptionsEnd();CHKERRQ(ierr); 4322 if (bs > 1) {ierr = MatSetBlockSize(newMat,bs);CHKERRQ(ierr);} 4323 4324 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 4325 ierr = PetscBinaryRead(fd,header,4,PETSC_INT);CHKERRQ(ierr); 4326 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file"); 4327 M = header[1]; N = header[2]; nz = header[3]; 4328 4329 if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ"); 4330 4331 /* read in row lengths */ 4332 ierr = PetscMalloc1(M,&rowlengths);CHKERRQ(ierr); 4333 ierr = PetscBinaryRead(fd,rowlengths,M,PETSC_INT);CHKERRQ(ierr); 4334 4335 /* check if sum of rowlengths is same as nz */ 4336 for (i=0,sum=0; i< M; i++) sum +=rowlengths[i]; 4337 if (sum != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_READ,"Inconsistant matrix data in file. no-nonzeros = %dD, sum-row-lengths = %D\n",nz,sum); 4338 4339 /* set global size if not set already*/ 4340 if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) { 4341 ierr = MatSetSizes(newMat,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr); 4342 } else { 4343 /* if sizes and type are already set, check if the vector global sizes are correct */ 4344 ierr = MatGetSize(newMat,&rows,&cols);CHKERRQ(ierr); 4345 if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */ 4346 ierr = MatGetLocalSize(newMat,&rows,&cols);CHKERRQ(ierr); 4347 } 4348 if (M != rows || N != cols) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different length (%D, %D) than the input matrix (%D, %D)",M,N,rows,cols); 4349 } 4350 ierr = MatSeqAIJSetPreallocation_SeqAIJ(newMat,0,rowlengths);CHKERRQ(ierr); 4351 a = (Mat_SeqAIJ*)newMat->data; 4352 4353 ierr = PetscBinaryRead(fd,a->j,nz,PETSC_INT);CHKERRQ(ierr); 4354 4355 /* read in nonzero values */ 4356 ierr = PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);CHKERRQ(ierr); 4357 4358 /* set matrix "i" values */ 4359 a->i[0] = 0; 4360 for (i=1; i<= M; i++) { 4361 a->i[i] = a->i[i-1] + rowlengths[i-1]; 4362 a->ilen[i-1] = rowlengths[i-1]; 4363 } 4364 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 4365 4366 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4367 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4368 PetscFunctionReturn(0); 4369 } 4370 4371 #undef __FUNCT__ 4372 #define __FUNCT__ "MatEqual_SeqAIJ" 4373 PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg) 4374 { 4375 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data; 4376 PetscErrorCode ierr; 4377 #if defined(PETSC_USE_COMPLEX) 4378 PetscInt k; 4379 #endif 4380 4381 PetscFunctionBegin; 4382 /* If the matrix dimensions are not equal,or no of nonzeros */ 4383 if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) { 4384 *flg = PETSC_FALSE; 4385 PetscFunctionReturn(0); 4386 } 4387 4388 /* if the a->i are the same */ 4389 ierr = PetscMemcmp(a->i,b->i,(A->rmap->n+1)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4390 if (!*flg) PetscFunctionReturn(0); 4391 4392 /* if a->j are the same */ 4393 ierr = PetscMemcmp(a->j,b->j,(a->nz)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4394 if (!*flg) PetscFunctionReturn(0); 4395 4396 /* if a->a are the same */ 4397 #if defined(PETSC_USE_COMPLEX) 4398 for (k=0; k<a->nz; k++) { 4399 if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) { 4400 *flg = PETSC_FALSE; 4401 PetscFunctionReturn(0); 4402 } 4403 } 4404 #else 4405 ierr = PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);CHKERRQ(ierr); 4406 #endif 4407 PetscFunctionReturn(0); 4408 } 4409 4410 #undef __FUNCT__ 4411 #define __FUNCT__ "MatCreateSeqAIJWithArrays" 4412 /*@ 4413 MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format) 4414 provided by the user. 4415 4416 Collective on MPI_Comm 4417 4418 Input Parameters: 4419 + comm - must be an MPI communicator of size 1 4420 . m - number of rows 4421 . n - number of columns 4422 . i - row indices 4423 . j - column indices 4424 - a - matrix values 4425 4426 Output Parameter: 4427 . mat - the matrix 4428 4429 Level: intermediate 4430 4431 Notes: 4432 The i, j, and a arrays are not copied by this routine, the user must free these arrays 4433 once the matrix is destroyed and not before 4434 4435 You cannot set new nonzero locations into this matrix, that will generate an error. 4436 4437 The i and j indices are 0 based 4438 4439 The format which is used for the sparse matrix input, is equivalent to a 4440 row-major ordering.. i.e for the following matrix, the input data expected is 4441 as shown: 4442 4443 1 0 0 4444 2 0 3 4445 4 5 6 4446 4447 i = {0,1,3,6} [size = nrow+1 = 3+1] 4448 j = {0,0,2,0,1,2} [size = nz = 6]; values must be sorted for each row 4449 v = {1,2,3,4,5,6} [size = nz = 6] 4450 4451 4452 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4453 4454 @*/ 4455 PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat) 4456 { 4457 PetscErrorCode ierr; 4458 PetscInt ii; 4459 Mat_SeqAIJ *aij; 4460 #if defined(PETSC_USE_DEBUG) 4461 PetscInt jj; 4462 #endif 4463 4464 PetscFunctionBegin; 4465 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 4466 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4467 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4468 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4469 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4470 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);CHKERRQ(ierr); 4471 aij = (Mat_SeqAIJ*)(*mat)->data; 4472 ierr = PetscMalloc2(m,&aij->imax,m,&aij->ilen);CHKERRQ(ierr); 4473 4474 aij->i = i; 4475 aij->j = j; 4476 aij->a = a; 4477 aij->singlemalloc = PETSC_FALSE; 4478 aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 4479 aij->free_a = PETSC_FALSE; 4480 aij->free_ij = PETSC_FALSE; 4481 4482 for (ii=0; ii<m; ii++) { 4483 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 4484 #if defined(PETSC_USE_DEBUG) 4485 if (i[ii+1] - i[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row length in i (row indices) row = %D length = %D",ii,i[ii+1] - i[ii]); 4486 for (jj=i[ii]+1; jj<i[ii+1]; jj++) { 4487 if (j[jj] < j[jj-1]) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %D (actual colum %D) in row %D is not sorted",jj-i[ii],j[jj],ii); 4488 if (j[jj] == j[jj]-1) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %D (actual colum %D) in row %D is identical to previous entry",jj-i[ii],j[jj],ii); 4489 } 4490 #endif 4491 } 4492 #if defined(PETSC_USE_DEBUG) 4493 for (ii=0; ii<aij->i[m]; ii++) { 4494 if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %D index = %D",ii,j[ii]); 4495 if (j[ii] > n - 1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column index to large at location = %D index = %D",ii,j[ii]); 4496 } 4497 #endif 4498 4499 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4500 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4501 PetscFunctionReturn(0); 4502 } 4503 #undef __FUNCT__ 4504 #define __FUNCT__ "MatCreateSeqAIJFromTriple" 4505 /*@C 4506 MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format) 4507 provided by the user. 4508 4509 Collective on MPI_Comm 4510 4511 Input Parameters: 4512 + comm - must be an MPI communicator of size 1 4513 . m - number of rows 4514 . n - number of columns 4515 . i - row indices 4516 . j - column indices 4517 . a - matrix values 4518 . nz - number of nonzeros 4519 - idx - 0 or 1 based 4520 4521 Output Parameter: 4522 . mat - the matrix 4523 4524 Level: intermediate 4525 4526 Notes: 4527 The i and j indices are 0 based 4528 4529 The format which is used for the sparse matrix input, is equivalent to a 4530 row-major ordering.. i.e for the following matrix, the input data expected is 4531 as shown: 4532 4533 1 0 0 4534 2 0 3 4535 4 5 6 4536 4537 i = {0,1,1,2,2,2} 4538 j = {0,0,2,0,1,2} 4539 v = {1,2,3,4,5,6} 4540 4541 4542 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4543 4544 @*/ 4545 PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat,PetscInt nz,PetscBool idx) 4546 { 4547 PetscErrorCode ierr; 4548 PetscInt ii, *nnz, one = 1,row,col; 4549 4550 4551 PetscFunctionBegin; 4552 ierr = PetscCalloc1(m,&nnz);CHKERRQ(ierr); 4553 for (ii = 0; ii < nz; ii++) { 4554 nnz[i[ii] - !!idx] += 1; 4555 } 4556 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4557 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4558 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4559 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);CHKERRQ(ierr); 4560 for (ii = 0; ii < nz; ii++) { 4561 if (idx) { 4562 row = i[ii] - 1; 4563 col = j[ii] - 1; 4564 } else { 4565 row = i[ii]; 4566 col = j[ii]; 4567 } 4568 ierr = MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);CHKERRQ(ierr); 4569 } 4570 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4571 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4572 ierr = PetscFree(nnz);CHKERRQ(ierr); 4573 PetscFunctionReturn(0); 4574 } 4575 4576 #undef __FUNCT__ 4577 #define __FUNCT__ "MatSetColoring_SeqAIJ" 4578 PetscErrorCode MatSetColoring_SeqAIJ(Mat A,ISColoring coloring) 4579 { 4580 PetscErrorCode ierr; 4581 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4582 4583 PetscFunctionBegin; 4584 if (coloring->ctype == IS_COLORING_GLOBAL) { 4585 ierr = ISColoringReference(coloring);CHKERRQ(ierr); 4586 a->coloring = coloring; 4587 } else if (coloring->ctype == IS_COLORING_GHOSTED) { 4588 PetscInt i,*larray; 4589 ISColoring ocoloring; 4590 ISColoringValue *colors; 4591 4592 /* set coloring for diagonal portion */ 4593 ierr = PetscMalloc1(A->cmap->n,&larray);CHKERRQ(ierr); 4594 for (i=0; i<A->cmap->n; i++) larray[i] = i; 4595 ierr = ISGlobalToLocalMappingApply(A->cmap->mapping,IS_GTOLM_MASK,A->cmap->n,larray,NULL,larray);CHKERRQ(ierr); 4596 ierr = PetscMalloc1(A->cmap->n,&colors);CHKERRQ(ierr); 4597 for (i=0; i<A->cmap->n; i++) colors[i] = coloring->colors[larray[i]]; 4598 ierr = PetscFree(larray);CHKERRQ(ierr); 4599 ierr = ISColoringCreate(PETSC_COMM_SELF,coloring->n,A->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4600 a->coloring = ocoloring; 4601 } 4602 PetscFunctionReturn(0); 4603 } 4604 4605 #undef __FUNCT__ 4606 #define __FUNCT__ "MatSetValuesAdifor_SeqAIJ" 4607 PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat A,PetscInt nl,void *advalues) 4608 { 4609 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4610 PetscInt m = A->rmap->n,*ii = a->i,*jj = a->j,nz,i,j; 4611 MatScalar *v = a->a; 4612 PetscScalar *values = (PetscScalar*)advalues; 4613 ISColoringValue *color; 4614 4615 PetscFunctionBegin; 4616 if (!a->coloring) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix"); 4617 color = a->coloring->colors; 4618 /* loop over rows */ 4619 for (i=0; i<m; i++) { 4620 nz = ii[i+1] - ii[i]; 4621 /* loop over columns putting computed value into matrix */ 4622 for (j=0; j<nz; j++) *v++ = values[color[*jj++]]; 4623 values += nl; /* jump to next row of derivatives */ 4624 } 4625 PetscFunctionReturn(0); 4626 } 4627 4628 #undef __FUNCT__ 4629 #define __FUNCT__ "MatSeqAIJInvalidateDiagonal" 4630 PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A) 4631 { 4632 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 4633 PetscErrorCode ierr; 4634 4635 PetscFunctionBegin; 4636 a->idiagvalid = PETSC_FALSE; 4637 a->ibdiagvalid = PETSC_FALSE; 4638 4639 ierr = MatSeqAIJInvalidateDiagonal_Inode(A);CHKERRQ(ierr); 4640 PetscFunctionReturn(0); 4641 } 4642 4643 /* 4644 Special version for direct calls from Fortran 4645 */ 4646 #include <petsc-private/fortranimpl.h> 4647 #if defined(PETSC_HAVE_FORTRAN_CAPS) 4648 #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ 4649 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 4650 #define matsetvaluesseqaij_ matsetvaluesseqaij 4651 #endif 4652 4653 /* Change these macros so can be used in void function */ 4654 #undef CHKERRQ 4655 #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr) 4656 #undef SETERRQ2 4657 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 4658 #undef SETERRQ3 4659 #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr) 4660 4661 #undef __FUNCT__ 4662 #define __FUNCT__ "matsetvaluesseqaij_" 4663 PETSC_EXTERN void PETSC_STDCALL matsetvaluesseqaij_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[],InsertMode *isis, PetscErrorCode *_ierr) 4664 { 4665 Mat A = *AA; 4666 PetscInt m = *mm, n = *nn; 4667 InsertMode is = *isis; 4668 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4669 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 4670 PetscInt *imax,*ai,*ailen; 4671 PetscErrorCode ierr; 4672 PetscInt *aj,nonew = a->nonew,lastcol = -1; 4673 MatScalar *ap,value,*aa; 4674 PetscBool ignorezeroentries = a->ignorezeroentries; 4675 PetscBool roworiented = a->roworiented; 4676 4677 PetscFunctionBegin; 4678 MatCheckPreallocated(A,1); 4679 imax = a->imax; 4680 ai = a->i; 4681 ailen = a->ilen; 4682 aj = a->j; 4683 aa = a->a; 4684 4685 for (k=0; k<m; k++) { /* loop over added rows */ 4686 row = im[k]; 4687 if (row < 0) continue; 4688 #if defined(PETSC_USE_DEBUG) 4689 if (row >= A->rmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large"); 4690 #endif 4691 rp = aj + ai[row]; ap = aa + ai[row]; 4692 rmax = imax[row]; nrow = ailen[row]; 4693 low = 0; 4694 high = nrow; 4695 for (l=0; l<n; l++) { /* loop over added columns */ 4696 if (in[l] < 0) continue; 4697 #if defined(PETSC_USE_DEBUG) 4698 if (in[l] >= A->cmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large"); 4699 #endif 4700 col = in[l]; 4701 if (roworiented) value = v[l + k*n]; 4702 else value = v[k + l*m]; 4703 4704 if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue; 4705 4706 if (col <= lastcol) low = 0; 4707 else high = nrow; 4708 lastcol = col; 4709 while (high-low > 5) { 4710 t = (low+high)/2; 4711 if (rp[t] > col) high = t; 4712 else low = t; 4713 } 4714 for (i=low; i<high; i++) { 4715 if (rp[i] > col) break; 4716 if (rp[i] == col) { 4717 if (is == ADD_VALUES) ap[i] += value; 4718 else ap[i] = value; 4719 goto noinsert; 4720 } 4721 } 4722 if (value == 0.0 && ignorezeroentries) goto noinsert; 4723 if (nonew == 1) goto noinsert; 4724 if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); 4725 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 4726 N = nrow++ - 1; a->nz++; high++; 4727 /* shift up all the later entries in this row */ 4728 for (ii=N; ii>=i; ii--) { 4729 rp[ii+1] = rp[ii]; 4730 ap[ii+1] = ap[ii]; 4731 } 4732 rp[i] = col; 4733 ap[i] = value; 4734 A->nonzerostate++; 4735 noinsert:; 4736 low = i + 1; 4737 } 4738 ailen[row] = nrow; 4739 } 4740 PetscFunctionReturnVoid(); 4741 } 4742 4743 4744