xref: /petsc/src/mat/impls/baij/seq/baijfact.c (revision b588c5a2f37da9e74d7d91cdaec80d3d0decb4d9)
1 #define PETSCMAT_DLL
2 
3 /*
4     Factorization code for BAIJ format.
5 */
6 #include "../src/mat/impls/baij/seq/baij.h"
7 #include "../src/mat/blockinvert.h"
8 
9 /* ------------------------------------------------------------*/
10 /*
11       Version for when blocks are 2 by 2
12 */
13 /* MatLUFactorNumeric_SeqBAIJ_2_newdatastruct -
14      copied from MatLUFactorNumeric_SeqBAIJ_N_newdatastruct() and re-implemented
15        Kernel_A_gets_A_times_B()
16        Kernel_A_gets_A_minus_B_times_C()
17        Kernel_A_gets_inverse_A()
18 */
19 #undef __FUNCT__
20 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2_newdatastruct"
21 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2_newdatastruct(Mat B,Mat A,const MatFactorInfo *info)
22 {
23   Mat            C=B;
24   Mat_SeqBAIJ    *a=(Mat_SeqBAIJ*)A->data,*b=(Mat_SeqBAIJ *)C->data;
25   IS             isrow = b->row,isicol = b->icol;
26   PetscErrorCode ierr;
27   const PetscInt *r,*ic,*ics;
28   PetscInt       i,j,k,n=a->mbs,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
29   PetscInt       *ajtmp,*bjtmp,nz,nzL,row,*bdiag=b->diag,*pj;
30   MatScalar      *rtmp,*pc,*mwork,*v,*pv,*aa=a->a;
31   PetscInt       bs=A->rmap->bs,bs2 = a->bs2,flg;
32   MatScalar      *x,m1,m2,m3,m4;
33   PetscReal      shift = info->shiftinblocks;
34 
35   PetscFunctionBegin;
36   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
37   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
38 
39   /* generate work space needed by LU factorization */
40   ierr = PetscMalloc((bs2*n+bs2+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
41   mwork = rtmp + bs2*n;
42   ierr = PetscMemzero(rtmp,(bs2*n+1)*sizeof(MatScalar));CHKERRQ(ierr);
43   ics  = ic;
44 
45   for (i=0; i<n; i++){
46     /* zero rtmp */
47     /* L part */
48     nz    = bi[i+1] - bi[i];
49     bjtmp = bj + bi[i];
50     for  (j=0; j<nz; j++){
51       ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
52     }
53 
54     /* U part */
55     nz = bi[2*n-i+1] - bi[2*n-i];
56     bjtmp = bj + bi[2*n-i];
57     for  (j=0; j<nz; j++){
58       ierr = PetscMemzero(rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
59     }
60 
61     /* load in initial (unfactored row) */
62     nz    = ai[r[i]+1] - ai[r[i]];
63     ajtmp = aj + ai[r[i]];
64     v     = aa + bs2*ai[r[i]];
65     for (j=0; j<nz; j++) {
66       ierr = PetscMemcpy(rtmp+bs2*ic[ajtmp[j]],v+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr);
67     }
68 
69     /* elimination */
70     bjtmp = bj + bi[i];
71     row   = *bjtmp++;
72     nzL   = bi[i+1] - bi[i];
73     k   = 0;
74     while  (k < nzL) {
75       pc = rtmp + bs2*row;
76       for (flg=0,j=0; j<bs2; j++) { if (pc[j]!=0.0) { flg = 1; break; }}
77       if (flg) {
78         pv         = b->a + bs2*bdiag[row];
79         /* Kernel_A_gets_A_times_B(bs,pc,pv,mwork); */
80         /* ---Kernel_A_gets_A_times_B_2(): *pc = *pc * (*pv); ------- */
81         ierr = PetscMemcpy(mwork,pc,bs2*sizeof(MatScalar));CHKERRQ(ierr);
82         pc[0] = m1 = mwork[0]*pv[0] + mwork[2]*pv[1];
83         pc[1] = m2 = mwork[1]*pv[0] + mwork[3]*pv[1];
84         pc[2] = m3 = mwork[0]*pv[2] + mwork[2]*pv[3];
85         pc[3] = m4 = mwork[1]*pv[2] + mwork[3]*pv[3];
86         /* ------------ endof Kernel_A_gets_A_times_B_2() ----------- */
87 
88         pj         = b->j + bi[2*n-row]; /* begining of U(row,:) */
89         pv         = b->a + bs2*bi[2*n-row];
90         nz         = bi[2*n-row+1] - bi[2*n-row] - 1; /* num of entries inU(row,:), excluding diag */
91         for (j=0; j<nz; j++) {
92           /* Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j); */
93           /* -- Kernel_A_gets_A_minus_B_times_C_2():  rtmp+bs2*pj[j] = rtmp+bs2*pj[j] - (*pc)*(pv+bs2*j) */
94           x    = rtmp + 4*pj[j];
95           x[0] -= m1*pv[0] + m3*pv[1];
96           x[1] -= m2*pv[0] + m4*pv[1];
97           x[2] -= m1*pv[2] + m3*pv[3];
98           x[3] -= m2*pv[2] + m4*pv[3];
99           pv   += 4;
100           /*------------ endof Kernel_A_gets_A_minus_B_times_C_2() -- */
101         }
102         ierr = PetscLogFlops(16*nz+12);CHKERRQ(ierr); /* flops = 2*bs^3*nz + 2*bs^3 - bs2) */
103       }
104       row = *bjtmp++; k++;
105     }
106 
107     /* finished row so stick it into b->a */
108     /* L part */
109     pv   = b->a + bs2*bi[i] ;
110     pj   = b->j + bi[i] ;
111     nz   = bi[i+1] - bi[i];
112     for (j=0; j<nz; j++) {
113       ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
114     }
115 
116     /* Mark diagonal and invert diagonal for simplier triangular solves */
117     pv  = b->a + bs2*bdiag[i];
118     pj  = b->j + bdiag[i];
119     ierr = PetscMemcpy(pv,rtmp+bs2*pj[0],bs2*sizeof(MatScalar));CHKERRQ(ierr);
120     // ierr = Kernel_A_gets_inverse_A(bs,pv,v_pivots,v_work);CHKERRQ(ierr);
121     ierr = Kernel_A_gets_inverse_A_2(pv,shift);CHKERRQ(ierr);
122 
123     /* U part */
124     pv = b->a + bs2*bi[2*n-i];
125     pj = b->j + bi[2*n-i];
126     nz = bi[2*n-i+1] - bi[2*n-i] - 1;
127     for (j=0; j<nz; j++){
128       ierr = PetscMemcpy(pv+bs2*j,rtmp+bs2*pj[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
129     }
130   }
131 
132   ierr = PetscFree(rtmp);CHKERRQ(ierr);
133   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
134   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
135 
136   C->assembled = PETSC_TRUE;
137   ierr = PetscLogFlops(1.3333*bs*bs2*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
138   PetscFunctionReturn(0);
139 }
140 
141 #undef __FUNCT__
142 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2"
143 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2(Mat B,Mat A,const MatFactorInfo *info)
144 {
145   Mat            C = B;
146   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data;
147   IS             isrow = b->row,isicol = b->icol;
148   PetscErrorCode ierr;
149   const PetscInt *r,*ic;
150   PetscInt       i,j,n = a->mbs,*bi = b->i,*bj = b->j;
151   PetscInt       *ajtmpold,*ajtmp,nz,row;
152   PetscInt       *diag_offset=b->diag,idx,*ai=a->i,*aj=a->j,*pj;
153   MatScalar      *pv,*v,*rtmp,m1,m2,m3,m4,*pc,*w,*x,x1,x2,x3,x4;
154   MatScalar      p1,p2,p3,p4;
155   MatScalar      *ba = b->a,*aa = a->a;
156   PetscReal      shift = info->shiftinblocks;
157 
158   PetscFunctionBegin;
159   ierr  = ISGetIndices(isrow,&r);CHKERRQ(ierr);
160   ierr  = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
161   ierr  = PetscMalloc(4*(n+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
162 
163   for (i=0; i<n; i++) {
164     nz    = bi[i+1] - bi[i];
165     ajtmp = bj + bi[i];
166     for  (j=0; j<nz; j++) {
167       x = rtmp+4*ajtmp[j]; x[0] = x[1] = x[2] = x[3] = 0.0;
168     }
169     /* load in initial (unfactored row) */
170     idx      = r[i];
171     nz       = ai[idx+1] - ai[idx];
172     ajtmpold = aj + ai[idx];
173     v        = aa + 4*ai[idx];
174     for (j=0; j<nz; j++) {
175       x    = rtmp+4*ic[ajtmpold[j]];
176       x[0] = v[0]; x[1] = v[1]; x[2] = v[2]; x[3] = v[3];
177       v    += 4;
178     }
179     row = *ajtmp++;
180     while (row < i) {
181       pc = rtmp + 4*row;
182       p1 = pc[0]; p2 = pc[1]; p3 = pc[2]; p4 = pc[3];
183       if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) {
184         pv = ba + 4*diag_offset[row];
185         pj = bj + diag_offset[row] + 1;
186         x1 = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3];
187         pc[0] = m1 = p1*x1 + p3*x2;
188         pc[1] = m2 = p2*x1 + p4*x2;
189         pc[2] = m3 = p1*x3 + p3*x4;
190         pc[3] = m4 = p2*x3 + p4*x4;
191         nz = bi[row+1] - diag_offset[row] - 1;
192         pv += 4;
193         for (j=0; j<nz; j++) {
194           x1   = pv[0]; x2 = pv[1]; x3 = pv[2]; x4 = pv[3];
195           x    = rtmp + 4*pj[j];
196           x[0] -= m1*x1 + m3*x2;
197           x[1] -= m2*x1 + m4*x2;
198           x[2] -= m1*x3 + m3*x4;
199           x[3] -= m2*x3 + m4*x4;
200           pv   += 4;
201         }
202         ierr = PetscLogFlops(16.0*nz+12.0);CHKERRQ(ierr);
203       }
204       row = *ajtmp++;
205     }
206     /* finished row so stick it into b->a */
207     pv = ba + 4*bi[i];
208     pj = bj + bi[i];
209     nz = bi[i+1] - bi[i];
210     for (j=0; j<nz; j++) {
211       x     = rtmp+4*pj[j];
212       pv[0] = x[0]; pv[1] = x[1]; pv[2] = x[2]; pv[3] = x[3];
213       pv   += 4;
214     }
215     /* invert diagonal block */
216     w = ba + 4*diag_offset[i];
217     ierr = Kernel_A_gets_inverse_A_2(w,shift);CHKERRQ(ierr);
218   }
219 
220   ierr = PetscFree(rtmp);CHKERRQ(ierr);
221   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
222   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
223   C->ops->solve          = MatSolve_SeqBAIJ_2;
224   C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_2;
225   C->assembled = PETSC_TRUE;
226   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
227   PetscFunctionReturn(0);
228 }
229 /*
230       Version for when blocks are 2 by 2 Using natural ordering
231 */
232 #undef __FUNCT__
233 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering"
234 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_2_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info)
235 {
236   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data;
237   PetscErrorCode ierr;
238   PetscInt       i,j,n = a->mbs,*bi = b->i,*bj = b->j;
239   PetscInt       *ajtmpold,*ajtmp,nz,row;
240   PetscInt       *diag_offset = b->diag,*ai=a->i,*aj=a->j,*pj;
241   MatScalar      *pv,*v,*rtmp,*pc,*w,*x;
242   MatScalar      p1,p2,p3,p4,m1,m2,m3,m4,x1,x2,x3,x4;
243   MatScalar      *ba = b->a,*aa = a->a;
244   PetscReal      shift = info->shiftinblocks;
245 
246   PetscFunctionBegin;
247   ierr = PetscMalloc(4*(n+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
248   for (i=0; i<n; i++) {
249     nz    = bi[i+1] - bi[i];
250     ajtmp = bj + bi[i];
251     for  (j=0; j<nz; j++) {
252       x = rtmp+4*ajtmp[j];
253       x[0]  = x[1]  = x[2]  = x[3]  = 0.0;
254     }
255     /* load in initial (unfactored row) */
256     nz       = ai[i+1] - ai[i];
257     ajtmpold = aj + ai[i];
258     v        = aa + 4*ai[i];
259     for (j=0; j<nz; j++) {
260       x    = rtmp+4*ajtmpold[j];
261       x[0]  = v[0];  x[1]  = v[1];  x[2]  = v[2];  x[3]  = v[3];
262       v    += 4;
263     }
264     row = *ajtmp++;
265     while (row < i) {
266       pc  = rtmp + 4*row;
267       p1  = pc[0];  p2  = pc[1];  p3  = pc[2];  p4  = pc[3];
268       if (p1 != 0.0 || p2 != 0.0 || p3 != 0.0 || p4 != 0.0) {
269         pv = ba + 4*diag_offset[row];
270         pj = bj + diag_offset[row] + 1;
271         x1  = pv[0];  x2  = pv[1];  x3  = pv[2];  x4  = pv[3];
272         pc[0] = m1 = p1*x1 + p3*x2;
273         pc[1] = m2 = p2*x1 + p4*x2;
274         pc[2] = m3 = p1*x3 + p3*x4;
275         pc[3] = m4 = p2*x3 + p4*x4;
276         nz = bi[row+1] - diag_offset[row] - 1;
277         pv += 4;
278         for (j=0; j<nz; j++) {
279           x1   = pv[0];  x2  = pv[1];   x3 = pv[2];  x4  = pv[3];
280           x    = rtmp + 4*pj[j];
281           x[0] -= m1*x1 + m3*x2;
282           x[1] -= m2*x1 + m4*x2;
283           x[2] -= m1*x3 + m3*x4;
284           x[3] -= m2*x3 + m4*x4;
285           pv   += 4;
286         }
287         ierr = PetscLogFlops(16.0*nz+12.0);CHKERRQ(ierr);
288       }
289       row = *ajtmp++;
290     }
291     /* finished row so stick it into b->a */
292     pv = ba + 4*bi[i];
293     pj = bj + bi[i];
294     nz = bi[i+1] - bi[i];
295     for (j=0; j<nz; j++) {
296       x      = rtmp+4*pj[j];
297       pv[0]  = x[0];  pv[1]  = x[1];  pv[2]  = x[2];  pv[3]  = x[3];
298       /*
299       printf(" col %d:",pj[j]);
300       PetscInt j1;
301       for (j1=0; j1<4; j1++) printf(" %g,",*(pv+j1));
302       printf("\n");
303       */
304       pv   += 4;
305     }
306     /* invert diagonal block */
307     w = ba + 4*diag_offset[i];
308     /*
309     printf(" \n%d -th: diag: ",i);
310     for (j=0; j<4; j++){
311       printf(" %g,",w[j]);
312     }
313     printf("\n----------------------------\n");
314     */
315     ierr = Kernel_A_gets_inverse_A_2(w,shift);CHKERRQ(ierr);
316   }
317 
318   ierr = PetscFree(rtmp);CHKERRQ(ierr);
319   C->ops->solve          = MatSolve_SeqBAIJ_2_NaturalOrdering;
320   C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_2_NaturalOrdering;
321   C->assembled = PETSC_TRUE;
322   ierr = PetscLogFlops(1.3333*8*b->mbs);CHKERRQ(ierr); /* from inverting diagonal blocks */
323   PetscFunctionReturn(0);
324 }
325 
326 /* ----------------------------------------------------------- */
327 /*
328      Version for when blocks are 1 by 1.
329 */
330 #undef __FUNCT__
331 #define __FUNCT__ "MatLUFactorNumeric_SeqBAIJ_1"
332 PetscErrorCode MatLUFactorNumeric_SeqBAIJ_1(Mat C,Mat A,const MatFactorInfo *info)
333 {
334   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data,*b = (Mat_SeqBAIJ *)C->data;
335   IS             isrow = b->row,isicol = b->icol;
336   PetscErrorCode ierr;
337   const PetscInt *r,*ic;
338   PetscInt       i,j,n = a->mbs,*bi = b->i,*bj = b->j;
339   PetscInt       *ajtmpold,*ajtmp,nz,row,*ai = a->i,*aj = a->j;
340   PetscInt       *diag_offset = b->diag,diag,*pj;
341   MatScalar      *pv,*v,*rtmp,multiplier,*pc;
342   MatScalar      *ba = b->a,*aa = a->a;
343   PetscTruth     row_identity, col_identity;
344 
345   PetscFunctionBegin;
346   ierr  = ISGetIndices(isrow,&r);CHKERRQ(ierr);
347   ierr  = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
348   ierr  = PetscMalloc((n+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
349 
350   for (i=0; i<n; i++) {
351     nz    = bi[i+1] - bi[i];
352     ajtmp = bj + bi[i];
353     for  (j=0; j<nz; j++) rtmp[ajtmp[j]] = 0.0;
354 
355     /* load in initial (unfactored row) */
356     nz       = ai[r[i]+1] - ai[r[i]];
357     ajtmpold = aj + ai[r[i]];
358     v        = aa + ai[r[i]];
359     for (j=0; j<nz; j++) rtmp[ic[ajtmpold[j]]] =  v[j];
360 
361     row = *ajtmp++;
362     while (row < i) {
363       pc = rtmp + row;
364       if (*pc != 0.0) {
365         pv         = ba + diag_offset[row];
366         pj         = bj + diag_offset[row] + 1;
367         multiplier = *pc * *pv++;
368         *pc        = multiplier;
369         nz         = bi[row+1] - diag_offset[row] - 1;
370         for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j];
371         ierr = PetscLogFlops(1.0+2.0*nz);CHKERRQ(ierr);
372       }
373       row = *ajtmp++;
374     }
375     /* finished row so stick it into b->a */
376     pv = ba + bi[i];
377     pj = bj + bi[i];
378     nz = bi[i+1] - bi[i];
379     for (j=0; j<nz; j++) {pv[j] = rtmp[pj[j]];}
380     diag = diag_offset[i] - bi[i];
381     /* check pivot entry for current row */
382     if (pv[diag] == 0.0) {
383       SETERRQ2(PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot: row in original ordering %D in permuted ordering %D",r[i],i);
384     }
385     pv[diag] = 1.0/pv[diag];
386   }
387 
388   ierr = PetscFree(rtmp);CHKERRQ(ierr);
389   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
390   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
391   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
392   ierr = ISIdentity(isicol,&col_identity);CHKERRQ(ierr);
393   if (row_identity && col_identity) {
394     C->ops->solve          = MatSolve_SeqBAIJ_1_NaturalOrdering;
395     C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_1_NaturalOrdering;
396   } else {
397     C->ops->solve          = MatSolve_SeqBAIJ_1;
398     C->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_1;
399   }
400   C->assembled = PETSC_TRUE;
401   ierr = PetscLogFlops(C->cmap->n);CHKERRQ(ierr);
402   PetscFunctionReturn(0);
403 }
404 
405 EXTERN_C_BEGIN
406 #undef __FUNCT__
407 #define __FUNCT__ "MatGetFactor_seqbaij_petsc"
408 PetscErrorCode MatGetFactor_seqbaij_petsc(Mat A,MatFactorType ftype,Mat *B)
409 {
410   PetscInt           n = A->rmap->n;
411   PetscErrorCode     ierr;
412 
413   PetscFunctionBegin;
414   ierr = MatCreate(((PetscObject)A)->comm,B);CHKERRQ(ierr);
415   ierr = MatSetSizes(*B,n,n,n,n);CHKERRQ(ierr);
416   if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU || ftype == MAT_FACTOR_ILUDT) {
417     ierr = MatSetType(*B,MATSEQBAIJ);CHKERRQ(ierr);
418     (*B)->ops->lufactorsymbolic  = MatLUFactorSymbolic_SeqBAIJ;
419     (*B)->ops->ilufactorsymbolic = MatILUFactorSymbolic_SeqBAIJ;
420     (*B)->ops->iludtfactor       = MatILUDTFactor_SeqBAIJ;
421   } else if (ftype == MAT_FACTOR_CHOLESKY || ftype == MAT_FACTOR_ICC) {
422     ierr = MatSetType(*B,MATSEQSBAIJ);CHKERRQ(ierr);
423     ierr = MatSeqSBAIJSetPreallocation(*B,1,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
424     (*B)->ops->iccfactorsymbolic      = MatICCFactorSymbolic_SeqBAIJ;
425     (*B)->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SeqBAIJ;
426   } else SETERRQ(PETSC_ERR_SUP,"Factor type not supported");
427   (*B)->factor = ftype;
428   PetscFunctionReturn(0);
429 }
430 EXTERN_C_END
431 
432 EXTERN_C_BEGIN
433 #undef __FUNCT__
434 #define __FUNCT__ "MatGetFactorAvailable_seqaij_petsc"
435 PetscErrorCode MatGetFactorAvailable_seqbaij_petsc(Mat A,MatFactorType ftype,PetscTruth *flg)
436 {
437   PetscFunctionBegin;
438   *flg = PETSC_TRUE;
439   PetscFunctionReturn(0);
440 }
441 EXTERN_C_END
442 
443 /* ----------------------------------------------------------- */
444 #undef __FUNCT__
445 #define __FUNCT__ "MatLUFactor_SeqBAIJ"
446 PetscErrorCode MatLUFactor_SeqBAIJ(Mat A,IS row,IS col,const MatFactorInfo *info)
447 {
448   PetscErrorCode ierr;
449   Mat            C;
450 
451   PetscFunctionBegin;
452   ierr = MatGetFactor(A,MAT_SOLVER_PETSC,MAT_FACTOR_LU,&C);CHKERRQ(ierr);
453   ierr = MatLUFactorSymbolic(C,A,row,col,info);CHKERRQ(ierr);
454   ierr = MatLUFactorNumeric(C,A,info);CHKERRQ(ierr);
455   A->ops->solve            = C->ops->solve;
456   A->ops->solvetranspose   = C->ops->solvetranspose;
457   ierr = MatHeaderCopy(A,C);CHKERRQ(ierr);
458   ierr = PetscLogObjectParent(A,((Mat_SeqBAIJ*)(A->data))->icol);CHKERRQ(ierr);
459   PetscFunctionReturn(0);
460 }
461 
462 #include "../src/mat/impls/sbaij/seq/sbaij.h"
463 #undef __FUNCT__
464 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqBAIJ_N"
465 PetscErrorCode MatCholeskyFactorNumeric_SeqBAIJ_N(Mat C,Mat A,const MatFactorInfo *info)
466 {
467   PetscErrorCode ierr;
468   Mat_SeqBAIJ    *a=(Mat_SeqBAIJ*)A->data;
469   Mat_SeqSBAIJ   *b=(Mat_SeqSBAIJ*)C->data;
470   IS             ip=b->row;
471   const PetscInt *rip;
472   PetscInt       i,j,mbs=a->mbs,bs=A->rmap->bs,*bi=b->i,*bj=b->j,*bcol;
473   PetscInt       *ai=a->i,*aj=a->j;
474   PetscInt       k,jmin,jmax,*jl,*il,col,nexti,ili,nz;
475   MatScalar      *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi;
476   PetscReal      zeropivot,rs,shiftnz;
477   PetscReal      shiftpd;
478   ChShift_Ctx    sctx;
479   PetscInt       newshift;
480 
481   PetscFunctionBegin;
482   if (bs > 1) {
483     if (!a->sbaijMat){
484       ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&a->sbaijMat);CHKERRQ(ierr);
485     }
486     ierr = (a->sbaijMat)->ops->choleskyfactornumeric(C,a->sbaijMat,info);CHKERRQ(ierr);
487     ierr = MatDestroy(a->sbaijMat);CHKERRQ(ierr);
488     a->sbaijMat = PETSC_NULL;
489     PetscFunctionReturn(0);
490   }
491 
492   /* initialization */
493   shiftnz   = info->shiftnz;
494   shiftpd   = info->shiftpd;
495   zeropivot = info->zeropivot;
496 
497   ierr  = ISGetIndices(ip,&rip);CHKERRQ(ierr);
498   nz   = (2*mbs+1)*sizeof(PetscInt)+mbs*sizeof(MatScalar);
499   ierr = PetscMalloc(nz,&il);CHKERRQ(ierr);
500   jl   = il + mbs;
501   rtmp = (MatScalar*)(jl + mbs);
502 
503   sctx.shift_amount = 0;
504   sctx.nshift       = 0;
505   do {
506     sctx.chshift = PETSC_FALSE;
507     for (i=0; i<mbs; i++) {
508       rtmp[i] = 0.0; jl[i] = mbs; il[0] = 0;
509     }
510 
511     for (k = 0; k<mbs; k++){
512       bval = ba + bi[k];
513       /* initialize k-th row by the perm[k]-th row of A */
514       jmin = ai[rip[k]]; jmax = ai[rip[k]+1];
515       for (j = jmin; j < jmax; j++){
516         col = rip[aj[j]];
517         if (col >= k){ /* only take upper triangular entry */
518           rtmp[col] = aa[j];
519           *bval++  = 0.0; /* for in-place factorization */
520         }
521       }
522 
523       /* shift the diagonal of the matrix */
524       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
525 
526       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
527       dk = rtmp[k];
528       i = jl[k]; /* first row to be added to k_th row  */
529 
530       while (i < k){
531         nexti = jl[i]; /* next row to be added to k_th row */
532 
533         /* compute multiplier, update diag(k) and U(i,k) */
534         ili = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
535         uikdi = - ba[ili]*ba[bi[i]];  /* diagonal(k) */
536         dk += uikdi*ba[ili];
537         ba[ili] = uikdi; /* -U(i,k) */
538 
539         /* add multiple of row i to k-th row */
540         jmin = ili + 1; jmax = bi[i+1];
541         if (jmin < jmax){
542           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
543           /* update il and jl for row i */
544           il[i] = jmin;
545           j = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
546         }
547         i = nexti;
548       }
549 
550       /* shift the diagonals when zero pivot is detected */
551       /* compute rs=sum of abs(off-diagonal) */
552       rs   = 0.0;
553       jmin = bi[k]+1;
554       nz   = bi[k+1] - jmin;
555       if (nz){
556         bcol = bj + jmin;
557         while (nz--){
558           rs += PetscAbsScalar(rtmp[*bcol]);
559           bcol++;
560         }
561       }
562 
563       sctx.rs = rs;
564       sctx.pv = dk;
565       ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr);
566       if (newshift == 1) break;
567 
568       /* copy data into U(k,:) */
569       ba[bi[k]] = 1.0/dk; /* U(k,k) */
570       jmin = bi[k]+1; jmax = bi[k+1];
571       if (jmin < jmax) {
572         for (j=jmin; j<jmax; j++){
573           col = bj[j]; ba[j] = rtmp[col]; rtmp[col] = 0.0;
574         }
575         /* add the k-th row into il and jl */
576         il[k] = jmin;
577         i = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
578       }
579     }
580   } while (sctx.chshift);
581   ierr = PetscFree(il);CHKERRQ(ierr);
582 
583   ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr);
584   C->assembled    = PETSC_TRUE;
585   C->preallocated = PETSC_TRUE;
586   ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr);
587   if (sctx.nshift){
588     if (shiftpd) {
589       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
590     } else if (shiftnz) {
591       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
592     }
593   }
594   PetscFunctionReturn(0);
595 }
596 
597 #undef __FUNCT__
598 #define __FUNCT__ "MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering"
599 PetscErrorCode MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering(Mat C,Mat A,const MatFactorInfo *info)
600 {
601   Mat_SeqBAIJ    *a=(Mat_SeqBAIJ*)A->data;
602   Mat_SeqSBAIJ   *b=(Mat_SeqSBAIJ*)C->data;
603   PetscErrorCode ierr;
604   PetscInt       i,j,am=a->mbs;
605   PetscInt       *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
606   PetscInt       k,jmin,*jl,*il,nexti,ili,*acol,*bcol,nz;
607   MatScalar      *rtmp,*ba=b->a,*aa=a->a,dk,uikdi,*aval,*bval;
608   PetscReal      zeropivot,rs,shiftnz;
609   PetscReal      shiftpd;
610   ChShift_Ctx    sctx;
611   PetscInt       newshift;
612 
613   PetscFunctionBegin;
614   /* initialization */
615   shiftnz   = info->shiftnz;
616   shiftpd   = info->shiftpd;
617   zeropivot = info->zeropivot;
618 
619   nz   = (2*am+1)*sizeof(PetscInt)+am*sizeof(MatScalar);
620   ierr = PetscMalloc(nz,&il);CHKERRQ(ierr);
621   jl   = il + am;
622   rtmp = (MatScalar*)(jl + am);
623 
624   sctx.shift_amount = 0;
625   sctx.nshift       = 0;
626   do {
627     sctx.chshift = PETSC_FALSE;
628     for (i=0; i<am; i++) {
629       rtmp[i] = 0.0; jl[i] = am; il[0] = 0;
630     }
631 
632     for (k = 0; k<am; k++){
633     /* initialize k-th row with elements nonzero in row perm(k) of A */
634       nz   = ai[k+1] - ai[k];
635       acol = aj + ai[k];
636       aval = aa + ai[k];
637       bval = ba + bi[k];
638       while (nz -- ){
639         if (*acol < k) { /* skip lower triangular entries */
640           acol++; aval++;
641         } else {
642           rtmp[*acol++] = *aval++;
643           *bval++       = 0.0; /* for in-place factorization */
644         }
645       }
646 
647       /* shift the diagonal of the matrix */
648       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
649 
650       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
651       dk = rtmp[k];
652       i  = jl[k]; /* first row to be added to k_th row  */
653 
654       while (i < k){
655         nexti = jl[i]; /* next row to be added to k_th row */
656         /* compute multiplier, update D(k) and U(i,k) */
657         ili   = il[i];  /* index of first nonzero element in U(i,k:bms-1) */
658         uikdi = - ba[ili]*ba[bi[i]];
659         dk   += uikdi*ba[ili];
660         ba[ili] = uikdi; /* -U(i,k) */
661 
662         /* add multiple of row i to k-th row ... */
663         jmin = ili + 1;
664         nz   = bi[i+1] - jmin;
665         if (nz > 0){
666           bcol = bj + jmin;
667           bval = ba + jmin;
668           while (nz --) rtmp[*bcol++] += uikdi*(*bval++);
669           /* update il and jl for i-th row */
670           il[i] = jmin;
671           j = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
672         }
673         i = nexti;
674       }
675 
676       /* shift the diagonals when zero pivot is detected */
677       /* compute rs=sum of abs(off-diagonal) */
678       rs   = 0.0;
679       jmin = bi[k]+1;
680       nz   = bi[k+1] - jmin;
681       if (nz){
682         bcol = bj + jmin;
683         while (nz--){
684           rs += PetscAbsScalar(rtmp[*bcol]);
685           bcol++;
686         }
687       }
688 
689       sctx.rs = rs;
690       sctx.pv = dk;
691       ierr = MatCholeskyCheckShift_inline(info,sctx,k,newshift);CHKERRQ(ierr);
692       if (newshift == 1) break;    /* sctx.shift_amount is updated */
693 
694       /* copy data into U(k,:) */
695       ba[bi[k]] = 1.0/dk;
696       jmin      = bi[k]+1;
697       nz        = bi[k+1] - jmin;
698       if (nz){
699         bcol = bj + jmin;
700         bval = ba + jmin;
701         while (nz--){
702           *bval++       = rtmp[*bcol];
703           rtmp[*bcol++] = 0.0;
704         }
705         /* add k-th row into il and jl */
706         il[k] = jmin;
707         i = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
708       }
709     }
710   } while (sctx.chshift);
711   ierr = PetscFree(il);CHKERRQ(ierr);
712 
713   C->ops->solve                 = MatSolve_SeqSBAIJ_1_NaturalOrdering;
714   C->ops->solvetranspose        = MatSolve_SeqSBAIJ_1_NaturalOrdering;
715   C->assembled    = PETSC_TRUE;
716   C->preallocated = PETSC_TRUE;
717   ierr = PetscLogFlops(C->rmap->N);CHKERRQ(ierr);
718     if (sctx.nshift){
719     if (shiftnz) {
720       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
721     } else if (shiftpd) {
722       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %G\n",sctx.nshift,sctx.shift_amount);CHKERRQ(ierr);
723     }
724   }
725   PetscFunctionReturn(0);
726 }
727 
728 #include "petscbt.h"
729 #include "../src/mat/utils/freespace.h"
730 #undef __FUNCT__
731 #define __FUNCT__ "MatICCFactorSymbolic_SeqBAIJ"
732 PetscErrorCode MatICCFactorSymbolic_SeqBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
733 {
734   Mat_SeqBAIJ        *a = (Mat_SeqBAIJ*)A->data;
735   Mat_SeqSBAIJ       *b;
736   Mat                B;
737   PetscErrorCode     ierr;
738   PetscTruth         perm_identity;
739   PetscInt           reallocs=0,i,*ai=a->i,*aj=a->j,am=a->mbs,bs=A->rmap->bs,*ui;
740   const PetscInt     *rip;
741   PetscInt           jmin,jmax,nzk,k,j,*jl,prow,*il,nextprow;
742   PetscInt           nlnk,*lnk,*lnk_lvl=PETSC_NULL,ncols,ncols_upper,*cols,*cols_lvl,*uj,**uj_ptr,**uj_lvl_ptr;
743   PetscReal          fill=info->fill,levels=info->levels;
744   PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL;
745   PetscFreeSpaceList free_space_lvl=PETSC_NULL,current_space_lvl=PETSC_NULL;
746   PetscBT            lnkbt;
747 
748   PetscFunctionBegin;
749   if (bs > 1){
750     if (!a->sbaijMat){
751       ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&a->sbaijMat);CHKERRQ(ierr);
752     }
753     (fact)->ops->iccfactorsymbolic = MatICCFactorSymbolic_SeqSBAIJ;  /* undue the change made in MatGetFactor_seqbaij_petsc */
754     ierr = MatICCFactorSymbolic(fact,a->sbaijMat,perm,info);CHKERRQ(ierr);
755     PetscFunctionReturn(0);
756   }
757 
758   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
759   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
760 
761   /* special case that simply copies fill pattern */
762   if (!levels && perm_identity) {
763     ierr = MatMarkDiagonal_SeqBAIJ(A);CHKERRQ(ierr);
764     ierr = PetscMalloc((am+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr);
765     for (i=0; i<am; i++) {
766       ui[i] = ai[i+1] - a->diag[i]; /* ui: rowlengths - changes when !perm_identity */
767     }
768     B = fact;
769     ierr = MatSeqSBAIJSetPreallocation(B,1,0,ui);CHKERRQ(ierr);
770 
771 
772     b  = (Mat_SeqSBAIJ*)B->data;
773     uj = b->j;
774     for (i=0; i<am; i++) {
775       aj = a->j + a->diag[i];
776       for (j=0; j<ui[i]; j++){
777         *uj++ = *aj++;
778       }
779       b->ilen[i] = ui[i];
780     }
781     ierr = PetscFree(ui);CHKERRQ(ierr);
782     B->factor = MAT_FACTOR_NONE;
783     ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
784     ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
785     B->factor = MAT_FACTOR_ICC;
786 
787     B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering;
788     PetscFunctionReturn(0);
789   }
790 
791   /* initialization */
792   ierr  = PetscMalloc((am+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr);
793   ui[0] = 0;
794   ierr  = PetscMalloc((2*am+1)*sizeof(PetscInt),&cols_lvl);CHKERRQ(ierr);
795 
796   /* jl: linked list for storing indices of the pivot rows
797      il: il[i] points to the 1st nonzero entry of U(i,k:am-1) */
798   ierr = PetscMalloc((2*am+1)*sizeof(PetscInt)+2*am*sizeof(PetscInt*),&jl);CHKERRQ(ierr);
799   il         = jl + am;
800   uj_ptr     = (PetscInt**)(il + am);
801   uj_lvl_ptr = (PetscInt**)(uj_ptr + am);
802   for (i=0; i<am; i++){
803     jl[i] = am; il[i] = 0;
804   }
805 
806   /* create and initialize a linked list for storing column indices of the active row k */
807   nlnk = am + 1;
808   ierr = PetscIncompleteLLCreate(am,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
809 
810   /* initial FreeSpace size is fill*(ai[am]+1) */
811   ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+1)),&free_space);CHKERRQ(ierr);
812   current_space = free_space;
813   ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[am]+1)),&free_space_lvl);CHKERRQ(ierr);
814   current_space_lvl = free_space_lvl;
815 
816   for (k=0; k<am; k++){  /* for each active row k */
817     /* initialize lnk by the column indices of row rip[k] of A */
818     nzk   = 0;
819     ncols = ai[rip[k]+1] - ai[rip[k]];
820     ncols_upper = 0;
821     cols        = cols_lvl + am;
822     for (j=0; j<ncols; j++){
823       i = rip[*(aj + ai[rip[k]] + j)];
824       if (i >= k){ /* only take upper triangular entry */
825         cols[ncols_upper] = i;
826         cols_lvl[ncols_upper] = -1;  /* initialize level for nonzero entries */
827         ncols_upper++;
828       }
829     }
830     ierr = PetscIncompleteLLAdd(ncols_upper,cols,levels,cols_lvl,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
831     nzk += nlnk;
832 
833     /* update lnk by computing fill-in for each pivot row to be merged in */
834     prow = jl[k]; /* 1st pivot row */
835 
836     while (prow < k){
837       nextprow = jl[prow];
838 
839       /* merge prow into k-th row */
840       jmin = il[prow] + 1;  /* index of the 2nd nzero entry in U(prow,k:am-1) */
841       jmax = ui[prow+1];
842       ncols = jmax-jmin;
843       i     = jmin - ui[prow];
844       cols = uj_ptr[prow] + i; /* points to the 2nd nzero entry in U(prow,k:am-1) */
845       for (j=0; j<ncols; j++) cols_lvl[j] = *(uj_lvl_ptr[prow] + i + j);
846       ierr = PetscIncompleteLLAddSorted(ncols,cols,levels,cols_lvl,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
847       nzk += nlnk;
848 
849       /* update il and jl for prow */
850       if (jmin < jmax){
851         il[prow] = jmin;
852         j = *cols; jl[prow] = jl[j]; jl[j] = prow;
853       }
854       prow = nextprow;
855     }
856 
857     /* if free space is not available, make more free space */
858     if (current_space->local_remaining<nzk) {
859       i = am - k + 1; /* num of unfactored rows */
860       i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
861       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
862       ierr = PetscFreeSpaceGet(i,&current_space_lvl);CHKERRQ(ierr);
863       reallocs++;
864     }
865 
866     /* copy data into free_space and free_space_lvl, then initialize lnk */
867     ierr = PetscIncompleteLLClean(am,am,nzk,lnk,lnk_lvl,current_space->array,current_space_lvl->array,lnkbt);CHKERRQ(ierr);
868 
869     /* add the k-th row into il and jl */
870     if (nzk-1 > 0){
871       i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:am-1) */
872       jl[k] = jl[i]; jl[i] = k;
873       il[k] = ui[k] + 1;
874     }
875     uj_ptr[k]     = current_space->array;
876     uj_lvl_ptr[k] = current_space_lvl->array;
877 
878     current_space->array           += nzk;
879     current_space->local_used      += nzk;
880     current_space->local_remaining -= nzk;
881 
882     current_space_lvl->array           += nzk;
883     current_space_lvl->local_used      += nzk;
884     current_space_lvl->local_remaining -= nzk;
885 
886     ui[k+1] = ui[k] + nzk;
887   }
888 
889 #if defined(PETSC_USE_INFO)
890   if (ai[am] != 0) {
891     PetscReal af = ((PetscReal)(2*ui[am]-am))/((PetscReal)ai[am]);
892     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr);
893     ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr);
894     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr);
895   } else {
896     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
897   }
898 #endif
899 
900   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
901   ierr = PetscFree(jl);CHKERRQ(ierr);
902   ierr = PetscFree(cols_lvl);CHKERRQ(ierr);
903 
904   /* destroy list of free space and other temporary array(s) */
905   ierr = PetscMalloc((ui[am]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr);
906   ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
907   ierr = PetscIncompleteLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
908   ierr = PetscFreeSpaceDestroy(free_space_lvl);CHKERRQ(ierr);
909 
910   /* put together the new matrix in MATSEQSBAIJ format */
911   B = fact;
912   ierr = MatSeqSBAIJSetPreallocation(B,1,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
913 
914   b = (Mat_SeqSBAIJ*)B->data;
915   b->singlemalloc = PETSC_FALSE;
916   b->free_a       = PETSC_TRUE;
917   b->free_ij       = PETSC_TRUE;
918   ierr = PetscMalloc((ui[am]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr);
919   b->j    = uj;
920   b->i    = ui;
921   b->diag = 0;
922   b->ilen = 0;
923   b->imax = 0;
924   b->row  = perm;
925   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
926   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
927   b->icol = perm;
928   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
929   ierr    = PetscMalloc((am+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
930   ierr    = PetscLogObjectMemory(B,(ui[am]-am)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
931   b->maxnz = b->nz = ui[am];
932 
933   B->info.factor_mallocs    = reallocs;
934   B->info.fill_ratio_given  = fill;
935   if (ai[am] != 0) {
936     B->info.fill_ratio_needed = ((PetscReal)ui[am])/((PetscReal)ai[am]);
937   } else {
938     B->info.fill_ratio_needed = 0.0;
939   }
940   if (perm_identity){
941     B->ops->solve           = MatSolve_SeqSBAIJ_1_NaturalOrdering;
942     B->ops->solvetranspose  = MatSolve_SeqSBAIJ_1_NaturalOrdering;
943     B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering;
944   } else {
945     (fact)->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N;
946   }
947   PetscFunctionReturn(0);
948 }
949 
950 #undef __FUNCT__
951 #define __FUNCT__ "MatCholeskyFactorSymbolic_SeqBAIJ"
952 PetscErrorCode MatCholeskyFactorSymbolic_SeqBAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
953 {
954   Mat_SeqBAIJ        *a = (Mat_SeqBAIJ*)A->data;
955   Mat_SeqSBAIJ       *b;
956   Mat                B;
957   PetscErrorCode     ierr;
958   PetscTruth         perm_identity;
959   PetscReal          fill = info->fill;
960   const PetscInt     *rip;
961   PetscInt           i,mbs=a->mbs,bs=A->rmap->bs,*ai=a->i,*aj=a->j,reallocs=0,prow;
962   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
963   PetscInt           nlnk,*lnk,ncols,ncols_upper,*cols,*uj,**ui_ptr,*uj_ptr;
964   PetscFreeSpaceList free_space=PETSC_NULL,current_space=PETSC_NULL;
965   PetscBT            lnkbt;
966 
967   PetscFunctionBegin;
968   if (bs > 1) { /* convert to seqsbaij */
969     if (!a->sbaijMat){
970       ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&a->sbaijMat);CHKERRQ(ierr);
971     }
972     (fact)->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SeqSBAIJ; /* undue the change made in MatGetFactor_seqbaij_petsc */
973     ierr = MatCholeskyFactorSymbolic(fact,a->sbaijMat,perm,info);CHKERRQ(ierr);
974     PetscFunctionReturn(0);
975   }
976 
977   /* check whether perm is the identity mapping */
978   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
979   if (!perm_identity) SETERRQ(PETSC_ERR_SUP,"Matrix reordering is not supported");
980   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
981 
982   /* initialization */
983   ierr  = PetscMalloc((mbs+1)*sizeof(PetscInt),&ui);CHKERRQ(ierr);
984   ui[0] = 0;
985 
986   /* jl: linked list for storing indices of the pivot rows
987      il: il[i] points to the 1st nonzero entry of U(i,k:mbs-1) */
988   ierr = PetscMalloc((3*mbs+1)*sizeof(PetscInt)+mbs*sizeof(PetscInt*),&jl);CHKERRQ(ierr);
989   il     = jl + mbs;
990   cols   = il + mbs;
991   ui_ptr = (PetscInt**)(cols + mbs);
992   for (i=0; i<mbs; i++){
993     jl[i] = mbs; il[i] = 0;
994   }
995 
996   /* create and initialize a linked list for storing column indices of the active row k */
997   nlnk = mbs + 1;
998   ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
999 
1000   /* initial FreeSpace size is fill*(ai[mbs]+1) */
1001   ierr = PetscFreeSpaceGet((PetscInt)(fill*(ai[mbs]+1)),&free_space);CHKERRQ(ierr);
1002   current_space = free_space;
1003 
1004   for (k=0; k<mbs; k++){  /* for each active row k */
1005     /* initialize lnk by the column indices of row rip[k] of A */
1006     nzk   = 0;
1007     ncols = ai[rip[k]+1] - ai[rip[k]];
1008     ncols_upper = 0;
1009     for (j=0; j<ncols; j++){
1010       i = rip[*(aj + ai[rip[k]] + j)];
1011       if (i >= k){ /* only take upper triangular entry */
1012         cols[ncols_upper] = i;
1013         ncols_upper++;
1014       }
1015     }
1016     ierr = PetscLLAdd(ncols_upper,cols,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
1017     nzk += nlnk;
1018 
1019     /* update lnk by computing fill-in for each pivot row to be merged in */
1020     prow = jl[k]; /* 1st pivot row */
1021 
1022     while (prow < k){
1023       nextprow = jl[prow];
1024       /* merge prow into k-th row */
1025       jmin = il[prow] + 1;  /* index of the 2nd nzero entry in U(prow,k:mbs-1) */
1026       jmax = ui[prow+1];
1027       ncols = jmax-jmin;
1028       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:mbs-1) */
1029       ierr = PetscLLAddSorted(ncols,uj_ptr,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
1030       nzk += nlnk;
1031 
1032       /* update il and jl for prow */
1033       if (jmin < jmax){
1034         il[prow] = jmin;
1035         j = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow;
1036       }
1037       prow = nextprow;
1038     }
1039 
1040     /* if free space is not available, make more free space */
1041     if (current_space->local_remaining<nzk) {
1042       i = mbs - k + 1; /* num of unfactored rows */
1043       i = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
1044       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
1045       reallocs++;
1046     }
1047 
1048     /* copy data into free space, then initialize lnk */
1049     ierr = PetscLLClean(mbs,mbs,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
1050 
1051     /* add the k-th row into il and jl */
1052     if (nzk-1 > 0){
1053       i = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:mbs-1) */
1054       jl[k] = jl[i]; jl[i] = k;
1055       il[k] = ui[k] + 1;
1056     }
1057     ui_ptr[k] = current_space->array;
1058     current_space->array           += nzk;
1059     current_space->local_used      += nzk;
1060     current_space->local_remaining -= nzk;
1061 
1062     ui[k+1] = ui[k] + nzk;
1063   }
1064 
1065 #if defined(PETSC_USE_INFO)
1066   if (ai[mbs] != 0) {
1067     PetscReal af = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]);
1068     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %G needed %G\n",reallocs,fill,af);CHKERRQ(ierr);
1069     ierr = PetscInfo1(A,"Run with -pc_factor_fill %G or use \n",af);CHKERRQ(ierr);
1070     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%G) for best performance.\n",af);CHKERRQ(ierr);
1071   } else {
1072     ierr = PetscInfo(A,"Empty matrix.\n");CHKERRQ(ierr);
1073   }
1074 #endif
1075 
1076   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
1077   ierr = PetscFree(jl);CHKERRQ(ierr);
1078 
1079   /* destroy list of free space and other temporary array(s) */
1080   ierr = PetscMalloc((ui[mbs]+1)*sizeof(PetscInt),&uj);CHKERRQ(ierr);
1081   ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
1082   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
1083 
1084   /* put together the new matrix in MATSEQSBAIJ format */
1085   B    = fact;
1086   ierr = MatSeqSBAIJSetPreallocation(B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
1087 
1088   b = (Mat_SeqSBAIJ*)B->data;
1089   b->singlemalloc = PETSC_FALSE;
1090   b->free_a       = PETSC_TRUE;
1091   b->free_ij      = PETSC_TRUE;
1092   ierr = PetscMalloc((ui[mbs]+1)*sizeof(MatScalar),&b->a);CHKERRQ(ierr);
1093   b->j    = uj;
1094   b->i    = ui;
1095   b->diag = 0;
1096   b->ilen = 0;
1097   b->imax = 0;
1098   b->row  = perm;
1099   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
1100   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
1101   b->icol = perm;
1102   ierr    = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
1103   ierr    = PetscMalloc((mbs+1)*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
1104   ierr    = PetscLogObjectMemory(B,(ui[mbs]-mbs)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
1105   b->maxnz = b->nz = ui[mbs];
1106 
1107   B->info.factor_mallocs    = reallocs;
1108   B->info.fill_ratio_given  = fill;
1109   if (ai[mbs] != 0) {
1110     B->info.fill_ratio_needed = ((PetscReal)ui[mbs])/((PetscReal)ai[mbs]);
1111   } else {
1112     B->info.fill_ratio_needed = 0.0;
1113   }
1114   if (perm_identity){
1115     B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N_NaturalOrdering;
1116   } else {
1117     B->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqBAIJ_N;
1118   }
1119   PetscFunctionReturn(0);
1120 }
1121 
1122 /* --------------------------------------------------------- */
1123 #undef __FUNCT__
1124 #define __FUNCT__ "MatSolve_SeqBAIJ_N_NaturalOrdering_newdatastruct"
1125 PetscErrorCode MatSolve_SeqBAIJ_N_NaturalOrdering_newdatastruct(Mat A,Vec bb,Vec xx)
1126 {
1127   Mat_SeqBAIJ    *a=(Mat_SeqBAIJ *)A->data;
1128   PetscErrorCode ierr;
1129   const PetscInt *ai=a->i,*aj=a->j,*vi;
1130   PetscInt       i,k,n=a->mbs;
1131   PetscInt       nz,bs=A->rmap->bs,bs2=a->bs2,*adiag=a->diag;
1132   MatScalar      *aa=a->a,*v;
1133   PetscScalar    *x,*b,*s,*t,*ls;
1134 
1135   PetscFunctionBegin;
1136   /* printf("MatSolve_SeqBAIJ_NaturalOrdering_iludt..bs %d\n",bs); */
1137   ierr = VecGetArray(bb,&b);CHKERRQ(ierr);
1138   ierr = VecGetArray(xx,&x);CHKERRQ(ierr);
1139   t  = a->solve_work;
1140 
1141   /* forward solve the lower triangular */
1142   ierr = PetscMemcpy(t,b,bs*sizeof(PetscScalar));CHKERRQ(ierr); /* copy 1st block of b to t */
1143 
1144   for (i=1; i<n; i++) {
1145     v   = aa + bs2*ai[i];
1146     vi  = aj + ai[i];
1147     nz = ai[i+1] - ai[i];
1148     s = t + bs*i;
1149     ierr = PetscMemcpy(s,b+bs*i,bs*sizeof(PetscScalar));CHKERRQ(ierr); /* copy i_th block of b to t */
1150     for(k=0;k<nz;k++){
1151       Kernel_v_gets_v_minus_A_times_w(bs,s,v,t+bs*vi[k]);
1152       v += bs2;
1153     }
1154   }
1155 
1156   /* backward solve the upper triangular */
1157   ls = a->solve_work + A->cmap->n;
1158   for (i=n-1; i>=0; i--){
1159     v  = aa + bs2*ai[2*n-i];
1160     vi = aj + ai[2*n-i];
1161     nz = ai[2*n-i +1] - ai[2*n-i]-1;
1162     ierr = PetscMemcpy(ls,t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr);
1163     for(k=0;k<nz;k++){
1164       Kernel_v_gets_v_minus_A_times_w(bs,ls,v,t+bs*vi[k]);
1165       v += bs2;
1166     }
1167     Kernel_w_gets_A_times_v(bs,ls,aa+bs2*adiag[i],t+i*bs); /* *inv(diagonal[i]) */
1168     ierr = PetscMemcpy(x+i*bs,t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr);
1169   }
1170 
1171   ierr = VecRestoreArray(bb,&b);CHKERRQ(ierr);
1172   ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr);
1173   ierr = PetscLogFlops(2.0*(a->bs2)*(a->nz) - A->rmap->bs*A->cmap->n);CHKERRQ(ierr);
1174   PetscFunctionReturn(0);
1175 }
1176 
1177 #undef __FUNCT__
1178 #define __FUNCT__ "MatSolve_SeqBAIJ_N_newdatastruct"
1179 PetscErrorCode MatSolve_SeqBAIJ_N_newdatastruct(Mat A,Vec bb,Vec xx)
1180 {
1181   Mat_SeqBAIJ    *a=(Mat_SeqBAIJ *)A->data;
1182   IS             iscol=a->col,isrow=a->row;
1183   PetscErrorCode ierr;
1184   const PetscInt *r,*c,*rout,*cout,*ai=a->i,*aj=a->j,*vi;
1185   PetscInt       i,m,n=a->mbs;
1186   PetscInt       nz,bs=A->rmap->bs,bs2=a->bs2,k;
1187   MatScalar      *aa=a->a,*v;
1188   PetscScalar    *x,*b,*s,*t,*ls;
1189 
1190   PetscFunctionBegin;
1191   ierr = VecGetArray(bb,&b);CHKERRQ(ierr);
1192   ierr = VecGetArray(xx,&x);CHKERRQ(ierr);
1193   t  = a->solve_work;
1194 
1195   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1196   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1197 
1198   /* forward solve the lower triangular */
1199   ierr = PetscMemcpy(t,b+bs*r[0],bs*sizeof(PetscScalar));CHKERRQ(ierr);
1200   for (i=1; i<n; i++) {
1201     v   = aa + bs2*ai[i];
1202     vi  = aj + ai[i];
1203     nz = ai[i+1] - ai[i];
1204     s = t + bs*i;
1205     ierr = PetscMemcpy(s,b+bs*r[i],bs*sizeof(PetscScalar));CHKERRQ(ierr);
1206     for(m=0;m<nz;m++){
1207       Kernel_v_gets_v_minus_A_times_w(bs,s,v,t+bs*vi[m]);
1208       v += bs2;
1209     }
1210   }
1211 
1212   /* backward solve the upper triangular */
1213   ls = a->solve_work + A->cmap->n;
1214   for (i=n-1; i>=0; i--){
1215     k  = 2*n-i;
1216     v  = aa + bs2*ai[k];
1217     vi = aj + ai[k];
1218     nz = ai[k+1] - ai[k] - 1;
1219     ierr = PetscMemcpy(ls,t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr);
1220     for(m=0;m<nz;m++){
1221       Kernel_v_gets_v_minus_A_times_w(bs,ls,v,t+bs*vi[m]);
1222       v += bs2;
1223     }
1224     Kernel_w_gets_A_times_v(bs,ls,v,t+i*bs); /* *inv(diagonal[i]) */
1225     ierr = PetscMemcpy(x + bs*c[i],t+i*bs,bs*sizeof(PetscScalar));CHKERRQ(ierr);
1226   }
1227   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1228   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1229   ierr = VecRestoreArray(bb,&b);CHKERRQ(ierr);
1230   ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr);
1231   ierr = PetscLogFlops(2.0*(a->bs2)*(a->nz) - A->rmap->bs*A->cmap->n);CHKERRQ(ierr);
1232   PetscFunctionReturn(0);
1233 }
1234 
1235 #undef __FUNCT__
1236 #define __FUNCT__ "BlockAbs_privat"
1237 PetscErrorCode BlockAbs_private(PetscInt nbs,PetscInt bs2,PetscScalar *blockarray,PetscReal *absarray)
1238 {
1239   PetscErrorCode     ierr;
1240   PetscInt           i,j;
1241   PetscFunctionBegin;
1242   ierr = PetscMemzero(absarray,(nbs+1)*sizeof(PetscReal));CHKERRQ(ierr);
1243   for (i=0; i<nbs; i++){
1244     for (j=0; j<bs2; j++){
1245       if (absarray[i] < PetscAbsScalar(blockarray[i*nbs+j])) absarray[i] = PetscAbsScalar(blockarray[i*nbs+j]);
1246     }
1247   }
1248   PetscFunctionReturn(0);
1249 }
1250 
1251 #undef __FUNCT__
1252 #define __FUNCT__ "MatILUDTFactor_SeqBAIJ"
1253 PetscErrorCode MatILUDTFactor_SeqBAIJ(Mat A,IS isrow,IS iscol,const MatFactorInfo *info,Mat *fact)
1254 {
1255   Mat                B = *fact;
1256   Mat_SeqBAIJ        *a=(Mat_SeqBAIJ*)A->data,*b;
1257   IS                 isicol;
1258   PetscErrorCode     ierr;
1259   const PetscInt     *r,*ic;
1260   PetscInt           i,mbs=a->mbs,bs=A->rmap->bs,bs2=a->bs2,*ai=a->i,*aj=a->j,*ajtmp,*adiag;
1261   PetscInt           *bi,*bj,*bdiag;
1262 
1263   PetscInt           row,nzi,nzi_bl,nzi_bu,*im,dtcount,nzi_al,nzi_au;
1264   PetscInt           nlnk,*lnk;
1265   PetscBT            lnkbt;
1266   PetscTruth         row_identity,icol_identity,both_identity;
1267   MatScalar          *aatmp,*pv,*batmp,*ba,*rtmp,*pc,*multiplier,*vtmp;
1268   const PetscInt     *ics;
1269   PetscInt           j,nz,*pj,*bjtmp,k,ncut,*jtmp;
1270 
1271   PetscReal          dt=info->dt; /* shift=info->shiftinblocks; */
1272   PetscInt           nnz_max;
1273   PetscTruth         missing;
1274   PetscReal          *vtmp_abs;
1275   MatScalar          *v_work;
1276   PetscInt           *v_pivots;
1277 
1278   PetscFunctionBegin;
1279   /* ------- symbolic factorization, can be reused ---------*/
1280   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
1281   if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
1282   adiag=a->diag;
1283 
1284   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
1285 
1286   /* bdiag is location of diagonal in factor */
1287   ierr = PetscMalloc((mbs+1)*sizeof(PetscInt),&bdiag);CHKERRQ(ierr);
1288 
1289   /* allocate row pointers bi */
1290   ierr = PetscMalloc((2*mbs+2)*sizeof(PetscInt),&bi);CHKERRQ(ierr);
1291 
1292   /* allocate bj and ba; max num of nonzero entries is (ai[n]+2*n*dtcount+2) */
1293   dtcount = (PetscInt)info->dtcount;
1294   if (dtcount > mbs-1) dtcount = mbs-1;
1295   nnz_max  = ai[mbs]+2*mbs*dtcount +2;
1296   /* printf("MatILUDTFactor_SeqBAIJ, bs %d, ai[mbs] %d, nnz_max  %d, dtcount %d\n",bs,ai[mbs],nnz_max,dtcount); */
1297   ierr = PetscMalloc(nnz_max*sizeof(PetscInt),&bj);CHKERRQ(ierr);
1298   nnz_max = nnz_max*bs2;
1299   ierr = PetscMalloc(nnz_max*sizeof(MatScalar),&ba);CHKERRQ(ierr);
1300 
1301   /* put together the new matrix */
1302   ierr = MatSeqBAIJSetPreallocation_SeqBAIJ(B,bs,MAT_SKIP_ALLOCATION,PETSC_NULL);CHKERRQ(ierr);
1303   ierr = PetscLogObjectParent(B,isicol);CHKERRQ(ierr);
1304   b    = (Mat_SeqBAIJ*)(B)->data;
1305   b->free_a       = PETSC_TRUE;
1306   b->free_ij      = PETSC_TRUE;
1307   b->singlemalloc = PETSC_FALSE;
1308   b->a          = ba;
1309   b->j          = bj;
1310   b->i          = bi;
1311   b->diag       = bdiag;
1312   b->ilen       = 0;
1313   b->imax       = 0;
1314   b->row        = isrow;
1315   b->col        = iscol;
1316   ierr          = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
1317   ierr          = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
1318   b->icol       = isicol;
1319   ierr = PetscMalloc((bs*(mbs+1))*sizeof(PetscScalar),&b->solve_work);CHKERRQ(ierr);
1320 
1321   ierr = PetscLogObjectMemory(B,nnz_max*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
1322   b->maxnz = nnz_max/bs2;
1323 
1324   (B)->factor                = MAT_FACTOR_ILUDT;
1325   (B)->info.factor_mallocs   = 0;
1326   (B)->info.fill_ratio_given = ((PetscReal)nnz_max)/((PetscReal)(ai[mbs]*bs2));
1327   CHKMEMQ;
1328   /* ------- end of symbolic factorization ---------*/
1329   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
1330   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
1331   ics  = ic;
1332 
1333   /* linked list for storing column indices of the active row */
1334   nlnk = mbs + 1;
1335   ierr = PetscLLCreate(mbs,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
1336 
1337   /* im: used by PetscLLAddSortedLU(); jtmp: working array for column indices of active row */
1338   ierr = PetscMalloc((2*mbs+1)*sizeof(PetscInt),&im);CHKERRQ(ierr);
1339   jtmp = im + mbs;
1340   /* rtmp, vtmp: working arrays for sparse and contiguous row entries of active row */
1341   ierr = PetscMalloc((2*mbs*bs2+1)*sizeof(MatScalar),&rtmp);CHKERRQ(ierr);
1342   vtmp = rtmp + bs2*mbs;
1343   ierr = PetscMalloc((mbs+1)*sizeof(PetscReal),&vtmp_abs);CHKERRQ(ierr);
1344 
1345   ierr       = PetscMalloc(bs*sizeof(PetscInt) + (bs+bs2)*sizeof(MatScalar),&v_work);CHKERRQ(ierr);
1346   multiplier = v_work + bs;
1347   v_pivots   = (PetscInt*)(multiplier + bs2);
1348 
1349   bi[0]    = 0;
1350   bdiag[0] = (nnz_max/bs2)-1; /* location of diagonal in factor B */
1351   bi[2*mbs+1] = bdiag[0]+1; /* endof bj and ba array */
1352   for (i=0; i<mbs; i++) {
1353     /* copy initial fill into linked list */
1354     nzi = 0; /* nonzeros for active row i */
1355     nzi = ai[r[i]+1] - ai[r[i]];
1356     if (!nzi) SETERRQ2(PETSC_ERR_MAT_LU_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",r[i],i);
1357     nzi_al = adiag[r[i]] - ai[r[i]];
1358     nzi_au = ai[r[i]+1] - adiag[r[i]] -1;
1359     /* printf("row %d, nzi_al/au %d %d\n",i,nzi_al,nzi_au); */
1360 
1361     /* load in initial unfactored row */
1362     ajtmp = aj + ai[r[i]];
1363     ierr = PetscLLAddPerm(nzi,ajtmp,ic,mbs,nlnk,lnk,lnkbt);CHKERRQ(ierr);
1364     ierr = PetscMemzero(rtmp,(mbs*bs2+1)*sizeof(PetscScalar));CHKERRQ(ierr);
1365     aatmp = a->a + bs2*ai[r[i]];
1366     for (j=0; j<nzi; j++) {
1367       ierr = PetscMemcpy(rtmp+bs2*ic[ajtmp[j]],aatmp+bs2*j,bs2*sizeof(MatScalar));CHKERRQ(ierr);
1368     }
1369 
1370     /* add pivot rows into linked list */
1371     row = lnk[mbs];
1372     while (row < i) {
1373       nzi_bl = bi[row+1] - bi[row] + 1;
1374       bjtmp = bj + bdiag[row+1]+1; /* points to 1st column next to the diagonal in U */
1375       ierr  = PetscLLAddSortedLU(bjtmp,row,nlnk,lnk,lnkbt,i,nzi_bl,im);CHKERRQ(ierr);
1376       nzi  += nlnk;
1377       row   = lnk[row];
1378     }
1379 
1380     /* copy data from lnk into jtmp, then initialize lnk */
1381     ierr = PetscLLClean(mbs,mbs,nzi,lnk,jtmp,lnkbt);CHKERRQ(ierr);
1382 
1383     /* numerical factorization */
1384     bjtmp = jtmp;
1385     row   = *bjtmp++; /* 1st pivot row */
1386 
1387     while  (row < i) {
1388       pc = rtmp + bs2*row;
1389       pv = ba + bs2*bdiag[row]; /* inv(diag) of the pivot row */
1390       Kernel_A_gets_A_times_B(bs,pc,pv,multiplier); /* pc= multiplier = pc*inv(diag[row]) */
1391       ierr = BlockAbs_private(1,bs2,pc,vtmp_abs);CHKERRQ(ierr);
1392       if (vtmp_abs[0] > dt){ /* apply tolerance dropping rule */
1393         pj         = bj + bdiag[row+1] + 1; /* point to 1st entry of U(row,:) */
1394         pv         = ba + bs2*(bdiag[row+1] + 1);
1395         nz         = bdiag[row] - bdiag[row+1] - 1; /* num of entries in U(row,:), excluding diagonal */
1396         for (j=0; j<nz; j++){
1397           Kernel_A_gets_A_minus_B_times_C(bs,rtmp+bs2*pj[j],pc,pv+bs2*j);
1398         }
1399         /* ierr = PetscLogFlops(bslog*(nz+1.0)-bs);CHKERRQ(ierr); */
1400       }
1401       row = *bjtmp++;
1402     }
1403 
1404     /* copy sparse rtmp into contiguous vtmp; separate L and U part */
1405     nzi_bl = 0; j = 0;
1406     while (jtmp[j] < i){ /* L-part. Note: jtmp is sorted */
1407       ierr = PetscMemcpy(vtmp+bs2*j,rtmp+bs2*jtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
1408       nzi_bl++; j++;
1409     }
1410     nzi_bu = nzi - nzi_bl -1;
1411     /* printf("nzi %d, nzi_bl %d, nzi_bu %d\n",nzi,nzi_bl,nzi_bu); */
1412 
1413     while (j < nzi){ /* U-part */
1414       ierr = PetscMemcpy(vtmp+bs2*j,rtmp+bs2*jtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
1415       /*
1416       printf(" col %d: ",jtmp[j]);
1417       for (j1=0; j1<bs2; j1++) printf(" %g",*(vtmp+bs2*j+j1));
1418       printf(" \n");
1419       */
1420       j++;
1421     }
1422 
1423     ierr = BlockAbs_private(nzi,bs2,vtmp,vtmp_abs);CHKERRQ(ierr);
1424     /*
1425     printf(" row %d, nzi %d, vtmp_abs\n",i,nzi);
1426     for (j1=0; j1<nzi; j1++) printf(" (%d %g),",jtmp[j1],vtmp_abs[j1]);
1427     printf(" \n");
1428     */
1429     bjtmp = bj + bi[i];
1430     batmp = ba + bs2*bi[i];
1431     /* apply level dropping rule to L part */
1432     ncut = nzi_al + dtcount;
1433     if (ncut < nzi_bl){
1434       ierr = PetscSortSplitReal(ncut,nzi_bl,vtmp_abs,jtmp);CHKERRQ(ierr);
1435       ierr = PetscSortIntWithScalarArray(ncut,jtmp,vtmp);CHKERRQ(ierr);
1436     } else {
1437       ncut = nzi_bl;
1438     }
1439     for (j=0; j<ncut; j++){
1440       bjtmp[j] = jtmp[j];
1441       ierr = PetscMemcpy(batmp+bs2*j,rtmp+bs2*bjtmp[j],bs2*sizeof(MatScalar));CHKERRQ(ierr);
1442       /*
1443       printf(" col %d: ",bjtmp[j]);
1444       for (j1=0; j1<bs2; j1++) printf(" %g,",*(batmp+bs2*j+j1));
1445       printf("\n");
1446       */
1447     }
1448     bi[i+1] = bi[i] + ncut;
1449     nzi = ncut + 1;
1450 
1451     /* apply level dropping rule to U part */
1452     ncut = nzi_au + dtcount;
1453     if (ncut < nzi_bu){
1454       ierr = PetscSortSplitReal(ncut,nzi_bu,vtmp_abs+nzi_bl+1,jtmp+nzi_bl+1);CHKERRQ(ierr);
1455       ierr = PetscSortIntWithScalarArray(ncut,jtmp+nzi_bl+1,vtmp+nzi_bl+1);CHKERRQ(ierr);
1456     } else {
1457       ncut = nzi_bu;
1458     }
1459     nzi += ncut;
1460 
1461     /* mark bdiagonal */
1462     bdiag[i+1]    = bdiag[i] - (ncut + 1);
1463     bi[2*mbs - i] = bi[2*mbs - i +1] - (ncut + 1);
1464 
1465     bjtmp = bj + bdiag[i];
1466     batmp = ba + bs2*bdiag[i];
1467     ierr = PetscMemcpy(batmp,rtmp+bs2*i,bs2*sizeof(MatScalar));CHKERRQ(ierr);
1468     *bjtmp = i;
1469     /*
1470     printf(" diag %d: ",*bjtmp);
1471     for (j=0; j<bs2; j++){
1472       printf(" %g,",batmp[j]);
1473     }
1474     printf("\n");
1475     */
1476     bjtmp = bj + bdiag[i+1]+1;
1477     batmp = ba + (bdiag[i+1]+1)*bs2;
1478 
1479     for (k=0; k<ncut; k++){
1480       bjtmp[k] = jtmp[nzi_bl+1+k];
1481       ierr = PetscMemcpy(batmp+bs2*k,rtmp+bs2*bjtmp[k],bs2*sizeof(MatScalar));CHKERRQ(ierr);
1482       /*
1483       printf(" col %d:",bjtmp[k]);
1484       for (j1=0; j1<bs2; j1++) printf(" %g,",*(batmp+bs2*k+j1));
1485       printf("\n");
1486       */
1487     }
1488 
1489     im[i] = nzi; /* used by PetscLLAddSortedLU() */
1490 
1491     /* invert diagonal block for simplier triangular solves - add shift??? */
1492     batmp = ba + bs2*bdiag[i];
1493     ierr = Kernel_A_gets_inverse_A(bs,batmp,v_pivots,v_work);CHKERRQ(ierr);
1494   } /* for (i=0; i<mbs; i++) */
1495 
1496   /* printf("end of L %d, beginning of U %d\n",bi[mbs],bdiag[mbs]); */
1497   if (bi[mbs] >= bdiag[mbs]) SETERRQ2(PETSC_ERR_ARG_SIZ,"end of L array %d cannot >= the beginning of U array %d",bi[mbs],bdiag[mbs]);
1498 
1499   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
1500   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
1501 
1502   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
1503 
1504   ierr = PetscFree(im);CHKERRQ(ierr);
1505   ierr = PetscFree(rtmp);CHKERRQ(ierr);
1506 
1507   ierr = PetscLogFlops(bs2*B->cmap->n);CHKERRQ(ierr);
1508   b->maxnz = b->nz = bi[mbs] + bdiag[0] - bdiag[mbs];
1509 
1510   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
1511   ierr = ISIdentity(isicol,&icol_identity);CHKERRQ(ierr);
1512   both_identity = (PetscTruth) (row_identity && icol_identity);
1513   if (row_identity && icol_identity) {
1514     B->ops->solve = MatSolve_SeqBAIJ_N_NaturalOrdering_newdatastruct;
1515   } else {
1516     B->ops->solve = MatSolve_SeqBAIJ_N_newdatastruct;
1517   }
1518 
1519   B->ops->solveadd          = 0;
1520   B->ops->solvetranspose    = 0;
1521   B->ops->solvetransposeadd = 0;
1522   B->ops->matsolve          = 0;
1523   B->assembled              = PETSC_TRUE;
1524   B->preallocated           = PETSC_TRUE;
1525   PetscFunctionReturn(0);
1526 }
1527