xref: /petsc/src/mat/impls/sbaij/seq/sbaijfact.c (revision 09f3b4e5628a00a1eaf17d80982cfbcc515cc9c1)
1 #define PETSCMAT_DLL
2 
3 #include "src/mat/impls/baij/seq/baij.h"
4 #include "src/mat/impls/sbaij/seq/sbaij.h"
5 #include "src/inline/ilu.h"
6 #include "include/petscis.h"
7 
8 #if !defined(PETSC_USE_COMPLEX)
9 /*
10   input:
11    F -- numeric factor
12   output:
13    nneg, nzero, npos: matrix inertia
14 */
15 
16 #undef __FUNCT__
17 #define __FUNCT__ "MatGetInertia_SeqSBAIJ"
18 PetscErrorCode MatGetInertia_SeqSBAIJ(Mat F,PetscInt *nneig,PetscInt *nzero,PetscInt *npos)
19 {
20   Mat_SeqSBAIJ *fact_ptr = (Mat_SeqSBAIJ*)F->data;
21   PetscScalar  *dd = fact_ptr->a;
22   PetscInt     mbs=fact_ptr->mbs,bs=F->bs,i,nneig_tmp,npos_tmp,*fi = fact_ptr->i;
23 
24   PetscFunctionBegin;
25   if (bs != 1) SETERRQ1(PETSC_ERR_SUP,"No support for bs: %D >1 yet",bs);
26   nneig_tmp = 0; npos_tmp = 0;
27   for (i=0; i<mbs; i++){
28     if (PetscRealPart(dd[*fi]) > 0.0){
29       npos_tmp++;
30     } else if (PetscRealPart(dd[*fi]) < 0.0){
31       nneig_tmp++;
32     }
33     fi++;
34   }
35   if (nneig) *nneig = nneig_tmp;
36   if (npos)  *npos  = npos_tmp;
37   if (nzero) *nzero = mbs - nneig_tmp - npos_tmp;
38 
39   PetscFunctionReturn(0);
40 }
41 #endif /* !defined(PETSC_USE_COMPLEX) */
42 
43 /*
44   Symbolic U^T*D*U factorization for SBAIJ format. Modified from SSF of YSMP.
45   Use Modified Sparse Row (MSR) storage for u and ju. See page 85, "Iterative Methods ..." by Saad.
46 */
47 #undef __FUNCT__
48 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ_MSR"
49 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(Mat A,IS perm,MatFactorInfo *info,Mat *B)
50 {
51   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b;
52   PetscErrorCode ierr;
53   PetscInt       *rip,i,mbs = a->mbs,*ai,*aj;
54   PetscInt       *jutmp,bs = A->bs,bs2=a->bs2;
55   PetscInt       m,reallocs = 0,prow;
56   PetscInt       *jl,*q,jmin,jmax,juidx,nzk,qm,*iu,*ju,k,j,vj,umax,maxadd;
57   PetscReal      f = info->fill;
58   PetscTruth     perm_identity;
59 
60   PetscFunctionBegin;
61   /* check whether perm is the identity mapping */
62   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
63   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
64 
65   if (perm_identity){ /* without permutation */
66     a->permute = PETSC_FALSE;
67     ai = a->i; aj = a->j;
68   } else {            /* non-trivial permutation */
69     a->permute = PETSC_TRUE;
70     ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr);
71     ai = a->inew; aj = a->jnew;
72   }
73 
74   /* initialization */
75   ierr  = PetscMalloc((mbs+1)*sizeof(PetscInt),&iu);CHKERRQ(ierr);
76   umax  = (PetscInt)(f*ai[mbs] + 1); umax += mbs + 1;
77   ierr  = PetscMalloc(umax*sizeof(PetscInt),&ju);CHKERRQ(ierr);
78   iu[0] = mbs+1;
79   juidx = mbs + 1; /* index for ju */
80   ierr  = PetscMalloc(2*mbs*sizeof(PetscInt),&jl);CHKERRQ(ierr); /* linked list for pivot row */
81   q     = jl + mbs;   /* linked list for col index */
82   for (i=0; i<mbs; i++){
83     jl[i] = mbs;
84     q[i] = 0;
85   }
86 
87   /* for each row k */
88   for (k=0; k<mbs; k++){
89     for (i=0; i<mbs; i++) q[i] = 0;  /* to be removed! */
90     nzk  = 0; /* num. of nz blocks in k-th block row with diagonal block excluded */
91     q[k] = mbs;
92     /* initialize nonzero structure of k-th row to row rip[k] of A */
93     jmin = ai[rip[k]] +1; /* exclude diag[k] */
94     jmax = ai[rip[k]+1];
95     for (j=jmin; j<jmax; j++){
96       vj = rip[aj[j]]; /* col. value */
97       if(vj > k){
98         qm = k;
99         do {
100           m  = qm; qm = q[m];
101         } while(qm < vj);
102         if (qm == vj) {
103           SETERRQ(PETSC_ERR_PLIB,"Duplicate entry in A\n");
104         }
105         nzk++;
106         q[m]  = vj;
107         q[vj] = qm;
108       } /* if(vj > k) */
109     } /* for (j=jmin; j<jmax; j++) */
110 
111     /* modify nonzero structure of k-th row by computing fill-in
112        for each row i to be merged in */
113     prow = k;
114     prow = jl[prow]; /* next pivot row (== mbs for symbolic factorization) */
115 
116     while (prow < k){
117       /* merge row prow into k-th row */
118       jmin = iu[prow] + 1; jmax = iu[prow+1];
119       qm = k;
120       for (j=jmin; j<jmax; j++){
121         vj = ju[j];
122         do {
123           m = qm; qm = q[m];
124         } while (qm < vj);
125         if (qm != vj){
126          nzk++; q[m] = vj; q[vj] = qm; qm = vj;
127         }
128       }
129       prow = jl[prow]; /* next pivot row */
130     }
131 
132     /* add k to row list for first nonzero element in k-th row */
133     if (nzk > 0){
134       i = q[k]; /* col value of first nonzero element in U(k, k+1:mbs-1) */
135       jl[k] = jl[i]; jl[i] = k;
136     }
137     iu[k+1] = iu[k] + nzk;
138 
139     /* allocate more space to ju if needed */
140     if (iu[k+1] > umax) {
141       /* estimate how much additional space we will need */
142       /* use the strategy suggested by David Hysom <hysom@perch-t.icase.edu> */
143       /* just double the memory each time */
144       maxadd = umax;
145       if (maxadd < nzk) maxadd = (mbs-k)*(nzk+1)/2;
146       umax += maxadd;
147 
148       /* allocate a longer ju */
149       ierr = PetscMalloc(umax*sizeof(PetscInt),&jutmp);CHKERRQ(ierr);
150       ierr = PetscMemcpy(jutmp,ju,iu[k]*sizeof(PetscInt));CHKERRQ(ierr);
151       ierr = PetscFree(ju);CHKERRQ(ierr);
152       ju   = jutmp;
153       reallocs++; /* count how many times we realloc */
154     }
155 
156     /* save nonzero structure of k-th row in ju */
157     i=k;
158     while (nzk --) {
159       i           = q[i];
160       ju[juidx++] = i;
161     }
162   }
163 
164 #if defined(PETSC_USE_VERBOSE)
165   if (ai[mbs] != 0) {
166     PetscReal af = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]);
167     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,f,af));CHKERRQ(ierr);
168     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af));CHKERRQ(ierr);
169     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g);\n",af));CHKERRQ(ierr);
170     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:for best performance.\n"));CHKERRQ(ierr);
171   } else {
172     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"));CHKERRQ(ierr);
173   }
174 #endif
175 
176   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
177   ierr = PetscFree(jl);CHKERRQ(ierr);
178 
179   /* put together the new matrix */
180   ierr = MatCreate(A->comm,B);CHKERRQ(ierr);
181   ierr = MatSetSizes(*B,bs*mbs,bs*mbs,bs*mbs,bs*mbs);CHKERRQ(ierr);
182   ierr = MatSetType(*B,A->type_name);CHKERRQ(ierr);
183   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(*B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
184 
185   /* ierr = PetscLogObjectParent(*B,iperm);CHKERRQ(ierr); */
186   b = (Mat_SeqSBAIJ*)(*B)->data;
187   b->singlemalloc = PETSC_FALSE;
188   ierr = PetscMalloc((iu[mbs]+1)*sizeof(MatScalar)*bs2,&b->a);CHKERRQ(ierr);
189   b->j    = ju;
190   b->i    = iu;
191   b->diag = 0;
192   b->ilen = 0;
193   b->imax = 0;
194   b->row  = perm;
195   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
196   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
197   b->icol = perm;
198   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
199   ierr    = PetscMalloc((bs*mbs+bs)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
200   /* In b structure:  Free imax, ilen, old a, old j.
201      Allocate idnew, solve_work, new a, new j */
202   ierr = PetscLogObjectMemory(*B,(iu[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
203   b->maxnz = b->nz = iu[mbs];
204 
205   (*B)->factor                 = FACTOR_CHOLESKY;
206   (*B)->info.factor_mallocs    = reallocs;
207   (*B)->info.fill_ratio_given  = f;
208   if (ai[mbs] != 0) {
209     (*B)->info.fill_ratio_needed = ((PetscReal)iu[mbs])/((PetscReal)ai[mbs]);
210   } else {
211     (*B)->info.fill_ratio_needed = 0.0;
212   }
213 
214   if (perm_identity){
215     switch (bs) {
216       case 1:
217         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering;
218         (*B)->ops->solve           = MatSolve_SeqSBAIJ_1_NaturalOrdering;
219         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"));CHKERRQ(ierr);
220         break;
221       case 2:
222         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering;
223         (*B)->ops->solve           = MatSolve_SeqSBAIJ_2_NaturalOrdering;
224         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"));CHKERRQ(ierr);
225         break;
226       case 3:
227         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering;
228         (*B)->ops->solve           = MatSolve_SeqSBAIJ_3_NaturalOrdering;
229         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"));CHKERRQ(ierr);
230         break;
231       case 4:
232         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering;
233         (*B)->ops->solve           = MatSolve_SeqSBAIJ_4_NaturalOrdering;
234         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"));CHKERRQ(ierr);
235         break;
236       case 5:
237         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering;
238         (*B)->ops->solve           = MatSolve_SeqSBAIJ_5_NaturalOrdering;
239         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"));CHKERRQ(ierr);
240         break;
241       case 6:
242         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering;
243         (*B)->ops->solve           = MatSolve_SeqSBAIJ_6_NaturalOrdering;
244         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"));CHKERRQ(ierr);
245         break;
246       case 7:
247         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering;
248         (*B)->ops->solve           = MatSolve_SeqSBAIJ_7_NaturalOrdering;
249         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"));CHKERRQ(ierr);
250       break;
251       default:
252         (*B)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering;
253         (*B)->ops->solve           = MatSolve_SeqSBAIJ_N_NaturalOrdering;
254         ierr = PetscVerboseInfo((A,"MatICCFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"));CHKERRQ(ierr);
255       break;
256     }
257   }
258   PetscFunctionReturn(0);
259 }
260 /*
261     Symbolic U^T*D*U factorization for SBAIJ format.
262 */
263 #include "petscbt.h"
264 #include "src/mat/utils/freespace.h"
265 #undef __FUNCT__
266 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqSBAIJ"
267 PetscErrorCode MatCholeskyFactorSymbolic_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info,Mat *fact)
268 {
269   Mat_SeqSBAIJ       *a = (Mat_SeqSBAIJ*)A->data;
270   Mat_SeqSBAIJ       *b;
271   Mat                B;
272   PetscErrorCode     ierr;
273   PetscTruth         perm_identity;
274   PetscReal          fill = info->fill;
275   PetscInt           *rip,i,mbs=a->mbs,bs=A->bs,*ai,*aj,reallocs=0,prow;
276   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
277   PetscInt           nlnk,*lnk,ncols,*cols,*uj,**ui_ptr,*uj_ptr;
278   PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL;
279   PetscBT            lnkbt;
280 
281   PetscFunctionBegin;
282   /*
283    This code originally uses Modified Sparse Row (MSR) storage
284    (see page 85, "Iterative Methods ..." by Saad) for the output matrix B - bad choise!
285    Then it is rewritten so the factor B takes seqsbaij format. However the associated
286    MatCholeskyFactorNumeric_() have not been modified for the cases of bs>1 or !perm_identity,
287    thus the original code in MSR format is still used for these cases.
288    The code below should replace MatCholeskyFactorSymbolic_SeqSBAIJ_MSR() whenever
289    MatCholeskyFactorNumeric_() is modified for using sbaij symbolic factor.
290   */
291   if (bs > 1){
292     ierr = MatCholeskyFactorSymbolic_SeqSBAIJ_MSR(A,perm,info,fact);CHKERRQ(ierr);
293     PetscFunctionReturn(0);
294   }
295 
296   /* check whether perm is the identity mapping */
297   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
298 
299   if (perm_identity){
300     a->permute = PETSC_FALSE;
301     ai = a->i; aj = a->j;
302   } else {
303     a->permute = PETSC_TRUE;
304     ierr = MatReorderingSeqSBAIJ(A,perm);CHKERRQ(ierr);
305     ai = a->inew; aj = a->jnew;
306   }
307   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
308 
309   /* initialization */
310   ierr  = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr);
311   ui[0] = 0;
312 
313   /* jl: linked list for storing indices of the pivot rows
314      il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */
315   ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt)+mbs*sizeof(PetscInt*),&jl);CHKERRQ(ierr);
316   il     = jl + mbs;
317   cols   = il + mbs;
318   ui_ptr = (PetscInt**)(cols + mbs);
319 
320   for (i=0; i<mbs; i++){
321     jl[i] = mbs; il[i] = 0;
322   }
323 
324   /* create and initialize a linked list for storing column indices of the active row k */
325   nlnk = mbs + 1;
326   ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
327 
328   /* initial FreeSpace size is fill*(ai[mbs]+1) */
329   ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr);
330   current_space = free_space;
331 
332   for (k=0; k<mbs; k++){  /* for each active row k */
333     /* initialize lnk by the column indices of row rip[k] of A */
334     nzk   = 0;
335     ncols = ai[rip[k]+1] - ai[rip[k]];
336     for (j=0; j<ncols; j++){
337       i = *(aj + ai[rip[k]] + j);
338       cols[j] = rip[i];
339     }
340     ierr = PetscLLAdd(ncols,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
341     nzk += nlnk;
342 
343     /* update lnk by computing fill-in for each pivot row to be merged in */
344     prow = jl[k]; /* 1st pivot row */
345 
346     while (prow < k){
347       nextprow = jl[prow];
348       /* merge prow into k-th row */
349       jmin = il[prow] + 1;  /* index of the 2nd nzero entry in U(prow,k:mbs-1) */
350       jmax = ui[prow+1];
351       ncols = jmax-jmin;
352       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */
353       ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
354       nzk += nlnk;
355 
356       /* update il and jl for prow */
357       if (jmin < jmax){
358         il[prow] = jmin;
359         j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow;
360       }
361       prow = nextprow;
362     }
363 
364     /* if free space is not available, make more free space */
365     if (current_space->local_remaining<nzk) {
366       i = mbs - k + 1; /* num of unfactored rows */
367       i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
368       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
369       reallocs++;
370     }
371 
372     /* copy data into free space, then initialize lnk */
373     ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
374 
375     /* add the k-th row into il and jl */
376     if (nzk-1 > 0){
377       i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */
378       jl[k] = jl[i]; jl[i] = k;
379       il[k] = ui[k] + 1;
380     }
381     ui_ptr[k] = current_space->array;
382     current_space->array           += nzk;
383     current_space->local_used      += nzk;
384     current_space->local_remaining -= nzk;
385 
386     ui[k+1] = ui[k] + nzk;
387   }
388 
389 #if defined(PETSC_USE_VERBOSE)
390   if (ai[mbs] != 0) {
391     PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]);
392     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Reallocs %D Fill ratio:given %g needed %g\n",reallocs,fill,af));CHKERRQ(ierr);
393     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Run with -pc_cholesky_fill %g or use \n",af));CHKERRQ(ierr);
394     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:PCCholeskySetFill(pc,%g) for best performance.\n",af));CHKERRQ(ierr);
395   } else {
396     ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Empty matrix.\n"));CHKERRQ(ierr);
397   }
398 #endif
399 
400   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
401   ierr = PetscFree(jl);CHKERRQ(ierr);
402 
403   /* destroy list of free space and other temporary array(s) */
404   ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr);
405   ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
406   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
407 
408   /* put together the new matrix in MATSEQSBAIJ format */
409   ierr = MatCreate(PETSC_COMM_SELF,fact);CHKERRQ(ierr);
410   ierr = MatSetSizes(*fact,mbs,mbs,mbs,mbs);CHKERRQ(ierr);
411   B    = *fact;
412   ierr = MatSetType(B,MATSEQSBAIJ);CHKERRQ(ierr);
413   ierr = MatSeqSBAIJSetPreallocation_SeqSBAIJ(B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
414 
415   b = (Mat_SeqSBAIJ*)B->data;
416   b->singlemalloc = PETSC_FALSE;
417   ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr);
418   b->j    = uj;
419   b->i    = ui;
420   b->diag = 0;
421   b->ilen = 0;
422   b->imax = 0;
423   b->row  = perm;
424   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
425   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
426   b->icol = perm;
427   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
428   ierr    = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
429   ierr    = PetscLogObjectMemory(B,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
430   b->maxnz = b->nz = ui[mbs];
431 
432   B->factor                 = FACTOR_CHOLESKY;
433   B->info.factor_mallocs    = reallocs;
434   B->info.fill_ratio_given  = fill;
435   if (ai[mbs] != 0) {
436     B->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]);
437   } else {
438     B->info.fill_ratio_needed = 0.0;
439   }
440 
441   if (perm_identity){
442     switch (bs) {
443       case 1:
444         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering;
445         B->ops->solve           = MatSolve_SeqSBAIJ_1_NaturalOrdering;
446         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=1\n"));CHKERRQ(ierr);
447         break;
448       case 2:
449         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering;
450         B->ops->solve           = MatSolve_SeqSBAIJ_2_NaturalOrdering;
451         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=2\n"));CHKERRQ(ierr);
452         break;
453       case 3:
454         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_3_NaturalOrdering;
455         B->ops->solve           = MatSolve_SeqSBAIJ_3_NaturalOrdering;
456         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:sing special in-place natural ordering factor and solve BS=3\n"));CHKERRQ(ierr);
457         break;
458       case 4:
459         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_4_NaturalOrdering;
460         B->ops->solve           = MatSolve_SeqSBAIJ_4_NaturalOrdering;
461         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=4\n"));CHKERRQ(ierr);
462         break;
463       case 5:
464         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_5_NaturalOrdering;
465         B->ops->solve           = MatSolve_SeqSBAIJ_5_NaturalOrdering;
466         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=5\n"));CHKERRQ(ierr);
467         break;
468       case 6:
469         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_6_NaturalOrdering;
470         B->ops->solve           = MatSolve_SeqSBAIJ_6_NaturalOrdering;
471         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=6\n"));CHKERRQ(ierr);
472         break;
473       case 7:
474         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_7_NaturalOrdering;
475         B->ops->solve           = MatSolve_SeqSBAIJ_7_NaturalOrdering;
476         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS=7\n"));CHKERRQ(ierr);
477       break;
478       default:
479         B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering;
480         B->ops->solve           = MatSolve_SeqSBAIJ_N_NaturalOrdering;
481         ierr = PetscVerboseInfo((A,"MatCholeskyFactorSymbolic_SeqSBAIJ:Using special in-place natural ordering factor and solve BS>7\n"));CHKERRQ(ierr);
482       break;
483     }
484   }
485   PetscFunctionReturn(0);
486 }
487 #undef __FUNCT__
488 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N"
489 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N(Mat A,MatFactorInfo *info,Mat *B)
490 {
491   Mat            C = *B;
492   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
493   IS             perm = b->row;
494   PetscErrorCode ierr;
495   PetscInt       *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
496   PetscInt       *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
497   PetscInt       bs=A->bs,bs2 = a->bs2;
498   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
499   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
500   MatScalar      *work;
501   PetscInt       *pivots;
502 
503   PetscFunctionBegin;
504   /* initialization */
505   ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
506   ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr);
507   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
508   jl   = il + mbs;
509   for (i=0; i<mbs; i++) {
510     jl[i] = mbs; il[0] = 0;
511   }
512   ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr);
513   uik  = dk + bs2;
514   work = uik + bs2;
515   ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr);
516 
517   ierr  = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr);
518 
519   /* check permutation */
520   if (!a->permute){
521     ai = a->i; aj = a->j; aa = a->a;
522   } else {
523     ai   = a->inew; aj = a->jnew;
524     ierr = PetscMalloc(bs2*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr);
525     ierr = PetscMemcpy(aa,a->a,bs2*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr);
526     ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr);
527     ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr);
528 
529     for (i=0; i<mbs; i++){
530       jmin = ai[i]; jmax = ai[i+1];
531       for (j=jmin; j<jmax; j++){
532         while (a2anew[j] != j){
533           k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k;
534           for (k1=0; k1<bs2; k1++){
535             dk[k1]       = aa[k*bs2+k1];
536             aa[k*bs2+k1] = aa[j*bs2+k1];
537             aa[j*bs2+k1] = dk[k1];
538           }
539         }
540         /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */
541         if (i > aj[j]){
542           /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */
543           ap = aa + j*bs2;                     /* ptr to the beginning of j-th block of aa */
544           for (k=0; k<bs2; k++) dk[k] = ap[k]; /* dk <- j-th block of aa */
545           for (k=0; k<bs; k++){               /* j-th block of aa <- dk^T */
546             for (k1=0; k1<bs; k1++) *ap++ = dk[k + bs*k1];
547           }
548         }
549       }
550     }
551     ierr = PetscFree(a2anew);CHKERRQ(ierr);
552   }
553 
554   /* for each row k */
555   for (k = 0; k<mbs; k++){
556 
557     /*initialize k-th row with elements nonzero in row perm(k) of A */
558     jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1];
559 
560     ap = aa + jmin*bs2;
561     for (j = jmin; j < jmax; j++){
562       vj = perm_ptr[aj[j]];         /* block col. index */
563       rtmp_ptr = rtmp + vj*bs2;
564       for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++;
565     }
566 
567     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
568     ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr);
569     i = jl[k]; /* first row to be added to k_th row  */
570 
571     while (i < k){
572       nexti = jl[i]; /* next row to be added to k_th row */
573 
574       /* compute multiplier */
575       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
576 
577       /* uik = -inv(Di)*U_bar(i,k) */
578       diag = ba + i*bs2;
579       u    = ba + ili*bs2;
580       ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
581       Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u);
582 
583       /* update D(k) += -U(i,k)^T * U_bar(i,k) */
584       Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u);
585 
586       /* update -U(i,k) */
587       ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
588 
589       /* add multiple of row i to k-th row ... */
590       jmin = ili + 1; jmax = bi[i+1];
591       if (jmin < jmax){
592         for (j=jmin; j<jmax; j++) {
593           /* rtmp += -U(i,k)^T * U_bar(i,j) */
594           rtmp_ptr = rtmp + bj[j]*bs2;
595           u = ba + j*bs2;
596           Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u);
597         }
598 
599         /* ... add i to row list for next nonzero entry */
600         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
601         j     = bj[jmin];
602         jl[i] = jl[j]; jl[j] = i; /* update jl */
603       }
604       i = nexti;
605     }
606 
607     /* save nonzero entries in k-th row of U ... */
608 
609     /* invert diagonal block */
610     diag = ba+k*bs2;
611     ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr);
612     ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr);
613 
614     jmin = bi[k]; jmax = bi[k+1];
615     if (jmin < jmax) {
616       for (j=jmin; j<jmax; j++){
617          vj = bj[j];           /* block col. index of U */
618          u   = ba + j*bs2;
619          rtmp_ptr = rtmp + vj*bs2;
620          for (k1=0; k1<bs2; k1++){
621            *u++        = *rtmp_ptr;
622            *rtmp_ptr++ = 0.0;
623          }
624       }
625 
626       /* ... add k to row list for first nonzero entry in k-th row */
627       il[k] = jmin;
628       i     = bj[jmin];
629       jl[k] = jl[i]; jl[i] = k;
630     }
631   }
632 
633   ierr = PetscFree(rtmp);CHKERRQ(ierr);
634   ierr = PetscFree(il);CHKERRQ(ierr);
635   ierr = PetscFree(dk);CHKERRQ(ierr);
636   ierr = PetscFree(pivots);CHKERRQ(ierr);
637   if (a->permute){
638     ierr = PetscFree(aa);CHKERRQ(ierr);
639   }
640 
641   ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr);
642   C->factor       = FACTOR_CHOLESKY;
643   C->assembled    = PETSC_TRUE;
644   C->preallocated = PETSC_TRUE;
645   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
646   PetscFunctionReturn(0);
647 }
648 
649 #undef __FUNCT__
650 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering"
651 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_N_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B)
652 {
653   Mat            C = *B;
654   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
655   PetscErrorCode ierr;
656   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
657   PetscInt       *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
658   PetscInt       bs=A->bs,bs2 = a->bs2;
659   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
660   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
661   MatScalar      *work;
662   PetscInt       *pivots;
663 
664   PetscFunctionBegin;
665   ierr = PetscMalloc(bs2*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
666   ierr = PetscMemzero(rtmp,bs2*mbs*sizeof(MatScalar));CHKERRQ(ierr);
667   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
668   jl   = il + mbs;
669   for (i=0; i<mbs; i++) {
670     jl[i] = mbs; il[0] = 0;
671   }
672   ierr = PetscMalloc((2*bs2+bs)*sizeof(MatScalar),&dk);CHKERRQ(ierr);
673   uik  = dk + bs2;
674   work = uik + bs2;
675   ierr = PetscMalloc(bs*sizeof(PetscInt),&pivots);CHKERRQ(ierr);
676 
677   ai = a->i; aj = a->j; aa = a->a;
678 
679   /* for each row k */
680   for (k = 0; k<mbs; k++){
681 
682     /*initialize k-th row with elements nonzero in row k of A */
683     jmin = ai[k]; jmax = ai[k+1];
684     ap = aa + jmin*bs2;
685     for (j = jmin; j < jmax; j++){
686       vj = aj[j];         /* block col. index */
687       rtmp_ptr = rtmp + vj*bs2;
688       for (i=0; i<bs2; i++) *rtmp_ptr++ = *ap++;
689     }
690 
691     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
692     ierr = PetscMemcpy(dk,rtmp+k*bs2,bs2*sizeof(MatScalar));CHKERRQ(ierr);
693     i = jl[k]; /* first row to be added to k_th row  */
694 
695     while (i < k){
696       nexti = jl[i]; /* next row to be added to k_th row */
697 
698       /* compute multiplier */
699       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
700 
701       /* uik = -inv(Di)*U_bar(i,k) */
702       diag = ba + i*bs2;
703       u    = ba + ili*bs2;
704       ierr = PetscMemzero(uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
705       Kernel_A_gets_A_minus_B_times_C(bs,uik,diag,u);
706 
707       /* update D(k) += -U(i,k)^T * U_bar(i,k) */
708       Kernel_A_gets_A_plus_Btranspose_times_C(bs,dk,uik,u);
709 
710       /* update -U(i,k) */
711       ierr = PetscMemcpy(ba+ili*bs2,uik,bs2*sizeof(MatScalar));CHKERRQ(ierr);
712 
713       /* add multiple of row i to k-th row ... */
714       jmin = ili + 1; jmax = bi[i+1];
715       if (jmin < jmax){
716         for (j=jmin; j<jmax; j++) {
717           /* rtmp += -U(i,k)^T * U_bar(i,j) */
718           rtmp_ptr = rtmp + bj[j]*bs2;
719           u = ba + j*bs2;
720           Kernel_A_gets_A_plus_Btranspose_times_C(bs,rtmp_ptr,uik,u);
721         }
722 
723         /* ... add i to row list for next nonzero entry */
724         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
725         j     = bj[jmin];
726         jl[i] = jl[j]; jl[j] = i; /* update jl */
727       }
728       i = nexti;
729     }
730 
731     /* save nonzero entries in k-th row of U ... */
732 
733     /* invert diagonal block */
734     diag = ba+k*bs2;
735     ierr = PetscMemcpy(diag,dk,bs2*sizeof(MatScalar));CHKERRQ(ierr);
736     ierr = Kernel_A_gets_inverse_A(bs,diag,pivots,work);CHKERRQ(ierr);
737 
738     jmin = bi[k]; jmax = bi[k+1];
739     if (jmin < jmax) {
740       for (j=jmin; j<jmax; j++){
741          vj = bj[j];           /* block col. index of U */
742          u   = ba + j*bs2;
743          rtmp_ptr = rtmp + vj*bs2;
744          for (k1=0; k1<bs2; k1++){
745            *u++        = *rtmp_ptr;
746            *rtmp_ptr++ = 0.0;
747          }
748       }
749 
750       /* ... add k to row list for first nonzero entry in k-th row */
751       il[k] = jmin;
752       i     = bj[jmin];
753       jl[k] = jl[i]; jl[i] = k;
754     }
755   }
756 
757   ierr = PetscFree(rtmp);CHKERRQ(ierr);
758   ierr = PetscFree(il);CHKERRQ(ierr);
759   ierr = PetscFree(dk);CHKERRQ(ierr);
760   ierr = PetscFree(pivots);CHKERRQ(ierr);
761 
762   C->factor    = FACTOR_CHOLESKY;
763   C->assembled = PETSC_TRUE;
764   C->preallocated = PETSC_TRUE;
765   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
766   PetscFunctionReturn(0);
767 }
768 
769 /*
770     Numeric U^T*D*U factorization for SBAIJ format. Modified from SNF of YSMP.
771     Version for blocks 2 by 2.
772 */
773 #undef __FUNCT__
774 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2"
775 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2(Mat A,MatFactorInfo *info,Mat *B)
776 {
777   Mat            C = *B;
778   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
779   IS             perm = b->row;
780   PetscErrorCode ierr;
781   PetscInt       *perm_ptr,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
782   PetscInt       *ai,*aj,*a2anew,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
783   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
784   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
785 
786   PetscFunctionBegin;
787   /* initialization */
788   /* il and jl record the first nonzero element in each row of the accessing
789      window U(0:k, k:mbs-1).
790      jl:    list of rows to be added to uneliminated rows
791             i>= k: jl(i) is the first row to be added to row i
792             i<  k: jl(i) is the row following row i in some list of rows
793             jl(i) = mbs indicates the end of a list
794      il(i): points to the first nonzero element in columns k,...,mbs-1 of
795             row i of U */
796   ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
797   ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr);
798   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
799   jl   = il + mbs;
800   for (i=0; i<mbs; i++) {
801     jl[i] = mbs; il[0] = 0;
802   }
803   ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr);
804   uik  = dk + 4;
805   ierr = ISGetIndices(perm,&perm_ptr);CHKERRQ(ierr);
806 
807   /* check permutation */
808   if (!a->permute){
809     ai = a->i; aj = a->j; aa = a->a;
810   } else {
811     ai   = a->inew; aj = a->jnew;
812     ierr = PetscMalloc(4*ai[mbs]*sizeof(MatScalar),&aa);CHKERRQ(ierr);
813     ierr = PetscMemcpy(aa,a->a,4*ai[mbs]*sizeof(MatScalar));CHKERRQ(ierr);
814     ierr = PetscMalloc(ai[mbs]*sizeof(PetscInt),&a2anew);CHKERRQ(ierr);
815     ierr = PetscMemcpy(a2anew,a->a2anew,(ai[mbs])*sizeof(PetscInt));CHKERRQ(ierr);
816 
817     for (i=0; i<mbs; i++){
818       jmin = ai[i]; jmax = ai[i+1];
819       for (j=jmin; j<jmax; j++){
820         while (a2anew[j] != j){
821           k = a2anew[j]; a2anew[j] = a2anew[k]; a2anew[k] = k;
822           for (k1=0; k1<4; k1++){
823             dk[k1]       = aa[k*4+k1];
824             aa[k*4+k1] = aa[j*4+k1];
825             aa[j*4+k1] = dk[k1];
826           }
827         }
828         /* transform columnoriented blocks that lie in the lower triangle to roworiented blocks */
829         if (i > aj[j]){
830           /* printf("change orientation, row: %d, col: %d\n",i,aj[j]); */
831           ap = aa + j*4;     /* ptr to the beginning of the block */
832           dk[1] = ap[1];     /* swap ap[1] and ap[2] */
833           ap[1] = ap[2];
834           ap[2] = dk[1];
835         }
836       }
837     }
838     ierr = PetscFree(a2anew);CHKERRQ(ierr);
839   }
840 
841   /* for each row k */
842   for (k = 0; k<mbs; k++){
843 
844     /*initialize k-th row with elements nonzero in row perm(k) of A */
845     jmin = ai[perm_ptr[k]]; jmax = ai[perm_ptr[k]+1];
846     ap = aa + jmin*4;
847     for (j = jmin; j < jmax; j++){
848       vj = perm_ptr[aj[j]];         /* block col. index */
849       rtmp_ptr = rtmp + vj*4;
850       for (i=0; i<4; i++) *rtmp_ptr++ = *ap++;
851     }
852 
853     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
854     ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr);
855     i = jl[k]; /* first row to be added to k_th row  */
856 
857     while (i < k){
858       nexti = jl[i]; /* next row to be added to k_th row */
859 
860       /* compute multiplier */
861       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
862 
863       /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */
864       diag = ba + i*4;
865       u    = ba + ili*4;
866       uik[0] = -(diag[0]*u[0] + diag[2]*u[1]);
867       uik[1] = -(diag[1]*u[0] + diag[3]*u[1]);
868       uik[2] = -(diag[0]*u[2] + diag[2]*u[3]);
869       uik[3] = -(diag[1]*u[2] + diag[3]*u[3]);
870 
871       /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */
872       dk[0] += uik[0]*u[0] + uik[1]*u[1];
873       dk[1] += uik[2]*u[0] + uik[3]*u[1];
874       dk[2] += uik[0]*u[2] + uik[1]*u[3];
875       dk[3] += uik[2]*u[2] + uik[3]*u[3];
876 
877       /* update -U(i,k): ba[ili] = uik */
878       ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr);
879 
880       /* add multiple of row i to k-th row ... */
881       jmin = ili + 1; jmax = bi[i+1];
882       if (jmin < jmax){
883         for (j=jmin; j<jmax; j++) {
884           /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */
885           rtmp_ptr = rtmp + bj[j]*4;
886           u = ba + j*4;
887           rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1];
888           rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1];
889           rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3];
890           rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3];
891         }
892 
893         /* ... add i to row list for next nonzero entry */
894         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
895         j     = bj[jmin];
896         jl[i] = jl[j]; jl[j] = i; /* update jl */
897       }
898       i = nexti;
899     }
900 
901     /* save nonzero entries in k-th row of U ... */
902 
903     /* invert diagonal block */
904     diag = ba+k*4;
905     ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr);
906     ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr);
907 
908     jmin = bi[k]; jmax = bi[k+1];
909     if (jmin < jmax) {
910       for (j=jmin; j<jmax; j++){
911          vj = bj[j];           /* block col. index of U */
912          u   = ba + j*4;
913          rtmp_ptr = rtmp + vj*4;
914          for (k1=0; k1<4; k1++){
915            *u++        = *rtmp_ptr;
916            *rtmp_ptr++ = 0.0;
917          }
918       }
919 
920       /* ... add k to row list for first nonzero entry in k-th row */
921       il[k] = jmin;
922       i     = bj[jmin];
923       jl[k] = jl[i]; jl[i] = k;
924     }
925   }
926 
927   ierr = PetscFree(rtmp);CHKERRQ(ierr);
928   ierr = PetscFree(il);CHKERRQ(ierr);
929   ierr = PetscFree(dk);CHKERRQ(ierr);
930   if (a->permute) {
931     ierr = PetscFree(aa);CHKERRQ(ierr);
932   }
933   ierr = ISRestoreIndices(perm,&perm_ptr);CHKERRQ(ierr);
934   C->factor    = FACTOR_CHOLESKY;
935   C->assembled = PETSC_TRUE;
936   C->preallocated = PETSC_TRUE;
937   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
938   PetscFunctionReturn(0);
939 }
940 
941 /*
942       Version for when blocks are 2 by 2 Using natural ordering
943 */
944 #undef __FUNCT__
945 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering"
946 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_2_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B)
947 {
948   Mat            C = *B;
949   Mat_SeqSBAIJ   *a = (Mat_SeqSBAIJ*)A->data,*b = (Mat_SeqSBAIJ *)C->data;
950   PetscErrorCode ierr;
951   PetscInt       i,j,mbs=a->mbs,*bi=b->i,*bj=b->j;
952   PetscInt       *ai,*aj,k,k1,jmin,jmax,*jl,*il,vj,nexti,ili;
953   MatScalar      *ba = b->a,*aa,*ap,*dk,*uik;
954   MatScalar      *u,*diag,*rtmp,*rtmp_ptr;
955 
956   PetscFunctionBegin;
957   /* initialization */
958   /* il and jl record the first nonzero element in each row of the accessing
959      window U(0:k, k:mbs-1).
960      jl:    list of rows to be added to uneliminated rows
961             i>= k: jl(i) is the first row to be added to row i
962             i<  k: jl(i) is the row following row i in some list of rows
963             jl(i) = mbs indicates the end of a list
964      il(i): points to the first nonzero element in columns k,...,mbs-1 of
965             row i of U */
966   ierr = PetscMalloc(4*mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
967   ierr = PetscMemzero(rtmp,4*mbs*sizeof(MatScalar));CHKERRQ(ierr);
968   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
969   jl   = il + mbs;
970   for (i=0; i<mbs; i++) {
971     jl[i] = mbs; il[0] = 0;
972   }
973   ierr = PetscMalloc(8*sizeof(MatScalar),&dk);CHKERRQ(ierr);
974   uik  = dk + 4;
975 
976   ai = a->i; aj = a->j; aa = a->a;
977 
978   /* for each row k */
979   for (k = 0; k<mbs; k++){
980 
981     /*initialize k-th row with elements nonzero in row k of A */
982     jmin = ai[k]; jmax = ai[k+1];
983     ap = aa + jmin*4;
984     for (j = jmin; j < jmax; j++){
985       vj = aj[j];         /* block col. index */
986       rtmp_ptr = rtmp + vj*4;
987       for (i=0; i<4; i++) *rtmp_ptr++ = *ap++;
988     }
989 
990     /* modify k-th row by adding in those rows i with U(i,k) != 0 */
991     ierr = PetscMemcpy(dk,rtmp+k*4,4*sizeof(MatScalar));CHKERRQ(ierr);
992     i = jl[k]; /* first row to be added to k_th row  */
993 
994     while (i < k){
995       nexti = jl[i]; /* next row to be added to k_th row */
996 
997       /* compute multiplier */
998       ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
999 
1000       /* uik = -inv(Di)*U_bar(i,k): - ba[ili]*ba[i] */
1001       diag = ba + i*4;
1002       u    = ba + ili*4;
1003       uik[0] = -(diag[0]*u[0] + diag[2]*u[1]);
1004       uik[1] = -(diag[1]*u[0] + diag[3]*u[1]);
1005       uik[2] = -(diag[0]*u[2] + diag[2]*u[3]);
1006       uik[3] = -(diag[1]*u[2] + diag[3]*u[3]);
1007 
1008       /* update D(k) += -U(i,k)^T * U_bar(i,k): dk += uik*ba[ili] */
1009       dk[0] += uik[0]*u[0] + uik[1]*u[1];
1010       dk[1] += uik[2]*u[0] + uik[3]*u[1];
1011       dk[2] += uik[0]*u[2] + uik[1]*u[3];
1012       dk[3] += uik[2]*u[2] + uik[3]*u[3];
1013 
1014       /* update -U(i,k): ba[ili] = uik */
1015       ierr = PetscMemcpy(ba+ili*4,uik,4*sizeof(MatScalar));CHKERRQ(ierr);
1016 
1017       /* add multiple of row i to k-th row ... */
1018       jmin = ili + 1; jmax = bi[i+1];
1019       if (jmin < jmax){
1020         for (j=jmin; j<jmax; j++) {
1021           /* rtmp += -U(i,k)^T * U_bar(i,j): rtmp[bj[j]] += uik*ba[j]; */
1022           rtmp_ptr = rtmp + bj[j]*4;
1023           u = ba + j*4;
1024           rtmp_ptr[0] += uik[0]*u[0] + uik[1]*u[1];
1025           rtmp_ptr[1] += uik[2]*u[0] + uik[3]*u[1];
1026           rtmp_ptr[2] += uik[0]*u[2] + uik[1]*u[3];
1027           rtmp_ptr[3] += uik[2]*u[2] + uik[3]*u[3];
1028         }
1029 
1030         /* ... add i to row list for next nonzero entry */
1031         il[i] = jmin;             /* update il(i) in column k+1, ... mbs-1 */
1032         j     = bj[jmin];
1033         jl[i] = jl[j]; jl[j] = i; /* update jl */
1034       }
1035       i = nexti;
1036     }
1037 
1038     /* save nonzero entries in k-th row of U ... */
1039 
1040     /* invert diagonal block */
1041     diag = ba+k*4;
1042     ierr = PetscMemcpy(diag,dk,4*sizeof(MatScalar));CHKERRQ(ierr);
1043     ierr = Kernel_A_gets_inverse_A_2(diag);CHKERRQ(ierr);
1044 
1045     jmin = bi[k]; jmax = bi[k+1];
1046     if (jmin < jmax) {
1047       for (j=jmin; j<jmax; j++){
1048          vj = bj[j];           /* block col. index of U */
1049          u   = ba + j*4;
1050          rtmp_ptr = rtmp + vj*4;
1051          for (k1=0; k1<4; k1++){
1052            *u++        = *rtmp_ptr;
1053            *rtmp_ptr++ = 0.0;
1054          }
1055       }
1056 
1057       /* ... add k to row list for first nonzero entry in k-th row */
1058       il[k] = jmin;
1059       i     = bj[jmin];
1060       jl[k] = jl[i]; jl[i] = k;
1061     }
1062   }
1063 
1064   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1065   ierr = PetscFree(il);CHKERRQ(ierr);
1066   ierr = PetscFree(dk);CHKERRQ(ierr);
1067 
1068   C->factor    = FACTOR_CHOLESKY;
1069   C->assembled = PETSC_TRUE;
1070   C->preallocated = PETSC_TRUE;
1071   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
1072   PetscFunctionReturn(0);
1073 }
1074 
1075 /*
1076     Numeric U^T*D*U factorization for SBAIJ format.
1077     Version for blocks are 1 by 1.
1078 */
1079 #undef __FUNCT__
1080 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1"
1081 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1(Mat A,MatFactorInfo *info,Mat *B)
1082 {
1083   Mat            C = *B;
1084   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data;
1085   IS             ip=b->row;
1086   PetscErrorCode ierr;
1087   PetscInt       *rip,i,j,mbs=a->mbs,*bi=b->i,*bj=b->j,*bcol;
1088   PetscInt       *ai,*aj,*a2anew;
1089   PetscInt       k,jmin,jmax,*jl,*il,col,nexti,ili,nz;
1090   MatScalar      *rtmp,*ba=b->a,*bval,*aa,dk,uikdi;
1091   PetscReal      zeropivot,rs,shiftnz;
1092   PetscReal      shiftpd;
1093   ChShift_Ctx    sctx;
1094   PetscInt       newshift;
1095 
1096   PetscFunctionBegin;
1097   /* initialization */
1098   shiftnz   = info->shiftnz;
1099   shiftpd   = info->shiftpd;
1100   zeropivot = info->zeropivot;
1101 
1102   ierr  = ISGetIndices(ip,&rip);CHKERRQ(ierr);
1103   if (!a->permute){
1104     ai = a->i; aj = a->j; aa = a->a;
1105   } else {
1106     ai = a->inew; aj = a->jnew;
1107     nz = ai[mbs];
1108     ierr = PetscMalloc(nz*sizeof(MatScalar),&aa);CHKERRQ(ierr);
1109     a2anew = a->a2anew;
1110     bval   = a->a;
1111     for (j=0; j<nz; j++){
1112       aa[a2anew[j]] = *(bval++);
1113     }
1114   }
1115 
1116   /* initialization */
1117   /* il and jl record the first nonzero element in each row of the accessing
1118      window U(0:k, k:mbs-1).
1119      jl:    list of rows to be added to uneliminated rows
1120             i>= k: jl(i) is the first row to be added to row i
1121             i<  k: jl(i) is the row following row i in some list of rows
1122             jl(i) = mbs indicates the end of a list
1123      il(i): points to the first nonzero element in columns k,...,mbs-1 of
1124             row i of U */
1125   nz   = (2*mbs+1)*sizeof(PetscInt)+mbs*sizeof(MatScalar);
1126   ierr = PetscMalloc(nz,&il);CHKERRQ(ierr);
1127   jl   = il + mbs;
1128   rtmp = (MatScalar*)(jl + mbs);
1129 
1130   sctx.shift_amount = 0;
1131   sctx.nshift       = 0;
1132   do {
1133     sctx.chshift = PETSC_FALSE;
1134     for (i=0; i<mbs; i++) {
1135       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
1136     }
1137 
1138     for (k = 0; k<mbs; k++){
1139       /*initialize k-th row by the perm[k]-th row of A */
1140       jmin = ai[rip[k]]; jmax = ai[rip[k]+1];
1141       bval = ba + bi[k];
1142       for (j = jmin; j < jmax; j++){
1143         col = rip[aj[j]];
1144         rtmp[col] = aa[j];
1145         *bval++  = 0.0; /* for in-place factorization */
1146       }
1147 
1148       /* shift the diagonal of the matrix */
1149       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
1150 
1151       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1152       dk = rtmp[k];
1153       i = jl[k]; /* first row to be added to k_th row  */
1154 
1155       while (i < k){
1156         nexti = jl[i]; /* next row to be added to k_th row */
1157 
1158         /* compute multiplier, update diag(k) and U(i,k) */
1159         ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
1160         uikdi = - ba[ili]*ba[bi[i]];  /* diagonal(k) */
1161         dk += uikdi*ba[ili];
1162         ba[ili] = uikdi; /* -U(i,k) */
1163 
1164         /* add multiple of row i to k-th row */
1165         jmin = ili + 1; jmax = bi[i+1];
1166         if (jmin < jmax){
1167           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
1168           /* update il and jl for row i */
1169           il[i] = jmin;
1170           j = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
1171         }
1172         i = nexti;
1173       }
1174 
1175       /* shift the diagonals when zero pivot is detected */
1176       /* compute rs=sum of abs(off-diagonal) */
1177       rs   = 0.0;
1178       jmin = bi[k]+1;
1179       nz   = bi[k+1] - jmin;
1180       if (nz){
1181         bcol = bj + jmin;
1182         while (nz--){
1183           rs += PetscAbsScalar(rtmp[*bcol]);
1184           bcol++;
1185         }
1186       }
1187 
1188       sctx.rs = rs;
1189       sctx.pv = dk;
1190       ierr = MatCholeskyCheckShift_inline(info,sctx,newshift);CHKERRQ(ierr);
1191       if (newshift == 1){
1192         break;    /* sctx.shift_amount is updated */
1193       } else if (newshift == -1){
1194         SETERRQ4(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g * rs %g",k,PetscAbsScalar(dk),zeropivot,rs);
1195       }
1196 
1197       /* copy data into U(k,:) */
1198       ba[bi[k]] = 1.0/dk; /* U(k,k) */
1199       jmin = bi[k]+1; jmax = bi[k+1];
1200       if (jmin < jmax) {
1201         for (j=jmin; j<jmax; j++){
1202           col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0;
1203         }
1204         /* add the k-th row into il and jl */
1205         il[k] = jmin;
1206         i = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
1207       }
1208     }
1209   } while (sctx.chshift);
1210   ierr = PetscFree(il);CHKERRQ(ierr);
1211   if (a->permute){ierr = PetscFree(aa);CHKERRQ(ierr);}
1212 
1213   ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr);
1214   C->factor       = FACTOR_CHOLESKY;
1215   C->assembled    = PETSC_TRUE;
1216   C->preallocated = PETSC_TRUE;
1217   ierr = PetscLogFlops(C->m);CHKERRQ(ierr);
1218     if (sctx.nshift){
1219     if (shiftnz) {
1220       ierr = PetscVerboseInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1: number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr);
1221     } else if (shiftpd) {
1222       ierr = PetscVerboseInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1: number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr);
1223     }
1224   }
1225   PetscFunctionReturn(0);
1226 }
1227 
1228 /*
1229   Version for when blocks are 1 by 1 Using natural ordering
1230 */
1231 #undef __FUNCT__
1232 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering"
1233 PetscErrorCode MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering(Mat A,MatFactorInfo *info,Mat *B)
1234 {
1235   Mat            C = *B;
1236   Mat_SeqSBAIJ   *a=(Mat_SeqSBAIJ*)A->data,*b=(Mat_SeqSBAIJ *)C->data;
1237   PetscErrorCode ierr;
1238   PetscInt       i,j,mbs = a->mbs;
1239   PetscInt       *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
1240   PetscInt       k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz;
1241   MatScalar      *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval;
1242   PetscReal      zeropivot,rs,shiftnz;
1243   PetscReal      shiftpd;
1244   ChShift_Ctx    sctx;
1245   PetscInt       newshift;
1246 
1247   PetscFunctionBegin;
1248   /* initialization */
1249   shiftnz   = info->shiftnz;
1250   shiftpd   = info->shiftpd;
1251   zeropivot = info->zeropivot;
1252 
1253   /* il and jl record the first nonzero element in each row of the accessing
1254      window U(0:k, k:mbs-1).
1255      jl:    list of rows to be added to uneliminated rows
1256             i>= k: jl(i) is the first row to be added to row i
1257             i<  k: jl(i) is the row following row i in some list of rows
1258             jl(i) = mbs indicates the end of a list
1259      il(i): points to the first nonzero element in U(i,k:mbs-1)
1260   */
1261   ierr = PetscMalloc(mbs*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
1262   ierr = PetscMalloc(2*mbs*sizeof(PetscInt),&il);CHKERRQ(ierr);
1263   jl   = il + mbs;
1264 
1265   sctx.shift_amount = 0;
1266   sctx.nshift       = 0;
1267   do {
1268     sctx.chshift = PETSC_FALSE;
1269     for (i=0; i<mbs; i++) {
1270       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
1271     }
1272 
1273     for (k = 0; k<mbs; k++){
1274       /*initialize k-th row with elements nonzero in row perm(k) of A */
1275       nz   = ai[k+1] - ai[k];
1276       acol = aj + ai[k];
1277       aval = aa + ai[k];
1278       bval = ba + bi[k];
1279       while (nz -- ){
1280         rtmp[*acol++] = *aval++;
1281         *bval++       = 0.0; /* for in-place factorization */
1282       }
1283 
1284       /* shift the diagonal of the matrix */
1285       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
1286 
1287       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
1288       dk = rtmp[k];
1289       i  = jl[k]; /* first row to be added to k_th row  */
1290 
1291       while (i < k){
1292         nexti = jl[i]; /* next row to be added to k_th row */
1293         /* compute multiplier, update D(k) and U(i,k) */
1294         ili   = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
1295         uikdi = - ba[ili]*ba[bi[i]];
1296         dk   += uikdi*ba[ili];
1297         ba[ili] = uikdi; /* -U(i,k) */
1298 
1299         /* add multiple of row i to k-th row ... */
1300         jmin = ili + 1;
1301         nz   = bi[i+1] - jmin;
1302         if (nz > 0){
1303           bcol = bj + jmin;
1304           bval = ba + jmin;
1305           while (nz --) rtmp[*bcol++] += uikdi*(*bval++);
1306           /* update il and jl for i-th row */
1307           il[i] = jmin;
1308           j = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
1309         }
1310         i = nexti;
1311       }
1312 
1313       /* shift the diagonals when zero pivot is detected */
1314       /* compute rs=sum of abs(off-diagonal) */
1315       rs   = 0.0;
1316       jmin = bi[k]+1;
1317       nz   = bi[k+1] - jmin;
1318       if (nz){
1319         bcol = bj + jmin;
1320         while (nz--){
1321           rs += PetscAbsScalar(rtmp[*bcol]);
1322           bcol++;
1323         }
1324       }
1325 
1326       sctx.rs = rs;
1327       sctx.pv = dk;
1328       ierr = MatCholeskyCheckShift_inline(info,sctx,newshift);CHKERRQ(ierr);
1329       if (newshift == 1){
1330         break;    /* sctx.shift_amount is updated */
1331       } else if (newshift == -1){
1332         SETERRQ4(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot row %D value %g tolerance %g * rs %g",k,PetscAbsScalar(dk),zeropivot,rs);
1333       }
1334 
1335       /* copy data into U(k,:) */
1336       ba[bi[k]] = 1.0/dk;
1337       jmin      = bi[k]+1;
1338       nz        = bi[k+1] - jmin;
1339       if (nz){
1340         bcol = bj + jmin;
1341         bval = ba + jmin;
1342         while (nz--){
1343           *bval++       = rtmp[*bcol];
1344           rtmp[*bcol++] = 0.0;
1345         }
1346         /* add k-th row into il and jl */
1347         il[k] = jmin;
1348         i = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
1349       }
1350     } /* end of for (k = 0; k<mbs; k++) */
1351   } while (sctx.chshift);
1352   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1353   ierr = PetscFree(il);CHKERRQ(ierr);
1354 
1355   C->factor       = FACTOR_CHOLESKY;
1356   C->assembled    = PETSC_TRUE;
1357   C->preallocated = PETSC_TRUE;
1358   ierr = PetscLogFlops(C->m);CHKERRQ(ierr);
1359   if (sctx.nshift){
1360     if (shiftnz) {
1361       ierr = PetscVerboseInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering: number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr);
1362     } else if (shiftpd) {
1363       ierr = PetscVerboseInfo((0,"MatCholeskyFactorNumeric_SeqSBAIJ_1_NaturalOrdering: number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,sctx.shift_amount));CHKERRQ(ierr);
1364     }
1365   }
1366   PetscFunctionReturn(0);
1367 }
1368 
1369 #undef __FUNCT__
1370 #define __FUNCT__ "MatCholeskyFactor_SeqSBAIJ"
1371 PetscErrorCode MatCholeskyFactor_SeqSBAIJ(Mat A,IS perm,MatFactorInfo *info)
1372 {
1373   PetscErrorCode ierr;
1374   Mat            C;
1375 
1376   PetscFunctionBegin;
1377   ierr = MatCholeskyFactorSymbolic(A,perm,info,&C);CHKERRQ(ierr);
1378   ierr = MatCholeskyFactorNumeric(A,info,&C);CHKERRQ(ierr);
1379   ierr = MatHeaderCopy(A,C);CHKERRQ(ierr);
1380   PetscFunctionReturn(0);
1381 }
1382 
1383 
1384