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,l,row,col,N; 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 Mat_SeqAIJ *csp=(Mat_SeqAIJ*)J->data; 2714 PetscScalar *ca=csp->a; 2715 const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows; 2716 PetscInt *den2sp=coloring->den2sp,*idx; 2717 PetscInt **rows=coloring->rows,**columns=coloring->columns,ncolumns_k,nrows_k,**columnsforrow=coloring->columnsforrow; 2718 #if defined(JACOBIANCOLOROPT) 2719 PetscLogDouble t0,t1,t_init=0.0,t_setvals=0.0,t_func=0.0,t_dx=0.0,t_kl=0.0,t00,t11; 2720 #endif 2721 PetscInt kl; 2722 2723 PetscFunctionBegin; 2724 #if defined(JACOBIANCOLOROPT) 2725 ierr = PetscTime(&t0);CHKERRQ(ierr); 2726 #endif 2727 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 2728 ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 2729 if (flg) { 2730 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 2731 } else { 2732 PetscBool assembled; 2733 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 2734 if (assembled) { 2735 ierr = MatZeroEntries(J);CHKERRQ(ierr); 2736 } 2737 } 2738 2739 if (!coloring->vscale) { 2740 ierr = VecDuplicate(x1,&coloring->vscale);CHKERRQ(ierr); 2741 } 2742 2743 if (coloring->htype[0] == 'w') { /* tacky test; need to make systematic if we add other approaches to computing h*/ 2744 ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr); 2745 } 2746 #if defined(JACOBIANCOLOROPT) 2747 ierr = PetscTime(&t1);CHKERRQ(ierr); 2748 t_init += t1 - t0; 2749 #endif 2750 2751 /* Set w1 = F(x1) */ 2752 #if defined(JACOBIANCOLOROPT) 2753 ierr = PetscTime(&t0);CHKERRQ(ierr); 2754 #endif 2755 if (!coloring->fset) { 2756 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2757 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 2758 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2759 } else { 2760 coloring->fset = PETSC_FALSE; 2761 } 2762 #if defined(JACOBIANCOLOROPT) 2763 ierr = PetscTime(&t1);CHKERRQ(ierr); 2764 t_setvals += t1 - t0; 2765 #endif 2766 2767 #if defined(JACOBIANCOLOROPT) 2768 ierr = PetscTime(&t0);CHKERRQ(ierr); 2769 #endif 2770 if (!coloring->w3) { 2771 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 2772 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 2773 } 2774 w3 = coloring->w3; 2775 2776 /* Compute scale factors: vscale = 1./dx = 1./(epsilon*xx) */ 2777 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr); 2778 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2779 ierr = VecGetSize(x1,&N);CHKERRQ(ierr); 2780 for (col=0; col<N; col++) { 2781 if (coloring->htype[0] == 'w') { 2782 dx = 1.0 + unorm; 2783 } else { 2784 dx = xx[col]; 2785 } 2786 if (dx == (PetscScalar)0.0) dx = 1.0; 2787 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2788 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2789 dx *= epsilon; 2790 vscale_array[col] = (PetscScalar)dx; 2791 } 2792 #if defined(JACOBIANCOLOROPT) 2793 ierr = PetscTime(&t1);CHKERRQ(ierr); 2794 t_init += t1 - t0; 2795 #endif 2796 2797 idx = den2sp; 2798 for (k=0; k<ncolors; k++) { /* loop over colors */ 2799 #if defined(JACOBIANCOLOROPT) 2800 ierr = PetscTime(&t0);CHKERRQ(ierr); 2801 #endif 2802 coloring->currentcolor = k; 2803 2804 /* 2805 Loop over each column associated with color 2806 adding the perturbation to the vector w3 = x1 + dx. 2807 */ 2808 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 2809 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 2810 ncolumns_k = ncolumns[k]; 2811 for (l=0; l<ncolumns_k; l++) { /* loop over columns */ 2812 col = columns[k][l]; 2813 w3_array[col] += vscale_array[col]; 2814 } 2815 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 2816 #if defined(JACOBIANCOLOROPT) 2817 ierr = PetscTime(&t1);CHKERRQ(ierr); 2818 t_dx += t1 - t0; 2819 #endif 2820 2821 /* 2822 Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 2823 w2 = F(x1 + dx) - F(x1) 2824 */ 2825 #if defined(JACOBIANCOLOROPT) 2826 ierr = PetscTime(&t0);CHKERRQ(ierr); 2827 #endif 2828 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2829 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 2830 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2831 #if defined(JACOBIANCOLOROPT) 2832 ierr = PetscTime(&t1);CHKERRQ(ierr); 2833 t_func += t1 - t0; 2834 #endif 2835 2836 #if defined(JACOBIANCOLOROPT) 2837 ierr = PetscTime(&t0);CHKERRQ(ierr); 2838 #endif 2839 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 2840 2841 /* 2842 Loop over rows of vector, putting w2/dx into Jacobian matrix 2843 */ 2844 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 2845 nrows_k = nrows[k]; 2846 for (l=0; l<nrows_k; l++) { /* loop over rows */ 2847 #if defined(JACOBIANCOLOROPT) 2848 ierr = PetscTime(&t00);CHKERRQ(ierr); 2849 #endif 2850 row = rows[k][l]; /* row index */ 2851 kl = columnsforrow[k][l]; 2852 #if defined(JACOBIANCOLOROPT) 2853 ierr = PetscTime(&t11);CHKERRQ(ierr); 2854 t_kl += t11 - t00; 2855 #endif 2856 2857 ca[idx[l]] = y[row]/vscale_array[kl];//y[row]/vscale_array[columnsforrow[k][l]]; 2858 } 2859 idx += nrows_k; 2860 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 2861 2862 #if defined(JACOBIANCOLOROPT) 2863 ierr = PetscTime(&t1);CHKERRQ(ierr); 2864 t_setvals += t1 - t0; 2865 #endif 2866 } /* endof for each color */ 2867 #if defined(JACOBIANCOLOROPT) 2868 printf(" FDColorApply time: init %g + func %g + setvalues %g + dx %g= %g\n",t_init,t_func,t_setvals,t_dx,t_init+t_func+t_setvals+t_dx); 2869 printf(" FDColorApply time: kl %g\n",t_kl); 2870 #endif 2871 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2872 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 2873 2874 coloring->currentcolor = -1; 2875 PetscFunctionReturn(0); 2876 } 2877 /* --------------------------------------------------------*/ 2878 2879 #undef __FUNCT__ 2880 #define __FUNCT__ "MatFDColoringApply_SeqAIJ_old" 2881 PetscErrorCode MatFDColoringApply_SeqAIJ_old(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 2882 { 2883 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 2884 PetscErrorCode ierr; 2885 PetscInt k,N,start,end,l,row,col,srow,**vscaleforrow; 2886 PetscScalar dx,*y,*xx,*w3_array; 2887 PetscScalar *vscale_array; 2888 PetscReal epsilon = coloring->error_rel,umin = coloring->umin; 2889 Vec w1,w2,w3; 2890 void *fctx = coloring->fctx; 2891 PetscBool flg = PETSC_FALSE; 2892 2893 PetscFunctionBegin; 2894 printf("MatFDColoringApply_SeqAIJ ...\n"); 2895 if (!coloring->w1) { 2896 ierr = VecDuplicate(x1,&coloring->w1);CHKERRQ(ierr); 2897 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w1);CHKERRQ(ierr); 2898 ierr = VecDuplicate(x1,&coloring->w2);CHKERRQ(ierr); 2899 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w2);CHKERRQ(ierr); 2900 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 2901 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 2902 } 2903 w1 = coloring->w1; w2 = coloring->w2; w3 = coloring->w3; 2904 2905 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 2906 ierr = PetscOptionsGetBool(((PetscObject)coloring)->prefix,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 2907 if (flg) { 2908 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 2909 } else { 2910 PetscBool assembled; 2911 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 2912 if (assembled) { 2913 ierr = MatZeroEntries(J);CHKERRQ(ierr); 2914 } 2915 } 2916 2917 ierr = VecGetOwnershipRange(x1,&start,&end);CHKERRQ(ierr); 2918 ierr = VecGetSize(x1,&N);CHKERRQ(ierr); 2919 2920 if (!coloring->fset) { 2921 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2922 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 2923 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2924 } else { 2925 coloring->fset = PETSC_FALSE; 2926 } 2927 2928 /* 2929 Compute all the scale factors and share with other processors 2930 */ 2931 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr);xx = xx - start; 2932 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);vscale_array = vscale_array - start; 2933 for (k=0; k<coloring->ncolors; k++) { 2934 /* 2935 Loop over each column associated with color adding the 2936 perturbation to the vector w3. 2937 */ 2938 for (l=0; l<coloring->ncolumns[k]; l++) { 2939 col = coloring->columns[k][l]; /* column of the matrix we are probing for */ 2940 dx = xx[col]; 2941 if (dx == 0.0) dx = 1.0; 2942 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2943 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2944 dx *= epsilon; 2945 vscale_array[col] = 1.0/dx; 2946 } 2947 } 2948 vscale_array = vscale_array + start; 2949 2950 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2951 ierr = VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2952 ierr = VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2953 2954 /* ierr = VecView(coloring->vscale,PETSC_VIEWER_STDOUT_WORLD); 2955 ierr = VecView(x1,PETSC_VIEWER_STDOUT_WORLD);*/ 2956 2957 if (coloring->vscaleforrow) vscaleforrow = coloring->vscaleforrow; 2958 else vscaleforrow = coloring->columnsforrow; 2959 2960 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2961 /* 2962 Loop over each color 2963 */ 2964 for (k=0; k<coloring->ncolors; k++) { 2965 coloring->currentcolor = k; 2966 2967 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 2968 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr);w3_array = w3_array - start; 2969 /* 2970 Loop over each column associated with color adding the 2971 perturbation to the vector w3. 2972 */ 2973 for (l=0; l<coloring->ncolumns[k]; l++) { 2974 col = coloring->columns[k][l]; /* column of the matrix we are probing for */ 2975 dx = xx[col]; 2976 if (dx == 0.0) dx = 1.0; 2977 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2978 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2979 dx *= epsilon; 2980 if (!PetscAbsScalar(dx)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Computed 0 differencing parameter"); 2981 w3_array[col] += dx; 2982 } 2983 w3_array = w3_array + start; ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 2984 2985 /* 2986 Evaluate function at x1 + dx (here dx is a vector of perturbations) 2987 */ 2988 2989 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2990 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 2991 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2992 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 2993 2994 /* 2995 Loop over rows of vector, putting results into Jacobian matrix 2996 */ 2997 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 2998 for (l=0; l<coloring->nrows[k]; l++) { 2999 row = coloring->rows[k][l]; 3000 col = coloring->columnsforrow[k][l]; 3001 y[row] *= vscale_array[vscaleforrow[k][l]]; 3002 srow = row + start; 3003 ierr = MatSetValues_SeqAIJ(J,1,&srow,1,&col,y+row,INSERT_VALUES);CHKERRQ(ierr); 3004 } 3005 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 3006 } 3007 coloring->currentcolor = k; 3008 3009 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 3010 xx = xx + start; 3011 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 3012 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3013 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3014 PetscFunctionReturn(0); 3015 } 3016 3017 /* 3018 Computes the number of nonzeros per row needed for preallocation when X and Y 3019 have different nonzero structure. 3020 */ 3021 #undef __FUNCT__ 3022 #define __FUNCT__ "MatAXPYGetPreallocation_SeqAIJ" 3023 PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz) 3024 { 3025 PetscInt i,m=Y->rmap->N; 3026 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 3027 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 3028 const PetscInt *xi = x->i,*yi = y->i; 3029 3030 PetscFunctionBegin; 3031 /* Set the number of nonzeros in the new matrix */ 3032 for (i=0; i<m; i++) { 3033 PetscInt j,k,nzx = xi[i+1] - xi[i],nzy = yi[i+1] - yi[i]; 3034 const PetscInt *xj = x->j+xi[i],*yj = y->j+yi[i]; 3035 nnz[i] = 0; 3036 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 3037 for (; k<nzy && yj[k]<xj[j]; k++) nnz[i]++; /* Catch up to X */ 3038 if (k<nzy && yj[k]==xj[j]) k++; /* Skip duplicate */ 3039 nnz[i]++; 3040 } 3041 for (; k<nzy; k++) nnz[i]++; 3042 } 3043 PetscFunctionReturn(0); 3044 } 3045 3046 #undef __FUNCT__ 3047 #define __FUNCT__ "MatAXPY_SeqAIJ" 3048 PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 3049 { 3050 PetscErrorCode ierr; 3051 PetscInt i; 3052 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data; 3053 PetscBLASInt one=1,bnz; 3054 3055 PetscFunctionBegin; 3056 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 3057 if (str == SAME_NONZERO_PATTERN) { 3058 PetscScalar alpha = a; 3059 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 3060 ierr = MatSeqAIJInvalidateDiagonal(Y);CHKERRQ(ierr); 3061 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 3062 if (y->xtoy && y->XtoY != X) { 3063 ierr = PetscFree(y->xtoy);CHKERRQ(ierr); 3064 ierr = MatDestroy(&y->XtoY);CHKERRQ(ierr); 3065 } 3066 if (!y->xtoy) { /* get xtoy */ 3067 ierr = MatAXPYGetxtoy_Private(X->rmap->n,x->i,x->j,NULL, y->i,y->j,NULL, &y->xtoy);CHKERRQ(ierr); 3068 y->XtoY = X; 3069 ierr = PetscObjectReference((PetscObject)X);CHKERRQ(ierr); 3070 } 3071 for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]); 3072 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); 3073 } else { 3074 Mat B; 3075 PetscInt *nnz; 3076 ierr = PetscMalloc(Y->rmap->N*sizeof(PetscInt),&nnz);CHKERRQ(ierr); 3077 ierr = MatCreate(PetscObjectComm((PetscObject)Y),&B);CHKERRQ(ierr); 3078 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 3079 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 3080 ierr = MatSetBlockSizes(B,Y->rmap->bs,Y->cmap->bs);CHKERRQ(ierr); 3081 ierr = MatSetType(B,(MatType) ((PetscObject)Y)->type_name);CHKERRQ(ierr); 3082 ierr = MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);CHKERRQ(ierr); 3083 ierr = MatSeqAIJSetPreallocation(B,0,nnz);CHKERRQ(ierr); 3084 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 3085 ierr = MatHeaderReplace(Y,B);CHKERRQ(ierr); 3086 ierr = PetscFree(nnz);CHKERRQ(ierr); 3087 } 3088 PetscFunctionReturn(0); 3089 } 3090 3091 #undef __FUNCT__ 3092 #define __FUNCT__ "MatConjugate_SeqAIJ" 3093 PetscErrorCode MatConjugate_SeqAIJ(Mat mat) 3094 { 3095 #if defined(PETSC_USE_COMPLEX) 3096 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3097 PetscInt i,nz; 3098 PetscScalar *a; 3099 3100 PetscFunctionBegin; 3101 nz = aij->nz; 3102 a = aij->a; 3103 for (i=0; i<nz; i++) a[i] = PetscConj(a[i]); 3104 #else 3105 PetscFunctionBegin; 3106 #endif 3107 PetscFunctionReturn(0); 3108 } 3109 3110 #undef __FUNCT__ 3111 #define __FUNCT__ "MatGetRowMaxAbs_SeqAIJ" 3112 PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3113 { 3114 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3115 PetscErrorCode ierr; 3116 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3117 PetscReal atmp; 3118 PetscScalar *x; 3119 MatScalar *aa; 3120 3121 PetscFunctionBegin; 3122 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3123 aa = a->a; 3124 ai = a->i; 3125 aj = a->j; 3126 3127 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3128 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3129 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3130 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3131 for (i=0; i<m; i++) { 3132 ncols = ai[1] - ai[0]; ai++; 3133 x[i] = 0.0; 3134 for (j=0; j<ncols; j++) { 3135 atmp = PetscAbsScalar(*aa); 3136 if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3137 aa++; aj++; 3138 } 3139 } 3140 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3141 PetscFunctionReturn(0); 3142 } 3143 3144 #undef __FUNCT__ 3145 #define __FUNCT__ "MatGetRowMax_SeqAIJ" 3146 PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3147 { 3148 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3149 PetscErrorCode ierr; 3150 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3151 PetscScalar *x; 3152 MatScalar *aa; 3153 3154 PetscFunctionBegin; 3155 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3156 aa = a->a; 3157 ai = a->i; 3158 aj = a->j; 3159 3160 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3161 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3162 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3163 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3164 for (i=0; i<m; i++) { 3165 ncols = ai[1] - ai[0]; ai++; 3166 if (ncols == A->cmap->n) { /* row is dense */ 3167 x[i] = *aa; if (idx) idx[i] = 0; 3168 } else { /* row is sparse so already KNOW maximum is 0.0 or higher */ 3169 x[i] = 0.0; 3170 if (idx) { 3171 idx[i] = 0; /* in case ncols is zero */ 3172 for (j=0;j<ncols;j++) { /* find first implicit 0.0 in the row */ 3173 if (aj[j] > j) { 3174 idx[i] = j; 3175 break; 3176 } 3177 } 3178 } 3179 } 3180 for (j=0; j<ncols; j++) { 3181 if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3182 aa++; aj++; 3183 } 3184 } 3185 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3186 PetscFunctionReturn(0); 3187 } 3188 3189 #undef __FUNCT__ 3190 #define __FUNCT__ "MatGetRowMinAbs_SeqAIJ" 3191 PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3192 { 3193 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3194 PetscErrorCode ierr; 3195 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3196 PetscReal atmp; 3197 PetscScalar *x; 3198 MatScalar *aa; 3199 3200 PetscFunctionBegin; 3201 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3202 aa = a->a; 3203 ai = a->i; 3204 aj = a->j; 3205 3206 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3207 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3208 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3209 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); 3210 for (i=0; i<m; i++) { 3211 ncols = ai[1] - ai[0]; ai++; 3212 if (ncols) { 3213 /* Get first nonzero */ 3214 for (j = 0; j < ncols; j++) { 3215 atmp = PetscAbsScalar(aa[j]); 3216 if (atmp > 1.0e-12) { 3217 x[i] = atmp; 3218 if (idx) idx[i] = aj[j]; 3219 break; 3220 } 3221 } 3222 if (j == ncols) {x[i] = PetscAbsScalar(*aa); if (idx) idx[i] = *aj;} 3223 } else { 3224 x[i] = 0.0; if (idx) idx[i] = 0; 3225 } 3226 for (j = 0; j < ncols; j++) { 3227 atmp = PetscAbsScalar(*aa); 3228 if (atmp > 1.0e-12 && PetscAbsScalar(x[i]) > atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3229 aa++; aj++; 3230 } 3231 } 3232 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3233 PetscFunctionReturn(0); 3234 } 3235 3236 #undef __FUNCT__ 3237 #define __FUNCT__ "MatGetRowMin_SeqAIJ" 3238 PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3239 { 3240 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3241 PetscErrorCode ierr; 3242 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3243 PetscScalar *x; 3244 MatScalar *aa; 3245 3246 PetscFunctionBegin; 3247 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3248 aa = a->a; 3249 ai = a->i; 3250 aj = a->j; 3251 3252 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3253 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3254 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3255 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3256 for (i=0; i<m; i++) { 3257 ncols = ai[1] - ai[0]; ai++; 3258 if (ncols == A->cmap->n) { /* row is dense */ 3259 x[i] = *aa; if (idx) idx[i] = 0; 3260 } else { /* row is sparse so already KNOW minimum is 0.0 or lower */ 3261 x[i] = 0.0; 3262 if (idx) { /* find first implicit 0.0 in the row */ 3263 idx[i] = 0; /* in case ncols is zero */ 3264 for (j=0; j<ncols; j++) { 3265 if (aj[j] > j) { 3266 idx[i] = j; 3267 break; 3268 } 3269 } 3270 } 3271 } 3272 for (j=0; j<ncols; j++) { 3273 if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3274 aa++; aj++; 3275 } 3276 } 3277 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3278 PetscFunctionReturn(0); 3279 } 3280 3281 #include <petscblaslapack.h> 3282 #include <petsc-private/kernels/blockinvert.h> 3283 3284 #undef __FUNCT__ 3285 #define __FUNCT__ "MatInvertBlockDiagonal_SeqAIJ" 3286 PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values) 3287 { 3288 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 3289 PetscErrorCode ierr; 3290 PetscInt i,bs = A->rmap->bs,mbs = A->rmap->n/A->rmap->bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j; 3291 MatScalar *diag,work[25],*v_work; 3292 PetscReal shift = 0.0; 3293 3294 PetscFunctionBegin; 3295 if (a->ibdiagvalid) { 3296 if (values) *values = a->ibdiag; 3297 PetscFunctionReturn(0); 3298 } 3299 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 3300 if (!a->ibdiag) { 3301 ierr = PetscMalloc(bs2*mbs*sizeof(PetscScalar),&a->ibdiag);CHKERRQ(ierr); 3302 ierr = PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));CHKERRQ(ierr); 3303 } 3304 diag = a->ibdiag; 3305 if (values) *values = a->ibdiag; 3306 /* factor and invert each block */ 3307 switch (bs) { 3308 case 1: 3309 for (i=0; i<mbs; i++) { 3310 ierr = MatGetValues(A,1,&i,1,&i,diag+i);CHKERRQ(ierr); 3311 diag[i] = (PetscScalar)1.0 / (diag[i] + shift); 3312 } 3313 break; 3314 case 2: 3315 for (i=0; i<mbs; i++) { 3316 ij[0] = 2*i; ij[1] = 2*i + 1; 3317 ierr = MatGetValues(A,2,ij,2,ij,diag);CHKERRQ(ierr); 3318 ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 3319 ierr = PetscKernel_A_gets_transpose_A_2(diag);CHKERRQ(ierr); 3320 diag += 4; 3321 } 3322 break; 3323 case 3: 3324 for (i=0; i<mbs; i++) { 3325 ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2; 3326 ierr = MatGetValues(A,3,ij,3,ij,diag);CHKERRQ(ierr); 3327 ierr = PetscKernel_A_gets_inverse_A_3(diag,shift);CHKERRQ(ierr); 3328 ierr = PetscKernel_A_gets_transpose_A_3(diag);CHKERRQ(ierr); 3329 diag += 9; 3330 } 3331 break; 3332 case 4: 3333 for (i=0; i<mbs; i++) { 3334 ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3; 3335 ierr = MatGetValues(A,4,ij,4,ij,diag);CHKERRQ(ierr); 3336 ierr = PetscKernel_A_gets_inverse_A_4(diag,shift);CHKERRQ(ierr); 3337 ierr = PetscKernel_A_gets_transpose_A_4(diag);CHKERRQ(ierr); 3338 diag += 16; 3339 } 3340 break; 3341 case 5: 3342 for (i=0; i<mbs; i++) { 3343 ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4; 3344 ierr = MatGetValues(A,5,ij,5,ij,diag);CHKERRQ(ierr); 3345 ierr = PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift);CHKERRQ(ierr); 3346 ierr = PetscKernel_A_gets_transpose_A_5(diag);CHKERRQ(ierr); 3347 diag += 25; 3348 } 3349 break; 3350 case 6: 3351 for (i=0; i<mbs; i++) { 3352 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; 3353 ierr = MatGetValues(A,6,ij,6,ij,diag);CHKERRQ(ierr); 3354 ierr = PetscKernel_A_gets_inverse_A_6(diag,shift);CHKERRQ(ierr); 3355 ierr = PetscKernel_A_gets_transpose_A_6(diag);CHKERRQ(ierr); 3356 diag += 36; 3357 } 3358 break; 3359 case 7: 3360 for (i=0; i<mbs; i++) { 3361 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; 3362 ierr = MatGetValues(A,7,ij,7,ij,diag);CHKERRQ(ierr); 3363 ierr = PetscKernel_A_gets_inverse_A_7(diag,shift);CHKERRQ(ierr); 3364 ierr = PetscKernel_A_gets_transpose_A_7(diag);CHKERRQ(ierr); 3365 diag += 49; 3366 } 3367 break; 3368 default: 3369 ierr = PetscMalloc3(bs,MatScalar,&v_work,bs,PetscInt,&v_pivots,bs,PetscInt,&IJ);CHKERRQ(ierr); 3370 for (i=0; i<mbs; i++) { 3371 for (j=0; j<bs; j++) { 3372 IJ[j] = bs*i + j; 3373 } 3374 ierr = MatGetValues(A,bs,IJ,bs,IJ,diag);CHKERRQ(ierr); 3375 ierr = PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work);CHKERRQ(ierr); 3376 ierr = PetscKernel_A_gets_transpose_A_N(diag,bs);CHKERRQ(ierr); 3377 diag += bs2; 3378 } 3379 ierr = PetscFree3(v_work,v_pivots,IJ);CHKERRQ(ierr); 3380 } 3381 a->ibdiagvalid = PETSC_TRUE; 3382 PetscFunctionReturn(0); 3383 } 3384 3385 #undef __FUNCT__ 3386 #define __FUNCT__ "MatSetRandom_SeqAIJ" 3387 static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx) 3388 { 3389 PetscErrorCode ierr; 3390 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data; 3391 PetscScalar a; 3392 PetscInt m,n,i,j,col; 3393 3394 PetscFunctionBegin; 3395 if (!x->assembled) { 3396 ierr = MatGetSize(x,&m,&n);CHKERRQ(ierr); 3397 for (i=0; i<m; i++) { 3398 for (j=0; j<aij->imax[i]; j++) { 3399 ierr = PetscRandomGetValue(rctx,&a);CHKERRQ(ierr); 3400 col = (PetscInt)(n*PetscRealPart(a)); 3401 ierr = MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);CHKERRQ(ierr); 3402 } 3403 } 3404 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not yet coded"); 3405 ierr = MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3406 ierr = MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3407 PetscFunctionReturn(0); 3408 } 3409 3410 extern PetscErrorCode MatFDColoringApply_AIJ(Mat,MatFDColoring,Vec,MatStructure*,void*); 3411 /* -------------------------------------------------------------------*/ 3412 static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ, 3413 MatGetRow_SeqAIJ, 3414 MatRestoreRow_SeqAIJ, 3415 MatMult_SeqAIJ, 3416 /* 4*/ MatMultAdd_SeqAIJ, 3417 MatMultTranspose_SeqAIJ, 3418 MatMultTransposeAdd_SeqAIJ, 3419 0, 3420 0, 3421 0, 3422 /* 10*/ 0, 3423 MatLUFactor_SeqAIJ, 3424 0, 3425 MatSOR_SeqAIJ, 3426 MatTranspose_SeqAIJ, 3427 /*1 5*/ MatGetInfo_SeqAIJ, 3428 MatEqual_SeqAIJ, 3429 MatGetDiagonal_SeqAIJ, 3430 MatDiagonalScale_SeqAIJ, 3431 MatNorm_SeqAIJ, 3432 /* 20*/ 0, 3433 MatAssemblyEnd_SeqAIJ, 3434 MatSetOption_SeqAIJ, 3435 MatZeroEntries_SeqAIJ, 3436 /* 24*/ MatZeroRows_SeqAIJ, 3437 0, 3438 0, 3439 0, 3440 0, 3441 /* 29*/ MatSetUp_SeqAIJ, 3442 0, 3443 0, 3444 0, 3445 0, 3446 /* 34*/ MatDuplicate_SeqAIJ, 3447 0, 3448 0, 3449 MatILUFactor_SeqAIJ, 3450 0, 3451 /* 39*/ MatAXPY_SeqAIJ, 3452 MatGetSubMatrices_SeqAIJ, 3453 MatIncreaseOverlap_SeqAIJ, 3454 MatGetValues_SeqAIJ, 3455 MatCopy_SeqAIJ, 3456 /* 44*/ MatGetRowMax_SeqAIJ, 3457 MatScale_SeqAIJ, 3458 0, 3459 MatDiagonalSet_SeqAIJ, 3460 MatZeroRowsColumns_SeqAIJ, 3461 /* 49*/ MatSetRandom_SeqAIJ, 3462 MatGetRowIJ_SeqAIJ, 3463 MatRestoreRowIJ_SeqAIJ, 3464 MatGetColumnIJ_SeqAIJ, 3465 MatRestoreColumnIJ_SeqAIJ, 3466 /* 54*/ MatFDColoringCreate_SeqAIJ, 3467 0, 3468 0, 3469 MatPermute_SeqAIJ, 3470 0, 3471 /* 59*/ 0, 3472 MatDestroy_SeqAIJ, 3473 MatView_SeqAIJ, 3474 0, 3475 MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ, 3476 /* 64*/ MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ, 3477 MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ, 3478 0, 3479 0, 3480 0, 3481 /* 69*/ MatGetRowMaxAbs_SeqAIJ, 3482 MatGetRowMinAbs_SeqAIJ, 3483 0, 3484 MatSetColoring_SeqAIJ, 3485 0, 3486 /* 74*/ MatSetValuesAdifor_SeqAIJ, 3487 MatFDColoringApply_AIJ, 3488 0, 3489 0, 3490 0, 3491 /* 79*/ MatFindZeroDiagonals_SeqAIJ, 3492 0, 3493 0, 3494 0, 3495 MatLoad_SeqAIJ, 3496 /* 84*/ MatIsSymmetric_SeqAIJ, 3497 MatIsHermitian_SeqAIJ, 3498 0, 3499 0, 3500 0, 3501 /* 89*/ MatMatMult_SeqAIJ_SeqAIJ, 3502 MatMatMultSymbolic_SeqAIJ_SeqAIJ, 3503 MatMatMultNumeric_SeqAIJ_SeqAIJ, 3504 MatPtAP_SeqAIJ_SeqAIJ, 3505 MatPtAPSymbolic_SeqAIJ_SeqAIJ_DenseAxpy, 3506 /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ, 3507 MatMatTransposeMult_SeqAIJ_SeqAIJ, 3508 MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ, 3509 MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ, 3510 0, 3511 /* 99*/ 0, 3512 0, 3513 0, 3514 MatConjugate_SeqAIJ, 3515 0, 3516 /*104*/ MatSetValuesRow_SeqAIJ, 3517 MatRealPart_SeqAIJ, 3518 MatImaginaryPart_SeqAIJ, 3519 0, 3520 0, 3521 /*109*/ MatMatSolve_SeqAIJ, 3522 0, 3523 MatGetRowMin_SeqAIJ, 3524 0, 3525 MatMissingDiagonal_SeqAIJ, 3526 /*114*/ 0, 3527 0, 3528 0, 3529 0, 3530 0, 3531 /*119*/ 0, 3532 0, 3533 0, 3534 0, 3535 MatGetMultiProcBlock_SeqAIJ, 3536 /*124*/ MatFindNonzeroRows_SeqAIJ, 3537 MatGetColumnNorms_SeqAIJ, 3538 MatInvertBlockDiagonal_SeqAIJ, 3539 0, 3540 0, 3541 /*129*/ 0, 3542 MatTransposeMatMult_SeqAIJ_SeqAIJ, 3543 MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ, 3544 MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ, 3545 MatTransposeColoringCreate_SeqAIJ, 3546 /*134*/ MatTransColoringApplySpToDen_SeqAIJ, 3547 MatTransColoringApplyDenToSp_SeqAIJ, 3548 MatRARt_SeqAIJ_SeqAIJ, 3549 MatRARtSymbolic_SeqAIJ_SeqAIJ, 3550 MatRARtNumeric_SeqAIJ_SeqAIJ, 3551 /*139*/0, 3552 0 3553 }; 3554 3555 #undef __FUNCT__ 3556 #define __FUNCT__ "MatSeqAIJSetColumnIndices_SeqAIJ" 3557 PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices) 3558 { 3559 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3560 PetscInt i,nz,n; 3561 3562 PetscFunctionBegin; 3563 nz = aij->maxnz; 3564 n = mat->rmap->n; 3565 for (i=0; i<nz; i++) { 3566 aij->j[i] = indices[i]; 3567 } 3568 aij->nz = nz; 3569 for (i=0; i<n; i++) { 3570 aij->ilen[i] = aij->imax[i]; 3571 } 3572 PetscFunctionReturn(0); 3573 } 3574 3575 #undef __FUNCT__ 3576 #define __FUNCT__ "MatSeqAIJSetColumnIndices" 3577 /*@ 3578 MatSeqAIJSetColumnIndices - Set the column indices for all the rows 3579 in the matrix. 3580 3581 Input Parameters: 3582 + mat - the SeqAIJ matrix 3583 - indices - the column indices 3584 3585 Level: advanced 3586 3587 Notes: 3588 This can be called if you have precomputed the nonzero structure of the 3589 matrix and want to provide it to the matrix object to improve the performance 3590 of the MatSetValues() operation. 3591 3592 You MUST have set the correct numbers of nonzeros per row in the call to 3593 MatCreateSeqAIJ(), and the columns indices MUST be sorted. 3594 3595 MUST be called before any calls to MatSetValues(); 3596 3597 The indices should start with zero, not one. 3598 3599 @*/ 3600 PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices) 3601 { 3602 PetscErrorCode ierr; 3603 3604 PetscFunctionBegin; 3605 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3606 PetscValidPointer(indices,2); 3607 ierr = PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));CHKERRQ(ierr); 3608 PetscFunctionReturn(0); 3609 } 3610 3611 /* ----------------------------------------------------------------------------------------*/ 3612 3613 #undef __FUNCT__ 3614 #define __FUNCT__ "MatStoreValues_SeqAIJ" 3615 PetscErrorCode MatStoreValues_SeqAIJ(Mat mat) 3616 { 3617 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3618 PetscErrorCode ierr; 3619 size_t nz = aij->i[mat->rmap->n]; 3620 3621 PetscFunctionBegin; 3622 if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3623 3624 /* allocate space for values if not already there */ 3625 if (!aij->saved_values) { 3626 ierr = PetscMalloc((nz+1)*sizeof(PetscScalar),&aij->saved_values);CHKERRQ(ierr); 3627 ierr = PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));CHKERRQ(ierr); 3628 } 3629 3630 /* copy values over */ 3631 ierr = PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3632 PetscFunctionReturn(0); 3633 } 3634 3635 #undef __FUNCT__ 3636 #define __FUNCT__ "MatStoreValues" 3637 /*@ 3638 MatStoreValues - Stashes a copy of the matrix values; this allows, for 3639 example, reuse of the linear part of a Jacobian, while recomputing the 3640 nonlinear portion. 3641 3642 Collect on Mat 3643 3644 Input Parameters: 3645 . mat - the matrix (currently only AIJ matrices support this option) 3646 3647 Level: advanced 3648 3649 Common Usage, with SNESSolve(): 3650 $ Create Jacobian matrix 3651 $ Set linear terms into matrix 3652 $ Apply boundary conditions to matrix, at this time matrix must have 3653 $ final nonzero structure (i.e. setting the nonlinear terms and applying 3654 $ boundary conditions again will not change the nonzero structure 3655 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3656 $ ierr = MatStoreValues(mat); 3657 $ Call SNESSetJacobian() with matrix 3658 $ In your Jacobian routine 3659 $ ierr = MatRetrieveValues(mat); 3660 $ Set nonlinear terms in matrix 3661 3662 Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself: 3663 $ // build linear portion of Jacobian 3664 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3665 $ ierr = MatStoreValues(mat); 3666 $ loop over nonlinear iterations 3667 $ ierr = MatRetrieveValues(mat); 3668 $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian 3669 $ // call MatAssemblyBegin/End() on matrix 3670 $ Solve linear system with Jacobian 3671 $ endloop 3672 3673 Notes: 3674 Matrix must already be assemblied before calling this routine 3675 Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before 3676 calling this routine. 3677 3678 When this is called multiple times it overwrites the previous set of stored values 3679 and does not allocated additional space. 3680 3681 .seealso: MatRetrieveValues() 3682 3683 @*/ 3684 PetscErrorCode MatStoreValues(Mat mat) 3685 { 3686 PetscErrorCode ierr; 3687 3688 PetscFunctionBegin; 3689 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3690 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3691 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3692 ierr = PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));CHKERRQ(ierr); 3693 PetscFunctionReturn(0); 3694 } 3695 3696 #undef __FUNCT__ 3697 #define __FUNCT__ "MatRetrieveValues_SeqAIJ" 3698 PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat) 3699 { 3700 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3701 PetscErrorCode ierr; 3702 PetscInt nz = aij->i[mat->rmap->n]; 3703 3704 PetscFunctionBegin; 3705 if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3706 if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first"); 3707 /* copy values over */ 3708 ierr = PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3709 PetscFunctionReturn(0); 3710 } 3711 3712 #undef __FUNCT__ 3713 #define __FUNCT__ "MatRetrieveValues" 3714 /*@ 3715 MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for 3716 example, reuse of the linear part of a Jacobian, while recomputing the 3717 nonlinear portion. 3718 3719 Collect on Mat 3720 3721 Input Parameters: 3722 . mat - the matrix (currently on AIJ matrices support this option) 3723 3724 Level: advanced 3725 3726 .seealso: MatStoreValues() 3727 3728 @*/ 3729 PetscErrorCode MatRetrieveValues(Mat mat) 3730 { 3731 PetscErrorCode ierr; 3732 3733 PetscFunctionBegin; 3734 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3735 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3736 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3737 ierr = PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));CHKERRQ(ierr); 3738 PetscFunctionReturn(0); 3739 } 3740 3741 3742 /* --------------------------------------------------------------------------------*/ 3743 #undef __FUNCT__ 3744 #define __FUNCT__ "MatCreateSeqAIJ" 3745 /*@C 3746 MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format 3747 (the default parallel PETSc format). For good matrix assembly performance 3748 the user should preallocate the matrix storage by setting the parameter nz 3749 (or the array nnz). By setting these parameters accurately, performance 3750 during matrix assembly can be increased by more than a factor of 50. 3751 3752 Collective on MPI_Comm 3753 3754 Input Parameters: 3755 + comm - MPI communicator, set to PETSC_COMM_SELF 3756 . m - number of rows 3757 . n - number of columns 3758 . nz - number of nonzeros per row (same for all rows) 3759 - nnz - array containing the number of nonzeros in the various rows 3760 (possibly different for each row) or NULL 3761 3762 Output Parameter: 3763 . A - the matrix 3764 3765 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3766 MatXXXXSetPreallocation() paradgm instead of this routine directly. 3767 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3768 3769 Notes: 3770 If nnz is given then nz is ignored 3771 3772 The AIJ format (also called the Yale sparse matrix format or 3773 compressed row storage), is fully compatible with standard Fortran 77 3774 storage. That is, the stored row and column indices can begin at 3775 either one (as in Fortran) or zero. See the users' manual for details. 3776 3777 Specify the preallocated storage with either nz or nnz (not both). 3778 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3779 allocation. For large problems you MUST preallocate memory or you 3780 will get TERRIBLE performance, see the users' manual chapter on matrices. 3781 3782 By default, this format uses inodes (identical nodes) when possible, to 3783 improve numerical efficiency of matrix-vector products and solves. We 3784 search for consecutive rows with the same nonzero structure, thereby 3785 reusing matrix information to achieve increased efficiency. 3786 3787 Options Database Keys: 3788 + -mat_no_inode - Do not use inodes 3789 - -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3790 3791 Level: intermediate 3792 3793 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays() 3794 3795 @*/ 3796 PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A) 3797 { 3798 PetscErrorCode ierr; 3799 3800 PetscFunctionBegin; 3801 ierr = MatCreate(comm,A);CHKERRQ(ierr); 3802 ierr = MatSetSizes(*A,m,n,m,n);CHKERRQ(ierr); 3803 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 3804 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);CHKERRQ(ierr); 3805 PetscFunctionReturn(0); 3806 } 3807 3808 #undef __FUNCT__ 3809 #define __FUNCT__ "MatSeqAIJSetPreallocation" 3810 /*@C 3811 MatSeqAIJSetPreallocation - For good matrix assembly performance 3812 the user should preallocate the matrix storage by setting the parameter nz 3813 (or the array nnz). By setting these parameters accurately, performance 3814 during matrix assembly can be increased by more than a factor of 50. 3815 3816 Collective on MPI_Comm 3817 3818 Input Parameters: 3819 + B - The matrix-free 3820 . nz - number of nonzeros per row (same for all rows) 3821 - nnz - array containing the number of nonzeros in the various rows 3822 (possibly different for each row) or NULL 3823 3824 Notes: 3825 If nnz is given then nz is ignored 3826 3827 The AIJ format (also called the Yale sparse matrix format or 3828 compressed row storage), is fully compatible with standard Fortran 77 3829 storage. That is, the stored row and column indices can begin at 3830 either one (as in Fortran) or zero. See the users' manual for details. 3831 3832 Specify the preallocated storage with either nz or nnz (not both). 3833 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3834 allocation. For large problems you MUST preallocate memory or you 3835 will get TERRIBLE performance, see the users' manual chapter on matrices. 3836 3837 You can call MatGetInfo() to get information on how effective the preallocation was; 3838 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3839 You can also run with the option -info and look for messages with the string 3840 malloc in them to see if additional memory allocation was needed. 3841 3842 Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix 3843 entries or columns indices 3844 3845 By default, this format uses inodes (identical nodes) when possible, to 3846 improve numerical efficiency of matrix-vector products and solves. We 3847 search for consecutive rows with the same nonzero structure, thereby 3848 reusing matrix information to achieve increased efficiency. 3849 3850 Options Database Keys: 3851 + -mat_no_inode - Do not use inodes 3852 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3853 - -mat_aij_oneindex - Internally use indexing starting at 1 3854 rather than 0. Note that when calling MatSetValues(), 3855 the user still MUST index entries starting at 0! 3856 3857 Level: intermediate 3858 3859 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo() 3860 3861 @*/ 3862 PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[]) 3863 { 3864 PetscErrorCode ierr; 3865 3866 PetscFunctionBegin; 3867 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3868 PetscValidType(B,1); 3869 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));CHKERRQ(ierr); 3870 PetscFunctionReturn(0); 3871 } 3872 3873 #undef __FUNCT__ 3874 #define __FUNCT__ "MatSeqAIJSetPreallocation_SeqAIJ" 3875 PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz) 3876 { 3877 Mat_SeqAIJ *b; 3878 PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE; 3879 PetscErrorCode ierr; 3880 PetscInt i; 3881 3882 PetscFunctionBegin; 3883 if (nz >= 0 || nnz) realalloc = PETSC_TRUE; 3884 if (nz == MAT_SKIP_ALLOCATION) { 3885 skipallocation = PETSC_TRUE; 3886 nz = 0; 3887 } 3888 3889 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3890 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3891 3892 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3893 if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %d",nz); 3894 if (nnz) { 3895 for (i=0; i<B->rmap->n; i++) { 3896 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]); 3897 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); 3898 } 3899 } 3900 3901 B->preallocated = PETSC_TRUE; 3902 3903 b = (Mat_SeqAIJ*)B->data; 3904 3905 if (!skipallocation) { 3906 if (!b->imax) { 3907 ierr = PetscMalloc2(B->rmap->n,PetscInt,&b->imax,B->rmap->n,PetscInt,&b->ilen);CHKERRQ(ierr); 3908 ierr = PetscLogObjectMemory((PetscObject)B,2*B->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 3909 } 3910 if (!nnz) { 3911 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10; 3912 else if (nz < 0) nz = 1; 3913 for (i=0; i<B->rmap->n; i++) b->imax[i] = nz; 3914 nz = nz*B->rmap->n; 3915 } else { 3916 nz = 0; 3917 for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz += nnz[i];} 3918 } 3919 /* b->ilen will count nonzeros in each row so far. */ 3920 for (i=0; i<B->rmap->n; i++) b->ilen[i] = 0; 3921 3922 /* allocate the matrix space */ 3923 ierr = MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);CHKERRQ(ierr); 3924 ierr = PetscMalloc3(nz,PetscScalar,&b->a,nz,PetscInt,&b->j,B->rmap->n+1,PetscInt,&b->i);CHKERRQ(ierr); 3925 ierr = PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));CHKERRQ(ierr); 3926 b->i[0] = 0; 3927 for (i=1; i<B->rmap->n+1; i++) { 3928 b->i[i] = b->i[i-1] + b->imax[i-1]; 3929 } 3930 b->singlemalloc = PETSC_TRUE; 3931 b->free_a = PETSC_TRUE; 3932 b->free_ij = PETSC_TRUE; 3933 #if defined(PETSC_THREADCOMM_ACTIVE) 3934 ierr = MatZeroEntries_SeqAIJ(B);CHKERRQ(ierr); 3935 #endif 3936 } else { 3937 b->free_a = PETSC_FALSE; 3938 b->free_ij = PETSC_FALSE; 3939 } 3940 3941 b->nz = 0; 3942 b->maxnz = nz; 3943 B->info.nz_unneeded = (double)b->maxnz; 3944 if (realalloc) { 3945 ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3946 } 3947 PetscFunctionReturn(0); 3948 } 3949 3950 #undef __FUNCT__ 3951 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR" 3952 /*@ 3953 MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format. 3954 3955 Input Parameters: 3956 + B - the matrix 3957 . i - the indices into j for the start of each row (starts with zero) 3958 . j - the column indices for each row (starts with zero) these must be sorted for each row 3959 - v - optional values in the matrix 3960 3961 Level: developer 3962 3963 The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays() 3964 3965 .keywords: matrix, aij, compressed row, sparse, sequential 3966 3967 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), SeqAIJ 3968 @*/ 3969 PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[]) 3970 { 3971 PetscErrorCode ierr; 3972 3973 PetscFunctionBegin; 3974 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3975 PetscValidType(B,1); 3976 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3977 PetscFunctionReturn(0); 3978 } 3979 3980 #undef __FUNCT__ 3981 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR_SeqAIJ" 3982 PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3983 { 3984 PetscInt i; 3985 PetscInt m,n; 3986 PetscInt nz; 3987 PetscInt *nnz, nz_max = 0; 3988 PetscScalar *values; 3989 PetscErrorCode ierr; 3990 3991 PetscFunctionBegin; 3992 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]); 3993 3994 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3995 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3996 3997 ierr = MatGetSize(B, &m, &n);CHKERRQ(ierr); 3998 ierr = PetscMalloc((m+1) * sizeof(PetscInt), &nnz);CHKERRQ(ierr); 3999 for (i = 0; i < m; i++) { 4000 nz = Ii[i+1]- Ii[i]; 4001 nz_max = PetscMax(nz_max, nz); 4002 if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz); 4003 nnz[i] = nz; 4004 } 4005 ierr = MatSeqAIJSetPreallocation(B, 0, nnz);CHKERRQ(ierr); 4006 ierr = PetscFree(nnz);CHKERRQ(ierr); 4007 4008 if (v) { 4009 values = (PetscScalar*) v; 4010 } else { 4011 ierr = PetscMalloc(nz_max*sizeof(PetscScalar), &values);CHKERRQ(ierr); 4012 ierr = PetscMemzero(values, nz_max*sizeof(PetscScalar));CHKERRQ(ierr); 4013 } 4014 4015 for (i = 0; i < m; i++) { 4016 nz = Ii[i+1] - Ii[i]; 4017 ierr = MatSetValues_SeqAIJ(B, 1, &i, nz, J+Ii[i], values + (v ? Ii[i] : 0), INSERT_VALUES);CHKERRQ(ierr); 4018 } 4019 4020 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4021 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4022 4023 if (!v) { 4024 ierr = PetscFree(values);CHKERRQ(ierr); 4025 } 4026 ierr = MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 4027 PetscFunctionReturn(0); 4028 } 4029 4030 #include <../src/mat/impls/dense/seq/dense.h> 4031 #include <petsc-private/kernels/petscaxpy.h> 4032 4033 #undef __FUNCT__ 4034 #define __FUNCT__ "MatMatMultNumeric_SeqDense_SeqAIJ" 4035 /* 4036 Computes (B'*A')' since computing B*A directly is untenable 4037 4038 n p p 4039 ( ) ( ) ( ) 4040 m ( A ) * n ( B ) = m ( C ) 4041 ( ) ( ) ( ) 4042 4043 */ 4044 PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C) 4045 { 4046 PetscErrorCode ierr; 4047 Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data; 4048 Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data; 4049 Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data; 4050 PetscInt i,n,m,q,p; 4051 const PetscInt *ii,*idx; 4052 const PetscScalar *b,*a,*a_q; 4053 PetscScalar *c,*c_q; 4054 4055 PetscFunctionBegin; 4056 m = A->rmap->n; 4057 n = A->cmap->n; 4058 p = B->cmap->n; 4059 a = sub_a->v; 4060 b = sub_b->a; 4061 c = sub_c->v; 4062 ierr = PetscMemzero(c,m*p*sizeof(PetscScalar));CHKERRQ(ierr); 4063 4064 ii = sub_b->i; 4065 idx = sub_b->j; 4066 for (i=0; i<n; i++) { 4067 q = ii[i+1] - ii[i]; 4068 while (q-->0) { 4069 c_q = c + m*(*idx); 4070 a_q = a + m*i; 4071 PetscKernelAXPY(c_q,*b,a_q,m); 4072 idx++; 4073 b++; 4074 } 4075 } 4076 PetscFunctionReturn(0); 4077 } 4078 4079 #undef __FUNCT__ 4080 #define __FUNCT__ "MatMatMultSymbolic_SeqDense_SeqAIJ" 4081 PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 4082 { 4083 PetscErrorCode ierr; 4084 PetscInt m=A->rmap->n,n=B->cmap->n; 4085 Mat Cmat; 4086 4087 PetscFunctionBegin; 4088 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); 4089 ierr = MatCreate(PetscObjectComm((PetscObject)A),&Cmat);CHKERRQ(ierr); 4090 ierr = MatSetSizes(Cmat,m,n,m,n);CHKERRQ(ierr); 4091 ierr = MatSetBlockSizes(Cmat,A->rmap->bs,B->cmap->bs);CHKERRQ(ierr); 4092 ierr = MatSetType(Cmat,MATSEQDENSE);CHKERRQ(ierr); 4093 ierr = MatSeqDenseSetPreallocation(Cmat,NULL);CHKERRQ(ierr); 4094 4095 Cmat->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ; 4096 4097 *C = Cmat; 4098 PetscFunctionReturn(0); 4099 } 4100 4101 /* ----------------------------------------------------------------*/ 4102 #undef __FUNCT__ 4103 #define __FUNCT__ "MatMatMult_SeqDense_SeqAIJ" 4104 PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 4105 { 4106 PetscErrorCode ierr; 4107 4108 PetscFunctionBegin; 4109 if (scall == MAT_INITIAL_MATRIX) { 4110 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 4111 ierr = MatMatMultSymbolic_SeqDense_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 4112 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 4113 } 4114 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 4115 ierr = MatMatMultNumeric_SeqDense_SeqAIJ(A,B,*C);CHKERRQ(ierr); 4116 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 4117 PetscFunctionReturn(0); 4118 } 4119 4120 4121 /*MC 4122 MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices, 4123 based on compressed sparse row format. 4124 4125 Options Database Keys: 4126 . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions() 4127 4128 Level: beginner 4129 4130 .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType 4131 M*/ 4132 4133 /*MC 4134 MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices. 4135 4136 This matrix type is identical to MATSEQAIJ when constructed with a single process communicator, 4137 and MATMPIAIJ otherwise. As a result, for single process communicators, 4138 MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation is supported 4139 for communicators controlling multiple processes. It is recommended that you call both of 4140 the above preallocation routines for simplicity. 4141 4142 Options Database Keys: 4143 . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions() 4144 4145 Developer Notes: Subclasses include MATAIJCUSP, MATAIJPERM, MATAIJCRL, and also automatically switches over to use inodes when 4146 enough exist. 4147 4148 Level: beginner 4149 4150 .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ 4151 M*/ 4152 4153 /*MC 4154 MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices. 4155 4156 This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator, 4157 and MATMPIAIJCRL otherwise. As a result, for single process communicators, 4158 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 4159 for communicators controlling multiple processes. It is recommended that you call both of 4160 the above preallocation routines for simplicity. 4161 4162 Options Database Keys: 4163 . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions() 4164 4165 Level: beginner 4166 4167 .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL 4168 M*/ 4169 4170 #if defined(PETSC_HAVE_PASTIX) 4171 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_pastix(Mat,MatFactorType,Mat*); 4172 #endif 4173 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4174 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_essl(Mat,MatFactorType,Mat*); 4175 #endif 4176 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*); 4177 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_petsc(Mat,MatFactorType,Mat*); 4178 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_bas(Mat,MatFactorType,Mat*); 4179 extern PetscErrorCode MatGetFactorAvailable_seqaij_petsc(Mat,MatFactorType,PetscBool*); 4180 #if defined(PETSC_HAVE_MUMPS) 4181 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mumps(Mat,MatFactorType,Mat*); 4182 #endif 4183 #if defined(PETSC_HAVE_SUPERLU) 4184 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat,MatFactorType,Mat*); 4185 #endif 4186 #if defined(PETSC_HAVE_SUPERLU_DIST) 4187 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu_dist(Mat,MatFactorType,Mat*); 4188 #endif 4189 #if defined(PETSC_HAVE_UMFPACK) 4190 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_umfpack(Mat,MatFactorType,Mat*); 4191 #endif 4192 #if defined(PETSC_HAVE_CHOLMOD) 4193 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat,MatFactorType,Mat*); 4194 #endif 4195 #if defined(PETSC_HAVE_LUSOL) 4196 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_lusol(Mat,MatFactorType,Mat*); 4197 #endif 4198 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4199 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_matlab(Mat,MatFactorType,Mat*); 4200 extern PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*); 4201 extern PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*); 4202 #endif 4203 #if defined(PETSC_HAVE_CLIQUE) 4204 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_clique(Mat,MatFactorType,Mat*); 4205 #endif 4206 4207 4208 #undef __FUNCT__ 4209 #define __FUNCT__ "MatSeqAIJGetArray" 4210 /*@C 4211 MatSeqAIJGetArray - gives access to the array where the data for a SeqSeqAIJ matrix is stored 4212 4213 Not Collective 4214 4215 Input Parameter: 4216 . mat - a MATSEQDENSE matrix 4217 4218 Output Parameter: 4219 . array - pointer to the data 4220 4221 Level: intermediate 4222 4223 .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90() 4224 @*/ 4225 PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array) 4226 { 4227 PetscErrorCode ierr; 4228 4229 PetscFunctionBegin; 4230 ierr = PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4231 PetscFunctionReturn(0); 4232 } 4233 4234 #undef __FUNCT__ 4235 #define __FUNCT__ "MatSeqAIJRestoreArray" 4236 /*@C 4237 MatSeqAIJRestoreArray - returns access to the array where the data for a SeqSeqAIJ matrix is stored obtained by MatSeqAIJGetArray() 4238 4239 Not Collective 4240 4241 Input Parameters: 4242 . mat - a MATSEQDENSE matrix 4243 . array - pointer to the data 4244 4245 Level: intermediate 4246 4247 .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90() 4248 @*/ 4249 PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array) 4250 { 4251 PetscErrorCode ierr; 4252 4253 PetscFunctionBegin; 4254 ierr = PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4255 PetscFunctionReturn(0); 4256 } 4257 4258 #undef __FUNCT__ 4259 #define __FUNCT__ "MatCreate_SeqAIJ" 4260 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B) 4261 { 4262 Mat_SeqAIJ *b; 4263 PetscErrorCode ierr; 4264 PetscMPIInt size; 4265 4266 PetscFunctionBegin; 4267 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);CHKERRQ(ierr); 4268 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1"); 4269 4270 ierr = PetscNewLog(B,Mat_SeqAIJ,&b);CHKERRQ(ierr); 4271 4272 B->data = (void*)b; 4273 4274 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 4275 4276 b->row = 0; 4277 b->col = 0; 4278 b->icol = 0; 4279 b->reallocs = 0; 4280 b->ignorezeroentries = PETSC_FALSE; 4281 b->roworiented = PETSC_TRUE; 4282 b->nonew = 0; 4283 b->diag = 0; 4284 b->solve_work = 0; 4285 B->spptr = 0; 4286 b->saved_values = 0; 4287 b->idiag = 0; 4288 b->mdiag = 0; 4289 b->ssor_work = 0; 4290 b->omega = 1.0; 4291 b->fshift = 0.0; 4292 b->idiagvalid = PETSC_FALSE; 4293 b->ibdiagvalid = PETSC_FALSE; 4294 b->keepnonzeropattern = PETSC_FALSE; 4295 b->xtoy = 0; 4296 b->XtoY = 0; 4297 B->same_nonzero = PETSC_FALSE; 4298 4299 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4300 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);CHKERRQ(ierr); 4301 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);CHKERRQ(ierr); 4302 4303 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4304 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_matlab_C",MatGetFactor_seqaij_matlab);CHKERRQ(ierr); 4305 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);CHKERRQ(ierr); 4306 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);CHKERRQ(ierr); 4307 #endif 4308 #if defined(PETSC_HAVE_PASTIX) 4309 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_pastix_C",MatGetFactor_seqaij_pastix);CHKERRQ(ierr); 4310 #endif 4311 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4312 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_essl_C",MatGetFactor_seqaij_essl);CHKERRQ(ierr); 4313 #endif 4314 #if defined(PETSC_HAVE_SUPERLU) 4315 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_C",MatGetFactor_seqaij_superlu);CHKERRQ(ierr); 4316 #endif 4317 #if defined(PETSC_HAVE_SUPERLU_DIST) 4318 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_dist_C",MatGetFactor_seqaij_superlu_dist);CHKERRQ(ierr); 4319 #endif 4320 #if defined(PETSC_HAVE_MUMPS) 4321 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mumps_C",MatGetFactor_aij_mumps);CHKERRQ(ierr); 4322 #endif 4323 #if defined(PETSC_HAVE_UMFPACK) 4324 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_umfpack_C",MatGetFactor_seqaij_umfpack);CHKERRQ(ierr); 4325 #endif 4326 #if defined(PETSC_HAVE_CHOLMOD) 4327 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_cholmod_C",MatGetFactor_seqaij_cholmod);CHKERRQ(ierr); 4328 #endif 4329 #if defined(PETSC_HAVE_LUSOL) 4330 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_lusol_C",MatGetFactor_seqaij_lusol);CHKERRQ(ierr); 4331 #endif 4332 #if defined(PETSC_HAVE_CLIQUE) 4333 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_clique_C",MatGetFactor_aij_clique);CHKERRQ(ierr); 4334 #endif 4335 4336 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_petsc_C",MatGetFactor_seqaij_petsc);CHKERRQ(ierr); 4337 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactorAvailable_petsc_C",MatGetFactorAvailable_seqaij_petsc);CHKERRQ(ierr); 4338 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_bas_C",MatGetFactor_seqaij_bas);CHKERRQ(ierr); 4339 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);CHKERRQ(ierr); 4340 ierr = PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);CHKERRQ(ierr); 4341 ierr = PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);CHKERRQ(ierr); 4342 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);CHKERRQ(ierr); 4343 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);CHKERRQ(ierr); 4344 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);CHKERRQ(ierr); 4345 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);CHKERRQ(ierr); 4346 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4347 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4348 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);CHKERRQ(ierr); 4349 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);CHKERRQ(ierr); 4350 ierr = PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);CHKERRQ(ierr); 4351 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMult_seqdense_seqaij_C",MatMatMult_SeqDense_SeqAIJ);CHKERRQ(ierr); 4352 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_seqdense_seqaij_C",MatMatMultSymbolic_SeqDense_SeqAIJ);CHKERRQ(ierr); 4353 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_seqdense_seqaij_C",MatMatMultNumeric_SeqDense_SeqAIJ);CHKERRQ(ierr); 4354 ierr = MatCreate_SeqAIJ_Inode(B);CHKERRQ(ierr); 4355 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4356 PetscFunctionReturn(0); 4357 } 4358 4359 #undef __FUNCT__ 4360 #define __FUNCT__ "MatDuplicateNoCreate_SeqAIJ" 4361 /* 4362 Given a matrix generated with MatGetFactor() duplicates all the information in A into B 4363 */ 4364 PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace) 4365 { 4366 Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data; 4367 PetscErrorCode ierr; 4368 PetscInt i,m = A->rmap->n; 4369 4370 PetscFunctionBegin; 4371 c = (Mat_SeqAIJ*)C->data; 4372 4373 C->factortype = A->factortype; 4374 c->row = 0; 4375 c->col = 0; 4376 c->icol = 0; 4377 c->reallocs = 0; 4378 4379 C->assembled = PETSC_TRUE; 4380 4381 ierr = PetscLayoutReference(A->rmap,&C->rmap);CHKERRQ(ierr); 4382 ierr = PetscLayoutReference(A->cmap,&C->cmap);CHKERRQ(ierr); 4383 4384 ierr = PetscMalloc2(m,PetscInt,&c->imax,m,PetscInt,&c->ilen);CHKERRQ(ierr); 4385 ierr = PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));CHKERRQ(ierr); 4386 for (i=0; i<m; i++) { 4387 c->imax[i] = a->imax[i]; 4388 c->ilen[i] = a->ilen[i]; 4389 } 4390 4391 /* allocate the matrix space */ 4392 if (mallocmatspace) { 4393 ierr = PetscMalloc3(a->i[m],PetscScalar,&c->a,a->i[m],PetscInt,&c->j,m+1,PetscInt,&c->i);CHKERRQ(ierr); 4394 ierr = PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4395 4396 c->singlemalloc = PETSC_TRUE; 4397 4398 ierr = PetscMemcpy(c->i,a->i,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4399 if (m > 0) { 4400 ierr = PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(PetscInt));CHKERRQ(ierr); 4401 if (cpvalues == MAT_COPY_VALUES) { 4402 ierr = PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4403 } else { 4404 ierr = PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4405 } 4406 } 4407 } 4408 4409 c->ignorezeroentries = a->ignorezeroentries; 4410 c->roworiented = a->roworiented; 4411 c->nonew = a->nonew; 4412 if (a->diag) { 4413 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&c->diag);CHKERRQ(ierr); 4414 ierr = PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4415 for (i=0; i<m; i++) { 4416 c->diag[i] = a->diag[i]; 4417 } 4418 } else c->diag = 0; 4419 4420 c->solve_work = 0; 4421 c->saved_values = 0; 4422 c->idiag = 0; 4423 c->ssor_work = 0; 4424 c->keepnonzeropattern = a->keepnonzeropattern; 4425 c->free_a = PETSC_TRUE; 4426 c->free_ij = PETSC_TRUE; 4427 c->xtoy = 0; 4428 c->XtoY = 0; 4429 4430 c->rmax = a->rmax; 4431 c->nz = a->nz; 4432 c->maxnz = a->nz; /* Since we allocate exactly the right amount */ 4433 C->preallocated = PETSC_TRUE; 4434 4435 c->compressedrow.use = a->compressedrow.use; 4436 c->compressedrow.nrows = a->compressedrow.nrows; 4437 c->compressedrow.check = a->compressedrow.check; 4438 if (a->compressedrow.use) { 4439 i = a->compressedrow.nrows; 4440 ierr = PetscMalloc2(i+1,PetscInt,&c->compressedrow.i,i,PetscInt,&c->compressedrow.rindex);CHKERRQ(ierr); 4441 ierr = PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));CHKERRQ(ierr); 4442 ierr = PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));CHKERRQ(ierr); 4443 } else { 4444 c->compressedrow.use = PETSC_FALSE; 4445 c->compressedrow.i = NULL; 4446 c->compressedrow.rindex = NULL; 4447 } 4448 C->same_nonzero = A->same_nonzero; 4449 4450 ierr = MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);CHKERRQ(ierr); 4451 ierr = PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);CHKERRQ(ierr); 4452 PetscFunctionReturn(0); 4453 } 4454 4455 #undef __FUNCT__ 4456 #define __FUNCT__ "MatDuplicate_SeqAIJ" 4457 PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B) 4458 { 4459 PetscErrorCode ierr; 4460 4461 PetscFunctionBegin; 4462 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 4463 ierr = MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);CHKERRQ(ierr); 4464 ierr = MatSetBlockSizes(*B,A->rmap->bs,A->cmap->bs);CHKERRQ(ierr); 4465 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 4466 ierr = MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);CHKERRQ(ierr); 4467 PetscFunctionReturn(0); 4468 } 4469 4470 #undef __FUNCT__ 4471 #define __FUNCT__ "MatLoad_SeqAIJ" 4472 PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer) 4473 { 4474 Mat_SeqAIJ *a; 4475 PetscErrorCode ierr; 4476 PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N,rows,cols; 4477 int fd; 4478 PetscMPIInt size; 4479 MPI_Comm comm; 4480 PetscInt bs = 1; 4481 4482 PetscFunctionBegin; 4483 ierr = PetscObjectGetComm((PetscObject)viewer,&comm);CHKERRQ(ierr); 4484 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4485 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"view must have one processor"); 4486 4487 ierr = PetscOptionsBegin(comm,NULL,"Options for loading SEQAIJ matrix","Mat");CHKERRQ(ierr); 4488 ierr = PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);CHKERRQ(ierr); 4489 ierr = PetscOptionsEnd();CHKERRQ(ierr); 4490 if (bs > 1) {ierr = MatSetBlockSize(newMat,bs);CHKERRQ(ierr);} 4491 4492 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 4493 ierr = PetscBinaryRead(fd,header,4,PETSC_INT);CHKERRQ(ierr); 4494 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file"); 4495 M = header[1]; N = header[2]; nz = header[3]; 4496 4497 if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ"); 4498 4499 /* read in row lengths */ 4500 ierr = PetscMalloc(M*sizeof(PetscInt),&rowlengths);CHKERRQ(ierr); 4501 ierr = PetscBinaryRead(fd,rowlengths,M,PETSC_INT);CHKERRQ(ierr); 4502 4503 /* check if sum of rowlengths is same as nz */ 4504 for (i=0,sum=0; i< M; i++) sum +=rowlengths[i]; 4505 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); 4506 4507 /* set global size if not set already*/ 4508 if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) { 4509 ierr = MatSetSizes(newMat,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr); 4510 } else { 4511 /* if sizes and type are already set, check if the vector global sizes are correct */ 4512 ierr = MatGetSize(newMat,&rows,&cols);CHKERRQ(ierr); 4513 if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */ 4514 ierr = MatGetLocalSize(newMat,&rows,&cols);CHKERRQ(ierr); 4515 } 4516 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); 4517 } 4518 ierr = MatSeqAIJSetPreallocation_SeqAIJ(newMat,0,rowlengths);CHKERRQ(ierr); 4519 a = (Mat_SeqAIJ*)newMat->data; 4520 4521 ierr = PetscBinaryRead(fd,a->j,nz,PETSC_INT);CHKERRQ(ierr); 4522 4523 /* read in nonzero values */ 4524 ierr = PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);CHKERRQ(ierr); 4525 4526 /* set matrix "i" values */ 4527 a->i[0] = 0; 4528 for (i=1; i<= M; i++) { 4529 a->i[i] = a->i[i-1] + rowlengths[i-1]; 4530 a->ilen[i-1] = rowlengths[i-1]; 4531 } 4532 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 4533 4534 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4535 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4536 PetscFunctionReturn(0); 4537 } 4538 4539 #undef __FUNCT__ 4540 #define __FUNCT__ "MatEqual_SeqAIJ" 4541 PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg) 4542 { 4543 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data; 4544 PetscErrorCode ierr; 4545 #if defined(PETSC_USE_COMPLEX) 4546 PetscInt k; 4547 #endif 4548 4549 PetscFunctionBegin; 4550 /* If the matrix dimensions are not equal,or no of nonzeros */ 4551 if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) { 4552 *flg = PETSC_FALSE; 4553 PetscFunctionReturn(0); 4554 } 4555 4556 /* if the a->i are the same */ 4557 ierr = PetscMemcmp(a->i,b->i,(A->rmap->n+1)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4558 if (!*flg) PetscFunctionReturn(0); 4559 4560 /* if a->j are the same */ 4561 ierr = PetscMemcmp(a->j,b->j,(a->nz)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4562 if (!*flg) PetscFunctionReturn(0); 4563 4564 /* if a->a are the same */ 4565 #if defined(PETSC_USE_COMPLEX) 4566 for (k=0; k<a->nz; k++) { 4567 if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) { 4568 *flg = PETSC_FALSE; 4569 PetscFunctionReturn(0); 4570 } 4571 } 4572 #else 4573 ierr = PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);CHKERRQ(ierr); 4574 #endif 4575 PetscFunctionReturn(0); 4576 } 4577 4578 #undef __FUNCT__ 4579 #define __FUNCT__ "MatCreateSeqAIJWithArrays" 4580 /*@ 4581 MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format) 4582 provided by the user. 4583 4584 Collective on MPI_Comm 4585 4586 Input Parameters: 4587 + comm - must be an MPI communicator of size 1 4588 . m - number of rows 4589 . n - number of columns 4590 . i - row indices 4591 . j - column indices 4592 - a - matrix values 4593 4594 Output Parameter: 4595 . mat - the matrix 4596 4597 Level: intermediate 4598 4599 Notes: 4600 The i, j, and a arrays are not copied by this routine, the user must free these arrays 4601 once the matrix is destroyed and not before 4602 4603 You cannot set new nonzero locations into this matrix, that will generate an error. 4604 4605 The i and j indices are 0 based 4606 4607 The format which is used for the sparse matrix input, is equivalent to a 4608 row-major ordering.. i.e for the following matrix, the input data expected is 4609 as shown: 4610 4611 1 0 0 4612 2 0 3 4613 4 5 6 4614 4615 i = {0,1,3,6} [size = nrow+1 = 3+1] 4616 j = {0,0,2,0,1,2} [size = nz = 6]; values must be sorted for each row 4617 v = {1,2,3,4,5,6} [size = nz = 6] 4618 4619 4620 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4621 4622 @*/ 4623 PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat) 4624 { 4625 PetscErrorCode ierr; 4626 PetscInt ii; 4627 Mat_SeqAIJ *aij; 4628 #if defined(PETSC_USE_DEBUG) 4629 PetscInt jj; 4630 #endif 4631 4632 PetscFunctionBegin; 4633 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 4634 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4635 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4636 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4637 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4638 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);CHKERRQ(ierr); 4639 aij = (Mat_SeqAIJ*)(*mat)->data; 4640 ierr = PetscMalloc2(m,PetscInt,&aij->imax,m,PetscInt,&aij->ilen);CHKERRQ(ierr); 4641 4642 aij->i = i; 4643 aij->j = j; 4644 aij->a = a; 4645 aij->singlemalloc = PETSC_FALSE; 4646 aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 4647 aij->free_a = PETSC_FALSE; 4648 aij->free_ij = PETSC_FALSE; 4649 4650 for (ii=0; ii<m; ii++) { 4651 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 4652 #if defined(PETSC_USE_DEBUG) 4653 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]); 4654 for (jj=i[ii]+1; jj<i[ii+1]; jj++) { 4655 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); 4656 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); 4657 } 4658 #endif 4659 } 4660 #if defined(PETSC_USE_DEBUG) 4661 for (ii=0; ii<aij->i[m]; ii++) { 4662 if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %d index = %d",ii,j[ii]); 4663 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]); 4664 } 4665 #endif 4666 4667 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4668 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4669 PetscFunctionReturn(0); 4670 } 4671 #undef __FUNCT__ 4672 #define __FUNCT__ "MatCreateSeqAIJFromTriple" 4673 /*@C 4674 MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format) 4675 provided by the user. 4676 4677 Collective on MPI_Comm 4678 4679 Input Parameters: 4680 + comm - must be an MPI communicator of size 1 4681 . m - number of rows 4682 . n - number of columns 4683 . i - row indices 4684 . j - column indices 4685 . a - matrix values 4686 . nz - number of nonzeros 4687 - idx - 0 or 1 based 4688 4689 Output Parameter: 4690 . mat - the matrix 4691 4692 Level: intermediate 4693 4694 Notes: 4695 The i and j indices are 0 based 4696 4697 The format which is used for the sparse matrix input, is equivalent to a 4698 row-major ordering.. i.e for the following matrix, the input data expected is 4699 as shown: 4700 4701 1 0 0 4702 2 0 3 4703 4 5 6 4704 4705 i = {0,1,1,2,2,2} 4706 j = {0,0,2,0,1,2} 4707 v = {1,2,3,4,5,6} 4708 4709 4710 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4711 4712 @*/ 4713 PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat,PetscInt nz,PetscBool idx) 4714 { 4715 PetscErrorCode ierr; 4716 PetscInt ii, *nnz, one = 1,row,col; 4717 4718 4719 PetscFunctionBegin; 4720 ierr = PetscMalloc(m*sizeof(PetscInt),&nnz);CHKERRQ(ierr); 4721 ierr = PetscMemzero(nnz,m*sizeof(PetscInt));CHKERRQ(ierr); 4722 for (ii = 0; ii < nz; ii++) { 4723 nnz[i[ii]] += 1; 4724 } 4725 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4726 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4727 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4728 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4729 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);CHKERRQ(ierr); 4730 for (ii = 0; ii < nz; ii++) { 4731 if (idx) { 4732 row = i[ii] - 1; 4733 col = j[ii] - 1; 4734 } else { 4735 row = i[ii]; 4736 col = j[ii]; 4737 } 4738 ierr = MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);CHKERRQ(ierr); 4739 } 4740 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4741 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4742 ierr = PetscFree(nnz);CHKERRQ(ierr); 4743 PetscFunctionReturn(0); 4744 } 4745 4746 #undef __FUNCT__ 4747 #define __FUNCT__ "MatSetColoring_SeqAIJ" 4748 PetscErrorCode MatSetColoring_SeqAIJ(Mat A,ISColoring coloring) 4749 { 4750 PetscErrorCode ierr; 4751 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4752 4753 PetscFunctionBegin; 4754 if (coloring->ctype == IS_COLORING_GLOBAL) { 4755 ierr = ISColoringReference(coloring);CHKERRQ(ierr); 4756 a->coloring = coloring; 4757 } else if (coloring->ctype == IS_COLORING_GHOSTED) { 4758 PetscInt i,*larray; 4759 ISColoring ocoloring; 4760 ISColoringValue *colors; 4761 4762 /* set coloring for diagonal portion */ 4763 ierr = PetscMalloc(A->cmap->n*sizeof(PetscInt),&larray);CHKERRQ(ierr); 4764 for (i=0; i<A->cmap->n; i++) larray[i] = i; 4765 ierr = ISGlobalToLocalMappingApply(A->cmap->mapping,IS_GTOLM_MASK,A->cmap->n,larray,NULL,larray);CHKERRQ(ierr); 4766 ierr = PetscMalloc(A->cmap->n*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 4767 for (i=0; i<A->cmap->n; i++) colors[i] = coloring->colors[larray[i]]; 4768 ierr = PetscFree(larray);CHKERRQ(ierr); 4769 ierr = ISColoringCreate(PETSC_COMM_SELF,coloring->n,A->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4770 a->coloring = ocoloring; 4771 } 4772 PetscFunctionReturn(0); 4773 } 4774 4775 #undef __FUNCT__ 4776 #define __FUNCT__ "MatSetValuesAdifor_SeqAIJ" 4777 PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat A,PetscInt nl,void *advalues) 4778 { 4779 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4780 PetscInt m = A->rmap->n,*ii = a->i,*jj = a->j,nz,i,j; 4781 MatScalar *v = a->a; 4782 PetscScalar *values = (PetscScalar*)advalues; 4783 ISColoringValue *color; 4784 4785 PetscFunctionBegin; 4786 if (!a->coloring) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix"); 4787 color = a->coloring->colors; 4788 /* loop over rows */ 4789 for (i=0; i<m; i++) { 4790 nz = ii[i+1] - ii[i]; 4791 /* loop over columns putting computed value into matrix */ 4792 for (j=0; j<nz; j++) *v++ = values[color[*jj++]]; 4793 values += nl; /* jump to next row of derivatives */ 4794 } 4795 PetscFunctionReturn(0); 4796 } 4797 4798 #undef __FUNCT__ 4799 #define __FUNCT__ "MatSeqAIJInvalidateDiagonal" 4800 PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A) 4801 { 4802 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 4803 PetscErrorCode ierr; 4804 4805 PetscFunctionBegin; 4806 a->idiagvalid = PETSC_FALSE; 4807 a->ibdiagvalid = PETSC_FALSE; 4808 4809 ierr = MatSeqAIJInvalidateDiagonal_Inode(A);CHKERRQ(ierr); 4810 PetscFunctionReturn(0); 4811 } 4812 4813 /* 4814 Special version for direct calls from Fortran 4815 */ 4816 #include <petsc-private/fortranimpl.h> 4817 #if defined(PETSC_HAVE_FORTRAN_CAPS) 4818 #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ 4819 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 4820 #define matsetvaluesseqaij_ matsetvaluesseqaij 4821 #endif 4822 4823 /* Change these macros so can be used in void function */ 4824 #undef CHKERRQ 4825 #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr) 4826 #undef SETERRQ2 4827 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 4828 #undef SETERRQ3 4829 #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr) 4830 4831 #undef __FUNCT__ 4832 #define __FUNCT__ "matsetvaluesseqaij_" 4833 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) 4834 { 4835 Mat A = *AA; 4836 PetscInt m = *mm, n = *nn; 4837 InsertMode is = *isis; 4838 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4839 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 4840 PetscInt *imax,*ai,*ailen; 4841 PetscErrorCode ierr; 4842 PetscInt *aj,nonew = a->nonew,lastcol = -1; 4843 MatScalar *ap,value,*aa; 4844 PetscBool ignorezeroentries = a->ignorezeroentries; 4845 PetscBool roworiented = a->roworiented; 4846 4847 PetscFunctionBegin; 4848 MatCheckPreallocated(A,1); 4849 imax = a->imax; 4850 ai = a->i; 4851 ailen = a->ilen; 4852 aj = a->j; 4853 aa = a->a; 4854 4855 for (k=0; k<m; k++) { /* loop over added rows */ 4856 row = im[k]; 4857 if (row < 0) continue; 4858 #if defined(PETSC_USE_DEBUG) 4859 if (row >= A->rmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large"); 4860 #endif 4861 rp = aj + ai[row]; ap = aa + ai[row]; 4862 rmax = imax[row]; nrow = ailen[row]; 4863 low = 0; 4864 high = nrow; 4865 for (l=0; l<n; l++) { /* loop over added columns */ 4866 if (in[l] < 0) continue; 4867 #if defined(PETSC_USE_DEBUG) 4868 if (in[l] >= A->cmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large"); 4869 #endif 4870 col = in[l]; 4871 if (roworiented) value = v[l + k*n]; 4872 else value = v[k + l*m]; 4873 4874 if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue; 4875 4876 if (col <= lastcol) low = 0; 4877 else high = nrow; 4878 lastcol = col; 4879 while (high-low > 5) { 4880 t = (low+high)/2; 4881 if (rp[t] > col) high = t; 4882 else low = t; 4883 } 4884 for (i=low; i<high; i++) { 4885 if (rp[i] > col) break; 4886 if (rp[i] == col) { 4887 if (is == ADD_VALUES) ap[i] += value; 4888 else ap[i] = value; 4889 goto noinsert; 4890 } 4891 } 4892 if (value == 0.0 && ignorezeroentries) goto noinsert; 4893 if (nonew == 1) goto noinsert; 4894 if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); 4895 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 4896 N = nrow++ - 1; a->nz++; high++; 4897 /* shift up all the later entries in this row */ 4898 for (ii=N; ii>=i; ii--) { 4899 rp[ii+1] = rp[ii]; 4900 ap[ii+1] = ap[ii]; 4901 } 4902 rp[i] = col; 4903 ap[i] = value; 4904 noinsert:; 4905 low = i + 1; 4906 } 4907 ailen[row] = nrow; 4908 } 4909 A->same_nonzero = PETSC_FALSE; 4910 PetscFunctionReturnVoid(); 4911 } 4912 4913 4914