1 2 #include "src/mat/impls/baij/seq/baij.h" 3 #include "src/mat/impls/sbaij/seq/sbaij.h" 4 #include "src/inline/ilu.h" 5 #include "include/petscis.h" 6 7 #if !defined(PETSC_USE_COMPLEX) 8 /* 9 input: 10 F -- numeric factor 11 output: 12 nneg, nzero, npos: matrix inertia 13 */ 14 15 #undef __FUNCT__ 16 #define __FUNCT__ "MatGetInertia_SeqSBAIJ" 17 PetscErrorCode MatGetInertia_SeqSBAIJ(Mat F,PetscInt *nneig,PetscInt *nzero,PetscInt *npos) 18 { 19 Mat_SeqSBAIJ *fact_ptr = (Mat_SeqSBAIJ*)F->data; 20 PetscScalar *dd = fact_ptr->a; 21 PetscInt mbs=fact_ptr->mbs,bs=F->bs,i,nneig_tmp,npos_tmp,*fi = fact_ptr->i; 22 23 PetscFunctionBegin; 24 if (bs != 1) SETERRQ1(PETSC_ERR_SUP,"No support for bs: %D >1 yet",bs); 25 nneig_tmp = 0; npos_tmp = 0; 26 for (i=0; i<mbs; i++){ 27 if (PetscRealPart(dd[*fi]) > 0.0){ 28 npos_tmp++; 29 } else if (PetscRealPart(dd[*fi]) < 0.0){ 30 nneig_tmp++; 31 } 32 fi++; 33 } 34 if (nneig) *nneig = nneig_tmp; 35 if (npos) *npos = npos_tmp; 36 if (nzero) *nzero = mbs - nneig_tmp - npos_tmp; 37 38 PetscFunctionReturn(0); 39 } 40 #endif /* !defined(PETSC_USE_COMPLEX) */ 41 42 /* Using Modified Sparse Row (MSR) storage. 43 See page 85, "Iterative Methods ..." by Saad. */ 44 /* 45 Symbolic U^T*D*U factorization for SBAIJ format. Modified from SSF of YSMP. 46 */ 47 /* Use Modified Sparse Row storage for u and ju, see Saad pp.85 */ 48 #undef __FUNCT__ 49 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ" 50 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info,Mat *B) 51 { 52 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b; 53 PetscErrorCode ierr; 54 PetscInt *rip,i,mbs = a->mbs,*ai,*aj; 55 PetscInt *jutmp,bs = A->bs,bs2=a->bs2; 56 PetscInt m,reallocs = 0,prow; 57 PetscInt *jl,*q,jmin,jmax,juidx,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd; 58 PetscInt *il,ili,nextprow; 59 PetscReal f = info->fill; 60 PetscTruth perm_identity; 61 62 PetscFunctionBegin; 63 /* check whether perm is the identity mapping */ 64 ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr); 65 66 /* -- inplace factorization, i.e., use sbaij for *B -- */ 67 if (perm_identity && bs==1 ){ 68 if (!perm_identity) a->permute = PETSC_TRUE; 69 70 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 71 72 if (perm_identity){ /* without permutation */ 73 ai = a->i; aj = a->j; 74 } else { /* non-trivial permutation */ 75 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 76 ai = a->inew; aj = a->jnew; 77 } 78 79 /* initialization */ 80 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&iu);CHKERRQ(ierr); 81 umax = (PetscInt)(f*ai[mbs] + 1); 82 ierr = PetscMalloc(umax*sizeof(PetscInt),&ju);CHKERRQ(ierr); 83 iu[0] = 0; 84 juidx = 0; /* index for ju */ 85 ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt),&jl);CHKERRQ(ierr); /* linked list for getting pivot row */ 86 q = jl + mbs; /* linked list for col index of active row */ 87 il = q + mbs; 88 for (i=0; i<mbs; i++){ 89 jl[i] = mbs; 90 q[i] = 0; 91 il[i] = 0; 92 } 93 94 /* for each row k */ 95 for (k=0; k<mbs; k++){ 96 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 97 q[k] = mbs; 98 /* initialize nonzero structure of k-th row to row rip[k] of A */ 99 jmin = ai[rip[k]] +1; /* exclude diag[k] */ 100 jmax = ai[rip[k]+1]; 101 for (j=jmin; j<jmax; j++){ 102 vj = rip[aj[j]]; /* col. value */ 103 if(vj > k){ 104 qm = k; 105 do { 106 m = qm; qm = q[m]; 107 } while(qm < vj); 108 if (qm == vj) { 109 SETERRQ(PETSC_ERR_PLIB,"Duplicate entry in A\n"); 110 } 111 nzk++; 112 q[m] = vj; 113 q[vj] = qm; 114 } /* if(vj > k) */ 115 } /* for (j=jmin; j<jmax; j++) */ 116 117 /* modify nonzero structure of k-th row by computing fill-in 118 for each row i to be merged in */ 119 prow = k; 120 prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */ 121 122 while (prow < k){ 123 nextprow = jl[prow]; 124 125 /* merge row prow into k-th row */ 126 ili = il[prow]; 127 jmin = ili + 1; /* points to 2nd nzero entry in U(prow,k:mbs-1) */ 128 jmax = iu[prow+1]; 129 qm = k; 130 for (j=jmin; j<jmax; j++){ 131 vj = ju[j]; 132 do { 133 m = qm; qm = q[m]; 134 } while (qm < vj); 135 if (qm != vj){ /* a fill */ 136 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 137 } 138 } /* end of for (j=jmin; j<jmax; j++) */ 139 if (jmin < jmax){ 140 il[prow] = jmin; 141 j = ju[jmin]; 142 jl[prow] = jl[j]; jl[j] = prow; /* update jl */ 143 } 144 prow = nextprow; 145 } 146 147 /* update il and jl */ 148 if (nzk > 0){ 149 i = q[k]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */ 150 jl[k] = jl[i]; jl[i] = k; 151 il[k] = iu[k] + 1; 152 } 153 iu[k+1] = iu[k] + nzk + 1; /* include diag[k] */ 154 155 /* allocate more space to ju if needed */ 156 if (iu[k+1] > umax) { 157 /* estimate how much additional space we will need */ 158 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 159 /* just double the memory each time */ 160 maxadd = umax; 161 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 162 umax += maxadd; 163 164 /* allocate a longer ju */ 165 ierr = PetscMalloc(umax*sizeof(PetscInt),&jutmp);CHKERRQ(ierr); 166 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr); 167 ierr = PetscFree(ju);CHKERRQ(ierr); 168 ju = jutmp; 169 reallocs++; /* count how many times we realloc */ 170 } 171 172 /* save nonzero structure of k-th row in ju */ 173 ju[juidx++] = k; /* diag[k] */ 174 i = k; 175 while (nzk --) { 176 i = q[i]; 177 ju[juidx++] = i; 178 } 179 } 180 181 if (ai[mbs] != 0) { 182 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 183 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,f,af); 184 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af); 185 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g);\n",af); 186 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"); 187 } else { 188 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"); 189 } 190 191 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 192 /* ierr = PetscFree(q);CHKERRQ(ierr); */ 193 ierr = PetscFree(jl);CHKERRQ(ierr); 194 195 /* put together the new matrix */ 196 ierr = MatCreate(A->comm,bs*mbs,bs*mbs,bs*mbs,bs*mbs,B);CHKERRQ(ierr); 197 ierr = MatSetType(*B,A->type_name);CHKERRQ(ierr); 198 ierr = MatSeqSBAIJSetPreallocation(*B,bs,0,PETSC_NULL);CHKERRQ(ierr); 199 200 /* PetscLogObjectParent(*B,iperm); */ 201 b = (Mat_SeqSBAIJ*)(*B)->data; 202 ierr = PetscFree(b->imax);CHKERRQ(ierr); 203 b->singlemalloc = PETSC_FALSE; 204 /* the next line frees the default space generated by the Create() */ 205 ierr = PetscFree(b->a);CHKERRQ(ierr); 206 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 207 ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr); 208 b->j = ju; 209 b->i = iu; 210 b->diag = 0; 211 b->ilen = 0; 212 b->imax = 0; 213 b->row = perm; 214 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 215 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 216 b->icol = perm; 217 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 218 ierr = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 219 /* In b structure: Free imax, ilen, old a, old j. 220 Allocate idnew, solve_work, new a, new j */ 221 PetscLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar))); 222 b->maxnz = b->nz = iu[mbs]; 223 224 (*B)->factor = FACTOR_CHOLESKY; 225 (*B)->info.factor_mallocs = reallocs; 226 (*B)->info.fill_ratio_given = f; 227 if (ai[mbs] != 0) { 228 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 229 } else { 230 (*B)->info.fill_ratio_needed = 0.0; 231 } 232 233 234 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 235 (*B)->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 236 (*B)->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering; 237 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"); 238 239 PetscFunctionReturn(0); 240 } 241 /* ----------- end of new code --------------------*/ 242 243 244 if (!perm_identity) a->permute = PETSC_TRUE; 245 246 ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr); 247 248 if (perm_identity){ /* without permutation */ 249 ai = a->i; aj = a->j; 250 } else { /* non-trivial permutation */ 251 ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr); 252 ai = a->inew; aj = a->jnew; 253 } 254 255 /* initialization */ 256 ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&iu);CHKERRQ(ierr); 257 umax = (PetscInt)(f*ai[mbs] + 1); umax += mbs + 1; 258 ierr = PetscMalloc(umax*sizeof(PetscInt),&ju);CHKERRQ(ierr); 259 iu[0] = mbs+1; 260 juidx = mbs + 1; /* index for ju */ 261 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&jl);CHKERRQ(ierr); /* linked list for pivot row */ 262 q = jl + mbs; /* linked list for col index */ 263 for (i=0; i<mbs; i++){ 264 jl[i] = mbs; 265 q[i] = 0; 266 } 267 268 /* for each row k */ 269 for (k=0; k<mbs; k++){ 270 for (i=0; i<mbs; i++) q[i] = 0; /* to be removed! */ 271 nzk = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */ 272 q[k] = mbs; 273 /* initialize nonzero structure of k-th row to row rip[k] of A */ 274 jmin = ai[rip[k]] +1; /* exclude diag[k] */ 275 jmax = ai[rip[k]+1]; 276 for (j=jmin; j<jmax; j++){ 277 vj = rip[aj[j]]; /* col. value */ 278 if(vj > k){ 279 qm = k; 280 do { 281 m = qm; qm = q[m]; 282 } while(qm < vj); 283 if (qm == vj) { 284 SETERRQ(PETSC_ERR_PLIB,"Duplicate entry in A\n"); 285 } 286 nzk++; 287 q[m] = vj; 288 q[vj] = qm; 289 } /* if(vj > k) */ 290 } /* for (j=jmin; j<jmax; j++) */ 291 292 /* modify nonzero structure of k-th row by computing fill-in 293 for each row i to be merged in */ 294 prow = k; 295 prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */ 296 297 while (prow < k){ 298 /* merge row prow into k-th row */ 299 jmin = iu[prow] + 1; jmax = iu[prow+1]; 300 qm = k; 301 for (j=jmin; j<jmax; j++){ 302 vj = ju[j]; 303 do { 304 m = qm; qm = q[m]; 305 } while (qm < vj); 306 if (qm != vj){ 307 nzk++; q[m] = vj; q[vj] = qm; qm = vj; 308 } 309 } 310 prow = jl[prow]; /* next pivot row */ 311 } 312 313 /* add k to row list for first nonzero element in k-th row */ 314 if (nzk > 0){ 315 i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */ 316 jl[k] = jl[i]; jl[i] = k; 317 } 318 iu[k+1] = iu[k] + nzk; 319 320 /* allocate more space to ju if needed */ 321 if (iu[k+1] > umax) { 322 /* estimate how much additional space we will need */ 323 /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */ 324 /* just double the memory each time */ 325 maxadd = umax; 326 if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2; 327 umax += maxadd; 328 329 /* allocate a longer ju */ 330 ierr = PetscMalloc(umax*sizeof(PetscInt),&jutmp);CHKERRQ(ierr); 331 ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr); 332 ierr = PetscFree(ju);CHKERRQ(ierr); 333 ju = jutmp; 334 reallocs++; /* count how many times we realloc */ 335 } 336 337 /* save nonzero structure of k-th row in ju */ 338 i=k; 339 while (nzk --) { 340 i = q[i]; 341 ju[juidx++] = i; 342 } 343 } 344 345 if (ai[mbs] != 0) { 346 PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 347 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,f,af); 348 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af); 349 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g);\n",af); 350 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"); 351 } else { 352 PetscLogInfo(A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"); 353 } 354 355 ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr); 356 /* ierr = PetscFree(q);CHKERRQ(ierr); */ 357 ierr = PetscFree(jl);CHKERRQ(ierr); 358 359 /* put together the new matrix */ 360 ierr = MatCreate(A->comm,bs*mbs,bs*mbs,bs*mbs,bs*mbs,B);CHKERRQ(ierr); 361 ierr = MatSetType(*B,A->type_name);CHKERRQ(ierr); 362 ierr = MatSeqSBAIJSetPreallocation(*B,bs,0,PETSC_NULL);CHKERRQ(ierr); 363 364 /* PetscLogObjectParent(*B,iperm); */ 365 b = (Mat_SeqSBAIJ*)(*B)->data; 366 ierr = PetscFree(b->imax);CHKERRQ(ierr); 367 b->singlemalloc = PETSC_FALSE; 368 /* the next line frees the default space generated by the Create() */ 369 ierr = PetscFree(b->a);CHKERRQ(ierr); 370 ierr = PetscFree(b->ilen);CHKERRQ(ierr); 371 ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr); 372 b->j = ju; 373 b->i = iu; 374 b->diag = 0; 375 b->ilen = 0; 376 b->imax = 0; 377 b->row = perm; 378 b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */ 379 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 380 b->icol = perm; 381 ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr); 382 ierr = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr); 383 /* In b structure: Free imax, ilen, old a, old j. 384 Allocate idnew, solve_work, new a, new j */ 385 PetscLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar))); 386 b->maxnz = b->nz = iu[mbs]; 387 388 (*B)->factor = FACTOR_CHOLESKY; 389 (*B)->info.factor_mallocs = reallocs; 390 (*B)->info.fill_ratio_given = f; 391 if (ai[mbs] != 0) { 392 (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]); 393 } else { 394 (*B)->info.fill_ratio_needed = 0.0; 395 } 396 397 if (perm_identity){ 398 switch (bs) { 399 case 1: 400 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering; 401 (*B)->ops->solve = MatSolve_SeqSBAIJ_1_NaturalOrdering; 402 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"); 403 break; 404 case 2: 405 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering; 406 (*B)->ops->solve = MatSolve_SeqSBAIJ_2_NaturalOrdering; 407 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"); 408 break; 409 case 3: 410 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering; 411 (*B)->ops->solve = MatSolve_SeqSBAIJ_3_NaturalOrdering; 412 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"); 413 break; 414 case 4: 415 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering; 416 (*B)->ops->solve = MatSolve_SeqSBAIJ_4_NaturalOrdering; 417 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"); 418 break; 419 case 5: 420 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering; 421 (*B)->ops->solve = MatSolve_SeqSBAIJ_5_NaturalOrdering; 422 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"); 423 break; 424 case 6: 425 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering; 426 (*B)->ops->solve = MatSolve_SeqSBAIJ_6_NaturalOrdering; 427 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"); 428 break; 429 case 7: 430 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering; 431 (*B)->ops->solve = MatSolve_SeqSBAIJ_7_NaturalOrdering; 432 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"); 433 break; 434 default: 435 (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering; 436 (*B)->ops->solve = MatSolve_SeqSBAIJ_N_NaturalOrdering; 437 PetscLogInfo(A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"); 438 break; 439 } 440 } 441 442 PetscFunctionReturn(0); 443 } 444 445 446 #undef __FUNCT__ 447 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N" 448 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,Mat *B) 449 { 450 Mat C = *B; 451 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 452 IS perm = b->row; 453 PetscErrorCode ierr; 454 PetscInt *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 455 PetscInt *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 456 PetscInt bs=A->bs,bs2 = a->bs2; 457 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 458 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 459 MatScalar *work; 460 PetscInt *pivots; 461 462 PetscFunctionBegin; 463 /* initialization */ 464 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 465 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 466 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 467 jl = il + mbs; 468 for (i=0; i<mbs; i++) { 469 jl[i] = mbs; il[0] = 0; 470 } 471 ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr); 472 uik = dk + bs2; 473 work = uik + bs2; 474 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 475 476 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 477 478 /* check permutation */ 479 if (!a->permute){ 480 ai = a->i; aj = a->j; aa = a->a; 481 } else { 482 ai = a->inew; aj = a->jnew; 483 ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 484 ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 485 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 486 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 487 488 for (i=0; i<mbs; i++){ 489 jmin = ai[i]; jmax = ai[i+1]; 490 for (j=jmin; j<jmax; j++){ 491 while (a2anew[j] != j){ 492 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 493 for (k1=0; k1<bs2; k1++){ 494 dk[k1] = aa[k*bs2+k1]; 495 aa[k*bs2+k1] = aa[j*bs2+k1]; 496 aa[j*bs2+k1] = dk[k1]; 497 } 498 } 499 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 500 if (i > aj[j]){ 501 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 502 ap = aa + j*bs2; /* ptr to the beginning of j-th block of aa */ 503 for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */ 504 for (k=0; k<bs; k++){ /* j-th block of aa <- dk^T */ 505 for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1]; 506 } 507 } 508 } 509 } 510 ierr = PetscFree(a2anew);CHKERRQ(ierr); 511 } 512 513 /* for each row k */ 514 for (k = 0; k<mbs; k++){ 515 516 /*initialize k-th row with elements nonzero in row perm(k) of A */ 517 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 518 519 ap = aa + jmin*bs2; 520 for (j = jmin; j < jmax; j++){ 521 vj = perm_ptr[aj[j]]; /* block col. index */ 522 rtmp_ptr = rtmp + vj*bs2; 523 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 524 } 525 526 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 527 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 528 i = jl[k]; /* first row to be added to k_th row */ 529 530 while (i < k){ 531 nexti = jl[i]; /* next row to be added to k_th row */ 532 533 /* compute multiplier */ 534 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 535 536 /* uik = -inv(Di)*U_bar(i,k) */ 537 diag = ba + i*bs2; 538 u = ba + ili*bs2; 539 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 540 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 541 542 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 543 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 544 545 /* update -U(i,k) */ 546 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 547 548 /* add multiple of row i to k-th row ... */ 549 jmin = ili + 1; jmax = bi[i+1]; 550 if (jmin < jmax){ 551 for (j=jmin; j<jmax; j++) { 552 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 553 rtmp_ptr = rtmp + bj[j]*bs2; 554 u = ba + j*bs2; 555 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 556 } 557 558 /* ... add i to row list for next nonzero entry */ 559 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 560 j = bj[jmin]; 561 jl[i] = jl[j]; jl[j] = i; /* update jl */ 562 } 563 i = nexti; 564 } 565 566 /* save nonzero entries in k-th row of U ... */ 567 568 /* invert diagonal block */ 569 diag = ba+k*bs2; 570 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 571 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 572 573 jmin = bi[k]; jmax = bi[k+1]; 574 if (jmin < jmax) { 575 for (j=jmin; j<jmax; j++){ 576 vj = bj[j]; /* block col. index of U */ 577 u = ba + j*bs2; 578 rtmp_ptr = rtmp + vj*bs2; 579 for (k1=0; k1<bs2; k1++){ 580 *u++ = *rtmp_ptr; 581 *rtmp_ptr++ = 0.0; 582 } 583 } 584 585 /* ... add k to row list for first nonzero entry in k-th row */ 586 il[k] = jmin; 587 i = bj[jmin]; 588 jl[k] = jl[i]; jl[i] = k; 589 } 590 } 591 592 ierr = PetscFree(rtmp);CHKERRQ(ierr); 593 ierr = PetscFree(il);CHKERRQ(ierr); 594 ierr = PetscFree(dk);CHKERRQ(ierr); 595 ierr = PetscFree(pivots);CHKERRQ(ierr); 596 if (a->permute){ 597 ierr = PetscFree(aa);CHKERRQ(ierr); 598 } 599 600 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 601 C->factor = FACTOR_CHOLESKY; 602 C->assembled = PETSC_TRUE; 603 C->preallocated = PETSC_TRUE; 604 PetscLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 605 PetscFunctionReturn(0); 606 } 607 608 #undef __FUNCT__ 609 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering" 610 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat A,Mat *B) 611 { 612 Mat C = *B; 613 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 614 PetscErrorCode ierr; 615 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 616 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 617 PetscInt bs=A->bs,bs2 = a->bs2; 618 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 619 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 620 MatScalar *work; 621 PetscInt *pivots; 622 623 PetscFunctionBegin; 624 /* initialization */ 625 626 ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 627 ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr); 628 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 629 jl = il + mbs; 630 for (i=0; i<mbs; i++) { 631 jl[i] = mbs; il[0] = 0; 632 } 633 ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr); 634 uik = dk + bs2; 635 work = uik + bs2; 636 ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr); 637 638 ai = a->i; aj = a->j; aa = a->a; 639 640 /* for each row k */ 641 for (k = 0; k<mbs; k++){ 642 643 /*initialize k-th row with elements nonzero in row k of A */ 644 jmin = ai[k]; jmax = ai[k+1]; 645 ap = aa + jmin*bs2; 646 for (j = jmin; j < jmax; j++){ 647 vj = aj[j]; /* block col. index */ 648 rtmp_ptr = rtmp + vj*bs2; 649 for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++; 650 } 651 652 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 653 ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr); 654 i = jl[k]; /* first row to be added to k_th row */ 655 656 while (i < k){ 657 nexti = jl[i]; /* next row to be added to k_th row */ 658 659 /* compute multiplier */ 660 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 661 662 /* uik = -inv(Di)*U_bar(i,k) */ 663 diag = ba + i*bs2; 664 u = ba + ili*bs2; 665 ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 666 Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u); 667 668 /* update D(k) += -U(i,k)^T * U_bar(i,k) */ 669 Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u); 670 671 /* update -U(i,k) */ 672 ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr); 673 674 /* add multiple of row i to k-th row ... */ 675 jmin = ili + 1; jmax = bi[i+1]; 676 if (jmin < jmax){ 677 for (j=jmin; j<jmax; j++) { 678 /* rtmp += -U(i,k)^T * U_bar(i,j) */ 679 rtmp_ptr = rtmp + bj[j]*bs2; 680 u = ba + j*bs2; 681 Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u); 682 } 683 684 /* ... add i to row list for next nonzero entry */ 685 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 686 j = bj[jmin]; 687 jl[i] = jl[j]; jl[j] = i; /* update jl */ 688 } 689 i = nexti; 690 } 691 692 /* save nonzero entries in k-th row of U ... */ 693 694 /* invert diagonal block */ 695 diag = ba+k*bs2; 696 ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr); 697 ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr); 698 699 jmin = bi[k]; jmax = bi[k+1]; 700 if (jmin < jmax) { 701 for (j=jmin; j<jmax; j++){ 702 vj = bj[j]; /* block col. index of U */ 703 u = ba + j*bs2; 704 rtmp_ptr = rtmp + vj*bs2; 705 for (k1=0; k1<bs2; k1++){ 706 *u++ = *rtmp_ptr; 707 *rtmp_ptr++ = 0.0; 708 } 709 } 710 711 /* ... add k to row list for first nonzero entry in k-th row */ 712 il[k] = jmin; 713 i = bj[jmin]; 714 jl[k] = jl[i]; jl[i] = k; 715 } 716 } 717 718 ierr = PetscFree(rtmp);CHKERRQ(ierr); 719 ierr = PetscFree(il);CHKERRQ(ierr); 720 ierr = PetscFree(dk);CHKERRQ(ierr); 721 ierr = PetscFree(pivots);CHKERRQ(ierr); 722 723 C->factor = FACTOR_CHOLESKY; 724 C->assembled = PETSC_TRUE; 725 C->preallocated = PETSC_TRUE; 726 PetscLogFlops(1.3333*bs*bs2*b->mbs); /* from inverting diagonal blocks */ 727 PetscFunctionReturn(0); 728 } 729 730 /* 731 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 732 Version for blocks 2 by 2. 733 */ 734 #undef __FUNCT__ 735 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2" 736 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,Mat *B) 737 { 738 Mat C = *B; 739 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 740 IS perm = b->row; 741 PetscErrorCode ierr; 742 PetscInt *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 743 PetscInt *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 744 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 745 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 746 747 PetscFunctionBegin; 748 /* initialization */ 749 /* il and jl record the first nonzero element in each row of the accessing 750 window U(0:k, k:mbs-1). 751 jl: list of rows to be added to uneliminated rows 752 i>= k: jl(i) is the first row to be added to row i 753 i< k: jl(i) is the row following row i in some list of rows 754 jl(i) = mbs indicates the end of a list 755 il(i): points to the first nonzero element in columns k,...,mbs-1 of 756 row i of U */ 757 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 758 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 759 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 760 jl = il + mbs; 761 for (i=0; i<mbs; i++) { 762 jl[i] = mbs; il[0] = 0; 763 } 764 ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr); 765 uik = dk + 4; 766 ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr); 767 768 /* check permutation */ 769 if (!a->permute){ 770 ai = a->i; aj = a->j; aa = a->a; 771 } else { 772 ai = a->inew; aj = a->jnew; 773 ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 774 ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 775 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr); 776 ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 777 778 for (i=0; i<mbs; i++){ 779 jmin = ai[i]; jmax = ai[i+1]; 780 for (j=jmin; j<jmax; j++){ 781 while (a2anew[j] != j){ 782 k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k; 783 for (k1=0; k1<4; k1++){ 784 dk[k1] = aa[k*4+k1]; 785 aa[k*4+k1] = aa[j*4+k1]; 786 aa[j*4+k1] = dk[k1]; 787 } 788 } 789 /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */ 790 if (i > aj[j]){ 791 /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */ 792 ap = aa + j*4; /* ptr to the beginning of the block */ 793 dk[1] = ap[1]; /* swap ap[1] and ap[2] */ 794 ap[1] = ap[2]; 795 ap[2] = dk[1]; 796 } 797 } 798 } 799 ierr = PetscFree(a2anew);CHKERRQ(ierr); 800 } 801 802 /* for each row k */ 803 for (k = 0; k<mbs; k++){ 804 805 /*initialize k-th row with elements nonzero in row perm(k) of A */ 806 jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1]; 807 ap = aa + jmin*4; 808 for (j = jmin; j < jmax; j++){ 809 vj = perm_ptr[aj[j]]; /* block col. index */ 810 rtmp_ptr = rtmp + vj*4; 811 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 812 } 813 814 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 815 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 816 i = jl[k]; /* first row to be added to k_th row */ 817 818 while (i < k){ 819 nexti = jl[i]; /* next row to be added to k_th row */ 820 821 /* compute multiplier */ 822 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 823 824 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 825 diag = ba + i*4; 826 u = ba + ili*4; 827 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 828 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 829 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 830 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 831 832 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 833 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 834 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 835 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 836 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 837 838 /* update -U(i,k): ba[ili] = uik */ 839 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 840 841 /* add multiple of row i to k-th row ... */ 842 jmin = ili + 1; jmax = bi[i+1]; 843 if (jmin < jmax){ 844 for (j=jmin; j<jmax; j++) { 845 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 846 rtmp_ptr = rtmp + bj[j]*4; 847 u = ba + j*4; 848 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 849 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 850 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 851 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 852 } 853 854 /* ... add i to row list for next nonzero entry */ 855 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 856 j = bj[jmin]; 857 jl[i] = jl[j]; jl[j] = i; /* update jl */ 858 } 859 i = nexti; 860 } 861 862 /* save nonzero entries in k-th row of U ... */ 863 864 /* invert diagonal block */ 865 diag = ba+k*4; 866 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 867 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 868 869 jmin = bi[k]; jmax = bi[k+1]; 870 if (jmin < jmax) { 871 for (j=jmin; j<jmax; j++){ 872 vj = bj[j]; /* block col. index of U */ 873 u = ba + j*4; 874 rtmp_ptr = rtmp + vj*4; 875 for (k1=0; k1<4; k1++){ 876 *u++ = *rtmp_ptr; 877 *rtmp_ptr++ = 0.0; 878 } 879 } 880 881 /* ... add k to row list for first nonzero entry in k-th row */ 882 il[k] = jmin; 883 i = bj[jmin]; 884 jl[k] = jl[i]; jl[i] = k; 885 } 886 } 887 888 ierr = PetscFree(rtmp);CHKERRQ(ierr); 889 ierr = PetscFree(il);CHKERRQ(ierr); 890 ierr = PetscFree(dk);CHKERRQ(ierr); 891 if (a->permute) { 892 ierr = PetscFree(aa);CHKERRQ(ierr); 893 } 894 ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr); 895 C->factor = FACTOR_CHOLESKY; 896 C->assembled = PETSC_TRUE; 897 C->preallocated = PETSC_TRUE; 898 PetscLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 899 PetscFunctionReturn(0); 900 } 901 902 /* 903 Version for when blocks are 2 by 2 Using natural ordering 904 */ 905 #undef __FUNCT__ 906 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering" 907 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,Mat *B) 908 { 909 Mat C = *B; 910 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 911 PetscErrorCode ierr; 912 PetscInt i,j,mbs=a->mbs,*bi=b->i,*bj=b->j; 913 PetscInt *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili; 914 MatScalar *ba = b->a,*aa,*ap,*dk,*uik; 915 MatScalar *u,*diag,*rtmp,*rtmp_ptr; 916 917 PetscFunctionBegin; 918 /* initialization */ 919 /* il and jl record the first nonzero element in each row of the accessing 920 window U(0:k, k:mbs-1). 921 jl: list of rows to be added to uneliminated rows 922 i>= k: jl(i) is the first row to be added to row i 923 i< k: jl(i) is the row following row i in some list of rows 924 jl(i) = mbs indicates the end of a list 925 il(i): points to the first nonzero element in columns k,...,mbs-1 of 926 row i of U */ 927 ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 928 ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr); 929 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 930 jl = il + mbs; 931 for (i=0; i<mbs; i++) { 932 jl[i] = mbs; il[0] = 0; 933 } 934 ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr); 935 uik = dk + 4; 936 937 ai = a->i; aj = a->j; aa = a->a; 938 939 /* for each row k */ 940 for (k = 0; k<mbs; k++){ 941 942 /*initialize k-th row with elements nonzero in row k of A */ 943 jmin = ai[k]; jmax = ai[k+1]; 944 ap = aa + jmin*4; 945 for (j = jmin; j < jmax; j++){ 946 vj = aj[j]; /* block col. index */ 947 rtmp_ptr = rtmp + vj*4; 948 for (i=0; i<4; i++) *rtmp_ptr++ = *ap++; 949 } 950 951 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 952 ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr); 953 i = jl[k]; /* first row to be added to k_th row */ 954 955 while (i < k){ 956 nexti = jl[i]; /* next row to be added to k_th row */ 957 958 /* compute multiplier */ 959 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 960 961 /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */ 962 diag = ba + i*4; 963 u = ba + ili*4; 964 uik[0] = -(diag[0]*u[0] + diag[2]*u[1]); 965 uik[1] = -(diag[1]*u[0] + diag[3]*u[1]); 966 uik[2] = -(diag[0]*u[2] + diag[2]*u[3]); 967 uik[3] = -(diag[1]*u[2] + diag[3]*u[3]); 968 969 /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */ 970 dk[0] += uik[0]*u[0] + uik[1]*u[1]; 971 dk[1] += uik[2]*u[0] + uik[3]*u[1]; 972 dk[2] += uik[0]*u[2] + uik[1]*u[3]; 973 dk[3] += uik[2]*u[2] + uik[3]*u[3]; 974 975 /* update -U(i,k): ba[ili] = uik */ 976 ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr); 977 978 /* add multiple of row i to k-th row ... */ 979 jmin = ili + 1; jmax = bi[i+1]; 980 if (jmin < jmax){ 981 for (j=jmin; j<jmax; j++) { 982 /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */ 983 rtmp_ptr = rtmp + bj[j]*4; 984 u = ba + j*4; 985 rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1]; 986 rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1]; 987 rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3]; 988 rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3]; 989 } 990 991 /* ... add i to row list for next nonzero entry */ 992 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 993 j = bj[jmin]; 994 jl[i] = jl[j]; jl[j] = i; /* update jl */ 995 } 996 i = nexti; 997 } 998 999 /* save nonzero entries in k-th row of U ... */ 1000 1001 /* invert diagonal block */ 1002 diag = ba+k*4; 1003 ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr); 1004 ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr); 1005 1006 jmin = bi[k]; jmax = bi[k+1]; 1007 if (jmin < jmax) { 1008 for (j=jmin; j<jmax; j++){ 1009 vj = bj[j]; /* block col. index of U */ 1010 u = ba + j*4; 1011 rtmp_ptr = rtmp + vj*4; 1012 for (k1=0; k1<4; k1++){ 1013 *u++ = *rtmp_ptr; 1014 *rtmp_ptr++ = 0.0; 1015 } 1016 } 1017 1018 /* ... add k to row list for first nonzero entry in k-th row */ 1019 il[k] = jmin; 1020 i = bj[jmin]; 1021 jl[k] = jl[i]; jl[i] = k; 1022 } 1023 } 1024 1025 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1026 ierr = PetscFree(il);CHKERRQ(ierr); 1027 ierr = PetscFree(dk);CHKERRQ(ierr); 1028 1029 C->factor = FACTOR_CHOLESKY; 1030 C->assembled = PETSC_TRUE; 1031 C->preallocated = PETSC_TRUE; 1032 PetscLogFlops(1.3333*8*b->mbs); /* from inverting diagonal blocks */ 1033 PetscFunctionReturn(0); 1034 } 1035 1036 /* 1037 Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP. 1038 Version for blocks are 1 by 1. 1039 */ 1040 #undef __FUNCT__ 1041 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1" 1042 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,Mat *B) 1043 { 1044 Mat C = *B; 1045 Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data; 1046 IS ip = b->row; 1047 PetscErrorCode ierr; 1048 PetscInt *rip,i,j,mbs = a->mbs,*bi = b->i,*bj = b->j; 1049 PetscInt *ai,*aj,*r; 1050 PetscInt k,jmin,jmax,*jl,*il,vj,nexti,ili; 1051 MatScalar *rtmp; 1052 MatScalar *ba = b->a,*aa,ak; 1053 MatScalar dk,uikdi; 1054 1055 PetscFunctionBegin; 1056 ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr); 1057 if (!a->permute){ 1058 ai = a->i; aj = a->j; aa = a->a; 1059 } else { 1060 ai = a->inew; aj = a->jnew; 1061 ierr = PetscMalloc(ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr); 1062 ierr = PetscMemcpy(aa,a->a,ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr); 1063 ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&r);CHKERRQ(ierr); 1064 ierr= PetscMemcpy(r,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr); 1065 1066 jmin = ai[0]; jmax = ai[mbs]; 1067 for (j=jmin; j<jmax; j++){ 1068 while (r[j] != j){ 1069 k = r[j]; r[j] = r[k]; r[k] = k; 1070 ak = aa[k]; aa[k] = aa[j]; aa[j] = ak; 1071 } 1072 } 1073 ierr = PetscFree(r);CHKERRQ(ierr); 1074 } 1075 1076 /* initialization */ 1077 /* il and jl record the first nonzero element in each row of the accessing 1078 window U(0:k, k:mbs-1). 1079 jl: list of rows to be added to uneliminated rows 1080 i>= k: jl(i) is the first row to be added to row i 1081 i< k: jl(i) is the row following row i in some list of rows 1082 jl(i) = mbs indicates the end of a list 1083 il(i): points to the first nonzero element in columns k,...,mbs-1 of 1084 row i of U */ 1085 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1086 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 1087 jl = il + mbs; 1088 for (i=0; i<mbs; i++) { 1089 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1090 } 1091 1092 /* for each row k */ 1093 for (k = 0; k<mbs; k++){ 1094 1095 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1096 jmin = ai[rip[k]]; jmax = ai[rip[k]+1]; 1097 1098 for (j = jmin; j < jmax; j++){ 1099 vj = rip[aj[j]]; 1100 rtmp[vj] = aa[j]; 1101 } 1102 1103 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 1104 dk = rtmp[k]; 1105 i = jl[k]; /* first row to be added to k_th row */ 1106 1107 while (i < k){ 1108 nexti = jl[i]; /* next row to be added to k_th row */ 1109 1110 /* compute multiplier, update D(k) and U(i,k) */ 1111 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1112 uikdi = - ba[ili]*ba[i]; 1113 dk += uikdi*ba[ili]; 1114 ba[ili] = uikdi; /* -U(i,k) */ 1115 1116 /* add multiple of row i to k-th row ... */ 1117 jmin = ili + 1; jmax = bi[i+1]; 1118 if (jmin < jmax){ 1119 for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j]; 1120 /* ... add i to row list for next nonzero entry */ 1121 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1122 j = bj[jmin]; 1123 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1124 } 1125 i = nexti; 1126 } 1127 1128 /* check for zero pivot and save diagoanl element */ 1129 if (dk == 0.0){ 1130 SETERRQ(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot"); 1131 /* 1132 } else if (PetscRealPart(dk) < 0.0){ 1133 SETERRQ2(PETSC_ERR_MAT_LU_ZRPVT,"Negative pivot: d[%d] = %g\n",k,dk); 1134 */ 1135 } 1136 1137 /* save nonzero entries in k-th row of U ... */ 1138 ba[k] = 1.0/dk; 1139 jmin = bi[k]; jmax = bi[k+1]; 1140 if (jmin < jmax) { 1141 for (j=jmin; j<jmax; j++){ 1142 vj = bj[j]; ba[j] = rtmp[vj]; rtmp[vj] = 0.0; 1143 } 1144 /* ... add k to row list for first nonzero entry in k-th row */ 1145 il[k] = jmin; 1146 i = bj[jmin]; 1147 jl[k] = jl[i]; jl[i] = k; 1148 } 1149 } 1150 1151 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1152 ierr = PetscFree(il);CHKERRQ(ierr); 1153 if (a->permute){ 1154 ierr = PetscFree(aa);CHKERRQ(ierr); 1155 } 1156 1157 ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr); 1158 C->factor = FACTOR_CHOLESKY; 1159 C->assembled = PETSC_TRUE; 1160 C->preallocated = PETSC_TRUE; 1161 PetscLogFlops(b->mbs); 1162 PetscFunctionReturn(0); 1163 } 1164 1165 /* 1166 Version for when blocks are 1 by 1 Using natural ordering 1167 */ 1168 #undef __FUNCT__ 1169 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering" 1170 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat A,Mat *B) 1171 { 1172 Mat C = *B; 1173 Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data; 1174 PetscErrorCode ierr; 1175 PetscInt i,j,mbs = a->mbs; 1176 PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j; 1177 PetscInt k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz,ndamp = 0; 1178 MatScalar *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval; 1179 PetscReal damping=b->factor_damping, zeropivot=b->factor_zeropivot,shift_amount; 1180 PetscTruth damp,chshift; 1181 PetscInt nshift=0; 1182 1183 PetscFunctionBegin; 1184 /* initialization */ 1185 /* il and jl record the first nonzero element in each row of the accessing 1186 window U(0:k, k:mbs-1). 1187 jl: list of rows to be added to uneliminated rows 1188 i>= k: jl(i) is the first row to be added to row i 1189 i< k: jl(i) is the row following row i in some list of rows 1190 jl(i) = mbs indicates the end of a list 1191 il(i): points to the first nonzero element in U(i,k:mbs-1) 1192 */ 1193 ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr); 1194 ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr); 1195 jl = il + mbs; 1196 1197 shift_amount = 0; 1198 do { 1199 damp = PETSC_FALSE; 1200 chshift = PETSC_FALSE; 1201 for (i=0; i<mbs; i++) { 1202 rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0; 1203 } 1204 1205 for (k = 0; k<mbs; k++){ /* row k */ 1206 /*initialize k-th row with elements nonzero in row perm(k) of A */ 1207 nz = ai[k+1] - ai[k]; 1208 acol = aj + ai[k]; 1209 aval = aa + ai[k]; 1210 bval = ba + bi[k]; 1211 while (nz -- ){ 1212 rtmp[*acol++] = *aval++; 1213 *bval++ = 0.0; /* for in-place factorization */ 1214 } 1215 /* damp the diagonal of the matrix */ 1216 if (ndamp||nshift) rtmp[k] += damping+shift_amount; 1217 1218 /* modify k-th row by adding in those rows i with U(i,k) != 0 */ 1219 dk = rtmp[k]; 1220 i = jl[k]; /* first row to be added to k_th row */ 1221 1222 while (i < k){ 1223 nexti = jl[i]; /* next row to be added to k_th row */ 1224 1225 /* compute multiplier, update D(k) and U(i,k) */ 1226 ili = il[i]; /* index of first nonzero element in U(i,k:bms-1) */ 1227 uikdi = - ba[ili]*ba[bi[i]]; 1228 dk += uikdi*ba[ili]; 1229 ba[ili] = uikdi; /* -U(i,k) */ 1230 1231 /* add multiple of row i to k-th row ... */ 1232 jmin = ili + 1; 1233 nz = bi[i+1] - jmin; 1234 if (nz > 0){ 1235 bcol = bj + jmin; 1236 bval = ba + jmin; 1237 while (nz --) rtmp[*bcol++] += uikdi*(*bval++); 1238 /* ... add i to row list for next nonzero entry */ 1239 il[i] = jmin; /* update il(i) in column k+1, ... mbs-1 */ 1240 j = bj[jmin]; 1241 jl[i] = jl[j]; jl[j] = i; /* update jl */ 1242 } 1243 i = nexti; 1244 } 1245 1246 if (PetscRealPart(dk) < zeropivot && b->factor_shift){ 1247 /* calculate a shift that would make this row diagonally dominant */ 1248 PetscReal rs = PetscAbs(PetscRealPart(dk)); 1249 jmin = bi[k]+1; 1250 nz = bi[k+1] - jmin; 1251 if (nz){ 1252 bcol = bj + jmin; 1253 bval = ba + jmin; 1254 while (nz--){ 1255 rs += PetscAbsScalar(rtmp[*bcol++]); 1256 } 1257 } 1258 /* if this shift is less than the previous, just up the previous 1259 one by a bit */ 1260 shift_amount = PetscMax(rs,1.1*shift_amount); 1261 chshift = PETSC_TRUE; 1262 /* Unlike in the ILU case there is no exit condition on nshift: 1263 we increase the shift until it converges. There is no guarantee that 1264 this algorithm converges faster or slower, or is better or worse 1265 than the ILU algorithm. */ 1266 nshift++; 1267 break; 1268 } 1269 if (PetscRealPart(dk) < zeropivot){ 1270 if (damping == (PetscReal) PETSC_DECIDE) damping = -PetscRealPart(dk)/(k+1); 1271 if (damping > 0.0) { 1272 if (ndamp) damping *= 2.0; 1273 damp = PETSC_TRUE; 1274 ndamp++; 1275 break; 1276 } else if (PetscAbsScalar(dk) < zeropivot){ 1277 SETERRQ3(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g",k,PetscRealPart(dk),zeropivot); 1278 } else { 1279 PetscLogInfo((PetscObject)A,"Negative pivot %g in row %D of Cholesky factorization\n",PetscRealPart(dk),k); 1280 } 1281 } 1282 1283 /* save nonzero entries in k-th row of U ... */ 1284 /* printf("%D, dk: %g, 1/dk: %g\n",k,dk,1/dk); */ 1285 ba[bi[k]] = 1.0/dk; 1286 jmin = bi[k]+1; 1287 nz = bi[k+1] - jmin; 1288 if (nz){ 1289 bcol = bj + jmin; 1290 bval = ba + jmin; 1291 while (nz--){ 1292 *bval++ = rtmp[*bcol]; 1293 rtmp[*bcol++] = 0.0; 1294 } 1295 /* ... add k to row list for first nonzero entry in k-th row */ 1296 il[k] = jmin; 1297 i = bj[jmin]; 1298 jl[k] = jl[i]; jl[i] = k; 1299 } 1300 } /* end of for (k = 0; k<mbs; k++) */ 1301 } while (damp||chshift); 1302 ierr = PetscFree(rtmp);CHKERRQ(ierr); 1303 ierr = PetscFree(il);CHKERRQ(ierr); 1304 1305 C->factor = FACTOR_CHOLESKY; 1306 C->assembled = PETSC_TRUE; 1307 C->preallocated = PETSC_TRUE; 1308 PetscLogFlops(b->mbs); 1309 if (ndamp) { 1310 PetscLogInfo(0,"MatCholeskyFactorNumerical_SeqSBAIJ_1_NaturalOrdering: number of damping tries %D damping value %g\n",ndamp,damping); 1311 } 1312 if (nshift) { 1313 PetscLogInfo(0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering diagonal shifted %D shifts\n",nshift); 1314 } 1315 1316 PetscFunctionReturn(0); 1317 } 1318 1319 #undef __FUNCT__ 1320 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ" 1321 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info) 1322 { 1323 PetscErrorCode ierr; 1324 Mat C; 1325 1326 PetscFunctionBegin; 1327 ierr = MatCholeskyFactorSymbolic(A,perm,info,&C);CHKERRQ(ierr); 1328 ierr = MatCholeskyFactorNumeric(A,&C);CHKERRQ(ierr); 1329 ierr = MatHeaderCopy(A,C);CHKERRQ(ierr); 1330 PetscFunctionReturn(0); 1331 } 1332 1333 1334