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(*B,bs,0,PETSC_NULL);CHKERRQ(ierr); 183 184 /* ierr = PetscLogObjectParent(*B,iperm);CHKERRQ(ierr); */ 185 b = (Mat_SeqSBAIJ*)(*B)->data; 186 ierr = PetscFree(b->imax);CHKERRQ(ierr); 187 b->singlemalloc = PETSC_FALSE; 188 /* the next line frees the default space generated by the Create() */ 189 ierr = PetscFree(b->a);CHKERRQ(ierr); 190 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 191 ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr); 192 b->j = ju; 193 b->i = iu; 194 b->diag = 0; 195 b->ilen = 0; 196 b->imax = 0; 197 b->row = perm; 198 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 199 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 200 b->icol = perm; 201 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 202 ierr = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 203 /* In b structure: Free imax, ilen, old a, old j. 204 Allocate idnew, solve_work, new a, new j */ 205 ierr = PetscLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 206 b->maxnz = b->nz = iu[mbs]; 207 208 (*B)->factor = FACTOR_CHOLESKY; 209 (*B)->info.factor_mallocs = reallocs; 210 (*B)->info.fill_ratio_given = f; 211 if (ai[mbs] != 0) { 212 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 213 } else { 214 (*B)->info.fill_ratio_needed = 0.0; 215 } 216 217 if (perm_identity){ 218 switch (bs) { 219 case 1: 220 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 221 (*B)->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 222 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"));CHKERRQ(ierr); 223 break; 224 case 2: 225 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering; 226 (*B)->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering; 227 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"));CHKERRQ(ierr); 228 break; 229 case 3: 230 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering; 231 (*B)->ops->solve = MatSolve_SeqSBAIJ_3_NaturalOrdering; 232 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"));CHKERRQ(ierr); 233 break; 234 case 4: 235 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering; 236 (*B)->ops->solve = MatSolve_SeqSBAIJ_4_NaturalOrdering; 237 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"));CHKERRQ(ierr); 238 break; 239 case 5: 240 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering; 241 (*B)->ops->solve = MatSolve_SeqSBAIJ_5_NaturalOrdering; 242 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"));CHKERRQ(ierr); 243 break; 244 case 6: 245 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering; 246 (*B)->ops->solve = MatSolve_SeqSBAIJ_6_NaturalOrdering; 247 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"));CHKERRQ(ierr); 248 break; 249 case 7: 250 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering; 251 (*B)->ops->solve = MatSolve_SeqSBAIJ_7_NaturalOrdering; 252 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"));CHKERRQ(ierr); 253 break; 254 default: 255 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering; 256 (*B)->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering; 257 ierr = PetscLogInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"));CHKERRQ(ierr); 258 break; 259 } 260 } 261 PetscFunctionReturn(0); 262 } 263 /* 264 Symbolic U^T*D*U factorization for SBAIJ format. 265 */ 266 #include "petscbt.h" 267 #include "src/mat/utils/freespace.h" 268 #undef __FUNCT__ 269 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 270 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info,Mat *fact) 271 { 272 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data; 273 Mat_SeqSBAIJ *b; 274 Mat B; 275 PetscErrorCode ierr; 276 PetscTruth perm_identity; 277 PetscReal fill = info->fill; 278 PetscInt *rip,i,mbs=a->mbs,bs=A->bs,*ai,*aj,reallocs=0,prow; 279 PetscInt *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow; 280 PetscInt nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr; 281 FreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL; 282 PetscBT lnkbt; 283 284 PetscFunctionBegin; 285 /* 286 This code originally uses Modified Sparse Row (MSR) storage 287 (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise! 288 Then it is rewritten so the factor B takes seqsbaij format. However the associated 289 MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity, 290 thus the original code in MSR format is still used for these cases. 291 The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever 292 MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor. 293 */ 294 if (bs > 1){ 295 ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(A,perm,info,fact);CHKERRQ(ierr); 296 PetscFunctionReturn(0); 297 } 298 299 /* check whether perm is the identity mapping */ 300 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 301 302 if (perm_identity){ 303 a->permute = PETSC_FALSE; 304 ai = a->i; aj = a->j; 305 } else { 306 a->permute = PETSC_TRUE; 307 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 308 ai = a->inew; aj = a->jnew; 309 } 310 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 311 312 /* initialization */ 313 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr); 314 ui[0] = 0; 315 316 /* jl: linked list for storing indices of the pivot rows 317 il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */ 318 ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt)+mbs*sizeof(PetscInt*),&jl);CHKERRQ(ierr); 319 il = jl + mbs; 320 cols = il + mbs; 321 ui_ptr = (PetscInt**)(cols + mbs); 322 323 for (i=0; i<mbs; i++){ 324 jl[i] = mbs; il[i] = 0; 325 } 326 327 /* create and initialize a linked list for storing column indices of the active row k */ 328 nlnk = mbs + 1; 329 ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 330 331 /* initial FreeSpace size is fill*(ai[mbs]+1) */ 332 ierr = GetMoreSpace((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr); 333 current_space = free_space; 334 335 for (k=0; k<mbs; k++){ /* for each active row k */ 336 /* initialize lnk by the column indices of row rip[k] of A */ 337 nzk = 0; 338 ncols = ai[rip[k]+1] - ai[rip[k]]; 339 for (j=0; j<ncols; j++){ 340 i = *(aj + ai[rip[k]] + j); 341 cols[j] = rip[i]; 342 } 343 ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 344 nzk += nlnk; 345 346 /* update lnk by computing fill-in for each pivot row to be merged in */ 347 prow = jl[k]; /* 1st pivot row */ 348 349 while (prow < k){ 350 nextprow = jl[prow]; 351 /* merge prow into k-th row */ 352 jmin = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */ 353 jmax = ui[prow+1]; 354 ncols = jmax-jmin; 355 uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */ 356 ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr); 357 nzk += nlnk; 358 359 /* update il and jl for prow */ 360 if (jmin < jmax){ 361 il[prow] = jmin; 362 j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow; 363 } 364 prow = nextprow; 365 } 366 367 /* if free space is not available, make more free space */ 368 if (current_space->local_remaining<nzk) { 369 i = mbs - k + 1; /* num of unfactored rows */ 370 i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */ 371 ierr = GetMoreSpace(i,¤t_space);CHKERRQ(ierr); 372 reallocs++; 373 } 374 375 /* copy data into free space, then initialize lnk */ 376 ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr); 377 378 /* add the k-th row into il and jl */ 379 if (nzk-1 > 0){ 380 i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 381 jl[k] = jl[i]; jl[i] = k; 382 il[k] = ui[k] + 1; 383 } 384 ui_ptr[k] = current_space->array; 385 current_space->array += nzk; 386 current_space->local_used += nzk; 387 current_space->local_remaining -= nzk; 388 389 ui[k+1] = ui[k] + nzk; 390 } 391 392 #if defined(PETSC_USE_DEBUG) 393 if (ai[mbs] != 0) { 394 PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 395 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,fill,af));CHKERRQ(ierr); 396 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af));CHKERRQ(ierr); 397 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g) for best performance.\n",af));CHKERRQ(ierr); 398 } else { 399 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"));CHKERRQ(ierr); 400 } 401 #endif 402 403 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 404 ierr = PetscFree(jl);CHKERRQ(ierr); 405 406 /* destroy list of free space and other temporary array(s) */ 407 ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr); 408 ierr = MakeSpaceContiguous(&free_space,uj);CHKERRQ(ierr); 409 ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr); 410 411 /* put together the new matrix in MATSEQSBAIJ format */ 412 ierr = MatCreate(PETSC_COMM_SELF,mbs,mbs,mbs,mbs,fact);CHKERRQ(ierr); 413 B = *fact; 414 ierr = MatSetType(B,MATSEQSBAIJ);CHKERRQ(ierr); 415 ierr = MatSeqSBAIJSetPreallocation(B,bs,0,PETSC_NULL);CHKERRQ(ierr); 416 417 b = (Mat_SeqSBAIJ*)B->data; 418 ierr = PetscFree(b->imax);CHKERRQ(ierr); 419 b->singlemalloc = PETSC_FALSE; 420 /* the next line frees the default space generated by the Create() */ 421 ierr = PetscFree(b->a);CHKERRQ(ierr); 422 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 423 ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr); 424 b->j = uj; 425 b->i = ui; 426 b->diag = 0; 427 b->ilen = 0; 428 b->imax = 0; 429 b->row = perm; 430 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 431 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 432 b->icol = perm; 433 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 434 ierr = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 435 ierr = PetscLogObjectMemory(B,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr); 436 b->maxnz = b->nz = ui[mbs]; 437 438 B->factor = FACTOR_CHOLESKY; 439 B->info.factor_mallocs = reallocs; 440 B->info.fill_ratio_given = fill; 441 if (ai[mbs] != 0) { 442 B->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]); 443 } else { 444 B->info.fill_ratio_needed = 0.0; 445 } 446 447 if (perm_identity){ 448 switch (bs) { 449 case 1: 450 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 451 B->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 452 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"));CHKERRQ(ierr); 453 break; 454 case 2: 455 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering; 456 B->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering; 457 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"));CHKERRQ(ierr); 458 break; 459 case 3: 460 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering; 461 B->ops->solve = MatSolve_SeqSBAIJ_3_NaturalOrdering; 462 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"));CHKERRQ(ierr); 463 break; 464 case 4: 465 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering; 466 B->ops->solve = MatSolve_SeqSBAIJ_4_NaturalOrdering; 467 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"));CHKERRQ(ierr); 468 break; 469 case 5: 470 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering; 471 B->ops->solve = MatSolve_SeqSBAIJ_5_NaturalOrdering; 472 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"));CHKERRQ(ierr); 473 break; 474 case 6: 475 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering; 476 B->ops->solve = MatSolve_SeqSBAIJ_6_NaturalOrdering; 477 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"));CHKERRQ(ierr); 478 break; 479 case 7: 480 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering; 481 B->ops->solve = MatSolve_SeqSBAIJ_7_NaturalOrdering; 482 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"));CHKERRQ(ierr); 483 break; 484 default: 485 B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering; 486 B->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering; 487 ierr = PetscLogInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"));CHKERRQ(ierr); 488 break; 489 } 490 } 491 PetscFunctionReturn(0); 492 } 493 #undef __FUNCT__ 494 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 495 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,MatFactorInfo *info,Mat *B) 496 { 497 Mat C = *B; 498 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 499 IS perm = b->row; 500 PetscErrorCode ierr; 501 PetscInt *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 502 PetscInt *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 503 PetscInt bs=A->bs,bs2 = a->bs2; 504 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 505 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 506 MatScalar *work; 507 PetscInt *pivots; 508 509 PetscFunctionBegin; 510 /* initialization */ 511 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 512 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 513 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 514 jl = il + mbs; 515 for (i=0; i<mbs; i++) { 516 jl[i] = mbs; il[0] = 0; 517 } 518 ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr); 519 uik = dk + bs2; 520 work = uik + bs2; 521 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 522 523 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 524 525 /* check permutation */ 526 if (!a->permute){ 527 ai = a->i; aj = a->j; aa = a->a; 528 } else { 529 ai = a->inew; aj = a->jnew; 530 ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 531 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 532 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 533 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 534 535 for (i=0; i<mbs; i++){ 536 jmin = ai[i]; jmax = ai[i+1]; 537 for (j=jmin; j<jmax; j++){ 538 while (a2anew[j] != j){ 539 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 540 for (k1=0; k1<bs2; k1++){ 541 dk[k1] = aa[k*bs2+k1]; 542 aa[k*bs2+k1] = aa[j*bs2+k1]; 543 aa[j*bs2+k1] = dk[k1]; 544 } 545 } 546 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 547 if (i > aj[j]){ 548 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 549 ap = aa + j*bs2; /* ptr to the beginning of j-th block of aa */ 550 for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 551 for (k=0; k<bs; k++){ /* j-th block of aa <- dk^T */ 552 for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1]; 553 } 554 } 555 } 556 } 557 ierr = PetscFree(a2anew);CHKERRQ(ierr); 558 } 559 560 /* for each row k */ 561 for (k = 0; k<mbs; k++){ 562 563 /*initialize k-th row with elements nonzero in row perm(k) of A */ 564 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 565 566 ap = aa + jmin*bs2; 567 for (j = jmin; j < jmax; j++){ 568 vj = perm_ptr[aj[j]]; /* block col. index */ 569 rtmp_ptr = rtmp + vj*bs2; 570 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 571 } 572 573 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 574 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 575 i = jl[k]; /* first row to be added to k_th row */ 576 577 while (i < k){ 578 nexti = jl[i]; /* next row to be added to k_th row */ 579 580 /* compute multiplier */ 581 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 582 583 /* uik = -inv(Di)*U_bar(i,k) */ 584 diag = ba + i*bs2; 585 u = ba + ili*bs2; 586 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 587 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 588 589 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 590 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 591 592 /* update -U(i,k) */ 593 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 594 595 /* add multiple of row i to k-th row ... */ 596 jmin = ili + 1; jmax = bi[i+1]; 597 if (jmin < jmax){ 598 for (j=jmin; j<jmax; j++) { 599 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 600 rtmp_ptr = rtmp + bj[j]*bs2; 601 u = ba + j*bs2; 602 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 603 } 604 605 /* ... add i to row list for next nonzero entry */ 606 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 607 j = bj[jmin]; 608 jl[i] = jl[j]; jl[j] = i; /* update jl */ 609 } 610 i = nexti; 611 } 612 613 /* save nonzero entries in k-th row of U ... */ 614 615 /* invert diagonal block */ 616 diag = ba+k*bs2; 617 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 618 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 619 620 jmin = bi[k]; jmax = bi[k+1]; 621 if (jmin < jmax) { 622 for (j=jmin; j<jmax; j++){ 623 vj = bj[j]; /* block col. index of U */ 624 u = ba + j*bs2; 625 rtmp_ptr = rtmp + vj*bs2; 626 for (k1=0; k1<bs2; k1++){ 627 *u++ = *rtmp_ptr; 628 *rtmp_ptr++ = 0.0; 629 } 630 } 631 632 /* ... add k to row list for first nonzero entry in k-th row */ 633 il[k] = jmin; 634 i = bj[jmin]; 635 jl[k] = jl[i]; jl[i] = k; 636 } 637 } 638 639 ierr = PetscFree(rtmp);CHKERRQ(ierr); 640 ierr = PetscFree(il);CHKERRQ(ierr); 641 ierr = PetscFree(dk);CHKERRQ(ierr); 642 ierr = PetscFree(pivots);CHKERRQ(ierr); 643 if (a->permute){ 644 ierr = PetscFree(aa);CHKERRQ(ierr); 645 } 646 647 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 648 C->factor = FACTOR_CHOLESKY; 649 C->assembled = PETSC_TRUE; 650 C->preallocated = PETSC_TRUE; 651 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 652 PetscFunctionReturn(0); 653 } 654 655 #undef __FUNCT__ 656 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering" 657 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B) 658 { 659 Mat C = *B; 660 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 661 PetscErrorCode ierr; 662 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 663 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 664 PetscInt bs=A->bs,bs2 = a->bs2; 665 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 666 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 667 MatScalar *work; 668 PetscInt *pivots; 669 670 PetscFunctionBegin; 671 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 672 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 673 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 674 jl = il + mbs; 675 for (i=0; i<mbs; i++) { 676 jl[i] = mbs; il[0] = 0; 677 } 678 ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr); 679 uik = dk + bs2; 680 work = uik + bs2; 681 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 682 683 ai = a->i; aj = a->j; aa = a->a; 684 685 /* for each row k */ 686 for (k = 0; k<mbs; k++){ 687 688 /*initialize k-th row with elements nonzero in row k of A */ 689 jmin = ai[k]; jmax = ai[k+1]; 690 ap = aa + jmin*bs2; 691 for (j = jmin; j < jmax; j++){ 692 vj = aj[j]; /* block col. index */ 693 rtmp_ptr = rtmp + vj*bs2; 694 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 695 } 696 697 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 698 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 699 i = jl[k]; /* first row to be added to k_th row */ 700 701 while (i < k){ 702 nexti = jl[i]; /* next row to be added to k_th row */ 703 704 /* compute multiplier */ 705 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 706 707 /* uik = -inv(Di)*U_bar(i,k) */ 708 diag = ba + i*bs2; 709 u = ba + ili*bs2; 710 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 711 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 712 713 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 714 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 715 716 /* update -U(i,k) */ 717 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 718 719 /* add multiple of row i to k-th row ... */ 720 jmin = ili + 1; jmax = bi[i+1]; 721 if (jmin < jmax){ 722 for (j=jmin; j<jmax; j++) { 723 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 724 rtmp_ptr = rtmp + bj[j]*bs2; 725 u = ba + j*bs2; 726 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 727 } 728 729 /* ... add i to row list for next nonzero entry */ 730 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 731 j = bj[jmin]; 732 jl[i] = jl[j]; jl[j] = i; /* update jl */ 733 } 734 i = nexti; 735 } 736 737 /* save nonzero entries in k-th row of U ... */ 738 739 /* invert diagonal block */ 740 diag = ba+k*bs2; 741 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 742 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 743 744 jmin = bi[k]; jmax = bi[k+1]; 745 if (jmin < jmax) { 746 for (j=jmin; j<jmax; j++){ 747 vj = bj[j]; /* block col. index of U */ 748 u = ba + j*bs2; 749 rtmp_ptr = rtmp + vj*bs2; 750 for (k1=0; k1<bs2; k1++){ 751 *u++ = *rtmp_ptr; 752 *rtmp_ptr++ = 0.0; 753 } 754 } 755 756 /* ... add k to row list for first nonzero entry in k-th row */ 757 il[k] = jmin; 758 i = bj[jmin]; 759 jl[k] = jl[i]; jl[i] = k; 760 } 761 } 762 763 ierr = PetscFree(rtmp);CHKERRQ(ierr); 764 ierr = PetscFree(il);CHKERRQ(ierr); 765 ierr = PetscFree(dk);CHKERRQ(ierr); 766 ierr = PetscFree(pivots);CHKERRQ(ierr); 767 768 C->factor = FACTOR_CHOLESKY; 769 C->assembled = PETSC_TRUE; 770 C->preallocated = PETSC_TRUE; 771 ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 772 PetscFunctionReturn(0); 773 } 774 775 /* 776 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 777 Version for blocks 2 by 2. 778 */ 779 #undef __FUNCT__ 780 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 781 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,MatFactorInfo *info,Mat *B) 782 { 783 Mat C = *B; 784 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 785 IS perm = b->row; 786 PetscErrorCode ierr; 787 PetscInt *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 788 PetscInt *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 789 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 790 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 791 792 PetscFunctionBegin; 793 /* initialization */ 794 /* il and jl record the first nonzero element in each row of the accessing 795 window U(0:k, k:mbs-1). 796 jl: list of rows to be added to uneliminated rows 797 i>= k: jl(i) is the first row to be added to row i 798 i< k: jl(i) is the row following row i in some list of rows 799 jl(i) = mbs indicates the end of a list 800 il(i): points to the first nonzero element in columns k,...,mbs-1 of 801 row i of U */ 802 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 803 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 804 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 805 jl = il + mbs; 806 for (i=0; i<mbs; i++) { 807 jl[i] = mbs; il[0] = 0; 808 } 809 ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr); 810 uik = dk + 4; 811 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 812 813 /* check permutation */ 814 if (!a->permute){ 815 ai = a->i; aj = a->j; aa = a->a; 816 } else { 817 ai = a->inew; aj = a->jnew; 818 ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 819 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 820 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 821 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 822 823 for (i=0; i<mbs; i++){ 824 jmin = ai[i]; jmax = ai[i+1]; 825 for (j=jmin; j<jmax; j++){ 826 while (a2anew[j] != j){ 827 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 828 for (k1=0; k1<4; k1++){ 829 dk[k1] = aa[k*4+k1]; 830 aa[k*4+k1] = aa[j*4+k1]; 831 aa[j*4+k1] = dk[k1]; 832 } 833 } 834 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 835 if (i > aj[j]){ 836 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 837 ap = aa + j*4; /* ptr to the beginning of the block */ 838 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 839 ap[1] = ap[2]; 840 ap[2] = dk[1]; 841 } 842 } 843 } 844 ierr = PetscFree(a2anew);CHKERRQ(ierr); 845 } 846 847 /* for each row k */ 848 for (k = 0; k<mbs; k++){ 849 850 /*initialize k-th row with elements nonzero in row perm(k) of A */ 851 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 852 ap = aa + jmin*4; 853 for (j = jmin; j < jmax; j++){ 854 vj = perm_ptr[aj[j]]; /* block col. index */ 855 rtmp_ptr = rtmp + vj*4; 856 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 857 } 858 859 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 860 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 861 i = jl[k]; /* first row to be added to k_th row */ 862 863 while (i < k){ 864 nexti = jl[i]; /* next row to be added to k_th row */ 865 866 /* compute multiplier */ 867 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 868 869 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 870 diag = ba + i*4; 871 u = ba + ili*4; 872 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 873 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 874 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 875 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 876 877 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 878 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 879 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 880 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 881 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 882 883 /* update -U(i,k): ba[ili] = uik */ 884 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 885 886 /* add multiple of row i to k-th row ... */ 887 jmin = ili + 1; jmax = bi[i+1]; 888 if (jmin < jmax){ 889 for (j=jmin; j<jmax; j++) { 890 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 891 rtmp_ptr = rtmp + bj[j]*4; 892 u = ba + j*4; 893 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 894 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 895 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 896 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 897 } 898 899 /* ... add i to row list for next nonzero entry */ 900 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 901 j = bj[jmin]; 902 jl[i] = jl[j]; jl[j] = i; /* update jl */ 903 } 904 i = nexti; 905 } 906 907 /* save nonzero entries in k-th row of U ... */ 908 909 /* invert diagonal block */ 910 diag = ba+k*4; 911 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 912 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 913 914 jmin = bi[k]; jmax = bi[k+1]; 915 if (jmin < jmax) { 916 for (j=jmin; j<jmax; j++){ 917 vj = bj[j]; /* block col. index of U */ 918 u = ba + j*4; 919 rtmp_ptr = rtmp + vj*4; 920 for (k1=0; k1<4; k1++){ 921 *u++ = *rtmp_ptr; 922 *rtmp_ptr++ = 0.0; 923 } 924 } 925 926 /* ... add k to row list for first nonzero entry in k-th row */ 927 il[k] = jmin; 928 i = bj[jmin]; 929 jl[k] = jl[i]; jl[i] = k; 930 } 931 } 932 933 ierr = PetscFree(rtmp);CHKERRQ(ierr); 934 ierr = PetscFree(il);CHKERRQ(ierr); 935 ierr = PetscFree(dk);CHKERRQ(ierr); 936 if (a->permute) { 937 ierr = PetscFree(aa);CHKERRQ(ierr); 938 } 939 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 940 C->factor = FACTOR_CHOLESKY; 941 C->assembled = PETSC_TRUE; 942 C->preallocated = PETSC_TRUE; 943 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 944 PetscFunctionReturn(0); 945 } 946 947 /* 948 Version for when blocks are 2 by 2 Using natural ordering 949 */ 950 #undef __FUNCT__ 951 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 952 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B) 953 { 954 Mat C = *B; 955 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 956 PetscErrorCode ierr; 957 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 958 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 959 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 960 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 961 962 PetscFunctionBegin; 963 /* initialization */ 964 /* il and jl record the first nonzero element in each row of the accessing 965 window U(0:k, k:mbs-1). 966 jl: list of rows to be added to uneliminated rows 967 i>= k: jl(i) is the first row to be added to row i 968 i< k: jl(i) is the row following row i in some list of rows 969 jl(i) = mbs indicates the end of a list 970 il(i): points to the first nonzero element in columns k,...,mbs-1 of 971 row i of U */ 972 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 973 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 974 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 975 jl = il + mbs; 976 for (i=0; i<mbs; i++) { 977 jl[i] = mbs; il[0] = 0; 978 } 979 ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr); 980 uik = dk + 4; 981 982 ai = a->i; aj = a->j; aa = a->a; 983 984 /* for each row k */ 985 for (k = 0; k<mbs; k++){ 986 987 /*initialize k-th row with elements nonzero in row k of A */ 988 jmin = ai[k]; jmax = ai[k+1]; 989 ap = aa + jmin*4; 990 for (j = jmin; j < jmax; j++){ 991 vj = aj[j]; /* block col. index */ 992 rtmp_ptr = rtmp + vj*4; 993 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 994 } 995 996 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 997 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 998 i = jl[k]; /* first row to be added to k_th row */ 999 1000 while (i < k){ 1001 nexti = jl[i]; /* next row to be added to k_th row */ 1002 1003 /* compute multiplier */ 1004 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1005 1006 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 1007 diag = ba + i*4; 1008 u = ba + ili*4; 1009 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 1010 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 1011 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 1012 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 1013 1014 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 1015 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 1016 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 1017 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 1018 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 1019 1020 /* update -U(i,k): ba[ili] = uik */ 1021 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 1022 1023 /* add multiple of row i to k-th row ... */ 1024 jmin = ili + 1; jmax = bi[i+1]; 1025 if (jmin < jmax){ 1026 for (j=jmin; j<jmax; j++) { 1027 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 1028 rtmp_ptr = rtmp + bj[j]*4; 1029 u = ba + j*4; 1030 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 1031 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 1032 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 1033 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 1034 } 1035 1036 /* ... add i to row list for next nonzero entry */ 1037 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1038 j = bj[jmin]; 1039 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1040 } 1041 i = nexti; 1042 } 1043 1044 /* save nonzero entries in k-th row of U ... */ 1045 1046 /* invert diagonal block */ 1047 diag = ba+k*4; 1048 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 1049 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 1050 1051 jmin = bi[k]; jmax = bi[k+1]; 1052 if (jmin < jmax) { 1053 for (j=jmin; j<jmax; j++){ 1054 vj = bj[j]; /* block col. index of U */ 1055 u = ba + j*4; 1056 rtmp_ptr = rtmp + vj*4; 1057 for (k1=0; k1<4; k1++){ 1058 *u++ = *rtmp_ptr; 1059 *rtmp_ptr++ = 0.0; 1060 } 1061 } 1062 1063 /* ... add k to row list for first nonzero entry in k-th row */ 1064 il[k] = jmin; 1065 i = bj[jmin]; 1066 jl[k] = jl[i]; jl[i] = k; 1067 } 1068 } 1069 1070 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1071 ierr = PetscFree(il);CHKERRQ(ierr); 1072 ierr = PetscFree(dk);CHKERRQ(ierr); 1073 1074 C->factor = FACTOR_CHOLESKY; 1075 C->assembled = PETSC_TRUE; 1076 C->preallocated = PETSC_TRUE; 1077 ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */ 1078 PetscFunctionReturn(0); 1079 } 1080 1081 /* 1082 Numeric U^T*D*U factorization for SBAIJ format. 1083 Version for blocks are 1 by 1. 1084 */ 1085 #undef __FUNCT__ 1086 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 1087 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,MatFactorInfo *info,Mat *B) 1088 { 1089 Mat C = *B; 1090 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1091 IS ip=b->row; 1092 PetscErrorCode ierr; 1093 PetscInt *rip,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol; 1094 PetscInt *ai,*aj,*a2anew; 1095 PetscInt k,jmin,jmax,*jl,*il,col,nexti,ili,nz; 1096 MatScalar *rtmp,*ba=b->a,*bval,*aa,dk,uikdi; 1097 PetscReal zeropivot,rs,shiftnz; 1098 PetscTruth shiftpd; 1099 ChShift_Ctx sctx; 1100 PetscInt newshift; 1101 1102 PetscFunctionBegin; 1103 /* initialization */ 1104 shiftnz = info->shiftnz; 1105 shiftpd = info->shiftpd; 1106 zeropivot = info->zeropivot; 1107 1108 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 1109 if (!a->permute){ 1110 ai = a->i; aj = a->j; aa = a->a; 1111 } else { 1112 ai = a->inew; aj = a->jnew; 1113 nz = ai[mbs]; 1114 ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr); 1115 a2anew = a->a2anew; 1116 bval = a->a; 1117 for (j=0; j<nz; j++){ 1118 aa[a2anew[j]] = *(bval++); 1119 } 1120 } 1121 1122 /* initialization */ 1123 /* il and jl record the first nonzero element in each row of the accessing 1124 window U(0:k, k:mbs-1). 1125 jl: list of rows to be added to uneliminated rows 1126 i>= k: jl(i) is the first row to be added to row i 1127 i< k: jl(i) is the row following row i in some list of rows 1128 jl(i) = mbs indicates the end of a list 1129 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1130 row i of U */ 1131 nz = (2*mbs+1)*sizeof(PetscInt)+mbs*sizeof(MatScalar); 1132 ierr = PetscMalloc(nz,&il);CHKERRQ(ierr); 1133 jl = il + mbs; 1134 rtmp = (MatScalar*)(jl + mbs); 1135 1136 sctx.shift_amount = 0; 1137 sctx.nshift = 0; 1138 do { 1139 sctx.chshift = PETSC_FALSE; 1140 for (i=0; i<mbs; i++) { 1141 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1142 } 1143 1144 for (k = 0; k<mbs; k++){ 1145 /*initialize k-th row by the perm[k]-th row of A */ 1146 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 1147 bval = ba + bi[k]; 1148 for (j = jmin; j < jmax; j++){ 1149 col = rip[aj[j]]; 1150 rtmp[col] = aa[j]; 1151 *bval++ = 0.0; /* for in-place factorization */ 1152 } 1153 1154 /* shift the diagonal of the matrix */ 1155 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1156 1157 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1158 dk = rtmp[k]; 1159 i = jl[k]; /* first row to be added to k_th row */ 1160 1161 while (i < k){ 1162 nexti = jl[i]; /* next row to be added to k_th row */ 1163 1164 /* compute multiplier, update diag(k) and U(i,k) */ 1165 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1166 uikdi = - ba[ili]*ba[bi[i]]; /* diagonal(k) */ 1167 dk += uikdi*ba[ili]; 1168 ba[ili] = uikdi; /* -U(i,k) */ 1169 1170 /* add multiple of row i to k-th row */ 1171 jmin = ili + 1; jmax = bi[i+1]; 1172 if (jmin < jmax){ 1173 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1174 /* update il and jl for row i */ 1175 il[i] = jmin; 1176 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1177 } 1178 i = nexti; 1179 } 1180 1181 /* shift the diagonals when zero pivot is detected */ 1182 /* compute rs=sum of abs(off-diagonal) */ 1183 rs = 0.0; 1184 jmin = bi[k]+1; 1185 nz = bi[k+1] - jmin; 1186 if (nz){ 1187 bcol = bj + jmin; 1188 while (nz--){ 1189 rs += PetscAbsScalar(rtmp[*bcol]); 1190 bcol++; 1191 } 1192 } 1193 1194 sctx.rs = rs; 1195 sctx.pv = dk; 1196 ierr = MatCholeskyCheckShift_inline(info,sctx,newshift);CHKERRQ(ierr); 1197 if (newshift == 1){ 1198 break; /* sctx.shift_amount is updated */ 1199 } else if (newshift == -1){ 1200 SETERRQ4(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g * rs %g",k,PetscAbsScalar(dk),zeropivot,rs); 1201 } 1202 1203 /* copy data into U(k,:) */ 1204 ba[bi[k]] = 1.0/dk; /* U(k,k) */ 1205 jmin = bi[k]+1; jmax = bi[k+1]; 1206 if (jmin < jmax) { 1207 for (j=jmin; j<jmax; j++){ 1208 col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0; 1209 } 1210 /* add the k-th row into il and jl */ 1211 il[k] = jmin; 1212 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1213 } 1214 } 1215 } while (sctx.chshift); 1216 ierr = PetscFree(il);CHKERRQ(ierr); 1217 if (a->permute){ierr = PetscFree(aa);CHKERRQ(ierr);} 1218 1219 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 1220 C->factor = FACTOR_CHOLESKY; 1221 C->assembled = PETSC_TRUE; 1222 C->preallocated = PETSC_TRUE; 1223 ierr = PetscLogFlops(C->m);CHKERRQ(ierr); 1224 if (sctx.nshift){ 1225 if (shiftnz) { 1226 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1: number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1227 } else if (shiftpd) { 1228 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1: number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1229 } 1230 } 1231 PetscFunctionReturn(0); 1232 } 1233 1234 /* 1235 Version for when blocks are 1 by 1 Using natural ordering 1236 */ 1237 #undef __FUNCT__ 1238 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering" 1239 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B) 1240 { 1241 Mat C = *B; 1242 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1243 PetscErrorCode ierr; 1244 PetscInt i,j,mbs = a->mbs; 1245 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 1246 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz; 1247 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 1248 PetscReal zeropivot,rs,shiftnz; 1249 PetscTruth shiftpd; 1250 ChShift_Ctx sctx; 1251 PetscInt newshift; 1252 1253 PetscFunctionBegin; 1254 /* initialization */ 1255 shiftnz = info->shiftnz; 1256 shiftpd = info->shiftpd; 1257 zeropivot = info->zeropivot; 1258 1259 /* il and jl record the first nonzero element in each row of the accessing 1260 window U(0:k, k:mbs-1). 1261 jl: list of rows to be added to uneliminated rows 1262 i>= k: jl(i) is the first row to be added to row i 1263 i< k: jl(i) is the row following row i in some list of rows 1264 jl(i) = mbs indicates the end of a list 1265 il(i): points to the first nonzero element in U(i,k:mbs-1) 1266 */ 1267 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1268 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 1269 jl = il + mbs; 1270 1271 sctx.shift_amount = 0; 1272 sctx.nshift = 0; 1273 do { 1274 sctx.chshift = PETSC_FALSE; 1275 for (i=0; i<mbs; i++) { 1276 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1277 } 1278 1279 for (k = 0; k<mbs; k++){ 1280 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1281 nz = ai[k+1] - ai[k]; 1282 acol = aj + ai[k]; 1283 aval = aa + ai[k]; 1284 bval = ba + bi[k]; 1285 while (nz -- ){ 1286 rtmp[*acol++] = *aval++; 1287 *bval++ = 0.0; /* for in-place factorization */ 1288 } 1289 1290 /* shift the diagonal of the matrix */ 1291 if (sctx.nshift) rtmp[k] += sctx.shift_amount; 1292 1293 /* modify k-th row by adding in those rows i with U(i,k)!=0 */ 1294 dk = rtmp[k]; 1295 i = jl[k]; /* first row to be added to k_th row */ 1296 1297 while (i < k){ 1298 nexti = jl[i]; /* next row to be added to k_th row */ 1299 /* compute multiplier, update D(k) and U(i,k) */ 1300 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1301 uikdi = - ba[ili]*ba[bi[i]]; 1302 dk += uikdi*ba[ili]; 1303 ba[ili] = uikdi; /* -U(i,k) */ 1304 1305 /* add multiple of row i to k-th row ... */ 1306 jmin = ili + 1; 1307 nz = bi[i+1] - jmin; 1308 if (nz > 0){ 1309 bcol = bj + jmin; 1310 bval = ba + jmin; 1311 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 1312 /* update il and jl for i-th row */ 1313 il[i] = jmin; 1314 j = bj[jmin]; jl[i] = jl[j]; jl[j] = i; 1315 } 1316 i = nexti; 1317 } 1318 1319 /* shift the diagonals when zero pivot is detected */ 1320 /* compute rs=sum of abs(off-diagonal) */ 1321 rs = 0.0; 1322 jmin = bi[k]+1; 1323 nz = bi[k+1] - jmin; 1324 if (nz){ 1325 bcol = bj + jmin; 1326 while (nz--){ 1327 rs += PetscAbsScalar(rtmp[*bcol]); 1328 bcol++; 1329 } 1330 } 1331 1332 sctx.rs = rs; 1333 sctx.pv = dk; 1334 ierr = MatCholeskyCheckShift_inline(info,sctx,newshift);CHKERRQ(ierr); 1335 if (newshift == 1){ 1336 break; /* sctx.shift_amount is updated */ 1337 } else if (newshift == -1){ 1338 SETERRQ4(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g * rs %g",k,PetscAbsScalar(dk),zeropivot,rs); 1339 } 1340 1341 /* copy data into U(k,:) */ 1342 ba[bi[k]] = 1.0/dk; 1343 jmin = bi[k]+1; 1344 nz = bi[k+1] - jmin; 1345 if (nz){ 1346 bcol = bj + jmin; 1347 bval = ba + jmin; 1348 while (nz--){ 1349 *bval++ = rtmp[*bcol]; 1350 rtmp[*bcol++] = 0.0; 1351 } 1352 /* add k-th row into il and jl */ 1353 il[k] = jmin; 1354 i = bj[jmin]; jl[k] = jl[i]; jl[i] = k; 1355 } 1356 } /* end of for (k = 0; k<mbs; k++) */ 1357 } while (sctx.chshift); 1358 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1359 ierr = PetscFree(il);CHKERRQ(ierr); 1360 1361 C->factor = FACTOR_CHOLESKY; 1362 C->assembled = PETSC_TRUE; 1363 C->preallocated = PETSC_TRUE; 1364 ierr = PetscLogFlops(C->m);CHKERRQ(ierr); 1365 if (sctx.nshift){ 1366 if (shiftnz) { 1367 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering: number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1368 } else if (shiftpd) { 1369 ierr = PetscLogInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering: number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr); 1370 } 1371 } 1372 PetscFunctionReturn(0); 1373 } 1374 1375 #undef __FUNCT__ 1376 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ" 1377 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info) 1378 { 1379 PetscErrorCode ierr; 1380 Mat C; 1381 1382 PetscFunctionBegin; 1383 ierr = MatCholeskyFactorSymbolic(A,perm,info,&C);CHKERRQ(ierr); 1384 ierr = MatCholeskyFactorNumeric(A,info,&C);CHKERRQ(ierr); 1385 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 1386 PetscFunctionReturn(0); 1387 } 1388 1389 1390