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/inline/ilu.h" 6 #include "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 PetscScalar *dd = fact_ptr->a; 22 PetscInt mbs=fact_ptr->mbs,bs=F->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 A,IS perm,MatFactorInfo *info,Mat *B) 50 { 51 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 52 PetscErrorCode ierr; 53 PetscInt *rip,i,mbs = a->mbs,*ai,*aj; 54 PetscInt *jutmp,bs = A->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 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&jl);CHKERRQ(ierr); /* linked list for pivot row */ 81 q = jl + mbs; /* linked list for col index */ 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_DEBUG) 165 if (ai[mbs] != 0) { 166 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 167 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,f,af));CHKERRQ(ierr); 168 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af));CHKERRQ(ierr); 169 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g);\n",af));CHKERRQ(ierr); 170 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"));CHKERRQ(ierr); 171 } else { 172 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"));CHKERRQ(ierr); 173 } 174 #endif 175 176 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 177 ierr = PetscFree(jl);CHKERRQ(ierr); 178 179 /* put together the new matrix */ 180 ierr = MatCreate(A->comm,bs*mbs,bs*mbs,bs*mbs,bs*mbs,B);CHKERRQ(ierr); 181 ierr = MatSetType(*B,A->type_name);CHKERRQ(ierr); 182 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(*B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 183 184 /* ierr = PetscLogObjectParent(*B,iperm);CHKERRQ(ierr); */ 185 b = (Mat_SeqSBAIJ*)(*B)->data; 186 b->singlemalloc = PETSC_FALSE; 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(*B,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 202 b->maxnz = b->nz = iu[mbs]; 203 204 (*B)->factor = FACTOR_CHOLESKY; 205 (*B)->info.factor_mallocs = reallocs; 206 (*B)->info.fill_ratio_given = f; 207 if (ai[mbs] != 0) { 208 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 209 } else { 210 (*B)->info.fill_ratio_needed = 0.0; 211 } 212 213 if (perm_identity){ 214 switch (bs) { 215 case 1: 216 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 217 (*B)->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 218 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"));CHKERRQ(ierr); 219 break; 220 case 2: 221 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering; 222 (*B)->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering; 223 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"));CHKERRQ(ierr); 224 break; 225 case 3: 226 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering; 227 (*B)->ops->solve = MatSolve_SeqSBAIJ_3_NaturalOrdering; 228 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"));CHKERRQ(ierr); 229 break; 230 case 4: 231 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering; 232 (*B)->ops->solve = MatSolve_SeqSBAIJ_4_NaturalOrdering; 233 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"));CHKERRQ(ierr); 234 break; 235 case 5: 236 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering; 237 (*B)->ops->solve = MatSolve_SeqSBAIJ_5_NaturalOrdering; 238 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"));CHKERRQ(ierr); 239 break; 240 case 6: 241 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering; 242 (*B)->ops->solve = MatSolve_SeqSBAIJ_6_NaturalOrdering; 243 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"));CHKERRQ(ierr); 244 break; 245 case 7: 246 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering; 247 (*B)->ops->solve = MatSolve_SeqSBAIJ_7_NaturalOrdering; 248 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"));CHKERRQ(ierr); 249 break; 250 default: 251 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering; 252 (*B)->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering; 253 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"));CHKERRQ(ierr); 254 break; 255 } 256 } 257 PetscFunctionReturn(0); 258 } 259 /* 260 Symbolic U^T*D*U factorization for SBAIJ format. 261 */ 262 #include "petscbt.h" 263 #include "src/mat/utils/freespace.h" 264 #undef __FUNCT__ 265 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 266 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info,Mat *fact) 267 { 268 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 269 Mat_SeqSBAIJ *b; 270 Mat B; 271 PetscErrorCode ierr; 272 PetscTruth perm_identity; 273 PetscReal fill = info->fill; 274 PetscInt *rip,i,mbs=a->mbs,bs=A->bs,*ai,*aj,reallocs=0,prow; 275 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 276 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr; 277 FreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 278 PetscBT lnkbt; 279 280 PetscFunctionBegin; 281 /* 282 This code originally uses Modified Sparse Row (MSR) storage 283 (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise! 284 Then it is rewritten so the factor B takes seqsbaij format. However the associated 285 MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity, 286 thus the original code in MSR format is still used for these cases. 287 The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever 288 MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor. 289 */ 290 if (bs > 1){ 291 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(A,perm,info,fact);CHKERRQ(ierr); 292 PetscFunctionReturn(0); 293 } 294 295 /* check whether perm is the identity mapping */ 296 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 297 298 if (perm_identity){ 299 a->permute = PETSC_FALSE; 300 ai = a->i; aj = a->j; 301 } else { 302 a->permute = PETSC_TRUE; 303 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 304 ai = a->inew; aj = a->jnew; 305 } 306 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 307 308 /* initialization */ 309 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 310 ui[0] = 0; 311 312 /* jl: linked list for storing indices of the pivot rows 313 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 314 ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt)+mbs*sizeof(PetscInt*),&jl);CHKERRQ(ierr); 315 il = jl + mbs; 316 cols = il + mbs; 317 ui_ptr = (PetscInt**)(cols + mbs); 318 319 for (i=0; i<mbs; i++){ 320 jl[i] = mbs; il[i] = 0; 321 } 322 323 /* create and initialize a linked list for storing column indices of the active row k */ 324 nlnk = mbs + 1; 325 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 326 327 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 328 ierr = GetMoreSpace((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 329 current_space = free_space; 330 331 for (k=0; k<mbs; k++){ /* for each active row k */ 332 /* initialize lnk by the column indices of row rip[k] of A */ 333 nzk = 0; 334 ncols = ai[rip[k]+1] - ai[rip[k]]; 335 for (j=0; j<ncols; j++){ 336 i = *(aj + ai[rip[k]] + j); 337 cols[j] = rip[i]; 338 } 339 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 340 nzk += nlnk; 341 342 /* update lnk by computing fill-in for each pivot row to be merged in */ 343 prow = jl[k]; /* 1st pivot row */ 344 345 while (prow < k){ 346 nextprow = jl[prow]; 347 /* merge prow into k-th row */ 348 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 349 jmax = ui[prow+1]; 350 ncols = jmax-jmin; 351 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 352 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 353 nzk += nlnk; 354 355 /* update il and jl for prow */ 356 if (jmin < jmax){ 357 il[prow] = jmin; 358 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 359 } 360 prow = nextprow; 361 } 362 363 /* if free space is not available, make more free space */ 364 if (current_space->local_remaining<nzk) { 365 i = mbs - k + 1; /* num of unfactored rows */ 366 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 367 ierr = GetMoreSpace(i,¤t_space);CHKERRQ(ierr); 368 reallocs++; 369 } 370 371 /* copy data into free space, then initialize lnk */ 372 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 373 374 /* add the k-th row into il and jl */ 375 if (nzk-1 > 0){ 376 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 377 jl[k] = jl[i]; jl[i] = k; 378 il[k] = ui[k] + 1; 379 } 380 ui_ptr[k] = current_space->array; 381 current_space->array += nzk; 382 current_space->local_used += nzk; 383 current_space->local_remaining -= nzk; 384 385 ui[k+1] = ui[k] + nzk; 386 } 387 388 #if defined(PETSC_USE_DEBUG) 389 if (ai[mbs] != 0) { 390 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 391 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,fill,af));CHKERRQ(ierr); 392 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af));CHKERRQ(ierr); 393 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g) for best performance.\n",af));CHKERRQ(ierr); 394 } else { 395 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"));CHKERRQ(ierr); 396 } 397 #endif 398 399 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 400 ierr = PetscFree(jl);CHKERRQ(ierr); 401 402 /* destroy list of free space and other temporary array(s) */ 403 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 404 ierr = MakeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 405 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 406 407 /* put together the new matrix in MATSEQSBAIJ format */ 408 ierr = MatCreate(PETSC_COMM_SELF,mbs,mbs,mbs,mbs,fact);CHKERRQ(ierr); 409 B = *fact; 410 ierr = MatSetType(B,MATSEQSBAIJ);CHKERRQ(ierr); 411 ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr); 412 413 b = (Mat_SeqSBAIJ*)B->data; 414 b->singlemalloc = PETSC_FALSE; 415 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 416 b->j = uj; 417 b->i = ui; 418 b->diag = 0; 419 b->ilen = 0; 420 b->imax = 0; 421 b->row = perm; 422 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 423 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 424 b->icol = perm; 425 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 426 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 427 ierr = PetscLogObjectMemory(B,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 428 b->maxnz = b->nz = ui[mbs]; 429 430 B->factor = FACTOR_CHOLESKY; 431 B->info.factor_mallocs = reallocs; 432 B->info.fill_ratio_given = fill; 433 if (ai[mbs] != 0) { 434 B->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 435 } else { 436 B->info.fill_ratio_needed = 0.0; 437 } 438 439 if (perm_identity){ 440 switch (bs) { 441 case 1: 442 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 443 B->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 444 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"));CHKERRQ(ierr); 445 break; 446 case 2: 447 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering; 448 B->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering; 449 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"));CHKERRQ(ierr); 450 break; 451 case 3: 452 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering; 453 B->ops->solve = MatSolve_SeqSBAIJ_3_NaturalOrdering; 454 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"));CHKERRQ(ierr); 455 break; 456 case 4: 457 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering; 458 B->ops->solve = MatSolve_SeqSBAIJ_4_NaturalOrdering; 459 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"));CHKERRQ(ierr); 460 break; 461 case 5: 462 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering; 463 B->ops->solve = MatSolve_SeqSBAIJ_5_NaturalOrdering; 464 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"));CHKERRQ(ierr); 465 break; 466 case 6: 467 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering; 468 B->ops->solve = MatSolve_SeqSBAIJ_6_NaturalOrdering; 469 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"));CHKERRQ(ierr); 470 break; 471 case 7: 472 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering; 473 B->ops->solve = MatSolve_SeqSBAIJ_7_NaturalOrdering; 474 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"));CHKERRQ(ierr); 475 break; 476 default: 477 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering; 478 B->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering; 479 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"));CHKERRQ(ierr); 480 break; 481 } 482 } 483 PetscFunctionReturn(0); 484 } 485 #undef __FUNCT__ 486 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 487 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,MatFactorInfo *info,Mat *B) 488 { 489 Mat C = *B; 490 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 491 IS perm = b->row; 492 PetscErrorCode ierr; 493 PetscInt *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 494 PetscInt *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 495 PetscInt bs=A->bs,bs2 = a->bs2; 496 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 497 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 498 MatScalar *work; 499 PetscInt *pivots; 500 501 PetscFunctionBegin; 502 /* initialization */ 503 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 504 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 505 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 506 jl = il + mbs; 507 for (i=0; i<mbs; i++) { 508 jl[i] = mbs; il[0] = 0; 509 } 510 ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr); 511 uik = dk + bs2; 512 work = uik + bs2; 513 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 514 515 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 516 517 /* check permutation */ 518 if (!a->permute){ 519 ai = a->i; aj = a->j; aa = a->a; 520 } else { 521 ai = a->inew; aj = a->jnew; 522 ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 523 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 524 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 525 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 526 527 for (i=0; i<mbs; i++){ 528 jmin = ai[i]; jmax = ai[i+1]; 529 for (j=jmin; j<jmax; j++){ 530 while (a2anew[j] != j){ 531 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 532 for (k1=0; k1<bs2; k1++){ 533 dk[k1] = aa[k*bs2+k1]; 534 aa[k*bs2+k1] = aa[j*bs2+k1]; 535 aa[j*bs2+k1] = dk[k1]; 536 } 537 } 538 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 539 if (i > aj[j]){ 540 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 541 ap = aa + j*bs2; /* ptr to the beginning of j-th block of aa */ 542 for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 543 for (k=0; k<bs; k++){ /* j-th block of aa <- dk^T */ 544 for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1]; 545 } 546 } 547 } 548 } 549 ierr = PetscFree(a2anew);CHKERRQ(ierr); 550 } 551 552 /* for each row k */ 553 for (k = 0; k<mbs; k++){ 554 555 /*initialize k-th row with elements nonzero in row perm(k) of A */ 556 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 557 558 ap = aa + jmin*bs2; 559 for (j = jmin; j < jmax; j++){ 560 vj = perm_ptr[aj[j]]; /* block col. index */ 561 rtmp_ptr = rtmp + vj*bs2; 562 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 563 } 564 565 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 566 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 567 i = jl[k]; /* first row to be added to k_th row */ 568 569 while (i < k){ 570 nexti = jl[i]; /* next row to be added to k_th row */ 571 572 /* compute multiplier */ 573 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 574 575 /* uik = -inv(Di)*U_bar(i,k) */ 576 diag = ba + i*bs2; 577 u = ba + ili*bs2; 578 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 579 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 580 581 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 582 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 583 584 /* update -U(i,k) */ 585 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 586 587 /* add multiple of row i to k-th row ... */ 588 jmin = ili + 1; jmax = bi[i+1]; 589 if (jmin < jmax){ 590 for (j=jmin; j<jmax; j++) { 591 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 592 rtmp_ptr = rtmp + bj[j]*bs2; 593 u = ba + j*bs2; 594 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 595 } 596 597 /* ... add i to row list for next nonzero entry */ 598 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 599 j = bj[jmin]; 600 jl[i] = jl[j]; jl[j] = i; /* update jl */ 601 } 602 i = nexti; 603 } 604 605 /* save nonzero entries in k-th row of U ... */ 606 607 /* invert diagonal block */ 608 diag = ba+k*bs2; 609 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 610 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 611 612 jmin = bi[k]; jmax = bi[k+1]; 613 if (jmin < jmax) { 614 for (j=jmin; j<jmax; j++){ 615 vj = bj[j]; /* block col. index of U */ 616 u = ba + j*bs2; 617 rtmp_ptr = rtmp + vj*bs2; 618 for (k1=0; k1<bs2; k1++){ 619 *u++ = *rtmp_ptr; 620 *rtmp_ptr++ = 0.0; 621 } 622 } 623 624 /* ... add k to row list for first nonzero entry in k-th row */ 625 il[k] = jmin; 626 i = bj[jmin]; 627 jl[k] = jl[i]; jl[i] = k; 628 } 629 } 630 631 ierr = PetscFree(rtmp);CHKERRQ(ierr); 632 ierr = PetscFree(il);CHKERRQ(ierr); 633 ierr = PetscFree(dk);CHKERRQ(ierr); 634 ierr = PetscFree(pivots);CHKERRQ(ierr); 635 if (a->permute){ 636 ierr = PetscFree(aa);CHKERRQ(ierr); 637 } 638 639 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 640 C->factor = FACTOR_CHOLESKY; 641 C->assembled = PETSC_TRUE; 642 C->preallocated = PETSC_TRUE; 643 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 644 PetscFunctionReturn(0); 645 } 646 647 #undef __FUNCT__ 648 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering" 649 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B) 650 { 651 Mat C = *B; 652 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 653 PetscErrorCode ierr; 654 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 655 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 656 PetscInt bs=A->bs,bs2 = a->bs2; 657 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 658 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 659 MatScalar *work; 660 PetscInt *pivots; 661 662 PetscFunctionBegin; 663 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 664 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 665 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 666 jl = il + mbs; 667 for (i=0; i<mbs; i++) { 668 jl[i] = mbs; il[0] = 0; 669 } 670 ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr); 671 uik = dk + bs2; 672 work = uik + bs2; 673 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 674 675 ai = a->i; aj = a->j; aa = a->a; 676 677 /* for each row k */ 678 for (k = 0; k<mbs; k++){ 679 680 /*initialize k-th row with elements nonzero in row k of A */ 681 jmin = ai[k]; jmax = ai[k+1]; 682 ap = aa + jmin*bs2; 683 for (j = jmin; j < jmax; j++){ 684 vj = aj[j]; /* block col. index */ 685 rtmp_ptr = rtmp + vj*bs2; 686 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 687 } 688 689 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 690 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 691 i = jl[k]; /* first row to be added to k_th row */ 692 693 while (i < k){ 694 nexti = jl[i]; /* next row to be added to k_th row */ 695 696 /* compute multiplier */ 697 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 698 699 /* uik = -inv(Di)*U_bar(i,k) */ 700 diag = ba + i*bs2; 701 u = ba + ili*bs2; 702 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 703 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 704 705 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 706 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 707 708 /* update -U(i,k) */ 709 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 710 711 /* add multiple of row i to k-th row ... */ 712 jmin = ili + 1; jmax = bi[i+1]; 713 if (jmin < jmax){ 714 for (j=jmin; j<jmax; j++) { 715 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 716 rtmp_ptr = rtmp + bj[j]*bs2; 717 u = ba + j*bs2; 718 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 719 } 720 721 /* ... add i to row list for next nonzero entry */ 722 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 723 j = bj[jmin]; 724 jl[i] = jl[j]; jl[j] = i; /* update jl */ 725 } 726 i = nexti; 727 } 728 729 /* save nonzero entries in k-th row of U ... */ 730 731 /* invert diagonal block */ 732 diag = ba+k*bs2; 733 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 734 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 735 736 jmin = bi[k]; jmax = bi[k+1]; 737 if (jmin < jmax) { 738 for (j=jmin; j<jmax; j++){ 739 vj = bj[j]; /* block col. index of U */ 740 u = ba + j*bs2; 741 rtmp_ptr = rtmp + vj*bs2; 742 for (k1=0; k1<bs2; k1++){ 743 *u++ = *rtmp_ptr; 744 *rtmp_ptr++ = 0.0; 745 } 746 } 747 748 /* ... add k to row list for first nonzero entry in k-th row */ 749 il[k] = jmin; 750 i = bj[jmin]; 751 jl[k] = jl[i]; jl[i] = k; 752 } 753 } 754 755 ierr = PetscFree(rtmp);CHKERRQ(ierr); 756 ierr = PetscFree(il);CHKERRQ(ierr); 757 ierr = PetscFree(dk);CHKERRQ(ierr); 758 ierr = PetscFree(pivots);CHKERRQ(ierr); 759 760 C->factor = FACTOR_CHOLESKY; 761 C->assembled = PETSC_TRUE; 762 C->preallocated = PETSC_TRUE; 763 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 764 PetscFunctionReturn(0); 765 } 766 767 /* 768 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 769 Version for blocks 2 by 2. 770 */ 771 #undef __FUNCT__ 772 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 773 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,MatFactorInfo *info,Mat *B) 774 { 775 Mat C = *B; 776 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 777 IS perm = b->row; 778 PetscErrorCode ierr; 779 PetscInt *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 780 PetscInt *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 781 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 782 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 783 784 PetscFunctionBegin; 785 /* initialization */ 786 /* il and jl record the first nonzero element in each row of the accessing 787 window U(0:k, k:mbs-1). 788 jl: list of rows to be added to uneliminated rows 789 i>= k: jl(i) is the first row to be added to row i 790 i< k: jl(i) is the row following row i in some list of rows 791 jl(i) = mbs indicates the end of a list 792 il(i): points to the first nonzero element in columns k,...,mbs-1 of 793 row i of U */ 794 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 795 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 796 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 797 jl = il + mbs; 798 for (i=0; i<mbs; i++) { 799 jl[i] = mbs; il[0] = 0; 800 } 801 ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr); 802 uik = dk + 4; 803 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 804 805 /* check permutation */ 806 if (!a->permute){ 807 ai = a->i; aj = a->j; aa = a->a; 808 } else { 809 ai = a->inew; aj = a->jnew; 810 ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 811 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 812 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 813 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 814 815 for (i=0; i<mbs; i++){ 816 jmin = ai[i]; jmax = ai[i+1]; 817 for (j=jmin; j<jmax; j++){ 818 while (a2anew[j] != j){ 819 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 820 for (k1=0; k1<4; k1++){ 821 dk[k1] = aa[k*4+k1]; 822 aa[k*4+k1] = aa[j*4+k1]; 823 aa[j*4+k1] = dk[k1]; 824 } 825 } 826 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 827 if (i > aj[j]){ 828 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 829 ap = aa + j*4; /* ptr to the beginning of the block */ 830 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 831 ap[1] = ap[2]; 832 ap[2] = dk[1]; 833 } 834 } 835 } 836 ierr = PetscFree(a2anew);CHKERRQ(ierr); 837 } 838 839 /* for each row k */ 840 for (k = 0; k<mbs; k++){ 841 842 /*initialize k-th row with elements nonzero in row perm(k) of A */ 843 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 844 ap = aa + jmin*4; 845 for (j = jmin; j < jmax; j++){ 846 vj = perm_ptr[aj[j]]; /* block col. index */ 847 rtmp_ptr = rtmp + vj*4; 848 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 849 } 850 851 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 852 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 853 i = jl[k]; /* first row to be added to k_th row */ 854 855 while (i < k){ 856 nexti = jl[i]; /* next row to be added to k_th row */ 857 858 /* compute multiplier */ 859 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 860 861 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 862 diag = ba + i*4; 863 u = ba + ili*4; 864 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 865 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 866 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 867 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 868 869 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 870 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 871 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 872 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 873 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 874 875 /* update -U(i,k): ba[ili] = uik */ 876 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 877 878 /* add multiple of row i to k-th row ... */ 879 jmin = ili + 1; jmax = bi[i+1]; 880 if (jmin < jmax){ 881 for (j=jmin; j<jmax; j++) { 882 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 883 rtmp_ptr = rtmp + bj[j]*4; 884 u = ba + j*4; 885 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 886 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 887 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 888 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 889 } 890 891 /* ... add i to row list for next nonzero entry */ 892 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 893 j = bj[jmin]; 894 jl[i] = jl[j]; jl[j] = i; /* update jl */ 895 } 896 i = nexti; 897 } 898 899 /* save nonzero entries in k-th row of U ... */ 900 901 /* invert diagonal block */ 902 diag = ba+k*4; 903 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 904 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 905 906 jmin = bi[k]; jmax = bi[k+1]; 907 if (jmin < jmax) { 908 for (j=jmin; j<jmax; j++){ 909 vj = bj[j]; /* block col. index of U */ 910 u = ba + j*4; 911 rtmp_ptr = rtmp + vj*4; 912 for (k1=0; k1<4; k1++){ 913 *u++ = *rtmp_ptr; 914 *rtmp_ptr++ = 0.0; 915 } 916 } 917 918 /* ... add k to row list for first nonzero entry in k-th row */ 919 il[k] = jmin; 920 i = bj[jmin]; 921 jl[k] = jl[i]; jl[i] = k; 922 } 923 } 924 925 ierr = PetscFree(rtmp);CHKERRQ(ierr); 926 ierr = PetscFree(il);CHKERRQ(ierr); 927 ierr = PetscFree(dk);CHKERRQ(ierr); 928 if (a->permute) { 929 ierr = PetscFree(aa);CHKERRQ(ierr); 930 } 931 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 932 C->factor = FACTOR_CHOLESKY; 933 C->assembled = PETSC_TRUE; 934 C->preallocated = PETSC_TRUE; 935 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 936 PetscFunctionReturn(0); 937 } 938 939 /* 940 Version for when blocks are 2 by 2 Using natural ordering 941 */ 942 #undef __FUNCT__ 943 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 944 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B) 945 { 946 Mat C = *B; 947 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 948 PetscErrorCode ierr; 949 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 950 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 951 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 952 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 953 954 PetscFunctionBegin; 955 /* initialization */ 956 /* il and jl record the first nonzero element in each row of the accessing 957 window U(0:k, k:mbs-1). 958 jl: list of rows to be added to uneliminated rows 959 i>= k: jl(i) is the first row to be added to row i 960 i< k: jl(i) is the row following row i in some list of rows 961 jl(i) = mbs indicates the end of a list 962 il(i): points to the first nonzero element in columns k,...,mbs-1 of 963 row i of U */ 964 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 965 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 966 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 967 jl = il + mbs; 968 for (i=0; i<mbs; i++) { 969 jl[i] = mbs; il[0] = 0; 970 } 971 ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr); 972 uik = dk + 4; 973 974 ai = a->i; aj = a->j; aa = a->a; 975 976 /* for each row k */ 977 for (k = 0; k<mbs; k++){ 978 979 /*initialize k-th row with elements nonzero in row k of A */ 980 jmin = ai[k]; jmax = ai[k+1]; 981 ap = aa + jmin*4; 982 for (j = jmin; j < jmax; j++){ 983 vj = aj[j]; /* block col. index */ 984 rtmp_ptr = rtmp + vj*4; 985 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 986 } 987 988 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 989 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 990 i = jl[k]; /* first row to be added to k_th row */ 991 992 while (i < k){ 993 nexti = jl[i]; /* next row to be added to k_th row */ 994 995 /* compute multiplier */ 996 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 997 998 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 999 diag = ba + i*4; 1000 u = ba + ili*4; 1001 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 1002 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 1003 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 1004 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 1005 1006 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 1007 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 1008 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 1009 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 1010 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 1011 1012 /* update -U(i,k): ba[ili] = uik */ 1013 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 1014 1015 /* add multiple of row i to k-th row ... */ 1016 jmin = ili + 1; jmax = bi[i+1]; 1017 if (jmin < jmax){ 1018 for (j=jmin; j<jmax; j++) { 1019 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 1020 rtmp_ptr = rtmp + bj[j]*4; 1021 u = ba + j*4; 1022 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 1023 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 1024 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 1025 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 1026 } 1027 1028 /* ... add i to row list for next nonzero entry */ 1029 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1030 j = bj[jmin]; 1031 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1032 } 1033 i = nexti; 1034 } 1035 1036 /* save nonzero entries in k-th row of U ... */ 1037 1038 /* invert diagonal block */ 1039 diag = ba+k*4; 1040 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 1041 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 1042 1043 jmin = bi[k]; jmax = bi[k+1]; 1044 if (jmin < jmax) { 1045 for (j=jmin; j<jmax; j++){ 1046 vj = bj[j]; /* block col. index of U */ 1047 u = ba + j*4; 1048 rtmp_ptr = rtmp + vj*4; 1049 for (k1=0; k1<4; k1++){ 1050 *u++ = *rtmp_ptr; 1051 *rtmp_ptr++ = 0.0; 1052 } 1053 } 1054 1055 /* ... add k to row list for first nonzero entry in k-th row */ 1056 il[k] = jmin; 1057 i = bj[jmin]; 1058 jl[k] = jl[i]; jl[i] = k; 1059 } 1060 } 1061 1062 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1063 ierr = PetscFree(il);CHKERRQ(ierr); 1064 ierr = PetscFree(dk);CHKERRQ(ierr); 1065 1066 C->factor = FACTOR_CHOLESKY; 1067 C->assembled = PETSC_TRUE; 1068 C->preallocated = PETSC_TRUE; 1069 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1070 PetscFunctionReturn(0); 1071 } 1072 1073 /* 1074 Numeric U^T*D*U factorization for SBAIJ format. 1075 Version for blocks are 1 by 1. 1076 */ 1077 #undef __FUNCT__ 1078 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 1079 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,MatFactorInfo *info,Mat *B) 1080 { 1081 Mat C = *B; 1082 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1083 IS ip=b->row; 1084 PetscErrorCode ierr; 1085 PetscInt *rip,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol; 1086 PetscInt *ai,*aj,*a2anew; 1087 PetscInt k,jmin,jmax,*jl,*il,col,nexti,ili,nz; 1088 MatScalar *rtmp,*ba=b->a,*bval,*aa,dk,uikdi; 1089 PetscReal zeropivot,rs,shiftnz; 1090 PetscTruth shiftpd; 1091 ChShift_Ctx sctx; 1092 PetscInt newshift; 1093 1094 PetscFunctionBegin; 1095 /* initialization */ 1096 shiftnz = info->shiftnz; 1097 shiftpd = info->shiftpd; 1098 zeropivot = info->zeropivot; 1099 1100 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 1101 if (!a->permute){ 1102 ai = a->i; aj = a->j; aa = a->a; 1103 } else { 1104 ai = a->inew; aj = a->jnew; 1105 nz = ai[mbs]; 1106 ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr); 1107 a2anew = a->a2anew; 1108 bval = a->a; 1109 for (j=0; j<nz; j++){ 1110 aa[a2anew[j]] = *(bval++); 1111 } 1112 } 1113 1114 /* initialization */ 1115 /* il and jl record the first nonzero element in each row of the accessing 1116 window U(0:k, k:mbs-1). 1117 jl: list of rows to be added to uneliminated rows 1118 i>= k: jl(i) is the first row to be added to row i 1119 i< k: jl(i) is the row following row i in some list of rows 1120 jl(i) = mbs indicates the end of a list 1121 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1122 row i of U */ 1123 nz = (2*mbs+1)*sizeof(PetscInt)+mbs*sizeof(MatScalar); 1124 ierr = PetscMalloc(nz,&il);CHKERRQ(ierr); 1125 jl = il + mbs; 1126 rtmp = (MatScalar*)(jl + mbs); 1127 1128 sctx.shift_amount = 0; 1129 sctx.nshift = 0; 1130 do { 1131 sctx.chshift = PETSC_FALSE; 1132 for (i=0; i<mbs; i++) { 1133 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1134 } 1135 1136 for (k = 0; k<mbs; k++){ 1137 /*initialize k-th row by the perm[k]-th row of A */ 1138 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 1139 bval = ba + bi[k]; 1140 for (j = jmin; j < jmax; j++){ 1141 col = rip[aj[j]]; 1142 rtmp[col] = aa[j]; 1143 *bval++ = 0.0; /* for in-place factorization */ 1144 } 1145 1146 /* shift the diagonal of the matrix */ 1147 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1148 1149 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1150 dk = rtmp[k]; 1151 i = jl[k]; /* first row to be added to k_th row */ 1152 1153 while (i < k){ 1154 nexti = jl[i]; /* next row to be added to k_th row */ 1155 1156 /* compute multiplier, update diag(k) and U(i,k) */ 1157 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1158 uikdi = - ba[ili]*ba[bi[i]]; /* diagonal(k) */ 1159 dk += uikdi*ba[ili]; 1160 ba[ili] = uikdi; /* -U(i,k) */ 1161 1162 /* add multiple of row i to k-th row */ 1163 jmin = ili + 1; jmax = bi[i+1]; 1164 if (jmin < jmax){ 1165 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1166 /* update il and jl for row i */ 1167 il[i] = jmin; 1168 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1169 } 1170 i = nexti; 1171 } 1172 1173 /* shift the diagonals when zero pivot is detected */ 1174 /* compute rs=sum of abs(off-diagonal) */ 1175 rs = 0.0; 1176 jmin = bi[k]+1; 1177 nz = bi[k+1] - jmin; 1178 if (nz){ 1179 bcol = bj + jmin; 1180 while (nz--){ 1181 rs += PetscAbsScalar(rtmp[*bcol]); 1182 bcol++; 1183 } 1184 } 1185 1186 sctx.rs = rs; 1187 sctx.pv = dk; 1188 ierr = MatCholeskyCheckShift_inline(info,sctx,newshift);CHKERRQ(ierr); 1189 if (newshift == 1){ 1190 break; /* sctx.shift_amount is updated */ 1191 } else if (newshift == -1){ 1192 SETERRQ4(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g * rs %g",k,PetscAbsScalar(dk),zeropivot,rs); 1193 } 1194 1195 /* copy data into U(k,:) */ 1196 ba[bi[k]] = 1.0/dk; /* U(k,k) */ 1197 jmin = bi[k]+1; jmax = bi[k+1]; 1198 if (jmin < jmax) { 1199 for (j=jmin; j<jmax; j++){ 1200 col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0; 1201 } 1202 /* add the k-th row into il and jl */ 1203 il[k] = jmin; 1204 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1205 } 1206 } 1207 } while (sctx.chshift); 1208 ierr = PetscFree(il);CHKERRQ(ierr); 1209 if (a->permute){ierr = PetscFree(aa);CHKERRQ(ierr);} 1210 1211 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 1212 C->factor = FACTOR_CHOLESKY; 1213 C->assembled = PETSC_TRUE; 1214 C->preallocated = PETSC_TRUE; 1215 ierr = PetscLogFlops(C->m);CHKERRQ(ierr); 1216 if (sctx.nshift){ 1217 if (shiftnz) { 1218 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1: number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1219 } else if (shiftpd) { 1220 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1: number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1221 } 1222 } 1223 PetscFunctionReturn(0); 1224 } 1225 1226 /* 1227 Version for when blocks are 1 by 1 Using natural ordering 1228 */ 1229 #undef __FUNCT__ 1230 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering" 1231 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B) 1232 { 1233 Mat C = *B; 1234 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1235 PetscErrorCode ierr; 1236 PetscInt i,j,mbs = a->mbs; 1237 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 1238 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz; 1239 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 1240 PetscReal zeropivot,rs,shiftnz; 1241 PetscTruth shiftpd; 1242 ChShift_Ctx sctx; 1243 PetscInt newshift; 1244 1245 PetscFunctionBegin; 1246 /* initialization */ 1247 shiftnz = info->shiftnz; 1248 shiftpd = info->shiftpd; 1249 zeropivot = info->zeropivot; 1250 1251 /* il and jl record the first nonzero element in each row of the accessing 1252 window U(0:k, k:mbs-1). 1253 jl: list of rows to be added to uneliminated rows 1254 i>= k: jl(i) is the first row to be added to row i 1255 i< k: jl(i) is the row following row i in some list of rows 1256 jl(i) = mbs indicates the end of a list 1257 il(i): points to the first nonzero element in U(i,k:mbs-1) 1258 */ 1259 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1260 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 1261 jl = il + mbs; 1262 1263 sctx.shift_amount = 0; 1264 sctx.nshift = 0; 1265 do { 1266 sctx.chshift = PETSC_FALSE; 1267 for (i=0; i<mbs; i++) { 1268 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1269 } 1270 1271 for (k = 0; k<mbs; k++){ 1272 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1273 nz = ai[k+1] - ai[k]; 1274 acol = aj + ai[k]; 1275 aval = aa + ai[k]; 1276 bval = ba + bi[k]; 1277 while (nz -- ){ 1278 rtmp[*acol++] = *aval++; 1279 *bval++ = 0.0; /* for in-place factorization */ 1280 } 1281 1282 /* shift the diagonal of the matrix */ 1283 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1284 1285 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1286 dk = rtmp[k]; 1287 i = jl[k]; /* first row to be added to k_th row */ 1288 1289 while (i < k){ 1290 nexti = jl[i]; /* next row to be added to k_th row */ 1291 /* compute multiplier, update D(k) and U(i,k) */ 1292 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1293 uikdi = - ba[ili]*ba[bi[i]]; 1294 dk += uikdi*ba[ili]; 1295 ba[ili] = uikdi; /* -U(i,k) */ 1296 1297 /* add multiple of row i to k-th row ... */ 1298 jmin = ili + 1; 1299 nz = bi[i+1] - jmin; 1300 if (nz > 0){ 1301 bcol = bj + jmin; 1302 bval = ba + jmin; 1303 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 1304 /* update il and jl for i-th row */ 1305 il[i] = jmin; 1306 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1307 } 1308 i = nexti; 1309 } 1310 1311 /* shift the diagonals when zero pivot is detected */ 1312 /* compute rs=sum of abs(off-diagonal) */ 1313 rs = 0.0; 1314 jmin = bi[k]+1; 1315 nz = bi[k+1] - jmin; 1316 if (nz){ 1317 bcol = bj + jmin; 1318 while (nz--){ 1319 rs += PetscAbsScalar(rtmp[*bcol]); 1320 bcol++; 1321 } 1322 } 1323 1324 sctx.rs = rs; 1325 sctx.pv = dk; 1326 ierr = MatCholeskyCheckShift_inline(info,sctx,newshift);CHKERRQ(ierr); 1327 if (newshift == 1){ 1328 break; /* sctx.shift_amount is updated */ 1329 } else if (newshift == -1){ 1330 SETERRQ4(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g * rs %g",k,PetscAbsScalar(dk),zeropivot,rs); 1331 } 1332 1333 /* copy data into U(k,:) */ 1334 ba[bi[k]] = 1.0/dk; 1335 jmin = bi[k]+1; 1336 nz = bi[k+1] - jmin; 1337 if (nz){ 1338 bcol = bj + jmin; 1339 bval = ba + jmin; 1340 while (nz--){ 1341 *bval++ = rtmp[*bcol]; 1342 rtmp[*bcol++] = 0.0; 1343 } 1344 /* add k-th row into il and jl */ 1345 il[k] = jmin; 1346 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1347 } 1348 } /* end of for (k = 0; k<mbs; k++) */ 1349 } while (sctx.chshift); 1350 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1351 ierr = PetscFree(il);CHKERRQ(ierr); 1352 1353 C->factor = FACTOR_CHOLESKY; 1354 C->assembled = PETSC_TRUE; 1355 C->preallocated = PETSC_TRUE; 1356 ierr = PetscLogFlops(C->m);CHKERRQ(ierr); 1357 if (sctx.nshift){ 1358 if (shiftnz) { 1359 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering: number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1360 } else if (shiftpd) { 1361 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering: number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1362 } 1363 } 1364 PetscFunctionReturn(0); 1365 } 1366 1367 #undef __FUNCT__ 1368 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ" 1369 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info) 1370 { 1371 PetscErrorCode ierr; 1372 Mat C; 1373 1374 PetscFunctionBegin; 1375 ierr = MatCholeskyFactorSymbolic(A,perm,info,&C);CHKERRQ(ierr); 1376 ierr = MatCholeskyFactorNumeric(A,info,&C);CHKERRQ(ierr); 1377 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 1378 PetscFunctionReturn(0); 1379 } 1380 1381 1382