1 2 #include <../src/mat/impls/baij/seq/baij.h> 3 #include <../src/mat/impls/sbaij/seq/sbaij.h> 4 #include <../src/mat/blockinvert.h> 5 #include <petscis.h> 6 7 /* 8 input: 9 F -- numeric factor 10 output: 11 nneg, nzero, npos: matrix inertia 12 */ 13 14 #undef __FUNCT__ 15 #define __FUNCT__ "MatGetInertia_SeqSBAIJ" 16 PetscErrorCode MatGetInertia_SeqSBAIJ(Mat F,PetscInt *nneig,PetscInt *nzero,PetscInt *npos) 17 { 18 Mat_SeqSBAIJ *fact_ptr = (Mat_SeqSBAIJ*)F->data; 19 MatScalar *dd = fact_ptr->a; 20 PetscInt mbs=fact_ptr->mbs,bs=F->rmap->bs,i,nneig_tmp,npos_tmp,*fi = fact_ptr->diag; 21 22 PetscFunctionBegin; 23 if (bs != 1) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for bs: %D >1 yet",bs); 24 nneig_tmp = 0; npos_tmp = 0; 25 for (i=0; i<mbs; i++) { 26 if (PetscRealPart(dd[*fi]) > 0.0) { 27 npos_tmp++; 28 } else if (PetscRealPart(dd[*fi]) < 0.0) { 29 nneig_tmp++; 30 } 31 fi++; 32 } 33 if (nneig) *nneig = nneig_tmp; 34 if (npos) *npos = npos_tmp; 35 if (nzero) *nzero = mbs - nneig_tmp - npos_tmp; 36 PetscFunctionReturn(0); 37 } 38 39 /* 40 Symbolic U^T*D*U factorization for SBAIJ format. Modified from SSF of YSMP. 41 Use Modified Sparse Row (MSR) storage for u and ju. See page 85, "Iterative Methods ..." by Saad. 42 */ 43 #undef __FUNCT__ 44 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_MSR" 45 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(Mat F,Mat A,IS perm,const MatFactorInfo *info) 46 { 47 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 48 PetscErrorCode ierr; 49 const PetscInt *rip,*ai,*aj; 50 PetscInt i,mbs = a->mbs,*jutmp,bs = A->rmap->bs,bs2=a->bs2; 51 PetscInt m,reallocs = 0,prow; 52 PetscInt *jl,*q,jmin,jmax,juidx,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 53 PetscReal f = info->fill; 54 PetscBool perm_identity; 55 56 PetscFunctionBegin; 57 /* check whether perm is the identity mapping */ 58 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 59 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 60 61 if (perm_identity) { /* without permutation */ 62 a->permute = PETSC_FALSE; 63 ai = a->i; aj = a->j; 64 } else { /* non-trivial permutation */ 65 a->permute = PETSC_TRUE; 66 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 67 ai = a->inew; aj = a->jnew; 68 } 69 70 /* initialization */ 71 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&iu);CHKERRQ(ierr); 72 umax = (PetscInt)(f*ai[mbs] + 1); umax += mbs + 1; 73 ierr = PetscMalloc(umax*sizeof(PetscInt),&ju);CHKERRQ(ierr); 74 iu[0] = mbs+1; 75 juidx = mbs + 1; /* index for ju */ 76 /* jl linked list for pivot row -- linked list for col index */ 77 ierr = PetscMalloc2(mbs,PetscInt,&jl,mbs,PetscInt,&q);CHKERRQ(ierr); 78 for (i=0; i<mbs; i++) { 79 jl[i] = mbs; 80 q[i] = 0; 81 } 82 83 /* for each row k */ 84 for (k=0; k<mbs; k++) { 85 for (i=0; i<mbs; i++) q[i] = 0; /* to be removed! */ 86 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 87 q[k] = mbs; 88 /* initialize nonzero structure of k-th row to row rip[k] of A */ 89 jmin = ai[rip[k]] +1; /* exclude diag[k] */ 90 jmax = ai[rip[k]+1]; 91 for (j=jmin; j<jmax; j++) { 92 vj = rip[aj[j]]; /* col. value */ 93 if (vj > k) { 94 qm = k; 95 do { 96 m = qm; qm = q[m]; 97 } while (qm < vj); 98 if (qm == vj) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Duplicate entry in A\n"); 99 nzk++; 100 q[m] = vj; 101 q[vj] = qm; 102 } /* if (vj > k) */ 103 } /* for (j=jmin; j<jmax; j++) */ 104 105 /* modify nonzero structure of k-th row by computing fill-in 106 for each row i to be merged in */ 107 prow = k; 108 prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */ 109 110 while (prow < k) { 111 /* merge row prow into k-th row */ 112 jmin = iu[prow] + 1; jmax = iu[prow+1]; 113 qm = k; 114 for (j=jmin; j<jmax; j++) { 115 vj = ju[j]; 116 do { 117 m = qm; qm = q[m]; 118 } while (qm < vj); 119 if (qm != vj) { 120 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 121 } 122 } 123 prow = jl[prow]; /* next pivot row */ 124 } 125 126 /* add k to row list for first nonzero element in k-th row */ 127 if (nzk > 0) { 128 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 129 jl[k] = jl[i]; jl[i] = k; 130 } 131 iu[k+1] = iu[k] + nzk; 132 133 /* allocate more space to ju if needed */ 134 if (iu[k+1] > umax) { 135 /* estimate how much additional space we will need */ 136 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 137 /* just double the memory each time */ 138 maxadd = umax; 139 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 140 umax += maxadd; 141 142 /* allocate a longer ju */ 143 ierr = PetscMalloc(umax*sizeof(PetscInt),&jutmp);CHKERRQ(ierr); 144 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr); 145 ierr = PetscFree(ju);CHKERRQ(ierr); 146 ju = jutmp; 147 reallocs++; /* count how many times we realloc */ 148 } 149 150 /* save nonzero structure of k-th row in ju */ 151 i=k; 152 while (nzk --) { 153 i = q[i]; 154 ju[juidx++] = i; 155 } 156 } 157 158 #if defined(PETSC_USE_INFO) 159 if (ai[mbs] != 0) { 160 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 161 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,f,af);CHKERRQ(ierr); 162 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 163 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G);\n",af);CHKERRQ(ierr); 164 ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr); 165 } else { 166 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 167 } 168 #endif 169 170 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 171 ierr = PetscFree2(jl,q);CHKERRQ(ierr); 172 173 /* put together the new matrix */ 174 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(F,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 175 176 /* ierr = PetscLogObjectParent(B,iperm);CHKERRQ(ierr); */ 177 b = (Mat_SeqSBAIJ*)(F)->data; 178 b->singlemalloc = PETSC_FALSE; 179 b->free_a = PETSC_TRUE; 180 b->free_ij = PETSC_TRUE; 181 ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr); 182 b->j = ju; 183 b->i = iu; 184 b->diag = 0; 185 b->ilen = 0; 186 b->imax = 0; 187 b->row = perm; 188 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 189 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 190 b->icol = perm; 191 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 192 ierr = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 193 /* In b structure: Free imax, ilen, old a, old j. 194 Allocate idnew, solve_work, new a, new j */ 195 ierr = PetscLogObjectMemory(F,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 196 b->maxnz = b->nz = iu[mbs]; 197 198 (F)->info.factor_mallocs = reallocs; 199 (F)->info.fill_ratio_given = f; 200 if (ai[mbs] != 0) { 201 (F)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 202 } else { 203 (F)->info.fill_ratio_needed = 0.0; 204 } 205 ierr = MatSeqSBAIJSetNumericFactorization_inplace(F,perm_identity);CHKERRQ(ierr); 206 PetscFunctionReturn(0); 207 } 208 /* 209 Symbolic U^T*D*U factorization for SBAIJ format. 210 See MatICCFactorSymbolic_SeqAIJ() for description of its data structure. 211 */ 212 #include <petscbt.h> 213 #include <../src/mat/utils/freespace.h> 214 #undef __FUNCT__ 215 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 216 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 217 { 218 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 219 Mat_SeqSBAIJ *b; 220 PetscErrorCode ierr; 221 PetscBool perm_identity,missing; 222 PetscReal fill = info->fill; 223 const PetscInt *rip,*ai=a->i,*aj=a->j; 224 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow,d; 225 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 226 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr,*udiag; 227 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 228 PetscBT lnkbt; 229 230 PetscFunctionBegin; 231 if (bs > 1) { 232 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(fact,A,perm,info);CHKERRQ(ierr); 233 PetscFunctionReturn(0); 234 } 235 if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n); 236 ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr); 237 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d); 238 239 /* check whether perm is the identity mapping */ 240 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 241 if (!perm_identity) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format"); 242 a->permute = PETSC_FALSE; 243 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 244 245 /* initialization */ 246 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 247 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&udiag);CHKERRQ(ierr); 248 ui[0] = 0; 249 250 /* jl: linked list for storing indices of the pivot rows 251 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 252 ierr = PetscMalloc4(mbs,PetscInt*,&ui_ptr,mbs,PetscInt,&il,mbs,PetscInt,&jl,mbs,PetscInt,&cols);CHKERRQ(ierr); 253 for (i=0; i<mbs; i++) { 254 jl[i] = mbs; il[i] = 0; 255 } 256 257 /* create and initialize a linked list for storing column indices of the active row k */ 258 nlnk = mbs + 1; 259 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 260 261 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 262 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 263 current_space = free_space; 264 265 for (k=0; k<mbs; k++) { /* for each active row k */ 266 /* initialize lnk by the column indices of row rip[k] of A */ 267 nzk = 0; 268 ncols = ai[k+1] - ai[k]; 269 if (!ncols) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Empty row %D in matrix ",k); 270 for (j=0; j<ncols; j++) { 271 i = *(aj + ai[k] + j); 272 cols[j] = i; 273 } 274 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 275 nzk += nlnk; 276 277 /* update lnk by computing fill-in for each pivot row to be merged in */ 278 prow = jl[k]; /* 1st pivot row */ 279 280 while (prow < k) { 281 nextprow = jl[prow]; 282 /* merge prow into k-th row */ 283 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 284 jmax = ui[prow+1]; 285 ncols = jmax-jmin; 286 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 287 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 288 nzk += nlnk; 289 290 /* update il and jl for prow */ 291 if (jmin < jmax) { 292 il[prow] = jmin; 293 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 294 } 295 prow = nextprow; 296 } 297 298 /* if free space is not available, make more free space */ 299 if (current_space->local_remaining<nzk) { 300 i = mbs - k + 1; /* num of unfactored rows */ 301 i *= PetscMin(nzk, i-1); /* i*nzk, i*(i-1): estimated and max additional space needed */ 302 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 303 reallocs++; 304 } 305 306 /* copy data into free space, then initialize lnk */ 307 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 308 309 /* add the k-th row into il and jl */ 310 if (nzk > 1) { 311 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 312 jl[k] = jl[i]; jl[i] = k; 313 il[k] = ui[k] + 1; 314 } 315 ui_ptr[k] = current_space->array; 316 current_space->array += nzk; 317 current_space->local_used += nzk; 318 current_space->local_remaining -= nzk; 319 320 ui[k+1] = ui[k] + nzk; 321 } 322 323 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 324 ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr); 325 326 /* destroy list of free space and other temporary array(s) */ 327 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 328 ierr = PetscFreeSpaceContiguous_Cholesky(&free_space,uj,mbs,ui,udiag);CHKERRQ(ierr); /* store matrix factor */ 329 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 330 331 /* put together the new matrix in MATSEQSBAIJ format */ 332 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 333 334 b = (Mat_SeqSBAIJ*)fact->data; 335 b->singlemalloc = PETSC_FALSE; 336 b->free_a = PETSC_TRUE; 337 b->free_ij = PETSC_TRUE; 338 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 339 b->j = uj; 340 b->i = ui; 341 b->diag = udiag; 342 b->free_diag = PETSC_TRUE; 343 b->ilen = 0; 344 b->imax = 0; 345 b->row = perm; 346 b->icol = perm; 347 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 348 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 349 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 350 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 351 ierr = PetscLogObjectMemory(fact,ui[mbs]*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 352 b->maxnz = b->nz = ui[mbs]; 353 354 fact->info.factor_mallocs = reallocs; 355 fact->info.fill_ratio_given = fill; 356 if (ai[mbs] != 0) { 357 fact->info.fill_ratio_needed = ((PetscReal)ui[mbs])/ai[mbs]; 358 } else { 359 fact->info.fill_ratio_needed = 0.0; 360 } 361 #if defined(PETSC_USE_INFO) 362 if (ai[mbs] != 0) { 363 PetscReal af = fact->info.fill_ratio_needed; 364 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 365 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 366 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 367 } else { 368 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 369 } 370 #endif 371 fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 372 PetscFunctionReturn(0); 373 } 374 375 #undef __FUNCT__ 376 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_inplace" 377 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(Mat fact,Mat A,IS perm,const MatFactorInfo *info) 378 { 379 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 380 Mat_SeqSBAIJ *b; 381 PetscErrorCode ierr; 382 PetscBool perm_identity,missing; 383 PetscReal fill = info->fill; 384 const PetscInt *rip,*ai,*aj; 385 PetscInt i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow,d; 386 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 387 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr; 388 PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 389 PetscBT lnkbt; 390 391 PetscFunctionBegin; 392 ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr); 393 if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d); 394 395 /* 396 This code originally uses Modified Sparse Row (MSR) storage 397 (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise! 398 Then it is rewritten so the factor B takes seqsbaij format. However the associated 399 MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity, 400 thus the original code in MSR format is still used for these cases. 401 The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever 402 MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor. 403 */ 404 if (bs > 1) { 405 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(fact,A,perm,info);CHKERRQ(ierr); 406 PetscFunctionReturn(0); 407 } 408 409 /* check whether perm is the identity mapping */ 410 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 411 412 if (perm_identity) { 413 a->permute = PETSC_FALSE; 414 ai = a->i; aj = a->j; 415 } else { 416 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format"); 417 /* There are bugs for reordeing. Needs further work. 418 MatReordering for sbaij cannot be efficient. User should use aij formt! */ 419 a->permute = PETSC_TRUE; 420 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 421 ai = a->inew; aj = a->jnew; 422 } 423 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 424 425 /* initialization */ 426 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 427 ui[0] = 0; 428 429 /* jl: linked list for storing indices of the pivot rows 430 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 431 ierr = PetscMalloc4(mbs,PetscInt*,&ui_ptr,mbs,PetscInt,&il,mbs,PetscInt,&jl,mbs,PetscInt,&cols);CHKERRQ(ierr); 432 for (i=0; i<mbs; i++) { 433 jl[i] = mbs; il[i] = 0; 434 } 435 436 /* create and initialize a linked list for storing column indices of the active row k */ 437 nlnk = mbs + 1; 438 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 439 440 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 441 ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 442 current_space = free_space; 443 444 for (k=0; k<mbs; k++) { /* for each active row k */ 445 /* initialize lnk by the column indices of row rip[k] of A */ 446 nzk = 0; 447 ncols = ai[rip[k]+1] - ai[rip[k]]; 448 for (j=0; j<ncols; j++) { 449 i = *(aj + ai[rip[k]] + j); 450 cols[j] = rip[i]; 451 } 452 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 453 nzk += nlnk; 454 455 /* update lnk by computing fill-in for each pivot row to be merged in */ 456 prow = jl[k]; /* 1st pivot row */ 457 458 while (prow < k) { 459 nextprow = jl[prow]; 460 /* merge prow into k-th row */ 461 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 462 jmax = ui[prow+1]; 463 ncols = jmax-jmin; 464 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 465 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 466 nzk += nlnk; 467 468 /* update il and jl for prow */ 469 if (jmin < jmax) { 470 il[prow] = jmin; 471 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 472 } 473 prow = nextprow; 474 } 475 476 /* if free space is not available, make more free space */ 477 if (current_space->local_remaining<nzk) { 478 i = mbs - k + 1; /* num of unfactored rows */ 479 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 480 ierr = PetscFreeSpaceGet(i,¤t_space);CHKERRQ(ierr); 481 reallocs++; 482 } 483 484 /* copy data into free space, then initialize lnk */ 485 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 486 487 /* add the k-th row into il and jl */ 488 if (nzk-1 > 0) { 489 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 490 jl[k] = jl[i]; jl[i] = k; 491 il[k] = ui[k] + 1; 492 } 493 ui_ptr[k] = current_space->array; 494 current_space->array += nzk; 495 current_space->local_used += nzk; 496 current_space->local_remaining -= nzk; 497 498 ui[k+1] = ui[k] + nzk; 499 } 500 501 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 502 ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr); 503 504 /* destroy list of free space and other temporary array(s) */ 505 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 506 ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 507 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 508 509 /* put together the new matrix in MATSEQSBAIJ format */ 510 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 511 512 b = (Mat_SeqSBAIJ*)fact->data; 513 b->singlemalloc = PETSC_FALSE; 514 b->free_a = PETSC_TRUE; 515 b->free_ij = PETSC_TRUE; 516 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 517 b->j = uj; 518 b->i = ui; 519 b->diag = 0; 520 b->ilen = 0; 521 b->imax = 0; 522 b->row = perm; 523 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 524 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 525 b->icol = perm; 526 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 527 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 528 ierr = PetscLogObjectMemory(fact,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 529 b->maxnz = b->nz = ui[mbs]; 530 531 fact->info.factor_mallocs = reallocs; 532 fact->info.fill_ratio_given = fill; 533 if (ai[mbs] != 0) { 534 fact->info.fill_ratio_needed = ((PetscReal)ui[mbs])/ai[mbs]; 535 } else { 536 fact->info.fill_ratio_needed = 0.0; 537 } 538 #if defined(PETSC_USE_INFO) 539 if (ai[mbs] != 0) { 540 PetscReal af = fact->info.fill_ratio_needed; 541 ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr); 542 ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr); 543 ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr); 544 } else { 545 ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr); 546 } 547 #endif 548 ierr = MatSeqSBAIJSetNumericFactorization_inplace(fact,perm_identity);CHKERRQ(ierr); 549 PetscFunctionReturn(0); 550 } 551 552 #undef __FUNCT__ 553 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 554 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat C,Mat A,const MatFactorInfo *info) 555 { 556 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 557 IS perm = b->row; 558 PetscErrorCode ierr; 559 const PetscInt *ai,*aj,*perm_ptr,mbs=a->mbs,*bi=b->i,*bj=b->j; 560 PetscInt i,j; 561 PetscInt *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 562 PetscInt bs=A->rmap->bs,bs2 = a->bs2,bslog = 0; 563 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 564 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 565 MatScalar *work; 566 PetscInt *pivots; 567 568 PetscFunctionBegin; 569 /* initialization */ 570 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 571 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 572 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 573 for (i=0; i<mbs; i++) { 574 jl[i] = mbs; il[0] = 0; 575 } 576 ierr = PetscMalloc3(bs2,MatScalar,&dk,bs2,MatScalar,&uik,bs,MatScalar,&work);CHKERRQ(ierr); 577 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 578 579 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 580 581 /* check permutation */ 582 if (!a->permute) { 583 ai = a->i; aj = a->j; aa = a->a; 584 } else { 585 ai = a->inew; aj = a->jnew; 586 ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 587 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 588 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 589 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 590 591 /* flops in while loop */ 592 bslog = 2*bs*bs2; 593 594 for (i=0; i<mbs; i++) { 595 jmin = ai[i]; jmax = ai[i+1]; 596 for (j=jmin; j<jmax; j++) { 597 while (a2anew[j] != j) { 598 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 599 for (k1=0; k1<bs2; k1++) { 600 dk[k1] = aa[k*bs2+k1]; 601 aa[k*bs2+k1] = aa[j*bs2+k1]; 602 aa[j*bs2+k1] = dk[k1]; 603 } 604 } 605 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 606 if (i > aj[j]) { 607 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 608 ap = aa + j*bs2; /* ptr to the beginning of j-th block of aa */ 609 for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 610 for (k=0; k<bs; k++) { /* j-th block of aa <- dk^T */ 611 for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1]; 612 } 613 } 614 } 615 } 616 ierr = PetscFree(a2anew);CHKERRQ(ierr); 617 } 618 619 /* for each row k */ 620 for (k = 0; k<mbs; k++) { 621 622 /*initialize k-th row with elements nonzero in row perm(k) of A */ 623 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 624 625 ap = aa + jmin*bs2; 626 for (j = jmin; j < jmax; j++) { 627 vj = perm_ptr[aj[j]]; /* block col. index */ 628 rtmp_ptr = rtmp + vj*bs2; 629 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 630 } 631 632 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 633 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 634 i = jl[k]; /* first row to be added to k_th row */ 635 636 while (i < k) { 637 nexti = jl[i]; /* next row to be added to k_th row */ 638 639 /* compute multiplier */ 640 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 641 642 /* uik = -inv(Di)*U_bar(i,k) */ 643 diag = ba + i*bs2; 644 u = ba + ili*bs2; 645 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 646 PetscKernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 647 648 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 649 PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 650 ierr = PetscLogFlops(bslog*2.0);CHKERRQ(ierr); 651 652 /* update -U(i,k) */ 653 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 654 655 /* add multiple of row i to k-th row ... */ 656 jmin = ili + 1; jmax = bi[i+1]; 657 if (jmin < jmax) { 658 for (j=jmin; j<jmax; j++) { 659 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 660 rtmp_ptr = rtmp + bj[j]*bs2; 661 u = ba + j*bs2; 662 PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 663 } 664 ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr); 665 666 /* ... add i to row list for next nonzero entry */ 667 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 668 j = bj[jmin]; 669 jl[i] = jl[j]; jl[j] = i; /* update jl */ 670 } 671 i = nexti; 672 } 673 674 /* save nonzero entries in k-th row of U ... */ 675 676 /* invert diagonal block */ 677 diag = ba+k*bs2; 678 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 679 ierr = PetscKernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 680 681 jmin = bi[k]; jmax = bi[k+1]; 682 if (jmin < jmax) { 683 for (j=jmin; j<jmax; j++) { 684 vj = bj[j]; /* block col. index of U */ 685 u = ba + j*bs2; 686 rtmp_ptr = rtmp + vj*bs2; 687 for (k1=0; k1<bs2; k1++) { 688 *u++ = *rtmp_ptr; 689 *rtmp_ptr++ = 0.0; 690 } 691 } 692 693 /* ... add k to row list for first nonzero entry in k-th row */ 694 il[k] = jmin; 695 i = bj[jmin]; 696 jl[k] = jl[i]; jl[i] = k; 697 } 698 } 699 700 ierr = PetscFree(rtmp);CHKERRQ(ierr); 701 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 702 ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr); 703 ierr = PetscFree(pivots);CHKERRQ(ierr); 704 if (a->permute) { 705 ierr = PetscFree(aa);CHKERRQ(ierr); 706 } 707 708 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 709 C->ops->solve = MatSolve_SeqSBAIJ_N_inplace; 710 C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_inplace; 711 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_N_inplace; 712 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_N_inplace; 713 714 C->assembled = PETSC_TRUE; 715 C->preallocated = PETSC_TRUE; 716 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 717 PetscFunctionReturn(0); 718 } 719 720 #undef __FUNCT__ 721 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering" 722 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 723 { 724 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 725 PetscErrorCode ierr; 726 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 727 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 728 PetscInt bs=A->rmap->bs,bs2 = a->bs2,bslog; 729 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 730 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 731 MatScalar *work; 732 PetscInt *pivots; 733 734 PetscFunctionBegin; 735 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 736 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 737 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 738 for (i=0; i<mbs; i++) { 739 jl[i] = mbs; il[0] = 0; 740 } 741 ierr = PetscMalloc3(bs2,MatScalar,&dk,bs2,MatScalar,&uik,bs,MatScalar,&work);CHKERRQ(ierr); 742 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 743 744 ai = a->i; aj = a->j; aa = a->a; 745 746 /* flops in while loop */ 747 bslog = 2*bs*bs2; 748 749 /* for each row k */ 750 for (k = 0; k<mbs; k++) { 751 752 /*initialize k-th row with elements nonzero in row k of A */ 753 jmin = ai[k]; jmax = ai[k+1]; 754 ap = aa + jmin*bs2; 755 for (j = jmin; j < jmax; j++) { 756 vj = aj[j]; /* block col. index */ 757 rtmp_ptr = rtmp + vj*bs2; 758 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 759 } 760 761 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 762 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 763 i = jl[k]; /* first row to be added to k_th row */ 764 765 while (i < k) { 766 nexti = jl[i]; /* next row to be added to k_th row */ 767 768 /* compute multiplier */ 769 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 770 771 /* uik = -inv(Di)*U_bar(i,k) */ 772 diag = ba + i*bs2; 773 u = ba + ili*bs2; 774 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 775 PetscKernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 776 777 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 778 PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 779 ierr = PetscLogFlops(bslog*2.0);CHKERRQ(ierr); 780 781 /* update -U(i,k) */ 782 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 783 784 /* add multiple of row i to k-th row ... */ 785 jmin = ili + 1; jmax = bi[i+1]; 786 if (jmin < jmax) { 787 for (j=jmin; j<jmax; j++) { 788 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 789 rtmp_ptr = rtmp + bj[j]*bs2; 790 u = ba + j*bs2; 791 PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 792 } 793 ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr); 794 795 /* ... add i to row list for next nonzero entry */ 796 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 797 j = bj[jmin]; 798 jl[i] = jl[j]; jl[j] = i; /* update jl */ 799 } 800 i = nexti; 801 } 802 803 /* save nonzero entries in k-th row of U ... */ 804 805 /* invert diagonal block */ 806 diag = ba+k*bs2; 807 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 808 ierr = PetscKernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 809 810 jmin = bi[k]; jmax = bi[k+1]; 811 if (jmin < jmax) { 812 for (j=jmin; j<jmax; j++) { 813 vj = bj[j]; /* block col. index of U */ 814 u = ba + j*bs2; 815 rtmp_ptr = rtmp + vj*bs2; 816 for (k1=0; k1<bs2; k1++) { 817 *u++ = *rtmp_ptr; 818 *rtmp_ptr++ = 0.0; 819 } 820 } 821 822 /* ... add k to row list for first nonzero entry in k-th row */ 823 il[k] = jmin; 824 i = bj[jmin]; 825 jl[k] = jl[i]; jl[i] = k; 826 } 827 } 828 829 ierr = PetscFree(rtmp);CHKERRQ(ierr); 830 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 831 ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr); 832 ierr = PetscFree(pivots);CHKERRQ(ierr); 833 834 C->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 835 C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 836 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 837 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace; 838 C->assembled = PETSC_TRUE; 839 C->preallocated = PETSC_TRUE; 840 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 841 PetscFunctionReturn(0); 842 } 843 844 /* 845 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 846 Version for blocks 2 by 2. 847 */ 848 #undef __FUNCT__ 849 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 850 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat C,Mat A,const MatFactorInfo *info) 851 { 852 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 853 IS perm = b->row; 854 PetscErrorCode ierr; 855 const PetscInt *ai,*aj,*perm_ptr; 856 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 857 PetscInt *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 858 MatScalar *ba = b->a,*aa,*ap; 859 MatScalar *u,*diag,*rtmp,*rtmp_ptr,dk[4],uik[4]; 860 PetscReal shift = info->shiftamount; 861 862 PetscFunctionBegin; 863 /* initialization */ 864 /* il and jl record the first nonzero element in each row of the accessing 865 window U(0:k, k:mbs-1). 866 jl: list of rows to be added to uneliminated rows 867 i>= k: jl(i) is the first row to be added to row i 868 i< k: jl(i) is the row following row i in some list of rows 869 jl(i) = mbs indicates the end of a list 870 il(i): points to the first nonzero element in columns k,...,mbs-1 of 871 row i of U */ 872 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 873 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 874 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 875 for (i=0; i<mbs; i++) { 876 jl[i] = mbs; il[0] = 0; 877 } 878 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 879 880 /* check permutation */ 881 if (!a->permute) { 882 ai = a->i; aj = a->j; aa = a->a; 883 } else { 884 ai = a->inew; aj = a->jnew; 885 ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 886 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 887 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 888 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 889 890 for (i=0; i<mbs; i++) { 891 jmin = ai[i]; jmax = ai[i+1]; 892 for (j=jmin; j<jmax; j++) { 893 while (a2anew[j] != j) { 894 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 895 for (k1=0; k1<4; k1++) { 896 dk[k1] = aa[k*4+k1]; 897 aa[k*4+k1] = aa[j*4+k1]; 898 aa[j*4+k1] = dk[k1]; 899 } 900 } 901 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 902 if (i > aj[j]) { 903 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 904 ap = aa + j*4; /* ptr to the beginning of the block */ 905 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 906 ap[1] = ap[2]; 907 ap[2] = dk[1]; 908 } 909 } 910 } 911 ierr = PetscFree(a2anew);CHKERRQ(ierr); 912 } 913 914 /* for each row k */ 915 for (k = 0; k<mbs; k++) { 916 917 /*initialize k-th row with elements nonzero in row perm(k) of A */ 918 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 919 ap = aa + jmin*4; 920 for (j = jmin; j < jmax; j++) { 921 vj = perm_ptr[aj[j]]; /* block col. index */ 922 rtmp_ptr = rtmp + vj*4; 923 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 924 } 925 926 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 927 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 928 i = jl[k]; /* first row to be added to k_th row */ 929 930 while (i < k) { 931 nexti = jl[i]; /* next row to be added to k_th row */ 932 933 /* compute multiplier */ 934 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 935 936 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 937 diag = ba + i*4; 938 u = ba + ili*4; 939 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 940 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 941 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 942 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 943 944 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 945 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 946 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 947 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 948 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 949 950 ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr); 951 952 /* update -U(i,k): ba[ili] = uik */ 953 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 954 955 /* add multiple of row i to k-th row ... */ 956 jmin = ili + 1; jmax = bi[i+1]; 957 if (jmin < jmax) { 958 for (j=jmin; j<jmax; j++) { 959 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 960 rtmp_ptr = rtmp + bj[j]*4; 961 u = ba + j*4; 962 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 963 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 964 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 965 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 966 } 967 ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr); 968 969 /* ... add i to row list for next nonzero entry */ 970 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 971 j = bj[jmin]; 972 jl[i] = jl[j]; jl[j] = i; /* update jl */ 973 } 974 i = nexti; 975 } 976 977 /* save nonzero entries in k-th row of U ... */ 978 979 /* invert diagonal block */ 980 diag = ba+k*4; 981 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 982 ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 983 984 jmin = bi[k]; jmax = bi[k+1]; 985 if (jmin < jmax) { 986 for (j=jmin; j<jmax; j++) { 987 vj = bj[j]; /* block col. index of U */ 988 u = ba + j*4; 989 rtmp_ptr = rtmp + vj*4; 990 for (k1=0; k1<4; k1++) { 991 *u++ = *rtmp_ptr; 992 *rtmp_ptr++ = 0.0; 993 } 994 } 995 996 /* ... add k to row list for first nonzero entry in k-th row */ 997 il[k] = jmin; 998 i = bj[jmin]; 999 jl[k] = jl[i]; jl[i] = k; 1000 } 1001 } 1002 1003 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1004 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1005 if (a->permute) { 1006 ierr = PetscFree(aa);CHKERRQ(ierr); 1007 } 1008 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 1009 C->ops->solve = MatSolve_SeqSBAIJ_2_inplace; 1010 C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_inplace; 1011 C->assembled = PETSC_TRUE; 1012 C->preallocated = PETSC_TRUE; 1013 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1014 PetscFunctionReturn(0); 1015 } 1016 1017 /* 1018 Version for when blocks are 2 by 2 Using natural ordering 1019 */ 1020 #undef __FUNCT__ 1021 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 1022 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info) 1023 { 1024 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 1025 PetscErrorCode ierr; 1026 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 1027 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 1028 MatScalar *ba = b->a,*aa,*ap,dk[8],uik[8]; 1029 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 1030 PetscReal shift = info->shiftamount; 1031 1032 PetscFunctionBegin; 1033 /* initialization */ 1034 /* il and jl record the first nonzero element in each row of the accessing 1035 window U(0:k, k:mbs-1). 1036 jl: list of rows to be added to uneliminated rows 1037 i>= k: jl(i) is the first row to be added to row i 1038 i< k: jl(i) is the row following row i in some list of rows 1039 jl(i) = mbs indicates the end of a list 1040 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1041 row i of U */ 1042 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1043 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 1044 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1045 for (i=0; i<mbs; i++) { 1046 jl[i] = mbs; il[0] = 0; 1047 } 1048 ai = a->i; aj = a->j; aa = a->a; 1049 1050 /* for each row k */ 1051 for (k = 0; k<mbs; k++) { 1052 1053 /*initialize k-th row with elements nonzero in row k of A */ 1054 jmin = ai[k]; jmax = ai[k+1]; 1055 ap = aa + jmin*4; 1056 for (j = jmin; j < jmax; j++) { 1057 vj = aj[j]; /* block col. index */ 1058 rtmp_ptr = rtmp + vj*4; 1059 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 1060 } 1061 1062 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 1063 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 1064 i = jl[k]; /* first row to be added to k_th row */ 1065 1066 while (i < k) { 1067 nexti = jl[i]; /* next row to be added to k_th row */ 1068 1069 /* compute multiplier */ 1070 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1071 1072 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 1073 diag = ba + i*4; 1074 u = ba + ili*4; 1075 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 1076 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 1077 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 1078 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 1079 1080 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 1081 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 1082 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 1083 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 1084 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 1085 1086 ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr); 1087 1088 /* update -U(i,k): ba[ili] = uik */ 1089 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 1090 1091 /* add multiple of row i to k-th row ... */ 1092 jmin = ili + 1; jmax = bi[i+1]; 1093 if (jmin < jmax) { 1094 for (j=jmin; j<jmax; j++) { 1095 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 1096 rtmp_ptr = rtmp + bj[j]*4; 1097 u = ba + j*4; 1098 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 1099 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 1100 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 1101 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 1102 } 1103 ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr); 1104 1105 /* ... add i to row list for next nonzero entry */ 1106 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1107 j = bj[jmin]; 1108 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1109 } 1110 i = nexti; 1111 } 1112 1113 /* save nonzero entries in k-th row of U ... */ 1114 1115 /* invert diagonal block */ 1116 diag = ba+k*4; 1117 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 1118 ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr); 1119 1120 jmin = bi[k]; jmax = bi[k+1]; 1121 if (jmin < jmax) { 1122 for (j=jmin; j<jmax; j++) { 1123 vj = bj[j]; /* block col. index of U */ 1124 u = ba + j*4; 1125 rtmp_ptr = rtmp + vj*4; 1126 for (k1=0; k1<4; k1++) { 1127 *u++ = *rtmp_ptr; 1128 *rtmp_ptr++ = 0.0; 1129 } 1130 } 1131 1132 /* ... add k to row list for first nonzero entry in k-th row */ 1133 il[k] = jmin; 1134 i = bj[jmin]; 1135 jl[k] = jl[i]; jl[i] = k; 1136 } 1137 } 1138 1139 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1140 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1141 1142 C->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1143 C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1144 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1145 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace; 1146 C->assembled = PETSC_TRUE; 1147 C->preallocated = PETSC_TRUE; 1148 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1149 PetscFunctionReturn(0); 1150 } 1151 1152 /* 1153 Numeric U^T*D*U factorization for SBAIJ format. 1154 Version for blocks are 1 by 1. 1155 */ 1156 #undef __FUNCT__ 1157 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace" 1158 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace(Mat C,Mat A,const MatFactorInfo *info) 1159 { 1160 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1161 IS ip=b->row; 1162 PetscErrorCode ierr; 1163 const PetscInt *ai,*aj,*rip; 1164 PetscInt *a2anew,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol; 1165 PetscInt k,jmin,jmax,*jl,*il,col,nexti,ili,nz; 1166 MatScalar *rtmp,*ba=b->a,*bval,*aa,dk,uikdi; 1167 PetscReal rs; 1168 FactorShiftCtx sctx; 1169 1170 PetscFunctionBegin; 1171 /* MatPivotSetUp(): initialize shift context sctx */ 1172 ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr); 1173 1174 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 1175 if (!a->permute) { 1176 ai = a->i; aj = a->j; aa = a->a; 1177 } else { 1178 ai = a->inew; aj = a->jnew; 1179 nz = ai[mbs]; 1180 ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr); 1181 a2anew = a->a2anew; 1182 bval = a->a; 1183 for (j=0; j<nz; j++) { 1184 aa[a2anew[j]] = *(bval++); 1185 } 1186 } 1187 1188 /* initialization */ 1189 /* il and jl record the first nonzero element in each row of the accessing 1190 window U(0:k, k:mbs-1). 1191 jl: list of rows to be added to uneliminated rows 1192 i>= k: jl(i) is the first row to be added to row i 1193 i< k: jl(i) is the row following row i in some list of rows 1194 jl(i) = mbs indicates the end of a list 1195 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1196 row i of U */ 1197 ierr = PetscMalloc3(mbs,MatScalar,&rtmp,mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1198 1199 do { 1200 sctx.newshift = PETSC_FALSE; 1201 for (i=0; i<mbs; i++) { 1202 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1203 } 1204 1205 for (k = 0; k<mbs; k++) { 1206 /*initialize k-th row by the perm[k]-th row of A */ 1207 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 1208 bval = ba + bi[k]; 1209 for (j = jmin; j < jmax; j++) { 1210 col = rip[aj[j]]; 1211 rtmp[col] = aa[j]; 1212 *bval++ = 0.0; /* for in-place factorization */ 1213 } 1214 1215 /* shift the diagonal of the matrix */ 1216 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1217 1218 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1219 dk = rtmp[k]; 1220 i = jl[k]; /* first row to be added to k_th row */ 1221 1222 while (i < k) { 1223 nexti = jl[i]; /* next row to be added to k_th row */ 1224 1225 /* compute multiplier, update diag(k) and U(i,k) */ 1226 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1227 uikdi = - ba[ili]*ba[bi[i]]; /* diagonal(k) */ 1228 dk += uikdi*ba[ili]; 1229 ba[ili] = uikdi; /* -U(i,k) */ 1230 1231 /* add multiple of row i to k-th row */ 1232 jmin = ili + 1; jmax = bi[i+1]; 1233 if (jmin < jmax) { 1234 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1235 ierr = PetscLogFlops(2.0*(jmax-jmin));CHKERRQ(ierr); 1236 1237 /* update il and jl for row i */ 1238 il[i] = jmin; 1239 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1240 } 1241 i = nexti; 1242 } 1243 1244 /* shift the diagonals when zero pivot is detected */ 1245 /* compute rs=sum of abs(off-diagonal) */ 1246 rs = 0.0; 1247 jmin = bi[k]+1; 1248 nz = bi[k+1] - jmin; 1249 if (nz) { 1250 bcol = bj + jmin; 1251 while (nz--) { 1252 rs += PetscAbsScalar(rtmp[*bcol]); 1253 bcol++; 1254 } 1255 } 1256 1257 sctx.rs = rs; 1258 sctx.pv = dk; 1259 ierr = MatPivotCheck(A,info,&sctx,k);CHKERRQ(ierr); 1260 if (sctx.newshift) break; /* sctx.shift_amount is updated */ 1261 dk = sctx.pv; 1262 1263 /* copy data into U(k,:) */ 1264 ba[bi[k]] = 1.0/dk; /* U(k,k) */ 1265 jmin = bi[k]+1; jmax = bi[k+1]; 1266 if (jmin < jmax) { 1267 for (j=jmin; j<jmax; j++) { 1268 col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0; 1269 } 1270 /* add the k-th row into il and jl */ 1271 il[k] = jmin; 1272 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1273 } 1274 } 1275 } while (sctx.newshift); 1276 ierr = PetscFree3(rtmp,il,jl);CHKERRQ(ierr); 1277 if (a->permute) {ierr = PetscFree(aa);CHKERRQ(ierr);} 1278 1279 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 1280 C->ops->solve = MatSolve_SeqSBAIJ_1_inplace; 1281 C->ops->solves = MatSolves_SeqSBAIJ_1_inplace; 1282 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_inplace; 1283 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_inplace; 1284 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_inplace; 1285 C->assembled = PETSC_TRUE; 1286 C->preallocated = PETSC_TRUE; 1287 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 1288 if (sctx.nshift) { 1289 if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) { 1290 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1291 } else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { 1292 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1293 } 1294 } 1295 PetscFunctionReturn(0); 1296 } 1297 1298 /* 1299 Version for when blocks are 1 by 1 Using natural ordering under new datastructure 1300 Modified from MatCholeskyFactorNumeric_SeqAIJ() 1301 */ 1302 #undef __FUNCT__ 1303 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering" 1304 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat B,Mat A,const MatFactorInfo *info) 1305 { 1306 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data; 1307 Mat_SeqSBAIJ *b=(Mat_SeqSBAIJ*)B->data; 1308 PetscErrorCode ierr; 1309 PetscInt i,j,mbs=A->rmap->n,*bi=b->i,*bj=b->j,*bdiag=b->diag,*bjtmp; 1310 PetscInt *ai=a->i,*aj=a->j,*ajtmp; 1311 PetscInt k,jmin,jmax,*c2r,*il,col,nexti,ili,nz; 1312 MatScalar *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi; 1313 FactorShiftCtx sctx; 1314 PetscReal rs; 1315 MatScalar d,*v; 1316 1317 PetscFunctionBegin; 1318 ierr = PetscMalloc3(mbs,MatScalar,&rtmp,mbs,PetscInt,&il,mbs,PetscInt,&c2r);CHKERRQ(ierr); 1319 1320 /* MatPivotSetUp(): initialize shift context sctx */ 1321 ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr); 1322 1323 if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */ 1324 sctx.shift_top = info->zeropivot; 1325 ierr = PetscMemzero(rtmp,mbs*sizeof(MatScalar));CHKERRQ(ierr); 1326 for (i=0; i<mbs; i++) { 1327 /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */ 1328 d = (aa)[a->diag[i]]; 1329 rtmp[i] += - PetscRealPart(d); /* diagonal entry */ 1330 ajtmp = aj + ai[i] + 1; /* exclude diagonal */ 1331 v = aa + ai[i] + 1; 1332 nz = ai[i+1] - ai[i] - 1 ; 1333 for (j=0; j<nz; j++) { 1334 rtmp[i] += PetscAbsScalar(v[j]); 1335 rtmp[ajtmp[j]] += PetscAbsScalar(v[j]); 1336 } 1337 if (PetscRealPart(rtmp[i]) > sctx.shift_top) sctx.shift_top = PetscRealPart(rtmp[i]); 1338 } 1339 sctx.shift_top *= 1.1; 1340 sctx.nshift_max = 5; 1341 sctx.shift_lo = 0.; 1342 sctx.shift_hi = 1.; 1343 } 1344 1345 /* allocate working arrays 1346 c2r: linked list, keep track of pivot rows for a given column. c2r[col]: head of the list for a given col 1347 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 1348 */ 1349 do { 1350 sctx.newshift = PETSC_FALSE; 1351 1352 for (i=0; i<mbs; i++) c2r[i] = mbs; 1353 if (mbs) il[0] = 0; 1354 1355 for (k = 0; k<mbs; k++) { 1356 /* zero rtmp */ 1357 nz = bi[k+1] - bi[k]; 1358 bjtmp = bj + bi[k]; 1359 for (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0; 1360 1361 /* load in initial unfactored row */ 1362 bval = ba + bi[k]; 1363 jmin = ai[k]; jmax = ai[k+1]; 1364 for (j = jmin; j < jmax; j++) { 1365 col = aj[j]; 1366 rtmp[col] = aa[j]; 1367 *bval++ = 0.0; /* for in-place factorization */ 1368 } 1369 /* shift the diagonal of the matrix: ZeropivotApply() */ 1370 rtmp[k] += sctx.shift_amount; /* shift the diagonal of the matrix */ 1371 1372 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1373 dk = rtmp[k]; 1374 i = c2r[k]; /* first row to be added to k_th row */ 1375 1376 while (i < k) { 1377 nexti = c2r[i]; /* next row to be added to k_th row */ 1378 1379 /* compute multiplier, update diag(k) and U(i,k) */ 1380 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1381 uikdi = - ba[ili]*ba[bdiag[i]]; /* diagonal(k) */ 1382 dk += uikdi*ba[ili]; /* update diag[k] */ 1383 ba[ili] = uikdi; /* -U(i,k) */ 1384 1385 /* add multiple of row i to k-th row */ 1386 jmin = ili + 1; jmax = bi[i+1]; 1387 if (jmin < jmax) { 1388 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1389 /* update il and c2r for row i */ 1390 il[i] = jmin; 1391 j = bj[jmin]; c2r[i] = c2r[j]; c2r[j] = i; 1392 } 1393 i = nexti; 1394 } 1395 1396 /* copy data into U(k,:) */ 1397 rs = 0.0; 1398 jmin = bi[k]; jmax = bi[k+1]-1; 1399 if (jmin < jmax) { 1400 for (j=jmin; j<jmax; j++) { 1401 col = bj[j]; ba[j] = rtmp[col]; rs += PetscAbsScalar(ba[j]); 1402 } 1403 /* add the k-th row into il and c2r */ 1404 il[k] = jmin; 1405 i = bj[jmin]; c2r[k] = c2r[i]; c2r[i] = k; 1406 } 1407 1408 sctx.rs = rs; 1409 sctx.pv = dk; 1410 ierr = MatPivotCheck(A,info,&sctx,k);CHKERRQ(ierr); 1411 if (sctx.newshift) break; 1412 dk = sctx.pv; 1413 1414 ba[bdiag[k]] = 1.0/dk; /* U(k,k) */ 1415 } 1416 } while (sctx.newshift); 1417 1418 ierr = PetscFree3(rtmp,il,c2r);CHKERRQ(ierr); 1419 1420 B->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1421 B->ops->solves = MatSolves_SeqSBAIJ_1; 1422 B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering; 1423 B->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering; 1424 B->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering; 1425 1426 B->assembled = PETSC_TRUE; 1427 B->preallocated = PETSC_TRUE; 1428 ierr = PetscLogFlops(B->rmap->n);CHKERRQ(ierr); 1429 1430 /* MatPivotView() */ 1431 if (sctx.nshift) { 1432 if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { 1433 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); 1434 } else if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) { 1435 ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1436 } else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) { 1437 ierr = PetscInfo2(A,"number of shift_inblocks applied %D, each shift_amount %G\n",sctx.nshift,info->shiftamount);CHKERRQ(ierr); 1438 } 1439 } 1440 PetscFunctionReturn(0); 1441 } 1442 1443 #undef __FUNCT__ 1444 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace" 1445 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace(Mat C,Mat A,const MatFactorInfo *info) 1446 { 1447 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1448 PetscErrorCode ierr; 1449 PetscInt i,j,mbs = a->mbs; 1450 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 1451 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz; 1452 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 1453 PetscReal rs; 1454 FactorShiftCtx sctx; 1455 1456 PetscFunctionBegin; 1457 /* MatPivotSetUp(): initialize shift context sctx */ 1458 ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr); 1459 1460 /* initialization */ 1461 /* il and jl record the first nonzero element in each row of the accessing 1462 window U(0:k, k:mbs-1). 1463 jl: list of rows to be added to uneliminated rows 1464 i>= k: jl(i) is the first row to be added to row i 1465 i< k: jl(i) is the row following row i in some list of rows 1466 jl(i) = mbs indicates the end of a list 1467 il(i): points to the first nonzero element in U(i,k:mbs-1) 1468 */ 1469 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1470 ierr = PetscMalloc2(mbs,PetscInt,&il,mbs,PetscInt,&jl);CHKERRQ(ierr); 1471 1472 do { 1473 sctx.newshift = PETSC_FALSE; 1474 for (i=0; i<mbs; i++) { 1475 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1476 } 1477 1478 for (k = 0; k<mbs; k++) { 1479 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1480 nz = ai[k+1] - ai[k]; 1481 acol = aj + ai[k]; 1482 aval = aa + ai[k]; 1483 bval = ba + bi[k]; 1484 while (nz--) { 1485 rtmp[*acol++] = *aval++; 1486 *bval++ = 0.0; /* for in-place factorization */ 1487 } 1488 1489 /* shift the diagonal of the matrix */ 1490 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1491 1492 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1493 dk = rtmp[k]; 1494 i = jl[k]; /* first row to be added to k_th row */ 1495 1496 while (i < k) { 1497 nexti = jl[i]; /* next row to be added to k_th row */ 1498 /* compute multiplier, update D(k) and U(i,k) */ 1499 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1500 uikdi = - ba[ili]*ba[bi[i]]; 1501 dk += uikdi*ba[ili]; 1502 ba[ili] = uikdi; /* -U(i,k) */ 1503 1504 /* add multiple of row i to k-th row ... */ 1505 jmin = ili + 1; 1506 nz = bi[i+1] - jmin; 1507 if (nz > 0) { 1508 bcol = bj + jmin; 1509 bval = ba + jmin; 1510 ierr = PetscLogFlops(2.0*nz);CHKERRQ(ierr); 1511 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 1512 1513 /* update il and jl for i-th row */ 1514 il[i] = jmin; 1515 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1516 } 1517 i = nexti; 1518 } 1519 1520 /* shift the diagonals when zero pivot is detected */ 1521 /* compute rs=sum of abs(off-diagonal) */ 1522 rs = 0.0; 1523 jmin = bi[k]+1; 1524 nz = bi[k+1] - jmin; 1525 if (nz) { 1526 bcol = bj + jmin; 1527 while (nz--) { 1528 rs += PetscAbsScalar(rtmp[*bcol]); 1529 bcol++; 1530 } 1531 } 1532 1533 sctx.rs = rs; 1534 sctx.pv = dk; 1535 ierr = MatPivotCheck(A,info,&sctx,k);CHKERRQ(ierr); 1536 if (sctx.newshift) break; /* sctx.shift_amount is updated */ 1537 dk = sctx.pv; 1538 1539 /* copy data into U(k,:) */ 1540 ba[bi[k]] = 1.0/dk; 1541 jmin = bi[k]+1; 1542 nz = bi[k+1] - jmin; 1543 if (nz) { 1544 bcol = bj + jmin; 1545 bval = ba + jmin; 1546 while (nz--) { 1547 *bval++ = rtmp[*bcol]; 1548 rtmp[*bcol++] = 0.0; 1549 } 1550 /* add k-th row into il and jl */ 1551 il[k] = jmin; 1552 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1553 } 1554 } /* end of for (k = 0; k<mbs; k++) */ 1555 } while (sctx.newshift); 1556 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1557 ierr = PetscFree2(il,jl);CHKERRQ(ierr); 1558 1559 C->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1560 C->ops->solves = MatSolves_SeqSBAIJ_1_inplace; 1561 C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1562 C->ops->forwardsolve = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1563 C->ops->backwardsolve = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace; 1564 1565 C->assembled = PETSC_TRUE; 1566 C->preallocated = PETSC_TRUE; 1567 ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr); 1568 if (sctx.nshift) { 1569 if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) { 1570 ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1571 } else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { 1572 ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr); 1573 } 1574 } 1575 PetscFunctionReturn(0); 1576 } 1577 1578 #undef __FUNCT__ 1579 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ" 1580 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,const MatFactorInfo *info) 1581 { 1582 PetscErrorCode ierr; 1583 Mat C; 1584 1585 PetscFunctionBegin; 1586 ierr = MatGetFactor(A,"petsc",MAT_FACTOR_CHOLESKY,&C);CHKERRQ(ierr); 1587 ierr = MatCholeskyFactorSymbolic(C,A,perm,info);CHKERRQ(ierr); 1588 ierr = MatCholeskyFactorNumeric(C,A,info);CHKERRQ(ierr); 1589 A->ops->solve = C->ops->solve; 1590 A->ops->solvetranspose = C->ops->solvetranspose; 1591 ierr = MatHeaderMerge(A,C);CHKERRQ(ierr); 1592 PetscFunctionReturn(0); 1593 } 1594 1595 1596