xref: /petsc/src/mat/impls/sbaij/seq/sbaijfact.c (revision 4f02bc6a7df4e4b03c783003a8e0899ee35fcb05)
1 
2 #include <../src/mat/impls/baij/seq/baij.h>
3 #include <../src/mat/impls/sbaij/seq/sbaij.h>
4 #include <petsc/private/kernels/blockinvert.h>
5 #include <petscis.h>
6 
7 /*
8   input:
9    F -- numeric factor
10   output:
11    nneg, nzero, npos: matrix inertia
12 */
13 
14 #undef __FUNCT__
15 #define __FUNCT__ "MatGetInertia_SeqSBAIJ"
16 PetscErrorCode MatGetInertia_SeqSBAIJ(Mat F,PetscInt *nneig,PetscInt *nzero,PetscInt *npos)
17 {
18   Mat_SeqSBAIJ *fact_ptr = (Mat_SeqSBAIJ*)F->data;
19   MatScalar    *dd       = fact_ptr->a;
20   PetscInt     mbs       =fact_ptr->mbs,bs=F->rmap->bs,i,nneig_tmp,npos_tmp,*fi = fact_ptr->diag;
21 
22   PetscFunctionBegin;
23   if (bs != 1) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for bs: %D >1 yet",bs);
24   nneig_tmp = 0; npos_tmp = 0;
25   for (i=0; i<mbs; i++) {
26     if (PetscRealPart(dd[*fi]) > 0.0) npos_tmp++;
27     else if (PetscRealPart(dd[*fi]) < 0.0) nneig_tmp++;
28     fi++;
29   }
30   if (nneig) *nneig = nneig_tmp;
31   if (npos)  *npos  = npos_tmp;
32   if (nzero) *nzero = mbs - nneig_tmp - npos_tmp;
33   PetscFunctionReturn(0);
34 }
35 
36 /*
37   Symbolic U^T*D*U factorization for SBAIJ format. Modified from SSF of YSMP.
38   Use Modified Sparse Row (MSR) storage for u and ju. See page 85, "Iterative Methods ..." by Saad.
39 */
40 #undef __FUNCT__
41 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_MSR"
42 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(Mat F,Mat A,IS perm,const MatFactorInfo *info)
43 {
44   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b;
45   PetscErrorCode ierr;
46   const PetscInt *rip,*ai,*aj;
47   PetscInt       i,mbs = a->mbs,*jutmp,bs = A->rmap->bs,bs2=a->bs2;
48   PetscInt       m,reallocs = 0,prow;
49   PetscInt       *jl,*q,jmin,jmax,juidx,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd;
50   PetscReal      f = info->fill;
51   PetscBool      perm_identity;
52 
53   PetscFunctionBegin;
54   /* check whether perm is the identity mapping */
55   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
56   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
57 
58   if (perm_identity) { /* without permutation */
59     a->permute = PETSC_FALSE;
60 
61     ai = a->i; aj = a->j;
62   } else {            /* non-trivial permutation */
63     a->permute = PETSC_TRUE;
64 
65     ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr);
66 
67     ai = a->inew; aj = a->jnew;
68   }
69 
70   /* initialization */
71   ierr  = PetscMalloc1(mbs+1,&iu);CHKERRQ(ierr);
72   umax  = (PetscInt)(f*ai[mbs] + 1); umax += mbs + 1;
73   ierr  = PetscMalloc1(umax,&ju);CHKERRQ(ierr);
74   iu[0] = mbs+1;
75   juidx = mbs + 1; /* index for ju */
76   /* jl linked list for pivot row -- linked list for col index */
77   ierr = PetscMalloc2(mbs,&jl,mbs,&q);CHKERRQ(ierr);
78   for (i=0; i<mbs; i++) {
79     jl[i] = mbs;
80     q[i]  = 0;
81   }
82 
83   /* for each row k */
84   for (k=0; k<mbs; k++) {
85     for (i=0; i<mbs; i++) q[i] = 0;  /* to be removed! */
86     nzk  = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */
87     q[k] = mbs;
88     /* initialize nonzero structure of k-th row to row rip[k] of A */
89     jmin = ai[rip[k]] +1; /* exclude diag[k] */
90     jmax = ai[rip[k]+1];
91     for (j=jmin; j<jmax; j++) {
92       vj = rip[aj[j]]; /* col. value */
93       if (vj > k) {
94         qm = k;
95         do {
96           m = qm; qm = q[m];
97         } while (qm < vj);
98         if (qm == vj) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Duplicate entry in A\n");
99         nzk++;
100         q[m]  = vj;
101         q[vj] = qm;
102       } /* if (vj > k) */
103     } /* for (j=jmin; j<jmax; j++) */
104 
105     /* modify nonzero structure of k-th row by computing fill-in
106        for each row i to be merged in */
107     prow = k;
108     prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */
109 
110     while (prow < k) {
111       /* merge row prow into k-th row */
112       jmin = iu[prow] + 1; jmax = iu[prow+1];
113       qm   = k;
114       for (j=jmin; j<jmax; j++) {
115         vj = ju[j];
116         do {
117           m = qm; qm = q[m];
118         } while (qm < vj);
119         if (qm != vj) {
120           nzk++; q[m] = vj; q[vj] = qm; qm = vj;
121         }
122       }
123       prow = jl[prow]; /* next pivot row */
124     }
125 
126     /* add k to row list for first nonzero element in k-th row */
127     if (nzk > 0) {
128       i     = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */
129       jl[k] = jl[i]; jl[i] = k;
130     }
131     iu[k+1] = iu[k] + nzk;
132 
133     /* allocate more space to ju if needed */
134     if (iu[k+1] > umax) {
135       /* estimate how much additional space we will need */
136       /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */
137       /* just double the memory each time */
138       maxadd = umax;
139       if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2;
140       umax += maxadd;
141 
142       /* allocate a longer ju */
143       ierr = PetscMalloc1(umax,&jutmp);CHKERRQ(ierr);
144       ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr);
145       ierr = PetscFree(ju);CHKERRQ(ierr);
146       ju   = jutmp;
147       reallocs++; /* count how many times we realloc */
148     }
149 
150     /* save nonzero structure of k-th row in ju */
151     i=k;
152     while (nzk--) {
153       i           = q[i];
154       ju[juidx++] = i;
155     }
156   }
157 
158 #if defined(PETSC_USE_INFO)
159   if (ai[mbs] != 0) {
160     PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]);
161     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)f,(double)af);CHKERRQ(ierr);
162     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
163     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g);\n",(double)af);CHKERRQ(ierr);
164     ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr);
165   } else {
166     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
167   }
168 #endif
169 
170   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
171   ierr = PetscFree2(jl,q);CHKERRQ(ierr);
172 
173   /* put together the new matrix */
174   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(F,bs,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
175 
176   /* ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)iperm);CHKERRQ(ierr); */
177   b                = (Mat_SeqSBAIJ*)(F)->data;
178   b->singlemalloc  = PETSC_FALSE;
179   b->free_a        = PETSC_TRUE;
180   b->free_ij       = PETSC_TRUE;
181 
182   ierr    = PetscMalloc1((iu[mbs]+1)*bs2,&b->a);CHKERRQ(ierr);
183   b->j    = ju;
184   b->i    = iu;
185   b->diag = 0;
186   b->ilen = 0;
187   b->imax = 0;
188   b->row  = perm;
189 
190   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
191 
192   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
193 
194   b->icol = perm;
195   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
196   ierr    = PetscMalloc1(bs*mbs+bs,&b->solve_work);CHKERRQ(ierr);
197   /* In b structure:  Free imax, ilen, old a, old j.
198      Allocate idnew, solve_work, new a, new j */
199   ierr     = PetscLogObjectMemory((PetscObject)F,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
200   b->maxnz = b->nz = iu[mbs];
201 
202   (F)->info.factor_mallocs   = reallocs;
203   (F)->info.fill_ratio_given = f;
204   if (ai[mbs] != 0) {
205     (F)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]);
206   } else {
207     (F)->info.fill_ratio_needed = 0.0;
208   }
209   ierr = MatSeqSBAIJSetNumericFactorization_inplace(F,perm_identity);CHKERRQ(ierr);
210   PetscFunctionReturn(0);
211 }
212 /*
213     Symbolic U^T*D*U factorization for SBAIJ format.
214     See MatICCFactorSymbolic_SeqAIJ() for description of its data structure.
215 */
216 #include <petscbt.h>
217 #include <../src/mat/utils/freespace.h>
218 #undef __FUNCT__
219 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ"
220 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
221 {
222   Mat_SeqSBAIJ       *a = (Mat_SeqSBAIJ*)A->data;
223   Mat_SeqSBAIJ       *b;
224   PetscErrorCode     ierr;
225   PetscBool          perm_identity,missing;
226   PetscReal          fill = info->fill;
227   const PetscInt     *rip,*ai=a->i,*aj=a->j;
228   PetscInt           i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow;
229   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
230   PetscInt           nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr,*udiag;
231   PetscFreeSpaceList free_space=NULL,current_space=NULL;
232   PetscBT            lnkbt;
233 
234   PetscFunctionBegin;
235   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
236   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
237   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
238   if (bs > 1) {
239     ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(fact,A,perm,info);CHKERRQ(ierr);
240     PetscFunctionReturn(0);
241   }
242 
243   /* check whether perm is the identity mapping */
244   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
245   if (!perm_identity) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format");
246   a->permute = PETSC_FALSE;
247   ierr       = ISGetIndices(perm,&rip);CHKERRQ(ierr);
248 
249   /* initialization */
250   ierr  = PetscMalloc1(mbs+1,&ui);CHKERRQ(ierr);
251   ierr  = PetscMalloc1(mbs+1,&udiag);CHKERRQ(ierr);
252   ui[0] = 0;
253 
254   /* jl: linked list for storing indices of the pivot rows
255      il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */
256   ierr = PetscMalloc4(mbs,&ui_ptr,mbs,&il,mbs,&jl,mbs,&cols);CHKERRQ(ierr);
257   for (i=0; i<mbs; i++) {
258     jl[i] = mbs; il[i] = 0;
259   }
260 
261   /* create and initialize a linked list for storing column indices of the active row k */
262   nlnk = mbs + 1;
263   ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
264 
265   /* initial FreeSpace size is fill*(ai[mbs]+1) */
266   ierr          = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,ai[mbs]+1),&free_space);CHKERRQ(ierr);
267   current_space = free_space;
268 
269   for (k=0; k<mbs; k++) {  /* for each active row k */
270     /* initialize lnk by the column indices of row rip[k] of A */
271     nzk   = 0;
272     ncols = ai[k+1] - ai[k];
273     if (!ncols) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Empty row %D in matrix ",k);
274     for (j=0; j<ncols; j++) {
275       i       = *(aj + ai[k] + j);
276       cols[j] = i;
277     }
278     ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
279     nzk += nlnk;
280 
281     /* update lnk by computing fill-in for each pivot row to be merged in */
282     prow = jl[k]; /* 1st pivot row */
283 
284     while (prow < k) {
285       nextprow = jl[prow];
286       /* merge prow into k-th row */
287       jmin   = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */
288       jmax   = ui[prow+1];
289       ncols  = jmax-jmin;
290       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */
291       ierr   = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
292       nzk   += nlnk;
293 
294       /* update il and jl for prow */
295       if (jmin < jmax) {
296         il[prow] = jmin;
297         j        = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow;
298       }
299       prow = nextprow;
300     }
301 
302     /* if free space is not available, make more free space */
303     if (current_space->local_remaining<nzk) {
304       i    = mbs - k + 1; /* num of unfactored rows */
305       i    = PetscIntMultTruncate(i,PetscMin(nzk, i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
306       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
307       reallocs++;
308     }
309 
310     /* copy data into free space, then initialize lnk */
311     ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
312 
313     /* add the k-th row into il and jl */
314     if (nzk > 1) {
315       i     = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */
316       jl[k] = jl[i]; jl[i] = k;
317       il[k] = ui[k] + 1;
318     }
319     ui_ptr[k] = current_space->array;
320 
321     current_space->array           += nzk;
322     current_space->local_used      += nzk;
323     current_space->local_remaining -= nzk;
324 
325     ui[k+1] = ui[k] + nzk;
326   }
327 
328   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
329   ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr);
330 
331   /* destroy list of free space and other temporary array(s) */
332   ierr = PetscMalloc1(ui[mbs]+1,&uj);CHKERRQ(ierr);
333   ierr = PetscFreeSpaceContiguous_Cholesky(&free_space,uj,mbs,ui,udiag);CHKERRQ(ierr); /* store matrix factor */
334   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
335 
336   /* put together the new matrix in MATSEQSBAIJ format */
337   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
338 
339   b               = (Mat_SeqSBAIJ*)fact->data;
340   b->singlemalloc = PETSC_FALSE;
341   b->free_a       = PETSC_TRUE;
342   b->free_ij      = PETSC_TRUE;
343 
344   ierr = PetscMalloc1(ui[mbs]+1,&b->a);CHKERRQ(ierr);
345 
346   b->j         = uj;
347   b->i         = ui;
348   b->diag      = udiag;
349   b->free_diag = PETSC_TRUE;
350   b->ilen      = 0;
351   b->imax      = 0;
352   b->row       = perm;
353   b->icol      = perm;
354 
355   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
356   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
357 
358   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
359 
360   ierr = PetscMalloc1(mbs+1,&b->solve_work);CHKERRQ(ierr);
361   ierr = PetscLogObjectMemory((PetscObject)fact,ui[mbs]*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
362 
363   b->maxnz = b->nz = ui[mbs];
364 
365   fact->info.factor_mallocs   = reallocs;
366   fact->info.fill_ratio_given = fill;
367   if (ai[mbs] != 0) {
368     fact->info.fill_ratio_needed = ((PetscReal)ui[mbs])/ai[mbs];
369   } else {
370     fact->info.fill_ratio_needed = 0.0;
371   }
372 #if defined(PETSC_USE_INFO)
373   if (ai[mbs] != 0) {
374     PetscReal af = fact->info.fill_ratio_needed;
375     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)fill,(double)af);CHKERRQ(ierr);
376     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
377     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g) for best performance.\n",(double)af);CHKERRQ(ierr);
378   } else {
379     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
380   }
381 #endif
382   fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering;
383   PetscFunctionReturn(0);
384 }
385 
386 #undef __FUNCT__
387 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_inplace"
388 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_inplace(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
389 {
390   Mat_SeqSBAIJ       *a = (Mat_SeqSBAIJ*)A->data;
391   Mat_SeqSBAIJ       *b;
392   PetscErrorCode     ierr;
393   PetscBool          perm_identity,missing;
394   PetscReal          fill = info->fill;
395   const PetscInt     *rip,*ai,*aj;
396   PetscInt           i,mbs=a->mbs,bs=A->rmap->bs,reallocs=0,prow,d;
397   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
398   PetscInt           nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr;
399   PetscFreeSpaceList free_space=NULL,current_space=NULL;
400   PetscBT            lnkbt;
401 
402   PetscFunctionBegin;
403   ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr);
404   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d);
405 
406   /*
407    This code originally uses Modified Sparse Row (MSR) storage
408    (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise!
409    Then it is rewritten so the factor B takes seqsbaij format. However the associated
410    MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity,
411    thus the original code in MSR format is still used for these cases.
412    The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever
413    MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor.
414   */
415   if (bs > 1) {
416     ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(fact,A,perm,info);CHKERRQ(ierr);
417     PetscFunctionReturn(0);
418   }
419 
420   /* check whether perm is the identity mapping */
421   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
422 
423   if (perm_identity) {
424     a->permute = PETSC_FALSE;
425 
426     ai = a->i; aj = a->j;
427   } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format");
428   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
429 
430   /* initialization */
431   ierr  = PetscMalloc1(mbs+1,&ui);CHKERRQ(ierr);
432   ui[0] = 0;
433 
434   /* jl: linked list for storing indices of the pivot rows
435      il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */
436   ierr = PetscMalloc4(mbs,&ui_ptr,mbs,&il,mbs,&jl,mbs,&cols);CHKERRQ(ierr);
437   for (i=0; i<mbs; i++) {
438     jl[i] = mbs; il[i] = 0;
439   }
440 
441   /* create and initialize a linked list for storing column indices of the active row k */
442   nlnk = mbs + 1;
443   ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
444 
445   /* initial FreeSpace size is fill*(ai[mbs]+1) */
446   ierr          = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,ai[mbs]+1),&free_space);CHKERRQ(ierr);
447   current_space = free_space;
448 
449   for (k=0; k<mbs; k++) {  /* for each active row k */
450     /* initialize lnk by the column indices of row rip[k] of A */
451     nzk   = 0;
452     ncols = ai[rip[k]+1] - ai[rip[k]];
453     for (j=0; j<ncols; j++) {
454       i       = *(aj + ai[rip[k]] + j);
455       cols[j] = rip[i];
456     }
457     ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
458     nzk += nlnk;
459 
460     /* update lnk by computing fill-in for each pivot row to be merged in */
461     prow = jl[k]; /* 1st pivot row */
462 
463     while (prow < k) {
464       nextprow = jl[prow];
465       /* merge prow into k-th row */
466       jmin   = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:mbs-1) */
467       jmax   = ui[prow+1];
468       ncols  = jmax-jmin;
469       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */
470       ierr   = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
471       nzk   += nlnk;
472 
473       /* update il and jl for prow */
474       if (jmin < jmax) {
475         il[prow] = jmin;
476 
477         j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow;
478       }
479       prow = nextprow;
480     }
481 
482     /* if free space is not available, make more free space */
483     if (current_space->local_remaining<nzk) {
484       i    = mbs - k + 1; /* num of unfactored rows */
485       i    = PetscMin(PetscIntMultTruncate(i,nzk), PetscIntMultTruncate(i,i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
486       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
487       reallocs++;
488     }
489 
490     /* copy data into free space, then initialize lnk */
491     ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
492 
493     /* add the k-th row into il and jl */
494     if (nzk-1 > 0) {
495       i     = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */
496       jl[k] = jl[i]; jl[i] = k;
497       il[k] = ui[k] + 1;
498     }
499     ui_ptr[k] = current_space->array;
500 
501     current_space->array           += nzk;
502     current_space->local_used      += nzk;
503     current_space->local_remaining -= nzk;
504 
505     ui[k+1] = ui[k] + nzk;
506   }
507 
508   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
509   ierr = PetscFree4(ui_ptr,il,jl,cols);CHKERRQ(ierr);
510 
511   /* destroy list of free space and other temporary array(s) */
512   ierr = PetscMalloc1(ui[mbs]+1,&uj);CHKERRQ(ierr);
513   ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
514   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
515 
516   /* put together the new matrix in MATSEQSBAIJ format */
517   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(fact,bs,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
518 
519   b               = (Mat_SeqSBAIJ*)fact->data;
520   b->singlemalloc = PETSC_FALSE;
521   b->free_a       = PETSC_TRUE;
522   b->free_ij      = PETSC_TRUE;
523 
524   ierr = PetscMalloc1(ui[mbs]+1,&b->a);CHKERRQ(ierr);
525 
526   b->j    = uj;
527   b->i    = ui;
528   b->diag = 0;
529   b->ilen = 0;
530   b->imax = 0;
531   b->row  = perm;
532 
533   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
534 
535   ierr     = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
536   b->icol  = perm;
537   ierr     = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
538   ierr     = PetscMalloc1(mbs+1,&b->solve_work);CHKERRQ(ierr);
539   ierr     = PetscLogObjectMemory((PetscObject)fact,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
540   b->maxnz = b->nz = ui[mbs];
541 
542   fact->info.factor_mallocs   = reallocs;
543   fact->info.fill_ratio_given = fill;
544   if (ai[mbs] != 0) {
545     fact->info.fill_ratio_needed = ((PetscReal)ui[mbs])/ai[mbs];
546   } else {
547     fact->info.fill_ratio_needed = 0.0;
548   }
549 #if defined(PETSC_USE_INFO)
550   if (ai[mbs] != 0) {
551     PetscReal af = fact->info.fill_ratio_needed;
552     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)fill,(double)af);CHKERRQ(ierr);
553     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
554     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g) for best performance.\n",(double)af);CHKERRQ(ierr);
555   } else {
556     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
557   }
558 #endif
559   ierr = MatSeqSBAIJSetNumericFactorization_inplace(fact,perm_identity);CHKERRQ(ierr);
560   PetscFunctionReturn(0);
561 }
562 
563 #undef __FUNCT__
564 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N"
565 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat C,Mat A,const MatFactorInfo *info)
566 {
567   Mat_SeqSBAIJ   *a   = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ*)C->data;
568   IS             perm = b->row;
569   PetscErrorCode ierr;
570   const PetscInt *ai,*aj,*perm_ptr,mbs=a->mbs,*bi=b->i,*bj=b->j;
571   PetscInt       i,j;
572   PetscInt       *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
573   PetscInt       bs  =A->rmap->bs,bs2 = a->bs2;
574   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
575   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
576   MatScalar      *work;
577   PetscInt       *pivots;
578 
579   PetscFunctionBegin;
580   /* initialization */
581   ierr = PetscCalloc1(bs2*mbs,&rtmp);CHKERRQ(ierr);
582   ierr = PetscMalloc2(mbs,&il,mbs,&jl);CHKERRQ(ierr);
583   for (i=0; i<mbs; i++) {
584     jl[i] = mbs; il[0] = 0;
585   }
586   ierr = PetscMalloc3(bs2,&dk,bs2,&uik,bs,&work);CHKERRQ(ierr);
587   ierr = PetscMalloc1(bs,&pivots);CHKERRQ(ierr);
588 
589   ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr);
590 
591   /* check permutation */
592   if (!a->permute) {
593     ai = a->i; aj = a->j; aa = a->a;
594   } else {
595     ai   = a->inew; aj = a->jnew;
596     ierr = PetscMalloc1(bs2*ai[mbs],&aa);CHKERRQ(ierr);
597     ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr);
598     ierr = PetscMalloc1(ai[mbs],&a2anew);CHKERRQ(ierr);
599     ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr);
600 
601     for (i=0; i<mbs; i++) {
602       jmin = ai[i]; jmax = ai[i+1];
603       for (j=jmin; j<jmax; j++) {
604         while (a2anew[j] != j) {
605           k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k;
606           for (k1=0; k1<bs2; k1++) {
607             dk[k1]       = aa[k*bs2+k1];
608             aa[k*bs2+k1] = aa[j*bs2+k1];
609             aa[j*bs2+k1] = dk[k1];
610           }
611         }
612         /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */
613         if (i > aj[j]) {
614           /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */
615           ap = aa + j*bs2;                     /* ptr to the beginning of j-th block of aa */
616           for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */
617           for (k=0; k<bs; k++) {               /* j-th block of aa <- dk^T */
618             for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1];
619           }
620         }
621       }
622     }
623     ierr = PetscFree(a2anew);CHKERRQ(ierr);
624   }
625 
626   /* for each row k */
627   for (k = 0; k<mbs; k++) {
628 
629     /*initialize k-th row with elements nonzero in row perm(k) of A */
630     jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1];
631 
632     ap = aa + jmin*bs2;
633     for (j = jmin; j < jmax; j++) {
634       vj       = perm_ptr[aj[j]];   /* block col. index */
635       rtmp_ptr = rtmp + vj*bs2;
636       for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++;
637     }
638 
639     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
640     ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr);
641     i    = jl[k]; /* first row to be added to k_th row  */
642 
643     while (i < k) {
644       nexti = jl[i]; /* next row to be added to k_th row */
645 
646       /* compute multiplier */
647       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
648 
649       /* uik = -inv(Di)*U_bar(i,k) */
650       diag = ba + i*bs2;
651       u    = ba + ili*bs2;
652       ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
653       PetscKernel_A_gets_A_minus_B_times_C(bs,uik,diag,u);
654 
655       /* update D(k) += -U(i,k)^T * U_bar(i,k) */
656       PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u);
657       ierr = PetscLogFlops(4.0*bs*bs2);CHKERRQ(ierr);
658 
659       /* update -U(i,k) */
660       ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
661 
662       /* add multiple of row i to k-th row ... */
663       jmin = ili + 1; jmax = bi[i+1];
664       if (jmin < jmax) {
665         for (j=jmin; j<jmax; j++) {
666           /* rtmp += -U(i,k)^T * U_bar(i,j) */
667           rtmp_ptr = rtmp + bj[j]*bs2;
668           u        = ba + j*bs2;
669           PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u);
670         }
671         ierr = PetscLogFlops(2.0*bs*bs2*(jmax-jmin));CHKERRQ(ierr);
672 
673         /* ... add i to row list for next nonzero entry */
674         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
675         j     = bj[jmin];
676         jl[i] = jl[j]; jl[j] = i; /* update jl */
677       }
678       i = nexti;
679     }
680 
681     /* save nonzero entries in k-th row of U ... */
682 
683     /* invert diagonal block */
684     diag = ba+k*bs2;
685     ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr);
686     ierr = PetscKernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr);
687 
688     jmin = bi[k]; jmax = bi[k+1];
689     if (jmin < jmax) {
690       for (j=jmin; j<jmax; j++) {
691         vj       = bj[j];      /* block col. index of U */
692         u        = ba + j*bs2;
693         rtmp_ptr = rtmp + vj*bs2;
694         for (k1=0; k1<bs2; k1++) {
695           *u++        = *rtmp_ptr;
696           *rtmp_ptr++ = 0.0;
697         }
698       }
699 
700       /* ... add k to row list for first nonzero entry in k-th row */
701       il[k] = jmin;
702       i     = bj[jmin];
703       jl[k] = jl[i]; jl[i] = k;
704     }
705   }
706 
707   ierr = PetscFree(rtmp);CHKERRQ(ierr);
708   ierr = PetscFree2(il,jl);CHKERRQ(ierr);
709   ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr);
710   ierr = PetscFree(pivots);CHKERRQ(ierr);
711   if (a->permute) {
712     ierr = PetscFree(aa);CHKERRQ(ierr);
713   }
714 
715   ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr);
716 
717   C->ops->solve          = MatSolve_SeqSBAIJ_N_inplace;
718   C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_inplace;
719   C->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_N_inplace;
720   C->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_N_inplace;
721 
722   C->assembled    = PETSC_TRUE;
723   C->preallocated = PETSC_TRUE;
724 
725   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
726   PetscFunctionReturn(0);
727 }
728 
729 #undef __FUNCT__
730 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering"
731 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info)
732 {
733   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ*)C->data;
734   PetscErrorCode ierr;
735   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
736   PetscInt       *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
737   PetscInt       bs  =A->rmap->bs,bs2 = a->bs2;
738   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
739   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
740   MatScalar      *work;
741   PetscInt       *pivots;
742 
743   PetscFunctionBegin;
744   ierr = PetscCalloc1(bs2*mbs,&rtmp);CHKERRQ(ierr);
745   ierr = PetscMalloc2(mbs,&il,mbs,&jl);CHKERRQ(ierr);
746   for (i=0; i<mbs; i++) {
747     jl[i] = mbs; il[0] = 0;
748   }
749   ierr = PetscMalloc3(bs2,&dk,bs2,&uik,bs,&work);CHKERRQ(ierr);
750   ierr = PetscMalloc1(bs,&pivots);CHKERRQ(ierr);
751 
752   ai = a->i; aj = a->j; aa = a->a;
753 
754   /* for each row k */
755   for (k = 0; k<mbs; k++) {
756 
757     /*initialize k-th row with elements nonzero in row k of A */
758     jmin = ai[k]; jmax = ai[k+1];
759     ap   = aa + jmin*bs2;
760     for (j = jmin; j < jmax; j++) {
761       vj       = aj[j];   /* block col. index */
762       rtmp_ptr = rtmp + vj*bs2;
763       for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++;
764     }
765 
766     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
767     ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr);
768     i    = jl[k]; /* first row to be added to k_th row  */
769 
770     while (i < k) {
771       nexti = jl[i]; /* next row to be added to k_th row */
772 
773       /* compute multiplier */
774       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
775 
776       /* uik = -inv(Di)*U_bar(i,k) */
777       diag = ba + i*bs2;
778       u    = ba + ili*bs2;
779       ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
780       PetscKernel_A_gets_A_minus_B_times_C(bs,uik,diag,u);
781 
782       /* update D(k) += -U(i,k)^T * U_bar(i,k) */
783       PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u);
784       ierr = PetscLogFlops(2.0*bs*bs2);CHKERRQ(ierr);
785 
786       /* update -U(i,k) */
787       ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
788 
789       /* add multiple of row i to k-th row ... */
790       jmin = ili + 1; jmax = bi[i+1];
791       if (jmin < jmax) {
792         for (j=jmin; j<jmax; j++) {
793           /* rtmp += -U(i,k)^T * U_bar(i,j) */
794           rtmp_ptr = rtmp + bj[j]*bs2;
795           u        = ba + j*bs2;
796           PetscKernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u);
797         }
798         ierr = PetscLogFlops(2.0*bs*bs2*(jmax-jmin));CHKERRQ(ierr);
799 
800         /* ... add i to row list for next nonzero entry */
801         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
802         j     = bj[jmin];
803         jl[i] = jl[j]; jl[j] = i; /* update jl */
804       }
805       i = nexti;
806     }
807 
808     /* save nonzero entries in k-th row of U ... */
809 
810     /* invert diagonal block */
811     diag = ba+k*bs2;
812     ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr);
813     ierr = PetscKernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr);
814 
815     jmin = bi[k]; jmax = bi[k+1];
816     if (jmin < jmax) {
817       for (j=jmin; j<jmax; j++) {
818         vj       = bj[j];      /* block col. index of U */
819         u        = ba + j*bs2;
820         rtmp_ptr = rtmp + vj*bs2;
821         for (k1=0; k1<bs2; k1++) {
822           *u++        = *rtmp_ptr;
823           *rtmp_ptr++ = 0.0;
824         }
825       }
826 
827       /* ... add k to row list for first nonzero entry in k-th row */
828       il[k] = jmin;
829       i     = bj[jmin];
830       jl[k] = jl[i]; jl[i] = k;
831     }
832   }
833 
834   ierr = PetscFree(rtmp);CHKERRQ(ierr);
835   ierr = PetscFree2(il,jl);CHKERRQ(ierr);
836   ierr = PetscFree3(dk,uik,work);CHKERRQ(ierr);
837   ierr = PetscFree(pivots);CHKERRQ(ierr);
838 
839   C->ops->solve          = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace;
840   C->ops->solvetranspose = MatSolve_SeqSBAIJ_N_NaturalOrdering_inplace;
841   C->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace;
842   C->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_N_NaturalOrdering_inplace;
843   C->assembled           = PETSC_TRUE;
844   C->preallocated        = PETSC_TRUE;
845 
846   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
847   PetscFunctionReturn(0);
848 }
849 
850 /*
851     Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP.
852     Version for blocks 2 by 2.
853 */
854 #undef __FUNCT__
855 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2"
856 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat C,Mat A,const MatFactorInfo *info)
857 {
858   Mat_SeqSBAIJ   *a   = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ*)C->data;
859   IS             perm = b->row;
860   PetscErrorCode ierr;
861   const PetscInt *ai,*aj,*perm_ptr;
862   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
863   PetscInt       *a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
864   MatScalar      *ba = b->a,*aa,*ap;
865   MatScalar      *u,*diag,*rtmp,*rtmp_ptr,dk[4],uik[4];
866   PetscReal      shift = info->shiftamount;
867 
868   PetscFunctionBegin;
869   /* initialization */
870   /* il and jl record the first nonzero element in each row of the accessing
871      window U(0:k, k:mbs-1).
872      jl:    list of rows to be added to uneliminated rows
873             i>= k: jl(i) is the first row to be added to row i
874             i<  k: jl(i) is the row following row i in some list of rows
875             jl(i) = mbs indicates the end of a list
876      il(i): points to the first nonzero element in columns k,...,mbs-1 of
877             row i of U */
878   ierr = PetscCalloc1(4*mbs,&rtmp);CHKERRQ(ierr);
879   ierr = PetscMalloc2(mbs,&il,mbs,&jl);CHKERRQ(ierr);
880   for (i=0; i<mbs; i++) {
881     jl[i] = mbs; il[0] = 0;
882   }
883   ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr);
884 
885   /* check permutation */
886   if (!a->permute) {
887     ai = a->i; aj = a->j; aa = a->a;
888   } else {
889     ai   = a->inew; aj = a->jnew;
890     ierr = PetscMalloc1(4*ai[mbs],&aa);CHKERRQ(ierr);
891     ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr);
892     ierr = PetscMalloc1(ai[mbs],&a2anew);CHKERRQ(ierr);
893     ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr);
894 
895     for (i=0; i<mbs; i++) {
896       jmin = ai[i]; jmax = ai[i+1];
897       for (j=jmin; j<jmax; j++) {
898         while (a2anew[j] != j) {
899           k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k;
900           for (k1=0; k1<4; k1++) {
901             dk[k1]     = aa[k*4+k1];
902             aa[k*4+k1] = aa[j*4+k1];
903             aa[j*4+k1] = dk[k1];
904           }
905         }
906         /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */
907         if (i > aj[j]) {
908           /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */
909           ap    = aa + j*4;  /* ptr to the beginning of the block */
910           dk[1] = ap[1];     /* swap ap[1] and ap[2] */
911           ap[1] = ap[2];
912           ap[2] = dk[1];
913         }
914       }
915     }
916     ierr = PetscFree(a2anew);CHKERRQ(ierr);
917   }
918 
919   /* for each row k */
920   for (k = 0; k<mbs; k++) {
921 
922     /*initialize k-th row with elements nonzero in row perm(k) of A */
923     jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1];
924     ap   = aa + jmin*4;
925     for (j = jmin; j < jmax; j++) {
926       vj       = perm_ptr[aj[j]];   /* block col. index */
927       rtmp_ptr = rtmp + vj*4;
928       for (i=0; i<4; i++) *rtmp_ptr++ = *ap++;
929     }
930 
931     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
932     ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr);
933     i    = jl[k]; /* first row to be added to k_th row  */
934 
935     while (i < k) {
936       nexti = jl[i]; /* next row to be added to k_th row */
937 
938       /* compute multiplier */
939       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
940 
941       /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */
942       diag   = ba + i*4;
943       u      = ba + ili*4;
944       uik[0] = -(diag[0]*u[0] + diag[2]*u[1]);
945       uik[1] = -(diag[1]*u[0] + diag[3]*u[1]);
946       uik[2] = -(diag[0]*u[2] + diag[2]*u[3]);
947       uik[3] = -(diag[1]*u[2] + diag[3]*u[3]);
948 
949       /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */
950       dk[0] += uik[0]*u[0] + uik[1]*u[1];
951       dk[1] += uik[2]*u[0] + uik[3]*u[1];
952       dk[2] += uik[0]*u[2] + uik[1]*u[3];
953       dk[3] += uik[2]*u[2] + uik[3]*u[3];
954 
955       ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr);
956 
957       /* update -U(i,k): ba[ili] = uik */
958       ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr);
959 
960       /* add multiple of row i to k-th row ... */
961       jmin = ili + 1; jmax = bi[i+1];
962       if (jmin < jmax) {
963         for (j=jmin; j<jmax; j++) {
964           /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */
965           rtmp_ptr     = rtmp + bj[j]*4;
966           u            = ba + j*4;
967           rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1];
968           rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1];
969           rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3];
970           rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3];
971         }
972         ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr);
973 
974         /* ... add i to row list for next nonzero entry */
975         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
976         j     = bj[jmin];
977         jl[i] = jl[j]; jl[j] = i; /* update jl */
978       }
979       i = nexti;
980     }
981 
982     /* save nonzero entries in k-th row of U ... */
983 
984     /* invert diagonal block */
985     diag = ba+k*4;
986     ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr);
987     ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr);
988 
989     jmin = bi[k]; jmax = bi[k+1];
990     if (jmin < jmax) {
991       for (j=jmin; j<jmax; j++) {
992         vj       = bj[j];      /* block col. index of U */
993         u        = ba + j*4;
994         rtmp_ptr = rtmp + vj*4;
995         for (k1=0; k1<4; k1++) {
996           *u++        = *rtmp_ptr;
997           *rtmp_ptr++ = 0.0;
998         }
999       }
1000 
1001       /* ... add k to row list for first nonzero entry in k-th row */
1002       il[k] = jmin;
1003       i     = bj[jmin];
1004       jl[k] = jl[i]; jl[i] = k;
1005     }
1006   }
1007 
1008   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1009   ierr = PetscFree2(il,jl);CHKERRQ(ierr);
1010   if (a->permute) {
1011     ierr = PetscFree(aa);CHKERRQ(ierr);
1012   }
1013   ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr);
1014 
1015   C->ops->solve          = MatSolve_SeqSBAIJ_2_inplace;
1016   C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_inplace;
1017   C->assembled           = PETSC_TRUE;
1018   C->preallocated        = PETSC_TRUE;
1019 
1020   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
1021   PetscFunctionReturn(0);
1022 }
1023 
1024 /*
1025       Version for when blocks are 2 by 2 Using natural ordering
1026 */
1027 #undef __FUNCT__
1028 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering"
1029 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info)
1030 {
1031   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ*)C->data;
1032   PetscErrorCode ierr;
1033   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
1034   PetscInt       *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
1035   MatScalar      *ba = b->a,*aa,*ap,dk[8],uik[8];
1036   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
1037   PetscReal      shift = info->shiftamount;
1038 
1039   PetscFunctionBegin;
1040   /* initialization */
1041   /* il and jl record the first nonzero element in each row of the accessing
1042      window U(0:k, k:mbs-1).
1043      jl:    list of rows to be added to uneliminated rows
1044             i>= k: jl(i) is the first row to be added to row i
1045             i<  k: jl(i) is the row following row i in some list of rows
1046             jl(i) = mbs indicates the end of a list
1047      il(i): points to the first nonzero element in columns k,...,mbs-1 of
1048             row i of U */
1049   ierr = PetscCalloc1(4*mbs,&rtmp);CHKERRQ(ierr);
1050   ierr = PetscMalloc2(mbs,&il,mbs,&jl);CHKERRQ(ierr);
1051   for (i=0; i<mbs; i++) {
1052     jl[i] = mbs; il[0] = 0;
1053   }
1054   ai = a->i; aj = a->j; aa = a->a;
1055 
1056   /* for each row k */
1057   for (k = 0; k<mbs; k++) {
1058 
1059     /*initialize k-th row with elements nonzero in row k of A */
1060     jmin = ai[k]; jmax = ai[k+1];
1061     ap   = aa + jmin*4;
1062     for (j = jmin; j < jmax; j++) {
1063       vj       = aj[j];   /* block col. index */
1064       rtmp_ptr = rtmp + vj*4;
1065       for (i=0; i<4; i++) *rtmp_ptr++ = *ap++;
1066     }
1067 
1068     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
1069     ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr);
1070     i    = jl[k]; /* first row to be added to k_th row  */
1071 
1072     while (i < k) {
1073       nexti = jl[i]; /* next row to be added to k_th row */
1074 
1075       /* compute multiplier */
1076       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
1077 
1078       /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */
1079       diag   = ba + i*4;
1080       u      = ba + ili*4;
1081       uik[0] = -(diag[0]*u[0] + diag[2]*u[1]);
1082       uik[1] = -(diag[1]*u[0] + diag[3]*u[1]);
1083       uik[2] = -(diag[0]*u[2] + diag[2]*u[3]);
1084       uik[3] = -(diag[1]*u[2] + diag[3]*u[3]);
1085 
1086       /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */
1087       dk[0] += uik[0]*u[0] + uik[1]*u[1];
1088       dk[1] += uik[2]*u[0] + uik[3]*u[1];
1089       dk[2] += uik[0]*u[2] + uik[1]*u[3];
1090       dk[3] += uik[2]*u[2] + uik[3]*u[3];
1091 
1092       ierr = PetscLogFlops(16.0*2.0);CHKERRQ(ierr);
1093 
1094       /* update -U(i,k): ba[ili] = uik */
1095       ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr);
1096 
1097       /* add multiple of row i to k-th row ... */
1098       jmin = ili + 1; jmax = bi[i+1];
1099       if (jmin < jmax) {
1100         for (j=jmin; j<jmax; j++) {
1101           /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */
1102           rtmp_ptr     = rtmp + bj[j]*4;
1103           u            = ba + j*4;
1104           rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1];
1105           rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1];
1106           rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3];
1107           rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3];
1108         }
1109         ierr = PetscLogFlops(16.0*(jmax-jmin));CHKERRQ(ierr);
1110 
1111         /* ... add i to row list for next nonzero entry */
1112         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
1113         j     = bj[jmin];
1114         jl[i] = jl[j]; jl[j] = i; /* update jl */
1115       }
1116       i = nexti;
1117     }
1118 
1119     /* save nonzero entries in k-th row of U ... */
1120 
1121     /* invert diagonal block */
1122     diag = ba+k*4;
1123     ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr);
1124     ierr = PetscKernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr);
1125 
1126     jmin = bi[k]; jmax = bi[k+1];
1127     if (jmin < jmax) {
1128       for (j=jmin; j<jmax; j++) {
1129         vj       = bj[j];      /* block col. index of U */
1130         u        = ba + j*4;
1131         rtmp_ptr = rtmp + vj*4;
1132         for (k1=0; k1<4; k1++) {
1133           *u++        = *rtmp_ptr;
1134           *rtmp_ptr++ = 0.0;
1135         }
1136       }
1137 
1138       /* ... add k to row list for first nonzero entry in k-th row */
1139       il[k] = jmin;
1140       i     = bj[jmin];
1141       jl[k] = jl[i]; jl[i] = k;
1142     }
1143   }
1144 
1145   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1146   ierr = PetscFree2(il,jl);CHKERRQ(ierr);
1147 
1148   C->ops->solve          = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace;
1149   C->ops->solvetranspose = MatSolve_SeqSBAIJ_2_NaturalOrdering_inplace;
1150   C->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace;
1151   C->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_2_NaturalOrdering_inplace;
1152   C->assembled           = PETSC_TRUE;
1153   C->preallocated        = PETSC_TRUE;
1154 
1155   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
1156   PetscFunctionReturn(0);
1157 }
1158 
1159 /*
1160     Numeric U^T*D*U factorization for SBAIJ format.
1161     Version for blocks are 1 by 1.
1162 */
1163 #undef __FUNCT__
1164 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace"
1165 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_inplace(Mat C,Mat A,const MatFactorInfo *info)
1166 {
1167   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ*)C->data;
1168   IS             ip=b->row;
1169   PetscErrorCode ierr;
1170   const PetscInt *ai,*aj,*rip;
1171   PetscInt       *a2anew,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol;
1172   PetscInt       k,jmin,jmax,*jl,*il,col,nexti,ili,nz;
1173   MatScalar      *rtmp,*ba=b->a,*bval,*aa,dk,uikdi;
1174   PetscReal      rs;
1175   FactorShiftCtx sctx;
1176 
1177   PetscFunctionBegin;
1178   /* MatPivotSetUp(): initialize shift context sctx */
1179   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
1180 
1181   ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr);
1182   if (!a->permute) {
1183     ai = a->i; aj = a->j; aa = a->a;
1184   } else {
1185     ai     = a->inew; aj = a->jnew;
1186     nz     = ai[mbs];
1187     ierr   = PetscMalloc1(nz,&aa);CHKERRQ(ierr);
1188     a2anew = a->a2anew;
1189     bval   = a->a;
1190     for (j=0; j<nz; j++) {
1191       aa[a2anew[j]] = *(bval++);
1192     }
1193   }
1194 
1195   /* initialization */
1196   /* il and jl record the first nonzero element in each row of the accessing
1197      window U(0:k, k:mbs-1).
1198      jl:    list of rows to be added to uneliminated rows
1199             i>= k: jl(i) is the first row to be added to row i
1200             i<  k: jl(i) is the row following row i in some list of rows
1201             jl(i) = mbs indicates the end of a list
1202      il(i): points to the first nonzero element in columns k,...,mbs-1 of
1203             row i of U */
1204   ierr = PetscMalloc3(mbs,&rtmp,mbs,&il,mbs,&jl);CHKERRQ(ierr);
1205 
1206   do {
1207     sctx.newshift = PETSC_FALSE;
1208     for (i=0; i<mbs; i++) {
1209       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
1210     }
1211 
1212     for (k = 0; k<mbs; k++) {
1213       /*initialize k-th row by the perm[k]-th row of A */
1214       jmin = ai[rip[k]]; jmax = ai[rip[k]+1];
1215       bval = ba + bi[k];
1216       for (j = jmin; j < jmax; j++) {
1217         col       = rip[aj[j]];
1218         rtmp[col] = aa[j];
1219         *bval++   = 0.0; /* for in-place factorization */
1220       }
1221 
1222       /* shift the diagonal of the matrix */
1223       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
1224 
1225       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1226       dk = rtmp[k];
1227       i  = jl[k]; /* first row to be added to k_th row  */
1228 
1229       while (i < k) {
1230         nexti = jl[i]; /* next row to be added to k_th row */
1231 
1232         /* compute multiplier, update diag(k) and U(i,k) */
1233         ili     = il[i]; /* index of first nonzero element in U(i,k:bms-1) */
1234         uikdi   = -ba[ili]*ba[bi[i]]; /* diagonal(k) */
1235         dk     += uikdi*ba[ili];
1236         ba[ili] = uikdi; /* -U(i,k) */
1237 
1238         /* add multiple of row i to k-th row */
1239         jmin = ili + 1; jmax = bi[i+1];
1240         if (jmin < jmax) {
1241           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
1242           ierr = PetscLogFlops(2.0*(jmax-jmin));CHKERRQ(ierr);
1243 
1244           /* update il and jl for row i */
1245           il[i] = jmin;
1246           j     = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
1247         }
1248         i = nexti;
1249       }
1250 
1251       /* shift the diagonals when zero pivot is detected */
1252       /* compute rs=sum of abs(off-diagonal) */
1253       rs   = 0.0;
1254       jmin = bi[k]+1;
1255       nz   = bi[k+1] - jmin;
1256       if (nz) {
1257         bcol = bj + jmin;
1258         while (nz--) {
1259           rs += PetscAbsScalar(rtmp[*bcol]);
1260           bcol++;
1261         }
1262       }
1263 
1264       sctx.rs = rs;
1265       sctx.pv = dk;
1266       ierr    = MatPivotCheck(C,A,info,&sctx,k);CHKERRQ(ierr);
1267       if (sctx.newshift) break;    /* sctx.shift_amount is updated */
1268       dk = sctx.pv;
1269 
1270       /* copy data into U(k,:) */
1271       ba[bi[k]] = 1.0/dk; /* U(k,k) */
1272       jmin      = bi[k]+1; jmax = bi[k+1];
1273       if (jmin < jmax) {
1274         for (j=jmin; j<jmax; j++) {
1275           col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0;
1276         }
1277         /* add the k-th row into il and jl */
1278         il[k] = jmin;
1279         i     = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
1280       }
1281     }
1282   } while (sctx.newshift);
1283   ierr = PetscFree3(rtmp,il,jl);CHKERRQ(ierr);
1284   if (a->permute) {ierr = PetscFree(aa);CHKERRQ(ierr);}
1285 
1286   ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr);
1287 
1288   C->ops->solve          = MatSolve_SeqSBAIJ_1_inplace;
1289   C->ops->solves         = MatSolves_SeqSBAIJ_1_inplace;
1290   C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_inplace;
1291   C->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1_inplace;
1292   C->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1_inplace;
1293   C->assembled           = PETSC_TRUE;
1294   C->preallocated        = PETSC_TRUE;
1295 
1296   ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr);
1297   if (sctx.nshift) {
1298     if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
1299       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
1300     } else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
1301       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
1302     }
1303   }
1304   PetscFunctionReturn(0);
1305 }
1306 
1307 /*
1308   Version for when blocks are 1 by 1 Using natural ordering under new datastructure
1309   Modified from MatCholeskyFactorNumeric_SeqAIJ()
1310 */
1311 #undef __FUNCT__
1312 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering"
1313 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat B,Mat A,const MatFactorInfo *info)
1314 {
1315   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data;
1316   Mat_SeqSBAIJ   *b=(Mat_SeqSBAIJ*)B->data;
1317   PetscErrorCode ierr;
1318   PetscInt       i,j,mbs=A->rmap->n,*bi=b->i,*bj=b->j,*bdiag=b->diag,*bjtmp;
1319   PetscInt       *ai=a->i,*aj=a->j,*ajtmp;
1320   PetscInt       k,jmin,jmax,*c2r,*il,col,nexti,ili,nz;
1321   MatScalar      *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi;
1322   FactorShiftCtx sctx;
1323   PetscReal      rs;
1324   MatScalar      d,*v;
1325 
1326   PetscFunctionBegin;
1327   ierr = PetscMalloc3(mbs,&rtmp,mbs,&il,mbs,&c2r);CHKERRQ(ierr);
1328 
1329   /* MatPivotSetUp(): initialize shift context sctx */
1330   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
1331 
1332   if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
1333     sctx.shift_top = info->zeropivot;
1334 
1335     ierr = PetscMemzero(rtmp,mbs*sizeof(MatScalar));CHKERRQ(ierr);
1336 
1337     for (i=0; i<mbs; i++) {
1338       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
1339       d        = (aa)[a->diag[i]];
1340       rtmp[i] += -PetscRealPart(d);  /* diagonal entry */
1341       ajtmp    = aj + ai[i] + 1;     /* exclude diagonal */
1342       v        = aa + ai[i] + 1;
1343       nz       = ai[i+1] - ai[i] - 1;
1344       for (j=0; j<nz; j++) {
1345         rtmp[i]        += PetscAbsScalar(v[j]);
1346         rtmp[ajtmp[j]] += PetscAbsScalar(v[j]);
1347       }
1348       if (PetscRealPart(rtmp[i]) > sctx.shift_top) sctx.shift_top = PetscRealPart(rtmp[i]);
1349     }
1350     sctx.shift_top *= 1.1;
1351     sctx.nshift_max = 5;
1352     sctx.shift_lo   = 0.;
1353     sctx.shift_hi   = 1.;
1354   }
1355 
1356   /* allocate working arrays
1357      c2r: linked list, keep track of pivot rows for a given column. c2r[col]: head of the list for a given col
1358      il:  for active k row, il[i] gives the index of the 1st nonzero entry in U[i,k:n-1] in bj and ba arrays
1359   */
1360   do {
1361     sctx.newshift = PETSC_FALSE;
1362 
1363     for (i=0; i<mbs; i++) c2r[i] = mbs;
1364     if (mbs) il[0] = 0;
1365 
1366     for (k = 0; k<mbs; k++) {
1367       /* zero rtmp */
1368       nz    = bi[k+1] - bi[k];
1369       bjtmp = bj + bi[k];
1370       for (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0;
1371 
1372       /* load in initial unfactored row */
1373       bval = ba + bi[k];
1374       jmin = ai[k]; jmax = ai[k+1];
1375       for (j = jmin; j < jmax; j++) {
1376         col       = aj[j];
1377         rtmp[col] = aa[j];
1378         *bval++   = 0.0; /* for in-place factorization */
1379       }
1380       /* shift the diagonal of the matrix: ZeropivotApply() */
1381       rtmp[k] += sctx.shift_amount;  /* shift the diagonal of the matrix */
1382 
1383       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1384       dk = rtmp[k];
1385       i  = c2r[k]; /* first row to be added to k_th row  */
1386 
1387       while (i < k) {
1388         nexti = c2r[i]; /* next row to be added to k_th row */
1389 
1390         /* compute multiplier, update diag(k) and U(i,k) */
1391         ili     = il[i]; /* index of first nonzero element in U(i,k:bms-1) */
1392         uikdi   = -ba[ili]*ba[bdiag[i]]; /* diagonal(k) */
1393         dk     += uikdi*ba[ili]; /* update diag[k] */
1394         ba[ili] = uikdi; /* -U(i,k) */
1395 
1396         /* add multiple of row i to k-th row */
1397         jmin = ili + 1; jmax = bi[i+1];
1398         if (jmin < jmax) {
1399           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
1400           /* update il and c2r for row i */
1401           il[i] = jmin;
1402           j     = bj[jmin]; c2r[i] = c2r[j]; c2r[j] = i;
1403         }
1404         i = nexti;
1405       }
1406 
1407       /* copy data into U(k,:) */
1408       rs   = 0.0;
1409       jmin = bi[k]; jmax = bi[k+1]-1;
1410       if (jmin < jmax) {
1411         for (j=jmin; j<jmax; j++) {
1412           col = bj[j]; ba[j] = rtmp[col]; rs += PetscAbsScalar(ba[j]);
1413         }
1414         /* add the k-th row into il and c2r */
1415         il[k] = jmin;
1416         i     = bj[jmin]; c2r[k] = c2r[i]; c2r[i] = k;
1417       }
1418 
1419       sctx.rs = rs;
1420       sctx.pv = dk;
1421       ierr    = MatPivotCheck(B,A,info,&sctx,k);CHKERRQ(ierr);
1422       if (sctx.newshift) break;
1423       dk = sctx.pv;
1424 
1425       ba[bdiag[k]] = 1.0/dk; /* U(k,k) */
1426     }
1427   } while (sctx.newshift);
1428 
1429   ierr = PetscFree3(rtmp,il,c2r);CHKERRQ(ierr);
1430 
1431   B->ops->solve          = MatSolve_SeqSBAIJ_1_NaturalOrdering;
1432   B->ops->solves         = MatSolves_SeqSBAIJ_1;
1433   B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering;
1434   B->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering;
1435   B->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering;
1436 
1437   B->assembled    = PETSC_TRUE;
1438   B->preallocated = PETSC_TRUE;
1439 
1440   ierr = PetscLogFlops(B->rmap->n);CHKERRQ(ierr);
1441 
1442   /* MatPivotView() */
1443   if (sctx.nshift) {
1444     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
1445       ierr = PetscInfo4(A,"number of shift_pd tries %D, shift_amount %g, diagonal shifted up by %e fraction top_value %e\n",sctx.nshift,(double)sctx.shift_amount,(double)sctx.shift_fraction,(double)sctx.shift_top);CHKERRQ(ierr);
1446     } else if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
1447       ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
1448     } else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) {
1449       ierr = PetscInfo2(A,"number of shift_inblocks applied %D, each shift_amount %g\n",sctx.nshift,(double)info->shiftamount);CHKERRQ(ierr);
1450     }
1451   }
1452   PetscFunctionReturn(0);
1453 }
1454 
1455 #undef __FUNCT__
1456 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace"
1457 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering_inplace(Mat C,Mat A,const MatFactorInfo *info)
1458 {
1459   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ*)C->data;
1460   PetscErrorCode ierr;
1461   PetscInt       i,j,mbs = a->mbs;
1462   PetscInt       *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
1463   PetscInt       k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz;
1464   MatScalar      *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval;
1465   PetscReal      rs;
1466   FactorShiftCtx sctx;
1467 
1468   PetscFunctionBegin;
1469   /* MatPivotSetUp(): initialize shift context sctx */
1470   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
1471 
1472   /* initialization */
1473   /* il and jl record the first nonzero element in each row of the accessing
1474      window U(0:k, k:mbs-1).
1475      jl:    list of rows to be added to uneliminated rows
1476             i>= k: jl(i) is the first row to be added to row i
1477             i<  k: jl(i) is the row following row i in some list of rows
1478             jl(i) = mbs indicates the end of a list
1479      il(i): points to the first nonzero element in U(i,k:mbs-1)
1480   */
1481   ierr = PetscMalloc1(mbs,&rtmp);CHKERRQ(ierr);
1482   ierr = PetscMalloc2(mbs,&il,mbs,&jl);CHKERRQ(ierr);
1483 
1484   do {
1485     sctx.newshift = PETSC_FALSE;
1486     for (i=0; i<mbs; i++) {
1487       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
1488     }
1489 
1490     for (k = 0; k<mbs; k++) {
1491       /*initialize k-th row with elements nonzero in row perm(k) of A */
1492       nz   = ai[k+1] - ai[k];
1493       acol = aj + ai[k];
1494       aval = aa + ai[k];
1495       bval = ba + bi[k];
1496       while (nz--) {
1497         rtmp[*acol++] = *aval++;
1498         *bval++       = 0.0; /* for in-place factorization */
1499       }
1500 
1501       /* shift the diagonal of the matrix */
1502       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
1503 
1504       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1505       dk = rtmp[k];
1506       i  = jl[k]; /* first row to be added to k_th row  */
1507 
1508       while (i < k) {
1509         nexti = jl[i]; /* next row to be added to k_th row */
1510         /* compute multiplier, update D(k) and U(i,k) */
1511         ili     = il[i]; /* index of first nonzero element in U(i,k:bms-1) */
1512         uikdi   = -ba[ili]*ba[bi[i]];
1513         dk     += uikdi*ba[ili];
1514         ba[ili] = uikdi; /* -U(i,k) */
1515 
1516         /* add multiple of row i to k-th row ... */
1517         jmin = ili + 1;
1518         nz   = bi[i+1] - jmin;
1519         if (nz > 0) {
1520           bcol = bj + jmin;
1521           bval = ba + jmin;
1522           ierr = PetscLogFlops(2.0*nz);CHKERRQ(ierr);
1523           while (nz--) rtmp[*bcol++] += uikdi*(*bval++);
1524 
1525           /* update il and jl for i-th row */
1526           il[i] = jmin;
1527           j     = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
1528         }
1529         i = nexti;
1530       }
1531 
1532       /* shift the diagonals when zero pivot is detected */
1533       /* compute rs=sum of abs(off-diagonal) */
1534       rs   = 0.0;
1535       jmin = bi[k]+1;
1536       nz   = bi[k+1] - jmin;
1537       if (nz) {
1538         bcol = bj + jmin;
1539         while (nz--) {
1540           rs += PetscAbsScalar(rtmp[*bcol]);
1541           bcol++;
1542         }
1543       }
1544 
1545       sctx.rs = rs;
1546       sctx.pv = dk;
1547       ierr    = MatPivotCheck(C,A,info,&sctx,k);CHKERRQ(ierr);
1548       if (sctx.newshift) break;    /* sctx.shift_amount is updated */
1549       dk = sctx.pv;
1550 
1551       /* copy data into U(k,:) */
1552       ba[bi[k]] = 1.0/dk;
1553       jmin      = bi[k]+1;
1554       nz        = bi[k+1] - jmin;
1555       if (nz) {
1556         bcol = bj + jmin;
1557         bval = ba + jmin;
1558         while (nz--) {
1559           *bval++       = rtmp[*bcol];
1560           rtmp[*bcol++] = 0.0;
1561         }
1562         /* add k-th row into il and jl */
1563         il[k] = jmin;
1564         i     = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
1565       }
1566     } /* end of for (k = 0; k<mbs; k++) */
1567   } while (sctx.newshift);
1568   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1569   ierr = PetscFree2(il,jl);CHKERRQ(ierr);
1570 
1571   C->ops->solve          = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
1572   C->ops->solves         = MatSolves_SeqSBAIJ_1_inplace;
1573   C->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
1574   C->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
1575   C->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
1576 
1577   C->assembled    = PETSC_TRUE;
1578   C->preallocated = PETSC_TRUE;
1579 
1580   ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr);
1581   if (sctx.nshift) {
1582     if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
1583       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
1584     } else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
1585       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
1586     }
1587   }
1588   PetscFunctionReturn(0);
1589 }
1590 
1591 #undef __FUNCT__
1592 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ"
1593 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,const MatFactorInfo *info)
1594 {
1595   PetscErrorCode ierr;
1596   Mat            C;
1597 
1598   PetscFunctionBegin;
1599   ierr = MatGetFactor(A,"petsc",MAT_FACTOR_CHOLESKY,&C);CHKERRQ(ierr);
1600   ierr = MatCholeskyFactorSymbolic(C,A,perm,info);CHKERRQ(ierr);
1601   ierr = MatCholeskyFactorNumeric(C,A,info);CHKERRQ(ierr);
1602 
1603   A->ops->solve          = C->ops->solve;
1604   A->ops->solvetranspose = C->ops->solvetranspose;
1605 
1606   ierr = MatHeaderMerge(A,&C);CHKERRQ(ierr);
1607   PetscFunctionReturn(0);
1608 }
1609 
1610 
1611