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 PetscTruth olddatastruct=PETSC_FALSE; 236 237 PetscFunctionBegin; 238 ierr = PetscOptionsGetTruth(PETSC_NULL,"-cholesky_old",&olddatastruct,PETSC_NULL);CHKERRQ(ierr); 239 if (olddatastruct || bs>1 ){ 240 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(fact,A,perm,info);CHKERRQ(ierr); 241 PetscFunctionReturn(0); 242 } 243 244 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); 245 ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr); 246 if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d); 247 248 /* check whether perm is the identity mapping */ 249 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 250 if (!perm_identity) SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format"); 251 a->permute = PETSC_FALSE; 252 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 253 254 /* initialization */ 255 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 256 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&udiag);CHKERRQ(ierr); 257 ui[0] = 0; 258 259 /* jl: linked list for storing indices of the pivot rows 260 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 261 ierr = PetscMalloc4(mbs,PetscInt*,&ui_ptr,mbs,PetscInt,&il,mbs,PetscInt,&jl,mbs,PetscInt,&cols);CHKERRQ(ierr); 262 for (i=0; i<mbs; i++){ 263 jl[i] = mbs; il[i] = 0; 264 } 265 266 /* create and initialize a linked list for storing column indices of the active row k */ 267 nlnk = mbs + 1; 268 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 269 270 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 271 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 272 current_space = free_space; 273 274 for (k=0; k<mbs; k++){ /* for each active row k */ 275 /* initialize lnk by the column indices of row rip[k] of A */ 276 nzk = 0; 277 ncols = ai[k+1] - ai[k]; 278 if (!ncols) SETERRQ1(PETSC_ERR_MAT_CH_ZRPVT,"Empty row %D in matrix ",k); 279 for (j=0; j<ncols; j++){ 280 i = *(aj + ai[k] + j); 281 cols[j] = i; 282 } 283 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 284 nzk += nlnk; 285 286 /* update lnk by computing fill-in for each pivot row to be merged in */ 287 prow = jl[k]; /* 1st pivot row */ 288 289 while (prow < k){ 290 nextprow = jl[prow]; 291 /* merge prow into k-th row */ 292 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 293 jmax = ui[prow+1]; 294 ncols = jmax-jmin; 295 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 296 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 297 nzk += nlnk; 298 299 /* update il and jl for prow */ 300 if (jmin < jmax){ 301 il[prow] = jmin; 302 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 303 } 304 prow = nextprow; 305 } 306 307 /* if free space is not available, make more free space */ 308 if (current_space->local_remaining<nzk) { 309 i = mbs - k + 1; /* num of unfactored rows */ 310 i *= PetscMin(nzk, i-1); /* i*nzk, i*(i-1): estimated and max additional space needed */ 311 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 312 reallocs++; 313 } 314 315 /* copy data into free space, then initialize lnk */ 316 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 317 318 /* add the k-th row into il and jl */ 319 if (nzk > 1){ 320 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 321 jl[k] = jl[i]; jl[i] = k; 322 il[k] = ui[k] + 1; 323 } 324 ui_ptr[k] = current_space->array; 325 current_space->array += nzk; 326 current_space->local_used += nzk; 327 current_space->local_remaining -= nzk; 328 329 ui[k+1] = ui[k] + nzk; 330 } 331 332 #if defined(PETSC_USE_INFO) 333 if (ai[mbs] != 0) { 334 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 335 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 336 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 337 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 338 } else { 339 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 340 } 341 #endif 342 343 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 344 ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr); 345 346 /* destroy list of free space and other temporary array(s) */ 347 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 348 ierr = PetscFreeSpaceContiguous_Cholesky(&free_space,uj,mbs,ui,udiag);CHKERRQ(ierr); /* store matrix factor */ 349 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 350 351 /* put together the new matrix in MATSEQSBAIJ format */ 352 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 353 354 b = (Mat_SeqSBAIJ*)fact->data; 355 b->singlemalloc = PETSC_FALSE; 356 b->free_a = PETSC_TRUE; 357 b->free_ij = PETSC_TRUE; 358 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 359 b->j = uj; 360 b->i = ui; 361 b->diag = udiag; 362 b->free_diag = PETSC_TRUE; 363 b->ilen = 0; 364 b->imax = 0; 365 b->row = perm; 366 b->icol = perm; 367 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 368 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 369 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 370 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 371 ierr = PetscLogObjectMemory(fact,ui[mbs]*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 372 b->maxnz = b->nz = ui[mbs]; 373 374 (fact)->info.factor_mallocs = reallocs; 375 (fact)->info.fill_ratio_given = fill; 376 if (ai[mbs] != 0) { 377 (fact)->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 378 } else { 379 (fact)->info.fill_ratio_needed = 0.0; 380 } 381 fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 382 PetscFunctionReturn(0); 383 } 384 385 #undef __FUNCT__ 386 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_inplace" 387 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 388 { 389 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 390 Mat_SeqSBAIJ *b; 391 PetscErrorCode ierr; 392 PetscTruth perm_identity,missing; 393 PetscReal fill = info->fill; 394 const PetscInt *rip,*ai,*aj; 395 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow,d; 396 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 397 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr; 398 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 399 PetscBT lnkbt; 400 401 PetscFunctionBegin; 402 ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr); 403 if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d); 404 405 /* 406 This code originally uses Modified Sparse Row (MSR) storage 407 (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise! 408 Then it is rewritten so the factor B takes seqsbaij format. However the associated 409 MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity, 410 thus the original code in MSR format is still used for these cases. 411 The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever 412 MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor. 413 */ 414 if (bs > 1){ 415 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(fact,A,perm,info);CHKERRQ(ierr); 416 PetscFunctionReturn(0); 417 } 418 419 /* check whether perm is the identity mapping */ 420 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 421 422 if (perm_identity){ 423 a->permute = PETSC_FALSE; 424 ai = a->i; aj = a->j; 425 } else { 426 SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format"); 427 /* There are bugs for reordeing. Needs further work. 428 MatReordering for sbaij cannot be efficient. User should use aij formt! */ 429 a->permute = PETSC_TRUE; 430 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 431 ai = a->inew; aj = a->jnew; 432 } 433 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 434 435 /* initialization */ 436 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 437 ui[0] = 0; 438 439 /* jl: linked list for storing indices of the pivot rows 440 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 441 ierr = PetscMalloc4(mbs,PetscInt*,&ui_ptr,mbs,PetscInt,&il,mbs,PetscInt,&jl,mbs,PetscInt,&cols);CHKERRQ(ierr); 442 for (i=0; i<mbs; i++){ 443 jl[i] = mbs; il[i] = 0; 444 } 445 446 /* create and initialize a linked list for storing column indices of the active row k */ 447 nlnk = mbs + 1; 448 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 449 450 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 451 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 452 current_space = free_space; 453 454 for (k=0; k<mbs; k++){ /* for each active row k */ 455 /* initialize lnk by the column indices of row rip[k] of A */ 456 nzk = 0; 457 ncols = ai[rip[k]+1] - ai[rip[k]]; 458 for (j=0; j<ncols; j++){ 459 i = *(aj + ai[rip[k]] + j); 460 cols[j] = rip[i]; 461 } 462 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 463 nzk += nlnk; 464 465 /* update lnk by computing fill-in for each pivot row to be merged in */ 466 prow = jl[k]; /* 1st pivot row */ 467 468 while (prow < k){ 469 nextprow = jl[prow]; 470 /* merge prow into k-th row */ 471 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 472 jmax = ui[prow+1]; 473 ncols = jmax-jmin; 474 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 475 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 476 nzk += nlnk; 477 478 /* update il and jl for prow */ 479 if (jmin < jmax){ 480 il[prow] = jmin; 481 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 482 } 483 prow = nextprow; 484 } 485 486 /* if free space is not available, make more free space */ 487 if (current_space->local_remaining<nzk) { 488 i = mbs - k + 1; /* num of unfactored rows */ 489 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 490 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 491 reallocs++; 492 } 493 494 /* copy data into free space, then initialize lnk */ 495 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 496 497 /* add the k-th row into il and jl */ 498 if (nzk-1 > 0){ 499 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 500 jl[k] = jl[i]; jl[i] = k; 501 il[k] = ui[k] + 1; 502 } 503 ui_ptr[k] = current_space->array; 504 current_space->array += nzk; 505 current_space->local_used += nzk; 506 current_space->local_remaining -= nzk; 507 508 ui[k+1] = ui[k] + nzk; 509 } 510 511 #if defined(PETSC_USE_INFO) 512 if (ai[mbs] != 0) { 513 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 514 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 515 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 516 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 517 } else { 518 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 519 } 520 #endif 521 522 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 523 ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr); 524 525 /* destroy list of free space and other temporary array(s) */ 526 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 527 ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 528 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 529 530 /* put together the new matrix in MATSEQSBAIJ format */ 531 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 532 533 b = (Mat_SeqSBAIJ*)(fact)->data; 534 b->singlemalloc = PETSC_FALSE; 535 b->free_a = PETSC_TRUE; 536 b->free_ij = PETSC_TRUE; 537 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 538 b->j = uj; 539 b->i = ui; 540 b->diag = 0; 541 b->ilen = 0; 542 b->imax = 0; 543 b->row = perm; 544 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 545 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 546 b->icol = perm; 547 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 548 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 549 ierr = PetscLogObjectMemory(fact,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 550 b->maxnz = b->nz = ui[mbs]; 551 552 (fact)->info.factor_mallocs = reallocs; 553 (fact)->info.fill_ratio_given = fill; 554 if (ai[mbs] != 0) { 555 (fact)->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 556 } else { 557 (fact)->info.fill_ratio_needed = 0.0; 558 } 559 ierr = MatSeqSBAIJSetNumericFactorization_inplace(fact,perm_identity);CHKERRQ(ierr); 560 PetscFunctionReturn(0); 561 } 562 563 #undef __FUNCT__ 564 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 565 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat C,Mat A,const MatFactorInfo *info) 566 { 567 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 568 IS perm = b->row; 569 PetscErrorCode ierr; 570 const PetscInt *ai,*aj,*perm_ptr,mbs=a->mbs,*bi=b->i,*bj=b->j; 571 PetscInt i,j; 572 PetscInt *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 573 PetscInt bs=A->rmap->bs,bs2 = a->bs2,bslog = 0; 574 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 575 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 576 MatScalar *work; 577 PetscInt *pivots; 578 579 PetscFunctionBegin; 580 /* initialization */ 581 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 582 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 583 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 584 for (i=0; i<mbs; i++) { 585 jl[i] = mbs; il[0] = 0; 586 } 587 ierr = PetscMalloc3(bs2,MatScalar,&dk,bs2,MatScalar,&uik,bs,MatScalar,&work);CHKERRQ(ierr); 588 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 589 590 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 591 592 /* check permutation */ 593 if (!a->permute){ 594 ai = a->i; aj = a->j; aa = a->a; 595 } else { 596 ai = a->inew; aj = a->jnew; 597 ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 598 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 599 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 600 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 601 602 /* flops in while loop */ 603 bslog = 2*bs*bs2; 604 605 for (i=0; i<mbs; i++){ 606 jmin = ai[i]; jmax = ai[i+1]; 607 for (j=jmin; j<jmax; j++){ 608 while (a2anew[j] != j){ 609 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 610 for (k1=0; k1<bs2; k1++){ 611 dk[k1] = aa[k*bs2+k1]; 612 aa[k*bs2+k1] = aa[j*bs2+k1]; 613 aa[j*bs2+k1] = dk[k1]; 614 } 615 } 616 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 617 if (i > aj[j]){ 618 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 619 ap = aa + j*bs2; /* ptr to the beginning of j-th block of aa */ 620 for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 621 for (k=0; k<bs; k++){ /* j-th block of aa <- dk^T */ 622 for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1]; 623 } 624 } 625 } 626 } 627 ierr = PetscFree(a2anew);CHKERRQ(ierr); 628 } 629 630 /* for each row k */ 631 for (k = 0; k<mbs; k++){ 632 633 /*initialize k-th row with elements nonzero in row perm(k) of A */ 634 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 635 636 ap = aa + jmin*bs2; 637 for (j = jmin; j < jmax; j++){ 638 vj = perm_ptr[aj[j]]; /* block col. index */ 639 rtmp_ptr = rtmp + vj*bs2; 640 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 641 } 642 643 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 644 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 645 i = jl[k]; /* first row to be added to k_th row */ 646 647 while (i < k){ 648 nexti = jl[i]; /* next row to be added to k_th row */ 649 650 /* compute multiplier */ 651 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 652 653 /* uik = -inv(Di)*U_bar(i,k) */ 654 diag = ba + i*bs2; 655 u = ba + ili*bs2; 656 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 657 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 658 659 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 660 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 661 ierr = PetscLogFlops(bslog*2.0);CHKERRQ(ierr); 662 663 /* update -U(i,k) */ 664 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 665 666 /* add multiple of row i to k-th row ... */ 667 jmin = ili + 1; jmax = bi[i+1]; 668 if (jmin < jmax){ 669 for (j=jmin; j<jmax; j++) { 670 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 671 rtmp_ptr = rtmp + bj[j]*bs2; 672 u = ba + j*bs2; 673 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 674 } 675 ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr); 676 677 /* ... add i to row list for next nonzero entry */ 678 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 679 j = bj[jmin]; 680 jl[i] = jl[j]; jl[j] = i; /* update jl */ 681 } 682 i = nexti; 683 } 684 685 /* save nonzero entries in k-th row of U ... */ 686 687 /* invert diagonal block */ 688 diag = ba+k*bs2; 689 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 690 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 691 692 jmin = bi[k]; jmax = bi[k+1]; 693 if (jmin < jmax) { 694 for (j=jmin; j<jmax; j++){ 695 vj = bj[j]; /* block col. index of U */ 696 u = ba + j*bs2; 697 rtmp_ptr = rtmp + vj*bs2; 698 for (k1=0; k1<bs2; k1++){ 699 *u++ = *rtmp_ptr; 700 *rtmp_ptr++ = 0.0; 701 } 702 } 703 704 /* ... add k to row list for first nonzero entry in k-th row */ 705 il[k] = jmin; 706 i = bj[jmin]; 707 jl[k] = jl[i]; jl[i] = k; 708 } 709 } 710 711 ierr = PetscFree(rtmp);CHKERRQ(ierr); 712 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 713 ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr); 714 ierr = PetscFree(pivots);CHKERRQ(ierr); 715 if (a->permute){ 716 ierr = PetscFree(aa);CHKERRQ(ierr); 717 } 718 719 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 720 C->ops->solve = MatSolve_SeqSBAIJ_N_inplace; 721 C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_inplace; 722 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_N_inplace; 723 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_N_inplace; 724 725 C->assembled = PETSC_TRUE; 726 C->preallocated = PETSC_TRUE; 727 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 728 PetscFunctionReturn(0); 729 } 730 731 #undef __FUNCT__ 732 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering" 733 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 734 { 735 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 736 PetscErrorCode ierr; 737 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 738 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 739 PetscInt bs=A->rmap->bs,bs2 = a->bs2,bslog; 740 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 741 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 742 MatScalar *work; 743 PetscInt *pivots; 744 745 PetscFunctionBegin; 746 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 747 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 748 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 749 for (i=0; i<mbs; i++) { 750 jl[i] = mbs; il[0] = 0; 751 } 752 ierr = PetscMalloc3(bs2,MatScalar,&dk,bs2,MatScalar,&uik,bs,MatScalar,&work);CHKERRQ(ierr); 753 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 754 755 ai = a->i; aj = a->j; aa = a->a; 756 757 /* flops in while loop */ 758 bslog = 2*bs*bs2; 759 760 /* for each row k */ 761 for (k = 0; k<mbs; k++){ 762 763 /*initialize k-th row with elements nonzero in row k of A */ 764 jmin = ai[k]; jmax = ai[k+1]; 765 ap = aa + jmin*bs2; 766 for (j = jmin; j < jmax; j++){ 767 vj = aj[j]; /* block col. index */ 768 rtmp_ptr = rtmp + vj*bs2; 769 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 770 } 771 772 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 773 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 774 i = jl[k]; /* first row to be added to k_th row */ 775 776 while (i < k){ 777 nexti = jl[i]; /* next row to be added to k_th row */ 778 779 /* compute multiplier */ 780 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 781 782 /* uik = -inv(Di)*U_bar(i,k) */ 783 diag = ba + i*bs2; 784 u = ba + ili*bs2; 785 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 786 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 787 788 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 789 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 790 ierr = PetscLogFlops(bslog*2.0);CHKERRQ(ierr); 791 792 /* update -U(i,k) */ 793 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 794 795 /* add multiple of row i to k-th row ... */ 796 jmin = ili + 1; jmax = bi[i+1]; 797 if (jmin < jmax){ 798 for (j=jmin; j<jmax; j++) { 799 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 800 rtmp_ptr = rtmp + bj[j]*bs2; 801 u = ba + j*bs2; 802 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 803 } 804 ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr); 805 806 /* ... add i to row list for next nonzero entry */ 807 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 808 j = bj[jmin]; 809 jl[i] = jl[j]; jl[j] = i; /* update jl */ 810 } 811 i = nexti; 812 } 813 814 /* save nonzero entries in k-th row of U ... */ 815 816 /* invert diagonal block */ 817 diag = ba+k*bs2; 818 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 819 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 820 821 jmin = bi[k]; jmax = bi[k+1]; 822 if (jmin < jmax) { 823 for (j=jmin; j<jmax; j++){ 824 vj = bj[j]; /* block col. index of U */ 825 u = ba + j*bs2; 826 rtmp_ptr = rtmp + vj*bs2; 827 for (k1=0; k1<bs2; k1++){ 828 *u++ = *rtmp_ptr; 829 *rtmp_ptr++ = 0.0; 830 } 831 } 832 833 /* ... add k to row list for first nonzero entry in k-th row */ 834 il[k] = jmin; 835 i = bj[jmin]; 836 jl[k] = jl[i]; jl[i] = k; 837 } 838 } 839 840 ierr = PetscFree(rtmp);CHKERRQ(ierr); 841 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 842 ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr); 843 ierr = PetscFree(pivots);CHKERRQ(ierr); 844 845 C->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 846 C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 847 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 848 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 849 C->assembled = PETSC_TRUE; 850 C->preallocated = PETSC_TRUE; 851 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 852 PetscFunctionReturn(0); 853 } 854 855 /* 856 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 857 Version for blocks 2 by 2. 858 */ 859 #undef __FUNCT__ 860 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 861 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat C,Mat A,const MatFactorInfo *info) 862 { 863 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 864 IS perm = b->row; 865 PetscErrorCode ierr; 866 const PetscInt *ai,*aj,*perm_ptr; 867 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 868 PetscInt *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 869 MatScalar *ba = b->a,*aa,*ap; 870 MatScalar *u,*diag,*rtmp,*rtmp_ptr,dk[4],uik[4]; 871 PetscReal shift = info->shiftinblocks; 872 873 PetscFunctionBegin; 874 /* initialization */ 875 /* il and jl record the first nonzero element in each row of the accessing 876 window U(0:k, k:mbs-1). 877 jl: list of rows to be added to uneliminated rows 878 i>= k: jl(i) is the first row to be added to row i 879 i< k: jl(i) is the row following row i in some list of rows 880 jl(i) = mbs indicates the end of a list 881 il(i): points to the first nonzero element in columns k,...,mbs-1 of 882 row i of U */ 883 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 884 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 885 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 886 for (i=0; i<mbs; i++) { 887 jl[i] = mbs; il[0] = 0; 888 } 889 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 890 891 /* check permutation */ 892 if (!a->permute){ 893 ai = a->i; aj = a->j; aa = a->a; 894 } else { 895 ai = a->inew; aj = a->jnew; 896 ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 897 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 898 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 899 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 900 901 for (i=0; i<mbs; i++){ 902 jmin = ai[i]; jmax = ai[i+1]; 903 for (j=jmin; j<jmax; j++){ 904 while (a2anew[j] != j){ 905 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 906 for (k1=0; k1<4; k1++){ 907 dk[k1] = aa[k*4+k1]; 908 aa[k*4+k1] = aa[j*4+k1]; 909 aa[j*4+k1] = dk[k1]; 910 } 911 } 912 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 913 if (i > aj[j]){ 914 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 915 ap = aa + j*4; /* ptr to the beginning of the block */ 916 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 917 ap[1] = ap[2]; 918 ap[2] = dk[1]; 919 } 920 } 921 } 922 ierr = PetscFree(a2anew);CHKERRQ(ierr); 923 } 924 925 /* for each row k */ 926 for (k = 0; k<mbs; k++){ 927 928 /*initialize k-th row with elements nonzero in row perm(k) of A */ 929 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 930 ap = aa + jmin*4; 931 for (j = jmin; j < jmax; j++){ 932 vj = perm_ptr[aj[j]]; /* block col. index */ 933 rtmp_ptr = rtmp + vj*4; 934 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 935 } 936 937 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 938 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 939 i = jl[k]; /* first row to be added to k_th row */ 940 941 while (i < k){ 942 nexti = jl[i]; /* next row to be added to k_th row */ 943 944 /* compute multiplier */ 945 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 946 947 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 948 diag = ba + i*4; 949 u = ba + ili*4; 950 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 951 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 952 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 953 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 954 955 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 956 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 957 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 958 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 959 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 960 961 ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr); 962 963 /* update -U(i,k): ba[ili] = uik */ 964 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 965 966 /* add multiple of row i to k-th row ... */ 967 jmin = ili + 1; jmax = bi[i+1]; 968 if (jmin < jmax){ 969 for (j=jmin; j<jmax; j++) { 970 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 971 rtmp_ptr = rtmp + bj[j]*4; 972 u = ba + j*4; 973 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 974 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 975 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 976 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 977 } 978 ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr); 979 980 /* ... add i to row list for next nonzero entry */ 981 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 982 j = bj[jmin]; 983 jl[i] = jl[j]; jl[j] = i; /* update jl */ 984 } 985 i = nexti; 986 } 987 988 /* save nonzero entries in k-th row of U ... */ 989 990 /* invert diagonal block */ 991 diag = ba+k*4; 992 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 993 ierr = Kernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 994 995 jmin = bi[k]; jmax = bi[k+1]; 996 if (jmin < jmax) { 997 for (j=jmin; j<jmax; j++){ 998 vj = bj[j]; /* block col. index of U */ 999 u = ba + j*4; 1000 rtmp_ptr = rtmp + vj*4; 1001 for (k1=0; k1<4; k1++){ 1002 *u++ = *rtmp_ptr; 1003 *rtmp_ptr++ = 0.0; 1004 } 1005 } 1006 1007 /* ... add k to row list for first nonzero entry in k-th row */ 1008 il[k] = jmin; 1009 i = bj[jmin]; 1010 jl[k] = jl[i]; jl[i] = k; 1011 } 1012 } 1013 1014 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1015 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1016 if (a->permute) { 1017 ierr = PetscFree(aa);CHKERRQ(ierr); 1018 } 1019 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 1020 C->ops->solve = MatSolve_SeqSBAIJ_2_inplace; 1021 C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_inplace; 1022 C->assembled = PETSC_TRUE; 1023 C->preallocated = PETSC_TRUE; 1024 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1025 PetscFunctionReturn(0); 1026 } 1027 1028 /* 1029 Version for when blocks are 2 by 2 Using natural ordering 1030 */ 1031 #undef __FUNCT__ 1032 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 1033 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 1034 { 1035 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 1036 PetscErrorCode ierr; 1037 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 1038 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 1039 MatScalar *ba = b->a,*aa,*ap,dk[8],uik[8]; 1040 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 1041 PetscReal shift = info->shiftinblocks; 1042 1043 PetscFunctionBegin; 1044 /* initialization */ 1045 /* il and jl record the first nonzero element in each row of the accessing 1046 window U(0:k, k:mbs-1). 1047 jl: list of rows to be added to uneliminated rows 1048 i>= k: jl(i) is the first row to be added to row i 1049 i< k: jl(i) is the row following row i in some list of rows 1050 jl(i) = mbs indicates the end of a list 1051 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1052 row i of U */ 1053 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1054 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 1055 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1056 for (i=0; i<mbs; i++) { 1057 jl[i] = mbs; il[0] = 0; 1058 } 1059 ai = a->i; aj = a->j; aa = a->a; 1060 1061 /* for each row k */ 1062 for (k = 0; k<mbs; k++){ 1063 1064 /*initialize k-th row with elements nonzero in row k of A */ 1065 jmin = ai[k]; jmax = ai[k+1]; 1066 ap = aa + jmin*4; 1067 for (j = jmin; j < jmax; j++){ 1068 vj = aj[j]; /* block col. index */ 1069 rtmp_ptr = rtmp + vj*4; 1070 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 1071 } 1072 1073 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 1074 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 1075 i = jl[k]; /* first row to be added to k_th row */ 1076 1077 while (i < k){ 1078 nexti = jl[i]; /* next row to be added to k_th row */ 1079 1080 /* compute multiplier */ 1081 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1082 1083 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 1084 diag = ba + i*4; 1085 u = ba + ili*4; 1086 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 1087 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 1088 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 1089 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 1090 1091 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 1092 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 1093 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 1094 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 1095 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 1096 1097 ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr); 1098 1099 /* update -U(i,k): ba[ili] = uik */ 1100 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 1101 1102 /* add multiple of row i to k-th row ... */ 1103 jmin = ili + 1; jmax = bi[i+1]; 1104 if (jmin < jmax){ 1105 for (j=jmin; j<jmax; j++) { 1106 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 1107 rtmp_ptr = rtmp + bj[j]*4; 1108 u = ba + j*4; 1109 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 1110 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 1111 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 1112 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 1113 } 1114 ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr); 1115 1116 /* ... add i to row list for next nonzero entry */ 1117 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1118 j = bj[jmin]; 1119 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1120 } 1121 i = nexti; 1122 } 1123 1124 /* save nonzero entries in k-th row of U ... */ 1125 1126 /* invert diagonal block */ 1127 diag = ba+k*4; 1128 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 1129 ierr = Kernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 1130 1131 jmin = bi[k]; jmax = bi[k+1]; 1132 if (jmin < jmax) { 1133 for (j=jmin; j<jmax; j++){ 1134 vj = bj[j]; /* block col. index of U */ 1135 u = ba + j*4; 1136 rtmp_ptr = rtmp + vj*4; 1137 for (k1=0; k1<4; k1++){ 1138 *u++ = *rtmp_ptr; 1139 *rtmp_ptr++ = 0.0; 1140 } 1141 } 1142 1143 /* ... add k to row list for first nonzero entry in k-th row */ 1144 il[k] = jmin; 1145 i = bj[jmin]; 1146 jl[k] = jl[i]; jl[i] = k; 1147 } 1148 } 1149 1150 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1151 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1152 1153 C->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1154 C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1155 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1156 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1157 C->assembled = PETSC_TRUE; 1158 C->preallocated = PETSC_TRUE; 1159 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1160 PetscFunctionReturn(0); 1161 } 1162 1163 /* 1164 Numeric U^T*D*U factorization for SBAIJ format. 1165 Version for blocks are 1 by 1. 1166 */ 1167 #undef __FUNCT__ 1168 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace" 1169 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace(Mat C,Mat A,const MatFactorInfo *info) 1170 { 1171 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1172 IS ip=b->row; 1173 PetscErrorCode ierr; 1174 const PetscInt *ai,*aj,*rip; 1175 PetscInt *a2anew,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol; 1176 PetscInt k,jmin,jmax,*jl,*il,col,nexti,ili,nz; 1177 MatScalar *rtmp,*ba=b->a,*bval,*aa,dk,uikdi; 1178 PetscReal zeropivot,rs,shiftnz; 1179 PetscReal shiftpd; 1180 ChShift_Ctx sctx; 1181 PetscInt newshift; 1182 1183 PetscFunctionBegin; 1184 /* initialization */ 1185 shiftnz = info->shiftnz; 1186 shiftpd = info->shiftpd; 1187 zeropivot = info->zeropivot; 1188 1189 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 1190 if (!a->permute){ 1191 ai = a->i; aj = a->j; aa = a->a; 1192 } else { 1193 ai = a->inew; aj = a->jnew; 1194 nz = ai[mbs]; 1195 ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr); 1196 a2anew = a->a2anew; 1197 bval = a->a; 1198 for (j=0; j<nz; j++){ 1199 aa[a2anew[j]] = *(bval++); 1200 } 1201 } 1202 1203 /* initialization */ 1204 /* il and jl record the first nonzero element in each row of the accessing 1205 window U(0:k, k:mbs-1). 1206 jl: list of rows to be added to uneliminated rows 1207 i>= k: jl(i) is the first row to be added to row i 1208 i< k: jl(i) is the row following row i in some list of rows 1209 jl(i) = mbs indicates the end of a list 1210 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1211 row i of U */ 1212 ierr = PetscMalloc3(mbs,MatScalar,&rtmp,mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1213 1214 sctx.shift_amount = 0; 1215 sctx.nshift = 0; 1216 do { 1217 sctx.chshift = PETSC_FALSE; 1218 for (i=0; i<mbs; i++) { 1219 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1220 } 1221 1222 for (k = 0; k<mbs; k++){ 1223 /*initialize k-th row by the perm[k]-th row of A */ 1224 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 1225 bval = ba + bi[k]; 1226 for (j = jmin; j < jmax; j++){ 1227 col = rip[aj[j]]; 1228 rtmp[col] = aa[j]; 1229 *bval++ = 0.0; /* for in-place factorization */ 1230 } 1231 1232 /* shift the diagonal of the matrix */ 1233 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1234 1235 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1236 dk = rtmp[k]; 1237 i = jl[k]; /* first row to be added to k_th row */ 1238 1239 while (i < k){ 1240 nexti = jl[i]; /* next row to be added to k_th row */ 1241 1242 /* compute multiplier, update diag(k) and U(i,k) */ 1243 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1244 uikdi = - ba[ili]*ba[bi[i]]; /* diagonal(k) */ 1245 dk += uikdi*ba[ili]; 1246 ba[ili] = uikdi; /* -U(i,k) */ 1247 1248 /* add multiple of row i to k-th row */ 1249 jmin = ili + 1; jmax = bi[i+1]; 1250 if (jmin < jmax){ 1251 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1252 ierr = PetscLogFlops(2.0*(jmax-jmin));CHKERRQ(ierr); 1253 1254 /* update il and jl for row i */ 1255 il[i] = jmin; 1256 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1257 } 1258 i = nexti; 1259 } 1260 1261 /* shift the diagonals when zero pivot is detected */ 1262 /* compute rs=sum of abs(off-diagonal) */ 1263 rs = 0.0; 1264 jmin = bi[k]+1; 1265 nz = bi[k+1] - jmin; 1266 if (nz){ 1267 bcol = bj + jmin; 1268 while (nz--){ 1269 rs += PetscAbsScalar(rtmp[*bcol]); 1270 bcol++; 1271 } 1272 } 1273 1274 sctx.rs = rs; 1275 sctx.pv = dk; 1276 ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr); 1277 if (newshift == 1) break; /* sctx.shift_amount is updated */ 1278 1279 /* copy data into U(k,:) */ 1280 ba[bi[k]] = 1.0/dk; /* U(k,k) */ 1281 jmin = bi[k]+1; jmax = bi[k+1]; 1282 if (jmin < jmax) { 1283 for (j=jmin; j<jmax; j++){ 1284 col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0; 1285 } 1286 /* add the k-th row into il and jl */ 1287 il[k] = jmin; 1288 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1289 } 1290 } 1291 } while (sctx.chshift); 1292 ierr = PetscFree3(rtmp,il,jl);CHKERRQ(ierr); 1293 if (a->permute){ierr = PetscFree(aa);CHKERRQ(ierr);} 1294 1295 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 1296 C->ops->solve = MatSolve_SeqSBAIJ_1_inplace; 1297 C->ops->solves = MatSolves_SeqSBAIJ_1_inplace; 1298 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_inplace; 1299 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_inplace; 1300 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_inplace; 1301 C->assembled = PETSC_TRUE; 1302 C->preallocated = PETSC_TRUE; 1303 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 1304 if (sctx.nshift){ 1305 if (shiftnz) { 1306 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1307 } else if (shiftpd) { 1308 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1309 } 1310 } 1311 PetscFunctionReturn(0); 1312 } 1313 1314 /* 1315 Version for when blocks are 1 by 1 Using natural ordering under new datastructure 1316 Modified from MatCholeskyFactorNumeric_SeqAIJ() 1317 */ 1318 #undef __FUNCT__ 1319 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering" 1320 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat B,Mat A,const MatFactorInfo *info) 1321 { 1322 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data; 1323 Mat_SeqSBAIJ *b=(Mat_SeqSBAIJ*)B->data; 1324 PetscErrorCode ierr; 1325 PetscInt i,j,mbs=A->rmap->n,*bi=b->i,*bj=b->j,*bdiag=b->diag,*bjtmp; 1326 PetscInt *ai=a->i,*aj=a->j; 1327 PetscInt k,jmin,jmax,*c2r,*il,col,nexti,ili,nz; 1328 MatScalar *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi; 1329 1330 LUShift_Ctx sctx; 1331 PetscInt newshift; 1332 PetscReal rs; 1333 MatScalar d,*v; 1334 1335 PetscFunctionBegin; 1336 /* MatPivotSetUp(): initialize shift context sctx */ 1337 ierr = PetscMemzero(&sctx,sizeof(LUShift_Ctx));CHKERRQ(ierr); 1338 1339 /* if both shift schemes are chosen by user, only use info->shiftpd */ 1340 if (info->shiftpd) { /* set sctx.shift_top=max{rs} */ 1341 sctx.shift_top = info->zeropivot; 1342 for (i=0; i<mbs; i++) { 1343 /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */ 1344 d = (aa)[a->diag[i]]; 1345 rs = -PetscAbsScalar(d) - PetscRealPart(d); 1346 v = aa+ai[i]; 1347 nz = ai[i+1] - ai[i]; 1348 for (j=0; j<nz; j++) 1349 rs += PetscAbsScalar(v[j]); 1350 if (rs>sctx.shift_top) sctx.shift_top = rs; 1351 } 1352 sctx.shift_top *= 1.1; 1353 sctx.nshift_max = 5; 1354 sctx.shift_lo = 0.; 1355 sctx.shift_hi = 1.; 1356 } 1357 1358 /* allocate working arrays 1359 c2r: linked list, keep track of pivot rows for a given column. c2r[col]: head of the list for a given col 1360 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 1361 */ 1362 ierr = PetscMalloc3(mbs,MatScalar,&rtmp,mbs,PetscInt,&il,mbs,PetscInt,&c2r);CHKERRQ(ierr); 1363 1364 do { 1365 sctx.lushift = PETSC_FALSE; 1366 1367 for (i=0; i<mbs; i++) c2r[i] = mbs; 1368 il[0] = 0; 1369 1370 for (k = 0; k<mbs; k++){ 1371 /* zero rtmp */ 1372 nz = bi[k+1] - bi[k]; 1373 bjtmp = bj + bi[k]; 1374 for (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0; 1375 1376 /* load in initial unfactored row */ 1377 bval = ba + bi[k]; 1378 jmin = ai[k]; jmax = ai[k+1]; 1379 for (j = jmin; j < jmax; j++){ 1380 col = aj[j]; 1381 rtmp[col] = aa[j]; 1382 *bval++ = 0.0; /* for in-place factorization */ 1383 } 1384 /* shift the diagonal of the matrix: ZeropivotApply() */ 1385 rtmp[k] += sctx.shift_amount; /* shift the diagonal of the matrix */ 1386 1387 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1388 dk = rtmp[k]; 1389 i = c2r[k]; /* first row to be added to k_th row */ 1390 1391 while (i < k){ 1392 nexti = c2r[i]; /* next row to be added to k_th row */ 1393 1394 /* compute multiplier, update diag(k) and U(i,k) */ 1395 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1396 uikdi = - ba[ili]*ba[bdiag[i]]; /* diagonal(k) */ 1397 dk += uikdi*ba[ili]; /* update diag[k] */ 1398 ba[ili] = uikdi; /* -U(i,k) */ 1399 1400 /* add multiple of row i to k-th row */ 1401 jmin = ili + 1; jmax = bi[i+1]; 1402 if (jmin < jmax){ 1403 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1404 /* update il and c2r for row i */ 1405 il[i] = jmin; 1406 j = bj[jmin]; c2r[i] = c2r[j]; c2r[j] = i; 1407 } 1408 i = nexti; 1409 } 1410 1411 /* copy data into U(k,:) */ 1412 rs = 0.0; 1413 jmin = bi[k]; jmax = bi[k+1]-1; 1414 if (jmin < jmax) { 1415 for (j=jmin; j<jmax; j++){ 1416 col = bj[j]; ba[j] = rtmp[col]; rs += PetscAbsScalar(ba[j]); 1417 } 1418 /* add the k-th row into il and c2r */ 1419 il[k] = jmin; 1420 i = bj[jmin]; c2r[k] = c2r[i]; c2r[i] = k; 1421 } 1422 1423 /* MatPivotCheck() */ 1424 sctx.rs = rs; 1425 sctx.pv = dk; 1426 if (info->shiftnz){ 1427 ierr = MatPivotCheck_nz(info,sctx,k,newshift);CHKERRQ(ierr); 1428 } else if (info->shiftpd){ 1429 ierr = MatPivotCheck_pd(info,sctx,k,newshift);CHKERRQ(ierr); 1430 } else if (info->shiftinblocks){ 1431 ierr = MatPivotCheck_inblocks(info,sctx,k,newshift);CHKERRQ(ierr); 1432 } else { 1433 ierr = MatPivotCheck_none(info,sctx,k,newshift);CHKERRQ(ierr); 1434 } 1435 dk = sctx.pv; 1436 if (newshift == 1) break; 1437 1438 ba[bdiag[k]] = 1.0/dk; /* U(k,k) */ 1439 } 1440 } while (sctx.lushift); 1441 1442 ierr = PetscFree3(rtmp,il,c2r);CHKERRQ(ierr); 1443 1444 B->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1445 B->ops->solves = MatSolves_SeqSBAIJ_1; 1446 B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1447 B->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering; 1448 B->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering; 1449 1450 B->assembled = PETSC_TRUE; 1451 B->preallocated = PETSC_TRUE; 1452 ierr = PetscLogFlops(B->rmap->n);CHKERRQ(ierr); 1453 1454 /* MatPivotView() */ 1455 if (sctx.nshift){ 1456 if (info->shiftpd) { 1457 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); 1458 } else if (info->shiftnz) { 1459 ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1460 } else if (info->shiftinblocks){ 1461 ierr = PetscInfo2(A,"number of shift_inblocks applied %D, each shift_amount %G\n",sctx.nshift,info->shiftinblocks);CHKERRQ(ierr); 1462 } 1463 } 1464 PetscFunctionReturn(0); 1465 } 1466 1467 #undef __FUNCT__ 1468 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace" 1469 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace(Mat C,Mat A,const MatFactorInfo *info) 1470 { 1471 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1472 PetscErrorCode ierr; 1473 PetscInt i,j,mbs = a->mbs; 1474 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 1475 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz; 1476 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 1477 PetscReal zeropivot,rs,shiftnz; 1478 PetscReal shiftpd; 1479 ChShift_Ctx sctx; 1480 PetscInt newshift; 1481 1482 PetscFunctionBegin; 1483 /* initialization */ 1484 shiftnz = info->shiftnz; 1485 shiftpd = info->shiftpd; 1486 zeropivot = info->zeropivot; 1487 1488 /* il and jl record the first nonzero element in each row of the accessing 1489 window U(0:k, k:mbs-1). 1490 jl: list of rows to be added to uneliminated rows 1491 i>= k: jl(i) is the first row to be added to row i 1492 i< k: jl(i) is the row following row i in some list of rows 1493 jl(i) = mbs indicates the end of a list 1494 il(i): points to the first nonzero element in U(i,k:mbs-1) 1495 */ 1496 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1497 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1498 1499 sctx.shift_amount = 0; 1500 sctx.nshift = 0; 1501 do { 1502 sctx.chshift = PETSC_FALSE; 1503 for (i=0; i<mbs; i++) { 1504 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1505 } 1506 1507 for (k = 0; k<mbs; k++){ 1508 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1509 nz = ai[k+1] - ai[k]; 1510 acol = aj + ai[k]; 1511 aval = aa + ai[k]; 1512 bval = ba + bi[k]; 1513 while (nz -- ){ 1514 rtmp[*acol++] = *aval++; 1515 *bval++ = 0.0; /* for in-place factorization */ 1516 } 1517 1518 /* shift the diagonal of the matrix */ 1519 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1520 1521 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1522 dk = rtmp[k]; 1523 i = jl[k]; /* first row to be added to k_th row */ 1524 1525 while (i < k){ 1526 nexti = jl[i]; /* next row to be added to k_th row */ 1527 /* compute multiplier, update D(k) and U(i,k) */ 1528 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1529 uikdi = - ba[ili]*ba[bi[i]]; 1530 dk += uikdi*ba[ili]; 1531 ba[ili] = uikdi; /* -U(i,k) */ 1532 1533 /* add multiple of row i to k-th row ... */ 1534 jmin = ili + 1; 1535 nz = bi[i+1] - jmin; 1536 if (nz > 0){ 1537 bcol = bj + jmin; 1538 bval = ba + jmin; 1539 ierr = PetscLogFlops(2.0*nz);CHKERRQ(ierr); 1540 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 1541 1542 /* update il and jl for i-th row */ 1543 il[i] = jmin; 1544 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1545 } 1546 i = nexti; 1547 } 1548 1549 /* shift the diagonals when zero pivot is detected */ 1550 /* compute rs=sum of abs(off-diagonal) */ 1551 rs = 0.0; 1552 jmin = bi[k]+1; 1553 nz = bi[k+1] - jmin; 1554 if (nz){ 1555 bcol = bj + jmin; 1556 while (nz--){ 1557 rs += PetscAbsScalar(rtmp[*bcol]); 1558 bcol++; 1559 } 1560 } 1561 1562 sctx.rs = rs; 1563 sctx.pv = dk; 1564 ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr); 1565 if (newshift == 1) break; /* sctx.shift_amount is updated */ 1566 1567 /* copy data into U(k,:) */ 1568 ba[bi[k]] = 1.0/dk; 1569 jmin = bi[k]+1; 1570 nz = bi[k+1] - jmin; 1571 if (nz){ 1572 bcol = bj + jmin; 1573 bval = ba + jmin; 1574 while (nz--){ 1575 *bval++ = rtmp[*bcol]; 1576 rtmp[*bcol++] = 0.0; 1577 } 1578 /* add k-th row into il and jl */ 1579 il[k] = jmin; 1580 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1581 } 1582 } /* end of for (k = 0; k<mbs; k++) */ 1583 } while (sctx.chshift); 1584 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1585 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1586 1587 C->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1588 C->ops->solves = MatSolves_SeqSBAIJ_1_inplace; 1589 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1590 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1591 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1592 1593 C->assembled = PETSC_TRUE; 1594 C->preallocated = PETSC_TRUE; 1595 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 1596 if (sctx.nshift){ 1597 if (shiftnz) { 1598 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1599 } else if (shiftpd) { 1600 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1601 } 1602 } 1603 PetscFunctionReturn(0); 1604 } 1605 1606 #undef __FUNCT__ 1607 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ" 1608 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,const MatFactorInfo *info) 1609 { 1610 PetscErrorCode ierr; 1611 Mat C; 1612 1613 PetscFunctionBegin; 1614 ierr = MatGetFactor(A,"petsc",MAT_FACTOR_CHOLESKY,&C);CHKERRQ(ierr); 1615 ierr = MatCholeskyFactorSymbolic(C,A,perm,info);CHKERRQ(ierr); 1616 ierr = MatCholeskyFactorNumeric(C,A,info);CHKERRQ(ierr); 1617 A->ops->solve = C->ops->solve; 1618 A->ops->solvetranspose = C->ops->solvetranspose; 1619 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 1620 PetscFunctionReturn(0); 1621 } 1622 1623 1624