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__ "MatFindZeroDiagonals_SeqAIJ_Private" 49 PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat A,PetscInt *nrows,PetscInt **zrows) 50 { 51 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 52 const MatScalar *aa = a->a; 53 PetscInt i,m=A->rmap->n,cnt = 0; 54 const PetscInt *jj = a->j,*diag; 55 PetscInt *rows; 56 PetscErrorCode ierr; 57 58 PetscFunctionBegin; 59 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 60 diag = a->diag; 61 for (i=0; i<m; i++) { 62 if ((jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) { 63 cnt++; 64 } 65 } 66 ierr = PetscMalloc(cnt*sizeof(PetscInt),&rows);CHKERRQ(ierr); 67 cnt = 0; 68 for (i=0; i<m; i++) { 69 if ((jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) { 70 rows[cnt++] = i; 71 } 72 } 73 *nrows = cnt; 74 *zrows = rows; 75 PetscFunctionReturn(0); 76 } 77 78 #undef __FUNCT__ 79 #define __FUNCT__ "MatFindZeroDiagonals_SeqAIJ" 80 PetscErrorCode MatFindZeroDiagonals_SeqAIJ(Mat A,IS *zrows) 81 { 82 PetscInt nrows,*rows; 83 PetscErrorCode ierr; 84 85 PetscFunctionBegin; 86 *zrows = NULL; 87 ierr = MatFindZeroDiagonals_SeqAIJ_Private(A,&nrows,&rows);CHKERRQ(ierr); 88 ierr = ISCreateGeneral(PetscObjectComm((PetscObject)A),nrows,rows,PETSC_OWN_POINTER,zrows);CHKERRQ(ierr); 89 PetscFunctionReturn(0); 90 } 91 92 #undef __FUNCT__ 93 #define __FUNCT__ "MatFindNonzeroRows_SeqAIJ" 94 PetscErrorCode MatFindNonzeroRows_SeqAIJ(Mat A,IS *keptrows) 95 { 96 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 97 const MatScalar *aa; 98 PetscInt m=A->rmap->n,cnt = 0; 99 const PetscInt *ii; 100 PetscInt n,i,j,*rows; 101 PetscErrorCode ierr; 102 103 PetscFunctionBegin; 104 *keptrows = 0; 105 ii = a->i; 106 for (i=0; i<m; i++) { 107 n = ii[i+1] - ii[i]; 108 if (!n) { 109 cnt++; 110 goto ok1; 111 } 112 aa = a->a + ii[i]; 113 for (j=0; j<n; j++) { 114 if (aa[j] != 0.0) goto ok1; 115 } 116 cnt++; 117 ok1:; 118 } 119 if (!cnt) PetscFunctionReturn(0); 120 ierr = PetscMalloc((A->rmap->n-cnt)*sizeof(PetscInt),&rows);CHKERRQ(ierr); 121 cnt = 0; 122 for (i=0; i<m; i++) { 123 n = ii[i+1] - ii[i]; 124 if (!n) continue; 125 aa = a->a + ii[i]; 126 for (j=0; j<n; j++) { 127 if (aa[j] != 0.0) { 128 rows[cnt++] = i; 129 break; 130 } 131 } 132 } 133 ierr = ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,keptrows);CHKERRQ(ierr); 134 PetscFunctionReturn(0); 135 } 136 137 #undef __FUNCT__ 138 #define __FUNCT__ "MatDiagonalSet_SeqAIJ" 139 PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is) 140 { 141 PetscErrorCode ierr; 142 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data; 143 PetscInt i,*diag, m = Y->rmap->n; 144 MatScalar *aa = aij->a; 145 PetscScalar *v; 146 PetscBool missing; 147 148 PetscFunctionBegin; 149 if (Y->assembled) { 150 ierr = MatMissingDiagonal_SeqAIJ(Y,&missing,NULL);CHKERRQ(ierr); 151 if (!missing) { 152 diag = aij->diag; 153 ierr = VecGetArray(D,&v);CHKERRQ(ierr); 154 if (is == INSERT_VALUES) { 155 for (i=0; i<m; i++) { 156 aa[diag[i]] = v[i]; 157 } 158 } else { 159 for (i=0; i<m; i++) { 160 aa[diag[i]] += v[i]; 161 } 162 } 163 ierr = VecRestoreArray(D,&v);CHKERRQ(ierr); 164 PetscFunctionReturn(0); 165 } 166 ierr = MatSeqAIJInvalidateDiagonal(Y);CHKERRQ(ierr); 167 } 168 ierr = MatDiagonalSet_Default(Y,D,is);CHKERRQ(ierr); 169 PetscFunctionReturn(0); 170 } 171 172 #undef __FUNCT__ 173 #define __FUNCT__ "MatGetRowIJ_SeqAIJ" 174 PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *m,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 175 { 176 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 177 PetscErrorCode ierr; 178 PetscInt i,ishift; 179 180 PetscFunctionBegin; 181 *m = A->rmap->n; 182 if (!ia) PetscFunctionReturn(0); 183 ishift = 0; 184 if (symmetric && !A->structurally_symmetric) { 185 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,ishift,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 186 } else if (oshift == 1) { 187 PetscInt *tia; 188 PetscInt nz = a->i[A->rmap->n]; 189 /* malloc space and add 1 to i and j indices */ 190 ierr = PetscMalloc((A->rmap->n+1)*sizeof(PetscInt),&tia);CHKERRQ(ierr); 191 for (i=0; i<A->rmap->n+1; i++) tia[i] = a->i[i] + 1; 192 *ia = tia; 193 if (ja) { 194 PetscInt *tja; 195 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&tja);CHKERRQ(ierr); 196 for (i=0; i<nz; i++) tja[i] = a->j[i] + 1; 197 *ja = tja; 198 } 199 } else { 200 *ia = a->i; 201 if (ja) *ja = a->j; 202 } 203 PetscFunctionReturn(0); 204 } 205 206 #undef __FUNCT__ 207 #define __FUNCT__ "MatRestoreRowIJ_SeqAIJ" 208 PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 209 { 210 PetscErrorCode ierr; 211 212 PetscFunctionBegin; 213 if (!ia) PetscFunctionReturn(0); 214 if ((symmetric && !A->structurally_symmetric) || oshift == 1) { 215 ierr = PetscFree(*ia);CHKERRQ(ierr); 216 if (ja) {ierr = PetscFree(*ja);CHKERRQ(ierr);} 217 } 218 PetscFunctionReturn(0); 219 } 220 221 #undef __FUNCT__ 222 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ" 223 PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 224 { 225 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 226 PetscErrorCode ierr; 227 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 228 PetscInt nz = a->i[m],row,*jj,mr,col; 229 230 PetscFunctionBegin; 231 *nn = n; 232 if (!ia) PetscFunctionReturn(0); 233 if (symmetric) { 234 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 235 } else { 236 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&collengths);CHKERRQ(ierr); 237 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 238 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&cia);CHKERRQ(ierr); 239 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cja);CHKERRQ(ierr); 240 jj = a->j; 241 for (i=0; i<nz; i++) { 242 collengths[jj[i]]++; 243 } 244 cia[0] = oshift; 245 for (i=0; i<n; i++) { 246 cia[i+1] = cia[i] + collengths[i]; 247 } 248 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 249 jj = a->j; 250 for (row=0; row<m; row++) { 251 mr = a->i[row+1] - a->i[row]; 252 for (i=0; i<mr; i++) { 253 col = *jj++; 254 255 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 256 } 257 } 258 ierr = PetscFree(collengths);CHKERRQ(ierr); 259 *ia = cia; *ja = cja; 260 } 261 PetscFunctionReturn(0); 262 } 263 264 #undef __FUNCT__ 265 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ" 266 PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done) 267 { 268 PetscErrorCode ierr; 269 270 PetscFunctionBegin; 271 if (!ia) PetscFunctionReturn(0); 272 273 ierr = PetscFree(*ia);CHKERRQ(ierr); 274 ierr = PetscFree(*ja);CHKERRQ(ierr); 275 PetscFunctionReturn(0); 276 } 277 278 /* 279 MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from 280 MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output 281 spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ() and MatFDColoringCreate_SeqAIJ() 282 */ 283 #undef __FUNCT__ 284 #define __FUNCT__ "MatGetColumnIJ_SeqAIJ_Color" 285 PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 286 { 287 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 288 PetscErrorCode ierr; 289 PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n; 290 PetscInt nz = a->i[m],row,*jj,mr,col; 291 PetscInt *cspidx; 292 293 PetscFunctionBegin; 294 *nn = n; 295 if (!ia) PetscFunctionReturn(0); 296 if (symmetric) { 297 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MatGetColumnIJ_SeqAIJ_Color() not supported for the case symmetric"); 298 ierr = MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);CHKERRQ(ierr); 299 } else { 300 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&collengths);CHKERRQ(ierr); 301 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 302 ierr = PetscMalloc((n+1)*sizeof(PetscInt),&cia);CHKERRQ(ierr); 303 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cja);CHKERRQ(ierr); 304 ierr = PetscMalloc((nz+1)*sizeof(PetscInt),&cspidx);CHKERRQ(ierr); 305 jj = a->j; 306 for (i=0; i<nz; i++) { 307 collengths[jj[i]]++; 308 } 309 cia[0] = oshift; 310 for (i=0; i<n; i++) { 311 cia[i+1] = cia[i] + collengths[i]; 312 } 313 ierr = PetscMemzero(collengths,n*sizeof(PetscInt));CHKERRQ(ierr); 314 jj = a->j; 315 for (row=0; row<m; row++) { 316 mr = a->i[row+1] - a->i[row]; 317 for (i=0; i<mr; i++) { 318 col = *jj++; 319 320 cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */ 321 cja[cia[col] + collengths[col]++ - oshift] = row + oshift; 322 } 323 } 324 ierr = PetscFree(collengths);CHKERRQ(ierr); 325 *ia = cia; *ja = cja; 326 *spidx = cspidx; 327 } 328 PetscFunctionReturn(0); 329 } 330 331 #undef __FUNCT__ 332 #define __FUNCT__ "MatRestoreColumnIJ_SeqAIJ_Color" 333 PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done) 334 { 335 PetscErrorCode ierr; 336 337 PetscFunctionBegin; 338 if (!ia) PetscFunctionReturn(0); 339 340 ierr = PetscFree(*ia);CHKERRQ(ierr); 341 ierr = PetscFree(*ja);CHKERRQ(ierr); 342 ierr = PetscFree(*spidx);CHKERRQ(ierr); 343 PetscFunctionReturn(0); 344 } 345 346 #undef __FUNCT__ 347 #define __FUNCT__ "MatSetValuesRow_SeqAIJ" 348 PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[]) 349 { 350 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 351 PetscInt *ai = a->i; 352 PetscErrorCode ierr; 353 354 PetscFunctionBegin; 355 ierr = PetscMemcpy(a->a+ai[row],v,(ai[row+1]-ai[row])*sizeof(PetscScalar));CHKERRQ(ierr); 356 PetscFunctionReturn(0); 357 } 358 359 #undef __FUNCT__ 360 #define __FUNCT__ "MatSetValues_SeqAIJ" 361 PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is) 362 { 363 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 364 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 365 PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen; 366 PetscErrorCode ierr; 367 PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1; 368 MatScalar *ap,value,*aa = a->a; 369 PetscBool ignorezeroentries = a->ignorezeroentries; 370 PetscBool roworiented = a->roworiented; 371 372 PetscFunctionBegin; 373 if (v) PetscValidScalarPointer(v,6); 374 for (k=0; k<m; k++) { /* loop over added rows */ 375 row = im[k]; 376 if (row < 0) continue; 377 #if defined(PETSC_USE_DEBUG) 378 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); 379 #endif 380 rp = aj + ai[row]; ap = aa + ai[row]; 381 rmax = imax[row]; nrow = ailen[row]; 382 low = 0; 383 high = nrow; 384 for (l=0; l<n; l++) { /* loop over added columns */ 385 if (in[l] < 0) continue; 386 #if defined(PETSC_USE_DEBUG) 387 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); 388 #endif 389 col = in[l]; 390 if (v) { 391 if (roworiented) { 392 value = v[l + k*n]; 393 } else { 394 value = v[k + l*m]; 395 } 396 } else { 397 value = 0.; 398 } 399 if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue; 400 401 if (col <= lastcol) low = 0; 402 else high = nrow; 403 lastcol = col; 404 while (high-low > 5) { 405 t = (low+high)/2; 406 if (rp[t] > col) high = t; 407 else low = t; 408 } 409 for (i=low; i<high; i++) { 410 if (rp[i] > col) break; 411 if (rp[i] == col) { 412 if (is == ADD_VALUES) ap[i] += value; 413 else ap[i] = value; 414 low = i + 1; 415 goto noinsert; 416 } 417 } 418 if (value == 0.0 && ignorezeroentries) goto noinsert; 419 if (nonew == 1) goto noinsert; 420 if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%D,%D) in the matrix",row,col); 421 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 422 N = nrow++ - 1; a->nz++; high++; 423 /* shift up all the later entries in this row */ 424 for (ii=N; ii>=i; ii--) { 425 rp[ii+1] = rp[ii]; 426 ap[ii+1] = ap[ii]; 427 } 428 rp[i] = col; 429 ap[i] = value; 430 low = i + 1; 431 noinsert:; 432 } 433 ailen[row] = nrow; 434 } 435 A->same_nonzero = PETSC_FALSE; 436 PetscFunctionReturn(0); 437 } 438 439 440 #undef __FUNCT__ 441 #define __FUNCT__ "MatGetValues_SeqAIJ" 442 PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[]) 443 { 444 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 445 PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j; 446 PetscInt *ai = a->i,*ailen = a->ilen; 447 MatScalar *ap,*aa = a->a; 448 449 PetscFunctionBegin; 450 for (k=0; k<m; k++) { /* loop over rows */ 451 row = im[k]; 452 if (row < 0) {v += n; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",row); */ 453 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); 454 rp = aj + ai[row]; ap = aa + ai[row]; 455 nrow = ailen[row]; 456 for (l=0; l<n; l++) { /* loop over columns */ 457 if (in[l] < 0) {v++; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",in[l]); */ 458 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); 459 col = in[l]; 460 high = nrow; low = 0; /* assume unsorted */ 461 while (high-low > 5) { 462 t = (low+high)/2; 463 if (rp[t] > col) high = t; 464 else low = t; 465 } 466 for (i=low; i<high; i++) { 467 if (rp[i] > col) break; 468 if (rp[i] == col) { 469 *v++ = ap[i]; 470 goto finished; 471 } 472 } 473 *v++ = 0.0; 474 finished:; 475 } 476 } 477 PetscFunctionReturn(0); 478 } 479 480 481 #undef __FUNCT__ 482 #define __FUNCT__ "MatView_SeqAIJ_Binary" 483 PetscErrorCode MatView_SeqAIJ_Binary(Mat A,PetscViewer viewer) 484 { 485 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 486 PetscErrorCode ierr; 487 PetscInt i,*col_lens; 488 int fd; 489 FILE *file; 490 491 PetscFunctionBegin; 492 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 493 ierr = PetscMalloc((4+A->rmap->n)*sizeof(PetscInt),&col_lens);CHKERRQ(ierr); 494 495 col_lens[0] = MAT_FILE_CLASSID; 496 col_lens[1] = A->rmap->n; 497 col_lens[2] = A->cmap->n; 498 col_lens[3] = a->nz; 499 500 /* store lengths of each row and write (including header) to file */ 501 for (i=0; i<A->rmap->n; i++) { 502 col_lens[4+i] = a->i[i+1] - a->i[i]; 503 } 504 ierr = PetscBinaryWrite(fd,col_lens,4+A->rmap->n,PETSC_INT,PETSC_TRUE);CHKERRQ(ierr); 505 ierr = PetscFree(col_lens);CHKERRQ(ierr); 506 507 /* store column indices (zero start index) */ 508 ierr = PetscBinaryWrite(fd,a->j,a->nz,PETSC_INT,PETSC_FALSE);CHKERRQ(ierr); 509 510 /* store nonzero values */ 511 ierr = PetscBinaryWrite(fd,a->a,a->nz,PETSC_SCALAR,PETSC_FALSE);CHKERRQ(ierr); 512 513 ierr = PetscViewerBinaryGetInfoPointer(viewer,&file);CHKERRQ(ierr); 514 if (file) { 515 fprintf(file,"-matload_block_size %d\n",(int)A->rmap->bs); 516 } 517 PetscFunctionReturn(0); 518 } 519 520 extern PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer); 521 522 #undef __FUNCT__ 523 #define __FUNCT__ "MatView_SeqAIJ_ASCII" 524 PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer) 525 { 526 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 527 PetscErrorCode ierr; 528 PetscInt i,j,m = A->rmap->n,shift=0; 529 const char *name; 530 PetscViewerFormat format; 531 532 PetscFunctionBegin; 533 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 534 if (format == PETSC_VIEWER_ASCII_MATLAB) { 535 PetscInt nofinalvalue = 0; 536 if (m && ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap->n-!shift))) { 537 nofinalvalue = 1; 538 } 539 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 540 ierr = PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",m,A->cmap->n);CHKERRQ(ierr); 541 ierr = PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %D \n",a->nz);CHKERRQ(ierr); 542 ierr = PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,3);\n",a->nz+nofinalvalue);CHKERRQ(ierr); 543 ierr = PetscViewerASCIIPrintf(viewer,"zzz = [\n");CHKERRQ(ierr); 544 545 for (i=0; i<m; i++) { 546 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 547 #if defined(PETSC_USE_COMPLEX) 548 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e + %18.16ei \n",i+1,a->j[j]+!shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 549 #else 550 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",i+1,a->j[j]+!shift,a->a[j]);CHKERRQ(ierr); 551 #endif 552 } 553 } 554 if (nofinalvalue) { 555 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",m,A->cmap->n,0.0);CHKERRQ(ierr); 556 } 557 ierr = PetscObjectGetName((PetscObject)A,&name);CHKERRQ(ierr); 558 ierr = PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);CHKERRQ(ierr); 559 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 560 } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO) { 561 PetscFunctionReturn(0); 562 } else if (format == PETSC_VIEWER_ASCII_COMMON) { 563 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 564 ierr = PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer);CHKERRQ(ierr); 565 for (i=0; i<m; i++) { 566 ierr = PetscViewerASCIIPrintf(viewer,"row %D:",i);CHKERRQ(ierr); 567 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 568 #if defined(PETSC_USE_COMPLEX) 569 if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) { 570 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 571 } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) { 572 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 573 } else if (PetscRealPart(a->a[j]) != 0.0) { 574 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,PetscRealPart(a->a[j]));CHKERRQ(ierr); 575 } 576 #else 577 if (a->a[j] != 0.0) {ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,(double)a->a[j]);CHKERRQ(ierr);} 578 #endif 579 } 580 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 581 } 582 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 583 } else if (format == PETSC_VIEWER_ASCII_SYMMODU) { 584 PetscInt nzd=0,fshift=1,*sptr; 585 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 586 ierr = PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer);CHKERRQ(ierr); 587 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&sptr);CHKERRQ(ierr); 588 for (i=0; i<m; i++) { 589 sptr[i] = nzd+1; 590 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 591 if (a->j[j] >= i) { 592 #if defined(PETSC_USE_COMPLEX) 593 if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++; 594 #else 595 if (a->a[j] != 0.0) nzd++; 596 #endif 597 } 598 } 599 } 600 sptr[m] = nzd+1; 601 ierr = PetscViewerASCIIPrintf(viewer," %D %D\n\n",m,nzd);CHKERRQ(ierr); 602 for (i=0; i<m+1; i+=6) { 603 if (i+4<m) { 604 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); 605 } else if (i+3<m) { 606 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); 607 } else if (i+2<m) { 608 ierr = PetscViewerASCIIPrintf(viewer," %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);CHKERRQ(ierr); 609 } else if (i+1<m) { 610 ierr = PetscViewerASCIIPrintf(viewer," %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2]);CHKERRQ(ierr); 611 } else if (i<m) { 612 ierr = PetscViewerASCIIPrintf(viewer," %D %D\n",sptr[i],sptr[i+1]);CHKERRQ(ierr); 613 } else { 614 ierr = PetscViewerASCIIPrintf(viewer," %D\n",sptr[i]);CHKERRQ(ierr); 615 } 616 } 617 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 618 ierr = PetscFree(sptr);CHKERRQ(ierr); 619 for (i=0; i<m; i++) { 620 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 621 if (a->j[j] >= i) {ierr = PetscViewerASCIIPrintf(viewer," %D ",a->j[j]+fshift);CHKERRQ(ierr);} 622 } 623 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 624 } 625 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 626 for (i=0; i<m; i++) { 627 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 628 if (a->j[j] >= i) { 629 #if defined(PETSC_USE_COMPLEX) 630 if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) { 631 ierr = PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 632 } 633 #else 634 if (a->a[j] != 0.0) {ierr = PetscViewerASCIIPrintf(viewer," %18.16e ",a->a[j]);CHKERRQ(ierr);} 635 #endif 636 } 637 } 638 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 639 } 640 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 641 } else if (format == PETSC_VIEWER_ASCII_DENSE) { 642 PetscInt cnt = 0,jcnt; 643 PetscScalar value; 644 645 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 646 ierr = PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer);CHKERRQ(ierr); 647 for (i=0; i<m; i++) { 648 jcnt = 0; 649 for (j=0; j<A->cmap->n; j++) { 650 if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) { 651 value = a->a[cnt++]; 652 jcnt++; 653 } else { 654 value = 0.0; 655 } 656 #if defined(PETSC_USE_COMPLEX) 657 ierr = PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",PetscRealPart(value),PetscImaginaryPart(value));CHKERRQ(ierr); 658 #else 659 ierr = PetscViewerASCIIPrintf(viewer," %7.5e ",value);CHKERRQ(ierr); 660 #endif 661 } 662 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 663 } 664 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 665 } else if (format == PETSC_VIEWER_ASCII_MATRIXMARKET) { 666 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 667 ierr = PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer);CHKERRQ(ierr); 668 #if defined(PETSC_USE_COMPLEX) 669 ierr = PetscViewerASCIIPrintf(viewer,"%%matrix complex general\n");CHKERRQ(ierr); 670 #else 671 ierr = PetscViewerASCIIPrintf(viewer,"%%matrix real general\n");CHKERRQ(ierr); 672 #endif 673 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %D\n", m, A->cmap->n, a->nz);CHKERRQ(ierr); 674 for (i=0; i<m; i++) { 675 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 676 #if defined(PETSC_USE_COMPLEX) 677 if (PetscImaginaryPart(a->a[j]) > 0.0) { 678 ierr = PetscViewerASCIIPrintf(viewer,"%D %D, %g %g\n", i+shift,a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 679 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 680 ierr = PetscViewerASCIIPrintf(viewer,"%D %D, %g -%g\n", i+shift,a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 681 } else { 682 ierr = PetscViewerASCIIPrintf(viewer,"%D %D, %g\n", i+shift,a->j[j]+shift,PetscRealPart(a->a[j]));CHKERRQ(ierr); 683 } 684 #else 685 ierr = PetscViewerASCIIPrintf(viewer,"%D %D %g\n", i+shift, a->j[j]+shift, (double)a->a[j]);CHKERRQ(ierr); 686 #endif 687 } 688 } 689 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 690 } else { 691 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);CHKERRQ(ierr); 692 ierr = PetscObjectPrintClassNamePrefixType((PetscObject)A,viewer);CHKERRQ(ierr); 693 if (A->factortype) { 694 for (i=0; i<m; i++) { 695 ierr = PetscViewerASCIIPrintf(viewer,"row %D:",i);CHKERRQ(ierr); 696 /* L part */ 697 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 698 #if defined(PETSC_USE_COMPLEX) 699 if (PetscImaginaryPart(a->a[j]) > 0.0) { 700 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 701 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 702 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 703 } else { 704 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,PetscRealPart(a->a[j]));CHKERRQ(ierr); 705 } 706 #else 707 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,(double)a->a[j]);CHKERRQ(ierr); 708 #endif 709 } 710 /* diagonal */ 711 j = a->diag[i]; 712 #if defined(PETSC_USE_COMPLEX) 713 if (PetscImaginaryPart(a->a[j]) > 0.0) { 714 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j]+shift,PetscRealPart(1.0/a->a[j]),PetscImaginaryPart(1.0/a->a[j]));CHKERRQ(ierr); 715 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 716 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j]+shift,PetscRealPart(1.0/a->a[j]),-PetscImaginaryPart(1.0/a->a[j]));CHKERRQ(ierr); 717 } else { 718 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,PetscRealPart(1.0/a->a[j]));CHKERRQ(ierr); 719 } 720 #else 721 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,(double)1.0/a->a[j]);CHKERRQ(ierr); 722 #endif 723 724 /* U part */ 725 for (j=a->diag[i+1]+1+shift; j<a->diag[i]+shift; j++) { 726 #if defined(PETSC_USE_COMPLEX) 727 if (PetscImaginaryPart(a->a[j]) > 0.0) { 728 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 729 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 730 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 731 } else { 732 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,PetscRealPart(a->a[j]));CHKERRQ(ierr); 733 } 734 #else 735 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,(double)a->a[j]);CHKERRQ(ierr); 736 #endif 737 } 738 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 739 } 740 } else { 741 for (i=0; i<m; i++) { 742 ierr = PetscViewerASCIIPrintf(viewer,"row %D:",i);CHKERRQ(ierr); 743 for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) { 744 #if defined(PETSC_USE_COMPLEX) 745 if (PetscImaginaryPart(a->a[j]) > 0.0) { 746 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 747 } else if (PetscImaginaryPart(a->a[j]) < 0.0) { 748 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));CHKERRQ(ierr); 749 } else { 750 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,PetscRealPart(a->a[j]));CHKERRQ(ierr); 751 } 752 #else 753 ierr = PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j]+shift,(double)a->a[j]);CHKERRQ(ierr); 754 #endif 755 } 756 ierr = PetscViewerASCIIPrintf(viewer,"\n");CHKERRQ(ierr); 757 } 758 } 759 ierr = PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);CHKERRQ(ierr); 760 } 761 ierr = PetscViewerFlush(viewer);CHKERRQ(ierr); 762 PetscFunctionReturn(0); 763 } 764 765 #include <petscdraw.h> 766 #undef __FUNCT__ 767 #define __FUNCT__ "MatView_SeqAIJ_Draw_Zoom" 768 PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa) 769 { 770 Mat A = (Mat) Aa; 771 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 772 PetscErrorCode ierr; 773 PetscInt i,j,m = A->rmap->n,color; 774 PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r,maxv = 0.0; 775 PetscViewer viewer; 776 PetscViewerFormat format; 777 778 PetscFunctionBegin; 779 ierr = PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);CHKERRQ(ierr); 780 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 781 782 ierr = PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);CHKERRQ(ierr); 783 /* loop over matrix elements drawing boxes */ 784 785 if (format != PETSC_VIEWER_DRAW_CONTOUR) { 786 /* Blue for negative, Cyan for zero and Red for positive */ 787 color = PETSC_DRAW_BLUE; 788 for (i=0; i<m; i++) { 789 y_l = m - i - 1.0; y_r = y_l + 1.0; 790 for (j=a->i[i]; j<a->i[i+1]; j++) { 791 x_l = a->j[j]; x_r = x_l + 1.0; 792 if (PetscRealPart(a->a[j]) >= 0.) continue; 793 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 794 } 795 } 796 color = PETSC_DRAW_CYAN; 797 for (i=0; i<m; i++) { 798 y_l = m - i - 1.0; y_r = y_l + 1.0; 799 for (j=a->i[i]; j<a->i[i+1]; j++) { 800 x_l = a->j[j]; x_r = x_l + 1.0; 801 if (a->a[j] != 0.) continue; 802 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 803 } 804 } 805 color = PETSC_DRAW_RED; 806 for (i=0; i<m; i++) { 807 y_l = m - i - 1.0; y_r = y_l + 1.0; 808 for (j=a->i[i]; j<a->i[i+1]; j++) { 809 x_l = a->j[j]; x_r = x_l + 1.0; 810 if (PetscRealPart(a->a[j]) <= 0.) continue; 811 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 812 } 813 } 814 } else { 815 /* use contour shading to indicate magnitude of values */ 816 /* first determine max of all nonzero values */ 817 PetscInt nz = a->nz,count; 818 PetscDraw popup; 819 PetscReal scale; 820 821 for (i=0; i<nz; i++) { 822 if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]); 823 } 824 scale = (245.0 - PETSC_DRAW_BASIC_COLORS)/maxv; 825 ierr = PetscDrawGetPopup(draw,&popup);CHKERRQ(ierr); 826 if (popup) { 827 ierr = PetscDrawScalePopup(popup,0.0,maxv);CHKERRQ(ierr); 828 } 829 count = 0; 830 for (i=0; i<m; i++) { 831 y_l = m - i - 1.0; y_r = y_l + 1.0; 832 for (j=a->i[i]; j<a->i[i+1]; j++) { 833 x_l = a->j[j]; x_r = x_l + 1.0; 834 color = PETSC_DRAW_BASIC_COLORS + (PetscInt)(scale*PetscAbsScalar(a->a[count])); 835 ierr = PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);CHKERRQ(ierr); 836 count++; 837 } 838 } 839 } 840 PetscFunctionReturn(0); 841 } 842 843 #include <petscdraw.h> 844 #undef __FUNCT__ 845 #define __FUNCT__ "MatView_SeqAIJ_Draw" 846 PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer) 847 { 848 PetscErrorCode ierr; 849 PetscDraw draw; 850 PetscReal xr,yr,xl,yl,h,w; 851 PetscBool isnull; 852 853 PetscFunctionBegin; 854 ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr); 855 ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); 856 if (isnull) PetscFunctionReturn(0); 857 858 ierr = PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);CHKERRQ(ierr); 859 xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0; 860 xr += w; yr += h; xl = -w; yl = -h; 861 ierr = PetscDrawSetCoordinates(draw,xl,yl,xr,yr);CHKERRQ(ierr); 862 ierr = PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);CHKERRQ(ierr); 863 ierr = PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL);CHKERRQ(ierr); 864 PetscFunctionReturn(0); 865 } 866 867 #undef __FUNCT__ 868 #define __FUNCT__ "MatView_SeqAIJ" 869 PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer) 870 { 871 PetscErrorCode ierr; 872 PetscBool iascii,isbinary,isdraw; 873 874 PetscFunctionBegin; 875 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 876 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);CHKERRQ(ierr); 877 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr); 878 if (iascii) { 879 ierr = MatView_SeqAIJ_ASCII(A,viewer);CHKERRQ(ierr); 880 } else if (isbinary) { 881 ierr = MatView_SeqAIJ_Binary(A,viewer);CHKERRQ(ierr); 882 } else if (isdraw) { 883 ierr = MatView_SeqAIJ_Draw(A,viewer);CHKERRQ(ierr); 884 } 885 ierr = MatView_SeqAIJ_Inode(A,viewer);CHKERRQ(ierr); 886 PetscFunctionReturn(0); 887 } 888 889 #undef __FUNCT__ 890 #define __FUNCT__ "MatAssemblyEnd_SeqAIJ" 891 PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode) 892 { 893 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 894 PetscErrorCode ierr; 895 PetscInt fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax; 896 PetscInt m = A->rmap->n,*ip,N,*ailen = a->ilen,rmax = 0; 897 MatScalar *aa = a->a,*ap; 898 PetscReal ratio = 0.6; 899 900 PetscFunctionBegin; 901 if (mode == MAT_FLUSH_ASSEMBLY) PetscFunctionReturn(0); 902 903 if (m) rmax = ailen[0]; /* determine row with most nonzeros */ 904 for (i=1; i<m; i++) { 905 /* move each row back by the amount of empty slots (fshift) before it*/ 906 fshift += imax[i-1] - ailen[i-1]; 907 rmax = PetscMax(rmax,ailen[i]); 908 if (fshift) { 909 ip = aj + ai[i]; 910 ap = aa + ai[i]; 911 N = ailen[i]; 912 for (j=0; j<N; j++) { 913 ip[j-fshift] = ip[j]; 914 ap[j-fshift] = ap[j]; 915 } 916 } 917 ai[i] = ai[i-1] + ailen[i-1]; 918 } 919 if (m) { 920 fshift += imax[m-1] - ailen[m-1]; 921 ai[m] = ai[m-1] + ailen[m-1]; 922 } 923 /* reset ilen and imax for each row */ 924 for (i=0; i<m; i++) { 925 ailen[i] = imax[i] = ai[i+1] - ai[i]; 926 } 927 a->nz = ai[m]; 928 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); 929 930 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 931 ierr = PetscInfo4(A,"Matrix size: %D X %D; storage space: %D unneeded,%D used\n",m,A->cmap->n,fshift,a->nz);CHKERRQ(ierr); 932 ierr = PetscInfo1(A,"Number of mallocs during MatSetValues() is %D\n",a->reallocs);CHKERRQ(ierr); 933 ierr = PetscInfo1(A,"Maximum nonzeros in any row is %D\n",rmax);CHKERRQ(ierr); 934 935 A->info.mallocs += a->reallocs; 936 a->reallocs = 0; 937 A->info.nz_unneeded = (double)fshift; 938 a->rmax = rmax; 939 940 ierr = MatCheckCompressedRow(A,&a->compressedrow,a->i,m,ratio);CHKERRQ(ierr); 941 942 A->same_nonzero = PETSC_TRUE; 943 944 ierr = MatAssemblyEnd_SeqAIJ_Inode(A,mode);CHKERRQ(ierr); 945 946 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 947 PetscFunctionReturn(0); 948 } 949 950 #undef __FUNCT__ 951 #define __FUNCT__ "MatRealPart_SeqAIJ" 952 PetscErrorCode MatRealPart_SeqAIJ(Mat A) 953 { 954 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 955 PetscInt i,nz = a->nz; 956 MatScalar *aa = a->a; 957 PetscErrorCode ierr; 958 959 PetscFunctionBegin; 960 for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]); 961 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 962 PetscFunctionReturn(0); 963 } 964 965 #undef __FUNCT__ 966 #define __FUNCT__ "MatImaginaryPart_SeqAIJ" 967 PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A) 968 { 969 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 970 PetscInt i,nz = a->nz; 971 MatScalar *aa = a->a; 972 PetscErrorCode ierr; 973 974 PetscFunctionBegin; 975 for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]); 976 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 977 PetscFunctionReturn(0); 978 } 979 980 #if defined(PETSC_THREADCOMM_ACTIVE) 981 PetscErrorCode MatZeroEntries_SeqAIJ_Kernel(PetscInt thread_id,Mat A) 982 { 983 PetscErrorCode ierr; 984 PetscInt *trstarts=A->rmap->trstarts; 985 PetscInt n,start,end; 986 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 987 988 start = trstarts[thread_id]; 989 end = trstarts[thread_id+1]; 990 n = a->i[end] - a->i[start]; 991 ierr = PetscMemzero(a->a+a->i[start],n*sizeof(PetscScalar));CHKERRQ(ierr); 992 return 0; 993 } 994 995 #undef __FUNCT__ 996 #define __FUNCT__ "MatZeroEntries_SeqAIJ" 997 PetscErrorCode MatZeroEntries_SeqAIJ(Mat A) 998 { 999 PetscErrorCode ierr; 1000 1001 PetscFunctionBegin; 1002 ierr = PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatZeroEntries_SeqAIJ_Kernel,1,A);CHKERRQ(ierr); 1003 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1004 PetscFunctionReturn(0); 1005 } 1006 #else 1007 #undef __FUNCT__ 1008 #define __FUNCT__ "MatZeroEntries_SeqAIJ" 1009 PetscErrorCode MatZeroEntries_SeqAIJ(Mat A) 1010 { 1011 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1012 PetscErrorCode ierr; 1013 1014 PetscFunctionBegin; 1015 ierr = PetscMemzero(a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));CHKERRQ(ierr); 1016 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 1017 PetscFunctionReturn(0); 1018 } 1019 #endif 1020 1021 #undef __FUNCT__ 1022 #define __FUNCT__ "MatDestroy_SeqAIJ" 1023 PetscErrorCode MatDestroy_SeqAIJ(Mat A) 1024 { 1025 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1026 PetscErrorCode ierr; 1027 1028 PetscFunctionBegin; 1029 #if defined(PETSC_USE_LOG) 1030 PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap->n,A->cmap->n,a->nz); 1031 #endif 1032 ierr = MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);CHKERRQ(ierr); 1033 ierr = ISDestroy(&a->row);CHKERRQ(ierr); 1034 ierr = ISDestroy(&a->col);CHKERRQ(ierr); 1035 ierr = PetscFree(a->diag);CHKERRQ(ierr); 1036 ierr = PetscFree(a->ibdiag);CHKERRQ(ierr); 1037 ierr = PetscFree2(a->imax,a->ilen);CHKERRQ(ierr); 1038 ierr = PetscFree3(a->idiag,a->mdiag,a->ssor_work);CHKERRQ(ierr); 1039 ierr = PetscFree(a->solve_work);CHKERRQ(ierr); 1040 ierr = ISDestroy(&a->icol);CHKERRQ(ierr); 1041 ierr = PetscFree(a->saved_values);CHKERRQ(ierr); 1042 ierr = ISColoringDestroy(&a->coloring);CHKERRQ(ierr); 1043 ierr = PetscFree(a->xtoy);CHKERRQ(ierr); 1044 ierr = MatDestroy(&a->XtoY);CHKERRQ(ierr); 1045 ierr = PetscFree2(a->compressedrow.i,a->compressedrow.rindex);CHKERRQ(ierr); 1046 ierr = PetscFree(a->matmult_abdense);CHKERRQ(ierr); 1047 1048 ierr = MatDestroy_SeqAIJ_Inode(A);CHKERRQ(ierr); 1049 ierr = PetscFree(A->data);CHKERRQ(ierr); 1050 1051 ierr = PetscObjectChangeTypeName((PetscObject)A,0);CHKERRQ(ierr); 1052 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetColumnIndices_C",NULL);CHKERRQ(ierr); 1053 ierr = PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL);CHKERRQ(ierr); 1054 ierr = PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL);CHKERRQ(ierr); 1055 ierr = PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsbaij_C",NULL);CHKERRQ(ierr); 1056 ierr = PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqbaij_C",NULL);CHKERRQ(ierr); 1057 ierr = PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijperm_C",NULL);CHKERRQ(ierr); 1058 ierr = PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL);CHKERRQ(ierr); 1059 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocation_C",NULL);CHKERRQ(ierr); 1060 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C",NULL);CHKERRQ(ierr); 1061 ierr = PetscObjectComposeFunction((PetscObject)A,"MatReorderForNonzeroDiagonal_C",NULL);CHKERRQ(ierr); 1062 PetscFunctionReturn(0); 1063 } 1064 1065 #undef __FUNCT__ 1066 #define __FUNCT__ "MatSetOption_SeqAIJ" 1067 PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op,PetscBool flg) 1068 { 1069 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1070 PetscErrorCode ierr; 1071 1072 PetscFunctionBegin; 1073 switch (op) { 1074 case MAT_ROW_ORIENTED: 1075 a->roworiented = flg; 1076 break; 1077 case MAT_KEEP_NONZERO_PATTERN: 1078 a->keepnonzeropattern = flg; 1079 break; 1080 case MAT_NEW_NONZERO_LOCATIONS: 1081 a->nonew = (flg ? 0 : 1); 1082 break; 1083 case MAT_NEW_NONZERO_LOCATION_ERR: 1084 a->nonew = (flg ? -1 : 0); 1085 break; 1086 case MAT_NEW_NONZERO_ALLOCATION_ERR: 1087 a->nonew = (flg ? -2 : 0); 1088 break; 1089 case MAT_UNUSED_NONZERO_LOCATION_ERR: 1090 a->nounused = (flg ? -1 : 0); 1091 break; 1092 case MAT_IGNORE_ZERO_ENTRIES: 1093 a->ignorezeroentries = flg; 1094 break; 1095 case MAT_CHECK_COMPRESSED_ROW: 1096 a->compressedrow.check = flg; 1097 break; 1098 case MAT_SPD: 1099 case MAT_SYMMETRIC: 1100 case MAT_STRUCTURALLY_SYMMETRIC: 1101 case MAT_HERMITIAN: 1102 case MAT_SYMMETRY_ETERNAL: 1103 /* These options are handled directly by MatSetOption() */ 1104 break; 1105 case MAT_NEW_DIAGONALS: 1106 case MAT_IGNORE_OFF_PROC_ENTRIES: 1107 case MAT_USE_HASH_TABLE: 1108 ierr = PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);CHKERRQ(ierr); 1109 break; 1110 case MAT_USE_INODES: 1111 /* Not an error because MatSetOption_SeqAIJ_Inode handles this one */ 1112 break; 1113 default: 1114 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op); 1115 } 1116 ierr = MatSetOption_SeqAIJ_Inode(A,op,flg);CHKERRQ(ierr); 1117 PetscFunctionReturn(0); 1118 } 1119 1120 #undef __FUNCT__ 1121 #define __FUNCT__ "MatGetDiagonal_SeqAIJ" 1122 PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v) 1123 { 1124 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1125 PetscErrorCode ierr; 1126 PetscInt i,j,n,*ai=a->i,*aj=a->j,nz; 1127 PetscScalar *aa=a->a,*x,zero=0.0; 1128 1129 PetscFunctionBegin; 1130 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 1131 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 1132 1133 if (A->factortype == MAT_FACTOR_ILU || A->factortype == MAT_FACTOR_LU) { 1134 PetscInt *diag=a->diag; 1135 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 1136 for (i=0; i<n; i++) x[i] = 1.0/aa[diag[i]]; 1137 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 1138 PetscFunctionReturn(0); 1139 } 1140 1141 ierr = VecSet(v,zero);CHKERRQ(ierr); 1142 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 1143 for (i=0; i<n; i++) { 1144 nz = ai[i+1] - ai[i]; 1145 if (!nz) x[i] = 0.0; 1146 for (j=ai[i]; j<ai[i+1]; j++) { 1147 if (aj[j] == i) { 1148 x[i] = aa[j]; 1149 break; 1150 } 1151 } 1152 } 1153 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 1154 PetscFunctionReturn(0); 1155 } 1156 1157 #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h> 1158 #undef __FUNCT__ 1159 #define __FUNCT__ "MatMultTransposeAdd_SeqAIJ" 1160 PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy) 1161 { 1162 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1163 PetscScalar *x,*y; 1164 PetscErrorCode ierr; 1165 PetscInt m = A->rmap->n; 1166 #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ) 1167 MatScalar *v; 1168 PetscScalar alpha; 1169 PetscInt n,i,j,*idx,*ii,*ridx=NULL; 1170 Mat_CompressedRow cprow = a->compressedrow; 1171 PetscBool usecprow = cprow.use; 1172 #endif 1173 1174 PetscFunctionBegin; 1175 if (zz != yy) {ierr = VecCopy(zz,yy);CHKERRQ(ierr);} 1176 ierr = VecGetArray(xx,&x);CHKERRQ(ierr); 1177 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1178 1179 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ) 1180 fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y); 1181 #else 1182 if (usecprow) { 1183 m = cprow.nrows; 1184 ii = cprow.i; 1185 ridx = cprow.rindex; 1186 } else { 1187 ii = a->i; 1188 } 1189 for (i=0; i<m; i++) { 1190 idx = a->j + ii[i]; 1191 v = a->a + ii[i]; 1192 n = ii[i+1] - ii[i]; 1193 if (usecprow) { 1194 alpha = x[ridx[i]]; 1195 } else { 1196 alpha = x[i]; 1197 } 1198 for (j=0; j<n; j++) y[idx[j]] += alpha*v[j]; 1199 } 1200 #endif 1201 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1202 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1203 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1204 PetscFunctionReturn(0); 1205 } 1206 1207 #undef __FUNCT__ 1208 #define __FUNCT__ "MatMultTranspose_SeqAIJ" 1209 PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy) 1210 { 1211 PetscErrorCode ierr; 1212 1213 PetscFunctionBegin; 1214 ierr = VecSet(yy,0.0);CHKERRQ(ierr); 1215 ierr = MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);CHKERRQ(ierr); 1216 PetscFunctionReturn(0); 1217 } 1218 1219 #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h> 1220 #if defined(PETSC_THREADCOMM_ACTIVE) 1221 PetscErrorCode MatMult_SeqAIJ_Kernel(PetscInt thread_id,Mat A,Vec xx,Vec yy) 1222 { 1223 PetscErrorCode ierr; 1224 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1225 PetscScalar *y; 1226 const PetscScalar *x; 1227 const MatScalar *aa; 1228 PetscInt *trstarts=A->rmap->trstarts; 1229 PetscInt n,start,end,i; 1230 const PetscInt *aj,*ai; 1231 PetscScalar sum; 1232 1233 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1234 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1235 start = trstarts[thread_id]; 1236 end = trstarts[thread_id+1]; 1237 aj = a->j; 1238 aa = a->a; 1239 ai = a->i; 1240 for (i=start; i<end; i++) { 1241 n = ai[i+1] - ai[i]; 1242 aj = a->j + ai[i]; 1243 aa = a->a + ai[i]; 1244 sum = 0.0; 1245 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1246 y[i] = sum; 1247 } 1248 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1249 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1250 return 0; 1251 } 1252 1253 #undef __FUNCT__ 1254 #define __FUNCT__ "MatMult_SeqAIJ" 1255 PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy) 1256 { 1257 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1258 PetscScalar *y; 1259 const PetscScalar *x; 1260 const MatScalar *aa; 1261 PetscErrorCode ierr; 1262 PetscInt m=A->rmap->n; 1263 const PetscInt *aj,*ii,*ridx=NULL; 1264 PetscInt n,i,nonzerorow=0; 1265 PetscScalar sum; 1266 PetscBool usecprow=a->compressedrow.use; 1267 1268 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1269 #pragma disjoint(*x,*y,*aa) 1270 #endif 1271 1272 PetscFunctionBegin; 1273 aj = a->j; 1274 aa = a->a; 1275 ii = a->i; 1276 if (usecprow) { /* use compressed row format */ 1277 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1278 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1279 m = a->compressedrow.nrows; 1280 ii = a->compressedrow.i; 1281 ridx = a->compressedrow.rindex; 1282 for (i=0; i<m; i++) { 1283 n = ii[i+1] - ii[i]; 1284 aj = a->j + ii[i]; 1285 aa = a->a + ii[i]; 1286 sum = 0.0; 1287 nonzerorow += (n>0); 1288 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1289 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1290 y[*ridx++] = sum; 1291 } 1292 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1293 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1294 } else { /* do not use compressed row format */ 1295 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ) 1296 fortranmultaij_(&m,x,ii,aj,aa,y); 1297 #else 1298 ierr = PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatMult_SeqAIJ_Kernel,3,A,xx,yy);CHKERRQ(ierr); 1299 #endif 1300 } 1301 ierr = PetscLogFlops(2.0*a->nz - nonzerorow);CHKERRQ(ierr); 1302 PetscFunctionReturn(0); 1303 } 1304 #else 1305 #undef __FUNCT__ 1306 #define __FUNCT__ "MatMult_SeqAIJ" 1307 PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy) 1308 { 1309 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1310 PetscScalar *y; 1311 const PetscScalar *x; 1312 const MatScalar *aa; 1313 PetscErrorCode ierr; 1314 PetscInt m=A->rmap->n; 1315 const PetscInt *aj,*ii,*ridx=NULL; 1316 PetscInt n,i,nonzerorow=0; 1317 PetscScalar sum; 1318 PetscBool usecprow=a->compressedrow.use; 1319 1320 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1321 #pragma disjoint(*x,*y,*aa) 1322 #endif 1323 1324 PetscFunctionBegin; 1325 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1326 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1327 aj = a->j; 1328 aa = a->a; 1329 ii = a->i; 1330 if (usecprow) { /* use compressed row format */ 1331 m = a->compressedrow.nrows; 1332 ii = a->compressedrow.i; 1333 ridx = a->compressedrow.rindex; 1334 for (i=0; i<m; i++) { 1335 n = ii[i+1] - ii[i]; 1336 aj = a->j + ii[i]; 1337 aa = a->a + ii[i]; 1338 sum = 0.0; 1339 nonzerorow += (n>0); 1340 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1341 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1342 y[*ridx++] = sum; 1343 } 1344 } else { /* do not use compressed row format */ 1345 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ) 1346 fortranmultaij_(&m,x,ii,aj,aa,y); 1347 #else 1348 #if defined(PETSC_THREADCOMM_ACTIVE) 1349 ierr = PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatMult_SeqAIJ_Kernel,3,A,xx,yy);CHKERRQ(ierr); 1350 #else 1351 for (i=0; i<m; i++) { 1352 n = ii[i+1] - ii[i]; 1353 aj = a->j + ii[i]; 1354 aa = a->a + ii[i]; 1355 sum = 0.0; 1356 nonzerorow += (n>0); 1357 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1358 y[i] = sum; 1359 } 1360 #endif 1361 #endif 1362 } 1363 ierr = PetscLogFlops(2.0*a->nz - nonzerorow);CHKERRQ(ierr); 1364 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1365 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1366 PetscFunctionReturn(0); 1367 } 1368 #endif 1369 1370 #undef __FUNCT__ 1371 #define __FUNCT__ "MatMultMax_SeqAIJ" 1372 PetscErrorCode MatMultMax_SeqAIJ(Mat A,Vec xx,Vec yy) 1373 { 1374 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1375 PetscScalar *y; 1376 const PetscScalar *x; 1377 const MatScalar *aa; 1378 PetscErrorCode ierr; 1379 PetscInt m=A->rmap->n; 1380 const PetscInt *aj,*ii,*ridx=NULL; 1381 PetscInt n,i,nonzerorow=0; 1382 PetscScalar sum; 1383 PetscBool usecprow=a->compressedrow.use; 1384 1385 #if defined(PETSC_HAVE_PRAGMA_DISJOINT) 1386 #pragma disjoint(*x,*y,*aa) 1387 #endif 1388 1389 PetscFunctionBegin; 1390 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1391 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1392 aj = a->j; 1393 aa = a->a; 1394 ii = a->i; 1395 if (usecprow) { /* use compressed row format */ 1396 m = a->compressedrow.nrows; 1397 ii = a->compressedrow.i; 1398 ridx = a->compressedrow.rindex; 1399 for (i=0; i<m; i++) { 1400 n = ii[i+1] - ii[i]; 1401 aj = a->j + ii[i]; 1402 aa = a->a + ii[i]; 1403 sum = 0.0; 1404 nonzerorow += (n>0); 1405 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1406 /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */ 1407 y[*ridx++] = sum; 1408 } 1409 } else { /* do not use compressed row format */ 1410 for (i=0; i<m; i++) { 1411 n = ii[i+1] - ii[i]; 1412 aj = a->j + ii[i]; 1413 aa = a->a + ii[i]; 1414 sum = 0.0; 1415 nonzerorow += (n>0); 1416 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1417 y[i] = sum; 1418 } 1419 } 1420 ierr = PetscLogFlops(2.0*a->nz - nonzerorow);CHKERRQ(ierr); 1421 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1422 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1423 PetscFunctionReturn(0); 1424 } 1425 1426 #undef __FUNCT__ 1427 #define __FUNCT__ "MatMultAddMax_SeqAIJ" 1428 PetscErrorCode MatMultAddMax_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1429 { 1430 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1431 PetscScalar *y,*z; 1432 const PetscScalar *x; 1433 const MatScalar *aa; 1434 PetscErrorCode ierr; 1435 PetscInt m = A->rmap->n,*aj,*ii; 1436 PetscInt n,i,*ridx=NULL; 1437 PetscScalar sum; 1438 PetscBool usecprow=a->compressedrow.use; 1439 1440 PetscFunctionBegin; 1441 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1442 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1443 if (zz != yy) { 1444 ierr = VecGetArray(zz,&z);CHKERRQ(ierr); 1445 } else { 1446 z = y; 1447 } 1448 1449 aj = a->j; 1450 aa = a->a; 1451 ii = a->i; 1452 if (usecprow) { /* use compressed row format */ 1453 if (zz != yy) { 1454 ierr = PetscMemcpy(z,y,m*sizeof(PetscScalar));CHKERRQ(ierr); 1455 } 1456 m = a->compressedrow.nrows; 1457 ii = a->compressedrow.i; 1458 ridx = a->compressedrow.rindex; 1459 for (i=0; i<m; i++) { 1460 n = ii[i+1] - ii[i]; 1461 aj = a->j + ii[i]; 1462 aa = a->a + ii[i]; 1463 sum = y[*ridx]; 1464 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1465 z[*ridx++] = sum; 1466 } 1467 } else { /* do not use compressed row format */ 1468 for (i=0; i<m; i++) { 1469 n = ii[i+1] - ii[i]; 1470 aj = a->j + ii[i]; 1471 aa = a->a + ii[i]; 1472 sum = y[i]; 1473 PetscSparseDenseMaxDot(sum,x,aa,aj,n); 1474 z[i] = sum; 1475 } 1476 } 1477 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1478 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1479 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1480 if (zz != yy) { 1481 ierr = VecRestoreArray(zz,&z);CHKERRQ(ierr); 1482 } 1483 PetscFunctionReturn(0); 1484 } 1485 1486 #include <../src/mat/impls/aij/seq/ftn-kernels/fmultadd.h> 1487 #undef __FUNCT__ 1488 #define __FUNCT__ "MatMultAdd_SeqAIJ" 1489 PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz) 1490 { 1491 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1492 PetscScalar *y,*z; 1493 const PetscScalar *x; 1494 const MatScalar *aa; 1495 PetscErrorCode ierr; 1496 PetscInt m = A->rmap->n,*aj,*ii; 1497 PetscInt n,i,*ridx=NULL; 1498 PetscScalar sum; 1499 PetscBool usecprow=a->compressedrow.use; 1500 1501 PetscFunctionBegin; 1502 ierr = VecGetArrayRead(xx,&x);CHKERRQ(ierr); 1503 ierr = VecGetArray(yy,&y);CHKERRQ(ierr); 1504 if (zz != yy) { 1505 ierr = VecGetArray(zz,&z);CHKERRQ(ierr); 1506 } else { 1507 z = y; 1508 } 1509 1510 aj = a->j; 1511 aa = a->a; 1512 ii = a->i; 1513 if (usecprow) { /* use compressed row format */ 1514 if (zz != yy) { 1515 ierr = PetscMemcpy(z,y,m*sizeof(PetscScalar));CHKERRQ(ierr); 1516 } 1517 m = a->compressedrow.nrows; 1518 ii = a->compressedrow.i; 1519 ridx = a->compressedrow.rindex; 1520 for (i=0; i<m; i++) { 1521 n = ii[i+1] - ii[i]; 1522 aj = a->j + ii[i]; 1523 aa = a->a + ii[i]; 1524 sum = y[*ridx]; 1525 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1526 z[*ridx++] = sum; 1527 } 1528 } else { /* do not use compressed row format */ 1529 #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ) 1530 fortranmultaddaij_(&m,x,ii,aj,aa,y,z); 1531 #else 1532 for (i=0; i<m; i++) { 1533 n = ii[i+1] - ii[i]; 1534 aj = a->j + ii[i]; 1535 aa = a->a + ii[i]; 1536 sum = y[i]; 1537 PetscSparseDensePlusDot(sum,x,aa,aj,n); 1538 z[i] = sum; 1539 } 1540 #endif 1541 } 1542 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1543 ierr = VecRestoreArrayRead(xx,&x);CHKERRQ(ierr); 1544 ierr = VecRestoreArray(yy,&y);CHKERRQ(ierr); 1545 if (zz != yy) { 1546 ierr = VecRestoreArray(zz,&z);CHKERRQ(ierr); 1547 } 1548 #if defined(PETSC_HAVE_CUSP) 1549 /* 1550 ierr = VecView(xx,0);CHKERRQ(ierr); 1551 ierr = VecView(zz,0);CHKERRQ(ierr); 1552 ierr = MatView(A,0);CHKERRQ(ierr); 1553 */ 1554 #endif 1555 PetscFunctionReturn(0); 1556 } 1557 1558 /* 1559 Adds diagonal pointers to sparse matrix structure. 1560 */ 1561 #undef __FUNCT__ 1562 #define __FUNCT__ "MatMarkDiagonal_SeqAIJ" 1563 PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A) 1564 { 1565 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1566 PetscErrorCode ierr; 1567 PetscInt i,j,m = A->rmap->n; 1568 1569 PetscFunctionBegin; 1570 if (!a->diag) { 1571 ierr = PetscMalloc(m*sizeof(PetscInt),&a->diag);CHKERRQ(ierr); 1572 ierr = PetscLogObjectMemory((PetscObject)A, m*sizeof(PetscInt));CHKERRQ(ierr); 1573 } 1574 for (i=0; i<A->rmap->n; i++) { 1575 a->diag[i] = a->i[i+1]; 1576 for (j=a->i[i]; j<a->i[i+1]; j++) { 1577 if (a->j[j] == i) { 1578 a->diag[i] = j; 1579 break; 1580 } 1581 } 1582 } 1583 PetscFunctionReturn(0); 1584 } 1585 1586 /* 1587 Checks for missing diagonals 1588 */ 1589 #undef __FUNCT__ 1590 #define __FUNCT__ "MatMissingDiagonal_SeqAIJ" 1591 PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscBool *missing,PetscInt *d) 1592 { 1593 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1594 PetscInt *diag,*jj = a->j,i; 1595 1596 PetscFunctionBegin; 1597 *missing = PETSC_FALSE; 1598 if (A->rmap->n > 0 && !jj) { 1599 *missing = PETSC_TRUE; 1600 if (d) *d = 0; 1601 PetscInfo(A,"Matrix has no entries therefore is missing diagonal"); 1602 } else { 1603 diag = a->diag; 1604 for (i=0; i<A->rmap->n; i++) { 1605 if (jj[diag[i]] != i) { 1606 *missing = PETSC_TRUE; 1607 if (d) *d = i; 1608 PetscInfo1(A,"Matrix is missing diagonal number %D",i); 1609 break; 1610 } 1611 } 1612 } 1613 PetscFunctionReturn(0); 1614 } 1615 1616 #undef __FUNCT__ 1617 #define __FUNCT__ "MatInvertDiagonal_SeqAIJ" 1618 PetscErrorCode MatInvertDiagonal_SeqAIJ(Mat A,PetscScalar omega,PetscScalar fshift) 1619 { 1620 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 1621 PetscErrorCode ierr; 1622 PetscInt i,*diag,m = A->rmap->n; 1623 MatScalar *v = a->a; 1624 PetscScalar *idiag,*mdiag; 1625 1626 PetscFunctionBegin; 1627 if (a->idiagvalid) PetscFunctionReturn(0); 1628 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 1629 diag = a->diag; 1630 if (!a->idiag) { 1631 ierr = PetscMalloc3(m,PetscScalar,&a->idiag,m,PetscScalar,&a->mdiag,m,PetscScalar,&a->ssor_work);CHKERRQ(ierr); 1632 ierr = PetscLogObjectMemory((PetscObject)A, 3*m*sizeof(PetscScalar));CHKERRQ(ierr); 1633 v = a->a; 1634 } 1635 mdiag = a->mdiag; 1636 idiag = a->idiag; 1637 1638 if (omega == 1.0 && !PetscAbsScalar(fshift)) { 1639 for (i=0; i<m; i++) { 1640 mdiag[i] = v[diag[i]]; 1641 if (!PetscAbsScalar(mdiag[i])) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %D",i); 1642 idiag[i] = 1.0/v[diag[i]]; 1643 } 1644 ierr = PetscLogFlops(m);CHKERRQ(ierr); 1645 } else { 1646 for (i=0; i<m; i++) { 1647 mdiag[i] = v[diag[i]]; 1648 idiag[i] = omega/(fshift + v[diag[i]]); 1649 } 1650 ierr = PetscLogFlops(2.0*m);CHKERRQ(ierr); 1651 } 1652 a->idiagvalid = PETSC_TRUE; 1653 PetscFunctionReturn(0); 1654 } 1655 1656 #include <../src/mat/impls/aij/seq/ftn-kernels/frelax.h> 1657 #undef __FUNCT__ 1658 #define __FUNCT__ "MatSOR_SeqAIJ" 1659 PetscErrorCode MatSOR_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx) 1660 { 1661 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1662 PetscScalar *x,d,sum,*t,scale; 1663 const MatScalar *v = a->a,*idiag=0,*mdiag; 1664 const PetscScalar *b, *bs,*xb, *ts; 1665 PetscErrorCode ierr; 1666 PetscInt n = A->cmap->n,m = A->rmap->n,i; 1667 const PetscInt *idx,*diag; 1668 1669 PetscFunctionBegin; 1670 its = its*lits; 1671 1672 if (fshift != a->fshift || omega != a->omega) a->idiagvalid = PETSC_FALSE; /* must recompute idiag[] */ 1673 if (!a->idiagvalid) {ierr = MatInvertDiagonal_SeqAIJ(A,omega,fshift);CHKERRQ(ierr);} 1674 a->fshift = fshift; 1675 a->omega = omega; 1676 1677 diag = a->diag; 1678 t = a->ssor_work; 1679 idiag = a->idiag; 1680 mdiag = a->mdiag; 1681 1682 ierr = VecGetArray(xx,&x);CHKERRQ(ierr); 1683 ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr); 1684 /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */ 1685 if (flag == SOR_APPLY_UPPER) { 1686 /* apply (U + D/omega) to the vector */ 1687 bs = b; 1688 for (i=0; i<m; i++) { 1689 d = fshift + mdiag[i]; 1690 n = a->i[i+1] - diag[i] - 1; 1691 idx = a->j + diag[i] + 1; 1692 v = a->a + diag[i] + 1; 1693 sum = b[i]*d/omega; 1694 PetscSparseDensePlusDot(sum,bs,v,idx,n); 1695 x[i] = sum; 1696 } 1697 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1698 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1699 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 1700 PetscFunctionReturn(0); 1701 } 1702 1703 if (flag == SOR_APPLY_LOWER) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented"); 1704 else if (flag & SOR_EISENSTAT) { 1705 /* Let A = L + U + D; where L is lower trianglar, 1706 U is upper triangular, E = D/omega; This routine applies 1707 1708 (L + E)^{-1} A (U + E)^{-1} 1709 1710 to a vector efficiently using Eisenstat's trick. 1711 */ 1712 scale = (2.0/omega) - 1.0; 1713 1714 /* x = (E + U)^{-1} b */ 1715 for (i=m-1; i>=0; i--) { 1716 n = a->i[i+1] - diag[i] - 1; 1717 idx = a->j + diag[i] + 1; 1718 v = a->a + diag[i] + 1; 1719 sum = b[i]; 1720 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1721 x[i] = sum*idiag[i]; 1722 } 1723 1724 /* t = b - (2*E - D)x */ 1725 v = a->a; 1726 for (i=0; i<m; i++) t[i] = b[i] - scale*(v[*diag++])*x[i]; 1727 1728 /* t = (E + L)^{-1}t */ 1729 ts = t; 1730 diag = a->diag; 1731 for (i=0; i<m; i++) { 1732 n = diag[i] - a->i[i]; 1733 idx = a->j + a->i[i]; 1734 v = a->a + a->i[i]; 1735 sum = t[i]; 1736 PetscSparseDenseMinusDot(sum,ts,v,idx,n); 1737 t[i] = sum*idiag[i]; 1738 /* x = x + t */ 1739 x[i] += t[i]; 1740 } 1741 1742 ierr = PetscLogFlops(6.0*m-1 + 2.0*a->nz);CHKERRQ(ierr); 1743 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1744 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1745 PetscFunctionReturn(0); 1746 } 1747 if (flag & SOR_ZERO_INITIAL_GUESS) { 1748 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 1749 for (i=0; i<m; i++) { 1750 n = diag[i] - a->i[i]; 1751 idx = a->j + a->i[i]; 1752 v = a->a + a->i[i]; 1753 sum = b[i]; 1754 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1755 t[i] = sum; 1756 x[i] = sum*idiag[i]; 1757 } 1758 xb = t; 1759 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 1760 } else xb = b; 1761 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 1762 for (i=m-1; i>=0; i--) { 1763 n = a->i[i+1] - diag[i] - 1; 1764 idx = a->j + diag[i] + 1; 1765 v = a->a + diag[i] + 1; 1766 sum = xb[i]; 1767 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1768 if (xb == b) { 1769 x[i] = sum*idiag[i]; 1770 } else { 1771 x[i] = (1-omega)*x[i] + sum*idiag[i]; 1772 } 1773 } 1774 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 1775 } 1776 its--; 1777 } 1778 while (its--) { 1779 if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) { 1780 for (i=0; i<m; i++) { 1781 n = a->i[i+1] - a->i[i]; 1782 idx = a->j + a->i[i]; 1783 v = a->a + a->i[i]; 1784 sum = b[i]; 1785 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1786 x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i]; 1787 } 1788 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1789 } 1790 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 1791 for (i=m-1; i>=0; i--) { 1792 n = a->i[i+1] - a->i[i]; 1793 idx = a->j + a->i[i]; 1794 v = a->a + a->i[i]; 1795 sum = b[i]; 1796 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1797 x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i]; 1798 } 1799 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1800 } 1801 } 1802 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1803 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1804 PetscFunctionReturn(0); 1805 } 1806 1807 1808 #undef __FUNCT__ 1809 #define __FUNCT__ "MatGetInfo_SeqAIJ" 1810 PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info) 1811 { 1812 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1813 1814 PetscFunctionBegin; 1815 info->block_size = 1.0; 1816 info->nz_allocated = (double)a->maxnz; 1817 info->nz_used = (double)a->nz; 1818 info->nz_unneeded = (double)(a->maxnz - a->nz); 1819 info->assemblies = (double)A->num_ass; 1820 info->mallocs = (double)A->info.mallocs; 1821 info->memory = ((PetscObject)A)->mem; 1822 if (A->factortype) { 1823 info->fill_ratio_given = A->info.fill_ratio_given; 1824 info->fill_ratio_needed = A->info.fill_ratio_needed; 1825 info->factor_mallocs = A->info.factor_mallocs; 1826 } else { 1827 info->fill_ratio_given = 0; 1828 info->fill_ratio_needed = 0; 1829 info->factor_mallocs = 0; 1830 } 1831 PetscFunctionReturn(0); 1832 } 1833 1834 #undef __FUNCT__ 1835 #define __FUNCT__ "MatZeroRows_SeqAIJ" 1836 PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 1837 { 1838 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1839 PetscInt i,m = A->rmap->n - 1,d = 0; 1840 PetscErrorCode ierr; 1841 const PetscScalar *xx; 1842 PetscScalar *bb; 1843 PetscBool missing; 1844 1845 PetscFunctionBegin; 1846 if (x && b) { 1847 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 1848 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 1849 for (i=0; i<N; i++) { 1850 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1851 bb[rows[i]] = diag*xx[rows[i]]; 1852 } 1853 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 1854 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 1855 } 1856 1857 if (a->keepnonzeropattern) { 1858 for (i=0; i<N; i++) { 1859 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1860 ierr = PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));CHKERRQ(ierr); 1861 } 1862 if (diag != 0.0) { 1863 ierr = MatMissingDiagonal_SeqAIJ(A,&missing,&d);CHKERRQ(ierr); 1864 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d); 1865 for (i=0; i<N; i++) { 1866 a->a[a->diag[rows[i]]] = diag; 1867 } 1868 } 1869 A->same_nonzero = PETSC_TRUE; 1870 } else { 1871 if (diag != 0.0) { 1872 for (i=0; i<N; i++) { 1873 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1874 if (a->ilen[rows[i]] > 0) { 1875 a->ilen[rows[i]] = 1; 1876 a->a[a->i[rows[i]]] = diag; 1877 a->j[a->i[rows[i]]] = rows[i]; 1878 } else { /* in case row was completely empty */ 1879 ierr = MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);CHKERRQ(ierr); 1880 } 1881 } 1882 } else { 1883 for (i=0; i<N; i++) { 1884 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1885 a->ilen[rows[i]] = 0; 1886 } 1887 } 1888 A->same_nonzero = PETSC_FALSE; 1889 } 1890 ierr = MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1891 PetscFunctionReturn(0); 1892 } 1893 1894 #undef __FUNCT__ 1895 #define __FUNCT__ "MatZeroRowsColumns_SeqAIJ" 1896 PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 1897 { 1898 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1899 PetscInt i,j,m = A->rmap->n - 1,d = 0; 1900 PetscErrorCode ierr; 1901 PetscBool missing,*zeroed,vecs = PETSC_FALSE; 1902 const PetscScalar *xx; 1903 PetscScalar *bb; 1904 1905 PetscFunctionBegin; 1906 if (x && b) { 1907 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 1908 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 1909 vecs = PETSC_TRUE; 1910 } 1911 ierr = PetscMalloc(A->rmap->n*sizeof(PetscBool),&zeroed);CHKERRQ(ierr); 1912 ierr = PetscMemzero(zeroed,A->rmap->n*sizeof(PetscBool));CHKERRQ(ierr); 1913 for (i=0; i<N; i++) { 1914 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1915 ierr = PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));CHKERRQ(ierr); 1916 1917 zeroed[rows[i]] = PETSC_TRUE; 1918 } 1919 for (i=0; i<A->rmap->n; i++) { 1920 if (!zeroed[i]) { 1921 for (j=a->i[i]; j<a->i[i+1]; j++) { 1922 if (zeroed[a->j[j]]) { 1923 if (vecs) bb[i] -= a->a[j]*xx[a->j[j]]; 1924 a->a[j] = 0.0; 1925 } 1926 } 1927 } else if (vecs) bb[i] = diag*xx[i]; 1928 } 1929 if (x && b) { 1930 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 1931 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 1932 } 1933 ierr = PetscFree(zeroed);CHKERRQ(ierr); 1934 if (diag != 0.0) { 1935 ierr = MatMissingDiagonal_SeqAIJ(A,&missing,&d);CHKERRQ(ierr); 1936 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d); 1937 for (i=0; i<N; i++) { 1938 a->a[a->diag[rows[i]]] = diag; 1939 } 1940 } 1941 A->same_nonzero = PETSC_TRUE; 1942 ierr = MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1943 PetscFunctionReturn(0); 1944 } 1945 1946 #undef __FUNCT__ 1947 #define __FUNCT__ "MatGetRow_SeqAIJ" 1948 PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1949 { 1950 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1951 PetscInt *itmp; 1952 1953 PetscFunctionBegin; 1954 if (row < 0 || row >= A->rmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row); 1955 1956 *nz = a->i[row+1] - a->i[row]; 1957 if (v) *v = a->a + a->i[row]; 1958 if (idx) { 1959 itmp = a->j + a->i[row]; 1960 if (*nz) *idx = itmp; 1961 else *idx = 0; 1962 } 1963 PetscFunctionReturn(0); 1964 } 1965 1966 /* remove this function? */ 1967 #undef __FUNCT__ 1968 #define __FUNCT__ "MatRestoreRow_SeqAIJ" 1969 PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1970 { 1971 PetscFunctionBegin; 1972 PetscFunctionReturn(0); 1973 } 1974 1975 #undef __FUNCT__ 1976 #define __FUNCT__ "MatNorm_SeqAIJ" 1977 PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm) 1978 { 1979 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1980 MatScalar *v = a->a; 1981 PetscReal sum = 0.0; 1982 PetscErrorCode ierr; 1983 PetscInt i,j; 1984 1985 PetscFunctionBegin; 1986 if (type == NORM_FROBENIUS) { 1987 for (i=0; i<a->nz; i++) { 1988 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 1989 } 1990 *nrm = PetscSqrtReal(sum); 1991 } else if (type == NORM_1) { 1992 PetscReal *tmp; 1993 PetscInt *jj = a->j; 1994 ierr = PetscMalloc((A->cmap->n+1)*sizeof(PetscReal),&tmp);CHKERRQ(ierr); 1995 ierr = PetscMemzero(tmp,A->cmap->n*sizeof(PetscReal));CHKERRQ(ierr); 1996 *nrm = 0.0; 1997 for (j=0; j<a->nz; j++) { 1998 tmp[*jj++] += PetscAbsScalar(*v); v++; 1999 } 2000 for (j=0; j<A->cmap->n; j++) { 2001 if (tmp[j] > *nrm) *nrm = tmp[j]; 2002 } 2003 ierr = PetscFree(tmp);CHKERRQ(ierr); 2004 } else if (type == NORM_INFINITY) { 2005 *nrm = 0.0; 2006 for (j=0; j<A->rmap->n; j++) { 2007 v = a->a + a->i[j]; 2008 sum = 0.0; 2009 for (i=0; i<a->i[j+1]-a->i[j]; i++) { 2010 sum += PetscAbsScalar(*v); v++; 2011 } 2012 if (sum > *nrm) *nrm = sum; 2013 } 2014 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm"); 2015 PetscFunctionReturn(0); 2016 } 2017 2018 /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */ 2019 #undef __FUNCT__ 2020 #define __FUNCT__ "MatTransposeSymbolic_SeqAIJ" 2021 PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B) 2022 { 2023 PetscErrorCode ierr; 2024 PetscInt i,j,anzj; 2025 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b; 2026 PetscInt an=A->cmap->N,am=A->rmap->N; 2027 PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j; 2028 2029 PetscFunctionBegin; 2030 /* Allocate space for symbolic transpose info and work array */ 2031 ierr = PetscMalloc((an+1)*sizeof(PetscInt),&ati);CHKERRQ(ierr); 2032 ierr = PetscMalloc(ai[am]*sizeof(PetscInt),&atj);CHKERRQ(ierr); 2033 ierr = PetscMalloc(an*sizeof(PetscInt),&atfill);CHKERRQ(ierr); 2034 ierr = PetscMemzero(ati,(an+1)*sizeof(PetscInt));CHKERRQ(ierr); 2035 2036 /* Walk through aj and count ## of non-zeros in each row of A^T. */ 2037 /* Note: offset by 1 for fast conversion into csr format. */ 2038 for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1; 2039 /* Form ati for csr format of A^T. */ 2040 for (i=0;i<an;i++) ati[i+1] += ati[i]; 2041 2042 /* Copy ati into atfill so we have locations of the next free space in atj */ 2043 ierr = PetscMemcpy(atfill,ati,an*sizeof(PetscInt));CHKERRQ(ierr); 2044 2045 /* Walk through A row-wise and mark nonzero entries of A^T. */ 2046 for (i=0;i<am;i++) { 2047 anzj = ai[i+1] - ai[i]; 2048 for (j=0;j<anzj;j++) { 2049 atj[atfill[*aj]] = i; 2050 atfill[*aj++] += 1; 2051 } 2052 } 2053 2054 /* Clean up temporary space and complete requests. */ 2055 ierr = PetscFree(atfill);CHKERRQ(ierr); 2056 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B);CHKERRQ(ierr); 2057 2058 (*B)->rmap->bs = A->cmap->bs; 2059 (*B)->cmap->bs = A->rmap->bs; 2060 2061 b = (Mat_SeqAIJ*)((*B)->data); 2062 b->free_a = PETSC_FALSE; 2063 b->free_ij = PETSC_TRUE; 2064 b->nonew = 0; 2065 PetscFunctionReturn(0); 2066 } 2067 2068 #undef __FUNCT__ 2069 #define __FUNCT__ "MatTranspose_SeqAIJ" 2070 PetscErrorCode MatTranspose_SeqAIJ(Mat A,MatReuse reuse,Mat *B) 2071 { 2072 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2073 Mat C; 2074 PetscErrorCode ierr; 2075 PetscInt i,*aj = a->j,*ai = a->i,m = A->rmap->n,len,*col; 2076 MatScalar *array = a->a; 2077 2078 PetscFunctionBegin; 2079 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"); 2080 2081 if (reuse == MAT_INITIAL_MATRIX || *B == A) { 2082 ierr = PetscMalloc((1+A->cmap->n)*sizeof(PetscInt),&col);CHKERRQ(ierr); 2083 ierr = PetscMemzero(col,(1+A->cmap->n)*sizeof(PetscInt));CHKERRQ(ierr); 2084 2085 for (i=0; i<ai[m]; i++) col[aj[i]] += 1; 2086 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2087 ierr = MatSetSizes(C,A->cmap->n,m,A->cmap->n,m);CHKERRQ(ierr); 2088 ierr = MatSetBlockSizes(C,A->cmap->bs,A->rmap->bs);CHKERRQ(ierr); 2089 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2090 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,col);CHKERRQ(ierr); 2091 ierr = PetscFree(col);CHKERRQ(ierr); 2092 } else { 2093 C = *B; 2094 } 2095 2096 for (i=0; i<m; i++) { 2097 len = ai[i+1]-ai[i]; 2098 ierr = MatSetValues_SeqAIJ(C,len,aj,1,&i,array,INSERT_VALUES);CHKERRQ(ierr); 2099 array += len; 2100 aj += len; 2101 } 2102 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2103 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2104 2105 if (reuse == MAT_INITIAL_MATRIX || *B != A) { 2106 *B = C; 2107 } else { 2108 ierr = MatHeaderMerge(A,C);CHKERRQ(ierr); 2109 } 2110 PetscFunctionReturn(0); 2111 } 2112 2113 #undef __FUNCT__ 2114 #define __FUNCT__ "MatIsTranspose_SeqAIJ" 2115 PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2116 { 2117 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data; 2118 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2119 MatScalar *va,*vb; 2120 PetscErrorCode ierr; 2121 PetscInt ma,na,mb,nb, i; 2122 2123 PetscFunctionBegin; 2124 bij = (Mat_SeqAIJ*) B->data; 2125 2126 ierr = MatGetSize(A,&ma,&na);CHKERRQ(ierr); 2127 ierr = MatGetSize(B,&mb,&nb);CHKERRQ(ierr); 2128 if (ma!=nb || na!=mb) { 2129 *f = PETSC_FALSE; 2130 PetscFunctionReturn(0); 2131 } 2132 aii = aij->i; bii = bij->i; 2133 adx = aij->j; bdx = bij->j; 2134 va = aij->a; vb = bij->a; 2135 ierr = PetscMalloc(ma*sizeof(PetscInt),&aptr);CHKERRQ(ierr); 2136 ierr = PetscMalloc(mb*sizeof(PetscInt),&bptr);CHKERRQ(ierr); 2137 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2138 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2139 2140 *f = PETSC_TRUE; 2141 for (i=0; i<ma; i++) { 2142 while (aptr[i]<aii[i+1]) { 2143 PetscInt idc,idr; 2144 PetscScalar vc,vr; 2145 /* column/row index/value */ 2146 idc = adx[aptr[i]]; 2147 idr = bdx[bptr[idc]]; 2148 vc = va[aptr[i]]; 2149 vr = vb[bptr[idc]]; 2150 if (i!=idr || PetscAbsScalar(vc-vr) > tol) { 2151 *f = PETSC_FALSE; 2152 goto done; 2153 } else { 2154 aptr[i]++; 2155 if (B || i!=idc) bptr[idc]++; 2156 } 2157 } 2158 } 2159 done: 2160 ierr = PetscFree(aptr);CHKERRQ(ierr); 2161 ierr = PetscFree(bptr);CHKERRQ(ierr); 2162 PetscFunctionReturn(0); 2163 } 2164 2165 #undef __FUNCT__ 2166 #define __FUNCT__ "MatIsHermitianTranspose_SeqAIJ" 2167 PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2168 { 2169 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data; 2170 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2171 MatScalar *va,*vb; 2172 PetscErrorCode ierr; 2173 PetscInt ma,na,mb,nb, i; 2174 2175 PetscFunctionBegin; 2176 bij = (Mat_SeqAIJ*) B->data; 2177 2178 ierr = MatGetSize(A,&ma,&na);CHKERRQ(ierr); 2179 ierr = MatGetSize(B,&mb,&nb);CHKERRQ(ierr); 2180 if (ma!=nb || na!=mb) { 2181 *f = PETSC_FALSE; 2182 PetscFunctionReturn(0); 2183 } 2184 aii = aij->i; bii = bij->i; 2185 adx = aij->j; bdx = bij->j; 2186 va = aij->a; vb = bij->a; 2187 ierr = PetscMalloc(ma*sizeof(PetscInt),&aptr);CHKERRQ(ierr); 2188 ierr = PetscMalloc(mb*sizeof(PetscInt),&bptr);CHKERRQ(ierr); 2189 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2190 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2191 2192 *f = PETSC_TRUE; 2193 for (i=0; i<ma; i++) { 2194 while (aptr[i]<aii[i+1]) { 2195 PetscInt idc,idr; 2196 PetscScalar vc,vr; 2197 /* column/row index/value */ 2198 idc = adx[aptr[i]]; 2199 idr = bdx[bptr[idc]]; 2200 vc = va[aptr[i]]; 2201 vr = vb[bptr[idc]]; 2202 if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) { 2203 *f = PETSC_FALSE; 2204 goto done; 2205 } else { 2206 aptr[i]++; 2207 if (B || i!=idc) bptr[idc]++; 2208 } 2209 } 2210 } 2211 done: 2212 ierr = PetscFree(aptr);CHKERRQ(ierr); 2213 ierr = PetscFree(bptr);CHKERRQ(ierr); 2214 PetscFunctionReturn(0); 2215 } 2216 2217 #undef __FUNCT__ 2218 #define __FUNCT__ "MatIsSymmetric_SeqAIJ" 2219 PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2220 { 2221 PetscErrorCode ierr; 2222 2223 PetscFunctionBegin; 2224 ierr = MatIsTranspose_SeqAIJ(A,A,tol,f);CHKERRQ(ierr); 2225 PetscFunctionReturn(0); 2226 } 2227 2228 #undef __FUNCT__ 2229 #define __FUNCT__ "MatIsHermitian_SeqAIJ" 2230 PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2231 { 2232 PetscErrorCode ierr; 2233 2234 PetscFunctionBegin; 2235 ierr = MatIsHermitianTranspose_SeqAIJ(A,A,tol,f);CHKERRQ(ierr); 2236 PetscFunctionReturn(0); 2237 } 2238 2239 #undef __FUNCT__ 2240 #define __FUNCT__ "MatDiagonalScale_SeqAIJ" 2241 PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr) 2242 { 2243 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2244 PetscScalar *l,*r,x; 2245 MatScalar *v; 2246 PetscErrorCode ierr; 2247 PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz,*jj; 2248 2249 PetscFunctionBegin; 2250 if (ll) { 2251 /* The local size is used so that VecMPI can be passed to this routine 2252 by MatDiagonalScale_MPIAIJ */ 2253 ierr = VecGetLocalSize(ll,&m);CHKERRQ(ierr); 2254 if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length"); 2255 ierr = VecGetArray(ll,&l);CHKERRQ(ierr); 2256 v = a->a; 2257 for (i=0; i<m; i++) { 2258 x = l[i]; 2259 M = a->i[i+1] - a->i[i]; 2260 for (j=0; j<M; j++) (*v++) *= x; 2261 } 2262 ierr = VecRestoreArray(ll,&l);CHKERRQ(ierr); 2263 ierr = PetscLogFlops(nz);CHKERRQ(ierr); 2264 } 2265 if (rr) { 2266 ierr = VecGetLocalSize(rr,&n);CHKERRQ(ierr); 2267 if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length"); 2268 ierr = VecGetArray(rr,&r);CHKERRQ(ierr); 2269 v = a->a; jj = a->j; 2270 for (i=0; i<nz; i++) (*v++) *= r[*jj++]; 2271 ierr = VecRestoreArray(rr,&r);CHKERRQ(ierr); 2272 ierr = PetscLogFlops(nz);CHKERRQ(ierr); 2273 } 2274 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 2275 PetscFunctionReturn(0); 2276 } 2277 2278 #undef __FUNCT__ 2279 #define __FUNCT__ "MatGetSubMatrix_SeqAIJ" 2280 PetscErrorCode MatGetSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B) 2281 { 2282 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c; 2283 PetscErrorCode ierr; 2284 PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens; 2285 PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi; 2286 const PetscInt *irow,*icol; 2287 PetscInt nrows,ncols; 2288 PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen; 2289 MatScalar *a_new,*mat_a; 2290 Mat C; 2291 PetscBool stride,sorted; 2292 2293 PetscFunctionBegin; 2294 ierr = ISSorted(isrow,&sorted);CHKERRQ(ierr); 2295 if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"ISrow is not sorted"); 2296 ierr = ISSorted(iscol,&sorted);CHKERRQ(ierr); 2297 if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"IScol is not sorted"); 2298 2299 ierr = ISGetIndices(isrow,&irow);CHKERRQ(ierr); 2300 ierr = ISGetLocalSize(isrow,&nrows);CHKERRQ(ierr); 2301 ierr = ISGetLocalSize(iscol,&ncols);CHKERRQ(ierr); 2302 2303 ierr = ISStrideGetInfo(iscol,&first,&step);CHKERRQ(ierr); 2304 ierr = PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride);CHKERRQ(ierr); 2305 if (stride && step == 1) { 2306 /* special case of contiguous rows */ 2307 ierr = PetscMalloc2(nrows,PetscInt,&lens,nrows,PetscInt,&starts);CHKERRQ(ierr); 2308 /* loop over new rows determining lens and starting points */ 2309 for (i=0; i<nrows; i++) { 2310 kstart = ai[irow[i]]; 2311 kend = kstart + ailen[irow[i]]; 2312 for (k=kstart; k<kend; k++) { 2313 if (aj[k] >= first) { 2314 starts[i] = k; 2315 break; 2316 } 2317 } 2318 sum = 0; 2319 while (k < kend) { 2320 if (aj[k++] >= first+ncols) break; 2321 sum++; 2322 } 2323 lens[i] = sum; 2324 } 2325 /* create submatrix */ 2326 if (scall == MAT_REUSE_MATRIX) { 2327 PetscInt n_cols,n_rows; 2328 ierr = MatGetSize(*B,&n_rows,&n_cols);CHKERRQ(ierr); 2329 if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size"); 2330 ierr = MatZeroEntries(*B);CHKERRQ(ierr); 2331 C = *B; 2332 } else { 2333 PetscInt rbs,cbs; 2334 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2335 ierr = MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2336 ierr = ISGetBlockSize(isrow,&rbs);CHKERRQ(ierr); 2337 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2338 ierr = MatSetBlockSizes(C,rbs,cbs);CHKERRQ(ierr); 2339 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2340 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);CHKERRQ(ierr); 2341 } 2342 c = (Mat_SeqAIJ*)C->data; 2343 2344 /* loop over rows inserting into submatrix */ 2345 a_new = c->a; 2346 j_new = c->j; 2347 i_new = c->i; 2348 2349 for (i=0; i<nrows; i++) { 2350 ii = starts[i]; 2351 lensi = lens[i]; 2352 for (k=0; k<lensi; k++) { 2353 *j_new++ = aj[ii+k] - first; 2354 } 2355 ierr = PetscMemcpy(a_new,a->a + starts[i],lensi*sizeof(PetscScalar));CHKERRQ(ierr); 2356 a_new += lensi; 2357 i_new[i+1] = i_new[i] + lensi; 2358 c->ilen[i] = lensi; 2359 } 2360 ierr = PetscFree2(lens,starts);CHKERRQ(ierr); 2361 } else { 2362 ierr = ISGetIndices(iscol,&icol);CHKERRQ(ierr); 2363 ierr = PetscMalloc(oldcols*sizeof(PetscInt),&smap);CHKERRQ(ierr); 2364 ierr = PetscMemzero(smap,oldcols*sizeof(PetscInt));CHKERRQ(ierr); 2365 ierr = PetscMalloc((1+nrows)*sizeof(PetscInt),&lens);CHKERRQ(ierr); 2366 for (i=0; i<ncols; i++) { 2367 #if defined(PETSC_USE_DEBUG) 2368 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); 2369 #endif 2370 smap[icol[i]] = i+1; 2371 } 2372 2373 /* determine lens of each row */ 2374 for (i=0; i<nrows; i++) { 2375 kstart = ai[irow[i]]; 2376 kend = kstart + a->ilen[irow[i]]; 2377 lens[i] = 0; 2378 for (k=kstart; k<kend; k++) { 2379 if (smap[aj[k]]) { 2380 lens[i]++; 2381 } 2382 } 2383 } 2384 /* Create and fill new matrix */ 2385 if (scall == MAT_REUSE_MATRIX) { 2386 PetscBool equal; 2387 2388 c = (Mat_SeqAIJ*)((*B)->data); 2389 if ((*B)->rmap->n != nrows || (*B)->cmap->n != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size"); 2390 ierr = PetscMemcmp(c->ilen,lens,(*B)->rmap->n*sizeof(PetscInt),&equal);CHKERRQ(ierr); 2391 if (!equal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros"); 2392 ierr = PetscMemzero(c->ilen,(*B)->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 2393 C = *B; 2394 } else { 2395 PetscInt rbs,cbs; 2396 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2397 ierr = MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2398 ierr = ISGetBlockSize(isrow,&rbs);CHKERRQ(ierr); 2399 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2400 ierr = MatSetBlockSizes(C,rbs,cbs);CHKERRQ(ierr); 2401 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2402 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);CHKERRQ(ierr); 2403 } 2404 c = (Mat_SeqAIJ*)(C->data); 2405 for (i=0; i<nrows; i++) { 2406 row = irow[i]; 2407 kstart = ai[row]; 2408 kend = kstart + a->ilen[row]; 2409 mat_i = c->i[i]; 2410 mat_j = c->j + mat_i; 2411 mat_a = c->a + mat_i; 2412 mat_ilen = c->ilen + i; 2413 for (k=kstart; k<kend; k++) { 2414 if ((tcol=smap[a->j[k]])) { 2415 *mat_j++ = tcol - 1; 2416 *mat_a++ = a->a[k]; 2417 (*mat_ilen)++; 2418 2419 } 2420 } 2421 } 2422 /* Free work space */ 2423 ierr = ISRestoreIndices(iscol,&icol);CHKERRQ(ierr); 2424 ierr = PetscFree(smap);CHKERRQ(ierr); 2425 ierr = PetscFree(lens);CHKERRQ(ierr); 2426 } 2427 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2428 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2429 2430 ierr = ISRestoreIndices(isrow,&irow);CHKERRQ(ierr); 2431 *B = C; 2432 PetscFunctionReturn(0); 2433 } 2434 2435 #undef __FUNCT__ 2436 #define __FUNCT__ "MatGetMultiProcBlock_SeqAIJ" 2437 PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat) 2438 { 2439 PetscErrorCode ierr; 2440 Mat B; 2441 2442 PetscFunctionBegin; 2443 if (scall == MAT_INITIAL_MATRIX) { 2444 ierr = MatCreate(subComm,&B);CHKERRQ(ierr); 2445 ierr = MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n);CHKERRQ(ierr); 2446 ierr = MatSetBlockSizes(B,mat->rmap->bs,mat->cmap->bs);CHKERRQ(ierr); 2447 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 2448 ierr = MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE);CHKERRQ(ierr); 2449 *subMat = B; 2450 } else { 2451 ierr = MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 2452 } 2453 PetscFunctionReturn(0); 2454 } 2455 2456 #undef __FUNCT__ 2457 #define __FUNCT__ "MatILUFactor_SeqAIJ" 2458 PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info) 2459 { 2460 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2461 PetscErrorCode ierr; 2462 Mat outA; 2463 PetscBool row_identity,col_identity; 2464 2465 PetscFunctionBegin; 2466 if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu"); 2467 2468 ierr = ISIdentity(row,&row_identity);CHKERRQ(ierr); 2469 ierr = ISIdentity(col,&col_identity);CHKERRQ(ierr); 2470 2471 outA = inA; 2472 outA->factortype = MAT_FACTOR_LU; 2473 2474 ierr = PetscObjectReference((PetscObject)row);CHKERRQ(ierr); 2475 ierr = ISDestroy(&a->row);CHKERRQ(ierr); 2476 2477 a->row = row; 2478 2479 ierr = PetscObjectReference((PetscObject)col);CHKERRQ(ierr); 2480 ierr = ISDestroy(&a->col);CHKERRQ(ierr); 2481 2482 a->col = col; 2483 2484 /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */ 2485 ierr = ISDestroy(&a->icol);CHKERRQ(ierr); 2486 ierr = ISInvertPermutation(col,PETSC_DECIDE,&a->icol);CHKERRQ(ierr); 2487 ierr = PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);CHKERRQ(ierr); 2488 2489 if (!a->solve_work) { /* this matrix may have been factored before */ 2490 ierr = PetscMalloc((inA->rmap->n+1)*sizeof(PetscScalar),&a->solve_work);CHKERRQ(ierr); 2491 ierr = PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar));CHKERRQ(ierr); 2492 } 2493 2494 ierr = MatMarkDiagonal_SeqAIJ(inA);CHKERRQ(ierr); 2495 if (row_identity && col_identity) { 2496 ierr = MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info);CHKERRQ(ierr); 2497 } else { 2498 ierr = MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info);CHKERRQ(ierr); 2499 } 2500 PetscFunctionReturn(0); 2501 } 2502 2503 #undef __FUNCT__ 2504 #define __FUNCT__ "MatScale_SeqAIJ" 2505 PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha) 2506 { 2507 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2508 PetscScalar oalpha = alpha; 2509 PetscErrorCode ierr; 2510 PetscBLASInt one = 1,bnz; 2511 2512 PetscFunctionBegin; 2513 ierr = PetscBLASIntCast(a->nz,&bnz);CHKERRQ(ierr); 2514 PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&oalpha,a->a,&one)); 2515 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 2516 ierr = MatSeqAIJInvalidateDiagonal(inA);CHKERRQ(ierr); 2517 PetscFunctionReturn(0); 2518 } 2519 2520 #undef __FUNCT__ 2521 #define __FUNCT__ "MatGetSubMatrices_SeqAIJ" 2522 PetscErrorCode MatGetSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[]) 2523 { 2524 PetscErrorCode ierr; 2525 PetscInt i; 2526 2527 PetscFunctionBegin; 2528 if (scall == MAT_INITIAL_MATRIX) { 2529 ierr = PetscMalloc((n+1)*sizeof(Mat),B);CHKERRQ(ierr); 2530 } 2531 2532 for (i=0; i<n; i++) { 2533 ierr = MatGetSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);CHKERRQ(ierr); 2534 } 2535 PetscFunctionReturn(0); 2536 } 2537 2538 #undef __FUNCT__ 2539 #define __FUNCT__ "MatIncreaseOverlap_SeqAIJ" 2540 PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov) 2541 { 2542 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2543 PetscErrorCode ierr; 2544 PetscInt row,i,j,k,l,m,n,*nidx,isz,val; 2545 const PetscInt *idx; 2546 PetscInt start,end,*ai,*aj; 2547 PetscBT table; 2548 2549 PetscFunctionBegin; 2550 m = A->rmap->n; 2551 ai = a->i; 2552 aj = a->j; 2553 2554 if (ov < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used"); 2555 2556 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&nidx);CHKERRQ(ierr); 2557 ierr = PetscBTCreate(m,&table);CHKERRQ(ierr); 2558 2559 for (i=0; i<is_max; i++) { 2560 /* Initialize the two local arrays */ 2561 isz = 0; 2562 ierr = PetscBTMemzero(m,table);CHKERRQ(ierr); 2563 2564 /* Extract the indices, assume there can be duplicate entries */ 2565 ierr = ISGetIndices(is[i],&idx);CHKERRQ(ierr); 2566 ierr = ISGetLocalSize(is[i],&n);CHKERRQ(ierr); 2567 2568 /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */ 2569 for (j=0; j<n; ++j) { 2570 if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j]; 2571 } 2572 ierr = ISRestoreIndices(is[i],&idx);CHKERRQ(ierr); 2573 ierr = ISDestroy(&is[i]);CHKERRQ(ierr); 2574 2575 k = 0; 2576 for (j=0; j<ov; j++) { /* for each overlap */ 2577 n = isz; 2578 for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */ 2579 row = nidx[k]; 2580 start = ai[row]; 2581 end = ai[row+1]; 2582 for (l = start; l<end; l++) { 2583 val = aj[l]; 2584 if (!PetscBTLookupSet(table,val)) nidx[isz++] = val; 2585 } 2586 } 2587 } 2588 ierr = ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i));CHKERRQ(ierr); 2589 } 2590 ierr = PetscBTDestroy(&table);CHKERRQ(ierr); 2591 ierr = PetscFree(nidx);CHKERRQ(ierr); 2592 PetscFunctionReturn(0); 2593 } 2594 2595 /* -------------------------------------------------------------- */ 2596 #undef __FUNCT__ 2597 #define __FUNCT__ "MatPermute_SeqAIJ" 2598 PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B) 2599 { 2600 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2601 PetscErrorCode ierr; 2602 PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n; 2603 const PetscInt *row,*col; 2604 PetscInt *cnew,j,*lens; 2605 IS icolp,irowp; 2606 PetscInt *cwork = NULL; 2607 PetscScalar *vwork = NULL; 2608 2609 PetscFunctionBegin; 2610 ierr = ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);CHKERRQ(ierr); 2611 ierr = ISGetIndices(irowp,&row);CHKERRQ(ierr); 2612 ierr = ISInvertPermutation(colp,PETSC_DECIDE,&icolp);CHKERRQ(ierr); 2613 ierr = ISGetIndices(icolp,&col);CHKERRQ(ierr); 2614 2615 /* determine lengths of permuted rows */ 2616 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&lens);CHKERRQ(ierr); 2617 for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i]; 2618 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 2619 ierr = MatSetSizes(*B,m,n,m,n);CHKERRQ(ierr); 2620 ierr = MatSetBlockSizes(*B,A->rmap->bs,A->cmap->bs);CHKERRQ(ierr); 2621 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 2622 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);CHKERRQ(ierr); 2623 ierr = PetscFree(lens);CHKERRQ(ierr); 2624 2625 ierr = PetscMalloc(n*sizeof(PetscInt),&cnew);CHKERRQ(ierr); 2626 for (i=0; i<m; i++) { 2627 ierr = MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);CHKERRQ(ierr); 2628 for (j=0; j<nz; j++) cnew[j] = col[cwork[j]]; 2629 ierr = MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);CHKERRQ(ierr); 2630 ierr = MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);CHKERRQ(ierr); 2631 } 2632 ierr = PetscFree(cnew);CHKERRQ(ierr); 2633 2634 (*B)->assembled = PETSC_FALSE; 2635 2636 ierr = MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2637 ierr = MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2638 ierr = ISRestoreIndices(irowp,&row);CHKERRQ(ierr); 2639 ierr = ISRestoreIndices(icolp,&col);CHKERRQ(ierr); 2640 ierr = ISDestroy(&irowp);CHKERRQ(ierr); 2641 ierr = ISDestroy(&icolp);CHKERRQ(ierr); 2642 PetscFunctionReturn(0); 2643 } 2644 2645 #undef __FUNCT__ 2646 #define __FUNCT__ "MatCopy_SeqAIJ" 2647 PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str) 2648 { 2649 PetscErrorCode ierr; 2650 2651 PetscFunctionBegin; 2652 /* If the two matrices have the same copy implementation, use fast copy. */ 2653 if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) { 2654 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2655 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 2656 2657 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"); 2658 ierr = PetscMemcpy(b->a,a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));CHKERRQ(ierr); 2659 } else { 2660 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 2661 } 2662 PetscFunctionReturn(0); 2663 } 2664 2665 #undef __FUNCT__ 2666 #define __FUNCT__ "MatSetUp_SeqAIJ" 2667 PetscErrorCode MatSetUp_SeqAIJ(Mat A) 2668 { 2669 PetscErrorCode ierr; 2670 2671 PetscFunctionBegin; 2672 ierr = MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,0);CHKERRQ(ierr); 2673 PetscFunctionReturn(0); 2674 } 2675 2676 #undef __FUNCT__ 2677 #define __FUNCT__ "MatSeqAIJGetArray_SeqAIJ" 2678 PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[]) 2679 { 2680 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2681 2682 PetscFunctionBegin; 2683 *array = a->a; 2684 PetscFunctionReturn(0); 2685 } 2686 2687 #undef __FUNCT__ 2688 #define __FUNCT__ "MatSeqAIJRestoreArray_SeqAIJ" 2689 PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[]) 2690 { 2691 PetscFunctionBegin; 2692 PetscFunctionReturn(0); 2693 } 2694 2695 /* Optimize MatFDColoringApply_AIJ() by using array den2sp to skip calling MatSetValues() */ 2696 /* #define JACOBIANCOLOROPT */ 2697 #if defined(JACOBIANCOLOROPT) 2698 #include <petsctime.h> 2699 #endif 2700 #undef __FUNCT__ 2701 #define __FUNCT__ "MatFDColoringApply_SeqAIJ" 2702 PetscErrorCode MatFDColoringApply_SeqAIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 2703 { 2704 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 2705 PetscErrorCode ierr; 2706 PetscInt k,start,end,l,row,col,**vscaleforrow; 2707 PetscScalar dx,*y,*xx,*w3_array; 2708 PetscScalar *vscale_array; 2709 PetscReal epsilon = coloring->error_rel,umin = coloring->umin,unorm; 2710 Vec w1 = coloring->w1,w2=coloring->w2,w3; 2711 void *fctx = coloring->fctx; 2712 PetscBool flg = PETSC_FALSE; 2713 PetscInt ctype = coloring->ctype,N,col_start=0,col_end=0; 2714 Vec x1_tmp; 2715 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)J->data; 2716 PetscInt *den2sp=coloring->den2sp,*idx=den2sp; 2717 PetscScalar *ca=csp->a; 2718 const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows; 2719 PetscInt **rows=coloring->rows,**columns=coloring->columns,ncolumns_k,nrows_k; 2720 #if defined(JACOBIANCOLOROPT) 2721 PetscLogDouble t0,t1,time_setvalues=0.0; 2722 #endif 2723 2724 PetscFunctionBegin; 2725 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 2726 ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 2727 if (flg) { 2728 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 2729 } else { 2730 PetscBool assembled; 2731 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 2732 if (assembled) { 2733 ierr = MatZeroEntries(J);CHKERRQ(ierr); 2734 } 2735 } 2736 2737 x1_tmp = x1; 2738 if (!coloring->vscale) { 2739 ierr = VecDuplicate(x1_tmp,&coloring->vscale);CHKERRQ(ierr); 2740 } 2741 2742 if (coloring->htype[0] == 'w') { /* tacky test; need to make systematic if we add other approaches to computing h*/ 2743 ierr = VecNorm(x1_tmp,NORM_2,&unorm);CHKERRQ(ierr); 2744 } 2745 ierr = VecGetOwnershipRange(w1,&start,&end);CHKERRQ(ierr); /* OwnershipRange is used by ghosted x! */ 2746 2747 /* Set w1 = F(x1) */ 2748 if (!coloring->fset) { 2749 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2750 ierr = (*f)(sctx,x1_tmp,w1,fctx);CHKERRQ(ierr); 2751 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2752 } else { 2753 coloring->fset = PETSC_FALSE; 2754 } 2755 2756 if (!coloring->w3) { 2757 ierr = VecDuplicate(x1_tmp,&coloring->w3);CHKERRQ(ierr); 2758 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 2759 } 2760 w3 = coloring->w3; 2761 2762 /* Compute all the local scale factors, including ghost points */ 2763 ierr = VecGetLocalSize(x1_tmp,&N);CHKERRQ(ierr); 2764 ierr = VecGetArray(x1_tmp,&xx);CHKERRQ(ierr); 2765 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2766 if (ctype == IS_COLORING_GHOSTED) { 2767 col_start = 0; col_end = N; 2768 } else if (ctype == IS_COLORING_GLOBAL) { 2769 xx = xx - start; 2770 vscale_array = vscale_array - start; 2771 col_start = start; col_end = N + start; 2772 } 2773 for (col=col_start; col<col_end; col++) { 2774 /* Loop over each local column, vscale[col] = 1./(epsilon*dx[col]) */ 2775 if (coloring->htype[0] == 'w') { 2776 dx = 1.0 + unorm; 2777 } else { 2778 dx = xx[col]; 2779 } 2780 if (dx == (PetscScalar)0.0) dx = 1.0; 2781 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2782 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2783 dx *= epsilon; 2784 vscale_array[col] = (PetscScalar)1.0/dx; 2785 } 2786 if (ctype == IS_COLORING_GLOBAL) vscale_array = vscale_array + start; 2787 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2788 if (ctype == IS_COLORING_GLOBAL) { 2789 ierr = VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2790 ierr = VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2791 } 2792 2793 if (coloring->vscaleforrow) { 2794 vscaleforrow = coloring->vscaleforrow; 2795 } else SETERRQ(PetscObjectComm((PetscObject)J),PETSC_ERR_ARG_NULL,"Null Object: coloring->vscaleforrow"); 2796 2797 /* 2798 Loop over each color 2799 */ 2800 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2801 for (k=0; k<ncolors; k++) { /* loop over colors */ 2802 coloring->currentcolor = k; 2803 2804 ierr = VecCopy(x1_tmp,w3);CHKERRQ(ierr); 2805 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 2806 if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array - start; 2807 /* 2808 Loop over each column associated with color 2809 adding the perturbation to the vector w3. 2810 */ 2811 ncolumns_k = ncolumns[k]; 2812 for (l=0; l<ncolumns_k; l++) { /* loop over columns */ 2813 col = columns[k][l]; /* local column of the matrix we are probing for */ 2814 if (coloring->htype[0] == 'w') { 2815 dx = 1.0 + unorm; 2816 } else { 2817 dx = xx[col]; 2818 } 2819 if (dx == (PetscScalar)0.0) dx = 1.0; 2820 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2821 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2822 dx *= epsilon; 2823 if (!PetscAbsScalar(dx)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Computed 0 differencing parameter"); 2824 w3_array[col] += dx; 2825 } 2826 if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array + start; 2827 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 2828 2829 /* 2830 Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 2831 w2 = F(x1 + dx) - F(x1) 2832 */ 2833 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2834 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 2835 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2836 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 2837 2838 /* 2839 Loop over rows of vector, putting results into Jacobian matrix 2840 */ 2841 #if defined(JACOBIANCOLOROPT) 2842 ierr = PetscTime(&t0);CHKERRQ(ierr); 2843 #endif 2844 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 2845 nrows_k = nrows[k]; 2846 for (l=0; l<nrows_k; l++) { /* loop over rows */ 2847 row = rows[k][l]; /* local row index */ 2848 y[row] *= vscale_array[vscaleforrow[k][l]]; 2849 ca[idx[l]] = y[row]; 2850 } 2851 idx += nrows_k; 2852 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 2853 #if defined(JACOBIANCOLOROPT) 2854 ierr = PetscTime(&t1);CHKERRQ(ierr); 2855 time_setvalues += t1-t0; 2856 #endif 2857 } /* endof for each color */ 2858 #if defined(JACOBIANCOLOROPT) 2859 printf(" MatFDColoringApply_SeqAIJ: time_setvalues %g\n",time_setvalues); 2860 #endif 2861 if (ctype == IS_COLORING_GLOBAL) xx = xx + start; 2862 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2863 ierr = VecRestoreArray(x1_tmp,&xx);CHKERRQ(ierr); 2864 2865 coloring->currentcolor = -1; 2866 PetscFunctionReturn(0); 2867 } 2868 /* --------------------------------------------------------*/ 2869 2870 #undef __FUNCT__ 2871 #define __FUNCT__ "MatFDColoringApply_SeqAIJ_old" 2872 PetscErrorCode MatFDColoringApply_SeqAIJ_old(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 2873 { 2874 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 2875 PetscErrorCode ierr; 2876 PetscInt k,N,start,end,l,row,col,srow,**vscaleforrow; 2877 PetscScalar dx,*y,*xx,*w3_array; 2878 PetscScalar *vscale_array; 2879 PetscReal epsilon = coloring->error_rel,umin = coloring->umin; 2880 Vec w1,w2,w3; 2881 void *fctx = coloring->fctx; 2882 PetscBool flg = PETSC_FALSE; 2883 2884 PetscFunctionBegin; 2885 printf("MatFDColoringApply_SeqAIJ ...\n"); 2886 if (!coloring->w1) { 2887 ierr = VecDuplicate(x1,&coloring->w1);CHKERRQ(ierr); 2888 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w1);CHKERRQ(ierr); 2889 ierr = VecDuplicate(x1,&coloring->w2);CHKERRQ(ierr); 2890 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w2);CHKERRQ(ierr); 2891 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 2892 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 2893 } 2894 w1 = coloring->w1; w2 = coloring->w2; w3 = coloring->w3; 2895 2896 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 2897 ierr = PetscOptionsGetBool(((PetscObject)coloring)->prefix,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 2898 if (flg) { 2899 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 2900 } else { 2901 PetscBool assembled; 2902 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 2903 if (assembled) { 2904 ierr = MatZeroEntries(J);CHKERRQ(ierr); 2905 } 2906 } 2907 2908 ierr = VecGetOwnershipRange(x1,&start,&end);CHKERRQ(ierr); 2909 ierr = VecGetSize(x1,&N);CHKERRQ(ierr); 2910 2911 if (!coloring->fset) { 2912 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2913 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 2914 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2915 } else { 2916 coloring->fset = PETSC_FALSE; 2917 } 2918 2919 /* 2920 Compute all the scale factors and share with other processors 2921 */ 2922 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr);xx = xx - start; 2923 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);vscale_array = vscale_array - start; 2924 for (k=0; k<coloring->ncolors; k++) { 2925 /* 2926 Loop over each column associated with color adding the 2927 perturbation to the vector w3. 2928 */ 2929 for (l=0; l<coloring->ncolumns[k]; l++) { 2930 col = coloring->columns[k][l]; /* column of the matrix we are probing for */ 2931 dx = xx[col]; 2932 if (dx == 0.0) dx = 1.0; 2933 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2934 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2935 dx *= epsilon; 2936 vscale_array[col] = 1.0/dx; 2937 } 2938 } 2939 vscale_array = vscale_array + start; 2940 2941 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2942 ierr = VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2943 ierr = VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2944 2945 /* ierr = VecView(coloring->vscale,PETSC_VIEWER_STDOUT_WORLD); 2946 ierr = VecView(x1,PETSC_VIEWER_STDOUT_WORLD);*/ 2947 2948 if (coloring->vscaleforrow) vscaleforrow = coloring->vscaleforrow; 2949 else vscaleforrow = coloring->columnsforrow; 2950 2951 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2952 /* 2953 Loop over each color 2954 */ 2955 for (k=0; k<coloring->ncolors; k++) { 2956 coloring->currentcolor = k; 2957 2958 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 2959 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr);w3_array = w3_array - start; 2960 /* 2961 Loop over each column associated with color adding the 2962 perturbation to the vector w3. 2963 */ 2964 for (l=0; l<coloring->ncolumns[k]; l++) { 2965 col = coloring->columns[k][l]; /* column of the matrix we are probing for */ 2966 dx = xx[col]; 2967 if (dx == 0.0) dx = 1.0; 2968 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2969 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2970 dx *= epsilon; 2971 if (!PetscAbsScalar(dx)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Computed 0 differencing parameter"); 2972 w3_array[col] += dx; 2973 } 2974 w3_array = w3_array + start; ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 2975 2976 /* 2977 Evaluate function at x1 + dx (here dx is a vector of perturbations) 2978 */ 2979 2980 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2981 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 2982 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2983 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 2984 2985 /* 2986 Loop over rows of vector, putting results into Jacobian matrix 2987 */ 2988 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 2989 for (l=0; l<coloring->nrows[k]; l++) { 2990 row = coloring->rows[k][l]; 2991 col = coloring->columnsforrow[k][l]; 2992 y[row] *= vscale_array[vscaleforrow[k][l]]; 2993 srow = row + start; 2994 ierr = MatSetValues_SeqAIJ(J,1,&srow,1,&col,y+row,INSERT_VALUES);CHKERRQ(ierr); 2995 } 2996 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 2997 } 2998 coloring->currentcolor = k; 2999 3000 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 3001 xx = xx + start; 3002 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 3003 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3004 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3005 PetscFunctionReturn(0); 3006 } 3007 3008 /* 3009 Computes the number of nonzeros per row needed for preallocation when X and Y 3010 have different nonzero structure. 3011 */ 3012 #undef __FUNCT__ 3013 #define __FUNCT__ "MatAXPYGetPreallocation_SeqAIJ" 3014 PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz) 3015 { 3016 PetscInt i,m=Y->rmap->N; 3017 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 3018 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 3019 const PetscInt *xi = x->i,*yi = y->i; 3020 3021 PetscFunctionBegin; 3022 /* Set the number of nonzeros in the new matrix */ 3023 for (i=0; i<m; i++) { 3024 PetscInt j,k,nzx = xi[i+1] - xi[i],nzy = yi[i+1] - yi[i]; 3025 const PetscInt *xj = x->j+xi[i],*yj = y->j+yi[i]; 3026 nnz[i] = 0; 3027 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 3028 for (; k<nzy && yj[k]<xj[j]; k++) nnz[i]++; /* Catch up to X */ 3029 if (k<nzy && yj[k]==xj[j]) k++; /* Skip duplicate */ 3030 nnz[i]++; 3031 } 3032 for (; k<nzy; k++) nnz[i]++; 3033 } 3034 PetscFunctionReturn(0); 3035 } 3036 3037 #undef __FUNCT__ 3038 #define __FUNCT__ "MatAXPY_SeqAIJ" 3039 PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 3040 { 3041 PetscErrorCode ierr; 3042 PetscInt i; 3043 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data; 3044 PetscBLASInt one=1,bnz; 3045 3046 PetscFunctionBegin; 3047 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 3048 if (str == SAME_NONZERO_PATTERN) { 3049 PetscScalar alpha = a; 3050 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 3051 ierr = MatSeqAIJInvalidateDiagonal(Y);CHKERRQ(ierr); 3052 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 3053 if (y->xtoy && y->XtoY != X) { 3054 ierr = PetscFree(y->xtoy);CHKERRQ(ierr); 3055 ierr = MatDestroy(&y->XtoY);CHKERRQ(ierr); 3056 } 3057 if (!y->xtoy) { /* get xtoy */ 3058 ierr = MatAXPYGetxtoy_Private(X->rmap->n,x->i,x->j,NULL, y->i,y->j,NULL, &y->xtoy);CHKERRQ(ierr); 3059 y->XtoY = X; 3060 ierr = PetscObjectReference((PetscObject)X);CHKERRQ(ierr); 3061 } 3062 for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]); 3063 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); 3064 } else { 3065 Mat B; 3066 PetscInt *nnz; 3067 ierr = PetscMalloc(Y->rmap->N*sizeof(PetscInt),&nnz);CHKERRQ(ierr); 3068 ierr = MatCreate(PetscObjectComm((PetscObject)Y),&B);CHKERRQ(ierr); 3069 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 3070 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 3071 ierr = MatSetBlockSizes(B,Y->rmap->bs,Y->cmap->bs);CHKERRQ(ierr); 3072 ierr = MatSetType(B,(MatType) ((PetscObject)Y)->type_name);CHKERRQ(ierr); 3073 ierr = MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);CHKERRQ(ierr); 3074 ierr = MatSeqAIJSetPreallocation(B,0,nnz);CHKERRQ(ierr); 3075 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 3076 ierr = MatHeaderReplace(Y,B);CHKERRQ(ierr); 3077 ierr = PetscFree(nnz);CHKERRQ(ierr); 3078 } 3079 PetscFunctionReturn(0); 3080 } 3081 3082 #undef __FUNCT__ 3083 #define __FUNCT__ "MatConjugate_SeqAIJ" 3084 PetscErrorCode MatConjugate_SeqAIJ(Mat mat) 3085 { 3086 #if defined(PETSC_USE_COMPLEX) 3087 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3088 PetscInt i,nz; 3089 PetscScalar *a; 3090 3091 PetscFunctionBegin; 3092 nz = aij->nz; 3093 a = aij->a; 3094 for (i=0; i<nz; i++) a[i] = PetscConj(a[i]); 3095 #else 3096 PetscFunctionBegin; 3097 #endif 3098 PetscFunctionReturn(0); 3099 } 3100 3101 #undef __FUNCT__ 3102 #define __FUNCT__ "MatGetRowMaxAbs_SeqAIJ" 3103 PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3104 { 3105 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3106 PetscErrorCode ierr; 3107 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3108 PetscReal atmp; 3109 PetscScalar *x; 3110 MatScalar *aa; 3111 3112 PetscFunctionBegin; 3113 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3114 aa = a->a; 3115 ai = a->i; 3116 aj = a->j; 3117 3118 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3119 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3120 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3121 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3122 for (i=0; i<m; i++) { 3123 ncols = ai[1] - ai[0]; ai++; 3124 x[i] = 0.0; 3125 for (j=0; j<ncols; j++) { 3126 atmp = PetscAbsScalar(*aa); 3127 if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3128 aa++; aj++; 3129 } 3130 } 3131 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3132 PetscFunctionReturn(0); 3133 } 3134 3135 #undef __FUNCT__ 3136 #define __FUNCT__ "MatGetRowMax_SeqAIJ" 3137 PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3138 { 3139 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3140 PetscErrorCode ierr; 3141 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3142 PetscScalar *x; 3143 MatScalar *aa; 3144 3145 PetscFunctionBegin; 3146 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3147 aa = a->a; 3148 ai = a->i; 3149 aj = a->j; 3150 3151 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3152 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3153 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3154 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3155 for (i=0; i<m; i++) { 3156 ncols = ai[1] - ai[0]; ai++; 3157 if (ncols == A->cmap->n) { /* row is dense */ 3158 x[i] = *aa; if (idx) idx[i] = 0; 3159 } else { /* row is sparse so already KNOW maximum is 0.0 or higher */ 3160 x[i] = 0.0; 3161 if (idx) { 3162 idx[i] = 0; /* in case ncols is zero */ 3163 for (j=0;j<ncols;j++) { /* find first implicit 0.0 in the row */ 3164 if (aj[j] > j) { 3165 idx[i] = j; 3166 break; 3167 } 3168 } 3169 } 3170 } 3171 for (j=0; j<ncols; j++) { 3172 if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3173 aa++; aj++; 3174 } 3175 } 3176 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3177 PetscFunctionReturn(0); 3178 } 3179 3180 #undef __FUNCT__ 3181 #define __FUNCT__ "MatGetRowMinAbs_SeqAIJ" 3182 PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3183 { 3184 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3185 PetscErrorCode ierr; 3186 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3187 PetscReal atmp; 3188 PetscScalar *x; 3189 MatScalar *aa; 3190 3191 PetscFunctionBegin; 3192 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3193 aa = a->a; 3194 ai = a->i; 3195 aj = a->j; 3196 3197 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3198 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3199 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3200 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); 3201 for (i=0; i<m; i++) { 3202 ncols = ai[1] - ai[0]; ai++; 3203 if (ncols) { 3204 /* Get first nonzero */ 3205 for (j = 0; j < ncols; j++) { 3206 atmp = PetscAbsScalar(aa[j]); 3207 if (atmp > 1.0e-12) { 3208 x[i] = atmp; 3209 if (idx) idx[i] = aj[j]; 3210 break; 3211 } 3212 } 3213 if (j == ncols) {x[i] = PetscAbsScalar(*aa); if (idx) idx[i] = *aj;} 3214 } else { 3215 x[i] = 0.0; if (idx) idx[i] = 0; 3216 } 3217 for (j = 0; j < ncols; j++) { 3218 atmp = PetscAbsScalar(*aa); 3219 if (atmp > 1.0e-12 && PetscAbsScalar(x[i]) > atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3220 aa++; aj++; 3221 } 3222 } 3223 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3224 PetscFunctionReturn(0); 3225 } 3226 3227 #undef __FUNCT__ 3228 #define __FUNCT__ "MatGetRowMin_SeqAIJ" 3229 PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3230 { 3231 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3232 PetscErrorCode ierr; 3233 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3234 PetscScalar *x; 3235 MatScalar *aa; 3236 3237 PetscFunctionBegin; 3238 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3239 aa = a->a; 3240 ai = a->i; 3241 aj = a->j; 3242 3243 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3244 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3245 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3246 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3247 for (i=0; i<m; i++) { 3248 ncols = ai[1] - ai[0]; ai++; 3249 if (ncols == A->cmap->n) { /* row is dense */ 3250 x[i] = *aa; if (idx) idx[i] = 0; 3251 } else { /* row is sparse so already KNOW minimum is 0.0 or lower */ 3252 x[i] = 0.0; 3253 if (idx) { /* find first implicit 0.0 in the row */ 3254 idx[i] = 0; /* in case ncols is zero */ 3255 for (j=0; j<ncols; j++) { 3256 if (aj[j] > j) { 3257 idx[i] = j; 3258 break; 3259 } 3260 } 3261 } 3262 } 3263 for (j=0; j<ncols; j++) { 3264 if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3265 aa++; aj++; 3266 } 3267 } 3268 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3269 PetscFunctionReturn(0); 3270 } 3271 3272 #include <petscblaslapack.h> 3273 #include <petsc-private/kernels/blockinvert.h> 3274 3275 #undef __FUNCT__ 3276 #define __FUNCT__ "MatInvertBlockDiagonal_SeqAIJ" 3277 PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values) 3278 { 3279 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 3280 PetscErrorCode ierr; 3281 PetscInt i,bs = A->rmap->bs,mbs = A->rmap->n/A->rmap->bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j; 3282 MatScalar *diag,work[25],*v_work; 3283 PetscReal shift = 0.0; 3284 3285 PetscFunctionBegin; 3286 if (a->ibdiagvalid) { 3287 if (values) *values = a->ibdiag; 3288 PetscFunctionReturn(0); 3289 } 3290 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 3291 if (!a->ibdiag) { 3292 ierr = PetscMalloc(bs2*mbs*sizeof(PetscScalar),&a->ibdiag);CHKERRQ(ierr); 3293 ierr = PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));CHKERRQ(ierr); 3294 } 3295 diag = a->ibdiag; 3296 if (values) *values = a->ibdiag; 3297 /* factor and invert each block */ 3298 switch (bs) { 3299 case 1: 3300 for (i=0; i<mbs; i++) { 3301 ierr = MatGetValues(A,1,&i,1,&i,diag+i);CHKERRQ(ierr); 3302 diag[i] = (PetscScalar)1.0 / (diag[i] + shift); 3303 } 3304 break; 3305 case 2: 3306 for (i=0; i<mbs; i++) { 3307 ij[0] = 2*i; ij[1] = 2*i + 1; 3308 ierr = MatGetValues(A,2,ij,2,ij,diag);CHKERRQ(ierr); 3309 ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 3310 ierr = PetscKernel_A_gets_transpose_A_2(diag);CHKERRQ(ierr); 3311 diag += 4; 3312 } 3313 break; 3314 case 3: 3315 for (i=0; i<mbs; i++) { 3316 ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2; 3317 ierr = MatGetValues(A,3,ij,3,ij,diag);CHKERRQ(ierr); 3318 ierr = PetscKernel_A_gets_inverse_A_3(diag,shift);CHKERRQ(ierr); 3319 ierr = PetscKernel_A_gets_transpose_A_3(diag);CHKERRQ(ierr); 3320 diag += 9; 3321 } 3322 break; 3323 case 4: 3324 for (i=0; i<mbs; i++) { 3325 ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3; 3326 ierr = MatGetValues(A,4,ij,4,ij,diag);CHKERRQ(ierr); 3327 ierr = PetscKernel_A_gets_inverse_A_4(diag,shift);CHKERRQ(ierr); 3328 ierr = PetscKernel_A_gets_transpose_A_4(diag);CHKERRQ(ierr); 3329 diag += 16; 3330 } 3331 break; 3332 case 5: 3333 for (i=0; i<mbs; i++) { 3334 ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4; 3335 ierr = MatGetValues(A,5,ij,5,ij,diag);CHKERRQ(ierr); 3336 ierr = PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift);CHKERRQ(ierr); 3337 ierr = PetscKernel_A_gets_transpose_A_5(diag);CHKERRQ(ierr); 3338 diag += 25; 3339 } 3340 break; 3341 case 6: 3342 for (i=0; i<mbs; i++) { 3343 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; 3344 ierr = MatGetValues(A,6,ij,6,ij,diag);CHKERRQ(ierr); 3345 ierr = PetscKernel_A_gets_inverse_A_6(diag,shift);CHKERRQ(ierr); 3346 ierr = PetscKernel_A_gets_transpose_A_6(diag);CHKERRQ(ierr); 3347 diag += 36; 3348 } 3349 break; 3350 case 7: 3351 for (i=0; i<mbs; i++) { 3352 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; 3353 ierr = MatGetValues(A,7,ij,7,ij,diag);CHKERRQ(ierr); 3354 ierr = PetscKernel_A_gets_inverse_A_7(diag,shift);CHKERRQ(ierr); 3355 ierr = PetscKernel_A_gets_transpose_A_7(diag);CHKERRQ(ierr); 3356 diag += 49; 3357 } 3358 break; 3359 default: 3360 ierr = PetscMalloc3(bs,MatScalar,&v_work,bs,PetscInt,&v_pivots,bs,PetscInt,&IJ);CHKERRQ(ierr); 3361 for (i=0; i<mbs; i++) { 3362 for (j=0; j<bs; j++) { 3363 IJ[j] = bs*i + j; 3364 } 3365 ierr = MatGetValues(A,bs,IJ,bs,IJ,diag);CHKERRQ(ierr); 3366 ierr = PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work);CHKERRQ(ierr); 3367 ierr = PetscKernel_A_gets_transpose_A_N(diag,bs);CHKERRQ(ierr); 3368 diag += bs2; 3369 } 3370 ierr = PetscFree3(v_work,v_pivots,IJ);CHKERRQ(ierr); 3371 } 3372 a->ibdiagvalid = PETSC_TRUE; 3373 PetscFunctionReturn(0); 3374 } 3375 3376 #undef __FUNCT__ 3377 #define __FUNCT__ "MatSetRandom_SeqAIJ" 3378 static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx) 3379 { 3380 PetscErrorCode ierr; 3381 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data; 3382 PetscScalar a; 3383 PetscInt m,n,i,j,col; 3384 3385 PetscFunctionBegin; 3386 if (!x->assembled) { 3387 ierr = MatGetSize(x,&m,&n);CHKERRQ(ierr); 3388 for (i=0; i<m; i++) { 3389 for (j=0; j<aij->imax[i]; j++) { 3390 ierr = PetscRandomGetValue(rctx,&a);CHKERRQ(ierr); 3391 col = (PetscInt)(n*PetscRealPart(a)); 3392 ierr = MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);CHKERRQ(ierr); 3393 } 3394 } 3395 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not yet coded"); 3396 ierr = MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3397 ierr = MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3398 PetscFunctionReturn(0); 3399 } 3400 3401 extern PetscErrorCode MatFDColoringApply_AIJ(Mat,MatFDColoring,Vec,MatStructure*,void*); 3402 /* -------------------------------------------------------------------*/ 3403 static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ, 3404 MatGetRow_SeqAIJ, 3405 MatRestoreRow_SeqAIJ, 3406 MatMult_SeqAIJ, 3407 /* 4*/ MatMultAdd_SeqAIJ, 3408 MatMultTranspose_SeqAIJ, 3409 MatMultTransposeAdd_SeqAIJ, 3410 0, 3411 0, 3412 0, 3413 /* 10*/ 0, 3414 MatLUFactor_SeqAIJ, 3415 0, 3416 MatSOR_SeqAIJ, 3417 MatTranspose_SeqAIJ, 3418 /*1 5*/ MatGetInfo_SeqAIJ, 3419 MatEqual_SeqAIJ, 3420 MatGetDiagonal_SeqAIJ, 3421 MatDiagonalScale_SeqAIJ, 3422 MatNorm_SeqAIJ, 3423 /* 20*/ 0, 3424 MatAssemblyEnd_SeqAIJ, 3425 MatSetOption_SeqAIJ, 3426 MatZeroEntries_SeqAIJ, 3427 /* 24*/ MatZeroRows_SeqAIJ, 3428 0, 3429 0, 3430 0, 3431 0, 3432 /* 29*/ MatSetUp_SeqAIJ, 3433 0, 3434 0, 3435 0, 3436 0, 3437 /* 34*/ MatDuplicate_SeqAIJ, 3438 0, 3439 0, 3440 MatILUFactor_SeqAIJ, 3441 0, 3442 /* 39*/ MatAXPY_SeqAIJ, 3443 MatGetSubMatrices_SeqAIJ, 3444 MatIncreaseOverlap_SeqAIJ, 3445 MatGetValues_SeqAIJ, 3446 MatCopy_SeqAIJ, 3447 /* 44*/ MatGetRowMax_SeqAIJ, 3448 MatScale_SeqAIJ, 3449 0, 3450 MatDiagonalSet_SeqAIJ, 3451 MatZeroRowsColumns_SeqAIJ, 3452 /* 49*/ MatSetRandom_SeqAIJ, 3453 MatGetRowIJ_SeqAIJ, 3454 MatRestoreRowIJ_SeqAIJ, 3455 MatGetColumnIJ_SeqAIJ, 3456 MatRestoreColumnIJ_SeqAIJ, 3457 /* 54*/ MatFDColoringCreate_SeqAIJ, 3458 0, 3459 0, 3460 MatPermute_SeqAIJ, 3461 0, 3462 /* 59*/ 0, 3463 MatDestroy_SeqAIJ, 3464 MatView_SeqAIJ, 3465 0, 3466 MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ, 3467 /* 64*/ MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ, 3468 MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ, 3469 0, 3470 0, 3471 0, 3472 /* 69*/ MatGetRowMaxAbs_SeqAIJ, 3473 MatGetRowMinAbs_SeqAIJ, 3474 0, 3475 MatSetColoring_SeqAIJ, 3476 0, 3477 /* 74*/ MatSetValuesAdifor_SeqAIJ, 3478 MatFDColoringApply_AIJ, 3479 0, 3480 0, 3481 0, 3482 /* 79*/ MatFindZeroDiagonals_SeqAIJ, 3483 0, 3484 0, 3485 0, 3486 MatLoad_SeqAIJ, 3487 /* 84*/ MatIsSymmetric_SeqAIJ, 3488 MatIsHermitian_SeqAIJ, 3489 0, 3490 0, 3491 0, 3492 /* 89*/ MatMatMult_SeqAIJ_SeqAIJ, 3493 MatMatMultSymbolic_SeqAIJ_SeqAIJ, 3494 MatMatMultNumeric_SeqAIJ_SeqAIJ, 3495 MatPtAP_SeqAIJ_SeqAIJ, 3496 MatPtAPSymbolic_SeqAIJ_SeqAIJ_DenseAxpy, 3497 /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ, 3498 MatMatTransposeMult_SeqAIJ_SeqAIJ, 3499 MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ, 3500 MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ, 3501 0, 3502 /* 99*/ 0, 3503 0, 3504 0, 3505 MatConjugate_SeqAIJ, 3506 0, 3507 /*104*/ MatSetValuesRow_SeqAIJ, 3508 MatRealPart_SeqAIJ, 3509 MatImaginaryPart_SeqAIJ, 3510 0, 3511 0, 3512 /*109*/ MatMatSolve_SeqAIJ, 3513 0, 3514 MatGetRowMin_SeqAIJ, 3515 0, 3516 MatMissingDiagonal_SeqAIJ, 3517 /*114*/ 0, 3518 0, 3519 0, 3520 0, 3521 0, 3522 /*119*/ 0, 3523 0, 3524 0, 3525 0, 3526 MatGetMultiProcBlock_SeqAIJ, 3527 /*124*/ MatFindNonzeroRows_SeqAIJ, 3528 MatGetColumnNorms_SeqAIJ, 3529 MatInvertBlockDiagonal_SeqAIJ, 3530 0, 3531 0, 3532 /*129*/ 0, 3533 MatTransposeMatMult_SeqAIJ_SeqAIJ, 3534 MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ, 3535 MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ, 3536 MatTransposeColoringCreate_SeqAIJ, 3537 /*134*/ MatTransColoringApplySpToDen_SeqAIJ, 3538 MatTransColoringApplyDenToSp_SeqAIJ, 3539 MatRARt_SeqAIJ_SeqAIJ, 3540 MatRARtSymbolic_SeqAIJ_SeqAIJ, 3541 MatRARtNumeric_SeqAIJ_SeqAIJ, 3542 /*139*/0, 3543 0 3544 }; 3545 3546 #undef __FUNCT__ 3547 #define __FUNCT__ "MatSeqAIJSetColumnIndices_SeqAIJ" 3548 PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices) 3549 { 3550 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3551 PetscInt i,nz,n; 3552 3553 PetscFunctionBegin; 3554 nz = aij->maxnz; 3555 n = mat->rmap->n; 3556 for (i=0; i<nz; i++) { 3557 aij->j[i] = indices[i]; 3558 } 3559 aij->nz = nz; 3560 for (i=0; i<n; i++) { 3561 aij->ilen[i] = aij->imax[i]; 3562 } 3563 PetscFunctionReturn(0); 3564 } 3565 3566 #undef __FUNCT__ 3567 #define __FUNCT__ "MatSeqAIJSetColumnIndices" 3568 /*@ 3569 MatSeqAIJSetColumnIndices - Set the column indices for all the rows 3570 in the matrix. 3571 3572 Input Parameters: 3573 + mat - the SeqAIJ matrix 3574 - indices - the column indices 3575 3576 Level: advanced 3577 3578 Notes: 3579 This can be called if you have precomputed the nonzero structure of the 3580 matrix and want to provide it to the matrix object to improve the performance 3581 of the MatSetValues() operation. 3582 3583 You MUST have set the correct numbers of nonzeros per row in the call to 3584 MatCreateSeqAIJ(), and the columns indices MUST be sorted. 3585 3586 MUST be called before any calls to MatSetValues(); 3587 3588 The indices should start with zero, not one. 3589 3590 @*/ 3591 PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices) 3592 { 3593 PetscErrorCode ierr; 3594 3595 PetscFunctionBegin; 3596 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3597 PetscValidPointer(indices,2); 3598 ierr = PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));CHKERRQ(ierr); 3599 PetscFunctionReturn(0); 3600 } 3601 3602 /* ----------------------------------------------------------------------------------------*/ 3603 3604 #undef __FUNCT__ 3605 #define __FUNCT__ "MatStoreValues_SeqAIJ" 3606 PetscErrorCode MatStoreValues_SeqAIJ(Mat mat) 3607 { 3608 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3609 PetscErrorCode ierr; 3610 size_t nz = aij->i[mat->rmap->n]; 3611 3612 PetscFunctionBegin; 3613 if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3614 3615 /* allocate space for values if not already there */ 3616 if (!aij->saved_values) { 3617 ierr = PetscMalloc((nz+1)*sizeof(PetscScalar),&aij->saved_values);CHKERRQ(ierr); 3618 ierr = PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));CHKERRQ(ierr); 3619 } 3620 3621 /* copy values over */ 3622 ierr = PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3623 PetscFunctionReturn(0); 3624 } 3625 3626 #undef __FUNCT__ 3627 #define __FUNCT__ "MatStoreValues" 3628 /*@ 3629 MatStoreValues - Stashes a copy of the matrix values; this allows, for 3630 example, reuse of the linear part of a Jacobian, while recomputing the 3631 nonlinear portion. 3632 3633 Collect on Mat 3634 3635 Input Parameters: 3636 . mat - the matrix (currently only AIJ matrices support this option) 3637 3638 Level: advanced 3639 3640 Common Usage, with SNESSolve(): 3641 $ Create Jacobian matrix 3642 $ Set linear terms into matrix 3643 $ Apply boundary conditions to matrix, at this time matrix must have 3644 $ final nonzero structure (i.e. setting the nonlinear terms and applying 3645 $ boundary conditions again will not change the nonzero structure 3646 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3647 $ ierr = MatStoreValues(mat); 3648 $ Call SNESSetJacobian() with matrix 3649 $ In your Jacobian routine 3650 $ ierr = MatRetrieveValues(mat); 3651 $ Set nonlinear terms in matrix 3652 3653 Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself: 3654 $ // build linear portion of Jacobian 3655 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3656 $ ierr = MatStoreValues(mat); 3657 $ loop over nonlinear iterations 3658 $ ierr = MatRetrieveValues(mat); 3659 $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian 3660 $ // call MatAssemblyBegin/End() on matrix 3661 $ Solve linear system with Jacobian 3662 $ endloop 3663 3664 Notes: 3665 Matrix must already be assemblied before calling this routine 3666 Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before 3667 calling this routine. 3668 3669 When this is called multiple times it overwrites the previous set of stored values 3670 and does not allocated additional space. 3671 3672 .seealso: MatRetrieveValues() 3673 3674 @*/ 3675 PetscErrorCode MatStoreValues(Mat mat) 3676 { 3677 PetscErrorCode ierr; 3678 3679 PetscFunctionBegin; 3680 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3681 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3682 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3683 ierr = PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));CHKERRQ(ierr); 3684 PetscFunctionReturn(0); 3685 } 3686 3687 #undef __FUNCT__ 3688 #define __FUNCT__ "MatRetrieveValues_SeqAIJ" 3689 PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat) 3690 { 3691 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3692 PetscErrorCode ierr; 3693 PetscInt nz = aij->i[mat->rmap->n]; 3694 3695 PetscFunctionBegin; 3696 if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3697 if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first"); 3698 /* copy values over */ 3699 ierr = PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3700 PetscFunctionReturn(0); 3701 } 3702 3703 #undef __FUNCT__ 3704 #define __FUNCT__ "MatRetrieveValues" 3705 /*@ 3706 MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for 3707 example, reuse of the linear part of a Jacobian, while recomputing the 3708 nonlinear portion. 3709 3710 Collect on Mat 3711 3712 Input Parameters: 3713 . mat - the matrix (currently on AIJ matrices support this option) 3714 3715 Level: advanced 3716 3717 .seealso: MatStoreValues() 3718 3719 @*/ 3720 PetscErrorCode MatRetrieveValues(Mat mat) 3721 { 3722 PetscErrorCode ierr; 3723 3724 PetscFunctionBegin; 3725 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3726 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3727 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3728 ierr = PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));CHKERRQ(ierr); 3729 PetscFunctionReturn(0); 3730 } 3731 3732 3733 /* --------------------------------------------------------------------------------*/ 3734 #undef __FUNCT__ 3735 #define __FUNCT__ "MatCreateSeqAIJ" 3736 /*@C 3737 MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format 3738 (the default parallel PETSc format). For good matrix assembly performance 3739 the user should preallocate the matrix storage by setting the parameter nz 3740 (or the array nnz). By setting these parameters accurately, performance 3741 during matrix assembly can be increased by more than a factor of 50. 3742 3743 Collective on MPI_Comm 3744 3745 Input Parameters: 3746 + comm - MPI communicator, set to PETSC_COMM_SELF 3747 . m - number of rows 3748 . n - number of columns 3749 . nz - number of nonzeros per row (same for all rows) 3750 - nnz - array containing the number of nonzeros in the various rows 3751 (possibly different for each row) or NULL 3752 3753 Output Parameter: 3754 . A - the matrix 3755 3756 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3757 MatXXXXSetPreallocation() paradgm instead of this routine directly. 3758 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3759 3760 Notes: 3761 If nnz is given then nz is ignored 3762 3763 The AIJ format (also called the Yale sparse matrix format or 3764 compressed row storage), is fully compatible with standard Fortran 77 3765 storage. That is, the stored row and column indices can begin at 3766 either one (as in Fortran) or zero. See the users' manual for details. 3767 3768 Specify the preallocated storage with either nz or nnz (not both). 3769 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3770 allocation. For large problems you MUST preallocate memory or you 3771 will get TERRIBLE performance, see the users' manual chapter on matrices. 3772 3773 By default, this format uses inodes (identical nodes) when possible, to 3774 improve numerical efficiency of matrix-vector products and solves. We 3775 search for consecutive rows with the same nonzero structure, thereby 3776 reusing matrix information to achieve increased efficiency. 3777 3778 Options Database Keys: 3779 + -mat_no_inode - Do not use inodes 3780 - -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3781 3782 Level: intermediate 3783 3784 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays() 3785 3786 @*/ 3787 PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A) 3788 { 3789 PetscErrorCode ierr; 3790 3791 PetscFunctionBegin; 3792 ierr = MatCreate(comm,A);CHKERRQ(ierr); 3793 ierr = MatSetSizes(*A,m,n,m,n);CHKERRQ(ierr); 3794 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 3795 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);CHKERRQ(ierr); 3796 PetscFunctionReturn(0); 3797 } 3798 3799 #undef __FUNCT__ 3800 #define __FUNCT__ "MatSeqAIJSetPreallocation" 3801 /*@C 3802 MatSeqAIJSetPreallocation - For good matrix assembly performance 3803 the user should preallocate the matrix storage by setting the parameter nz 3804 (or the array nnz). By setting these parameters accurately, performance 3805 during matrix assembly can be increased by more than a factor of 50. 3806 3807 Collective on MPI_Comm 3808 3809 Input Parameters: 3810 + B - The matrix-free 3811 . nz - number of nonzeros per row (same for all rows) 3812 - nnz - array containing the number of nonzeros in the various rows 3813 (possibly different for each row) or NULL 3814 3815 Notes: 3816 If nnz is given then nz is ignored 3817 3818 The AIJ format (also called the Yale sparse matrix format or 3819 compressed row storage), is fully compatible with standard Fortran 77 3820 storage. That is, the stored row and column indices can begin at 3821 either one (as in Fortran) or zero. See the users' manual for details. 3822 3823 Specify the preallocated storage with either nz or nnz (not both). 3824 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3825 allocation. For large problems you MUST preallocate memory or you 3826 will get TERRIBLE performance, see the users' manual chapter on matrices. 3827 3828 You can call MatGetInfo() to get information on how effective the preallocation was; 3829 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3830 You can also run with the option -info and look for messages with the string 3831 malloc in them to see if additional memory allocation was needed. 3832 3833 Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix 3834 entries or columns indices 3835 3836 By default, this format uses inodes (identical nodes) when possible, to 3837 improve numerical efficiency of matrix-vector products and solves. We 3838 search for consecutive rows with the same nonzero structure, thereby 3839 reusing matrix information to achieve increased efficiency. 3840 3841 Options Database Keys: 3842 + -mat_no_inode - Do not use inodes 3843 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3844 - -mat_aij_oneindex - Internally use indexing starting at 1 3845 rather than 0. Note that when calling MatSetValues(), 3846 the user still MUST index entries starting at 0! 3847 3848 Level: intermediate 3849 3850 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo() 3851 3852 @*/ 3853 PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[]) 3854 { 3855 PetscErrorCode ierr; 3856 3857 PetscFunctionBegin; 3858 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3859 PetscValidType(B,1); 3860 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));CHKERRQ(ierr); 3861 PetscFunctionReturn(0); 3862 } 3863 3864 #undef __FUNCT__ 3865 #define __FUNCT__ "MatSeqAIJSetPreallocation_SeqAIJ" 3866 PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz) 3867 { 3868 Mat_SeqAIJ *b; 3869 PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE; 3870 PetscErrorCode ierr; 3871 PetscInt i; 3872 3873 PetscFunctionBegin; 3874 if (nz >= 0 || nnz) realalloc = PETSC_TRUE; 3875 if (nz == MAT_SKIP_ALLOCATION) { 3876 skipallocation = PETSC_TRUE; 3877 nz = 0; 3878 } 3879 3880 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3881 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3882 3883 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3884 if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %d",nz); 3885 if (nnz) { 3886 for (i=0; i<B->rmap->n; i++) { 3887 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]); 3888 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); 3889 } 3890 } 3891 3892 B->preallocated = PETSC_TRUE; 3893 3894 b = (Mat_SeqAIJ*)B->data; 3895 3896 if (!skipallocation) { 3897 if (!b->imax) { 3898 ierr = PetscMalloc2(B->rmap->n,PetscInt,&b->imax,B->rmap->n,PetscInt,&b->ilen);CHKERRQ(ierr); 3899 ierr = PetscLogObjectMemory((PetscObject)B,2*B->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 3900 } 3901 if (!nnz) { 3902 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10; 3903 else if (nz < 0) nz = 1; 3904 for (i=0; i<B->rmap->n; i++) b->imax[i] = nz; 3905 nz = nz*B->rmap->n; 3906 } else { 3907 nz = 0; 3908 for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz += nnz[i];} 3909 } 3910 /* b->ilen will count nonzeros in each row so far. */ 3911 for (i=0; i<B->rmap->n; i++) b->ilen[i] = 0; 3912 3913 /* allocate the matrix space */ 3914 ierr = MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);CHKERRQ(ierr); 3915 ierr = PetscMalloc3(nz,PetscScalar,&b->a,nz,PetscInt,&b->j,B->rmap->n+1,PetscInt,&b->i);CHKERRQ(ierr); 3916 ierr = PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));CHKERRQ(ierr); 3917 b->i[0] = 0; 3918 for (i=1; i<B->rmap->n+1; i++) { 3919 b->i[i] = b->i[i-1] + b->imax[i-1]; 3920 } 3921 b->singlemalloc = PETSC_TRUE; 3922 b->free_a = PETSC_TRUE; 3923 b->free_ij = PETSC_TRUE; 3924 #if defined(PETSC_THREADCOMM_ACTIVE) 3925 ierr = MatZeroEntries_SeqAIJ(B);CHKERRQ(ierr); 3926 #endif 3927 } else { 3928 b->free_a = PETSC_FALSE; 3929 b->free_ij = PETSC_FALSE; 3930 } 3931 3932 b->nz = 0; 3933 b->maxnz = nz; 3934 B->info.nz_unneeded = (double)b->maxnz; 3935 if (realalloc) { 3936 ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3937 } 3938 PetscFunctionReturn(0); 3939 } 3940 3941 #undef __FUNCT__ 3942 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR" 3943 /*@ 3944 MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format. 3945 3946 Input Parameters: 3947 + B - the matrix 3948 . i - the indices into j for the start of each row (starts with zero) 3949 . j - the column indices for each row (starts with zero) these must be sorted for each row 3950 - v - optional values in the matrix 3951 3952 Level: developer 3953 3954 The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays() 3955 3956 .keywords: matrix, aij, compressed row, sparse, sequential 3957 3958 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), SeqAIJ 3959 @*/ 3960 PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[]) 3961 { 3962 PetscErrorCode ierr; 3963 3964 PetscFunctionBegin; 3965 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3966 PetscValidType(B,1); 3967 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3968 PetscFunctionReturn(0); 3969 } 3970 3971 #undef __FUNCT__ 3972 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR_SeqAIJ" 3973 PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3974 { 3975 PetscInt i; 3976 PetscInt m,n; 3977 PetscInt nz; 3978 PetscInt *nnz, nz_max = 0; 3979 PetscScalar *values; 3980 PetscErrorCode ierr; 3981 3982 PetscFunctionBegin; 3983 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]); 3984 3985 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3986 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3987 3988 ierr = MatGetSize(B, &m, &n);CHKERRQ(ierr); 3989 ierr = PetscMalloc((m+1) * sizeof(PetscInt), &nnz);CHKERRQ(ierr); 3990 for (i = 0; i < m; i++) { 3991 nz = Ii[i+1]- Ii[i]; 3992 nz_max = PetscMax(nz_max, nz); 3993 if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz); 3994 nnz[i] = nz; 3995 } 3996 ierr = MatSeqAIJSetPreallocation(B, 0, nnz);CHKERRQ(ierr); 3997 ierr = PetscFree(nnz);CHKERRQ(ierr); 3998 3999 if (v) { 4000 values = (PetscScalar*) v; 4001 } else { 4002 ierr = PetscMalloc(nz_max*sizeof(PetscScalar), &values);CHKERRQ(ierr); 4003 ierr = PetscMemzero(values, nz_max*sizeof(PetscScalar));CHKERRQ(ierr); 4004 } 4005 4006 for (i = 0; i < m; i++) { 4007 nz = Ii[i+1] - Ii[i]; 4008 ierr = MatSetValues_SeqAIJ(B, 1, &i, nz, J+Ii[i], values + (v ? Ii[i] : 0), INSERT_VALUES);CHKERRQ(ierr); 4009 } 4010 4011 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4012 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4013 4014 if (!v) { 4015 ierr = PetscFree(values);CHKERRQ(ierr); 4016 } 4017 ierr = MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 4018 PetscFunctionReturn(0); 4019 } 4020 4021 #include <../src/mat/impls/dense/seq/dense.h> 4022 #include <petsc-private/kernels/petscaxpy.h> 4023 4024 #undef __FUNCT__ 4025 #define __FUNCT__ "MatMatMultNumeric_SeqDense_SeqAIJ" 4026 /* 4027 Computes (B'*A')' since computing B*A directly is untenable 4028 4029 n p p 4030 ( ) ( ) ( ) 4031 m ( A ) * n ( B ) = m ( C ) 4032 ( ) ( ) ( ) 4033 4034 */ 4035 PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C) 4036 { 4037 PetscErrorCode ierr; 4038 Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data; 4039 Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data; 4040 Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data; 4041 PetscInt i,n,m,q,p; 4042 const PetscInt *ii,*idx; 4043 const PetscScalar *b,*a,*a_q; 4044 PetscScalar *c,*c_q; 4045 4046 PetscFunctionBegin; 4047 m = A->rmap->n; 4048 n = A->cmap->n; 4049 p = B->cmap->n; 4050 a = sub_a->v; 4051 b = sub_b->a; 4052 c = sub_c->v; 4053 ierr = PetscMemzero(c,m*p*sizeof(PetscScalar));CHKERRQ(ierr); 4054 4055 ii = sub_b->i; 4056 idx = sub_b->j; 4057 for (i=0; i<n; i++) { 4058 q = ii[i+1] - ii[i]; 4059 while (q-->0) { 4060 c_q = c + m*(*idx); 4061 a_q = a + m*i; 4062 PetscKernelAXPY(c_q,*b,a_q,m); 4063 idx++; 4064 b++; 4065 } 4066 } 4067 PetscFunctionReturn(0); 4068 } 4069 4070 #undef __FUNCT__ 4071 #define __FUNCT__ "MatMatMultSymbolic_SeqDense_SeqAIJ" 4072 PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 4073 { 4074 PetscErrorCode ierr; 4075 PetscInt m=A->rmap->n,n=B->cmap->n; 4076 Mat Cmat; 4077 4078 PetscFunctionBegin; 4079 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); 4080 ierr = MatCreate(PetscObjectComm((PetscObject)A),&Cmat);CHKERRQ(ierr); 4081 ierr = MatSetSizes(Cmat,m,n,m,n);CHKERRQ(ierr); 4082 ierr = MatSetBlockSizes(Cmat,A->rmap->bs,B->cmap->bs);CHKERRQ(ierr); 4083 ierr = MatSetType(Cmat,MATSEQDENSE);CHKERRQ(ierr); 4084 ierr = MatSeqDenseSetPreallocation(Cmat,NULL);CHKERRQ(ierr); 4085 4086 Cmat->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ; 4087 4088 *C = Cmat; 4089 PetscFunctionReturn(0); 4090 } 4091 4092 /* ----------------------------------------------------------------*/ 4093 #undef __FUNCT__ 4094 #define __FUNCT__ "MatMatMult_SeqDense_SeqAIJ" 4095 PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 4096 { 4097 PetscErrorCode ierr; 4098 4099 PetscFunctionBegin; 4100 if (scall == MAT_INITIAL_MATRIX) { 4101 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 4102 ierr = MatMatMultSymbolic_SeqDense_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 4103 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 4104 } 4105 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 4106 ierr = MatMatMultNumeric_SeqDense_SeqAIJ(A,B,*C);CHKERRQ(ierr); 4107 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 4108 PetscFunctionReturn(0); 4109 } 4110 4111 4112 /*MC 4113 MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices, 4114 based on compressed sparse row format. 4115 4116 Options Database Keys: 4117 . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions() 4118 4119 Level: beginner 4120 4121 .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType 4122 M*/ 4123 4124 /*MC 4125 MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices. 4126 4127 This matrix type is identical to MATSEQAIJ when constructed with a single process communicator, 4128 and MATMPIAIJ otherwise. As a result, for single process communicators, 4129 MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation is supported 4130 for communicators controlling multiple processes. It is recommended that you call both of 4131 the above preallocation routines for simplicity. 4132 4133 Options Database Keys: 4134 . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions() 4135 4136 Developer Notes: Subclasses include MATAIJCUSP, MATAIJPERM, MATAIJCRL, and also automatically switches over to use inodes when 4137 enough exist. 4138 4139 Level: beginner 4140 4141 .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ 4142 M*/ 4143 4144 /*MC 4145 MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices. 4146 4147 This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator, 4148 and MATMPIAIJCRL otherwise. As a result, for single process communicators, 4149 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 4150 for communicators controlling multiple processes. It is recommended that you call both of 4151 the above preallocation routines for simplicity. 4152 4153 Options Database Keys: 4154 . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions() 4155 4156 Level: beginner 4157 4158 .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL 4159 M*/ 4160 4161 #if defined(PETSC_HAVE_PASTIX) 4162 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_pastix(Mat,MatFactorType,Mat*); 4163 #endif 4164 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4165 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_essl(Mat,MatFactorType,Mat*); 4166 #endif 4167 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*); 4168 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_petsc(Mat,MatFactorType,Mat*); 4169 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_bas(Mat,MatFactorType,Mat*); 4170 extern PetscErrorCode MatGetFactorAvailable_seqaij_petsc(Mat,MatFactorType,PetscBool*); 4171 #if defined(PETSC_HAVE_MUMPS) 4172 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mumps(Mat,MatFactorType,Mat*); 4173 #endif 4174 #if defined(PETSC_HAVE_SUPERLU) 4175 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat,MatFactorType,Mat*); 4176 #endif 4177 #if defined(PETSC_HAVE_SUPERLU_DIST) 4178 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu_dist(Mat,MatFactorType,Mat*); 4179 #endif 4180 #if defined(PETSC_HAVE_UMFPACK) 4181 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_umfpack(Mat,MatFactorType,Mat*); 4182 #endif 4183 #if defined(PETSC_HAVE_CHOLMOD) 4184 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat,MatFactorType,Mat*); 4185 #endif 4186 #if defined(PETSC_HAVE_LUSOL) 4187 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_lusol(Mat,MatFactorType,Mat*); 4188 #endif 4189 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4190 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_matlab(Mat,MatFactorType,Mat*); 4191 extern PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*); 4192 extern PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*); 4193 #endif 4194 #if defined(PETSC_HAVE_CLIQUE) 4195 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_clique(Mat,MatFactorType,Mat*); 4196 #endif 4197 4198 4199 #undef __FUNCT__ 4200 #define __FUNCT__ "MatSeqAIJGetArray" 4201 /*@C 4202 MatSeqAIJGetArray - gives access to the array where the data for a SeqSeqAIJ matrix is stored 4203 4204 Not Collective 4205 4206 Input Parameter: 4207 . mat - a MATSEQDENSE matrix 4208 4209 Output Parameter: 4210 . array - pointer to the data 4211 4212 Level: intermediate 4213 4214 .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90() 4215 @*/ 4216 PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array) 4217 { 4218 PetscErrorCode ierr; 4219 4220 PetscFunctionBegin; 4221 ierr = PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4222 PetscFunctionReturn(0); 4223 } 4224 4225 #undef __FUNCT__ 4226 #define __FUNCT__ "MatSeqAIJRestoreArray" 4227 /*@C 4228 MatSeqAIJRestoreArray - returns access to the array where the data for a SeqSeqAIJ matrix is stored obtained by MatSeqAIJGetArray() 4229 4230 Not Collective 4231 4232 Input Parameters: 4233 . mat - a MATSEQDENSE matrix 4234 . array - pointer to the data 4235 4236 Level: intermediate 4237 4238 .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90() 4239 @*/ 4240 PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array) 4241 { 4242 PetscErrorCode ierr; 4243 4244 PetscFunctionBegin; 4245 ierr = PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4246 PetscFunctionReturn(0); 4247 } 4248 4249 #undef __FUNCT__ 4250 #define __FUNCT__ "MatCreate_SeqAIJ" 4251 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B) 4252 { 4253 Mat_SeqAIJ *b; 4254 PetscErrorCode ierr; 4255 PetscMPIInt size; 4256 4257 PetscFunctionBegin; 4258 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);CHKERRQ(ierr); 4259 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1"); 4260 4261 ierr = PetscNewLog(B,Mat_SeqAIJ,&b);CHKERRQ(ierr); 4262 4263 B->data = (void*)b; 4264 4265 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 4266 4267 b->row = 0; 4268 b->col = 0; 4269 b->icol = 0; 4270 b->reallocs = 0; 4271 b->ignorezeroentries = PETSC_FALSE; 4272 b->roworiented = PETSC_TRUE; 4273 b->nonew = 0; 4274 b->diag = 0; 4275 b->solve_work = 0; 4276 B->spptr = 0; 4277 b->saved_values = 0; 4278 b->idiag = 0; 4279 b->mdiag = 0; 4280 b->ssor_work = 0; 4281 b->omega = 1.0; 4282 b->fshift = 0.0; 4283 b->idiagvalid = PETSC_FALSE; 4284 b->ibdiagvalid = PETSC_FALSE; 4285 b->keepnonzeropattern = PETSC_FALSE; 4286 b->xtoy = 0; 4287 b->XtoY = 0; 4288 B->same_nonzero = PETSC_FALSE; 4289 4290 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4291 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);CHKERRQ(ierr); 4292 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);CHKERRQ(ierr); 4293 4294 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4295 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_matlab_C",MatGetFactor_seqaij_matlab);CHKERRQ(ierr); 4296 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);CHKERRQ(ierr); 4297 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);CHKERRQ(ierr); 4298 #endif 4299 #if defined(PETSC_HAVE_PASTIX) 4300 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_pastix_C",MatGetFactor_seqaij_pastix);CHKERRQ(ierr); 4301 #endif 4302 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4303 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_essl_C",MatGetFactor_seqaij_essl);CHKERRQ(ierr); 4304 #endif 4305 #if defined(PETSC_HAVE_SUPERLU) 4306 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_C",MatGetFactor_seqaij_superlu);CHKERRQ(ierr); 4307 #endif 4308 #if defined(PETSC_HAVE_SUPERLU_DIST) 4309 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_dist_C",MatGetFactor_seqaij_superlu_dist);CHKERRQ(ierr); 4310 #endif 4311 #if defined(PETSC_HAVE_MUMPS) 4312 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mumps_C",MatGetFactor_aij_mumps);CHKERRQ(ierr); 4313 #endif 4314 #if defined(PETSC_HAVE_UMFPACK) 4315 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_umfpack_C",MatGetFactor_seqaij_umfpack);CHKERRQ(ierr); 4316 #endif 4317 #if defined(PETSC_HAVE_CHOLMOD) 4318 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_cholmod_C",MatGetFactor_seqaij_cholmod);CHKERRQ(ierr); 4319 #endif 4320 #if defined(PETSC_HAVE_LUSOL) 4321 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_lusol_C",MatGetFactor_seqaij_lusol);CHKERRQ(ierr); 4322 #endif 4323 #if defined(PETSC_HAVE_CLIQUE) 4324 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_clique_C",MatGetFactor_aij_clique);CHKERRQ(ierr); 4325 #endif 4326 4327 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_petsc_C",MatGetFactor_seqaij_petsc);CHKERRQ(ierr); 4328 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactorAvailable_petsc_C",MatGetFactorAvailable_seqaij_petsc);CHKERRQ(ierr); 4329 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_bas_C",MatGetFactor_seqaij_bas);CHKERRQ(ierr); 4330 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);CHKERRQ(ierr); 4331 ierr = PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);CHKERRQ(ierr); 4332 ierr = PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);CHKERRQ(ierr); 4333 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);CHKERRQ(ierr); 4334 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);CHKERRQ(ierr); 4335 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);CHKERRQ(ierr); 4336 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);CHKERRQ(ierr); 4337 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4338 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4339 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);CHKERRQ(ierr); 4340 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);CHKERRQ(ierr); 4341 ierr = PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);CHKERRQ(ierr); 4342 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMult_seqdense_seqaij_C",MatMatMult_SeqDense_SeqAIJ);CHKERRQ(ierr); 4343 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_seqdense_seqaij_C",MatMatMultSymbolic_SeqDense_SeqAIJ);CHKERRQ(ierr); 4344 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_seqdense_seqaij_C",MatMatMultNumeric_SeqDense_SeqAIJ);CHKERRQ(ierr); 4345 ierr = MatCreate_SeqAIJ_Inode(B);CHKERRQ(ierr); 4346 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4347 PetscFunctionReturn(0); 4348 } 4349 4350 #undef __FUNCT__ 4351 #define __FUNCT__ "MatDuplicateNoCreate_SeqAIJ" 4352 /* 4353 Given a matrix generated with MatGetFactor() duplicates all the information in A into B 4354 */ 4355 PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace) 4356 { 4357 Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data; 4358 PetscErrorCode ierr; 4359 PetscInt i,m = A->rmap->n; 4360 4361 PetscFunctionBegin; 4362 c = (Mat_SeqAIJ*)C->data; 4363 4364 C->factortype = A->factortype; 4365 c->row = 0; 4366 c->col = 0; 4367 c->icol = 0; 4368 c->reallocs = 0; 4369 4370 C->assembled = PETSC_TRUE; 4371 4372 ierr = PetscLayoutReference(A->rmap,&C->rmap);CHKERRQ(ierr); 4373 ierr = PetscLayoutReference(A->cmap,&C->cmap);CHKERRQ(ierr); 4374 4375 ierr = PetscMalloc2(m,PetscInt,&c->imax,m,PetscInt,&c->ilen);CHKERRQ(ierr); 4376 ierr = PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));CHKERRQ(ierr); 4377 for (i=0; i<m; i++) { 4378 c->imax[i] = a->imax[i]; 4379 c->ilen[i] = a->ilen[i]; 4380 } 4381 4382 /* allocate the matrix space */ 4383 if (mallocmatspace) { 4384 ierr = PetscMalloc3(a->i[m],PetscScalar,&c->a,a->i[m],PetscInt,&c->j,m+1,PetscInt,&c->i);CHKERRQ(ierr); 4385 ierr = PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4386 4387 c->singlemalloc = PETSC_TRUE; 4388 4389 ierr = PetscMemcpy(c->i,a->i,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4390 if (m > 0) { 4391 ierr = PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(PetscInt));CHKERRQ(ierr); 4392 if (cpvalues == MAT_COPY_VALUES) { 4393 ierr = PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4394 } else { 4395 ierr = PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4396 } 4397 } 4398 } 4399 4400 c->ignorezeroentries = a->ignorezeroentries; 4401 c->roworiented = a->roworiented; 4402 c->nonew = a->nonew; 4403 if (a->diag) { 4404 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&c->diag);CHKERRQ(ierr); 4405 ierr = PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4406 for (i=0; i<m; i++) { 4407 c->diag[i] = a->diag[i]; 4408 } 4409 } else c->diag = 0; 4410 4411 c->solve_work = 0; 4412 c->saved_values = 0; 4413 c->idiag = 0; 4414 c->ssor_work = 0; 4415 c->keepnonzeropattern = a->keepnonzeropattern; 4416 c->free_a = PETSC_TRUE; 4417 c->free_ij = PETSC_TRUE; 4418 c->xtoy = 0; 4419 c->XtoY = 0; 4420 4421 c->rmax = a->rmax; 4422 c->nz = a->nz; 4423 c->maxnz = a->nz; /* Since we allocate exactly the right amount */ 4424 C->preallocated = PETSC_TRUE; 4425 4426 c->compressedrow.use = a->compressedrow.use; 4427 c->compressedrow.nrows = a->compressedrow.nrows; 4428 c->compressedrow.check = a->compressedrow.check; 4429 if (a->compressedrow.use) { 4430 i = a->compressedrow.nrows; 4431 ierr = PetscMalloc2(i+1,PetscInt,&c->compressedrow.i,i,PetscInt,&c->compressedrow.rindex);CHKERRQ(ierr); 4432 ierr = PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));CHKERRQ(ierr); 4433 ierr = PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));CHKERRQ(ierr); 4434 } else { 4435 c->compressedrow.use = PETSC_FALSE; 4436 c->compressedrow.i = NULL; 4437 c->compressedrow.rindex = NULL; 4438 } 4439 C->same_nonzero = A->same_nonzero; 4440 4441 ierr = MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);CHKERRQ(ierr); 4442 ierr = PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);CHKERRQ(ierr); 4443 PetscFunctionReturn(0); 4444 } 4445 4446 #undef __FUNCT__ 4447 #define __FUNCT__ "MatDuplicate_SeqAIJ" 4448 PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B) 4449 { 4450 PetscErrorCode ierr; 4451 4452 PetscFunctionBegin; 4453 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 4454 ierr = MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);CHKERRQ(ierr); 4455 ierr = MatSetBlockSizes(*B,A->rmap->bs,A->cmap->bs);CHKERRQ(ierr); 4456 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 4457 ierr = MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);CHKERRQ(ierr); 4458 PetscFunctionReturn(0); 4459 } 4460 4461 #undef __FUNCT__ 4462 #define __FUNCT__ "MatLoad_SeqAIJ" 4463 PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer) 4464 { 4465 Mat_SeqAIJ *a; 4466 PetscErrorCode ierr; 4467 PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N,rows,cols; 4468 int fd; 4469 PetscMPIInt size; 4470 MPI_Comm comm; 4471 PetscInt bs = 1; 4472 4473 PetscFunctionBegin; 4474 ierr = PetscObjectGetComm((PetscObject)viewer,&comm);CHKERRQ(ierr); 4475 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4476 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"view must have one processor"); 4477 4478 ierr = PetscOptionsBegin(comm,NULL,"Options for loading SEQAIJ matrix","Mat");CHKERRQ(ierr); 4479 ierr = PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);CHKERRQ(ierr); 4480 ierr = PetscOptionsEnd();CHKERRQ(ierr); 4481 if (bs > 1) {ierr = MatSetBlockSize(newMat,bs);CHKERRQ(ierr);} 4482 4483 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 4484 ierr = PetscBinaryRead(fd,header,4,PETSC_INT);CHKERRQ(ierr); 4485 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file"); 4486 M = header[1]; N = header[2]; nz = header[3]; 4487 4488 if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ"); 4489 4490 /* read in row lengths */ 4491 ierr = PetscMalloc(M*sizeof(PetscInt),&rowlengths);CHKERRQ(ierr); 4492 ierr = PetscBinaryRead(fd,rowlengths,M,PETSC_INT);CHKERRQ(ierr); 4493 4494 /* check if sum of rowlengths is same as nz */ 4495 for (i=0,sum=0; i< M; i++) sum +=rowlengths[i]; 4496 if (sum != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_READ,"Inconsistant matrix data in file. no-nonzeros = %d, sum-row-lengths = %d\n",nz,sum); 4497 4498 /* set global size if not set already*/ 4499 if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) { 4500 ierr = MatSetSizes(newMat,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr); 4501 } else { 4502 /* if sizes and type are already set, check if the vector global sizes are correct */ 4503 ierr = MatGetSize(newMat,&rows,&cols);CHKERRQ(ierr); 4504 if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */ 4505 ierr = MatGetLocalSize(newMat,&rows,&cols);CHKERRQ(ierr); 4506 } 4507 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); 4508 } 4509 ierr = MatSeqAIJSetPreallocation_SeqAIJ(newMat,0,rowlengths);CHKERRQ(ierr); 4510 a = (Mat_SeqAIJ*)newMat->data; 4511 4512 ierr = PetscBinaryRead(fd,a->j,nz,PETSC_INT);CHKERRQ(ierr); 4513 4514 /* read in nonzero values */ 4515 ierr = PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);CHKERRQ(ierr); 4516 4517 /* set matrix "i" values */ 4518 a->i[0] = 0; 4519 for (i=1; i<= M; i++) { 4520 a->i[i] = a->i[i-1] + rowlengths[i-1]; 4521 a->ilen[i-1] = rowlengths[i-1]; 4522 } 4523 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 4524 4525 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4526 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4527 PetscFunctionReturn(0); 4528 } 4529 4530 #undef __FUNCT__ 4531 #define __FUNCT__ "MatEqual_SeqAIJ" 4532 PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg) 4533 { 4534 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data; 4535 PetscErrorCode ierr; 4536 #if defined(PETSC_USE_COMPLEX) 4537 PetscInt k; 4538 #endif 4539 4540 PetscFunctionBegin; 4541 /* If the matrix dimensions are not equal,or no of nonzeros */ 4542 if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) { 4543 *flg = PETSC_FALSE; 4544 PetscFunctionReturn(0); 4545 } 4546 4547 /* if the a->i are the same */ 4548 ierr = PetscMemcmp(a->i,b->i,(A->rmap->n+1)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4549 if (!*flg) PetscFunctionReturn(0); 4550 4551 /* if a->j are the same */ 4552 ierr = PetscMemcmp(a->j,b->j,(a->nz)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4553 if (!*flg) PetscFunctionReturn(0); 4554 4555 /* if a->a are the same */ 4556 #if defined(PETSC_USE_COMPLEX) 4557 for (k=0; k<a->nz; k++) { 4558 if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) { 4559 *flg = PETSC_FALSE; 4560 PetscFunctionReturn(0); 4561 } 4562 } 4563 #else 4564 ierr = PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);CHKERRQ(ierr); 4565 #endif 4566 PetscFunctionReturn(0); 4567 } 4568 4569 #undef __FUNCT__ 4570 #define __FUNCT__ "MatCreateSeqAIJWithArrays" 4571 /*@ 4572 MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format) 4573 provided by the user. 4574 4575 Collective on MPI_Comm 4576 4577 Input Parameters: 4578 + comm - must be an MPI communicator of size 1 4579 . m - number of rows 4580 . n - number of columns 4581 . i - row indices 4582 . j - column indices 4583 - a - matrix values 4584 4585 Output Parameter: 4586 . mat - the matrix 4587 4588 Level: intermediate 4589 4590 Notes: 4591 The i, j, and a arrays are not copied by this routine, the user must free these arrays 4592 once the matrix is destroyed and not before 4593 4594 You cannot set new nonzero locations into this matrix, that will generate an error. 4595 4596 The i and j indices are 0 based 4597 4598 The format which is used for the sparse matrix input, is equivalent to a 4599 row-major ordering.. i.e for the following matrix, the input data expected is 4600 as shown: 4601 4602 1 0 0 4603 2 0 3 4604 4 5 6 4605 4606 i = {0,1,3,6} [size = nrow+1 = 3+1] 4607 j = {0,0,2,0,1,2} [size = nz = 6]; values must be sorted for each row 4608 v = {1,2,3,4,5,6} [size = nz = 6] 4609 4610 4611 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4612 4613 @*/ 4614 PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat) 4615 { 4616 PetscErrorCode ierr; 4617 PetscInt ii; 4618 Mat_SeqAIJ *aij; 4619 #if defined(PETSC_USE_DEBUG) 4620 PetscInt jj; 4621 #endif 4622 4623 PetscFunctionBegin; 4624 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 4625 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4626 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4627 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4628 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4629 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);CHKERRQ(ierr); 4630 aij = (Mat_SeqAIJ*)(*mat)->data; 4631 ierr = PetscMalloc2(m,PetscInt,&aij->imax,m,PetscInt,&aij->ilen);CHKERRQ(ierr); 4632 4633 aij->i = i; 4634 aij->j = j; 4635 aij->a = a; 4636 aij->singlemalloc = PETSC_FALSE; 4637 aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 4638 aij->free_a = PETSC_FALSE; 4639 aij->free_ij = PETSC_FALSE; 4640 4641 for (ii=0; ii<m; ii++) { 4642 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 4643 #if defined(PETSC_USE_DEBUG) 4644 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]); 4645 for (jj=i[ii]+1; jj<i[ii+1]; jj++) { 4646 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); 4647 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); 4648 } 4649 #endif 4650 } 4651 #if defined(PETSC_USE_DEBUG) 4652 for (ii=0; ii<aij->i[m]; ii++) { 4653 if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %d index = %d",ii,j[ii]); 4654 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]); 4655 } 4656 #endif 4657 4658 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4659 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4660 PetscFunctionReturn(0); 4661 } 4662 #undef __FUNCT__ 4663 #define __FUNCT__ "MatCreateSeqAIJFromTriple" 4664 /*@C 4665 MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format) 4666 provided by the user. 4667 4668 Collective on MPI_Comm 4669 4670 Input Parameters: 4671 + comm - must be an MPI communicator of size 1 4672 . m - number of rows 4673 . n - number of columns 4674 . i - row indices 4675 . j - column indices 4676 . a - matrix values 4677 . nz - number of nonzeros 4678 - idx - 0 or 1 based 4679 4680 Output Parameter: 4681 . mat - the matrix 4682 4683 Level: intermediate 4684 4685 Notes: 4686 The i and j indices are 0 based 4687 4688 The format which is used for the sparse matrix input, is equivalent to a 4689 row-major ordering.. i.e for the following matrix, the input data expected is 4690 as shown: 4691 4692 1 0 0 4693 2 0 3 4694 4 5 6 4695 4696 i = {0,1,1,2,2,2} 4697 j = {0,0,2,0,1,2} 4698 v = {1,2,3,4,5,6} 4699 4700 4701 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4702 4703 @*/ 4704 PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat,PetscInt nz,PetscBool idx) 4705 { 4706 PetscErrorCode ierr; 4707 PetscInt ii, *nnz, one = 1,row,col; 4708 4709 4710 PetscFunctionBegin; 4711 ierr = PetscMalloc(m*sizeof(PetscInt),&nnz);CHKERRQ(ierr); 4712 ierr = PetscMemzero(nnz,m*sizeof(PetscInt));CHKERRQ(ierr); 4713 for (ii = 0; ii < nz; ii++) { 4714 nnz[i[ii]] += 1; 4715 } 4716 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4717 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4718 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4719 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4720 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);CHKERRQ(ierr); 4721 for (ii = 0; ii < nz; ii++) { 4722 if (idx) { 4723 row = i[ii] - 1; 4724 col = j[ii] - 1; 4725 } else { 4726 row = i[ii]; 4727 col = j[ii]; 4728 } 4729 ierr = MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);CHKERRQ(ierr); 4730 } 4731 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4732 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4733 ierr = PetscFree(nnz);CHKERRQ(ierr); 4734 PetscFunctionReturn(0); 4735 } 4736 4737 #undef __FUNCT__ 4738 #define __FUNCT__ "MatSetColoring_SeqAIJ" 4739 PetscErrorCode MatSetColoring_SeqAIJ(Mat A,ISColoring coloring) 4740 { 4741 PetscErrorCode ierr; 4742 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4743 4744 PetscFunctionBegin; 4745 if (coloring->ctype == IS_COLORING_GLOBAL) { 4746 ierr = ISColoringReference(coloring);CHKERRQ(ierr); 4747 a->coloring = coloring; 4748 } else if (coloring->ctype == IS_COLORING_GHOSTED) { 4749 PetscInt i,*larray; 4750 ISColoring ocoloring; 4751 ISColoringValue *colors; 4752 4753 /* set coloring for diagonal portion */ 4754 ierr = PetscMalloc(A->cmap->n*sizeof(PetscInt),&larray);CHKERRQ(ierr); 4755 for (i=0; i<A->cmap->n; i++) larray[i] = i; 4756 ierr = ISGlobalToLocalMappingApply(A->cmap->mapping,IS_GTOLM_MASK,A->cmap->n,larray,NULL,larray);CHKERRQ(ierr); 4757 ierr = PetscMalloc(A->cmap->n*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 4758 for (i=0; i<A->cmap->n; i++) colors[i] = coloring->colors[larray[i]]; 4759 ierr = PetscFree(larray);CHKERRQ(ierr); 4760 ierr = ISColoringCreate(PETSC_COMM_SELF,coloring->n,A->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4761 a->coloring = ocoloring; 4762 } 4763 PetscFunctionReturn(0); 4764 } 4765 4766 #undef __FUNCT__ 4767 #define __FUNCT__ "MatSetValuesAdifor_SeqAIJ" 4768 PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat A,PetscInt nl,void *advalues) 4769 { 4770 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4771 PetscInt m = A->rmap->n,*ii = a->i,*jj = a->j,nz,i,j; 4772 MatScalar *v = a->a; 4773 PetscScalar *values = (PetscScalar*)advalues; 4774 ISColoringValue *color; 4775 4776 PetscFunctionBegin; 4777 if (!a->coloring) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix"); 4778 color = a->coloring->colors; 4779 /* loop over rows */ 4780 for (i=0; i<m; i++) { 4781 nz = ii[i+1] - ii[i]; 4782 /* loop over columns putting computed value into matrix */ 4783 for (j=0; j<nz; j++) *v++ = values[color[*jj++]]; 4784 values += nl; /* jump to next row of derivatives */ 4785 } 4786 PetscFunctionReturn(0); 4787 } 4788 4789 #undef __FUNCT__ 4790 #define __FUNCT__ "MatSeqAIJInvalidateDiagonal" 4791 PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A) 4792 { 4793 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 4794 PetscErrorCode ierr; 4795 4796 PetscFunctionBegin; 4797 a->idiagvalid = PETSC_FALSE; 4798 a->ibdiagvalid = PETSC_FALSE; 4799 4800 ierr = MatSeqAIJInvalidateDiagonal_Inode(A);CHKERRQ(ierr); 4801 PetscFunctionReturn(0); 4802 } 4803 4804 /* 4805 Special version for direct calls from Fortran 4806 */ 4807 #include <petsc-private/fortranimpl.h> 4808 #if defined(PETSC_HAVE_FORTRAN_CAPS) 4809 #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ 4810 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 4811 #define matsetvaluesseqaij_ matsetvaluesseqaij 4812 #endif 4813 4814 /* Change these macros so can be used in void function */ 4815 #undef CHKERRQ 4816 #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr) 4817 #undef SETERRQ2 4818 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 4819 #undef SETERRQ3 4820 #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr) 4821 4822 #undef __FUNCT__ 4823 #define __FUNCT__ "matsetvaluesseqaij_" 4824 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) 4825 { 4826 Mat A = *AA; 4827 PetscInt m = *mm, n = *nn; 4828 InsertMode is = *isis; 4829 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4830 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 4831 PetscInt *imax,*ai,*ailen; 4832 PetscErrorCode ierr; 4833 PetscInt *aj,nonew = a->nonew,lastcol = -1; 4834 MatScalar *ap,value,*aa; 4835 PetscBool ignorezeroentries = a->ignorezeroentries; 4836 PetscBool roworiented = a->roworiented; 4837 4838 PetscFunctionBegin; 4839 MatCheckPreallocated(A,1); 4840 imax = a->imax; 4841 ai = a->i; 4842 ailen = a->ilen; 4843 aj = a->j; 4844 aa = a->a; 4845 4846 for (k=0; k<m; k++) { /* loop over added rows */ 4847 row = im[k]; 4848 if (row < 0) continue; 4849 #if defined(PETSC_USE_DEBUG) 4850 if (row >= A->rmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large"); 4851 #endif 4852 rp = aj + ai[row]; ap = aa + ai[row]; 4853 rmax = imax[row]; nrow = ailen[row]; 4854 low = 0; 4855 high = nrow; 4856 for (l=0; l<n; l++) { /* loop over added columns */ 4857 if (in[l] < 0) continue; 4858 #if defined(PETSC_USE_DEBUG) 4859 if (in[l] >= A->cmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large"); 4860 #endif 4861 col = in[l]; 4862 if (roworiented) value = v[l + k*n]; 4863 else value = v[k + l*m]; 4864 4865 if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue; 4866 4867 if (col <= lastcol) low = 0; 4868 else high = nrow; 4869 lastcol = col; 4870 while (high-low > 5) { 4871 t = (low+high)/2; 4872 if (rp[t] > col) high = t; 4873 else low = t; 4874 } 4875 for (i=low; i<high; i++) { 4876 if (rp[i] > col) break; 4877 if (rp[i] == col) { 4878 if (is == ADD_VALUES) ap[i] += value; 4879 else ap[i] = value; 4880 goto noinsert; 4881 } 4882 } 4883 if (value == 0.0 && ignorezeroentries) goto noinsert; 4884 if (nonew == 1) goto noinsert; 4885 if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); 4886 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 4887 N = nrow++ - 1; a->nz++; high++; 4888 /* shift up all the later entries in this row */ 4889 for (ii=N; ii>=i; ii--) { 4890 rp[ii+1] = rp[ii]; 4891 ap[ii+1] = ap[ii]; 4892 } 4893 rp[i] = col; 4894 ap[i] = value; 4895 noinsert:; 4896 low = i + 1; 4897 } 4898 ailen[row] = nrow; 4899 } 4900 A->same_nonzero = PETSC_FALSE; 4901 PetscFunctionReturnVoid(); 4902 } 4903 4904 4905