xref: /petsc/src/mat/impls/sbaij/seq/sbaijfact.c (revision d32f9abdbc052d6e1fd06679b17a55415c3aae30)
1 #define PETSCMAT_DLL
2 
3 #include "src/mat/impls/baij/seq/baij.h"
4 #include "src/mat/impls/sbaij/seq/sbaij.h"
5 #include "src/inline/ilu.h"
6 #include "include/petscis.h"
7 
8 #if !defined(PETSC_USE_COMPLEX)
9 /*
10   input:
11    F -- numeric factor
12   output:
13    nneg, nzero, npos: matrix inertia
14 */
15 
16 #undef __FUNCT__
17 #define __FUNCT__ "MatGetInertia_SeqSBAIJ"
18 PetscErrorCode MatGetInertia_SeqSBAIJ(Mat F,PetscInt *nneig,PetscInt *nzero,PetscInt *npos)
19 {
20   Mat_SeqSBAIJ *fact_ptr = (Mat_SeqSBAIJ*)F->data;
21   MatScalar    *dd = fact_ptr->a;
22   PetscInt     mbs=fact_ptr->mbs,bs=F->rmap.bs,i,nneig_tmp,npos_tmp,*fi = fact_ptr->i;
23 
24   PetscFunctionBegin;
25   if (bs != 1) SETERRQ1(PETSC_ERR_SUP,"No support for bs: %D >1 yet",bs);
26   nneig_tmp = 0; npos_tmp = 0;
27   for (i=0; i<mbs; i++){
28     if (PetscRealPart(dd[*fi]) > 0.0){
29       npos_tmp++;
30     } else if (PetscRealPart(dd[*fi]) < 0.0){
31       nneig_tmp++;
32     }
33     fi++;
34   }
35   if (nneig) *nneig = nneig_tmp;
36   if (npos)  *npos  = npos_tmp;
37   if (nzero) *nzero = mbs - nneig_tmp - npos_tmp;
38 
39   PetscFunctionReturn(0);
40 }
41 #endif /* !defined(PETSC_USE_COMPLEX) */
42 
43 /*
44   Symbolic U^T*D*U factorization for SBAIJ format. Modified from SSF of YSMP.
45   Use Modified Sparse Row (MSR) storage for u and ju. See page 85, "Iterative Methods ..." by Saad.
46 */
47 #undef __FUNCT__
48 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_MSR"
49 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(Mat A,IS perm,MatFactorInfo *info,Mat *B)
50 {
51   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b;
52   PetscErrorCode ierr;
53   PetscInt       *rip,i,mbs = a->mbs,*ai,*aj;
54   PetscInt       *jutmp,bs = A->rmap.bs,bs2=a->bs2;
55   PetscInt       m,reallocs = 0,prow;
56   PetscInt       *jl,*q,jmin,jmax,juidx,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd;
57   PetscReal      f = info->fill;
58   PetscTruth     perm_identity;
59 
60   PetscFunctionBegin;
61   /* check whether perm is the identity mapping */
62   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
63   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
64 
65   if (perm_identity){ /* without permutation */
66     a->permute = PETSC_FALSE;
67     ai = a->i; aj = a->j;
68   } else {            /* non-trivial permutation */
69     a->permute = PETSC_TRUE;
70     ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr);
71     ai = a->inew; aj = a->jnew;
72   }
73 
74   /* initialization */
75   ierr  = PetscMalloc((mbs+1)*sizeof(PetscInt),&iu);CHKERRQ(ierr);
76   umax  = (PetscInt)(f*ai[mbs] + 1); umax += mbs + 1;
77   ierr  = PetscMalloc(umax*sizeof(PetscInt),&ju);CHKERRQ(ierr);
78   iu[0] = mbs+1;
79   juidx = mbs + 1; /* index for ju */
80   ierr  = PetscMalloc(2*mbs*sizeof(PetscInt),&jl);CHKERRQ(ierr); /* linked list for pivot row */
81   q     = jl + mbs;   /* linked list for col index */
82   for (i=0; i<mbs; i++){
83     jl[i] = mbs;
84     q[i] = 0;
85   }
86 
87   /* for each row k */
88   for (k=0; k<mbs; k++){
89     for (i=0; i<mbs; i++) q[i] = 0;  /* to be removed! */
90     nzk  = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */
91     q[k] = mbs;
92     /* initialize nonzero structure of k-th row to row rip[k] of A */
93     jmin = ai[rip[k]] +1; /* exclude diag[k] */
94     jmax = ai[rip[k]+1];
95     for (j=jmin; j<jmax; j++){
96       vj = rip[aj[j]]; /* col. value */
97       if(vj > k){
98         qm = k;
99         do {
100           m  = qm; qm = q[m];
101         } while(qm < vj);
102         if (qm == vj) {
103           SETERRQ(PETSC_ERR_PLIB,"Duplicate entry in A\n");
104         }
105         nzk++;
106         q[m]  = vj;
107         q[vj] = qm;
108       } /* if(vj > k) */
109     } /* for (j=jmin; j<jmax; j++) */
110 
111     /* modify nonzero structure of k-th row by computing fill-in
112        for each row i to be merged in */
113     prow = k;
114     prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */
115 
116     while (prow < k){
117       /* merge row prow into k-th row */
118       jmin = iu[prow] + 1; jmax = iu[prow+1];
119       qm = k;
120       for (j=jmin; j<jmax; j++){
121         vj = ju[j];
122         do {
123           m = qm; qm = q[m];
124         } while (qm < vj);
125         if (qm != vj){
126          nzk++; q[m] = vj; q[vj] = qm; qm = vj;
127         }
128       }
129       prow = jl[prow]; /* next pivot row */
130     }
131 
132     /* add k to row list for first nonzero element in k-th row */
133     if (nzk > 0){
134       i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */
135       jl[k] = jl[i]; jl[i] = k;
136     }
137     iu[k+1] = iu[k] + nzk;
138 
139     /* allocate more space to ju if needed */
140     if (iu[k+1] > umax) {
141       /* estimate how much additional space we will need */
142       /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */
143       /* just double the memory each time */
144       maxadd = umax;
145       if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2;
146       umax += maxadd;
147 
148       /* allocate a longer ju */
149       ierr = PetscMalloc(umax*sizeof(PetscInt),&jutmp);CHKERRQ(ierr);
150       ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr);
151       ierr = PetscFree(ju);CHKERRQ(ierr);
152       ju   = jutmp;
153       reallocs++; /* count how many times we realloc */
154     }
155 
156     /* save nonzero structure of k-th row in ju */
157     i=k;
158     while (nzk --) {
159       i           = q[i];
160       ju[juidx++] = i;
161     }
162   }
163 
164 #if defined(PETSC_USE_INFO)
165   if (ai[mbs] != 0) {
166     PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]);
167     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,f,af);CHKERRQ(ierr);
168     ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr);
169     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G);\n",af);CHKERRQ(ierr);
170     ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr);
171   } else {
172     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
173   }
174 #endif
175 
176   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
177   ierr = PetscFree(jl);CHKERRQ(ierr);
178 
179   /* put together the new matrix */
180   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(*B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
181 
182   /* ierr = PetscLogObjectParent(*B,iperm);CHKERRQ(ierr); */
183   b = (Mat_SeqSBAIJ*)(*B)->data;
184   b->singlemalloc = PETSC_FALSE;
185   b->free_a       = PETSC_TRUE;
186   b->free_ij       = PETSC_TRUE;
187   ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr);
188   b->j    = ju;
189   b->i    = iu;
190   b->diag = 0;
191   b->ilen = 0;
192   b->imax = 0;
193   b->row  = perm;
194   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
195   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
196   b->icol = perm;
197   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
198   ierr    = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
199   /* In b structure:  Free imax, ilen, old a, old j.
200      Allocate idnew, solve_work, new a, new j */
201   ierr = PetscLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
202   b->maxnz = b->nz = iu[mbs];
203 
204   (*B)->factor                 = MAT_FACTOR_CHOLESKY;
205   (*B)->info.factor_mallocs    = reallocs;
206   (*B)->info.fill_ratio_given  = f;
207   if (ai[mbs] != 0) {
208     (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]);
209   } else {
210     (*B)->info.fill_ratio_needed = 0.0;
211   }
212 
213   if (perm_identity){
214     switch (bs) {
215       case 1:
216         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering;
217         (*B)->ops->solve                 = MatSolve_SeqSBAIJ_1_NaturalOrdering;
218         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering;
219         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering;
220         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=1\n");CHKERRQ(ierr);
221         break;
222       case 2:
223         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering;
224         (*B)->ops->solve           = MatSolve_SeqSBAIJ_2_NaturalOrdering;
225         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_2_NaturalOrdering;
226         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_2_NaturalOrdering;
227         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=2\n");CHKERRQ(ierr);
228         break;
229       case 3:
230         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering;
231         (*B)->ops->solve           = MatSolve_SeqSBAIJ_3_NaturalOrdering;
232         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_3_NaturalOrdering;
233         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_3_NaturalOrdering;
234         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=3\n");CHKERRQ(ierr);
235         break;
236       case 4:
237         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering;
238         (*B)->ops->solve           = MatSolve_SeqSBAIJ_4_NaturalOrdering;
239         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_4_NaturalOrdering;
240         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_4_NaturalOrdering;
241         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=4\n");CHKERRQ(ierr);
242         break;
243       case 5:
244         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering;
245         (*B)->ops->solve           = MatSolve_SeqSBAIJ_5_NaturalOrdering;
246         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_5_NaturalOrdering;
247         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_5_NaturalOrdering;
248         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=5\n");CHKERRQ(ierr);
249         break;
250       case 6:
251         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering;
252         (*B)->ops->solve           = MatSolve_SeqSBAIJ_6_NaturalOrdering;
253         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_6_NaturalOrdering;
254         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_6_NaturalOrdering;
255         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=6\n");CHKERRQ(ierr);
256         break;
257       case 7:
258         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering;
259         (*B)->ops->solve           = MatSolve_SeqSBAIJ_7_NaturalOrdering;
260         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_7_NaturalOrdering;
261         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_7_NaturalOrdering;
262         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=7\n");CHKERRQ(ierr);
263       break;
264       default:
265         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering;
266         (*B)->ops->solve           = MatSolve_SeqSBAIJ_N_NaturalOrdering;
267         (*B)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_N_NaturalOrdering;
268         (*B)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_N_NaturalOrdering;
269         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS>7\n");CHKERRQ(ierr);
270       break;
271     }
272   }
273   PetscFunctionReturn(0);
274 }
275 /*
276     Symbolic U^T*D*U factorization for SBAIJ format.
277 */
278 #include "petscbt.h"
279 #include "src/mat/utils/freespace.h"
280 #undef __FUNCT__
281 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ"
282 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info,Mat *fact)
283 {
284   Mat_SeqSBAIJ       *a = (Mat_SeqSBAIJ*)A->data;
285   Mat_SeqSBAIJ       *b;
286   PetscErrorCode     ierr;
287   PetscTruth         perm_identity,missing;
288   PetscReal          fill = info->fill;
289   PetscInt           *rip,i,mbs=a->mbs,bs=A->rmap.bs,*ai,*aj,reallocs=0,prow,d;
290   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
291   PetscInt           nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr;
292   PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL;
293   PetscBT            lnkbt;
294 
295   PetscFunctionBegin;
296   ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr);
297   if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d);
298 
299   /*
300    This code originally uses Modified Sparse Row (MSR) storage
301    (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise!
302    Then it is rewritten so the factor B takes seqsbaij format. However the associated
303    MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity,
304    thus the original code in MSR format is still used for these cases.
305    The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever
306    MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor.
307   */
308   if (bs > 1){
309     ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(A,perm,info,fact);CHKERRQ(ierr);
310     PetscFunctionReturn(0);
311   }
312 
313   /* check whether perm is the identity mapping */
314   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
315 
316   if (perm_identity){
317     a->permute = PETSC_FALSE;
318     ai = a->i; aj = a->j;
319   } else {
320     SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported for sbaij matrix. Use aij format");
321     /* There are bugs for reordeing. Needs further work.
322        MatReordering for sbaij cannot be efficient. User should use aij formt! */
323     a->permute = PETSC_TRUE;
324     ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr);
325     ai = a->inew; aj = a->jnew;
326   }
327   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
328 
329   /* initialization */
330   ierr  = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr);
331   ui[0] = 0;
332 
333   /* jl: linked list for storing indices of the pivot rows
334      il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */
335   ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt)+mbs*sizeof(PetscInt*),&jl);CHKERRQ(ierr);
336   il     = jl + mbs;
337   cols   = il + mbs;
338   ui_ptr = (PetscInt**)(cols + mbs);
339 
340   for (i=0; i<mbs; i++){
341     jl[i] = mbs; il[i] = 0;
342   }
343 
344   /* create and initialize a linked list for storing column indices of the active row k */
345   nlnk = mbs + 1;
346   ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
347 
348   /* initial FreeSpace size is fill*(ai[mbs]+1) */
349   ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr);
350   current_space = free_space;
351 
352   for (k=0; k<mbs; k++){  /* for each active row k */
353     /* initialize lnk by the column indices of row rip[k] of A */
354     nzk   = 0;
355     ncols = ai[rip[k]+1] - ai[rip[k]];
356     for (j=0; j<ncols; j++){
357       i = *(aj + ai[rip[k]] + j);
358       cols[j] = rip[i];
359     }
360     ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
361     nzk += nlnk;
362 
363     /* update lnk by computing fill-in for each pivot row to be merged in */
364     prow = jl[k]; /* 1st pivot row */
365 
366     while (prow < k){
367       nextprow = jl[prow];
368       /* merge prow into k-th row */
369       jmin = il[prow] + 1;  /* index of the 2nd nzero entry in U(prow,k:mbs-1) */
370       jmax = ui[prow+1];
371       ncols = jmax-jmin;
372       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */
373       ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
374       nzk += nlnk;
375 
376       /* update il and jl for prow */
377       if (jmin < jmax){
378         il[prow] = jmin;
379         j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow;
380       }
381       prow = nextprow;
382     }
383 
384     /* if free space is not available, make more free space */
385     if (current_space->local_remaining<nzk) {
386       i = mbs - k + 1; /* num of unfactored rows */
387       i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
388       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
389       reallocs++;
390     }
391 
392     /* copy data into free space, then initialize lnk */
393     ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
394 
395     /* add the k-th row into il and jl */
396     if (nzk-1 > 0){
397       i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */
398       jl[k] = jl[i]; jl[i] = k;
399       il[k] = ui[k] + 1;
400     }
401     ui_ptr[k] = current_space->array;
402     current_space->array           += nzk;
403     current_space->local_used      += nzk;
404     current_space->local_remaining -= nzk;
405 
406     ui[k+1] = ui[k] + nzk;
407   }
408 
409 #if defined(PETSC_USE_INFO)
410   if (ai[mbs] != 0) {
411     PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]);
412     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr);
413     ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr);
414     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr);
415   } else {
416     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
417   }
418 #endif
419 
420   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
421   ierr = PetscFree(jl);CHKERRQ(ierr);
422 
423   /* destroy list of free space and other temporary array(s) */
424   ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr);
425   ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
426   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
427 
428   /* put together the new matrix in MATSEQSBAIJ format */
429   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(*fact,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
430 
431   b = (Mat_SeqSBAIJ*)(*fact)->data;
432   b->singlemalloc = PETSC_FALSE;
433   b->free_a       = PETSC_TRUE;
434   b->free_ij      = PETSC_TRUE;
435   ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr);
436   b->j    = uj;
437   b->i    = ui;
438   b->diag = 0;
439   b->ilen = 0;
440   b->imax = 0;
441   b->row  = perm;
442   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
443   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
444   b->icol = perm;
445   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
446   ierr    = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
447   ierr    = PetscLogObjectMemory(*fact,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
448   b->maxnz = b->nz = ui[mbs];
449 
450   (*fact)->factor                 = MAT_FACTOR_CHOLESKY;
451   (*fact)->info.factor_mallocs    = reallocs;
452   (*fact)->info.fill_ratio_given  = fill;
453   if (ai[mbs] != 0) {
454     (*fact)->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]);
455   } else {
456     (*fact)->info.fill_ratio_needed = 0.0;
457   }
458 
459   if (perm_identity){
460     switch (bs) {
461       case 1:
462         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering;
463         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_1_NaturalOrdering;
464         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering;
465         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering;
466         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=1\n");CHKERRQ(ierr);
467         break;
468       case 2:
469         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering;
470         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_2_NaturalOrdering;
471         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_2_NaturalOrdering;
472         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_2_NaturalOrdering;
473         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=2\n");CHKERRQ(ierr);
474         break;
475       case 3:
476         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering;
477         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_3_NaturalOrdering;
478         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_3_NaturalOrdering;
479         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_3_NaturalOrdering;
480         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=3\n");CHKERRQ(ierr);
481         break;
482       case 4:
483         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering;
484         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_4_NaturalOrdering;
485         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_4_NaturalOrdering;
486         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_4_NaturalOrdering;
487         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=4\n");CHKERRQ(ierr);
488         break;
489       case 5:
490         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering;
491         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_5_NaturalOrdering;
492         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_5_NaturalOrdering;
493         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_5_NaturalOrdering;
494         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=5\n");CHKERRQ(ierr);
495         break;
496       case 6:
497         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering;
498         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_6_NaturalOrdering;
499         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_6_NaturalOrdering;
500         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_6_NaturalOrdering;
501         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=6\n");CHKERRQ(ierr);
502         break;
503       case 7:
504         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering;
505         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_7_NaturalOrdering;
506         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_7_NaturalOrdering;
507         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_7_NaturalOrdering;
508         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS=7\n");CHKERRQ(ierr);
509       break;
510       default:
511         (*fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering;
512         (*fact)->ops->solve           = MatSolve_SeqSBAIJ_N_NaturalOrdering;
513         (*fact)->ops->forwardsolve    = MatForwardSolve_SeqSBAIJ_N_NaturalOrdering;
514         (*fact)->ops->backwardsolve   = MatBackwardSolve_SeqSBAIJ_N_NaturalOrdering;
515         ierr = PetscInfo(A,"Using special in-place natural ordering factor and solve BS>7\n");CHKERRQ(ierr);
516       break;
517     }
518   }
519   PetscFunctionReturn(0);
520 }
521 #undef __FUNCT__
522 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N"
523 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,MatFactorInfo *info,Mat *B)
524 {
525   Mat            C = *B;
526   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
527   IS             perm = b->row;
528   PetscErrorCode ierr;
529   PetscInt       *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
530   PetscInt       *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
531   PetscInt       bs=A->rmap.bs,bs2 = a->bs2,bslog = 0;
532   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
533   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
534   MatScalar      *work;
535   PetscInt       *pivots;
536 
537   PetscFunctionBegin;
538   /* initialization */
539   ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
540   ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr);
541   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
542   jl   = il + mbs;
543   for (i=0; i<mbs; i++) {
544     jl[i] = mbs; il[0] = 0;
545   }
546   ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr);
547   uik  = dk + bs2;
548   work = uik + bs2;
549   ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr);
550 
551   ierr  = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr);
552 
553   /* check permutation */
554   if (!a->permute){
555     ai = a->i; aj = a->j; aa = a->a;
556   } else {
557     ai   = a->inew; aj = a->jnew;
558     ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr);
559     ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr);
560     ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr);
561     ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr);
562 
563     /* flops in while loop */
564     bslog = 2*bs*bs2;
565 
566     for (i=0; i<mbs; i++){
567       jmin = ai[i]; jmax = ai[i+1];
568       for (j=jmin; j<jmax; j++){
569         while (a2anew[j] != j){
570           k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k;
571           for (k1=0; k1<bs2; k1++){
572             dk[k1]       = aa[k*bs2+k1];
573             aa[k*bs2+k1] = aa[j*bs2+k1];
574             aa[j*bs2+k1] = dk[k1];
575           }
576         }
577         /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */
578         if (i > aj[j]){
579           /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */
580           ap = aa + j*bs2;                     /* ptr to the beginning of j-th block of aa */
581           for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */
582           for (k=0; k<bs; k++){               /* j-th block of aa <- dk^T */
583             for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1];
584           }
585         }
586       }
587     }
588     ierr = PetscFree(a2anew);CHKERRQ(ierr);
589   }
590 
591   /* for each row k */
592   for (k = 0; k<mbs; k++){
593 
594     /*initialize k-th row with elements nonzero in row perm(k) of A */
595     jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1];
596 
597     ap = aa + jmin*bs2;
598     for (j = jmin; j < jmax; j++){
599       vj = perm_ptr[aj[j]];         /* block col. index */
600       rtmp_ptr = rtmp + vj*bs2;
601       for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++;
602     }
603 
604     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
605     ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr);
606     i = jl[k]; /* first row to be added to k_th row  */
607 
608     while (i < k){
609       nexti = jl[i]; /* next row to be added to k_th row */
610 
611       /* compute multiplier */
612       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
613 
614       /* uik = -inv(Di)*U_bar(i,k) */
615       diag = ba + i*bs2;
616       u    = ba + ili*bs2;
617       ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
618       Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u);
619 
620       /* update D(k) += -U(i,k)^T * U_bar(i,k) */
621       Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u);
622       ierr = PetscLogFlops(bslog*2);CHKERRQ(ierr);
623 
624       /* update -U(i,k) */
625       ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
626 
627       /* add multiple of row i to k-th row ... */
628       jmin = ili + 1; jmax = bi[i+1];
629       if (jmin < jmax){
630         for (j=jmin; j<jmax; j++) {
631           /* rtmp += -U(i,k)^T * U_bar(i,j) */
632           rtmp_ptr = rtmp + bj[j]*bs2;
633           u = ba + j*bs2;
634           Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u);
635         }
636         ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr);
637 
638         /* ... add i to row list for next nonzero entry */
639         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
640         j     = bj[jmin];
641         jl[i] = jl[j]; jl[j] = i; /* update jl */
642       }
643       i = nexti;
644     }
645 
646     /* save nonzero entries in k-th row of U ... */
647 
648     /* invert diagonal block */
649     diag = ba+k*bs2;
650     ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr);
651     ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr);
652 
653     jmin = bi[k]; jmax = bi[k+1];
654     if (jmin < jmax) {
655       for (j=jmin; j<jmax; j++){
656          vj = bj[j];           /* block col. index of U */
657          u   = ba + j*bs2;
658          rtmp_ptr = rtmp + vj*bs2;
659          for (k1=0; k1<bs2; k1++){
660            *u++        = *rtmp_ptr;
661            *rtmp_ptr++ = 0.0;
662          }
663       }
664 
665       /* ... add k to row list for first nonzero entry in k-th row */
666       il[k] = jmin;
667       i     = bj[jmin];
668       jl[k] = jl[i]; jl[i] = k;
669     }
670   }
671 
672   ierr = PetscFree(rtmp);CHKERRQ(ierr);
673   ierr = PetscFree(il);CHKERRQ(ierr);
674   ierr = PetscFree(dk);CHKERRQ(ierr);
675   ierr = PetscFree(pivots);CHKERRQ(ierr);
676   if (a->permute){
677     ierr = PetscFree(aa);CHKERRQ(ierr);
678   }
679 
680   ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr);
681   C->factor       = MAT_FACTOR_CHOLESKY;
682   C->assembled    = PETSC_TRUE;
683   C->preallocated = PETSC_TRUE;
684   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
685   PetscFunctionReturn(0);
686 }
687 
688 #undef __FUNCT__
689 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering"
690 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B)
691 {
692   Mat            C = *B;
693   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
694   PetscErrorCode ierr;
695   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
696   PetscInt       *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
697   PetscInt       bs=A->rmap.bs,bs2 = a->bs2,bslog;
698   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
699   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
700   MatScalar      *work;
701   PetscInt       *pivots;
702 
703   PetscFunctionBegin;
704   ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
705   ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr);
706   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
707   jl   = il + mbs;
708   for (i=0; i<mbs; i++) {
709     jl[i] = mbs; il[0] = 0;
710   }
711   ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr);
712   uik  = dk + bs2;
713   work = uik + bs2;
714   ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr);
715 
716   ai = a->i; aj = a->j; aa = a->a;
717 
718   /* flops in while loop */
719   bslog = 2*bs*bs2;
720 
721   /* for each row k */
722   for (k = 0; k<mbs; k++){
723 
724     /*initialize k-th row with elements nonzero in row k of A */
725     jmin = ai[k]; jmax = ai[k+1];
726     ap = aa + jmin*bs2;
727     for (j = jmin; j < jmax; j++){
728       vj = aj[j];         /* block col. index */
729       rtmp_ptr = rtmp + vj*bs2;
730       for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++;
731     }
732 
733     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
734     ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr);
735     i = jl[k]; /* first row to be added to k_th row  */
736 
737     while (i < k){
738       nexti = jl[i]; /* next row to be added to k_th row */
739 
740       /* compute multiplier */
741       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
742 
743       /* uik = -inv(Di)*U_bar(i,k) */
744       diag = ba + i*bs2;
745       u    = ba + ili*bs2;
746       ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
747       Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u);
748 
749       /* update D(k) += -U(i,k)^T * U_bar(i,k) */
750       Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u);
751       ierr = PetscLogFlops(bslog*2);CHKERRQ(ierr);
752 
753       /* update -U(i,k) */
754       ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
755 
756       /* add multiple of row i to k-th row ... */
757       jmin = ili + 1; jmax = bi[i+1];
758       if (jmin < jmax){
759         for (j=jmin; j<jmax; j++) {
760           /* rtmp += -U(i,k)^T * U_bar(i,j) */
761           rtmp_ptr = rtmp + bj[j]*bs2;
762           u = ba + j*bs2;
763           Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u);
764         }
765         ierr = PetscLogFlops(bslog*(jmax-jmin));CHKERRQ(ierr);
766 
767         /* ... add i to row list for next nonzero entry */
768         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
769         j     = bj[jmin];
770         jl[i] = jl[j]; jl[j] = i; /* update jl */
771       }
772       i = nexti;
773     }
774 
775     /* save nonzero entries in k-th row of U ... */
776 
777     /* invert diagonal block */
778     diag = ba+k*bs2;
779     ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr);
780     ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr);
781 
782     jmin = bi[k]; jmax = bi[k+1];
783     if (jmin < jmax) {
784       for (j=jmin; j<jmax; j++){
785          vj = bj[j];           /* block col. index of U */
786          u   = ba + j*bs2;
787          rtmp_ptr = rtmp + vj*bs2;
788          for (k1=0; k1<bs2; k1++){
789            *u++        = *rtmp_ptr;
790            *rtmp_ptr++ = 0.0;
791          }
792       }
793 
794       /* ... add k to row list for first nonzero entry in k-th row */
795       il[k] = jmin;
796       i     = bj[jmin];
797       jl[k] = jl[i]; jl[i] = k;
798     }
799   }
800 
801   ierr = PetscFree(rtmp);CHKERRQ(ierr);
802   ierr = PetscFree(il);CHKERRQ(ierr);
803   ierr = PetscFree(dk);CHKERRQ(ierr);
804   ierr = PetscFree(pivots);CHKERRQ(ierr);
805 
806   C->factor    = MAT_FACTOR_CHOLESKY;
807   C->assembled = PETSC_TRUE;
808   C->preallocated = PETSC_TRUE;
809   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
810   PetscFunctionReturn(0);
811 }
812 
813 /*
814     Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP.
815     Version for blocks 2 by 2.
816 */
817 #undef __FUNCT__
818 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2"
819 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,MatFactorInfo *info,Mat *B)
820 {
821   Mat            C = *B;
822   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
823   IS             perm = b->row;
824   PetscErrorCode ierr;
825   PetscInt       *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
826   PetscInt       *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
827   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
828   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
829   PetscReal      shift = info->shiftinblocks;
830 
831   PetscFunctionBegin;
832   /* initialization */
833   /* il and jl record the first nonzero element in each row of the accessing
834      window U(0:k, k:mbs-1).
835      jl:    list of rows to be added to uneliminated rows
836             i>= k: jl(i) is the first row to be added to row i
837             i<  k: jl(i) is the row following row i in some list of rows
838             jl(i) = mbs indicates the end of a list
839      il(i): points to the first nonzero element in columns k,...,mbs-1 of
840             row i of U */
841   ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
842   ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr);
843   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
844   jl   = il + mbs;
845   for (i=0; i<mbs; i++) {
846     jl[i] = mbs; il[0] = 0;
847   }
848   ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr);
849   uik  = dk + 4;
850   ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr);
851 
852   /* check permutation */
853   if (!a->permute){
854     ai = a->i; aj = a->j; aa = a->a;
855   } else {
856     ai   = a->inew; aj = a->jnew;
857     ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr);
858     ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr);
859     ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr);
860     ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr);
861 
862     for (i=0; i<mbs; i++){
863       jmin = ai[i]; jmax = ai[i+1];
864       for (j=jmin; j<jmax; j++){
865         while (a2anew[j] != j){
866           k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k;
867           for (k1=0; k1<4; k1++){
868             dk[k1]       = aa[k*4+k1];
869             aa[k*4+k1] = aa[j*4+k1];
870             aa[j*4+k1] = dk[k1];
871           }
872         }
873         /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */
874         if (i > aj[j]){
875           /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */
876           ap = aa + j*4;     /* ptr to the beginning of the block */
877           dk[1] = ap[1];     /* swap ap[1] and ap[2] */
878           ap[1] = ap[2];
879           ap[2] = dk[1];
880         }
881       }
882     }
883     ierr = PetscFree(a2anew);CHKERRQ(ierr);
884   }
885 
886   /* for each row k */
887   for (k = 0; k<mbs; k++){
888 
889     /*initialize k-th row with elements nonzero in row perm(k) of A */
890     jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1];
891     ap = aa + jmin*4;
892     for (j = jmin; j < jmax; j++){
893       vj = perm_ptr[aj[j]];         /* block col. index */
894       rtmp_ptr = rtmp + vj*4;
895       for (i=0; i<4; i++) *rtmp_ptr++ = *ap++;
896     }
897 
898     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
899     ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr);
900     i = jl[k]; /* first row to be added to k_th row  */
901 
902     while (i < k){
903       nexti = jl[i]; /* next row to be added to k_th row */
904 
905       /* compute multiplier */
906       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
907 
908       /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */
909       diag = ba + i*4;
910       u    = ba + ili*4;
911       uik[0] = -(diag[0]*u[0] + diag[2]*u[1]);
912       uik[1] = -(diag[1]*u[0] + diag[3]*u[1]);
913       uik[2] = -(diag[0]*u[2] + diag[2]*u[3]);
914       uik[3] = -(diag[1]*u[2] + diag[3]*u[3]);
915 
916       /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */
917       dk[0] += uik[0]*u[0] + uik[1]*u[1];
918       dk[1] += uik[2]*u[0] + uik[3]*u[1];
919       dk[2] += uik[0]*u[2] + uik[1]*u[3];
920       dk[3] += uik[2]*u[2] + uik[3]*u[3];
921 
922       ierr = PetscLogFlops(16*2);CHKERRQ(ierr);
923 
924       /* update -U(i,k): ba[ili] = uik */
925       ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr);
926 
927       /* add multiple of row i to k-th row ... */
928       jmin = ili + 1; jmax = bi[i+1];
929       if (jmin < jmax){
930         for (j=jmin; j<jmax; j++) {
931           /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */
932           rtmp_ptr = rtmp + bj[j]*4;
933           u = ba + j*4;
934           rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1];
935           rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1];
936           rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3];
937           rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3];
938         }
939         ierr = PetscLogFlops(16*(jmax-jmin));CHKERRQ(ierr);
940 
941         /* ... add i to row list for next nonzero entry */
942         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
943         j     = bj[jmin];
944         jl[i] = jl[j]; jl[j] = i; /* update jl */
945       }
946       i = nexti;
947     }
948 
949     /* save nonzero entries in k-th row of U ... */
950 
951     /* invert diagonal block */
952     diag = ba+k*4;
953     ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr);
954     ierr = Kernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr);
955 
956     jmin = bi[k]; jmax = bi[k+1];
957     if (jmin < jmax) {
958       for (j=jmin; j<jmax; j++){
959          vj = bj[j];           /* block col. index of U */
960          u   = ba + j*4;
961          rtmp_ptr = rtmp + vj*4;
962          for (k1=0; k1<4; k1++){
963            *u++        = *rtmp_ptr;
964            *rtmp_ptr++ = 0.0;
965          }
966       }
967 
968       /* ... add k to row list for first nonzero entry in k-th row */
969       il[k] = jmin;
970       i     = bj[jmin];
971       jl[k] = jl[i]; jl[i] = k;
972     }
973   }
974 
975   ierr = PetscFree(rtmp);CHKERRQ(ierr);
976   ierr = PetscFree(il);CHKERRQ(ierr);
977   ierr = PetscFree(dk);CHKERRQ(ierr);
978   if (a->permute) {
979     ierr = PetscFree(aa);CHKERRQ(ierr);
980   }
981   ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr);
982   C->factor    = MAT_FACTOR_CHOLESKY;
983   C->assembled = PETSC_TRUE;
984   C->preallocated = PETSC_TRUE;
985   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
986   PetscFunctionReturn(0);
987 }
988 
989 /*
990       Version for when blocks are 2 by 2 Using natural ordering
991 */
992 #undef __FUNCT__
993 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering"
994 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B)
995 {
996   Mat            C = *B;
997   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
998   PetscErrorCode ierr;
999   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
1000   PetscInt       *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
1001   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
1002   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
1003   PetscReal      shift = info->shiftinblocks;
1004 
1005   PetscFunctionBegin;
1006   /* initialization */
1007   /* il and jl record the first nonzero element in each row of the accessing
1008      window U(0:k, k:mbs-1).
1009      jl:    list of rows to be added to uneliminated rows
1010             i>= k: jl(i) is the first row to be added to row i
1011             i<  k: jl(i) is the row following row i in some list of rows
1012             jl(i) = mbs indicates the end of a list
1013      il(i): points to the first nonzero element in columns k,...,mbs-1 of
1014             row i of U */
1015   ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
1016   ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr);
1017   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
1018   jl   = il + mbs;
1019   for (i=0; i<mbs; i++) {
1020     jl[i] = mbs; il[0] = 0;
1021   }
1022   ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr);
1023   uik  = dk + 4;
1024 
1025   ai = a->i; aj = a->j; aa = a->a;
1026 
1027   /* for each row k */
1028   for (k = 0; k<mbs; k++){
1029 
1030     /*initialize k-th row with elements nonzero in row k of A */
1031     jmin = ai[k]; jmax = ai[k+1];
1032     ap = aa + jmin*4;
1033     for (j = jmin; j < jmax; j++){
1034       vj = aj[j];         /* block col. index */
1035       rtmp_ptr = rtmp + vj*4;
1036       for (i=0; i<4; i++) *rtmp_ptr++ = *ap++;
1037     }
1038 
1039     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
1040     ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr);
1041     i = jl[k]; /* first row to be added to k_th row  */
1042 
1043     while (i < k){
1044       nexti = jl[i]; /* next row to be added to k_th row */
1045 
1046       /* compute multiplier */
1047       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
1048 
1049       /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */
1050       diag = ba + i*4;
1051       u    = ba + ili*4;
1052       uik[0] = -(diag[0]*u[0] + diag[2]*u[1]);
1053       uik[1] = -(diag[1]*u[0] + diag[3]*u[1]);
1054       uik[2] = -(diag[0]*u[2] + diag[2]*u[3]);
1055       uik[3] = -(diag[1]*u[2] + diag[3]*u[3]);
1056 
1057       /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */
1058       dk[0] += uik[0]*u[0] + uik[1]*u[1];
1059       dk[1] += uik[2]*u[0] + uik[3]*u[1];
1060       dk[2] += uik[0]*u[2] + uik[1]*u[3];
1061       dk[3] += uik[2]*u[2] + uik[3]*u[3];
1062 
1063       ierr = PetscLogFlops(16*2);CHKERRQ(ierr);
1064 
1065       /* update -U(i,k): ba[ili] = uik */
1066       ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr);
1067 
1068       /* add multiple of row i to k-th row ... */
1069       jmin = ili + 1; jmax = bi[i+1];
1070       if (jmin < jmax){
1071         for (j=jmin; j<jmax; j++) {
1072           /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */
1073           rtmp_ptr = rtmp + bj[j]*4;
1074           u = ba + j*4;
1075           rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1];
1076           rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1];
1077           rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3];
1078           rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3];
1079         }
1080         ierr = PetscLogFlops(16*(jmax-jmin));CHKERRQ(ierr);
1081 
1082         /* ... add i to row list for next nonzero entry */
1083         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
1084         j     = bj[jmin];
1085         jl[i] = jl[j]; jl[j] = i; /* update jl */
1086       }
1087       i = nexti;
1088     }
1089 
1090     /* save nonzero entries in k-th row of U ... */
1091 
1092     /* invert diagonal block */
1093     diag = ba+k*4;
1094     ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr);
1095     ierr = Kernel_A_gets_inverse_A_2(diag,shift);CHKERRQ(ierr);
1096 
1097     jmin = bi[k]; jmax = bi[k+1];
1098     if (jmin < jmax) {
1099       for (j=jmin; j<jmax; j++){
1100          vj = bj[j];           /* block col. index of U */
1101          u   = ba + j*4;
1102          rtmp_ptr = rtmp + vj*4;
1103          for (k1=0; k1<4; k1++){
1104            *u++        = *rtmp_ptr;
1105            *rtmp_ptr++ = 0.0;
1106          }
1107       }
1108 
1109       /* ... add k to row list for first nonzero entry in k-th row */
1110       il[k] = jmin;
1111       i     = bj[jmin];
1112       jl[k] = jl[i]; jl[i] = k;
1113     }
1114   }
1115 
1116   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1117   ierr = PetscFree(il);CHKERRQ(ierr);
1118   ierr = PetscFree(dk);CHKERRQ(ierr);
1119 
1120   C->factor    = MAT_FACTOR_CHOLESKY;
1121   C->assembled = PETSC_TRUE;
1122   C->preallocated = PETSC_TRUE;
1123   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
1124   PetscFunctionReturn(0);
1125 }
1126 
1127 /*
1128     Numeric U^T*D*U factorization for SBAIJ format.
1129     Version for blocks are 1 by 1.
1130 */
1131 #undef __FUNCT__
1132 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1"
1133 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,MatFactorInfo *info,Mat *B)
1134 {
1135   Mat            C = *B;
1136   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data;
1137   IS             ip=b->row;
1138   PetscErrorCode ierr;
1139   PetscInt       *rip,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol;
1140   PetscInt       *ai,*aj,*a2anew;
1141   PetscInt       k,jmin,jmax,*jl,*il,col,nexti,ili,nz;
1142   MatScalar      *rtmp,*ba=b->a,*bval,*aa,dk,uikdi;
1143   PetscReal      zeropivot,rs,shiftnz;
1144   PetscReal      shiftpd;
1145   ChShift_Ctx    sctx;
1146   PetscInt       newshift;
1147 
1148   PetscFunctionBegin;
1149   /* initialization */
1150   shiftnz   = info->shiftnz;
1151   shiftpd   = info->shiftpd;
1152   zeropivot = info->zeropivot;
1153 
1154   ierr  = ISGetIndices(ip,&rip);CHKERRQ(ierr);
1155   if (!a->permute){
1156     ai = a->i; aj = a->j; aa = a->a;
1157   } else {
1158     ai = a->inew; aj = a->jnew;
1159     nz = ai[mbs];
1160     ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr);
1161     a2anew = a->a2anew;
1162     bval   = a->a;
1163     for (j=0; j<nz; j++){
1164       aa[a2anew[j]] = *(bval++);
1165     }
1166   }
1167 
1168   /* initialization */
1169   /* il and jl record the first nonzero element in each row of the accessing
1170      window U(0:k, k:mbs-1).
1171      jl:    list of rows to be added to uneliminated rows
1172             i>= k: jl(i) is the first row to be added to row i
1173             i<  k: jl(i) is the row following row i in some list of rows
1174             jl(i) = mbs indicates the end of a list
1175      il(i): points to the first nonzero element in columns k,...,mbs-1 of
1176             row i of U */
1177   nz   = (2*mbs+1)*sizeof(PetscInt)+mbs*sizeof(MatScalar);
1178   ierr = PetscMalloc(nz,&il);CHKERRQ(ierr);
1179   jl   = il + mbs;
1180   rtmp = (MatScalar*)(jl + mbs);
1181 
1182   sctx.shift_amount = 0;
1183   sctx.nshift       = 0;
1184   do {
1185     sctx.chshift = PETSC_FALSE;
1186     for (i=0; i<mbs; i++) {
1187       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
1188     }
1189 
1190     for (k = 0; k<mbs; k++){
1191       /*initialize k-th row by the perm[k]-th row of A */
1192       jmin = ai[rip[k]]; jmax = ai[rip[k]+1];
1193       bval = ba + bi[k];
1194       for (j = jmin; j < jmax; j++){
1195         col = rip[aj[j]];
1196         rtmp[col] = aa[j];
1197         *bval++  = 0.0; /* for in-place factorization */
1198       }
1199 
1200       /* shift the diagonal of the matrix */
1201       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
1202 
1203       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1204       dk = rtmp[k];
1205       i = jl[k]; /* first row to be added to k_th row  */
1206 
1207       while (i < k){
1208         nexti = jl[i]; /* next row to be added to k_th row */
1209 
1210         /* compute multiplier, update diag(k) and U(i,k) */
1211         ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
1212         uikdi = - ba[ili]*ba[bi[i]];  /* diagonal(k) */
1213         dk += uikdi*ba[ili];
1214         ba[ili] = uikdi; /* -U(i,k) */
1215 
1216         /* add multiple of row i to k-th row */
1217         jmin = ili + 1; jmax = bi[i+1];
1218         if (jmin < jmax){
1219           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
1220           ierr = PetscLogFlops(2*(jmax-jmin));CHKERRQ(ierr);
1221 
1222           /* update il and jl for row i */
1223           il[i] = jmin;
1224           j = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
1225         }
1226         i = nexti;
1227       }
1228 
1229       /* shift the diagonals when zero pivot is detected */
1230       /* compute rs=sum of abs(off-diagonal) */
1231       rs   = 0.0;
1232       jmin = bi[k]+1;
1233       nz   = bi[k+1] - jmin;
1234       if (nz){
1235         bcol = bj + jmin;
1236         while (nz--){
1237           rs += PetscAbsScalar(rtmp[*bcol]);
1238           bcol++;
1239         }
1240       }
1241 
1242       sctx.rs = rs;
1243       sctx.pv = dk;
1244       ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr);
1245       if (newshift == 1) break;    /* sctx.shift_amount is updated */
1246 
1247       /* copy data into U(k,:) */
1248       ba[bi[k]] = 1.0/dk; /* U(k,k) */
1249       jmin = bi[k]+1; jmax = bi[k+1];
1250       if (jmin < jmax) {
1251         for (j=jmin; j<jmax; j++){
1252           col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0;
1253         }
1254         /* add the k-th row into il and jl */
1255         il[k] = jmin;
1256         i = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
1257       }
1258     }
1259   } while (sctx.chshift);
1260   ierr = PetscFree(il);CHKERRQ(ierr);
1261   if (a->permute){ierr = PetscFree(aa);CHKERRQ(ierr);}
1262 
1263   ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr);
1264   C->factor       = MAT_FACTOR_CHOLESKY;
1265   C->assembled    = PETSC_TRUE;
1266   C->preallocated = PETSC_TRUE;
1267   ierr = PetscLogFlops(C->rmap.N);CHKERRQ(ierr);
1268   if (sctx.nshift){
1269     if (shiftnz) {
1270       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
1271     } else if (shiftpd) {
1272       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
1273     }
1274   }
1275   PetscFunctionReturn(0);
1276 }
1277 
1278 /*
1279   Version for when blocks are 1 by 1 Using natural ordering
1280 */
1281 #undef __FUNCT__
1282 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering"
1283 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B)
1284 {
1285   Mat            C = *B;
1286   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data;
1287   PetscErrorCode ierr;
1288   PetscInt       i,j,mbs = a->mbs;
1289   PetscInt       *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
1290   PetscInt       k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz;
1291   MatScalar      *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval;
1292   PetscReal      zeropivot,rs,shiftnz;
1293   PetscReal      shiftpd;
1294   ChShift_Ctx    sctx;
1295   PetscInt       newshift;
1296 
1297   PetscFunctionBegin;
1298   /* initialization */
1299   shiftnz   = info->shiftnz;
1300   shiftpd   = info->shiftpd;
1301   zeropivot = info->zeropivot;
1302 
1303   /* il and jl record the first nonzero element in each row of the accessing
1304      window U(0:k, k:mbs-1).
1305      jl:    list of rows to be added to uneliminated rows
1306             i>= k: jl(i) is the first row to be added to row i
1307             i<  k: jl(i) is the row following row i in some list of rows
1308             jl(i) = mbs indicates the end of a list
1309      il(i): points to the first nonzero element in U(i,k:mbs-1)
1310   */
1311   ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
1312   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
1313   jl   = il + mbs;
1314 
1315   sctx.shift_amount = 0;
1316   sctx.nshift       = 0;
1317   do {
1318     sctx.chshift = PETSC_FALSE;
1319     for (i=0; i<mbs; i++) {
1320       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
1321     }
1322 
1323     for (k = 0; k<mbs; k++){
1324       /*initialize k-th row with elements nonzero in row perm(k) of A */
1325       nz   = ai[k+1] - ai[k];
1326       acol = aj + ai[k];
1327       aval = aa + ai[k];
1328       bval = ba + bi[k];
1329       while (nz -- ){
1330         rtmp[*acol++] = *aval++;
1331         *bval++       = 0.0; /* for in-place factorization */
1332       }
1333 
1334       /* shift the diagonal of the matrix */
1335       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
1336 
1337       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1338       dk = rtmp[k];
1339       i  = jl[k]; /* first row to be added to k_th row  */
1340 
1341       while (i < k){
1342         nexti = jl[i]; /* next row to be added to k_th row */
1343         /* compute multiplier, update D(k) and U(i,k) */
1344         ili   = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
1345         uikdi = - ba[ili]*ba[bi[i]];
1346         dk   += uikdi*ba[ili];
1347         ba[ili] = uikdi; /* -U(i,k) */
1348 
1349         /* add multiple of row i to k-th row ... */
1350         jmin = ili + 1;
1351         nz   = bi[i+1] - jmin;
1352         if (nz > 0){
1353           bcol = bj + jmin;
1354           bval = ba + jmin;
1355           ierr = PetscLogFlops(2*nz);CHKERRQ(ierr);
1356           while (nz --) rtmp[*bcol++] += uikdi*(*bval++);
1357 
1358           /* update il and jl for i-th row */
1359           il[i] = jmin;
1360           j = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
1361         }
1362         i = nexti;
1363       }
1364 
1365       /* shift the diagonals when zero pivot is detected */
1366       /* compute rs=sum of abs(off-diagonal) */
1367       rs   = 0.0;
1368       jmin = bi[k]+1;
1369       nz   = bi[k+1] - jmin;
1370       if (nz){
1371         bcol = bj + jmin;
1372         while (nz--){
1373           rs += PetscAbsScalar(rtmp[*bcol]);
1374           bcol++;
1375         }
1376       }
1377 
1378       sctx.rs = rs;
1379       sctx.pv = dk;
1380       ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr);
1381       if (newshift == 1) break;    /* sctx.shift_amount is updated */
1382 
1383       /* copy data into U(k,:) */
1384       ba[bi[k]] = 1.0/dk;
1385       jmin      = bi[k]+1;
1386       nz        = bi[k+1] - jmin;
1387       if (nz){
1388         bcol = bj + jmin;
1389         bval = ba + jmin;
1390         while (nz--){
1391           *bval++       = rtmp[*bcol];
1392           rtmp[*bcol++] = 0.0;
1393         }
1394         /* add k-th row into il and jl */
1395         il[k] = jmin;
1396         i = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
1397       }
1398     } /* end of for (k = 0; k<mbs; k++) */
1399   } while (sctx.chshift);
1400   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1401   ierr = PetscFree(il);CHKERRQ(ierr);
1402 
1403   C->factor       = MAT_FACTOR_CHOLESKY;
1404   C->assembled    = PETSC_TRUE;
1405   C->preallocated = PETSC_TRUE;
1406   ierr = PetscLogFlops(C->rmap.N);CHKERRQ(ierr);
1407   if (sctx.nshift){
1408     if (shiftnz) {
1409       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
1410     } else if (shiftpd) {
1411       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
1412     }
1413   }
1414   PetscFunctionReturn(0);
1415 }
1416 
1417 #undef __FUNCT__
1418 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ"
1419 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info)
1420 {
1421   PetscErrorCode ierr;
1422   Mat            C;
1423 
1424   PetscFunctionBegin;
1425   ierr = MatCholeskyFactorSymbolic(A,perm,info,&C);CHKERRQ(ierr);
1426   ierr = MatCholeskyFactorNumeric(A,info,&C);CHKERRQ(ierr);
1427   ierr = MatHeaderCopy(A,C);CHKERRQ(ierr);
1428   PetscFunctionReturn(0);
1429 }
1430 
1431 
1432