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]; /* omega in idiag */ 1772 } 1773 } 1774 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); /* assumes 1/2 in upper */ 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 /* lower */ 1782 n = diag[i] - a->i[i]; 1783 idx = a->j + a->i[i]; 1784 v = a->a + a->i[i]; 1785 sum = b[i]; 1786 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1787 t[i] = sum; /* save application of the lower-triangular part */ 1788 /* upper */ 1789 n = a->i[i+1] - diag[i] - 1; 1790 idx = a->j + diag[i] + 1; 1791 v = a->a + diag[i] + 1; 1792 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1793 x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 1794 } 1795 xb = t; 1796 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1797 } else xb = b; 1798 if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) { 1799 for (i=m-1; i>=0; i--) { 1800 sum = xb[i]; 1801 if (xb == b) { 1802 /* whole matrix (no checkpointing available) */ 1803 n = a->i[i+1] - a->i[i]; 1804 idx = a->j + a->i[i]; 1805 v = a->a + a->i[i]; 1806 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1807 x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i]; 1808 } else { /* lower-triangular part has been saved, so only apply upper-triangular */ 1809 n = a->i[i+1] - diag[i] - 1; 1810 idx = a->j + diag[i] + 1; 1811 v = a->a + diag[i] + 1; 1812 PetscSparseDenseMinusDot(sum,x,v,idx,n); 1813 x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */ 1814 } 1815 } 1816 if (xb == b) { 1817 ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr); 1818 } else { 1819 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); /* assumes 1/2 in upper */ 1820 } 1821 } 1822 } 1823 ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr); 1824 ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr); 1825 PetscFunctionReturn(0); 1826 } 1827 1828 1829 #undef __FUNCT__ 1830 #define __FUNCT__ "MatGetInfo_SeqAIJ" 1831 PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info) 1832 { 1833 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1834 1835 PetscFunctionBegin; 1836 info->block_size = 1.0; 1837 info->nz_allocated = (double)a->maxnz; 1838 info->nz_used = (double)a->nz; 1839 info->nz_unneeded = (double)(a->maxnz - a->nz); 1840 info->assemblies = (double)A->num_ass; 1841 info->mallocs = (double)A->info.mallocs; 1842 info->memory = ((PetscObject)A)->mem; 1843 if (A->factortype) { 1844 info->fill_ratio_given = A->info.fill_ratio_given; 1845 info->fill_ratio_needed = A->info.fill_ratio_needed; 1846 info->factor_mallocs = A->info.factor_mallocs; 1847 } else { 1848 info->fill_ratio_given = 0; 1849 info->fill_ratio_needed = 0; 1850 info->factor_mallocs = 0; 1851 } 1852 PetscFunctionReturn(0); 1853 } 1854 1855 #undef __FUNCT__ 1856 #define __FUNCT__ "MatZeroRows_SeqAIJ" 1857 PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 1858 { 1859 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1860 PetscInt i,m = A->rmap->n - 1,d = 0; 1861 PetscErrorCode ierr; 1862 const PetscScalar *xx; 1863 PetscScalar *bb; 1864 PetscBool missing; 1865 1866 PetscFunctionBegin; 1867 if (x && b) { 1868 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 1869 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 1870 for (i=0; i<N; i++) { 1871 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1872 bb[rows[i]] = diag*xx[rows[i]]; 1873 } 1874 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 1875 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 1876 } 1877 1878 if (a->keepnonzeropattern) { 1879 for (i=0; i<N; i++) { 1880 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1881 ierr = PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));CHKERRQ(ierr); 1882 } 1883 if (diag != 0.0) { 1884 ierr = MatMissingDiagonal_SeqAIJ(A,&missing,&d);CHKERRQ(ierr); 1885 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d); 1886 for (i=0; i<N; i++) { 1887 a->a[a->diag[rows[i]]] = diag; 1888 } 1889 } 1890 A->same_nonzero = PETSC_TRUE; 1891 } else { 1892 if (diag != 0.0) { 1893 for (i=0; i<N; i++) { 1894 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1895 if (a->ilen[rows[i]] > 0) { 1896 a->ilen[rows[i]] = 1; 1897 a->a[a->i[rows[i]]] = diag; 1898 a->j[a->i[rows[i]]] = rows[i]; 1899 } else { /* in case row was completely empty */ 1900 ierr = MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);CHKERRQ(ierr); 1901 } 1902 } 1903 } else { 1904 for (i=0; i<N; i++) { 1905 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1906 a->ilen[rows[i]] = 0; 1907 } 1908 } 1909 A->same_nonzero = PETSC_FALSE; 1910 } 1911 ierr = MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1912 PetscFunctionReturn(0); 1913 } 1914 1915 #undef __FUNCT__ 1916 #define __FUNCT__ "MatZeroRowsColumns_SeqAIJ" 1917 PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b) 1918 { 1919 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1920 PetscInt i,j,m = A->rmap->n - 1,d = 0; 1921 PetscErrorCode ierr; 1922 PetscBool missing,*zeroed,vecs = PETSC_FALSE; 1923 const PetscScalar *xx; 1924 PetscScalar *bb; 1925 1926 PetscFunctionBegin; 1927 if (x && b) { 1928 ierr = VecGetArrayRead(x,&xx);CHKERRQ(ierr); 1929 ierr = VecGetArray(b,&bb);CHKERRQ(ierr); 1930 vecs = PETSC_TRUE; 1931 } 1932 ierr = PetscMalloc(A->rmap->n*sizeof(PetscBool),&zeroed);CHKERRQ(ierr); 1933 ierr = PetscMemzero(zeroed,A->rmap->n*sizeof(PetscBool));CHKERRQ(ierr); 1934 for (i=0; i<N; i++) { 1935 if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]); 1936 ierr = PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));CHKERRQ(ierr); 1937 1938 zeroed[rows[i]] = PETSC_TRUE; 1939 } 1940 for (i=0; i<A->rmap->n; i++) { 1941 if (!zeroed[i]) { 1942 for (j=a->i[i]; j<a->i[i+1]; j++) { 1943 if (zeroed[a->j[j]]) { 1944 if (vecs) bb[i] -= a->a[j]*xx[a->j[j]]; 1945 a->a[j] = 0.0; 1946 } 1947 } 1948 } else if (vecs) bb[i] = diag*xx[i]; 1949 } 1950 if (x && b) { 1951 ierr = VecRestoreArrayRead(x,&xx);CHKERRQ(ierr); 1952 ierr = VecRestoreArray(b,&bb);CHKERRQ(ierr); 1953 } 1954 ierr = PetscFree(zeroed);CHKERRQ(ierr); 1955 if (diag != 0.0) { 1956 ierr = MatMissingDiagonal_SeqAIJ(A,&missing,&d);CHKERRQ(ierr); 1957 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d); 1958 for (i=0; i<N; i++) { 1959 a->a[a->diag[rows[i]]] = diag; 1960 } 1961 } 1962 A->same_nonzero = PETSC_TRUE; 1963 ierr = MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 1964 PetscFunctionReturn(0); 1965 } 1966 1967 #undef __FUNCT__ 1968 #define __FUNCT__ "MatGetRow_SeqAIJ" 1969 PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1970 { 1971 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 1972 PetscInt *itmp; 1973 1974 PetscFunctionBegin; 1975 if (row < 0 || row >= A->rmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row); 1976 1977 *nz = a->i[row+1] - a->i[row]; 1978 if (v) *v = a->a + a->i[row]; 1979 if (idx) { 1980 itmp = a->j + a->i[row]; 1981 if (*nz) *idx = itmp; 1982 else *idx = 0; 1983 } 1984 PetscFunctionReturn(0); 1985 } 1986 1987 /* remove this function? */ 1988 #undef __FUNCT__ 1989 #define __FUNCT__ "MatRestoreRow_SeqAIJ" 1990 PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v) 1991 { 1992 PetscFunctionBegin; 1993 PetscFunctionReturn(0); 1994 } 1995 1996 #undef __FUNCT__ 1997 #define __FUNCT__ "MatNorm_SeqAIJ" 1998 PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm) 1999 { 2000 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2001 MatScalar *v = a->a; 2002 PetscReal sum = 0.0; 2003 PetscErrorCode ierr; 2004 PetscInt i,j; 2005 2006 PetscFunctionBegin; 2007 if (type == NORM_FROBENIUS) { 2008 for (i=0; i<a->nz; i++) { 2009 sum += PetscRealPart(PetscConj(*v)*(*v)); v++; 2010 } 2011 *nrm = PetscSqrtReal(sum); 2012 } else if (type == NORM_1) { 2013 PetscReal *tmp; 2014 PetscInt *jj = a->j; 2015 ierr = PetscMalloc((A->cmap->n+1)*sizeof(PetscReal),&tmp);CHKERRQ(ierr); 2016 ierr = PetscMemzero(tmp,A->cmap->n*sizeof(PetscReal));CHKERRQ(ierr); 2017 *nrm = 0.0; 2018 for (j=0; j<a->nz; j++) { 2019 tmp[*jj++] += PetscAbsScalar(*v); v++; 2020 } 2021 for (j=0; j<A->cmap->n; j++) { 2022 if (tmp[j] > *nrm) *nrm = tmp[j]; 2023 } 2024 ierr = PetscFree(tmp);CHKERRQ(ierr); 2025 } else if (type == NORM_INFINITY) { 2026 *nrm = 0.0; 2027 for (j=0; j<A->rmap->n; j++) { 2028 v = a->a + a->i[j]; 2029 sum = 0.0; 2030 for (i=0; i<a->i[j+1]-a->i[j]; i++) { 2031 sum += PetscAbsScalar(*v); v++; 2032 } 2033 if (sum > *nrm) *nrm = sum; 2034 } 2035 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm"); 2036 PetscFunctionReturn(0); 2037 } 2038 2039 /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */ 2040 #undef __FUNCT__ 2041 #define __FUNCT__ "MatTransposeSymbolic_SeqAIJ" 2042 PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B) 2043 { 2044 PetscErrorCode ierr; 2045 PetscInt i,j,anzj; 2046 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b; 2047 PetscInt an=A->cmap->N,am=A->rmap->N; 2048 PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j; 2049 2050 PetscFunctionBegin; 2051 /* Allocate space for symbolic transpose info and work array */ 2052 ierr = PetscMalloc((an+1)*sizeof(PetscInt),&ati);CHKERRQ(ierr); 2053 ierr = PetscMalloc(ai[am]*sizeof(PetscInt),&atj);CHKERRQ(ierr); 2054 ierr = PetscMalloc(an*sizeof(PetscInt),&atfill);CHKERRQ(ierr); 2055 ierr = PetscMemzero(ati,(an+1)*sizeof(PetscInt));CHKERRQ(ierr); 2056 2057 /* Walk through aj and count ## of non-zeros in each row of A^T. */ 2058 /* Note: offset by 1 for fast conversion into csr format. */ 2059 for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1; 2060 /* Form ati for csr format of A^T. */ 2061 for (i=0;i<an;i++) ati[i+1] += ati[i]; 2062 2063 /* Copy ati into atfill so we have locations of the next free space in atj */ 2064 ierr = PetscMemcpy(atfill,ati,an*sizeof(PetscInt));CHKERRQ(ierr); 2065 2066 /* Walk through A row-wise and mark nonzero entries of A^T. */ 2067 for (i=0;i<am;i++) { 2068 anzj = ai[i+1] - ai[i]; 2069 for (j=0;j<anzj;j++) { 2070 atj[atfill[*aj]] = i; 2071 atfill[*aj++] += 1; 2072 } 2073 } 2074 2075 /* Clean up temporary space and complete requests. */ 2076 ierr = PetscFree(atfill);CHKERRQ(ierr); 2077 ierr = MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B);CHKERRQ(ierr); 2078 2079 (*B)->rmap->bs = A->cmap->bs; 2080 (*B)->cmap->bs = A->rmap->bs; 2081 2082 b = (Mat_SeqAIJ*)((*B)->data); 2083 b->free_a = PETSC_FALSE; 2084 b->free_ij = PETSC_TRUE; 2085 b->nonew = 0; 2086 PetscFunctionReturn(0); 2087 } 2088 2089 #undef __FUNCT__ 2090 #define __FUNCT__ "MatTranspose_SeqAIJ" 2091 PetscErrorCode MatTranspose_SeqAIJ(Mat A,MatReuse reuse,Mat *B) 2092 { 2093 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2094 Mat C; 2095 PetscErrorCode ierr; 2096 PetscInt i,*aj = a->j,*ai = a->i,m = A->rmap->n,len,*col; 2097 MatScalar *array = a->a; 2098 2099 PetscFunctionBegin; 2100 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"); 2101 2102 if (reuse == MAT_INITIAL_MATRIX || *B == A) { 2103 ierr = PetscMalloc((1+A->cmap->n)*sizeof(PetscInt),&col);CHKERRQ(ierr); 2104 ierr = PetscMemzero(col,(1+A->cmap->n)*sizeof(PetscInt));CHKERRQ(ierr); 2105 2106 for (i=0; i<ai[m]; i++) col[aj[i]] += 1; 2107 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2108 ierr = MatSetSizes(C,A->cmap->n,m,A->cmap->n,m);CHKERRQ(ierr); 2109 ierr = MatSetBlockSizes(C,A->cmap->bs,A->rmap->bs);CHKERRQ(ierr); 2110 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2111 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,col);CHKERRQ(ierr); 2112 ierr = PetscFree(col);CHKERRQ(ierr); 2113 } else { 2114 C = *B; 2115 } 2116 2117 for (i=0; i<m; i++) { 2118 len = ai[i+1]-ai[i]; 2119 ierr = MatSetValues_SeqAIJ(C,len,aj,1,&i,array,INSERT_VALUES);CHKERRQ(ierr); 2120 array += len; 2121 aj += len; 2122 } 2123 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2124 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2125 2126 if (reuse == MAT_INITIAL_MATRIX || *B != A) { 2127 *B = C; 2128 } else { 2129 ierr = MatHeaderMerge(A,C);CHKERRQ(ierr); 2130 } 2131 PetscFunctionReturn(0); 2132 } 2133 2134 #undef __FUNCT__ 2135 #define __FUNCT__ "MatIsTranspose_SeqAIJ" 2136 PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2137 { 2138 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data; 2139 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2140 MatScalar *va,*vb; 2141 PetscErrorCode ierr; 2142 PetscInt ma,na,mb,nb, i; 2143 2144 PetscFunctionBegin; 2145 bij = (Mat_SeqAIJ*) B->data; 2146 2147 ierr = MatGetSize(A,&ma,&na);CHKERRQ(ierr); 2148 ierr = MatGetSize(B,&mb,&nb);CHKERRQ(ierr); 2149 if (ma!=nb || na!=mb) { 2150 *f = PETSC_FALSE; 2151 PetscFunctionReturn(0); 2152 } 2153 aii = aij->i; bii = bij->i; 2154 adx = aij->j; bdx = bij->j; 2155 va = aij->a; vb = bij->a; 2156 ierr = PetscMalloc(ma*sizeof(PetscInt),&aptr);CHKERRQ(ierr); 2157 ierr = PetscMalloc(mb*sizeof(PetscInt),&bptr);CHKERRQ(ierr); 2158 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2159 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2160 2161 *f = PETSC_TRUE; 2162 for (i=0; i<ma; i++) { 2163 while (aptr[i]<aii[i+1]) { 2164 PetscInt idc,idr; 2165 PetscScalar vc,vr; 2166 /* column/row index/value */ 2167 idc = adx[aptr[i]]; 2168 idr = bdx[bptr[idc]]; 2169 vc = va[aptr[i]]; 2170 vr = vb[bptr[idc]]; 2171 if (i!=idr || PetscAbsScalar(vc-vr) > tol) { 2172 *f = PETSC_FALSE; 2173 goto done; 2174 } else { 2175 aptr[i]++; 2176 if (B || i!=idc) bptr[idc]++; 2177 } 2178 } 2179 } 2180 done: 2181 ierr = PetscFree(aptr);CHKERRQ(ierr); 2182 ierr = PetscFree(bptr);CHKERRQ(ierr); 2183 PetscFunctionReturn(0); 2184 } 2185 2186 #undef __FUNCT__ 2187 #define __FUNCT__ "MatIsHermitianTranspose_SeqAIJ" 2188 PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f) 2189 { 2190 Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data; 2191 PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; 2192 MatScalar *va,*vb; 2193 PetscErrorCode ierr; 2194 PetscInt ma,na,mb,nb, i; 2195 2196 PetscFunctionBegin; 2197 bij = (Mat_SeqAIJ*) B->data; 2198 2199 ierr = MatGetSize(A,&ma,&na);CHKERRQ(ierr); 2200 ierr = MatGetSize(B,&mb,&nb);CHKERRQ(ierr); 2201 if (ma!=nb || na!=mb) { 2202 *f = PETSC_FALSE; 2203 PetscFunctionReturn(0); 2204 } 2205 aii = aij->i; bii = bij->i; 2206 adx = aij->j; bdx = bij->j; 2207 va = aij->a; vb = bij->a; 2208 ierr = PetscMalloc(ma*sizeof(PetscInt),&aptr);CHKERRQ(ierr); 2209 ierr = PetscMalloc(mb*sizeof(PetscInt),&bptr);CHKERRQ(ierr); 2210 for (i=0; i<ma; i++) aptr[i] = aii[i]; 2211 for (i=0; i<mb; i++) bptr[i] = bii[i]; 2212 2213 *f = PETSC_TRUE; 2214 for (i=0; i<ma; i++) { 2215 while (aptr[i]<aii[i+1]) { 2216 PetscInt idc,idr; 2217 PetscScalar vc,vr; 2218 /* column/row index/value */ 2219 idc = adx[aptr[i]]; 2220 idr = bdx[bptr[idc]]; 2221 vc = va[aptr[i]]; 2222 vr = vb[bptr[idc]]; 2223 if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) { 2224 *f = PETSC_FALSE; 2225 goto done; 2226 } else { 2227 aptr[i]++; 2228 if (B || i!=idc) bptr[idc]++; 2229 } 2230 } 2231 } 2232 done: 2233 ierr = PetscFree(aptr);CHKERRQ(ierr); 2234 ierr = PetscFree(bptr);CHKERRQ(ierr); 2235 PetscFunctionReturn(0); 2236 } 2237 2238 #undef __FUNCT__ 2239 #define __FUNCT__ "MatIsSymmetric_SeqAIJ" 2240 PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2241 { 2242 PetscErrorCode ierr; 2243 2244 PetscFunctionBegin; 2245 ierr = MatIsTranspose_SeqAIJ(A,A,tol,f);CHKERRQ(ierr); 2246 PetscFunctionReturn(0); 2247 } 2248 2249 #undef __FUNCT__ 2250 #define __FUNCT__ "MatIsHermitian_SeqAIJ" 2251 PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f) 2252 { 2253 PetscErrorCode ierr; 2254 2255 PetscFunctionBegin; 2256 ierr = MatIsHermitianTranspose_SeqAIJ(A,A,tol,f);CHKERRQ(ierr); 2257 PetscFunctionReturn(0); 2258 } 2259 2260 #undef __FUNCT__ 2261 #define __FUNCT__ "MatDiagonalScale_SeqAIJ" 2262 PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr) 2263 { 2264 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2265 PetscScalar *l,*r,x; 2266 MatScalar *v; 2267 PetscErrorCode ierr; 2268 PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz,*jj; 2269 2270 PetscFunctionBegin; 2271 if (ll) { 2272 /* The local size is used so that VecMPI can be passed to this routine 2273 by MatDiagonalScale_MPIAIJ */ 2274 ierr = VecGetLocalSize(ll,&m);CHKERRQ(ierr); 2275 if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length"); 2276 ierr = VecGetArray(ll,&l);CHKERRQ(ierr); 2277 v = a->a; 2278 for (i=0; i<m; i++) { 2279 x = l[i]; 2280 M = a->i[i+1] - a->i[i]; 2281 for (j=0; j<M; j++) (*v++) *= x; 2282 } 2283 ierr = VecRestoreArray(ll,&l);CHKERRQ(ierr); 2284 ierr = PetscLogFlops(nz);CHKERRQ(ierr); 2285 } 2286 if (rr) { 2287 ierr = VecGetLocalSize(rr,&n);CHKERRQ(ierr); 2288 if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length"); 2289 ierr = VecGetArray(rr,&r);CHKERRQ(ierr); 2290 v = a->a; jj = a->j; 2291 for (i=0; i<nz; i++) (*v++) *= r[*jj++]; 2292 ierr = VecRestoreArray(rr,&r);CHKERRQ(ierr); 2293 ierr = PetscLogFlops(nz);CHKERRQ(ierr); 2294 } 2295 ierr = MatSeqAIJInvalidateDiagonal(A);CHKERRQ(ierr); 2296 PetscFunctionReturn(0); 2297 } 2298 2299 #undef __FUNCT__ 2300 #define __FUNCT__ "MatGetSubMatrix_SeqAIJ" 2301 PetscErrorCode MatGetSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B) 2302 { 2303 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c; 2304 PetscErrorCode ierr; 2305 PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens; 2306 PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi; 2307 const PetscInt *irow,*icol; 2308 PetscInt nrows,ncols; 2309 PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen; 2310 MatScalar *a_new,*mat_a; 2311 Mat C; 2312 PetscBool stride,sorted; 2313 2314 PetscFunctionBegin; 2315 ierr = ISSorted(isrow,&sorted);CHKERRQ(ierr); 2316 if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"ISrow is not sorted"); 2317 ierr = ISSorted(iscol,&sorted);CHKERRQ(ierr); 2318 if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"IScol is not sorted"); 2319 2320 ierr = ISGetIndices(isrow,&irow);CHKERRQ(ierr); 2321 ierr = ISGetLocalSize(isrow,&nrows);CHKERRQ(ierr); 2322 ierr = ISGetLocalSize(iscol,&ncols);CHKERRQ(ierr); 2323 2324 ierr = ISStrideGetInfo(iscol,&first,&step);CHKERRQ(ierr); 2325 ierr = PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride);CHKERRQ(ierr); 2326 if (stride && step == 1) { 2327 /* special case of contiguous rows */ 2328 ierr = PetscMalloc2(nrows,PetscInt,&lens,nrows,PetscInt,&starts);CHKERRQ(ierr); 2329 /* loop over new rows determining lens and starting points */ 2330 for (i=0; i<nrows; i++) { 2331 kstart = ai[irow[i]]; 2332 kend = kstart + ailen[irow[i]]; 2333 for (k=kstart; k<kend; k++) { 2334 if (aj[k] >= first) { 2335 starts[i] = k; 2336 break; 2337 } 2338 } 2339 sum = 0; 2340 while (k < kend) { 2341 if (aj[k++] >= first+ncols) break; 2342 sum++; 2343 } 2344 lens[i] = sum; 2345 } 2346 /* create submatrix */ 2347 if (scall == MAT_REUSE_MATRIX) { 2348 PetscInt n_cols,n_rows; 2349 ierr = MatGetSize(*B,&n_rows,&n_cols);CHKERRQ(ierr); 2350 if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size"); 2351 ierr = MatZeroEntries(*B);CHKERRQ(ierr); 2352 C = *B; 2353 } else { 2354 PetscInt rbs,cbs; 2355 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2356 ierr = MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2357 ierr = ISGetBlockSize(isrow,&rbs);CHKERRQ(ierr); 2358 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2359 ierr = MatSetBlockSizes(C,rbs,cbs);CHKERRQ(ierr); 2360 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2361 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);CHKERRQ(ierr); 2362 } 2363 c = (Mat_SeqAIJ*)C->data; 2364 2365 /* loop over rows inserting into submatrix */ 2366 a_new = c->a; 2367 j_new = c->j; 2368 i_new = c->i; 2369 2370 for (i=0; i<nrows; i++) { 2371 ii = starts[i]; 2372 lensi = lens[i]; 2373 for (k=0; k<lensi; k++) { 2374 *j_new++ = aj[ii+k] - first; 2375 } 2376 ierr = PetscMemcpy(a_new,a->a + starts[i],lensi*sizeof(PetscScalar));CHKERRQ(ierr); 2377 a_new += lensi; 2378 i_new[i+1] = i_new[i] + lensi; 2379 c->ilen[i] = lensi; 2380 } 2381 ierr = PetscFree2(lens,starts);CHKERRQ(ierr); 2382 } else { 2383 ierr = ISGetIndices(iscol,&icol);CHKERRQ(ierr); 2384 ierr = PetscMalloc(oldcols*sizeof(PetscInt),&smap);CHKERRQ(ierr); 2385 ierr = PetscMemzero(smap,oldcols*sizeof(PetscInt));CHKERRQ(ierr); 2386 ierr = PetscMalloc((1+nrows)*sizeof(PetscInt),&lens);CHKERRQ(ierr); 2387 for (i=0; i<ncols; i++) { 2388 #if defined(PETSC_USE_DEBUG) 2389 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); 2390 #endif 2391 smap[icol[i]] = i+1; 2392 } 2393 2394 /* determine lens of each row */ 2395 for (i=0; i<nrows; i++) { 2396 kstart = ai[irow[i]]; 2397 kend = kstart + a->ilen[irow[i]]; 2398 lens[i] = 0; 2399 for (k=kstart; k<kend; k++) { 2400 if (smap[aj[k]]) { 2401 lens[i]++; 2402 } 2403 } 2404 } 2405 /* Create and fill new matrix */ 2406 if (scall == MAT_REUSE_MATRIX) { 2407 PetscBool equal; 2408 2409 c = (Mat_SeqAIJ*)((*B)->data); 2410 if ((*B)->rmap->n != nrows || (*B)->cmap->n != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size"); 2411 ierr = PetscMemcmp(c->ilen,lens,(*B)->rmap->n*sizeof(PetscInt),&equal);CHKERRQ(ierr); 2412 if (!equal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros"); 2413 ierr = PetscMemzero(c->ilen,(*B)->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 2414 C = *B; 2415 } else { 2416 PetscInt rbs,cbs; 2417 ierr = MatCreate(PetscObjectComm((PetscObject)A),&C);CHKERRQ(ierr); 2418 ierr = MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 2419 ierr = ISGetBlockSize(isrow,&rbs);CHKERRQ(ierr); 2420 ierr = ISGetBlockSize(iscol,&cbs);CHKERRQ(ierr); 2421 ierr = MatSetBlockSizes(C,rbs,cbs);CHKERRQ(ierr); 2422 ierr = MatSetType(C,((PetscObject)A)->type_name);CHKERRQ(ierr); 2423 ierr = MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);CHKERRQ(ierr); 2424 } 2425 c = (Mat_SeqAIJ*)(C->data); 2426 for (i=0; i<nrows; i++) { 2427 row = irow[i]; 2428 kstart = ai[row]; 2429 kend = kstart + a->ilen[row]; 2430 mat_i = c->i[i]; 2431 mat_j = c->j + mat_i; 2432 mat_a = c->a + mat_i; 2433 mat_ilen = c->ilen + i; 2434 for (k=kstart; k<kend; k++) { 2435 if ((tcol=smap[a->j[k]])) { 2436 *mat_j++ = tcol - 1; 2437 *mat_a++ = a->a[k]; 2438 (*mat_ilen)++; 2439 2440 } 2441 } 2442 } 2443 /* Free work space */ 2444 ierr = ISRestoreIndices(iscol,&icol);CHKERRQ(ierr); 2445 ierr = PetscFree(smap);CHKERRQ(ierr); 2446 ierr = PetscFree(lens);CHKERRQ(ierr); 2447 } 2448 ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2449 ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2450 2451 ierr = ISRestoreIndices(isrow,&irow);CHKERRQ(ierr); 2452 *B = C; 2453 PetscFunctionReturn(0); 2454 } 2455 2456 #undef __FUNCT__ 2457 #define __FUNCT__ "MatGetMultiProcBlock_SeqAIJ" 2458 PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat) 2459 { 2460 PetscErrorCode ierr; 2461 Mat B; 2462 2463 PetscFunctionBegin; 2464 if (scall == MAT_INITIAL_MATRIX) { 2465 ierr = MatCreate(subComm,&B);CHKERRQ(ierr); 2466 ierr = MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n);CHKERRQ(ierr); 2467 ierr = MatSetBlockSizes(B,mat->rmap->bs,mat->cmap->bs);CHKERRQ(ierr); 2468 ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr); 2469 ierr = MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE);CHKERRQ(ierr); 2470 *subMat = B; 2471 } else { 2472 ierr = MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 2473 } 2474 PetscFunctionReturn(0); 2475 } 2476 2477 #undef __FUNCT__ 2478 #define __FUNCT__ "MatILUFactor_SeqAIJ" 2479 PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info) 2480 { 2481 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2482 PetscErrorCode ierr; 2483 Mat outA; 2484 PetscBool row_identity,col_identity; 2485 2486 PetscFunctionBegin; 2487 if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu"); 2488 2489 ierr = ISIdentity(row,&row_identity);CHKERRQ(ierr); 2490 ierr = ISIdentity(col,&col_identity);CHKERRQ(ierr); 2491 2492 outA = inA; 2493 outA->factortype = MAT_FACTOR_LU; 2494 2495 ierr = PetscObjectReference((PetscObject)row);CHKERRQ(ierr); 2496 ierr = ISDestroy(&a->row);CHKERRQ(ierr); 2497 2498 a->row = row; 2499 2500 ierr = PetscObjectReference((PetscObject)col);CHKERRQ(ierr); 2501 ierr = ISDestroy(&a->col);CHKERRQ(ierr); 2502 2503 a->col = col; 2504 2505 /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */ 2506 ierr = ISDestroy(&a->icol);CHKERRQ(ierr); 2507 ierr = ISInvertPermutation(col,PETSC_DECIDE,&a->icol);CHKERRQ(ierr); 2508 ierr = PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);CHKERRQ(ierr); 2509 2510 if (!a->solve_work) { /* this matrix may have been factored before */ 2511 ierr = PetscMalloc((inA->rmap->n+1)*sizeof(PetscScalar),&a->solve_work);CHKERRQ(ierr); 2512 ierr = PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar));CHKERRQ(ierr); 2513 } 2514 2515 ierr = MatMarkDiagonal_SeqAIJ(inA);CHKERRQ(ierr); 2516 if (row_identity && col_identity) { 2517 ierr = MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info);CHKERRQ(ierr); 2518 } else { 2519 ierr = MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info);CHKERRQ(ierr); 2520 } 2521 PetscFunctionReturn(0); 2522 } 2523 2524 #undef __FUNCT__ 2525 #define __FUNCT__ "MatScale_SeqAIJ" 2526 PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha) 2527 { 2528 Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data; 2529 PetscScalar oalpha = alpha; 2530 PetscErrorCode ierr; 2531 PetscBLASInt one = 1,bnz; 2532 2533 PetscFunctionBegin; 2534 ierr = PetscBLASIntCast(a->nz,&bnz);CHKERRQ(ierr); 2535 PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&oalpha,a->a,&one)); 2536 ierr = PetscLogFlops(a->nz);CHKERRQ(ierr); 2537 ierr = MatSeqAIJInvalidateDiagonal(inA);CHKERRQ(ierr); 2538 PetscFunctionReturn(0); 2539 } 2540 2541 #undef __FUNCT__ 2542 #define __FUNCT__ "MatGetSubMatrices_SeqAIJ" 2543 PetscErrorCode MatGetSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[]) 2544 { 2545 PetscErrorCode ierr; 2546 PetscInt i; 2547 2548 PetscFunctionBegin; 2549 if (scall == MAT_INITIAL_MATRIX) { 2550 ierr = PetscMalloc((n+1)*sizeof(Mat),B);CHKERRQ(ierr); 2551 } 2552 2553 for (i=0; i<n; i++) { 2554 ierr = MatGetSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);CHKERRQ(ierr); 2555 } 2556 PetscFunctionReturn(0); 2557 } 2558 2559 #undef __FUNCT__ 2560 #define __FUNCT__ "MatIncreaseOverlap_SeqAIJ" 2561 PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov) 2562 { 2563 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2564 PetscErrorCode ierr; 2565 PetscInt row,i,j,k,l,m,n,*nidx,isz,val; 2566 const PetscInt *idx; 2567 PetscInt start,end,*ai,*aj; 2568 PetscBT table; 2569 2570 PetscFunctionBegin; 2571 m = A->rmap->n; 2572 ai = a->i; 2573 aj = a->j; 2574 2575 if (ov < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used"); 2576 2577 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&nidx);CHKERRQ(ierr); 2578 ierr = PetscBTCreate(m,&table);CHKERRQ(ierr); 2579 2580 for (i=0; i<is_max; i++) { 2581 /* Initialize the two local arrays */ 2582 isz = 0; 2583 ierr = PetscBTMemzero(m,table);CHKERRQ(ierr); 2584 2585 /* Extract the indices, assume there can be duplicate entries */ 2586 ierr = ISGetIndices(is[i],&idx);CHKERRQ(ierr); 2587 ierr = ISGetLocalSize(is[i],&n);CHKERRQ(ierr); 2588 2589 /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */ 2590 for (j=0; j<n; ++j) { 2591 if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j]; 2592 } 2593 ierr = ISRestoreIndices(is[i],&idx);CHKERRQ(ierr); 2594 ierr = ISDestroy(&is[i]);CHKERRQ(ierr); 2595 2596 k = 0; 2597 for (j=0; j<ov; j++) { /* for each overlap */ 2598 n = isz; 2599 for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */ 2600 row = nidx[k]; 2601 start = ai[row]; 2602 end = ai[row+1]; 2603 for (l = start; l<end; l++) { 2604 val = aj[l]; 2605 if (!PetscBTLookupSet(table,val)) nidx[isz++] = val; 2606 } 2607 } 2608 } 2609 ierr = ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i));CHKERRQ(ierr); 2610 } 2611 ierr = PetscBTDestroy(&table);CHKERRQ(ierr); 2612 ierr = PetscFree(nidx);CHKERRQ(ierr); 2613 PetscFunctionReturn(0); 2614 } 2615 2616 /* -------------------------------------------------------------- */ 2617 #undef __FUNCT__ 2618 #define __FUNCT__ "MatPermute_SeqAIJ" 2619 PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B) 2620 { 2621 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2622 PetscErrorCode ierr; 2623 PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n; 2624 const PetscInt *row,*col; 2625 PetscInt *cnew,j,*lens; 2626 IS icolp,irowp; 2627 PetscInt *cwork = NULL; 2628 PetscScalar *vwork = NULL; 2629 2630 PetscFunctionBegin; 2631 ierr = ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);CHKERRQ(ierr); 2632 ierr = ISGetIndices(irowp,&row);CHKERRQ(ierr); 2633 ierr = ISInvertPermutation(colp,PETSC_DECIDE,&icolp);CHKERRQ(ierr); 2634 ierr = ISGetIndices(icolp,&col);CHKERRQ(ierr); 2635 2636 /* determine lengths of permuted rows */ 2637 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&lens);CHKERRQ(ierr); 2638 for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i]; 2639 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 2640 ierr = MatSetSizes(*B,m,n,m,n);CHKERRQ(ierr); 2641 ierr = MatSetBlockSizes(*B,A->rmap->bs,A->cmap->bs);CHKERRQ(ierr); 2642 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 2643 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);CHKERRQ(ierr); 2644 ierr = PetscFree(lens);CHKERRQ(ierr); 2645 2646 ierr = PetscMalloc(n*sizeof(PetscInt),&cnew);CHKERRQ(ierr); 2647 for (i=0; i<m; i++) { 2648 ierr = MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);CHKERRQ(ierr); 2649 for (j=0; j<nz; j++) cnew[j] = col[cwork[j]]; 2650 ierr = MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);CHKERRQ(ierr); 2651 ierr = MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);CHKERRQ(ierr); 2652 } 2653 ierr = PetscFree(cnew);CHKERRQ(ierr); 2654 2655 (*B)->assembled = PETSC_FALSE; 2656 2657 ierr = MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2658 ierr = MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2659 ierr = ISRestoreIndices(irowp,&row);CHKERRQ(ierr); 2660 ierr = ISRestoreIndices(icolp,&col);CHKERRQ(ierr); 2661 ierr = ISDestroy(&irowp);CHKERRQ(ierr); 2662 ierr = ISDestroy(&icolp);CHKERRQ(ierr); 2663 PetscFunctionReturn(0); 2664 } 2665 2666 #undef __FUNCT__ 2667 #define __FUNCT__ "MatCopy_SeqAIJ" 2668 PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str) 2669 { 2670 PetscErrorCode ierr; 2671 2672 PetscFunctionBegin; 2673 /* If the two matrices have the same copy implementation, use fast copy. */ 2674 if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) { 2675 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2676 Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data; 2677 2678 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"); 2679 ierr = PetscMemcpy(b->a,a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));CHKERRQ(ierr); 2680 } else { 2681 ierr = MatCopy_Basic(A,B,str);CHKERRQ(ierr); 2682 } 2683 PetscFunctionReturn(0); 2684 } 2685 2686 #undef __FUNCT__ 2687 #define __FUNCT__ "MatSetUp_SeqAIJ" 2688 PetscErrorCode MatSetUp_SeqAIJ(Mat A) 2689 { 2690 PetscErrorCode ierr; 2691 2692 PetscFunctionBegin; 2693 ierr = MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,0);CHKERRQ(ierr); 2694 PetscFunctionReturn(0); 2695 } 2696 2697 #undef __FUNCT__ 2698 #define __FUNCT__ "MatSeqAIJGetArray_SeqAIJ" 2699 PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[]) 2700 { 2701 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 2702 2703 PetscFunctionBegin; 2704 *array = a->a; 2705 PetscFunctionReturn(0); 2706 } 2707 2708 #undef __FUNCT__ 2709 #define __FUNCT__ "MatSeqAIJRestoreArray_SeqAIJ" 2710 PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[]) 2711 { 2712 PetscFunctionBegin; 2713 PetscFunctionReturn(0); 2714 } 2715 2716 /* Optimize MatFDColoringApply_AIJ() by using array den2sp to skip calling MatSetValues() */ 2717 /* #define JACOBIANCOLOROPT */ 2718 #if defined(JACOBIANCOLOROPT) 2719 #include <petsctime.h> 2720 #endif 2721 #undef __FUNCT__ 2722 #define __FUNCT__ "MatFDColoringApply_SeqAIJ" 2723 PetscErrorCode MatFDColoringApply_SeqAIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 2724 { 2725 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 2726 PetscErrorCode ierr; 2727 PetscInt k,start,end,l,row,col,**vscaleforrow; 2728 PetscScalar dx,*y,*xx,*w3_array; 2729 PetscScalar *vscale_array; 2730 PetscReal epsilon = coloring->error_rel,umin = coloring->umin,unorm; 2731 Vec w1 = coloring->w1,w2=coloring->w2,w3; 2732 void *fctx = coloring->fctx; 2733 PetscBool flg = PETSC_FALSE; 2734 PetscInt ctype = coloring->ctype,N,col_start=0,col_end=0; 2735 Vec x1_tmp; 2736 Mat_SeqAIJ *csp = (Mat_SeqAIJ*)J->data; 2737 PetscInt *den2sp=coloring->den2sp,*idx=den2sp; 2738 PetscScalar *ca=csp->a; 2739 const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows; 2740 PetscInt **rows=coloring->rows,**columns=coloring->columns,ncolumns_k,nrows_k; 2741 #if defined(JACOBIANCOLOROPT) 2742 PetscLogDouble t0,t1,time_setvalues=0.0; 2743 #endif 2744 2745 PetscFunctionBegin; 2746 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 2747 ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 2748 if (flg) { 2749 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 2750 } else { 2751 PetscBool assembled; 2752 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 2753 if (assembled) { 2754 ierr = MatZeroEntries(J);CHKERRQ(ierr); 2755 } 2756 } 2757 2758 x1_tmp = x1; 2759 if (!coloring->vscale) { 2760 ierr = VecDuplicate(x1_tmp,&coloring->vscale);CHKERRQ(ierr); 2761 } 2762 2763 if (coloring->htype[0] == 'w') { /* tacky test; need to make systematic if we add other approaches to computing h*/ 2764 ierr = VecNorm(x1_tmp,NORM_2,&unorm);CHKERRQ(ierr); 2765 } 2766 ierr = VecGetOwnershipRange(w1,&start,&end);CHKERRQ(ierr); /* OwnershipRange is used by ghosted x! */ 2767 2768 /* Set w1 = F(x1) */ 2769 if (!coloring->fset) { 2770 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2771 ierr = (*f)(sctx,x1_tmp,w1,fctx);CHKERRQ(ierr); 2772 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2773 } else { 2774 coloring->fset = PETSC_FALSE; 2775 } 2776 2777 if (!coloring->w3) { 2778 ierr = VecDuplicate(x1_tmp,&coloring->w3);CHKERRQ(ierr); 2779 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 2780 } 2781 w3 = coloring->w3; 2782 2783 /* Compute all the local scale factors, including ghost points */ 2784 ierr = VecGetLocalSize(x1_tmp,&N);CHKERRQ(ierr); 2785 ierr = VecGetArray(x1_tmp,&xx);CHKERRQ(ierr); 2786 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2787 if (ctype == IS_COLORING_GHOSTED) { 2788 col_start = 0; col_end = N; 2789 } else if (ctype == IS_COLORING_GLOBAL) { 2790 xx = xx - start; 2791 vscale_array = vscale_array - start; 2792 col_start = start; col_end = N + start; 2793 } 2794 for (col=col_start; col<col_end; col++) { 2795 /* Loop over each local column, vscale[col] = 1./(epsilon*dx[col]) */ 2796 if (coloring->htype[0] == 'w') { 2797 dx = 1.0 + unorm; 2798 } else { 2799 dx = xx[col]; 2800 } 2801 if (dx == (PetscScalar)0.0) dx = 1.0; 2802 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2803 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2804 dx *= epsilon; 2805 vscale_array[col] = (PetscScalar)1.0/dx; 2806 } 2807 if (ctype == IS_COLORING_GLOBAL) vscale_array = vscale_array + start; 2808 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2809 if (ctype == IS_COLORING_GLOBAL) { 2810 ierr = VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2811 ierr = VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2812 } 2813 2814 if (coloring->vscaleforrow) { 2815 vscaleforrow = coloring->vscaleforrow; 2816 } else SETERRQ(PetscObjectComm((PetscObject)J),PETSC_ERR_ARG_NULL,"Null Object: coloring->vscaleforrow"); 2817 2818 /* 2819 Loop over each color 2820 */ 2821 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2822 for (k=0; k<ncolors; k++) { /* loop over colors */ 2823 coloring->currentcolor = k; 2824 2825 ierr = VecCopy(x1_tmp,w3);CHKERRQ(ierr); 2826 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr); 2827 if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array - start; 2828 /* 2829 Loop over each column associated with color 2830 adding the perturbation to the vector w3. 2831 */ 2832 ncolumns_k = ncolumns[k]; 2833 for (l=0; l<ncolumns_k; l++) { /* loop over columns */ 2834 col = columns[k][l]; /* local column of the matrix we are probing for */ 2835 if (coloring->htype[0] == 'w') { 2836 dx = 1.0 + unorm; 2837 } else { 2838 dx = xx[col]; 2839 } 2840 if (dx == (PetscScalar)0.0) dx = 1.0; 2841 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2842 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2843 dx *= epsilon; 2844 if (!PetscAbsScalar(dx)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Computed 0 differencing parameter"); 2845 w3_array[col] += dx; 2846 } 2847 if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array + start; 2848 ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 2849 2850 /* 2851 Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations) 2852 w2 = F(x1 + dx) - F(x1) 2853 */ 2854 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2855 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 2856 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2857 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 2858 2859 /* 2860 Loop over rows of vector, putting results into Jacobian matrix 2861 */ 2862 #if defined(JACOBIANCOLOROPT) 2863 ierr = PetscTime(&t0);CHKERRQ(ierr); 2864 #endif 2865 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 2866 nrows_k = nrows[k]; 2867 for (l=0; l<nrows_k; l++) { /* loop over rows */ 2868 row = rows[k][l]; /* local row index */ 2869 y[row] *= vscale_array[vscaleforrow[k][l]]; 2870 ca[idx[l]] = y[row]; 2871 } 2872 idx += nrows_k; 2873 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 2874 #if defined(JACOBIANCOLOROPT) 2875 ierr = PetscTime(&t1);CHKERRQ(ierr); 2876 time_setvalues += t1-t0; 2877 #endif 2878 } /* endof for each color */ 2879 #if defined(JACOBIANCOLOROPT) 2880 printf(" MatFDColoringApply_SeqAIJ: time_setvalues %g\n",time_setvalues); 2881 #endif 2882 if (ctype == IS_COLORING_GLOBAL) xx = xx + start; 2883 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2884 ierr = VecRestoreArray(x1_tmp,&xx);CHKERRQ(ierr); 2885 2886 coloring->currentcolor = -1; 2887 PetscFunctionReturn(0); 2888 } 2889 /* --------------------------------------------------------*/ 2890 2891 #undef __FUNCT__ 2892 #define __FUNCT__ "MatFDColoringApply_SeqAIJ_old" 2893 PetscErrorCode MatFDColoringApply_SeqAIJ_old(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx) 2894 { 2895 PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f; 2896 PetscErrorCode ierr; 2897 PetscInt k,N,start,end,l,row,col,srow,**vscaleforrow; 2898 PetscScalar dx,*y,*xx,*w3_array; 2899 PetscScalar *vscale_array; 2900 PetscReal epsilon = coloring->error_rel,umin = coloring->umin; 2901 Vec w1,w2,w3; 2902 void *fctx = coloring->fctx; 2903 PetscBool flg = PETSC_FALSE; 2904 2905 PetscFunctionBegin; 2906 printf("MatFDColoringApply_SeqAIJ ...\n"); 2907 if (!coloring->w1) { 2908 ierr = VecDuplicate(x1,&coloring->w1);CHKERRQ(ierr); 2909 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w1);CHKERRQ(ierr); 2910 ierr = VecDuplicate(x1,&coloring->w2);CHKERRQ(ierr); 2911 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w2);CHKERRQ(ierr); 2912 ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr); 2913 ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr); 2914 } 2915 w1 = coloring->w1; w2 = coloring->w2; w3 = coloring->w3; 2916 2917 ierr = MatSetUnfactored(J);CHKERRQ(ierr); 2918 ierr = PetscOptionsGetBool(((PetscObject)coloring)->prefix,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr); 2919 if (flg) { 2920 ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr); 2921 } else { 2922 PetscBool assembled; 2923 ierr = MatAssembled(J,&assembled);CHKERRQ(ierr); 2924 if (assembled) { 2925 ierr = MatZeroEntries(J);CHKERRQ(ierr); 2926 } 2927 } 2928 2929 ierr = VecGetOwnershipRange(x1,&start,&end);CHKERRQ(ierr); 2930 ierr = VecGetSize(x1,&N);CHKERRQ(ierr); 2931 2932 if (!coloring->fset) { 2933 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2934 ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr); 2935 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 2936 } else { 2937 coloring->fset = PETSC_FALSE; 2938 } 2939 2940 /* 2941 Compute all the scale factors and share with other processors 2942 */ 2943 ierr = VecGetArray(x1,&xx);CHKERRQ(ierr);xx = xx - start; 2944 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);vscale_array = vscale_array - start; 2945 for (k=0; k<coloring->ncolors; k++) { 2946 /* 2947 Loop over each column associated with color adding the 2948 perturbation to the vector w3. 2949 */ 2950 for (l=0; l<coloring->ncolumns[k]; l++) { 2951 col = coloring->columns[k][l]; /* column of the matrix we are probing for */ 2952 dx = xx[col]; 2953 if (dx == 0.0) dx = 1.0; 2954 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2955 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2956 dx *= epsilon; 2957 vscale_array[col] = 1.0/dx; 2958 } 2959 } 2960 vscale_array = vscale_array + start; 2961 2962 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2963 ierr = VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2964 ierr = VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 2965 2966 /* ierr = VecView(coloring->vscale,PETSC_VIEWER_STDOUT_WORLD); 2967 ierr = VecView(x1,PETSC_VIEWER_STDOUT_WORLD);*/ 2968 2969 if (coloring->vscaleforrow) vscaleforrow = coloring->vscaleforrow; 2970 else vscaleforrow = coloring->columnsforrow; 2971 2972 ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 2973 /* 2974 Loop over each color 2975 */ 2976 for (k=0; k<coloring->ncolors; k++) { 2977 coloring->currentcolor = k; 2978 2979 ierr = VecCopy(x1,w3);CHKERRQ(ierr); 2980 ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr);w3_array = w3_array - start; 2981 /* 2982 Loop over each column associated with color adding the 2983 perturbation to the vector w3. 2984 */ 2985 for (l=0; l<coloring->ncolumns[k]; l++) { 2986 col = coloring->columns[k][l]; /* column of the matrix we are probing for */ 2987 dx = xx[col]; 2988 if (dx == 0.0) dx = 1.0; 2989 if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin; 2990 else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin; 2991 dx *= epsilon; 2992 if (!PetscAbsScalar(dx)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Computed 0 differencing parameter"); 2993 w3_array[col] += dx; 2994 } 2995 w3_array = w3_array + start; ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr); 2996 2997 /* 2998 Evaluate function at x1 + dx (here dx is a vector of perturbations) 2999 */ 3000 3001 ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 3002 ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr); 3003 ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr); 3004 ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr); 3005 3006 /* 3007 Loop over rows of vector, putting results into Jacobian matrix 3008 */ 3009 ierr = VecGetArray(w2,&y);CHKERRQ(ierr); 3010 for (l=0; l<coloring->nrows[k]; l++) { 3011 row = coloring->rows[k][l]; 3012 col = coloring->columnsforrow[k][l]; 3013 y[row] *= vscale_array[vscaleforrow[k][l]]; 3014 srow = row + start; 3015 ierr = MatSetValues_SeqAIJ(J,1,&srow,1,&col,y+row,INSERT_VALUES);CHKERRQ(ierr); 3016 } 3017 ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr); 3018 } 3019 coloring->currentcolor = k; 3020 3021 ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr); 3022 xx = xx + start; 3023 ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr); 3024 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3025 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3026 PetscFunctionReturn(0); 3027 } 3028 3029 /* 3030 Computes the number of nonzeros per row needed for preallocation when X and Y 3031 have different nonzero structure. 3032 */ 3033 #undef __FUNCT__ 3034 #define __FUNCT__ "MatAXPYGetPreallocation_SeqAIJ" 3035 PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz) 3036 { 3037 PetscInt i,m=Y->rmap->N; 3038 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data; 3039 Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data; 3040 const PetscInt *xi = x->i,*yi = y->i; 3041 3042 PetscFunctionBegin; 3043 /* Set the number of nonzeros in the new matrix */ 3044 for (i=0; i<m; i++) { 3045 PetscInt j,k,nzx = xi[i+1] - xi[i],nzy = yi[i+1] - yi[i]; 3046 const PetscInt *xj = x->j+xi[i],*yj = y->j+yi[i]; 3047 nnz[i] = 0; 3048 for (j=0,k=0; j<nzx; j++) { /* Point in X */ 3049 for (; k<nzy && yj[k]<xj[j]; k++) nnz[i]++; /* Catch up to X */ 3050 if (k<nzy && yj[k]==xj[j]) k++; /* Skip duplicate */ 3051 nnz[i]++; 3052 } 3053 for (; k<nzy; k++) nnz[i]++; 3054 } 3055 PetscFunctionReturn(0); 3056 } 3057 3058 #undef __FUNCT__ 3059 #define __FUNCT__ "MatAXPY_SeqAIJ" 3060 PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str) 3061 { 3062 PetscErrorCode ierr; 3063 PetscInt i; 3064 Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data; 3065 PetscBLASInt one=1,bnz; 3066 3067 PetscFunctionBegin; 3068 ierr = PetscBLASIntCast(x->nz,&bnz);CHKERRQ(ierr); 3069 if (str == SAME_NONZERO_PATTERN) { 3070 PetscScalar alpha = a; 3071 PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one)); 3072 ierr = MatSeqAIJInvalidateDiagonal(Y);CHKERRQ(ierr); 3073 } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */ 3074 if (y->xtoy && y->XtoY != X) { 3075 ierr = PetscFree(y->xtoy);CHKERRQ(ierr); 3076 ierr = MatDestroy(&y->XtoY);CHKERRQ(ierr); 3077 } 3078 if (!y->xtoy) { /* get xtoy */ 3079 ierr = MatAXPYGetxtoy_Private(X->rmap->n,x->i,x->j,NULL, y->i,y->j,NULL, &y->xtoy);CHKERRQ(ierr); 3080 y->XtoY = X; 3081 ierr = PetscObjectReference((PetscObject)X);CHKERRQ(ierr); 3082 } 3083 for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]); 3084 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); 3085 } else { 3086 Mat B; 3087 PetscInt *nnz; 3088 ierr = PetscMalloc(Y->rmap->N*sizeof(PetscInt),&nnz);CHKERRQ(ierr); 3089 ierr = MatCreate(PetscObjectComm((PetscObject)Y),&B);CHKERRQ(ierr); 3090 ierr = PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);CHKERRQ(ierr); 3091 ierr = MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);CHKERRQ(ierr); 3092 ierr = MatSetBlockSizes(B,Y->rmap->bs,Y->cmap->bs);CHKERRQ(ierr); 3093 ierr = MatSetType(B,(MatType) ((PetscObject)Y)->type_name);CHKERRQ(ierr); 3094 ierr = MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);CHKERRQ(ierr); 3095 ierr = MatSeqAIJSetPreallocation(B,0,nnz);CHKERRQ(ierr); 3096 ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); 3097 ierr = MatHeaderReplace(Y,B);CHKERRQ(ierr); 3098 ierr = PetscFree(nnz);CHKERRQ(ierr); 3099 } 3100 PetscFunctionReturn(0); 3101 } 3102 3103 #undef __FUNCT__ 3104 #define __FUNCT__ "MatConjugate_SeqAIJ" 3105 PetscErrorCode MatConjugate_SeqAIJ(Mat mat) 3106 { 3107 #if defined(PETSC_USE_COMPLEX) 3108 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3109 PetscInt i,nz; 3110 PetscScalar *a; 3111 3112 PetscFunctionBegin; 3113 nz = aij->nz; 3114 a = aij->a; 3115 for (i=0; i<nz; i++) a[i] = PetscConj(a[i]); 3116 #else 3117 PetscFunctionBegin; 3118 #endif 3119 PetscFunctionReturn(0); 3120 } 3121 3122 #undef __FUNCT__ 3123 #define __FUNCT__ "MatGetRowMaxAbs_SeqAIJ" 3124 PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3125 { 3126 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3127 PetscErrorCode ierr; 3128 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3129 PetscReal atmp; 3130 PetscScalar *x; 3131 MatScalar *aa; 3132 3133 PetscFunctionBegin; 3134 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3135 aa = a->a; 3136 ai = a->i; 3137 aj = a->j; 3138 3139 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3140 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3141 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3142 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3143 for (i=0; i<m; i++) { 3144 ncols = ai[1] - ai[0]; ai++; 3145 x[i] = 0.0; 3146 for (j=0; j<ncols; j++) { 3147 atmp = PetscAbsScalar(*aa); 3148 if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3149 aa++; aj++; 3150 } 3151 } 3152 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3153 PetscFunctionReturn(0); 3154 } 3155 3156 #undef __FUNCT__ 3157 #define __FUNCT__ "MatGetRowMax_SeqAIJ" 3158 PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3159 { 3160 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3161 PetscErrorCode ierr; 3162 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3163 PetscScalar *x; 3164 MatScalar *aa; 3165 3166 PetscFunctionBegin; 3167 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3168 aa = a->a; 3169 ai = a->i; 3170 aj = a->j; 3171 3172 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3173 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3174 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3175 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3176 for (i=0; i<m; i++) { 3177 ncols = ai[1] - ai[0]; ai++; 3178 if (ncols == A->cmap->n) { /* row is dense */ 3179 x[i] = *aa; if (idx) idx[i] = 0; 3180 } else { /* row is sparse so already KNOW maximum is 0.0 or higher */ 3181 x[i] = 0.0; 3182 if (idx) { 3183 idx[i] = 0; /* in case ncols is zero */ 3184 for (j=0;j<ncols;j++) { /* find first implicit 0.0 in the row */ 3185 if (aj[j] > j) { 3186 idx[i] = j; 3187 break; 3188 } 3189 } 3190 } 3191 } 3192 for (j=0; j<ncols; j++) { 3193 if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3194 aa++; aj++; 3195 } 3196 } 3197 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3198 PetscFunctionReturn(0); 3199 } 3200 3201 #undef __FUNCT__ 3202 #define __FUNCT__ "MatGetRowMinAbs_SeqAIJ" 3203 PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3204 { 3205 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3206 PetscErrorCode ierr; 3207 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3208 PetscReal atmp; 3209 PetscScalar *x; 3210 MatScalar *aa; 3211 3212 PetscFunctionBegin; 3213 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3214 aa = a->a; 3215 ai = a->i; 3216 aj = a->j; 3217 3218 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3219 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3220 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3221 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); 3222 for (i=0; i<m; i++) { 3223 ncols = ai[1] - ai[0]; ai++; 3224 if (ncols) { 3225 /* Get first nonzero */ 3226 for (j = 0; j < ncols; j++) { 3227 atmp = PetscAbsScalar(aa[j]); 3228 if (atmp > 1.0e-12) { 3229 x[i] = atmp; 3230 if (idx) idx[i] = aj[j]; 3231 break; 3232 } 3233 } 3234 if (j == ncols) {x[i] = PetscAbsScalar(*aa); if (idx) idx[i] = *aj;} 3235 } else { 3236 x[i] = 0.0; if (idx) idx[i] = 0; 3237 } 3238 for (j = 0; j < ncols; j++) { 3239 atmp = PetscAbsScalar(*aa); 3240 if (atmp > 1.0e-12 && PetscAbsScalar(x[i]) > atmp) {x[i] = atmp; if (idx) idx[i] = *aj;} 3241 aa++; aj++; 3242 } 3243 } 3244 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3245 PetscFunctionReturn(0); 3246 } 3247 3248 #undef __FUNCT__ 3249 #define __FUNCT__ "MatGetRowMin_SeqAIJ" 3250 PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[]) 3251 { 3252 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 3253 PetscErrorCode ierr; 3254 PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n; 3255 PetscScalar *x; 3256 MatScalar *aa; 3257 3258 PetscFunctionBegin; 3259 if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3260 aa = a->a; 3261 ai = a->i; 3262 aj = a->j; 3263 3264 ierr = VecSet(v,0.0);CHKERRQ(ierr); 3265 ierr = VecGetArray(v,&x);CHKERRQ(ierr); 3266 ierr = VecGetLocalSize(v,&n);CHKERRQ(ierr); 3267 if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector"); 3268 for (i=0; i<m; i++) { 3269 ncols = ai[1] - ai[0]; ai++; 3270 if (ncols == A->cmap->n) { /* row is dense */ 3271 x[i] = *aa; if (idx) idx[i] = 0; 3272 } else { /* row is sparse so already KNOW minimum is 0.0 or lower */ 3273 x[i] = 0.0; 3274 if (idx) { /* find first implicit 0.0 in the row */ 3275 idx[i] = 0; /* in case ncols is zero */ 3276 for (j=0; j<ncols; j++) { 3277 if (aj[j] > j) { 3278 idx[i] = j; 3279 break; 3280 } 3281 } 3282 } 3283 } 3284 for (j=0; j<ncols; j++) { 3285 if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;} 3286 aa++; aj++; 3287 } 3288 } 3289 ierr = VecRestoreArray(v,&x);CHKERRQ(ierr); 3290 PetscFunctionReturn(0); 3291 } 3292 3293 #include <petscblaslapack.h> 3294 #include <petsc-private/kernels/blockinvert.h> 3295 3296 #undef __FUNCT__ 3297 #define __FUNCT__ "MatInvertBlockDiagonal_SeqAIJ" 3298 PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values) 3299 { 3300 Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data; 3301 PetscErrorCode ierr; 3302 PetscInt i,bs = A->rmap->bs,mbs = A->rmap->n/A->rmap->bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j; 3303 MatScalar *diag,work[25],*v_work; 3304 PetscReal shift = 0.0; 3305 3306 PetscFunctionBegin; 3307 if (a->ibdiagvalid) { 3308 if (values) *values = a->ibdiag; 3309 PetscFunctionReturn(0); 3310 } 3311 ierr = MatMarkDiagonal_SeqAIJ(A);CHKERRQ(ierr); 3312 if (!a->ibdiag) { 3313 ierr = PetscMalloc(bs2*mbs*sizeof(PetscScalar),&a->ibdiag);CHKERRQ(ierr); 3314 ierr = PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));CHKERRQ(ierr); 3315 } 3316 diag = a->ibdiag; 3317 if (values) *values = a->ibdiag; 3318 /* factor and invert each block */ 3319 switch (bs) { 3320 case 1: 3321 for (i=0; i<mbs; i++) { 3322 ierr = MatGetValues(A,1,&i,1,&i,diag+i);CHKERRQ(ierr); 3323 diag[i] = (PetscScalar)1.0 / (diag[i] + shift); 3324 } 3325 break; 3326 case 2: 3327 for (i=0; i<mbs; i++) { 3328 ij[0] = 2*i; ij[1] = 2*i + 1; 3329 ierr = MatGetValues(A,2,ij,2,ij,diag);CHKERRQ(ierr); 3330 ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 3331 ierr = PetscKernel_A_gets_transpose_A_2(diag);CHKERRQ(ierr); 3332 diag += 4; 3333 } 3334 break; 3335 case 3: 3336 for (i=0; i<mbs; i++) { 3337 ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2; 3338 ierr = MatGetValues(A,3,ij,3,ij,diag);CHKERRQ(ierr); 3339 ierr = PetscKernel_A_gets_inverse_A_3(diag,shift);CHKERRQ(ierr); 3340 ierr = PetscKernel_A_gets_transpose_A_3(diag);CHKERRQ(ierr); 3341 diag += 9; 3342 } 3343 break; 3344 case 4: 3345 for (i=0; i<mbs; i++) { 3346 ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3; 3347 ierr = MatGetValues(A,4,ij,4,ij,diag);CHKERRQ(ierr); 3348 ierr = PetscKernel_A_gets_inverse_A_4(diag,shift);CHKERRQ(ierr); 3349 ierr = PetscKernel_A_gets_transpose_A_4(diag);CHKERRQ(ierr); 3350 diag += 16; 3351 } 3352 break; 3353 case 5: 3354 for (i=0; i<mbs; i++) { 3355 ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4; 3356 ierr = MatGetValues(A,5,ij,5,ij,diag);CHKERRQ(ierr); 3357 ierr = PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift);CHKERRQ(ierr); 3358 ierr = PetscKernel_A_gets_transpose_A_5(diag);CHKERRQ(ierr); 3359 diag += 25; 3360 } 3361 break; 3362 case 6: 3363 for (i=0; i<mbs; i++) { 3364 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; 3365 ierr = MatGetValues(A,6,ij,6,ij,diag);CHKERRQ(ierr); 3366 ierr = PetscKernel_A_gets_inverse_A_6(diag,shift);CHKERRQ(ierr); 3367 ierr = PetscKernel_A_gets_transpose_A_6(diag);CHKERRQ(ierr); 3368 diag += 36; 3369 } 3370 break; 3371 case 7: 3372 for (i=0; i<mbs; i++) { 3373 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; 3374 ierr = MatGetValues(A,7,ij,7,ij,diag);CHKERRQ(ierr); 3375 ierr = PetscKernel_A_gets_inverse_A_7(diag,shift);CHKERRQ(ierr); 3376 ierr = PetscKernel_A_gets_transpose_A_7(diag);CHKERRQ(ierr); 3377 diag += 49; 3378 } 3379 break; 3380 default: 3381 ierr = PetscMalloc3(bs,MatScalar,&v_work,bs,PetscInt,&v_pivots,bs,PetscInt,&IJ);CHKERRQ(ierr); 3382 for (i=0; i<mbs; i++) { 3383 for (j=0; j<bs; j++) { 3384 IJ[j] = bs*i + j; 3385 } 3386 ierr = MatGetValues(A,bs,IJ,bs,IJ,diag);CHKERRQ(ierr); 3387 ierr = PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work);CHKERRQ(ierr); 3388 ierr = PetscKernel_A_gets_transpose_A_N(diag,bs);CHKERRQ(ierr); 3389 diag += bs2; 3390 } 3391 ierr = PetscFree3(v_work,v_pivots,IJ);CHKERRQ(ierr); 3392 } 3393 a->ibdiagvalid = PETSC_TRUE; 3394 PetscFunctionReturn(0); 3395 } 3396 3397 #undef __FUNCT__ 3398 #define __FUNCT__ "MatSetRandom_SeqAIJ" 3399 static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx) 3400 { 3401 PetscErrorCode ierr; 3402 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data; 3403 PetscScalar a; 3404 PetscInt m,n,i,j,col; 3405 3406 PetscFunctionBegin; 3407 if (!x->assembled) { 3408 ierr = MatGetSize(x,&m,&n);CHKERRQ(ierr); 3409 for (i=0; i<m; i++) { 3410 for (j=0; j<aij->imax[i]; j++) { 3411 ierr = PetscRandomGetValue(rctx,&a);CHKERRQ(ierr); 3412 col = (PetscInt)(n*PetscRealPart(a)); 3413 ierr = MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);CHKERRQ(ierr); 3414 } 3415 } 3416 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not yet coded"); 3417 ierr = MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3418 ierr = MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 3419 PetscFunctionReturn(0); 3420 } 3421 3422 extern PetscErrorCode MatFDColoringApply_AIJ(Mat,MatFDColoring,Vec,MatStructure*,void*); 3423 /* -------------------------------------------------------------------*/ 3424 static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ, 3425 MatGetRow_SeqAIJ, 3426 MatRestoreRow_SeqAIJ, 3427 MatMult_SeqAIJ, 3428 /* 4*/ MatMultAdd_SeqAIJ, 3429 MatMultTranspose_SeqAIJ, 3430 MatMultTransposeAdd_SeqAIJ, 3431 0, 3432 0, 3433 0, 3434 /* 10*/ 0, 3435 MatLUFactor_SeqAIJ, 3436 0, 3437 MatSOR_SeqAIJ, 3438 MatTranspose_SeqAIJ, 3439 /*1 5*/ MatGetInfo_SeqAIJ, 3440 MatEqual_SeqAIJ, 3441 MatGetDiagonal_SeqAIJ, 3442 MatDiagonalScale_SeqAIJ, 3443 MatNorm_SeqAIJ, 3444 /* 20*/ 0, 3445 MatAssemblyEnd_SeqAIJ, 3446 MatSetOption_SeqAIJ, 3447 MatZeroEntries_SeqAIJ, 3448 /* 24*/ MatZeroRows_SeqAIJ, 3449 0, 3450 0, 3451 0, 3452 0, 3453 /* 29*/ MatSetUp_SeqAIJ, 3454 0, 3455 0, 3456 0, 3457 0, 3458 /* 34*/ MatDuplicate_SeqAIJ, 3459 0, 3460 0, 3461 MatILUFactor_SeqAIJ, 3462 0, 3463 /* 39*/ MatAXPY_SeqAIJ, 3464 MatGetSubMatrices_SeqAIJ, 3465 MatIncreaseOverlap_SeqAIJ, 3466 MatGetValues_SeqAIJ, 3467 MatCopy_SeqAIJ, 3468 /* 44*/ MatGetRowMax_SeqAIJ, 3469 MatScale_SeqAIJ, 3470 0, 3471 MatDiagonalSet_SeqAIJ, 3472 MatZeroRowsColumns_SeqAIJ, 3473 /* 49*/ MatSetRandom_SeqAIJ, 3474 MatGetRowIJ_SeqAIJ, 3475 MatRestoreRowIJ_SeqAIJ, 3476 MatGetColumnIJ_SeqAIJ, 3477 MatRestoreColumnIJ_SeqAIJ, 3478 /* 54*/ MatFDColoringCreate_SeqAIJ, 3479 0, 3480 0, 3481 MatPermute_SeqAIJ, 3482 0, 3483 /* 59*/ 0, 3484 MatDestroy_SeqAIJ, 3485 MatView_SeqAIJ, 3486 0, 3487 MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ, 3488 /* 64*/ MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ, 3489 MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ, 3490 0, 3491 0, 3492 0, 3493 /* 69*/ MatGetRowMaxAbs_SeqAIJ, 3494 MatGetRowMinAbs_SeqAIJ, 3495 0, 3496 MatSetColoring_SeqAIJ, 3497 0, 3498 /* 74*/ MatSetValuesAdifor_SeqAIJ, 3499 MatFDColoringApply_AIJ, 3500 0, 3501 0, 3502 0, 3503 /* 79*/ MatFindZeroDiagonals_SeqAIJ, 3504 0, 3505 0, 3506 0, 3507 MatLoad_SeqAIJ, 3508 /* 84*/ MatIsSymmetric_SeqAIJ, 3509 MatIsHermitian_SeqAIJ, 3510 0, 3511 0, 3512 0, 3513 /* 89*/ MatMatMult_SeqAIJ_SeqAIJ, 3514 MatMatMultSymbolic_SeqAIJ_SeqAIJ, 3515 MatMatMultNumeric_SeqAIJ_SeqAIJ, 3516 MatPtAP_SeqAIJ_SeqAIJ, 3517 MatPtAPSymbolic_SeqAIJ_SeqAIJ_DenseAxpy, 3518 /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ, 3519 MatMatTransposeMult_SeqAIJ_SeqAIJ, 3520 MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ, 3521 MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ, 3522 0, 3523 /* 99*/ 0, 3524 0, 3525 0, 3526 MatConjugate_SeqAIJ, 3527 0, 3528 /*104*/ MatSetValuesRow_SeqAIJ, 3529 MatRealPart_SeqAIJ, 3530 MatImaginaryPart_SeqAIJ, 3531 0, 3532 0, 3533 /*109*/ MatMatSolve_SeqAIJ, 3534 0, 3535 MatGetRowMin_SeqAIJ, 3536 0, 3537 MatMissingDiagonal_SeqAIJ, 3538 /*114*/ 0, 3539 0, 3540 0, 3541 0, 3542 0, 3543 /*119*/ 0, 3544 0, 3545 0, 3546 0, 3547 MatGetMultiProcBlock_SeqAIJ, 3548 /*124*/ MatFindNonzeroRows_SeqAIJ, 3549 MatGetColumnNorms_SeqAIJ, 3550 MatInvertBlockDiagonal_SeqAIJ, 3551 0, 3552 0, 3553 /*129*/ 0, 3554 MatTransposeMatMult_SeqAIJ_SeqAIJ, 3555 MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ, 3556 MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ, 3557 MatTransposeColoringCreate_SeqAIJ, 3558 /*134*/ MatTransColoringApplySpToDen_SeqAIJ, 3559 MatTransColoringApplyDenToSp_SeqAIJ, 3560 MatRARt_SeqAIJ_SeqAIJ, 3561 MatRARtSymbolic_SeqAIJ_SeqAIJ, 3562 MatRARtNumeric_SeqAIJ_SeqAIJ, 3563 /*139*/0, 3564 0 3565 }; 3566 3567 #undef __FUNCT__ 3568 #define __FUNCT__ "MatSeqAIJSetColumnIndices_SeqAIJ" 3569 PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices) 3570 { 3571 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3572 PetscInt i,nz,n; 3573 3574 PetscFunctionBegin; 3575 nz = aij->maxnz; 3576 n = mat->rmap->n; 3577 for (i=0; i<nz; i++) { 3578 aij->j[i] = indices[i]; 3579 } 3580 aij->nz = nz; 3581 for (i=0; i<n; i++) { 3582 aij->ilen[i] = aij->imax[i]; 3583 } 3584 PetscFunctionReturn(0); 3585 } 3586 3587 #undef __FUNCT__ 3588 #define __FUNCT__ "MatSeqAIJSetColumnIndices" 3589 /*@ 3590 MatSeqAIJSetColumnIndices - Set the column indices for all the rows 3591 in the matrix. 3592 3593 Input Parameters: 3594 + mat - the SeqAIJ matrix 3595 - indices - the column indices 3596 3597 Level: advanced 3598 3599 Notes: 3600 This can be called if you have precomputed the nonzero structure of the 3601 matrix and want to provide it to the matrix object to improve the performance 3602 of the MatSetValues() operation. 3603 3604 You MUST have set the correct numbers of nonzeros per row in the call to 3605 MatCreateSeqAIJ(), and the columns indices MUST be sorted. 3606 3607 MUST be called before any calls to MatSetValues(); 3608 3609 The indices should start with zero, not one. 3610 3611 @*/ 3612 PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices) 3613 { 3614 PetscErrorCode ierr; 3615 3616 PetscFunctionBegin; 3617 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3618 PetscValidPointer(indices,2); 3619 ierr = PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));CHKERRQ(ierr); 3620 PetscFunctionReturn(0); 3621 } 3622 3623 /* ----------------------------------------------------------------------------------------*/ 3624 3625 #undef __FUNCT__ 3626 #define __FUNCT__ "MatStoreValues_SeqAIJ" 3627 PetscErrorCode MatStoreValues_SeqAIJ(Mat mat) 3628 { 3629 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3630 PetscErrorCode ierr; 3631 size_t nz = aij->i[mat->rmap->n]; 3632 3633 PetscFunctionBegin; 3634 if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3635 3636 /* allocate space for values if not already there */ 3637 if (!aij->saved_values) { 3638 ierr = PetscMalloc((nz+1)*sizeof(PetscScalar),&aij->saved_values);CHKERRQ(ierr); 3639 ierr = PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));CHKERRQ(ierr); 3640 } 3641 3642 /* copy values over */ 3643 ierr = PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3644 PetscFunctionReturn(0); 3645 } 3646 3647 #undef __FUNCT__ 3648 #define __FUNCT__ "MatStoreValues" 3649 /*@ 3650 MatStoreValues - Stashes a copy of the matrix values; this allows, for 3651 example, reuse of the linear part of a Jacobian, while recomputing the 3652 nonlinear portion. 3653 3654 Collect on Mat 3655 3656 Input Parameters: 3657 . mat - the matrix (currently only AIJ matrices support this option) 3658 3659 Level: advanced 3660 3661 Common Usage, with SNESSolve(): 3662 $ Create Jacobian matrix 3663 $ Set linear terms into matrix 3664 $ Apply boundary conditions to matrix, at this time matrix must have 3665 $ final nonzero structure (i.e. setting the nonlinear terms and applying 3666 $ boundary conditions again will not change the nonzero structure 3667 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3668 $ ierr = MatStoreValues(mat); 3669 $ Call SNESSetJacobian() with matrix 3670 $ In your Jacobian routine 3671 $ ierr = MatRetrieveValues(mat); 3672 $ Set nonlinear terms in matrix 3673 3674 Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself: 3675 $ // build linear portion of Jacobian 3676 $ ierr = MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); 3677 $ ierr = MatStoreValues(mat); 3678 $ loop over nonlinear iterations 3679 $ ierr = MatRetrieveValues(mat); 3680 $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian 3681 $ // call MatAssemblyBegin/End() on matrix 3682 $ Solve linear system with Jacobian 3683 $ endloop 3684 3685 Notes: 3686 Matrix must already be assemblied before calling this routine 3687 Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before 3688 calling this routine. 3689 3690 When this is called multiple times it overwrites the previous set of stored values 3691 and does not allocated additional space. 3692 3693 .seealso: MatRetrieveValues() 3694 3695 @*/ 3696 PetscErrorCode MatStoreValues(Mat mat) 3697 { 3698 PetscErrorCode ierr; 3699 3700 PetscFunctionBegin; 3701 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3702 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3703 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3704 ierr = PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));CHKERRQ(ierr); 3705 PetscFunctionReturn(0); 3706 } 3707 3708 #undef __FUNCT__ 3709 #define __FUNCT__ "MatRetrieveValues_SeqAIJ" 3710 PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat) 3711 { 3712 Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data; 3713 PetscErrorCode ierr; 3714 PetscInt nz = aij->i[mat->rmap->n]; 3715 3716 PetscFunctionBegin; 3717 if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first"); 3718 if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first"); 3719 /* copy values over */ 3720 ierr = PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));CHKERRQ(ierr); 3721 PetscFunctionReturn(0); 3722 } 3723 3724 #undef __FUNCT__ 3725 #define __FUNCT__ "MatRetrieveValues" 3726 /*@ 3727 MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for 3728 example, reuse of the linear part of a Jacobian, while recomputing the 3729 nonlinear portion. 3730 3731 Collect on Mat 3732 3733 Input Parameters: 3734 . mat - the matrix (currently on AIJ matrices support this option) 3735 3736 Level: advanced 3737 3738 .seealso: MatStoreValues() 3739 3740 @*/ 3741 PetscErrorCode MatRetrieveValues(Mat mat) 3742 { 3743 PetscErrorCode ierr; 3744 3745 PetscFunctionBegin; 3746 PetscValidHeaderSpecific(mat,MAT_CLASSID,1); 3747 if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix"); 3748 if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix"); 3749 ierr = PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));CHKERRQ(ierr); 3750 PetscFunctionReturn(0); 3751 } 3752 3753 3754 /* --------------------------------------------------------------------------------*/ 3755 #undef __FUNCT__ 3756 #define __FUNCT__ "MatCreateSeqAIJ" 3757 /*@C 3758 MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format 3759 (the default parallel PETSc format). For good matrix assembly performance 3760 the user should preallocate the matrix storage by setting the parameter nz 3761 (or the array nnz). By setting these parameters accurately, performance 3762 during matrix assembly can be increased by more than a factor of 50. 3763 3764 Collective on MPI_Comm 3765 3766 Input Parameters: 3767 + comm - MPI communicator, set to PETSC_COMM_SELF 3768 . m - number of rows 3769 . n - number of columns 3770 . nz - number of nonzeros per row (same for all rows) 3771 - nnz - array containing the number of nonzeros in the various rows 3772 (possibly different for each row) or NULL 3773 3774 Output Parameter: 3775 . A - the matrix 3776 3777 It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(), 3778 MatXXXXSetPreallocation() paradgm instead of this routine directly. 3779 [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation] 3780 3781 Notes: 3782 If nnz is given then nz is ignored 3783 3784 The AIJ format (also called the Yale sparse matrix format or 3785 compressed row storage), is fully compatible with standard Fortran 77 3786 storage. That is, the stored row and column indices can begin at 3787 either one (as in Fortran) or zero. See the users' manual for details. 3788 3789 Specify the preallocated storage with either nz or nnz (not both). 3790 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3791 allocation. For large problems you MUST preallocate memory or you 3792 will get TERRIBLE performance, see the users' manual chapter on matrices. 3793 3794 By default, this format uses inodes (identical nodes) when possible, to 3795 improve numerical efficiency of matrix-vector products and solves. We 3796 search for consecutive rows with the same nonzero structure, thereby 3797 reusing matrix information to achieve increased efficiency. 3798 3799 Options Database Keys: 3800 + -mat_no_inode - Do not use inodes 3801 - -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3802 3803 Level: intermediate 3804 3805 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays() 3806 3807 @*/ 3808 PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A) 3809 { 3810 PetscErrorCode ierr; 3811 3812 PetscFunctionBegin; 3813 ierr = MatCreate(comm,A);CHKERRQ(ierr); 3814 ierr = MatSetSizes(*A,m,n,m,n);CHKERRQ(ierr); 3815 ierr = MatSetType(*A,MATSEQAIJ);CHKERRQ(ierr); 3816 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);CHKERRQ(ierr); 3817 PetscFunctionReturn(0); 3818 } 3819 3820 #undef __FUNCT__ 3821 #define __FUNCT__ "MatSeqAIJSetPreallocation" 3822 /*@C 3823 MatSeqAIJSetPreallocation - For good matrix assembly performance 3824 the user should preallocate the matrix storage by setting the parameter nz 3825 (or the array nnz). By setting these parameters accurately, performance 3826 during matrix assembly can be increased by more than a factor of 50. 3827 3828 Collective on MPI_Comm 3829 3830 Input Parameters: 3831 + B - The matrix-free 3832 . nz - number of nonzeros per row (same for all rows) 3833 - nnz - array containing the number of nonzeros in the various rows 3834 (possibly different for each row) or NULL 3835 3836 Notes: 3837 If nnz is given then nz is ignored 3838 3839 The AIJ format (also called the Yale sparse matrix format or 3840 compressed row storage), is fully compatible with standard Fortran 77 3841 storage. That is, the stored row and column indices can begin at 3842 either one (as in Fortran) or zero. See the users' manual for details. 3843 3844 Specify the preallocated storage with either nz or nnz (not both). 3845 Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory 3846 allocation. For large problems you MUST preallocate memory or you 3847 will get TERRIBLE performance, see the users' manual chapter on matrices. 3848 3849 You can call MatGetInfo() to get information on how effective the preallocation was; 3850 for example the fields mallocs,nz_allocated,nz_used,nz_unneeded; 3851 You can also run with the option -info and look for messages with the string 3852 malloc in them to see if additional memory allocation was needed. 3853 3854 Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix 3855 entries or columns indices 3856 3857 By default, this format uses inodes (identical nodes) when possible, to 3858 improve numerical efficiency of matrix-vector products and solves. We 3859 search for consecutive rows with the same nonzero structure, thereby 3860 reusing matrix information to achieve increased efficiency. 3861 3862 Options Database Keys: 3863 + -mat_no_inode - Do not use inodes 3864 . -mat_inode_limit <limit> - Sets inode limit (max limit=5) 3865 - -mat_aij_oneindex - Internally use indexing starting at 1 3866 rather than 0. Note that when calling MatSetValues(), 3867 the user still MUST index entries starting at 0! 3868 3869 Level: intermediate 3870 3871 .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo() 3872 3873 @*/ 3874 PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[]) 3875 { 3876 PetscErrorCode ierr; 3877 3878 PetscFunctionBegin; 3879 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3880 PetscValidType(B,1); 3881 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));CHKERRQ(ierr); 3882 PetscFunctionReturn(0); 3883 } 3884 3885 #undef __FUNCT__ 3886 #define __FUNCT__ "MatSeqAIJSetPreallocation_SeqAIJ" 3887 PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz) 3888 { 3889 Mat_SeqAIJ *b; 3890 PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE; 3891 PetscErrorCode ierr; 3892 PetscInt i; 3893 3894 PetscFunctionBegin; 3895 if (nz >= 0 || nnz) realalloc = PETSC_TRUE; 3896 if (nz == MAT_SKIP_ALLOCATION) { 3897 skipallocation = PETSC_TRUE; 3898 nz = 0; 3899 } 3900 3901 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 3902 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 3903 3904 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5; 3905 if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %d",nz); 3906 if (nnz) { 3907 for (i=0; i<B->rmap->n; i++) { 3908 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]); 3909 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); 3910 } 3911 } 3912 3913 B->preallocated = PETSC_TRUE; 3914 3915 b = (Mat_SeqAIJ*)B->data; 3916 3917 if (!skipallocation) { 3918 if (!b->imax) { 3919 ierr = PetscMalloc2(B->rmap->n,PetscInt,&b->imax,B->rmap->n,PetscInt,&b->ilen);CHKERRQ(ierr); 3920 ierr = PetscLogObjectMemory((PetscObject)B,2*B->rmap->n*sizeof(PetscInt));CHKERRQ(ierr); 3921 } 3922 if (!nnz) { 3923 if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10; 3924 else if (nz < 0) nz = 1; 3925 for (i=0; i<B->rmap->n; i++) b->imax[i] = nz; 3926 nz = nz*B->rmap->n; 3927 } else { 3928 nz = 0; 3929 for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz += nnz[i];} 3930 } 3931 /* b->ilen will count nonzeros in each row so far. */ 3932 for (i=0; i<B->rmap->n; i++) b->ilen[i] = 0; 3933 3934 /* allocate the matrix space */ 3935 ierr = MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);CHKERRQ(ierr); 3936 ierr = PetscMalloc3(nz,PetscScalar,&b->a,nz,PetscInt,&b->j,B->rmap->n+1,PetscInt,&b->i);CHKERRQ(ierr); 3937 ierr = PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));CHKERRQ(ierr); 3938 b->i[0] = 0; 3939 for (i=1; i<B->rmap->n+1; i++) { 3940 b->i[i] = b->i[i-1] + b->imax[i-1]; 3941 } 3942 b->singlemalloc = PETSC_TRUE; 3943 b->free_a = PETSC_TRUE; 3944 b->free_ij = PETSC_TRUE; 3945 #if defined(PETSC_THREADCOMM_ACTIVE) 3946 ierr = MatZeroEntries_SeqAIJ(B);CHKERRQ(ierr); 3947 #endif 3948 } else { 3949 b->free_a = PETSC_FALSE; 3950 b->free_ij = PETSC_FALSE; 3951 } 3952 3953 b->nz = 0; 3954 b->maxnz = nz; 3955 B->info.nz_unneeded = (double)b->maxnz; 3956 if (realalloc) { 3957 ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 3958 } 3959 PetscFunctionReturn(0); 3960 } 3961 3962 #undef __FUNCT__ 3963 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR" 3964 /*@ 3965 MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format. 3966 3967 Input Parameters: 3968 + B - the matrix 3969 . i - the indices into j for the start of each row (starts with zero) 3970 . j - the column indices for each row (starts with zero) these must be sorted for each row 3971 - v - optional values in the matrix 3972 3973 Level: developer 3974 3975 The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays() 3976 3977 .keywords: matrix, aij, compressed row, sparse, sequential 3978 3979 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), SeqAIJ 3980 @*/ 3981 PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[]) 3982 { 3983 PetscErrorCode ierr; 3984 3985 PetscFunctionBegin; 3986 PetscValidHeaderSpecific(B,MAT_CLASSID,1); 3987 PetscValidType(B,1); 3988 ierr = PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));CHKERRQ(ierr); 3989 PetscFunctionReturn(0); 3990 } 3991 3992 #undef __FUNCT__ 3993 #define __FUNCT__ "MatSeqAIJSetPreallocationCSR_SeqAIJ" 3994 PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[]) 3995 { 3996 PetscInt i; 3997 PetscInt m,n; 3998 PetscInt nz; 3999 PetscInt *nnz, nz_max = 0; 4000 PetscScalar *values; 4001 PetscErrorCode ierr; 4002 4003 PetscFunctionBegin; 4004 if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]); 4005 4006 ierr = PetscLayoutSetUp(B->rmap);CHKERRQ(ierr); 4007 ierr = PetscLayoutSetUp(B->cmap);CHKERRQ(ierr); 4008 4009 ierr = MatGetSize(B, &m, &n);CHKERRQ(ierr); 4010 ierr = PetscMalloc((m+1) * sizeof(PetscInt), &nnz);CHKERRQ(ierr); 4011 for (i = 0; i < m; i++) { 4012 nz = Ii[i+1]- Ii[i]; 4013 nz_max = PetscMax(nz_max, nz); 4014 if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz); 4015 nnz[i] = nz; 4016 } 4017 ierr = MatSeqAIJSetPreallocation(B, 0, nnz);CHKERRQ(ierr); 4018 ierr = PetscFree(nnz);CHKERRQ(ierr); 4019 4020 if (v) { 4021 values = (PetscScalar*) v; 4022 } else { 4023 ierr = PetscMalloc(nz_max*sizeof(PetscScalar), &values);CHKERRQ(ierr); 4024 ierr = PetscMemzero(values, nz_max*sizeof(PetscScalar));CHKERRQ(ierr); 4025 } 4026 4027 for (i = 0; i < m; i++) { 4028 nz = Ii[i+1] - Ii[i]; 4029 ierr = MatSetValues_SeqAIJ(B, 1, &i, nz, J+Ii[i], values + (v ? Ii[i] : 0), INSERT_VALUES);CHKERRQ(ierr); 4030 } 4031 4032 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4033 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4034 4035 if (!v) { 4036 ierr = PetscFree(values);CHKERRQ(ierr); 4037 } 4038 ierr = MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr); 4039 PetscFunctionReturn(0); 4040 } 4041 4042 #include <../src/mat/impls/dense/seq/dense.h> 4043 #include <petsc-private/kernels/petscaxpy.h> 4044 4045 #undef __FUNCT__ 4046 #define __FUNCT__ "MatMatMultNumeric_SeqDense_SeqAIJ" 4047 /* 4048 Computes (B'*A')' since computing B*A directly is untenable 4049 4050 n p p 4051 ( ) ( ) ( ) 4052 m ( A ) * n ( B ) = m ( C ) 4053 ( ) ( ) ( ) 4054 4055 */ 4056 PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C) 4057 { 4058 PetscErrorCode ierr; 4059 Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data; 4060 Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data; 4061 Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data; 4062 PetscInt i,n,m,q,p; 4063 const PetscInt *ii,*idx; 4064 const PetscScalar *b,*a,*a_q; 4065 PetscScalar *c,*c_q; 4066 4067 PetscFunctionBegin; 4068 m = A->rmap->n; 4069 n = A->cmap->n; 4070 p = B->cmap->n; 4071 a = sub_a->v; 4072 b = sub_b->a; 4073 c = sub_c->v; 4074 ierr = PetscMemzero(c,m*p*sizeof(PetscScalar));CHKERRQ(ierr); 4075 4076 ii = sub_b->i; 4077 idx = sub_b->j; 4078 for (i=0; i<n; i++) { 4079 q = ii[i+1] - ii[i]; 4080 while (q-->0) { 4081 c_q = c + m*(*idx); 4082 a_q = a + m*i; 4083 PetscKernelAXPY(c_q,*b,a_q,m); 4084 idx++; 4085 b++; 4086 } 4087 } 4088 PetscFunctionReturn(0); 4089 } 4090 4091 #undef __FUNCT__ 4092 #define __FUNCT__ "MatMatMultSymbolic_SeqDense_SeqAIJ" 4093 PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C) 4094 { 4095 PetscErrorCode ierr; 4096 PetscInt m=A->rmap->n,n=B->cmap->n; 4097 Mat Cmat; 4098 4099 PetscFunctionBegin; 4100 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); 4101 ierr = MatCreate(PetscObjectComm((PetscObject)A),&Cmat);CHKERRQ(ierr); 4102 ierr = MatSetSizes(Cmat,m,n,m,n);CHKERRQ(ierr); 4103 ierr = MatSetBlockSizes(Cmat,A->rmap->bs,B->cmap->bs);CHKERRQ(ierr); 4104 ierr = MatSetType(Cmat,MATSEQDENSE);CHKERRQ(ierr); 4105 ierr = MatSeqDenseSetPreallocation(Cmat,NULL);CHKERRQ(ierr); 4106 4107 Cmat->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ; 4108 4109 *C = Cmat; 4110 PetscFunctionReturn(0); 4111 } 4112 4113 /* ----------------------------------------------------------------*/ 4114 #undef __FUNCT__ 4115 #define __FUNCT__ "MatMatMult_SeqDense_SeqAIJ" 4116 PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C) 4117 { 4118 PetscErrorCode ierr; 4119 4120 PetscFunctionBegin; 4121 if (scall == MAT_INITIAL_MATRIX) { 4122 ierr = PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 4123 ierr = MatMatMultSymbolic_SeqDense_SeqAIJ(A,B,fill,C);CHKERRQ(ierr); 4124 ierr = PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);CHKERRQ(ierr); 4125 } 4126 ierr = PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 4127 ierr = MatMatMultNumeric_SeqDense_SeqAIJ(A,B,*C);CHKERRQ(ierr); 4128 ierr = PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);CHKERRQ(ierr); 4129 PetscFunctionReturn(0); 4130 } 4131 4132 4133 /*MC 4134 MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices, 4135 based on compressed sparse row format. 4136 4137 Options Database Keys: 4138 . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions() 4139 4140 Level: beginner 4141 4142 .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType 4143 M*/ 4144 4145 /*MC 4146 MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices. 4147 4148 This matrix type is identical to MATSEQAIJ when constructed with a single process communicator, 4149 and MATMPIAIJ otherwise. As a result, for single process communicators, 4150 MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation is supported 4151 for communicators controlling multiple processes. It is recommended that you call both of 4152 the above preallocation routines for simplicity. 4153 4154 Options Database Keys: 4155 . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions() 4156 4157 Developer Notes: Subclasses include MATAIJCUSP, MATAIJPERM, MATAIJCRL, and also automatically switches over to use inodes when 4158 enough exist. 4159 4160 Level: beginner 4161 4162 .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ 4163 M*/ 4164 4165 /*MC 4166 MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices. 4167 4168 This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator, 4169 and MATMPIAIJCRL otherwise. As a result, for single process communicators, 4170 MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported 4171 for communicators controlling multiple processes. It is recommended that you call both of 4172 the above preallocation routines for simplicity. 4173 4174 Options Database Keys: 4175 . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions() 4176 4177 Level: beginner 4178 4179 .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL 4180 M*/ 4181 4182 #if defined(PETSC_HAVE_PASTIX) 4183 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_pastix(Mat,MatFactorType,Mat*); 4184 #endif 4185 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4186 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_essl(Mat,MatFactorType,Mat*); 4187 #endif 4188 PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*); 4189 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_petsc(Mat,MatFactorType,Mat*); 4190 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_bas(Mat,MatFactorType,Mat*); 4191 extern PetscErrorCode MatGetFactorAvailable_seqaij_petsc(Mat,MatFactorType,PetscBool*); 4192 #if defined(PETSC_HAVE_MUMPS) 4193 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mumps(Mat,MatFactorType,Mat*); 4194 #endif 4195 #if defined(PETSC_HAVE_SUPERLU) 4196 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat,MatFactorType,Mat*); 4197 #endif 4198 #if defined(PETSC_HAVE_SUPERLU_DIST) 4199 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu_dist(Mat,MatFactorType,Mat*); 4200 #endif 4201 #if defined(PETSC_HAVE_UMFPACK) 4202 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_umfpack(Mat,MatFactorType,Mat*); 4203 #endif 4204 #if defined(PETSC_HAVE_CHOLMOD) 4205 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat,MatFactorType,Mat*); 4206 #endif 4207 #if defined(PETSC_HAVE_LUSOL) 4208 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_lusol(Mat,MatFactorType,Mat*); 4209 #endif 4210 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4211 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_matlab(Mat,MatFactorType,Mat*); 4212 extern PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*); 4213 extern PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*); 4214 #endif 4215 #if defined(PETSC_HAVE_CLIQUE) 4216 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_clique(Mat,MatFactorType,Mat*); 4217 #endif 4218 4219 4220 #undef __FUNCT__ 4221 #define __FUNCT__ "MatSeqAIJGetArray" 4222 /*@C 4223 MatSeqAIJGetArray - gives access to the array where the data for a SeqSeqAIJ matrix is stored 4224 4225 Not Collective 4226 4227 Input Parameter: 4228 . mat - a MATSEQDENSE matrix 4229 4230 Output Parameter: 4231 . array - pointer to the data 4232 4233 Level: intermediate 4234 4235 .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90() 4236 @*/ 4237 PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array) 4238 { 4239 PetscErrorCode ierr; 4240 4241 PetscFunctionBegin; 4242 ierr = PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4243 PetscFunctionReturn(0); 4244 } 4245 4246 #undef __FUNCT__ 4247 #define __FUNCT__ "MatSeqAIJRestoreArray" 4248 /*@C 4249 MatSeqAIJRestoreArray - returns access to the array where the data for a SeqSeqAIJ matrix is stored obtained by MatSeqAIJGetArray() 4250 4251 Not Collective 4252 4253 Input Parameters: 4254 . mat - a MATSEQDENSE matrix 4255 . array - pointer to the data 4256 4257 Level: intermediate 4258 4259 .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90() 4260 @*/ 4261 PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array) 4262 { 4263 PetscErrorCode ierr; 4264 4265 PetscFunctionBegin; 4266 ierr = PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));CHKERRQ(ierr); 4267 PetscFunctionReturn(0); 4268 } 4269 4270 #undef __FUNCT__ 4271 #define __FUNCT__ "MatCreate_SeqAIJ" 4272 PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B) 4273 { 4274 Mat_SeqAIJ *b; 4275 PetscErrorCode ierr; 4276 PetscMPIInt size; 4277 4278 PetscFunctionBegin; 4279 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);CHKERRQ(ierr); 4280 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1"); 4281 4282 ierr = PetscNewLog(B,Mat_SeqAIJ,&b);CHKERRQ(ierr); 4283 4284 B->data = (void*)b; 4285 4286 ierr = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr); 4287 4288 b->row = 0; 4289 b->col = 0; 4290 b->icol = 0; 4291 b->reallocs = 0; 4292 b->ignorezeroentries = PETSC_FALSE; 4293 b->roworiented = PETSC_TRUE; 4294 b->nonew = 0; 4295 b->diag = 0; 4296 b->solve_work = 0; 4297 B->spptr = 0; 4298 b->saved_values = 0; 4299 b->idiag = 0; 4300 b->mdiag = 0; 4301 b->ssor_work = 0; 4302 b->omega = 1.0; 4303 b->fshift = 0.0; 4304 b->idiagvalid = PETSC_FALSE; 4305 b->ibdiagvalid = PETSC_FALSE; 4306 b->keepnonzeropattern = PETSC_FALSE; 4307 b->xtoy = 0; 4308 b->XtoY = 0; 4309 B->same_nonzero = PETSC_FALSE; 4310 4311 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4312 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);CHKERRQ(ierr); 4313 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);CHKERRQ(ierr); 4314 4315 #if defined(PETSC_HAVE_MATLAB_ENGINE) 4316 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_matlab_C",MatGetFactor_seqaij_matlab);CHKERRQ(ierr); 4317 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);CHKERRQ(ierr); 4318 ierr = PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);CHKERRQ(ierr); 4319 #endif 4320 #if defined(PETSC_HAVE_PASTIX) 4321 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_pastix_C",MatGetFactor_seqaij_pastix);CHKERRQ(ierr); 4322 #endif 4323 #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128) 4324 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_essl_C",MatGetFactor_seqaij_essl);CHKERRQ(ierr); 4325 #endif 4326 #if defined(PETSC_HAVE_SUPERLU) 4327 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_C",MatGetFactor_seqaij_superlu);CHKERRQ(ierr); 4328 #endif 4329 #if defined(PETSC_HAVE_SUPERLU_DIST) 4330 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_dist_C",MatGetFactor_seqaij_superlu_dist);CHKERRQ(ierr); 4331 #endif 4332 #if defined(PETSC_HAVE_MUMPS) 4333 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mumps_C",MatGetFactor_aij_mumps);CHKERRQ(ierr); 4334 #endif 4335 #if defined(PETSC_HAVE_UMFPACK) 4336 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_umfpack_C",MatGetFactor_seqaij_umfpack);CHKERRQ(ierr); 4337 #endif 4338 #if defined(PETSC_HAVE_CHOLMOD) 4339 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_cholmod_C",MatGetFactor_seqaij_cholmod);CHKERRQ(ierr); 4340 #endif 4341 #if defined(PETSC_HAVE_LUSOL) 4342 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_lusol_C",MatGetFactor_seqaij_lusol);CHKERRQ(ierr); 4343 #endif 4344 #if defined(PETSC_HAVE_CLIQUE) 4345 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_clique_C",MatGetFactor_aij_clique);CHKERRQ(ierr); 4346 #endif 4347 4348 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_petsc_C",MatGetFactor_seqaij_petsc);CHKERRQ(ierr); 4349 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactorAvailable_petsc_C",MatGetFactorAvailable_seqaij_petsc);CHKERRQ(ierr); 4350 ierr = PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_bas_C",MatGetFactor_seqaij_bas);CHKERRQ(ierr); 4351 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);CHKERRQ(ierr); 4352 ierr = PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);CHKERRQ(ierr); 4353 ierr = PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);CHKERRQ(ierr); 4354 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);CHKERRQ(ierr); 4355 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);CHKERRQ(ierr); 4356 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);CHKERRQ(ierr); 4357 ierr = PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);CHKERRQ(ierr); 4358 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4359 ierr = PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);CHKERRQ(ierr); 4360 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);CHKERRQ(ierr); 4361 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);CHKERRQ(ierr); 4362 ierr = PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);CHKERRQ(ierr); 4363 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMult_seqdense_seqaij_C",MatMatMult_SeqDense_SeqAIJ);CHKERRQ(ierr); 4364 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_seqdense_seqaij_C",MatMatMultSymbolic_SeqDense_SeqAIJ);CHKERRQ(ierr); 4365 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_seqdense_seqaij_C",MatMatMultNumeric_SeqDense_SeqAIJ);CHKERRQ(ierr); 4366 ierr = MatCreate_SeqAIJ_Inode(B);CHKERRQ(ierr); 4367 ierr = PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);CHKERRQ(ierr); 4368 PetscFunctionReturn(0); 4369 } 4370 4371 #undef __FUNCT__ 4372 #define __FUNCT__ "MatDuplicateNoCreate_SeqAIJ" 4373 /* 4374 Given a matrix generated with MatGetFactor() duplicates all the information in A into B 4375 */ 4376 PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace) 4377 { 4378 Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data; 4379 PetscErrorCode ierr; 4380 PetscInt i,m = A->rmap->n; 4381 4382 PetscFunctionBegin; 4383 c = (Mat_SeqAIJ*)C->data; 4384 4385 C->factortype = A->factortype; 4386 c->row = 0; 4387 c->col = 0; 4388 c->icol = 0; 4389 c->reallocs = 0; 4390 4391 C->assembled = PETSC_TRUE; 4392 4393 ierr = PetscLayoutReference(A->rmap,&C->rmap);CHKERRQ(ierr); 4394 ierr = PetscLayoutReference(A->cmap,&C->cmap);CHKERRQ(ierr); 4395 4396 ierr = PetscMalloc2(m,PetscInt,&c->imax,m,PetscInt,&c->ilen);CHKERRQ(ierr); 4397 ierr = PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));CHKERRQ(ierr); 4398 for (i=0; i<m; i++) { 4399 c->imax[i] = a->imax[i]; 4400 c->ilen[i] = a->ilen[i]; 4401 } 4402 4403 /* allocate the matrix space */ 4404 if (mallocmatspace) { 4405 ierr = PetscMalloc3(a->i[m],PetscScalar,&c->a,a->i[m],PetscInt,&c->j,m+1,PetscInt,&c->i);CHKERRQ(ierr); 4406 ierr = PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4407 4408 c->singlemalloc = PETSC_TRUE; 4409 4410 ierr = PetscMemcpy(c->i,a->i,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4411 if (m > 0) { 4412 ierr = PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(PetscInt));CHKERRQ(ierr); 4413 if (cpvalues == MAT_COPY_VALUES) { 4414 ierr = PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4415 } else { 4416 ierr = PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));CHKERRQ(ierr); 4417 } 4418 } 4419 } 4420 4421 c->ignorezeroentries = a->ignorezeroentries; 4422 c->roworiented = a->roworiented; 4423 c->nonew = a->nonew; 4424 if (a->diag) { 4425 ierr = PetscMalloc((m+1)*sizeof(PetscInt),&c->diag);CHKERRQ(ierr); 4426 ierr = PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));CHKERRQ(ierr); 4427 for (i=0; i<m; i++) { 4428 c->diag[i] = a->diag[i]; 4429 } 4430 } else c->diag = 0; 4431 4432 c->solve_work = 0; 4433 c->saved_values = 0; 4434 c->idiag = 0; 4435 c->ssor_work = 0; 4436 c->keepnonzeropattern = a->keepnonzeropattern; 4437 c->free_a = PETSC_TRUE; 4438 c->free_ij = PETSC_TRUE; 4439 c->xtoy = 0; 4440 c->XtoY = 0; 4441 4442 c->rmax = a->rmax; 4443 c->nz = a->nz; 4444 c->maxnz = a->nz; /* Since we allocate exactly the right amount */ 4445 C->preallocated = PETSC_TRUE; 4446 4447 c->compressedrow.use = a->compressedrow.use; 4448 c->compressedrow.nrows = a->compressedrow.nrows; 4449 c->compressedrow.check = a->compressedrow.check; 4450 if (a->compressedrow.use) { 4451 i = a->compressedrow.nrows; 4452 ierr = PetscMalloc2(i+1,PetscInt,&c->compressedrow.i,i,PetscInt,&c->compressedrow.rindex);CHKERRQ(ierr); 4453 ierr = PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));CHKERRQ(ierr); 4454 ierr = PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));CHKERRQ(ierr); 4455 } else { 4456 c->compressedrow.use = PETSC_FALSE; 4457 c->compressedrow.i = NULL; 4458 c->compressedrow.rindex = NULL; 4459 } 4460 C->same_nonzero = A->same_nonzero; 4461 4462 ierr = MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);CHKERRQ(ierr); 4463 ierr = PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);CHKERRQ(ierr); 4464 PetscFunctionReturn(0); 4465 } 4466 4467 #undef __FUNCT__ 4468 #define __FUNCT__ "MatDuplicate_SeqAIJ" 4469 PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B) 4470 { 4471 PetscErrorCode ierr; 4472 4473 PetscFunctionBegin; 4474 ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr); 4475 ierr = MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);CHKERRQ(ierr); 4476 ierr = MatSetBlockSizes(*B,A->rmap->bs,A->cmap->bs);CHKERRQ(ierr); 4477 ierr = MatSetType(*B,((PetscObject)A)->type_name);CHKERRQ(ierr); 4478 ierr = MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);CHKERRQ(ierr); 4479 PetscFunctionReturn(0); 4480 } 4481 4482 #undef __FUNCT__ 4483 #define __FUNCT__ "MatLoad_SeqAIJ" 4484 PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer) 4485 { 4486 Mat_SeqAIJ *a; 4487 PetscErrorCode ierr; 4488 PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N,rows,cols; 4489 int fd; 4490 PetscMPIInt size; 4491 MPI_Comm comm; 4492 PetscInt bs = 1; 4493 4494 PetscFunctionBegin; 4495 ierr = PetscObjectGetComm((PetscObject)viewer,&comm);CHKERRQ(ierr); 4496 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 4497 if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"view must have one processor"); 4498 4499 ierr = PetscOptionsBegin(comm,NULL,"Options for loading SEQAIJ matrix","Mat");CHKERRQ(ierr); 4500 ierr = PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);CHKERRQ(ierr); 4501 ierr = PetscOptionsEnd();CHKERRQ(ierr); 4502 if (bs > 1) {ierr = MatSetBlockSize(newMat,bs);CHKERRQ(ierr);} 4503 4504 ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr); 4505 ierr = PetscBinaryRead(fd,header,4,PETSC_INT);CHKERRQ(ierr); 4506 if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file"); 4507 M = header[1]; N = header[2]; nz = header[3]; 4508 4509 if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ"); 4510 4511 /* read in row lengths */ 4512 ierr = PetscMalloc(M*sizeof(PetscInt),&rowlengths);CHKERRQ(ierr); 4513 ierr = PetscBinaryRead(fd,rowlengths,M,PETSC_INT);CHKERRQ(ierr); 4514 4515 /* check if sum of rowlengths is same as nz */ 4516 for (i=0,sum=0; i< M; i++) sum +=rowlengths[i]; 4517 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); 4518 4519 /* set global size if not set already*/ 4520 if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) { 4521 ierr = MatSetSizes(newMat,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr); 4522 } else { 4523 /* if sizes and type are already set, check if the vector global sizes are correct */ 4524 ierr = MatGetSize(newMat,&rows,&cols);CHKERRQ(ierr); 4525 if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */ 4526 ierr = MatGetLocalSize(newMat,&rows,&cols);CHKERRQ(ierr); 4527 } 4528 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); 4529 } 4530 ierr = MatSeqAIJSetPreallocation_SeqAIJ(newMat,0,rowlengths);CHKERRQ(ierr); 4531 a = (Mat_SeqAIJ*)newMat->data; 4532 4533 ierr = PetscBinaryRead(fd,a->j,nz,PETSC_INT);CHKERRQ(ierr); 4534 4535 /* read in nonzero values */ 4536 ierr = PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);CHKERRQ(ierr); 4537 4538 /* set matrix "i" values */ 4539 a->i[0] = 0; 4540 for (i=1; i<= M; i++) { 4541 a->i[i] = a->i[i-1] + rowlengths[i-1]; 4542 a->ilen[i-1] = rowlengths[i-1]; 4543 } 4544 ierr = PetscFree(rowlengths);CHKERRQ(ierr); 4545 4546 ierr = MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4547 ierr = MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4548 PetscFunctionReturn(0); 4549 } 4550 4551 #undef __FUNCT__ 4552 #define __FUNCT__ "MatEqual_SeqAIJ" 4553 PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg) 4554 { 4555 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data; 4556 PetscErrorCode ierr; 4557 #if defined(PETSC_USE_COMPLEX) 4558 PetscInt k; 4559 #endif 4560 4561 PetscFunctionBegin; 4562 /* If the matrix dimensions are not equal,or no of nonzeros */ 4563 if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) { 4564 *flg = PETSC_FALSE; 4565 PetscFunctionReturn(0); 4566 } 4567 4568 /* if the a->i are the same */ 4569 ierr = PetscMemcmp(a->i,b->i,(A->rmap->n+1)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4570 if (!*flg) PetscFunctionReturn(0); 4571 4572 /* if a->j are the same */ 4573 ierr = PetscMemcmp(a->j,b->j,(a->nz)*sizeof(PetscInt),flg);CHKERRQ(ierr); 4574 if (!*flg) PetscFunctionReturn(0); 4575 4576 /* if a->a are the same */ 4577 #if defined(PETSC_USE_COMPLEX) 4578 for (k=0; k<a->nz; k++) { 4579 if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) { 4580 *flg = PETSC_FALSE; 4581 PetscFunctionReturn(0); 4582 } 4583 } 4584 #else 4585 ierr = PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);CHKERRQ(ierr); 4586 #endif 4587 PetscFunctionReturn(0); 4588 } 4589 4590 #undef __FUNCT__ 4591 #define __FUNCT__ "MatCreateSeqAIJWithArrays" 4592 /*@ 4593 MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format) 4594 provided by the user. 4595 4596 Collective on MPI_Comm 4597 4598 Input Parameters: 4599 + comm - must be an MPI communicator of size 1 4600 . m - number of rows 4601 . n - number of columns 4602 . i - row indices 4603 . j - column indices 4604 - a - matrix values 4605 4606 Output Parameter: 4607 . mat - the matrix 4608 4609 Level: intermediate 4610 4611 Notes: 4612 The i, j, and a arrays are not copied by this routine, the user must free these arrays 4613 once the matrix is destroyed and not before 4614 4615 You cannot set new nonzero locations into this matrix, that will generate an error. 4616 4617 The i and j indices are 0 based 4618 4619 The format which is used for the sparse matrix input, is equivalent to a 4620 row-major ordering.. i.e for the following matrix, the input data expected is 4621 as shown: 4622 4623 1 0 0 4624 2 0 3 4625 4 5 6 4626 4627 i = {0,1,3,6} [size = nrow+1 = 3+1] 4628 j = {0,0,2,0,1,2} [size = nz = 6]; values must be sorted for each row 4629 v = {1,2,3,4,5,6} [size = nz = 6] 4630 4631 4632 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4633 4634 @*/ 4635 PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat) 4636 { 4637 PetscErrorCode ierr; 4638 PetscInt ii; 4639 Mat_SeqAIJ *aij; 4640 #if defined(PETSC_USE_DEBUG) 4641 PetscInt jj; 4642 #endif 4643 4644 PetscFunctionBegin; 4645 if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0"); 4646 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4647 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4648 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4649 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4650 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);CHKERRQ(ierr); 4651 aij = (Mat_SeqAIJ*)(*mat)->data; 4652 ierr = PetscMalloc2(m,PetscInt,&aij->imax,m,PetscInt,&aij->ilen);CHKERRQ(ierr); 4653 4654 aij->i = i; 4655 aij->j = j; 4656 aij->a = a; 4657 aij->singlemalloc = PETSC_FALSE; 4658 aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/ 4659 aij->free_a = PETSC_FALSE; 4660 aij->free_ij = PETSC_FALSE; 4661 4662 for (ii=0; ii<m; ii++) { 4663 aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii]; 4664 #if defined(PETSC_USE_DEBUG) 4665 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]); 4666 for (jj=i[ii]+1; jj<i[ii+1]; jj++) { 4667 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); 4668 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); 4669 } 4670 #endif 4671 } 4672 #if defined(PETSC_USE_DEBUG) 4673 for (ii=0; ii<aij->i[m]; ii++) { 4674 if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %d index = %d",ii,j[ii]); 4675 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]); 4676 } 4677 #endif 4678 4679 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4680 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4681 PetscFunctionReturn(0); 4682 } 4683 #undef __FUNCT__ 4684 #define __FUNCT__ "MatCreateSeqAIJFromTriple" 4685 /*@C 4686 MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format) 4687 provided by the user. 4688 4689 Collective on MPI_Comm 4690 4691 Input Parameters: 4692 + comm - must be an MPI communicator of size 1 4693 . m - number of rows 4694 . n - number of columns 4695 . i - row indices 4696 . j - column indices 4697 . a - matrix values 4698 . nz - number of nonzeros 4699 - idx - 0 or 1 based 4700 4701 Output Parameter: 4702 . mat - the matrix 4703 4704 Level: intermediate 4705 4706 Notes: 4707 The i and j indices are 0 based 4708 4709 The format which is used for the sparse matrix input, is equivalent to a 4710 row-major ordering.. i.e for the following matrix, the input data expected is 4711 as shown: 4712 4713 1 0 0 4714 2 0 3 4715 4 5 6 4716 4717 i = {0,1,1,2,2,2} 4718 j = {0,0,2,0,1,2} 4719 v = {1,2,3,4,5,6} 4720 4721 4722 .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR() 4723 4724 @*/ 4725 PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat,PetscInt nz,PetscBool idx) 4726 { 4727 PetscErrorCode ierr; 4728 PetscInt ii, *nnz, one = 1,row,col; 4729 4730 4731 PetscFunctionBegin; 4732 ierr = PetscMalloc(m*sizeof(PetscInt),&nnz);CHKERRQ(ierr); 4733 ierr = PetscMemzero(nnz,m*sizeof(PetscInt));CHKERRQ(ierr); 4734 for (ii = 0; ii < nz; ii++) { 4735 nnz[i[ii]] += 1; 4736 } 4737 ierr = MatCreate(comm,mat);CHKERRQ(ierr); 4738 ierr = MatSetSizes(*mat,m,n,m,n);CHKERRQ(ierr); 4739 /* ierr = MatSetBlockSizes(*mat,,);CHKERRQ(ierr); */ 4740 ierr = MatSetType(*mat,MATSEQAIJ);CHKERRQ(ierr); 4741 ierr = MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);CHKERRQ(ierr); 4742 for (ii = 0; ii < nz; ii++) { 4743 if (idx) { 4744 row = i[ii] - 1; 4745 col = j[ii] - 1; 4746 } else { 4747 row = i[ii]; 4748 col = j[ii]; 4749 } 4750 ierr = MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);CHKERRQ(ierr); 4751 } 4752 ierr = MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4753 ierr = MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 4754 ierr = PetscFree(nnz);CHKERRQ(ierr); 4755 PetscFunctionReturn(0); 4756 } 4757 4758 #undef __FUNCT__ 4759 #define __FUNCT__ "MatSetColoring_SeqAIJ" 4760 PetscErrorCode MatSetColoring_SeqAIJ(Mat A,ISColoring coloring) 4761 { 4762 PetscErrorCode ierr; 4763 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4764 4765 PetscFunctionBegin; 4766 if (coloring->ctype == IS_COLORING_GLOBAL) { 4767 ierr = ISColoringReference(coloring);CHKERRQ(ierr); 4768 a->coloring = coloring; 4769 } else if (coloring->ctype == IS_COLORING_GHOSTED) { 4770 PetscInt i,*larray; 4771 ISColoring ocoloring; 4772 ISColoringValue *colors; 4773 4774 /* set coloring for diagonal portion */ 4775 ierr = PetscMalloc(A->cmap->n*sizeof(PetscInt),&larray);CHKERRQ(ierr); 4776 for (i=0; i<A->cmap->n; i++) larray[i] = i; 4777 ierr = ISGlobalToLocalMappingApply(A->cmap->mapping,IS_GTOLM_MASK,A->cmap->n,larray,NULL,larray);CHKERRQ(ierr); 4778 ierr = PetscMalloc(A->cmap->n*sizeof(ISColoringValue),&colors);CHKERRQ(ierr); 4779 for (i=0; i<A->cmap->n; i++) colors[i] = coloring->colors[larray[i]]; 4780 ierr = PetscFree(larray);CHKERRQ(ierr); 4781 ierr = ISColoringCreate(PETSC_COMM_SELF,coloring->n,A->cmap->n,colors,&ocoloring);CHKERRQ(ierr); 4782 a->coloring = ocoloring; 4783 } 4784 PetscFunctionReturn(0); 4785 } 4786 4787 #undef __FUNCT__ 4788 #define __FUNCT__ "MatSetValuesAdifor_SeqAIJ" 4789 PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat A,PetscInt nl,void *advalues) 4790 { 4791 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4792 PetscInt m = A->rmap->n,*ii = a->i,*jj = a->j,nz,i,j; 4793 MatScalar *v = a->a; 4794 PetscScalar *values = (PetscScalar*)advalues; 4795 ISColoringValue *color; 4796 4797 PetscFunctionBegin; 4798 if (!a->coloring) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix"); 4799 color = a->coloring->colors; 4800 /* loop over rows */ 4801 for (i=0; i<m; i++) { 4802 nz = ii[i+1] - ii[i]; 4803 /* loop over columns putting computed value into matrix */ 4804 for (j=0; j<nz; j++) *v++ = values[color[*jj++]]; 4805 values += nl; /* jump to next row of derivatives */ 4806 } 4807 PetscFunctionReturn(0); 4808 } 4809 4810 #undef __FUNCT__ 4811 #define __FUNCT__ "MatSeqAIJInvalidateDiagonal" 4812 PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A) 4813 { 4814 Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data; 4815 PetscErrorCode ierr; 4816 4817 PetscFunctionBegin; 4818 a->idiagvalid = PETSC_FALSE; 4819 a->ibdiagvalid = PETSC_FALSE; 4820 4821 ierr = MatSeqAIJInvalidateDiagonal_Inode(A);CHKERRQ(ierr); 4822 PetscFunctionReturn(0); 4823 } 4824 4825 /* 4826 Special version for direct calls from Fortran 4827 */ 4828 #include <petsc-private/fortranimpl.h> 4829 #if defined(PETSC_HAVE_FORTRAN_CAPS) 4830 #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ 4831 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE) 4832 #define matsetvaluesseqaij_ matsetvaluesseqaij 4833 #endif 4834 4835 /* Change these macros so can be used in void function */ 4836 #undef CHKERRQ 4837 #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr) 4838 #undef SETERRQ2 4839 #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr) 4840 #undef SETERRQ3 4841 #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr) 4842 4843 #undef __FUNCT__ 4844 #define __FUNCT__ "matsetvaluesseqaij_" 4845 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) 4846 { 4847 Mat A = *AA; 4848 PetscInt m = *mm, n = *nn; 4849 InsertMode is = *isis; 4850 Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data; 4851 PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N; 4852 PetscInt *imax,*ai,*ailen; 4853 PetscErrorCode ierr; 4854 PetscInt *aj,nonew = a->nonew,lastcol = -1; 4855 MatScalar *ap,value,*aa; 4856 PetscBool ignorezeroentries = a->ignorezeroentries; 4857 PetscBool roworiented = a->roworiented; 4858 4859 PetscFunctionBegin; 4860 MatCheckPreallocated(A,1); 4861 imax = a->imax; 4862 ai = a->i; 4863 ailen = a->ilen; 4864 aj = a->j; 4865 aa = a->a; 4866 4867 for (k=0; k<m; k++) { /* loop over added rows */ 4868 row = im[k]; 4869 if (row < 0) continue; 4870 #if defined(PETSC_USE_DEBUG) 4871 if (row >= A->rmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large"); 4872 #endif 4873 rp = aj + ai[row]; ap = aa + ai[row]; 4874 rmax = imax[row]; nrow = ailen[row]; 4875 low = 0; 4876 high = nrow; 4877 for (l=0; l<n; l++) { /* loop over added columns */ 4878 if (in[l] < 0) continue; 4879 #if defined(PETSC_USE_DEBUG) 4880 if (in[l] >= A->cmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large"); 4881 #endif 4882 col = in[l]; 4883 if (roworiented) value = v[l + k*n]; 4884 else value = v[k + l*m]; 4885 4886 if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue; 4887 4888 if (col <= lastcol) low = 0; 4889 else high = nrow; 4890 lastcol = col; 4891 while (high-low > 5) { 4892 t = (low+high)/2; 4893 if (rp[t] > col) high = t; 4894 else low = t; 4895 } 4896 for (i=low; i<high; i++) { 4897 if (rp[i] > col) break; 4898 if (rp[i] == col) { 4899 if (is == ADD_VALUES) ap[i] += value; 4900 else ap[i] = value; 4901 goto noinsert; 4902 } 4903 } 4904 if (value == 0.0 && ignorezeroentries) goto noinsert; 4905 if (nonew == 1) goto noinsert; 4906 if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); 4907 MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar); 4908 N = nrow++ - 1; a->nz++; high++; 4909 /* shift up all the later entries in this row */ 4910 for (ii=N; ii>=i; ii--) { 4911 rp[ii+1] = rp[ii]; 4912 ap[ii+1] = ap[ii]; 4913 } 4914 rp[i] = col; 4915 ap[i] = value; 4916 noinsert:; 4917 low = i + 1; 4918 } 4919 ailen[row] = nrow; 4920 } 4921 A->same_nonzero = PETSC_FALSE; 4922 PetscFunctionReturnVoid(); 4923 } 4924 4925 4926