1 #define PETSCMAT_DLL 2 3 #include "../src/mat/impls/baij/seq/baij.h" 4 #include "../src/mat/impls/sbaij/seq/sbaij.h" 5 #include "../src/mat/blockinvert.h" 6 #include "petscis.h" 7 8 #if !defined(PETSC_USE_COMPLEX) 9 /* 10 input: 11 F -- numeric factor 12 output: 13 nneg, nzero, npos: matrix inertia 14 */ 15 16 #undef __FUNCT__ 17 #define __FUNCT__ "MatGetInertia_SeqSBAIJ" 18 PetscErrorCode MatGetInertia_SeqSBAIJ(Mat F,PetscInt *nneig,PetscInt *nzero,PetscInt *npos) 19 { 20 Mat_SeqSBAIJ *fact_ptr = (Mat_SeqSBAIJ*)F->data; 21 MatScalar *dd = fact_ptr->a; 22 PetscInt mbs=fact_ptr->mbs,bs=F->rmap->bs,i,nneig_tmp,npos_tmp,*fi = fact_ptr->i; 23 24 PetscFunctionBegin; 25 if (bs != 1) SETERRQ1(PETSC_ERR_SUP,"No support for bs: %D >1 yet",bs); 26 nneig_tmp = 0; npos_tmp = 0; 27 for (i=0; i<mbs; i++){ 28 if (PetscRealPart(dd[*fi]) > 0.0){ 29 npos_tmp++; 30 } else if (PetscRealPart(dd[*fi]) < 0.0){ 31 nneig_tmp++; 32 } 33 fi++; 34 } 35 if (nneig) *nneig = nneig_tmp; 36 if (npos) *npos = npos_tmp; 37 if (nzero) *nzero = mbs - nneig_tmp - npos_tmp; 38 39 PetscFunctionReturn(0); 40 } 41 #endif /* !defined(PETSC_USE_COMPLEX) */ 42 43 /* 44 Symbolic U^T*D*U factorization for SBAIJ format. Modified from SSF of YSMP. 45 Use Modified Sparse Row (MSR) storage for u and ju. See page 85, "Iterative Methods ..." by Saad. 46 */ 47 #undef __FUNCT__ 48 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_MSR" 49 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(Mat F,Mat A,IS perm,const MatFactorInfo *info) 50 { 51 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 52 PetscErrorCode ierr; 53 const PetscInt *rip,*ai,*aj; 54 PetscInt i,mbs = a->mbs,*jutmp,bs = A->rmap->bs,bs2=a->bs2; 55 PetscInt m,reallocs = 0,prow; 56 PetscInt *jl,*q,jmin,jmax,juidx,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 57 PetscReal f = info->fill; 58 PetscTruth perm_identity; 59 60 PetscFunctionBegin; 61 /* check whether perm is the identity mapping */ 62 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 63 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 64 65 if (perm_identity){ /* without permutation */ 66 a->permute = PETSC_FALSE; 67 ai = a->i; aj = a->j; 68 } else { /* non-trivial permutation */ 69 a->permute = PETSC_TRUE; 70 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 71 ai = a->inew; aj = a->jnew; 72 } 73 74 /* initialization */ 75 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&iu);CHKERRQ(ierr); 76 umax = (PetscInt)(f*ai[mbs] + 1); umax += mbs + 1; 77 ierr = PetscMalloc(umax*sizeof(PetscInt),&ju);CHKERRQ(ierr); 78 iu[0] = mbs+1; 79 juidx = mbs + 1; /* index for ju */ 80 /* jl linked list for pivot row -- linked list for col index */ 81 ierr = PetscMalloc2(mbs,PetscInt,&jl,mbs,PetscInt,&q);CHKERRQ(ierr); 82 for (i=0; i<mbs; i++){ 83 jl[i] = mbs; 84 q[i] = 0; 85 } 86 87 /* for each row k */ 88 for (k=0; k<mbs; k++){ 89 for (i=0; i<mbs; i++) q[i] = 0; /* to be removed! */ 90 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 91 q[k] = mbs; 92 /* initialize nonzero structure of k-th row to row rip[k] of A */ 93 jmin = ai[rip[k]] +1; /* exclude diag[k] */ 94 jmax = ai[rip[k]+1]; 95 for (j=jmin; j<jmax; j++){ 96 vj = rip[aj[j]]; /* col. value */ 97 if(vj > k){ 98 qm = k; 99 do { 100 m = qm; qm = q[m]; 101 } while(qm < vj); 102 if (qm == vj) { 103 SETERRQ(PETSC_ERR_PLIB,"Duplicate entry in A\n"); 104 } 105 nzk++; 106 q[m] = vj; 107 q[vj] = qm; 108 } /* if(vj > k) */ 109 } /* for (j=jmin; j<jmax; j++) */ 110 111 /* modify nonzero structure of k-th row by computing fill-in 112 for each row i to be merged in */ 113 prow = k; 114 prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */ 115 116 while (prow < k){ 117 /* merge row prow into k-th row */ 118 jmin = iu[prow] + 1; jmax = iu[prow+1]; 119 qm = k; 120 for (j=jmin; j<jmax; j++){ 121 vj = ju[j]; 122 do { 123 m = qm; qm = q[m]; 124 } while (qm < vj); 125 if (qm != vj){ 126 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 127 } 128 } 129 prow = jl[prow]; /* next pivot row */ 130 } 131 132 /* add k to row list for first nonzero element in k-th row */ 133 if (nzk > 0){ 134 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 135 jl[k] = jl[i]; jl[i] = k; 136 } 137 iu[k+1] = iu[k] + nzk; 138 139 /* allocate more space to ju if needed */ 140 if (iu[k+1] > umax) { 141 /* estimate how much additional space we will need */ 142 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 143 /* just double the memory each time */ 144 maxadd = umax; 145 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 146 umax += maxadd; 147 148 /* allocate a longer ju */ 149 ierr = PetscMalloc(umax*sizeof(PetscInt),&jutmp);CHKERRQ(ierr); 150 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr); 151 ierr = PetscFree(ju);CHKERRQ(ierr); 152 ju = jutmp; 153 reallocs++; /* count how many times we realloc */ 154 } 155 156 /* save nonzero structure of k-th row in ju */ 157 i=k; 158 while (nzk --) { 159 i = q[i]; 160 ju[juidx++] = i; 161 } 162 } 163 164 #if defined(PETSC_USE_INFO) 165 if (ai[mbs] != 0) { 166 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 167 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,f,af);CHKERRQ(ierr); 168 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 169 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G);\n",af);CHKERRQ(ierr); 170 ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr); 171 } else { 172 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 173 } 174 #endif 175 176 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 177 ierr = PetscFree2(jl,q);CHKERRQ(ierr); 178 179 /* put together the new matrix */ 180 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(F,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 181 182 /* ierr = PetscLogObjectParent(B,iperm);CHKERRQ(ierr); */ 183 b = (Mat_SeqSBAIJ*)(F)->data; 184 b->singlemalloc = PETSC_FALSE; 185 b->free_a = PETSC_TRUE; 186 b->free_ij = PETSC_TRUE; 187 ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr); 188 b->j = ju; 189 b->i = iu; 190 b->diag = 0; 191 b->ilen = 0; 192 b->imax = 0; 193 b->row = perm; 194 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 195 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 196 b->icol = perm; 197 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 198 ierr = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 199 /* In b structure: Free imax, ilen, old a, old j. 200 Allocate idnew, solve_work, new a, new j */ 201 ierr = PetscLogObjectMemory(F,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 202 b->maxnz = b->nz = iu[mbs]; 203 204 (F)->info.factor_mallocs = reallocs; 205 (F)->info.fill_ratio_given = f; 206 if (ai[mbs] != 0) { 207 (F)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 208 } else { 209 (F)->info.fill_ratio_needed = 0.0; 210 } 211 ierr = MatSeqSBAIJSetNumericFactorization_inplace(F,perm_identity);CHKERRQ(ierr); 212 PetscFunctionReturn(0); 213 } 214 /* 215 Symbolic U^T*D*U factorization for SBAIJ format. 216 See MatICCFactorSymbolic_SeqAIJ() for description of its data structure. 217 */ 218 #include "petscbt.h" 219 #include "../src/mat/utils/freespace.h" 220 #undef __FUNCT__ 221 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 222 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 223 { 224 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 225 Mat_SeqSBAIJ *b; 226 PetscErrorCode ierr; 227 PetscTruth perm_identity,missing; 228 PetscReal fill = info->fill; 229 const PetscInt *rip,*ai=a->i,*aj=a->j; 230 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow,d; 231 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 232 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr,*udiag; 233 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 234 PetscBT lnkbt; 235 236 PetscFunctionBegin; 237 if (bs > 1){ 238 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(fact,A,perm,info);CHKERRQ(ierr); 239 PetscFunctionReturn(0); 240 } 241 if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n); 242 ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr); 243 if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d); 244 245 /* check whether perm is the identity mapping */ 246 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 247 if (!perm_identity) SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format"); 248 a->permute = PETSC_FALSE; 249 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 250 251 /* initialization */ 252 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 253 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&udiag);CHKERRQ(ierr); 254 ui[0] = 0; 255 256 /* jl: linked list for storing indices of the pivot rows 257 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 258 ierr = PetscMalloc4(mbs,PetscInt*,&ui_ptr,mbs,PetscInt,&il,mbs,PetscInt,&jl,mbs,PetscInt,&cols);CHKERRQ(ierr); 259 for (i=0; i<mbs; i++){ 260 jl[i] = mbs; il[i] = 0; 261 } 262 263 /* create and initialize a linked list for storing column indices of the active row k */ 264 nlnk = mbs + 1; 265 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 266 267 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 268 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 269 current_space = free_space; 270 271 for (k=0; k<mbs; k++){ /* for each active row k */ 272 /* initialize lnk by the column indices of row rip[k] of A */ 273 nzk = 0; 274 ncols = ai[k+1] - ai[k]; 275 if (!ncols) SETERRQ1(PETSC_ERR_MAT_CH_ZRPVT,"Empty row %D in matrix ",k); 276 for (j=0; j<ncols; j++){ 277 i = *(aj + ai[k] + j); 278 cols[j] = i; 279 } 280 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 281 nzk += nlnk; 282 283 /* update lnk by computing fill-in for each pivot row to be merged in */ 284 prow = jl[k]; /* 1st pivot row */ 285 286 while (prow < k){ 287 nextprow = jl[prow]; 288 /* merge prow into k-th row */ 289 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 290 jmax = ui[prow+1]; 291 ncols = jmax-jmin; 292 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 293 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 294 nzk += nlnk; 295 296 /* update il and jl for prow */ 297 if (jmin < jmax){ 298 il[prow] = jmin; 299 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 300 } 301 prow = nextprow; 302 } 303 304 /* if free space is not available, make more free space */ 305 if (current_space->local_remaining<nzk) { 306 i = mbs - k + 1; /* num of unfactored rows */ 307 i *= PetscMin(nzk, i-1); /* i*nzk, i*(i-1): estimated and max additional space needed */ 308 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 309 reallocs++; 310 } 311 312 /* copy data into free space, then initialize lnk */ 313 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 314 315 /* add the k-th row into il and jl */ 316 if (nzk > 1){ 317 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 318 jl[k] = jl[i]; jl[i] = k; 319 il[k] = ui[k] + 1; 320 } 321 ui_ptr[k] = current_space->array; 322 current_space->array += nzk; 323 current_space->local_used += nzk; 324 current_space->local_remaining -= nzk; 325 326 ui[k+1] = ui[k] + nzk; 327 } 328 329 #if defined(PETSC_USE_INFO) 330 if (ai[mbs] != 0) { 331 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 332 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 333 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 334 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 335 } else { 336 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 337 } 338 #endif 339 340 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 341 ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr); 342 343 /* destroy list of free space and other temporary array(s) */ 344 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 345 ierr = PetscFreeSpaceContiguous_Cholesky(&free_space,uj,mbs,ui,udiag);CHKERRQ(ierr); /* store matrix factor */ 346 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 347 348 /* put together the new matrix in MATSEQSBAIJ format */ 349 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 350 351 b = (Mat_SeqSBAIJ*)fact->data; 352 b->singlemalloc = PETSC_FALSE; 353 b->free_a = PETSC_TRUE; 354 b->free_ij = PETSC_TRUE; 355 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 356 b->j = uj; 357 b->i = ui; 358 b->diag = udiag; 359 b->free_diag = PETSC_TRUE; 360 b->ilen = 0; 361 b->imax = 0; 362 b->row = perm; 363 b->icol = perm; 364 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 365 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 366 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 367 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 368 ierr = PetscLogObjectMemory(fact,ui[mbs]*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 369 b->maxnz = b->nz = ui[mbs]; 370 371 (fact)->info.factor_mallocs = reallocs; 372 (fact)->info.fill_ratio_given = fill; 373 if (ai[mbs] != 0) { 374 (fact)->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 375 } else { 376 (fact)->info.fill_ratio_needed = 0.0; 377 } 378 fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 379 PetscFunctionReturn(0); 380 } 381 382 #undef __FUNCT__ 383 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_inplace" 384 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 385 { 386 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 387 Mat_SeqSBAIJ *b; 388 PetscErrorCode ierr; 389 PetscTruth perm_identity,missing; 390 PetscReal fill = info->fill; 391 const PetscInt *rip,*ai,*aj; 392 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow,d; 393 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 394 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr; 395 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 396 PetscBT lnkbt; 397 398 PetscFunctionBegin; 399 ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr); 400 if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d); 401 402 /* 403 This code originally uses Modified Sparse Row (MSR) storage 404 (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise! 405 Then it is rewritten so the factor B takes seqsbaij format. However the associated 406 MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity, 407 thus the original code in MSR format is still used for these cases. 408 The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever 409 MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor. 410 */ 411 if (bs > 1){ 412 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(fact,A,perm,info);CHKERRQ(ierr); 413 PetscFunctionReturn(0); 414 } 415 416 /* check whether perm is the identity mapping */ 417 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 418 419 if (perm_identity){ 420 a->permute = PETSC_FALSE; 421 ai = a->i; aj = a->j; 422 } else { 423 SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format"); 424 /* There are bugs for reordeing. Needs further work. 425 MatReordering for sbaij cannot be efficient. User should use aij formt! */ 426 a->permute = PETSC_TRUE; 427 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 428 ai = a->inew; aj = a->jnew; 429 } 430 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 431 432 /* initialization */ 433 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 434 ui[0] = 0; 435 436 /* jl: linked list for storing indices of the pivot rows 437 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 438 ierr = PetscMalloc4(mbs,PetscInt*,&ui_ptr,mbs,PetscInt,&il,mbs,PetscInt,&jl,mbs,PetscInt,&cols);CHKERRQ(ierr); 439 for (i=0; i<mbs; i++){ 440 jl[i] = mbs; il[i] = 0; 441 } 442 443 /* create and initialize a linked list for storing column indices of the active row k */ 444 nlnk = mbs + 1; 445 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 446 447 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 448 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 449 current_space = free_space; 450 451 for (k=0; k<mbs; k++){ /* for each active row k */ 452 /* initialize lnk by the column indices of row rip[k] of A */ 453 nzk = 0; 454 ncols = ai[rip[k]+1] - ai[rip[k]]; 455 for (j=0; j<ncols; j++){ 456 i = *(aj + ai[rip[k]] + j); 457 cols[j] = rip[i]; 458 } 459 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 460 nzk += nlnk; 461 462 /* update lnk by computing fill-in for each pivot row to be merged in */ 463 prow = jl[k]; /* 1st pivot row */ 464 465 while (prow < k){ 466 nextprow = jl[prow]; 467 /* merge prow into k-th row */ 468 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 469 jmax = ui[prow+1]; 470 ncols = jmax-jmin; 471 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 472 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 473 nzk += nlnk; 474 475 /* update il and jl for prow */ 476 if (jmin < jmax){ 477 il[prow] = jmin; 478 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 479 } 480 prow = nextprow; 481 } 482 483 /* if free space is not available, make more free space */ 484 if (current_space->local_remaining<nzk) { 485 i = mbs - k + 1; /* num of unfactored rows */ 486 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 487 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 488 reallocs++; 489 } 490 491 /* copy data into free space, then initialize lnk */ 492 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 493 494 /* add the k-th row into il and jl */ 495 if (nzk-1 > 0){ 496 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 497 jl[k] = jl[i]; jl[i] = k; 498 il[k] = ui[k] + 1; 499 } 500 ui_ptr[k] = current_space->array; 501 current_space->array += nzk; 502 current_space->local_used += nzk; 503 current_space->local_remaining -= nzk; 504 505 ui[k+1] = ui[k] + nzk; 506 } 507 508 #if defined(PETSC_USE_INFO) 509 if (ai[mbs] != 0) { 510 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 511 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 512 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 513 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 514 } else { 515 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 516 } 517 #endif 518 519 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 520 ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr); 521 522 /* destroy list of free space and other temporary array(s) */ 523 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 524 ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 525 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 526 527 /* put together the new matrix in MATSEQSBAIJ format */ 528 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 529 530 b = (Mat_SeqSBAIJ*)(fact)->data; 531 b->singlemalloc = PETSC_FALSE; 532 b->free_a = PETSC_TRUE; 533 b->free_ij = PETSC_TRUE; 534 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 535 b->j = uj; 536 b->i = ui; 537 b->diag = 0; 538 b->ilen = 0; 539 b->imax = 0; 540 b->row = perm; 541 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 542 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 543 b->icol = perm; 544 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 545 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 546 ierr = PetscLogObjectMemory(fact,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 547 b->maxnz = b->nz = ui[mbs]; 548 549 (fact)->info.factor_mallocs = reallocs; 550 (fact)->info.fill_ratio_given = fill; 551 if (ai[mbs] != 0) { 552 (fact)->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 553 } else { 554 (fact)->info.fill_ratio_needed = 0.0; 555 } 556 ierr = MatSeqSBAIJSetNumericFactorization_inplace(fact,perm_identity);CHKERRQ(ierr); 557 PetscFunctionReturn(0); 558 } 559 560 #undef __FUNCT__ 561 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 562 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat C,Mat A,const MatFactorInfo *info) 563 { 564 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 565 IS perm = b->row; 566 PetscErrorCode ierr; 567 const PetscInt *ai,*aj,*perm_ptr,mbs=a->mbs,*bi=b->i,*bj=b->j; 568 PetscInt i,j; 569 PetscInt *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 570 PetscInt bs=A->rmap->bs,bs2 = a->bs2,bslog = 0; 571 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 572 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 573 MatScalar *work; 574 PetscInt *pivots; 575 576 PetscFunctionBegin; 577 /* initialization */ 578 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 579 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 580 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 581 for (i=0; i<mbs; i++) { 582 jl[i] = mbs; il[0] = 0; 583 } 584 ierr = PetscMalloc3(bs2,MatScalar,&dk,bs2,MatScalar,&uik,bs,MatScalar,&work);CHKERRQ(ierr); 585 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 586 587 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 588 589 /* check permutation */ 590 if (!a->permute){ 591 ai = a->i; aj = a->j; aa = a->a; 592 } else { 593 ai = a->inew; aj = a->jnew; 594 ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 595 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 596 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 597 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 598 599 /* flops in while loop */ 600 bslog = 2*bs*bs2; 601 602 for (i=0; i<mbs; i++){ 603 jmin = ai[i]; jmax = ai[i+1]; 604 for (j=jmin; j<jmax; j++){ 605 while (a2anew[j] != j){ 606 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 607 for (k1=0; k1<bs2; k1++){ 608 dk[k1] = aa[k*bs2+k1]; 609 aa[k*bs2+k1] = aa[j*bs2+k1]; 610 aa[j*bs2+k1] = dk[k1]; 611 } 612 } 613 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 614 if (i > aj[j]){ 615 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 616 ap = aa + j*bs2; /* ptr to the beginning of j-th block of aa */ 617 for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 618 for (k=0; k<bs; k++){ /* j-th block of aa <- dk^T */ 619 for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1]; 620 } 621 } 622 } 623 } 624 ierr = PetscFree(a2anew);CHKERRQ(ierr); 625 } 626 627 /* for each row k */ 628 for (k = 0; k<mbs; k++){ 629 630 /*initialize k-th row with elements nonzero in row perm(k) of A */ 631 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 632 633 ap = aa + jmin*bs2; 634 for (j = jmin; j < jmax; j++){ 635 vj = perm_ptr[aj[j]]; /* block col. index */ 636 rtmp_ptr = rtmp + vj*bs2; 637 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 638 } 639 640 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 641 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 642 i = jl[k]; /* first row to be added to k_th row */ 643 644 while (i < k){ 645 nexti = jl[i]; /* next row to be added to k_th row */ 646 647 /* compute multiplier */ 648 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 649 650 /* uik = -inv(Di)*U_bar(i,k) */ 651 diag = ba + i*bs2; 652 u = ba + ili*bs2; 653 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 654 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 655 656 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 657 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 658 ierr = PetscLogFlops(bslog*2.0);CHKERRQ(ierr); 659 660 /* update -U(i,k) */ 661 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 662 663 /* add multiple of row i to k-th row ... */ 664 jmin = ili + 1; jmax = bi[i+1]; 665 if (jmin < jmax){ 666 for (j=jmin; j<jmax; j++) { 667 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 668 rtmp_ptr = rtmp + bj[j]*bs2; 669 u = ba + j*bs2; 670 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 671 } 672 ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr); 673 674 /* ... add i to row list for next nonzero entry */ 675 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 676 j = bj[jmin]; 677 jl[i] = jl[j]; jl[j] = i; /* update jl */ 678 } 679 i = nexti; 680 } 681 682 /* save nonzero entries in k-th row of U ... */ 683 684 /* invert diagonal block */ 685 diag = ba+k*bs2; 686 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 687 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 688 689 jmin = bi[k]; jmax = bi[k+1]; 690 if (jmin < jmax) { 691 for (j=jmin; j<jmax; j++){ 692 vj = bj[j]; /* block col. index of U */ 693 u = ba + j*bs2; 694 rtmp_ptr = rtmp + vj*bs2; 695 for (k1=0; k1<bs2; k1++){ 696 *u++ = *rtmp_ptr; 697 *rtmp_ptr++ = 0.0; 698 } 699 } 700 701 /* ... add k to row list for first nonzero entry in k-th row */ 702 il[k] = jmin; 703 i = bj[jmin]; 704 jl[k] = jl[i]; jl[i] = k; 705 } 706 } 707 708 ierr = PetscFree(rtmp);CHKERRQ(ierr); 709 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 710 ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr); 711 ierr = PetscFree(pivots);CHKERRQ(ierr); 712 if (a->permute){ 713 ierr = PetscFree(aa);CHKERRQ(ierr); 714 } 715 716 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 717 C->ops->solve = MatSolve_SeqSBAIJ_N_inplace; 718 C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_inplace; 719 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_N_inplace; 720 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_N_inplace; 721 722 C->assembled = PETSC_TRUE; 723 C->preallocated = PETSC_TRUE; 724 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 725 PetscFunctionReturn(0); 726 } 727 728 #undef __FUNCT__ 729 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering" 730 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 731 { 732 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 733 PetscErrorCode ierr; 734 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 735 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 736 PetscInt bs=A->rmap->bs,bs2 = a->bs2,bslog; 737 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 738 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 739 MatScalar *work; 740 PetscInt *pivots; 741 742 PetscFunctionBegin; 743 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 744 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 745 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 746 for (i=0; i<mbs; i++) { 747 jl[i] = mbs; il[0] = 0; 748 } 749 ierr = PetscMalloc3(bs2,MatScalar,&dk,bs2,MatScalar,&uik,bs,MatScalar,&work);CHKERRQ(ierr); 750 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 751 752 ai = a->i; aj = a->j; aa = a->a; 753 754 /* flops in while loop */ 755 bslog = 2*bs*bs2; 756 757 /* for each row k */ 758 for (k = 0; k<mbs; k++){ 759 760 /*initialize k-th row with elements nonzero in row k of A */ 761 jmin = ai[k]; jmax = ai[k+1]; 762 ap = aa + jmin*bs2; 763 for (j = jmin; j < jmax; j++){ 764 vj = aj[j]; /* block col. index */ 765 rtmp_ptr = rtmp + vj*bs2; 766 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 767 } 768 769 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 770 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 771 i = jl[k]; /* first row to be added to k_th row */ 772 773 while (i < k){ 774 nexti = jl[i]; /* next row to be added to k_th row */ 775 776 /* compute multiplier */ 777 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 778 779 /* uik = -inv(Di)*U_bar(i,k) */ 780 diag = ba + i*bs2; 781 u = ba + ili*bs2; 782 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 783 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 784 785 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 786 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 787 ierr = PetscLogFlops(bslog*2.0);CHKERRQ(ierr); 788 789 /* update -U(i,k) */ 790 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 791 792 /* add multiple of row i to k-th row ... */ 793 jmin = ili + 1; jmax = bi[i+1]; 794 if (jmin < jmax){ 795 for (j=jmin; j<jmax; j++) { 796 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 797 rtmp_ptr = rtmp + bj[j]*bs2; 798 u = ba + j*bs2; 799 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 800 } 801 ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr); 802 803 /* ... add i to row list for next nonzero entry */ 804 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 805 j = bj[jmin]; 806 jl[i] = jl[j]; jl[j] = i; /* update jl */ 807 } 808 i = nexti; 809 } 810 811 /* save nonzero entries in k-th row of U ... */ 812 813 /* invert diagonal block */ 814 diag = ba+k*bs2; 815 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 816 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 817 818 jmin = bi[k]; jmax = bi[k+1]; 819 if (jmin < jmax) { 820 for (j=jmin; j<jmax; j++){ 821 vj = bj[j]; /* block col. index of U */ 822 u = ba + j*bs2; 823 rtmp_ptr = rtmp + vj*bs2; 824 for (k1=0; k1<bs2; k1++){ 825 *u++ = *rtmp_ptr; 826 *rtmp_ptr++ = 0.0; 827 } 828 } 829 830 /* ... add k to row list for first nonzero entry in k-th row */ 831 il[k] = jmin; 832 i = bj[jmin]; 833 jl[k] = jl[i]; jl[i] = k; 834 } 835 } 836 837 ierr = PetscFree(rtmp);CHKERRQ(ierr); 838 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 839 ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr); 840 ierr = PetscFree(pivots);CHKERRQ(ierr); 841 842 C->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 843 C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 844 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 845 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 846 C->assembled = PETSC_TRUE; 847 C->preallocated = PETSC_TRUE; 848 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 849 PetscFunctionReturn(0); 850 } 851 852 /* 853 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 854 Version for blocks 2 by 2. 855 */ 856 #undef __FUNCT__ 857 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 858 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat C,Mat A,const MatFactorInfo *info) 859 { 860 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 861 IS perm = b->row; 862 PetscErrorCode ierr; 863 const PetscInt *ai,*aj,*perm_ptr; 864 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 865 PetscInt *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 866 MatScalar *ba = b->a,*aa,*ap; 867 MatScalar *u,*diag,*rtmp,*rtmp_ptr,dk[4],uik[4]; 868 PetscReal shift = info->shiftamount; 869 870 PetscFunctionBegin; 871 /* initialization */ 872 /* il and jl record the first nonzero element in each row of the accessing 873 window U(0:k, k:mbs-1). 874 jl: list of rows to be added to uneliminated rows 875 i>= k: jl(i) is the first row to be added to row i 876 i< k: jl(i) is the row following row i in some list of rows 877 jl(i) = mbs indicates the end of a list 878 il(i): points to the first nonzero element in columns k,...,mbs-1 of 879 row i of U */ 880 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 881 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 882 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 883 for (i=0; i<mbs; i++) { 884 jl[i] = mbs; il[0] = 0; 885 } 886 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 887 888 /* check permutation */ 889 if (!a->permute){ 890 ai = a->i; aj = a->j; aa = a->a; 891 } else { 892 ai = a->inew; aj = a->jnew; 893 ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 894 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 895 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 896 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 897 898 for (i=0; i<mbs; i++){ 899 jmin = ai[i]; jmax = ai[i+1]; 900 for (j=jmin; j<jmax; j++){ 901 while (a2anew[j] != j){ 902 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 903 for (k1=0; k1<4; k1++){ 904 dk[k1] = aa[k*4+k1]; 905 aa[k*4+k1] = aa[j*4+k1]; 906 aa[j*4+k1] = dk[k1]; 907 } 908 } 909 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 910 if (i > aj[j]){ 911 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 912 ap = aa + j*4; /* ptr to the beginning of the block */ 913 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 914 ap[1] = ap[2]; 915 ap[2] = dk[1]; 916 } 917 } 918 } 919 ierr = PetscFree(a2anew);CHKERRQ(ierr); 920 } 921 922 /* for each row k */ 923 for (k = 0; k<mbs; k++){ 924 925 /*initialize k-th row with elements nonzero in row perm(k) of A */ 926 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 927 ap = aa + jmin*4; 928 for (j = jmin; j < jmax; j++){ 929 vj = perm_ptr[aj[j]]; /* block col. index */ 930 rtmp_ptr = rtmp + vj*4; 931 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 932 } 933 934 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 935 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 936 i = jl[k]; /* first row to be added to k_th row */ 937 938 while (i < k){ 939 nexti = jl[i]; /* next row to be added to k_th row */ 940 941 /* compute multiplier */ 942 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 943 944 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 945 diag = ba + i*4; 946 u = ba + ili*4; 947 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 948 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 949 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 950 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 951 952 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 953 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 954 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 955 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 956 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 957 958 ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr); 959 960 /* update -U(i,k): ba[ili] = uik */ 961 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 962 963 /* add multiple of row i to k-th row ... */ 964 jmin = ili + 1; jmax = bi[i+1]; 965 if (jmin < jmax){ 966 for (j=jmin; j<jmax; j++) { 967 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 968 rtmp_ptr = rtmp + bj[j]*4; 969 u = ba + j*4; 970 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 971 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 972 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 973 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 974 } 975 ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr); 976 977 /* ... add i to row list for next nonzero entry */ 978 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 979 j = bj[jmin]; 980 jl[i] = jl[j]; jl[j] = i; /* update jl */ 981 } 982 i = nexti; 983 } 984 985 /* save nonzero entries in k-th row of U ... */ 986 987 /* invert diagonal block */ 988 diag = ba+k*4; 989 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 990 ierr = Kernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 991 992 jmin = bi[k]; jmax = bi[k+1]; 993 if (jmin < jmax) { 994 for (j=jmin; j<jmax; j++){ 995 vj = bj[j]; /* block col. index of U */ 996 u = ba + j*4; 997 rtmp_ptr = rtmp + vj*4; 998 for (k1=0; k1<4; k1++){ 999 *u++ = *rtmp_ptr; 1000 *rtmp_ptr++ = 0.0; 1001 } 1002 } 1003 1004 /* ... add k to row list for first nonzero entry in k-th row */ 1005 il[k] = jmin; 1006 i = bj[jmin]; 1007 jl[k] = jl[i]; jl[i] = k; 1008 } 1009 } 1010 1011 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1012 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1013 if (a->permute) { 1014 ierr = PetscFree(aa);CHKERRQ(ierr); 1015 } 1016 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 1017 C->ops->solve = MatSolve_SeqSBAIJ_2_inplace; 1018 C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_inplace; 1019 C->assembled = PETSC_TRUE; 1020 C->preallocated = PETSC_TRUE; 1021 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1022 PetscFunctionReturn(0); 1023 } 1024 1025 /* 1026 Version for when blocks are 2 by 2 Using natural ordering 1027 */ 1028 #undef __FUNCT__ 1029 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 1030 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 1031 { 1032 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 1033 PetscErrorCode ierr; 1034 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 1035 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 1036 MatScalar *ba = b->a,*aa,*ap,dk[8],uik[8]; 1037 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 1038 PetscReal shift = info->shiftamount; 1039 1040 PetscFunctionBegin; 1041 /* initialization */ 1042 /* il and jl record the first nonzero element in each row of the accessing 1043 window U(0:k, k:mbs-1). 1044 jl: list of rows to be added to uneliminated rows 1045 i>= k: jl(i) is the first row to be added to row i 1046 i< k: jl(i) is the row following row i in some list of rows 1047 jl(i) = mbs indicates the end of a list 1048 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1049 row i of U */ 1050 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1051 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 1052 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1053 for (i=0; i<mbs; i++) { 1054 jl[i] = mbs; il[0] = 0; 1055 } 1056 ai = a->i; aj = a->j; aa = a->a; 1057 1058 /* for each row k */ 1059 for (k = 0; k<mbs; k++){ 1060 1061 /*initialize k-th row with elements nonzero in row k of A */ 1062 jmin = ai[k]; jmax = ai[k+1]; 1063 ap = aa + jmin*4; 1064 for (j = jmin; j < jmax; j++){ 1065 vj = aj[j]; /* block col. index */ 1066 rtmp_ptr = rtmp + vj*4; 1067 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 1068 } 1069 1070 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 1071 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 1072 i = jl[k]; /* first row to be added to k_th row */ 1073 1074 while (i < k){ 1075 nexti = jl[i]; /* next row to be added to k_th row */ 1076 1077 /* compute multiplier */ 1078 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1079 1080 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 1081 diag = ba + i*4; 1082 u = ba + ili*4; 1083 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 1084 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 1085 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 1086 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 1087 1088 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 1089 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 1090 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 1091 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 1092 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 1093 1094 ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr); 1095 1096 /* update -U(i,k): ba[ili] = uik */ 1097 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 1098 1099 /* add multiple of row i to k-th row ... */ 1100 jmin = ili + 1; jmax = bi[i+1]; 1101 if (jmin < jmax){ 1102 for (j=jmin; j<jmax; j++) { 1103 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 1104 rtmp_ptr = rtmp + bj[j]*4; 1105 u = ba + j*4; 1106 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 1107 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 1108 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 1109 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 1110 } 1111 ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr); 1112 1113 /* ... add i to row list for next nonzero entry */ 1114 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1115 j = bj[jmin]; 1116 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1117 } 1118 i = nexti; 1119 } 1120 1121 /* save nonzero entries in k-th row of U ... */ 1122 1123 /* invert diagonal block */ 1124 diag = ba+k*4; 1125 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 1126 ierr = Kernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 1127 1128 jmin = bi[k]; jmax = bi[k+1]; 1129 if (jmin < jmax) { 1130 for (j=jmin; j<jmax; j++){ 1131 vj = bj[j]; /* block col. index of U */ 1132 u = ba + j*4; 1133 rtmp_ptr = rtmp + vj*4; 1134 for (k1=0; k1<4; k1++){ 1135 *u++ = *rtmp_ptr; 1136 *rtmp_ptr++ = 0.0; 1137 } 1138 } 1139 1140 /* ... add k to row list for first nonzero entry in k-th row */ 1141 il[k] = jmin; 1142 i = bj[jmin]; 1143 jl[k] = jl[i]; jl[i] = k; 1144 } 1145 } 1146 1147 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1148 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1149 1150 C->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1151 C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1152 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1153 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1154 C->assembled = PETSC_TRUE; 1155 C->preallocated = PETSC_TRUE; 1156 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1157 PetscFunctionReturn(0); 1158 } 1159 1160 /* 1161 Numeric U^T*D*U factorization for SBAIJ format. 1162 Version for blocks are 1 by 1. 1163 */ 1164 #undef __FUNCT__ 1165 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace" 1166 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace(Mat C,Mat A,const MatFactorInfo *info) 1167 { 1168 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1169 IS ip=b->row; 1170 PetscErrorCode ierr; 1171 const PetscInt *ai,*aj,*rip; 1172 PetscInt *a2anew,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol; 1173 PetscInt k,jmin,jmax,*jl,*il,col,nexti,ili,nz; 1174 MatScalar *rtmp,*ba=b->a,*bval,*aa,dk,uikdi; 1175 PetscReal rs,shift; 1176 ChShift_Ctx sctx; 1177 PetscInt newshift; 1178 1179 PetscFunctionBegin; 1180 /* initialization */ 1181 shift = info->shiftamount; 1182 1183 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 1184 if (!a->permute){ 1185 ai = a->i; aj = a->j; aa = a->a; 1186 } else { 1187 ai = a->inew; aj = a->jnew; 1188 nz = ai[mbs]; 1189 ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr); 1190 a2anew = a->a2anew; 1191 bval = a->a; 1192 for (j=0; j<nz; j++){ 1193 aa[a2anew[j]] = *(bval++); 1194 } 1195 } 1196 1197 /* initialization */ 1198 /* il and jl record the first nonzero element in each row of the accessing 1199 window U(0:k, k:mbs-1). 1200 jl: list of rows to be added to uneliminated rows 1201 i>= k: jl(i) is the first row to be added to row i 1202 i< k: jl(i) is the row following row i in some list of rows 1203 jl(i) = mbs indicates the end of a list 1204 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1205 row i of U */ 1206 ierr = PetscMalloc3(mbs,MatScalar,&rtmp,mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1207 1208 sctx.shift_amount = 0; 1209 sctx.nshift = 0; 1210 do { 1211 sctx.chshift = PETSC_FALSE; 1212 for (i=0; i<mbs; i++) { 1213 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1214 } 1215 1216 for (k = 0; k<mbs; k++){ 1217 /*initialize k-th row by the perm[k]-th row of A */ 1218 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 1219 bval = ba + bi[k]; 1220 for (j = jmin; j < jmax; j++){ 1221 col = rip[aj[j]]; 1222 rtmp[col] = aa[j]; 1223 *bval++ = 0.0; /* for in-place factorization */ 1224 } 1225 1226 /* shift the diagonal of the matrix */ 1227 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1228 1229 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1230 dk = rtmp[k]; 1231 i = jl[k]; /* first row to be added to k_th row */ 1232 1233 while (i < k){ 1234 nexti = jl[i]; /* next row to be added to k_th row */ 1235 1236 /* compute multiplier, update diag(k) and U(i,k) */ 1237 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1238 uikdi = - ba[ili]*ba[bi[i]]; /* diagonal(k) */ 1239 dk += uikdi*ba[ili]; 1240 ba[ili] = uikdi; /* -U(i,k) */ 1241 1242 /* add multiple of row i to k-th row */ 1243 jmin = ili + 1; jmax = bi[i+1]; 1244 if (jmin < jmax){ 1245 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1246 ierr = PetscLogFlops(2.0*(jmax-jmin));CHKERRQ(ierr); 1247 1248 /* update il and jl for row i */ 1249 il[i] = jmin; 1250 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1251 } 1252 i = nexti; 1253 } 1254 1255 /* shift the diagonals when zero pivot is detected */ 1256 /* compute rs=sum of abs(off-diagonal) */ 1257 rs = 0.0; 1258 jmin = bi[k]+1; 1259 nz = bi[k+1] - jmin; 1260 if (nz){ 1261 bcol = bj + jmin; 1262 while (nz--){ 1263 rs += PetscAbsScalar(rtmp[*bcol]); 1264 bcol++; 1265 } 1266 } 1267 1268 sctx.rs = rs; 1269 sctx.pv = dk; 1270 ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr); 1271 if (newshift == 1) break; /* sctx.shift_amount is updated */ 1272 1273 /* copy data into U(k,:) */ 1274 ba[bi[k]] = 1.0/dk; /* U(k,k) */ 1275 jmin = bi[k]+1; jmax = bi[k+1]; 1276 if (jmin < jmax) { 1277 for (j=jmin; j<jmax; j++){ 1278 col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0; 1279 } 1280 /* add the k-th row into il and jl */ 1281 il[k] = jmin; 1282 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1283 } 1284 } 1285 } while (sctx.chshift); 1286 ierr = PetscFree3(rtmp,il,jl);CHKERRQ(ierr); 1287 if (a->permute){ierr = PetscFree(aa);CHKERRQ(ierr);} 1288 1289 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 1290 C->ops->solve = MatSolve_SeqSBAIJ_1_inplace; 1291 C->ops->solves = MatSolves_SeqSBAIJ_1_inplace; 1292 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_inplace; 1293 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_inplace; 1294 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_inplace; 1295 C->assembled = PETSC_TRUE; 1296 C->preallocated = PETSC_TRUE; 1297 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 1298 if (sctx.nshift){ 1299 if (info->shifttype == MAT_SHIFT_NONZERO) { 1300 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1301 } else if (info->shifttype == MAT_SHIFT_POSITIVE_DEFINITE) { 1302 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1303 } 1304 } 1305 PetscFunctionReturn(0); 1306 } 1307 1308 /* 1309 Version for when blocks are 1 by 1 Using natural ordering under new datastructure 1310 Modified from MatCholeskyFactorNumeric_SeqAIJ() 1311 */ 1312 #undef __FUNCT__ 1313 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering" 1314 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat B,Mat A,const MatFactorInfo *info) 1315 { 1316 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data; 1317 Mat_SeqSBAIJ *b=(Mat_SeqSBAIJ*)B->data; 1318 PetscErrorCode ierr; 1319 PetscInt i,j,mbs=A->rmap->n,*bi=b->i,*bj=b->j,*bdiag=b->diag,*bjtmp; 1320 PetscInt *ai=a->i,*aj=a->j,*ajtmp; 1321 PetscInt k,jmin,jmax,*c2r,*il,col,nexti,ili,nz; 1322 MatScalar *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi; 1323 FactorShiftCtx sctx; 1324 PetscReal rs; 1325 MatScalar d,*v; 1326 1327 PetscFunctionBegin; 1328 ierr = PetscMalloc3(mbs,MatScalar,&rtmp,mbs,PetscInt,&il,mbs,PetscInt,&c2r);CHKERRQ(ierr); 1329 1330 /* MatPivotSetUp(): initialize shift context sctx */ 1331 ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr); 1332 1333 if (info->shifttype == MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */ 1334 sctx.shift_top = info->zeropivot; 1335 ierr = PetscMemzero(rtmp,mbs*sizeof(MatScalar));CHKERRQ(ierr); 1336 for (i=0; i<mbs; i++) { 1337 /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */ 1338 d = (aa)[a->diag[i]]; 1339 rtmp[i] += - PetscRealPart(d); /* diagonal entry */ 1340 ajtmp = aj + ai[i] + 1; /* exclude diagonal */ 1341 v = aa + ai[i] + 1; 1342 nz = ai[i+1] - ai[i] - 1 ; 1343 for (j=0; j<nz; j++){ 1344 rtmp[i] += PetscAbsScalar(v[j]); 1345 rtmp[ajtmp[j]] += PetscAbsScalar(v[j]); 1346 } 1347 if (PetscRealPart(rtmp[i]) > sctx.shift_top) sctx.shift_top = PetscRealPart(rtmp[i]); 1348 } 1349 sctx.shift_top *= 1.1; 1350 sctx.nshift_max = 5; 1351 sctx.shift_lo = 0.; 1352 sctx.shift_hi = 1.; 1353 } 1354 1355 /* allocate working arrays 1356 c2r: linked list, keep track of pivot rows for a given column. c2r[col]: head of the list for a given col 1357 il: for active k row, il[i] gives the index of the 1st nonzero entry in U[i,k:n-1] in bj and ba arrays 1358 */ 1359 do { 1360 sctx.useshift = PETSC_FALSE; 1361 1362 for (i=0; i<mbs; i++) c2r[i] = mbs; 1363 il[0] = 0; 1364 1365 for (k = 0; k<mbs; k++){ 1366 /* zero rtmp */ 1367 nz = bi[k+1] - bi[k]; 1368 bjtmp = bj + bi[k]; 1369 for (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0; 1370 1371 /* load in initial unfactored row */ 1372 bval = ba + bi[k]; 1373 jmin = ai[k]; jmax = ai[k+1]; 1374 for (j = jmin; j < jmax; j++){ 1375 col = aj[j]; 1376 rtmp[col] = aa[j]; 1377 *bval++ = 0.0; /* for in-place factorization */ 1378 } 1379 /* shift the diagonal of the matrix: ZeropivotApply() */ 1380 rtmp[k] += sctx.shift_amount; /* shift the diagonal of the matrix */ 1381 1382 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1383 dk = rtmp[k]; 1384 i = c2r[k]; /* first row to be added to k_th row */ 1385 1386 while (i < k){ 1387 nexti = c2r[i]; /* next row to be added to k_th row */ 1388 1389 /* compute multiplier, update diag(k) and U(i,k) */ 1390 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1391 uikdi = - ba[ili]*ba[bdiag[i]]; /* diagonal(k) */ 1392 dk += uikdi*ba[ili]; /* update diag[k] */ 1393 ba[ili] = uikdi; /* -U(i,k) */ 1394 1395 /* add multiple of row i to k-th row */ 1396 jmin = ili + 1; jmax = bi[i+1]; 1397 if (jmin < jmax){ 1398 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1399 /* update il and c2r for row i */ 1400 il[i] = jmin; 1401 j = bj[jmin]; c2r[i] = c2r[j]; c2r[j] = i; 1402 } 1403 i = nexti; 1404 } 1405 1406 /* copy data into U(k,:) */ 1407 rs = 0.0; 1408 jmin = bi[k]; jmax = bi[k+1]-1; 1409 if (jmin < jmax) { 1410 for (j=jmin; j<jmax; j++){ 1411 col = bj[j]; ba[j] = rtmp[col]; rs += PetscAbsScalar(ba[j]); 1412 } 1413 /* add the k-th row into il and c2r */ 1414 il[k] = jmin; 1415 i = bj[jmin]; c2r[k] = c2r[i]; c2r[i] = k; 1416 } 1417 1418 /* MatPivotCheck() */ 1419 sctx.rs = rs; 1420 sctx.pv = dk; 1421 if (info->shifttype == MAT_SHIFT_NONZERO){ 1422 ierr = MatPivotCheck_nz(info,sctx,k);CHKERRQ(ierr); 1423 } else if (info->shifttype == MAT_SHIFT_POSITIVE_DEFINITE){ 1424 ierr = MatPivotCheck_pd(info,sctx,k);CHKERRQ(ierr); 1425 } else if (info->shifttype == MAT_SHIFT_INBLOCKS){ 1426 ierr = MatPivotCheck_inblocks(info,sctx,k);CHKERRQ(ierr); 1427 } else { 1428 ierr = MatPivotCheck_none(info,sctx,k);CHKERRQ(ierr); 1429 } 1430 dk = sctx.pv; 1431 1432 ba[bdiag[k]] = 1.0/dk; /* U(k,k) */ 1433 } 1434 } while (sctx.useshift); 1435 1436 ierr = PetscFree3(rtmp,il,c2r);CHKERRQ(ierr); 1437 1438 B->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1439 B->ops->solves = MatSolves_SeqSBAIJ_1; 1440 B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1441 B->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering; 1442 B->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering; 1443 1444 B->assembled = PETSC_TRUE; 1445 B->preallocated = PETSC_TRUE; 1446 ierr = PetscLogFlops(B->rmap->n);CHKERRQ(ierr); 1447 1448 /* MatPivotView() */ 1449 if (sctx.nshift){ 1450 if (info->shifttype == MAT_SHIFT_POSITIVE_DEFINITE) { 1451 ierr = PetscInfo4(A,"number of shift_pd tries %D, shift_amount %G, diagonal shifted up by %e fraction top_value %e\n",sctx.nshift,sctx.shift_amount,sctx.shift_fraction,sctx.shift_top);CHKERRQ(ierr); 1452 } else if (info->shifttype == MAT_SHIFT_NONZERO) { 1453 ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1454 } else if (info->shifttype == MAT_SHIFT_INBLOCKS){ 1455 ierr = PetscInfo2(A,"number of shift_inblocks applied %D, each shift_amount %G\n",sctx.nshift,info->shiftamount);CHKERRQ(ierr); 1456 } 1457 } 1458 PetscFunctionReturn(0); 1459 } 1460 1461 #undef __FUNCT__ 1462 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace" 1463 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace(Mat C,Mat A,const MatFactorInfo *info) 1464 { 1465 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1466 PetscErrorCode ierr; 1467 PetscInt i,j,mbs = a->mbs; 1468 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 1469 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz; 1470 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 1471 PetscReal rs; 1472 ChShift_Ctx sctx; 1473 PetscInt newshift; 1474 1475 PetscFunctionBegin; 1476 /* initialization */ 1477 /* il and jl record the first nonzero element in each row of the accessing 1478 window U(0:k, k:mbs-1). 1479 jl: list of rows to be added to uneliminated rows 1480 i>= k: jl(i) is the first row to be added to row i 1481 i< k: jl(i) is the row following row i in some list of rows 1482 jl(i) = mbs indicates the end of a list 1483 il(i): points to the first nonzero element in U(i,k:mbs-1) 1484 */ 1485 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1486 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1487 1488 sctx.shift_amount = 0; 1489 sctx.nshift = 0; 1490 do { 1491 sctx.chshift = PETSC_FALSE; 1492 for (i=0; i<mbs; i++) { 1493 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1494 } 1495 1496 for (k = 0; k<mbs; k++){ 1497 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1498 nz = ai[k+1] - ai[k]; 1499 acol = aj + ai[k]; 1500 aval = aa + ai[k]; 1501 bval = ba + bi[k]; 1502 while (nz -- ){ 1503 rtmp[*acol++] = *aval++; 1504 *bval++ = 0.0; /* for in-place factorization */ 1505 } 1506 1507 /* shift the diagonal of the matrix */ 1508 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1509 1510 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1511 dk = rtmp[k]; 1512 i = jl[k]; /* first row to be added to k_th row */ 1513 1514 while (i < k){ 1515 nexti = jl[i]; /* next row to be added to k_th row */ 1516 /* compute multiplier, update D(k) and U(i,k) */ 1517 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1518 uikdi = - ba[ili]*ba[bi[i]]; 1519 dk += uikdi*ba[ili]; 1520 ba[ili] = uikdi; /* -U(i,k) */ 1521 1522 /* add multiple of row i to k-th row ... */ 1523 jmin = ili + 1; 1524 nz = bi[i+1] - jmin; 1525 if (nz > 0){ 1526 bcol = bj + jmin; 1527 bval = ba + jmin; 1528 ierr = PetscLogFlops(2.0*nz);CHKERRQ(ierr); 1529 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 1530 1531 /* update il and jl for i-th row */ 1532 il[i] = jmin; 1533 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1534 } 1535 i = nexti; 1536 } 1537 1538 /* shift the diagonals when zero pivot is detected */ 1539 /* compute rs=sum of abs(off-diagonal) */ 1540 rs = 0.0; 1541 jmin = bi[k]+1; 1542 nz = bi[k+1] - jmin; 1543 if (nz){ 1544 bcol = bj + jmin; 1545 while (nz--){ 1546 rs += PetscAbsScalar(rtmp[*bcol]); 1547 bcol++; 1548 } 1549 } 1550 1551 sctx.rs = rs; 1552 sctx.pv = dk; 1553 ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr); 1554 if (newshift == 1) break; /* sctx.shift_amount is updated */ 1555 1556 /* copy data into U(k,:) */ 1557 ba[bi[k]] = 1.0/dk; 1558 jmin = bi[k]+1; 1559 nz = bi[k+1] - jmin; 1560 if (nz){ 1561 bcol = bj + jmin; 1562 bval = ba + jmin; 1563 while (nz--){ 1564 *bval++ = rtmp[*bcol]; 1565 rtmp[*bcol++] = 0.0; 1566 } 1567 /* add k-th row into il and jl */ 1568 il[k] = jmin; 1569 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1570 } 1571 } /* end of for (k = 0; k<mbs; k++) */ 1572 } while (sctx.chshift); 1573 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1574 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1575 1576 C->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1577 C->ops->solves = MatSolves_SeqSBAIJ_1_inplace; 1578 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1579 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1580 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1581 1582 C->assembled = PETSC_TRUE; 1583 C->preallocated = PETSC_TRUE; 1584 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 1585 if (sctx.nshift){ 1586 if (info->shifttype == MAT_SHIFT_NONZERO) { 1587 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1588 } else if (info->shifttype == MAT_SHIFT_POSITIVE_DEFINITE) { 1589 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1590 } 1591 } 1592 PetscFunctionReturn(0); 1593 } 1594 1595 #undef __FUNCT__ 1596 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ" 1597 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,const MatFactorInfo *info) 1598 { 1599 PetscErrorCode ierr; 1600 Mat C; 1601 1602 PetscFunctionBegin; 1603 ierr = MatGetFactor(A,"petsc",MAT_FACTOR_CHOLESKY,&C);CHKERRQ(ierr); 1604 ierr = MatCholeskyFactorSymbolic(C,A,perm,info);CHKERRQ(ierr); 1605 ierr = MatCholeskyFactorNumeric(C,A,info);CHKERRQ(ierr); 1606 A->ops->solve = C->ops->solve; 1607 A->ops->solvetranspose = C->ops->solvetranspose; 1608 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 1609 PetscFunctionReturn(0); 1610 } 1611 1612 1613