xref: /petsc/src/mat/impls/aij/seq/aijfact.c (revision e19f88df7cd0bcfe73faf98683db6f77794e28aa)
1 
2 #include <../src/mat/impls/aij/seq/aij.h>
3 #include <../src/mat/impls/sbaij/seq/sbaij.h>
4 #include <petscbt.h>
5 #include <../src/mat/utils/freespace.h>
6 
7 /*
8       Computes an ordering to get most of the large numerical values in the lower triangular part of the matrix
9 
10       This code does not work and is not called anywhere. It would be registered with MatOrderingRegisterAll()
11 */
12 PetscErrorCode MatGetOrdering_Flow_SeqAIJ(Mat mat,MatOrderingType type,IS *irow,IS *icol)
13 {
14   Mat_SeqAIJ        *a = (Mat_SeqAIJ*)mat->data;
15   PetscErrorCode    ierr;
16   PetscInt          i,j,jj,k, kk,n = mat->rmap->n, current = 0, newcurrent = 0,*order;
17   const PetscInt    *ai = a->i, *aj = a->j;
18   const PetscScalar *aa = a->a;
19   PetscBool         *done;
20   PetscReal         best,past = 0,future;
21 
22   PetscFunctionBegin;
23   /* pick initial row */
24   best = -1;
25   for (i=0; i<n; i++) {
26     future = 0.0;
27     for (j=ai[i]; j<ai[i+1]; j++) {
28       if (aj[j] != i) future += PetscAbsScalar(aa[j]);
29       else              past  = PetscAbsScalar(aa[j]);
30     }
31     if (!future) future = 1.e-10; /* if there is zero in the upper diagonal part want to rank this row high */
32     if (past/future > best) {
33       best    = past/future;
34       current = i;
35     }
36   }
37 
38   ierr     = PetscMalloc1(n,&done);CHKERRQ(ierr);
39   ierr     = PetscArrayzero(done,n);CHKERRQ(ierr);
40   ierr     = PetscMalloc1(n,&order);CHKERRQ(ierr);
41   order[0] = current;
42   for (i=0; i<n-1; i++) {
43     done[current] = PETSC_TRUE;
44     best          = -1;
45     /* loop over all neighbors of current pivot */
46     for (j=ai[current]; j<ai[current+1]; j++) {
47       jj = aj[j];
48       if (done[jj]) continue;
49       /* loop over columns of potential next row computing weights for below and above diagonal */
50       past = future = 0.0;
51       for (k=ai[jj]; k<ai[jj+1]; k++) {
52         kk = aj[k];
53         if (done[kk]) past += PetscAbsScalar(aa[k]);
54         else if (kk != jj) future += PetscAbsScalar(aa[k]);
55       }
56       if (!future) future = 1.e-10; /* if there is zero in the upper diagonal part want to rank this row high */
57       if (past/future > best) {
58         best       = past/future;
59         newcurrent = jj;
60       }
61     }
62     if (best == -1) { /* no neighbors to select from so select best of all that remain */
63       best = -1;
64       for (k=0; k<n; k++) {
65         if (done[k]) continue;
66         future = 0.0;
67         past   = 0.0;
68         for (j=ai[k]; j<ai[k+1]; j++) {
69           kk = aj[j];
70           if (done[kk])       past += PetscAbsScalar(aa[j]);
71           else if (kk != k) future += PetscAbsScalar(aa[j]);
72         }
73         if (!future) future = 1.e-10; /* if there is zero in the upper diagonal part want to rank this row high */
74         if (past/future > best) {
75           best       = past/future;
76           newcurrent = k;
77         }
78       }
79     }
80     if (current == newcurrent) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"newcurrent cannot be current");
81     current    = newcurrent;
82     order[i+1] = current;
83   }
84   ierr  = ISCreateGeneral(PETSC_COMM_SELF,n,order,PETSC_COPY_VALUES,irow);CHKERRQ(ierr);
85   *icol = *irow;
86   ierr  = PetscObjectReference((PetscObject)*irow);CHKERRQ(ierr);
87   ierr  = PetscFree(done);CHKERRQ(ierr);
88   ierr  = PetscFree(order);CHKERRQ(ierr);
89   PetscFunctionReturn(0);
90 }
91 
92 PETSC_INTERN PetscErrorCode MatGetFactor_seqaij_petsc(Mat A,MatFactorType ftype,Mat *B)
93 {
94   PetscInt       n = A->rmap->n;
95   PetscErrorCode ierr;
96 
97   PetscFunctionBegin;
98 #if defined(PETSC_USE_COMPLEX)
99   if (A->hermitian && !A->symmetric && (ftype == MAT_FACTOR_CHOLESKY||ftype == MAT_FACTOR_ICC)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Hermitian CHOLESKY or ICC Factor is not supported");
100 #endif
101   ierr = MatCreate(PetscObjectComm((PetscObject)A),B);CHKERRQ(ierr);
102   ierr = MatSetSizes(*B,n,n,n,n);CHKERRQ(ierr);
103   if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU || ftype == MAT_FACTOR_ILUDT) {
104     ierr = MatSetType(*B,MATSEQAIJ);CHKERRQ(ierr);
105 
106     (*B)->ops->ilufactorsymbolic = MatILUFactorSymbolic_SeqAIJ;
107     (*B)->ops->lufactorsymbolic  = MatLUFactorSymbolic_SeqAIJ;
108 
109     ierr = MatSetBlockSizesFromMats(*B,A,A);CHKERRQ(ierr);
110   } else if (ftype == MAT_FACTOR_CHOLESKY || ftype == MAT_FACTOR_ICC) {
111     ierr = MatSetType(*B,MATSEQSBAIJ);CHKERRQ(ierr);
112     ierr = MatSeqSBAIJSetPreallocation(*B,1,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
113 
114     (*B)->ops->iccfactorsymbolic      = MatICCFactorSymbolic_SeqAIJ;
115     (*B)->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SeqAIJ;
116   } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported");
117   (*B)->factortype = ftype;
118 
119   ierr = PetscFree((*B)->solvertype);CHKERRQ(ierr);
120   ierr = PetscStrallocpy(MATSOLVERPETSC,&(*B)->solvertype);CHKERRQ(ierr);
121   PetscFunctionReturn(0);
122 }
123 
124 PetscErrorCode MatLUFactorSymbolic_SeqAIJ_inplace(Mat B,Mat A,IS isrow,IS iscol,const MatFactorInfo *info)
125 {
126   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data,*b;
127   IS                 isicol;
128   PetscErrorCode     ierr;
129   const PetscInt     *r,*ic;
130   PetscInt           i,n=A->rmap->n,*ai=a->i,*aj=a->j;
131   PetscInt           *bi,*bj,*ajtmp;
132   PetscInt           *bdiag,row,nnz,nzi,reallocs=0,nzbd,*im;
133   PetscReal          f;
134   PetscInt           nlnk,*lnk,k,**bi_ptr;
135   PetscFreeSpaceList free_space=NULL,current_space=NULL;
136   PetscBT            lnkbt;
137   PetscBool          missing;
138 
139   PetscFunctionBegin;
140   if (A->rmap->N != A->cmap->N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"matrix must be square");
141   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
142   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
143 
144   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
145   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
146   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
147 
148   /* get new row pointers */
149   ierr  = PetscMalloc1(n+1,&bi);CHKERRQ(ierr);
150   bi[0] = 0;
151 
152   /* bdiag is location of diagonal in factor */
153   ierr     = PetscMalloc1(n+1,&bdiag);CHKERRQ(ierr);
154   bdiag[0] = 0;
155 
156   /* linked list for storing column indices of the active row */
157   nlnk = n + 1;
158   ierr = PetscLLCreate(n,n,nlnk,lnk,lnkbt);CHKERRQ(ierr);
159 
160   ierr = PetscMalloc2(n+1,&bi_ptr,n+1,&im);CHKERRQ(ierr);
161 
162   /* initial FreeSpace size is f*(ai[n]+1) */
163   f             = info->fill;
164   ierr          = PetscFreeSpaceGet(PetscRealIntMultTruncate(f,ai[n]+1),&free_space);CHKERRQ(ierr);
165   current_space = free_space;
166 
167   for (i=0; i<n; i++) {
168     /* copy previous fill into linked list */
169     nzi = 0;
170     nnz = ai[r[i]+1] - ai[r[i]];
171     ajtmp = aj + ai[r[i]];
172     ierr  = PetscLLAddPerm(nnz,ajtmp,ic,n,nlnk,lnk,lnkbt);CHKERRQ(ierr);
173     nzi  += nlnk;
174 
175     /* add pivot rows into linked list */
176     row = lnk[n];
177     while (row < i) {
178       nzbd  = bdiag[row] - bi[row] + 1;   /* num of entries in the row with column index <= row */
179       ajtmp = bi_ptr[row] + nzbd;   /* points to the entry next to the diagonal */
180       ierr  = PetscLLAddSortedLU(ajtmp,row,nlnk,lnk,lnkbt,i,nzbd,im);CHKERRQ(ierr);
181       nzi  += nlnk;
182       row   = lnk[row];
183     }
184     bi[i+1] = bi[i] + nzi;
185     im[i]   = nzi;
186 
187     /* mark bdiag */
188     nzbd = 0;
189     nnz  = nzi;
190     k    = lnk[n];
191     while (nnz-- && k < i) {
192       nzbd++;
193       k = lnk[k];
194     }
195     bdiag[i] = bi[i] + nzbd;
196 
197     /* if free space is not available, make more free space */
198     if (current_space->local_remaining<nzi) {
199       nnz  = PetscIntMultTruncate(n - i,nzi); /* estimated and max additional space needed */
200       ierr = PetscFreeSpaceGet(nnz,&current_space);CHKERRQ(ierr);
201       reallocs++;
202     }
203 
204     /* copy data into free space, then initialize lnk */
205     ierr = PetscLLClean(n,n,nzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
206 
207     bi_ptr[i]                       = current_space->array;
208     current_space->array           += nzi;
209     current_space->local_used      += nzi;
210     current_space->local_remaining -= nzi;
211   }
212 #if defined(PETSC_USE_INFO)
213   if (ai[n] != 0) {
214     PetscReal af = ((PetscReal)bi[n])/((PetscReal)ai[n]);
215     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)f,(double)af);CHKERRQ(ierr);
216     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
217     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g);\n",(double)af);CHKERRQ(ierr);
218     ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr);
219   } else {
220     ierr = PetscInfo(A,"Empty matrix\n");CHKERRQ(ierr);
221   }
222 #endif
223 
224   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
225   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
226 
227   /* destroy list of free space and other temporary array(s) */
228   ierr = PetscMalloc1(bi[n]+1,&bj);CHKERRQ(ierr);
229   ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr);
230   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
231   ierr = PetscFree2(bi_ptr,im);CHKERRQ(ierr);
232 
233   /* put together the new matrix */
234   ierr = MatSeqAIJSetPreallocation_SeqAIJ(B,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
235   ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)isicol);CHKERRQ(ierr);
236   b    = (Mat_SeqAIJ*)(B)->data;
237 
238   b->free_a       = PETSC_TRUE;
239   b->free_ij      = PETSC_TRUE;
240   b->singlemalloc = PETSC_FALSE;
241 
242   ierr    = PetscMalloc1(bi[n]+1,&b->a);CHKERRQ(ierr);
243   b->j    = bj;
244   b->i    = bi;
245   b->diag = bdiag;
246   b->ilen = 0;
247   b->imax = 0;
248   b->row  = isrow;
249   b->col  = iscol;
250   ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
251   ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
252   b->icol = isicol;
253   ierr    = PetscMalloc1(n+1,&b->solve_work);CHKERRQ(ierr);
254 
255   /* In b structure:  Free imax, ilen, old a, old j.  Allocate solve_work, new a, new j */
256   ierr     = PetscLogObjectMemory((PetscObject)B,(bi[n]-n)*(sizeof(PetscInt)+sizeof(PetscScalar)));CHKERRQ(ierr);
257   b->maxnz = b->nz = bi[n];
258 
259   (B)->factortype            = MAT_FACTOR_LU;
260   (B)->info.factor_mallocs   = reallocs;
261   (B)->info.fill_ratio_given = f;
262 
263   if (ai[n]) {
264     (B)->info.fill_ratio_needed = ((PetscReal)bi[n])/((PetscReal)ai[n]);
265   } else {
266     (B)->info.fill_ratio_needed = 0.0;
267   }
268   (B)->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_inplace;
269   if (a->inode.size) {
270     (B)->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_Inode_inplace;
271   }
272   PetscFunctionReturn(0);
273 }
274 
275 PetscErrorCode MatLUFactorSymbolic_SeqAIJ(Mat B,Mat A,IS isrow,IS iscol,const MatFactorInfo *info)
276 {
277   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data,*b;
278   IS                 isicol;
279   PetscErrorCode     ierr;
280   const PetscInt     *r,*ic,*ai=a->i,*aj=a->j,*ajtmp;
281   PetscInt           i,n=A->rmap->n;
282   PetscInt           *bi,*bj;
283   PetscInt           *bdiag,row,nnz,nzi,reallocs=0,nzbd,*im;
284   PetscReal          f;
285   PetscInt           nlnk,*lnk,k,**bi_ptr;
286   PetscFreeSpaceList free_space=NULL,current_space=NULL;
287   PetscBT            lnkbt;
288   PetscBool          missing;
289 
290   PetscFunctionBegin;
291   if (A->rmap->N != A->cmap->N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"matrix must be square");
292   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
293   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
294 
295   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
296   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
297   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
298 
299   /* get new row and diagonal pointers, must be allocated separately because they will be given to the Mat_SeqAIJ and freed separately */
300   ierr  = PetscMalloc1(n+1,&bi);CHKERRQ(ierr);
301   ierr  = PetscMalloc1(n+1,&bdiag);CHKERRQ(ierr);
302   bi[0] = bdiag[0] = 0;
303 
304   /* linked list for storing column indices of the active row */
305   nlnk = n + 1;
306   ierr = PetscLLCreate(n,n,nlnk,lnk,lnkbt);CHKERRQ(ierr);
307 
308   ierr = PetscMalloc2(n+1,&bi_ptr,n+1,&im);CHKERRQ(ierr);
309 
310   /* initial FreeSpace size is f*(ai[n]+1) */
311   f             = info->fill;
312   ierr          = PetscFreeSpaceGet(PetscRealIntMultTruncate(f,ai[n]+1),&free_space);CHKERRQ(ierr);
313   current_space = free_space;
314 
315   for (i=0; i<n; i++) {
316     /* copy previous fill into linked list */
317     nzi = 0;
318     nnz = ai[r[i]+1] - ai[r[i]];
319     ajtmp = aj + ai[r[i]];
320     ierr  = PetscLLAddPerm(nnz,ajtmp,ic,n,nlnk,lnk,lnkbt);CHKERRQ(ierr);
321     nzi  += nlnk;
322 
323     /* add pivot rows into linked list */
324     row = lnk[n];
325     while (row < i) {
326       nzbd  = bdiag[row] + 1; /* num of entries in the row with column index <= row */
327       ajtmp = bi_ptr[row] + nzbd; /* points to the entry next to the diagonal */
328       ierr  = PetscLLAddSortedLU(ajtmp,row,nlnk,lnk,lnkbt,i,nzbd,im);CHKERRQ(ierr);
329       nzi  += nlnk;
330       row   = lnk[row];
331     }
332     bi[i+1] = bi[i] + nzi;
333     im[i]   = nzi;
334 
335     /* mark bdiag */
336     nzbd = 0;
337     nnz  = nzi;
338     k    = lnk[n];
339     while (nnz-- && k < i) {
340       nzbd++;
341       k = lnk[k];
342     }
343     bdiag[i] = nzbd; /* note: bdiag[i] = nnzL as input for PetscFreeSpaceContiguous_LU() */
344 
345     /* if free space is not available, make more free space */
346     if (current_space->local_remaining<nzi) {
347       /* estimated additional space needed */
348       nnz  = PetscIntMultTruncate(2,PetscIntMultTruncate(n-1,nzi));
349       ierr = PetscFreeSpaceGet(nnz,&current_space);CHKERRQ(ierr);
350       reallocs++;
351     }
352 
353     /* copy data into free space, then initialize lnk */
354     ierr = PetscLLClean(n,n,nzi,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
355 
356     bi_ptr[i]                       = current_space->array;
357     current_space->array           += nzi;
358     current_space->local_used      += nzi;
359     current_space->local_remaining -= nzi;
360   }
361 
362   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
363   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
364 
365   /*   copy free_space into bj and free free_space; set bi, bj, bdiag in new datastructure; */
366   ierr = PetscMalloc1(bi[n]+1,&bj);CHKERRQ(ierr);
367   ierr = PetscFreeSpaceContiguous_LU(&free_space,bj,n,bi,bdiag);CHKERRQ(ierr);
368   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
369   ierr = PetscFree2(bi_ptr,im);CHKERRQ(ierr);
370 
371   /* put together the new matrix */
372   ierr = MatSeqAIJSetPreallocation_SeqAIJ(B,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
373   ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)isicol);CHKERRQ(ierr);
374   b    = (Mat_SeqAIJ*)(B)->data;
375 
376   b->free_a       = PETSC_TRUE;
377   b->free_ij      = PETSC_TRUE;
378   b->singlemalloc = PETSC_FALSE;
379 
380   ierr = PetscMalloc1(bdiag[0]+1,&b->a);CHKERRQ(ierr);
381 
382   b->j    = bj;
383   b->i    = bi;
384   b->diag = bdiag;
385   b->ilen = 0;
386   b->imax = 0;
387   b->row  = isrow;
388   b->col  = iscol;
389   ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
390   ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
391   b->icol = isicol;
392   ierr    = PetscMalloc1(n+1,&b->solve_work);CHKERRQ(ierr);
393 
394   /* In b structure:  Free imax, ilen, old a, old j.  Allocate solve_work, new a, new j */
395   ierr     = PetscLogObjectMemory((PetscObject)B,(bdiag[0]+1)*(sizeof(PetscInt)+sizeof(PetscScalar)));CHKERRQ(ierr);
396   b->maxnz = b->nz = bdiag[0]+1;
397 
398   B->factortype            = MAT_FACTOR_LU;
399   B->info.factor_mallocs   = reallocs;
400   B->info.fill_ratio_given = f;
401 
402   if (ai[n]) {
403     B->info.fill_ratio_needed = ((PetscReal)(bdiag[0]+1))/((PetscReal)ai[n]);
404   } else {
405     B->info.fill_ratio_needed = 0.0;
406   }
407 #if defined(PETSC_USE_INFO)
408   if (ai[n] != 0) {
409     PetscReal af = B->info.fill_ratio_needed;
410     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)f,(double)af);CHKERRQ(ierr);
411     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
412     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g);\n",(double)af);CHKERRQ(ierr);
413     ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr);
414   } else {
415     ierr = PetscInfo(A,"Empty matrix\n");CHKERRQ(ierr);
416   }
417 #endif
418   B->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ;
419   if (a->inode.size) {
420     B->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_Inode;
421   }
422   ierr = MatSeqAIJCheckInode_FactorLU(B);CHKERRQ(ierr);
423   PetscFunctionReturn(0);
424 }
425 
426 /*
427     Trouble in factorization, should we dump the original matrix?
428 */
429 PetscErrorCode MatFactorDumpMatrix(Mat A)
430 {
431   PetscErrorCode ierr;
432   PetscBool      flg = PETSC_FALSE;
433 
434   PetscFunctionBegin;
435   ierr = PetscOptionsGetBool(((PetscObject)A)->options,NULL,"-mat_factor_dump_on_error",&flg,NULL);CHKERRQ(ierr);
436   if (flg) {
437     PetscViewer viewer;
438     char        filename[PETSC_MAX_PATH_LEN];
439 
440     ierr = PetscSNPrintf(filename,PETSC_MAX_PATH_LEN,"matrix_factor_error.%d",PetscGlobalRank);CHKERRQ(ierr);
441     ierr = PetscViewerBinaryOpen(PetscObjectComm((PetscObject)A),filename,FILE_MODE_WRITE,&viewer);CHKERRQ(ierr);
442     ierr = MatView(A,viewer);CHKERRQ(ierr);
443     ierr = PetscViewerDestroy(&viewer);CHKERRQ(ierr);
444   }
445   PetscFunctionReturn(0);
446 }
447 
448 PetscErrorCode MatLUFactorNumeric_SeqAIJ(Mat B,Mat A,const MatFactorInfo *info)
449 {
450   Mat             C     =B;
451   Mat_SeqAIJ      *a    =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)C->data;
452   IS              isrow = b->row,isicol = b->icol;
453   PetscErrorCode  ierr;
454   const PetscInt  *r,*ic,*ics;
455   const PetscInt  n=A->rmap->n,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j,*bdiag=b->diag;
456   PetscInt        i,j,k,nz,nzL,row,*pj;
457   const PetscInt  *ajtmp,*bjtmp;
458   MatScalar       *rtmp,*pc,multiplier,*pv;
459   const MatScalar *aa=a->a,*v;
460   PetscBool       row_identity,col_identity;
461   FactorShiftCtx  sctx;
462   const PetscInt  *ddiag;
463   PetscReal       rs;
464   MatScalar       d;
465 
466   PetscFunctionBegin;
467   /* MatPivotSetUp(): initialize shift context sctx */
468   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
469 
470   if (info->shifttype == (PetscReal) MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
471     ddiag          = a->diag;
472     sctx.shift_top = info->zeropivot;
473     for (i=0; i<n; i++) {
474       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
475       d  = (aa)[ddiag[i]];
476       rs = -PetscAbsScalar(d) - PetscRealPart(d);
477       v  = aa+ai[i];
478       nz = ai[i+1] - ai[i];
479       for (j=0; j<nz; j++) rs += PetscAbsScalar(v[j]);
480       if (rs>sctx.shift_top) sctx.shift_top = rs;
481     }
482     sctx.shift_top *= 1.1;
483     sctx.nshift_max = 5;
484     sctx.shift_lo   = 0.;
485     sctx.shift_hi   = 1.;
486   }
487 
488   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
489   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
490   ierr = PetscMalloc1(n+1,&rtmp);CHKERRQ(ierr);
491   ics  = ic;
492 
493   do {
494     sctx.newshift = PETSC_FALSE;
495     for (i=0; i<n; i++) {
496       /* zero rtmp */
497       /* L part */
498       nz    = bi[i+1] - bi[i];
499       bjtmp = bj + bi[i];
500       for  (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0;
501 
502       /* U part */
503       nz    = bdiag[i]-bdiag[i+1];
504       bjtmp = bj + bdiag[i+1]+1;
505       for  (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0;
506 
507       /* load in initial (unfactored row) */
508       nz    = ai[r[i]+1] - ai[r[i]];
509       ajtmp = aj + ai[r[i]];
510       v     = aa + ai[r[i]];
511       for (j=0; j<nz; j++) {
512         rtmp[ics[ajtmp[j]]] = v[j];
513       }
514       /* ZeropivotApply() */
515       rtmp[i] += sctx.shift_amount;  /* shift the diagonal of the matrix */
516 
517       /* elimination */
518       bjtmp = bj + bi[i];
519       row   = *bjtmp++;
520       nzL   = bi[i+1] - bi[i];
521       for (k=0; k < nzL; k++) {
522         pc = rtmp + row;
523         if (*pc != 0.0) {
524           pv         = b->a + bdiag[row];
525           multiplier = *pc * (*pv);
526           *pc        = multiplier;
527 
528           pj = b->j + bdiag[row+1]+1; /* beginning of U(row,:) */
529           pv = b->a + bdiag[row+1]+1;
530           nz = bdiag[row]-bdiag[row+1]-1; /* num of entries in U(row,:) excluding diag */
531 
532           for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j];
533           ierr = PetscLogFlops(1+2*nz);CHKERRQ(ierr);
534         }
535         row = *bjtmp++;
536       }
537 
538       /* finished row so stick it into b->a */
539       rs = 0.0;
540       /* L part */
541       pv = b->a + bi[i];
542       pj = b->j + bi[i];
543       nz = bi[i+1] - bi[i];
544       for (j=0; j<nz; j++) {
545         pv[j] = rtmp[pj[j]]; rs += PetscAbsScalar(pv[j]);
546       }
547 
548       /* U part */
549       pv = b->a + bdiag[i+1]+1;
550       pj = b->j + bdiag[i+1]+1;
551       nz = bdiag[i] - bdiag[i+1]-1;
552       for (j=0; j<nz; j++) {
553         pv[j] = rtmp[pj[j]]; rs += PetscAbsScalar(pv[j]);
554       }
555 
556       sctx.rs = rs;
557       sctx.pv = rtmp[i];
558       ierr    = MatPivotCheck(B,A,info,&sctx,i);CHKERRQ(ierr);
559       if (sctx.newshift) break; /* break for-loop */
560       rtmp[i] = sctx.pv; /* sctx.pv might be updated in the case of MAT_SHIFT_INBLOCKS */
561 
562       /* Mark diagonal and invert diagonal for simplier triangular solves */
563       pv  = b->a + bdiag[i];
564       *pv = 1.0/rtmp[i];
565 
566     } /* endof for (i=0; i<n; i++) { */
567 
568     /* MatPivotRefine() */
569     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE && !sctx.newshift && sctx.shift_fraction>0 && sctx.nshift<sctx.nshift_max) {
570       /*
571        * if no shift in this attempt & shifting & started shifting & can refine,
572        * then try lower shift
573        */
574       sctx.shift_hi       = sctx.shift_fraction;
575       sctx.shift_fraction = (sctx.shift_hi+sctx.shift_lo)/2.;
576       sctx.shift_amount   = sctx.shift_fraction * sctx.shift_top;
577       sctx.newshift       = PETSC_TRUE;
578       sctx.nshift++;
579     }
580   } while (sctx.newshift);
581 
582   ierr = PetscFree(rtmp);CHKERRQ(ierr);
583   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
584   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
585 
586   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
587   ierr = ISIdentity(isicol,&col_identity);CHKERRQ(ierr);
588   if (b->inode.size) {
589     C->ops->solve = MatSolve_SeqAIJ_Inode;
590   } else if (row_identity && col_identity) {
591     C->ops->solve = MatSolve_SeqAIJ_NaturalOrdering;
592   } else {
593     C->ops->solve = MatSolve_SeqAIJ;
594   }
595   C->ops->solveadd          = MatSolveAdd_SeqAIJ;
596   C->ops->solvetranspose    = MatSolveTranspose_SeqAIJ;
597   C->ops->solvetransposeadd = MatSolveTransposeAdd_SeqAIJ;
598   C->ops->matsolve          = MatMatSolve_SeqAIJ;
599   C->assembled              = PETSC_TRUE;
600   C->preallocated           = PETSC_TRUE;
601 
602   ierr = PetscLogFlops(C->cmap->n);CHKERRQ(ierr);
603 
604   /* MatShiftView(A,info,&sctx) */
605   if (sctx.nshift) {
606     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
607       ierr = PetscInfo4(A,"number of shift_pd tries %D, shift_amount %g, diagonal shifted up by %e fraction top_value %e\n",sctx.nshift,(double)sctx.shift_amount,(double)sctx.shift_fraction,(double)sctx.shift_top);CHKERRQ(ierr);
608     } else if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
609       ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
610     } else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) {
611       ierr = PetscInfo2(A,"number of shift_inblocks applied %D, each shift_amount %g\n",sctx.nshift,(double)info->shiftamount);CHKERRQ(ierr);
612     }
613   }
614   PetscFunctionReturn(0);
615 }
616 
617 PetscErrorCode MatLUFactorNumeric_SeqAIJ_inplace(Mat B,Mat A,const MatFactorInfo *info)
618 {
619   Mat             C     =B;
620   Mat_SeqAIJ      *a    =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)C->data;
621   IS              isrow = b->row,isicol = b->icol;
622   PetscErrorCode  ierr;
623   const PetscInt  *r,*ic,*ics;
624   PetscInt        nz,row,i,j,n=A->rmap->n,diag;
625   const PetscInt  *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
626   const PetscInt  *ajtmp,*bjtmp,*diag_offset = b->diag,*pj;
627   MatScalar       *pv,*rtmp,*pc,multiplier,d;
628   const MatScalar *v,*aa=a->a;
629   PetscReal       rs=0.0;
630   FactorShiftCtx  sctx;
631   const PetscInt  *ddiag;
632   PetscBool       row_identity, col_identity;
633 
634   PetscFunctionBegin;
635   /* MatPivotSetUp(): initialize shift context sctx */
636   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
637 
638   if (info->shifttype == (PetscReal) MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
639     ddiag          = a->diag;
640     sctx.shift_top = info->zeropivot;
641     for (i=0; i<n; i++) {
642       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
643       d  = (aa)[ddiag[i]];
644       rs = -PetscAbsScalar(d) - PetscRealPart(d);
645       v  = aa+ai[i];
646       nz = ai[i+1] - ai[i];
647       for (j=0; j<nz; j++) rs += PetscAbsScalar(v[j]);
648       if (rs>sctx.shift_top) sctx.shift_top = rs;
649     }
650     sctx.shift_top *= 1.1;
651     sctx.nshift_max = 5;
652     sctx.shift_lo   = 0.;
653     sctx.shift_hi   = 1.;
654   }
655 
656   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
657   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
658   ierr = PetscMalloc1(n+1,&rtmp);CHKERRQ(ierr);
659   ics  = ic;
660 
661   do {
662     sctx.newshift = PETSC_FALSE;
663     for (i=0; i<n; i++) {
664       nz    = bi[i+1] - bi[i];
665       bjtmp = bj + bi[i];
666       for  (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0;
667 
668       /* load in initial (unfactored row) */
669       nz    = ai[r[i]+1] - ai[r[i]];
670       ajtmp = aj + ai[r[i]];
671       v     = aa + ai[r[i]];
672       for (j=0; j<nz; j++) {
673         rtmp[ics[ajtmp[j]]] = v[j];
674       }
675       rtmp[ics[r[i]]] += sctx.shift_amount; /* shift the diagonal of the matrix */
676 
677       row = *bjtmp++;
678       while  (row < i) {
679         pc = rtmp + row;
680         if (*pc != 0.0) {
681           pv         = b->a + diag_offset[row];
682           pj         = b->j + diag_offset[row] + 1;
683           multiplier = *pc / *pv++;
684           *pc        = multiplier;
685           nz         = bi[row+1] - diag_offset[row] - 1;
686           for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j];
687           ierr = PetscLogFlops(1+2*nz);CHKERRQ(ierr);
688         }
689         row = *bjtmp++;
690       }
691       /* finished row so stick it into b->a */
692       pv   = b->a + bi[i];
693       pj   = b->j + bi[i];
694       nz   = bi[i+1] - bi[i];
695       diag = diag_offset[i] - bi[i];
696       rs   = 0.0;
697       for (j=0; j<nz; j++) {
698         pv[j] = rtmp[pj[j]];
699         rs   += PetscAbsScalar(pv[j]);
700       }
701       rs -= PetscAbsScalar(pv[diag]);
702 
703       sctx.rs = rs;
704       sctx.pv = pv[diag];
705       ierr    = MatPivotCheck(B,A,info,&sctx,i);CHKERRQ(ierr);
706       if (sctx.newshift) break;
707       pv[diag] = sctx.pv;
708     }
709 
710     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE && !sctx.newshift && sctx.shift_fraction>0 && sctx.nshift<sctx.nshift_max) {
711       /*
712        * if no shift in this attempt & shifting & started shifting & can refine,
713        * then try lower shift
714        */
715       sctx.shift_hi       = sctx.shift_fraction;
716       sctx.shift_fraction = (sctx.shift_hi+sctx.shift_lo)/2.;
717       sctx.shift_amount   = sctx.shift_fraction * sctx.shift_top;
718       sctx.newshift       = PETSC_TRUE;
719       sctx.nshift++;
720     }
721   } while (sctx.newshift);
722 
723   /* invert diagonal entries for simplier triangular solves */
724   for (i=0; i<n; i++) {
725     b->a[diag_offset[i]] = 1.0/b->a[diag_offset[i]];
726   }
727   ierr = PetscFree(rtmp);CHKERRQ(ierr);
728   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
729   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
730 
731   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
732   ierr = ISIdentity(isicol,&col_identity);CHKERRQ(ierr);
733   if (row_identity && col_identity) {
734     C->ops->solve = MatSolve_SeqAIJ_NaturalOrdering_inplace;
735   } else {
736     C->ops->solve = MatSolve_SeqAIJ_inplace;
737   }
738   C->ops->solveadd          = MatSolveAdd_SeqAIJ_inplace;
739   C->ops->solvetranspose    = MatSolveTranspose_SeqAIJ_inplace;
740   C->ops->solvetransposeadd = MatSolveTransposeAdd_SeqAIJ_inplace;
741   C->ops->matsolve          = MatMatSolve_SeqAIJ_inplace;
742 
743   C->assembled    = PETSC_TRUE;
744   C->preallocated = PETSC_TRUE;
745 
746   ierr = PetscLogFlops(C->cmap->n);CHKERRQ(ierr);
747   if (sctx.nshift) {
748     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
749       ierr = PetscInfo4(A,"number of shift_pd tries %D, shift_amount %g, diagonal shifted up by %e fraction top_value %e\n",sctx.nshift,(double)sctx.shift_amount,(double)sctx.shift_fraction,(double)sctx.shift_top);CHKERRQ(ierr);
750     } else if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
751       ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
752     }
753   }
754   (C)->ops->solve          = MatSolve_SeqAIJ_inplace;
755   (C)->ops->solvetranspose = MatSolveTranspose_SeqAIJ_inplace;
756 
757   ierr = MatSeqAIJCheckInode(C);CHKERRQ(ierr);
758   PetscFunctionReturn(0);
759 }
760 
761 /*
762    This routine implements inplace ILU(0) with row or/and column permutations.
763    Input:
764      A - original matrix
765    Output;
766      A - a->i (rowptr) is same as original rowptr, but factored i-the row is stored in rowperm[i]
767          a->j (col index) is permuted by the inverse of colperm, then sorted
768          a->a reordered accordingly with a->j
769          a->diag (ptr to diagonal elements) is updated.
770 */
771 PetscErrorCode MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(Mat B,Mat A,const MatFactorInfo *info)
772 {
773   Mat_SeqAIJ      *a    =(Mat_SeqAIJ*)A->data;
774   IS              isrow = a->row,isicol = a->icol;
775   PetscErrorCode  ierr;
776   const PetscInt  *r,*ic,*ics;
777   PetscInt        i,j,n=A->rmap->n,*ai=a->i,*aj=a->j;
778   PetscInt        *ajtmp,nz,row;
779   PetscInt        *diag = a->diag,nbdiag,*pj;
780   PetscScalar     *rtmp,*pc,multiplier,d;
781   MatScalar       *pv,*v;
782   PetscReal       rs;
783   FactorShiftCtx  sctx;
784   const MatScalar *aa=a->a,*vtmp;
785 
786   PetscFunctionBegin;
787   if (A != B) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"input and output matrix must have same address");
788 
789   /* MatPivotSetUp(): initialize shift context sctx */
790   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
791 
792   if (info->shifttype == (PetscReal) MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
793     const PetscInt *ddiag = a->diag;
794     sctx.shift_top = info->zeropivot;
795     for (i=0; i<n; i++) {
796       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
797       d    = (aa)[ddiag[i]];
798       rs   = -PetscAbsScalar(d) - PetscRealPart(d);
799       vtmp = aa+ai[i];
800       nz   = ai[i+1] - ai[i];
801       for (j=0; j<nz; j++) rs += PetscAbsScalar(vtmp[j]);
802       if (rs>sctx.shift_top) sctx.shift_top = rs;
803     }
804     sctx.shift_top *= 1.1;
805     sctx.nshift_max = 5;
806     sctx.shift_lo   = 0.;
807     sctx.shift_hi   = 1.;
808   }
809 
810   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
811   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
812   ierr = PetscMalloc1(n+1,&rtmp);CHKERRQ(ierr);
813   ierr = PetscArrayzero(rtmp,n+1);CHKERRQ(ierr);
814   ics  = ic;
815 
816 #if defined(MV)
817   sctx.shift_top      = 0.;
818   sctx.nshift_max     = 0;
819   sctx.shift_lo       = 0.;
820   sctx.shift_hi       = 0.;
821   sctx.shift_fraction = 0.;
822 
823   if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
824     sctx.shift_top = 0.;
825     for (i=0; i<n; i++) {
826       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
827       d  = (a->a)[diag[i]];
828       rs = -PetscAbsScalar(d) - PetscRealPart(d);
829       v  = a->a+ai[i];
830       nz = ai[i+1] - ai[i];
831       for (j=0; j<nz; j++) rs += PetscAbsScalar(v[j]);
832       if (rs>sctx.shift_top) sctx.shift_top = rs;
833     }
834     if (sctx.shift_top < info->zeropivot) sctx.shift_top = info->zeropivot;
835     sctx.shift_top *= 1.1;
836     sctx.nshift_max = 5;
837     sctx.shift_lo   = 0.;
838     sctx.shift_hi   = 1.;
839   }
840 
841   sctx.shift_amount = 0.;
842   sctx.nshift       = 0;
843 #endif
844 
845   do {
846     sctx.newshift = PETSC_FALSE;
847     for (i=0; i<n; i++) {
848       /* load in initial unfactored row */
849       nz    = ai[r[i]+1] - ai[r[i]];
850       ajtmp = aj + ai[r[i]];
851       v     = a->a + ai[r[i]];
852       /* sort permuted ajtmp and values v accordingly */
853       for (j=0; j<nz; j++) ajtmp[j] = ics[ajtmp[j]];
854       ierr = PetscSortIntWithScalarArray(nz,ajtmp,v);CHKERRQ(ierr);
855 
856       diag[r[i]] = ai[r[i]];
857       for (j=0; j<nz; j++) {
858         rtmp[ajtmp[j]] = v[j];
859         if (ajtmp[j] < i) diag[r[i]]++; /* update a->diag */
860       }
861       rtmp[r[i]] += sctx.shift_amount; /* shift the diagonal of the matrix */
862 
863       row = *ajtmp++;
864       while  (row < i) {
865         pc = rtmp + row;
866         if (*pc != 0.0) {
867           pv = a->a + diag[r[row]];
868           pj = aj + diag[r[row]] + 1;
869 
870           multiplier = *pc / *pv++;
871           *pc        = multiplier;
872           nz         = ai[r[row]+1] - diag[r[row]] - 1;
873           for (j=0; j<nz; j++) rtmp[pj[j]] -= multiplier * pv[j];
874           ierr = PetscLogFlops(1+2*nz);CHKERRQ(ierr);
875         }
876         row = *ajtmp++;
877       }
878       /* finished row so overwrite it onto a->a */
879       pv     = a->a + ai[r[i]];
880       pj     = aj + ai[r[i]];
881       nz     = ai[r[i]+1] - ai[r[i]];
882       nbdiag = diag[r[i]] - ai[r[i]]; /* num of entries before the diagonal */
883 
884       rs = 0.0;
885       for (j=0; j<nz; j++) {
886         pv[j] = rtmp[pj[j]];
887         if (j != nbdiag) rs += PetscAbsScalar(pv[j]);
888       }
889 
890       sctx.rs = rs;
891       sctx.pv = pv[nbdiag];
892       ierr    = MatPivotCheck(B,A,info,&sctx,i);CHKERRQ(ierr);
893       if (sctx.newshift) break;
894       pv[nbdiag] = sctx.pv;
895     }
896 
897     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE && !sctx.newshift && sctx.shift_fraction>0 && sctx.nshift<sctx.nshift_max) {
898       /*
899        * if no shift in this attempt & shifting & started shifting & can refine,
900        * then try lower shift
901        */
902       sctx.shift_hi       = sctx.shift_fraction;
903       sctx.shift_fraction = (sctx.shift_hi+sctx.shift_lo)/2.;
904       sctx.shift_amount   = sctx.shift_fraction * sctx.shift_top;
905       sctx.newshift       = PETSC_TRUE;
906       sctx.nshift++;
907     }
908   } while (sctx.newshift);
909 
910   /* invert diagonal entries for simplier triangular solves */
911   for (i=0; i<n; i++) {
912     a->a[diag[r[i]]] = 1.0/a->a[diag[r[i]]];
913   }
914 
915   ierr = PetscFree(rtmp);CHKERRQ(ierr);
916   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
917   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
918 
919   A->ops->solve             = MatSolve_SeqAIJ_InplaceWithPerm;
920   A->ops->solveadd          = MatSolveAdd_SeqAIJ_inplace;
921   A->ops->solvetranspose    = MatSolveTranspose_SeqAIJ_inplace;
922   A->ops->solvetransposeadd = MatSolveTransposeAdd_SeqAIJ_inplace;
923 
924   A->assembled    = PETSC_TRUE;
925   A->preallocated = PETSC_TRUE;
926 
927   ierr = PetscLogFlops(A->cmap->n);CHKERRQ(ierr);
928   if (sctx.nshift) {
929     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
930       ierr = PetscInfo4(A,"number of shift_pd tries %D, shift_amount %g, diagonal shifted up by %e fraction top_value %e\n",sctx.nshift,(double)sctx.shift_amount,(double)sctx.shift_fraction,(double)sctx.shift_top);CHKERRQ(ierr);
931     } else if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
932       ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
933     }
934   }
935   PetscFunctionReturn(0);
936 }
937 
938 /* ----------------------------------------------------------- */
939 PetscErrorCode MatLUFactor_SeqAIJ(Mat A,IS row,IS col,const MatFactorInfo *info)
940 {
941   PetscErrorCode ierr;
942   Mat            C;
943 
944   PetscFunctionBegin;
945   ierr = MatGetFactor(A,MATSOLVERPETSC,MAT_FACTOR_LU,&C);CHKERRQ(ierr);
946   ierr = MatLUFactorSymbolic(C,A,row,col,info);CHKERRQ(ierr);
947   ierr = MatLUFactorNumeric(C,A,info);CHKERRQ(ierr);
948 
949   A->ops->solve          = C->ops->solve;
950   A->ops->solvetranspose = C->ops->solvetranspose;
951 
952   ierr = MatHeaderMerge(A,&C);CHKERRQ(ierr);
953   ierr = PetscLogObjectParent((PetscObject)A,(PetscObject)((Mat_SeqAIJ*)(A->data))->icol);CHKERRQ(ierr);
954   PetscFunctionReturn(0);
955 }
956 /* ----------------------------------------------------------- */
957 
958 
959 PetscErrorCode MatSolve_SeqAIJ_inplace(Mat A,Vec bb,Vec xx)
960 {
961   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
962   IS                iscol = a->col,isrow = a->row;
963   PetscErrorCode    ierr;
964   PetscInt          i, n = A->rmap->n,*vi,*ai = a->i,*aj = a->j;
965   PetscInt          nz;
966   const PetscInt    *rout,*cout,*r,*c;
967   PetscScalar       *x,*tmp,*tmps,sum;
968   const PetscScalar *b;
969   const MatScalar   *aa = a->a,*v;
970 
971   PetscFunctionBegin;
972   if (!n) PetscFunctionReturn(0);
973 
974   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
975   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
976   tmp  = a->solve_work;
977 
978   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
979   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout + (n-1);
980 
981   /* forward solve the lower triangular */
982   tmp[0] = b[*r++];
983   tmps   = tmp;
984   for (i=1; i<n; i++) {
985     v   = aa + ai[i];
986     vi  = aj + ai[i];
987     nz  = a->diag[i] - ai[i];
988     sum = b[*r++];
989     PetscSparseDenseMinusDot(sum,tmps,v,vi,nz);
990     tmp[i] = sum;
991   }
992 
993   /* backward solve the upper triangular */
994   for (i=n-1; i>=0; i--) {
995     v   = aa + a->diag[i] + 1;
996     vi  = aj + a->diag[i] + 1;
997     nz  = ai[i+1] - a->diag[i] - 1;
998     sum = tmp[i];
999     PetscSparseDenseMinusDot(sum,tmps,v,vi,nz);
1000     x[*c--] = tmp[i] = sum*aa[a->diag[i]];
1001   }
1002 
1003   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1004   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1005   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1006   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
1007   ierr = PetscLogFlops(2.0*a->nz - A->cmap->n);CHKERRQ(ierr);
1008   PetscFunctionReturn(0);
1009 }
1010 
1011 PetscErrorCode MatMatSolve_SeqAIJ_inplace(Mat A,Mat B,Mat X)
1012 {
1013   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1014   IS                iscol = a->col,isrow = a->row;
1015   PetscErrorCode    ierr;
1016   PetscInt          i, n = A->rmap->n,*vi,*ai = a->i,*aj = a->j;
1017   PetscInt          nz,neq;
1018   const PetscInt    *rout,*cout,*r,*c;
1019   PetscScalar       *x,*tmp,*tmps,sum;
1020   const PetscScalar *aa = a->a,*v;
1021   const PetscScalar *b;
1022   PetscBool         bisdense,xisdense;
1023 
1024   PetscFunctionBegin;
1025   if (!n) PetscFunctionReturn(0);
1026 
1027   ierr = PetscObjectTypeCompare((PetscObject)B,MATSEQDENSE,&bisdense);CHKERRQ(ierr);
1028   if (!bisdense) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"B matrix must be a SeqDense matrix");
1029   if (X != B) {
1030     ierr = PetscObjectTypeCompare((PetscObject)X,MATSEQDENSE,&xisdense);CHKERRQ(ierr);
1031     if (!xisdense) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"X matrix must be a SeqDense matrix");
1032   }
1033 
1034   ierr = MatDenseGetArrayRead(B,&b);CHKERRQ(ierr);
1035   ierr = MatDenseGetArray(X,&x);CHKERRQ(ierr);
1036 
1037   tmp  = a->solve_work;
1038   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1039   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1040 
1041   for (neq=0; neq<B->cmap->n; neq++) {
1042     /* forward solve the lower triangular */
1043     tmp[0] = b[r[0]];
1044     tmps   = tmp;
1045     for (i=1; i<n; i++) {
1046       v   = aa + ai[i];
1047       vi  = aj + ai[i];
1048       nz  = a->diag[i] - ai[i];
1049       sum = b[r[i]];
1050       PetscSparseDenseMinusDot(sum,tmps,v,vi,nz);
1051       tmp[i] = sum;
1052     }
1053     /* backward solve the upper triangular */
1054     for (i=n-1; i>=0; i--) {
1055       v   = aa + a->diag[i] + 1;
1056       vi  = aj + a->diag[i] + 1;
1057       nz  = ai[i+1] - a->diag[i] - 1;
1058       sum = tmp[i];
1059       PetscSparseDenseMinusDot(sum,tmps,v,vi,nz);
1060       x[c[i]] = tmp[i] = sum*aa[a->diag[i]];
1061     }
1062 
1063     b += n;
1064     x += n;
1065   }
1066   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1067   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1068   ierr = MatDenseRestoreArrayRead(B,&b);CHKERRQ(ierr);
1069   ierr = MatDenseRestoreArray(X,&x);CHKERRQ(ierr);
1070   ierr = PetscLogFlops(B->cmap->n*(2.0*a->nz - n));CHKERRQ(ierr);
1071   PetscFunctionReturn(0);
1072 }
1073 
1074 PetscErrorCode MatMatSolve_SeqAIJ(Mat A,Mat B,Mat X)
1075 {
1076   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1077   IS                iscol = a->col,isrow = a->row;
1078   PetscErrorCode    ierr;
1079   PetscInt          i, n = A->rmap->n,*vi,*ai = a->i,*aj = a->j,*adiag = a->diag;
1080   PetscInt          nz,neq;
1081   const PetscInt    *rout,*cout,*r,*c;
1082   PetscScalar       *x,*tmp,sum;
1083   const PetscScalar *b;
1084   const PetscScalar *aa = a->a,*v;
1085   PetscBool         bisdense,xisdense;
1086 
1087   PetscFunctionBegin;
1088   if (!n) PetscFunctionReturn(0);
1089 
1090   ierr = PetscObjectTypeCompare((PetscObject)B,MATSEQDENSE,&bisdense);CHKERRQ(ierr);
1091   if (!bisdense) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"B matrix must be a SeqDense matrix");
1092   if (X != B) {
1093     ierr = PetscObjectTypeCompare((PetscObject)X,MATSEQDENSE,&xisdense);CHKERRQ(ierr);
1094     if (!xisdense) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"X matrix must be a SeqDense matrix");
1095   }
1096 
1097   ierr = MatDenseGetArrayRead(B,&b);CHKERRQ(ierr);
1098   ierr = MatDenseGetArray(X,&x);CHKERRQ(ierr);
1099 
1100   tmp  = a->solve_work;
1101   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1102   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1103 
1104   for (neq=0; neq<B->cmap->n; neq++) {
1105     /* forward solve the lower triangular */
1106     tmp[0] = b[r[0]];
1107     v      = aa;
1108     vi     = aj;
1109     for (i=1; i<n; i++) {
1110       nz  = ai[i+1] - ai[i];
1111       sum = b[r[i]];
1112       PetscSparseDenseMinusDot(sum,tmp,v,vi,nz);
1113       tmp[i] = sum;
1114       v     += nz; vi += nz;
1115     }
1116 
1117     /* backward solve the upper triangular */
1118     for (i=n-1; i>=0; i--) {
1119       v   = aa + adiag[i+1]+1;
1120       vi  = aj + adiag[i+1]+1;
1121       nz  = adiag[i]-adiag[i+1]-1;
1122       sum = tmp[i];
1123       PetscSparseDenseMinusDot(sum,tmp,v,vi,nz);
1124       x[c[i]] = tmp[i] = sum*v[nz]; /* v[nz] = aa[adiag[i]] */
1125     }
1126 
1127     b += n;
1128     x += n;
1129   }
1130   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1131   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1132   ierr = MatDenseRestoreArrayRead(B,&b);CHKERRQ(ierr);
1133   ierr = MatDenseRestoreArray(X,&x);CHKERRQ(ierr);
1134   ierr = PetscLogFlops(B->cmap->n*(2.0*a->nz - n));CHKERRQ(ierr);
1135   PetscFunctionReturn(0);
1136 }
1137 
1138 PetscErrorCode MatSolve_SeqAIJ_InplaceWithPerm(Mat A,Vec bb,Vec xx)
1139 {
1140   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1141   IS                iscol = a->col,isrow = a->row;
1142   PetscErrorCode    ierr;
1143   const PetscInt    *r,*c,*rout,*cout;
1144   PetscInt          i, n = A->rmap->n,*vi,*ai = a->i,*aj = a->j;
1145   PetscInt          nz,row;
1146   PetscScalar       *x,*tmp,*tmps,sum;
1147   const PetscScalar *b;
1148   const MatScalar   *aa = a->a,*v;
1149 
1150   PetscFunctionBegin;
1151   if (!n) PetscFunctionReturn(0);
1152 
1153   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1154   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
1155   tmp  = a->solve_work;
1156 
1157   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1158   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout + (n-1);
1159 
1160   /* forward solve the lower triangular */
1161   tmp[0] = b[*r++];
1162   tmps   = tmp;
1163   for (row=1; row<n; row++) {
1164     i   = rout[row]; /* permuted row */
1165     v   = aa + ai[i];
1166     vi  = aj + ai[i];
1167     nz  = a->diag[i] - ai[i];
1168     sum = b[*r++];
1169     PetscSparseDenseMinusDot(sum,tmps,v,vi,nz);
1170     tmp[row] = sum;
1171   }
1172 
1173   /* backward solve the upper triangular */
1174   for (row=n-1; row>=0; row--) {
1175     i   = rout[row]; /* permuted row */
1176     v   = aa + a->diag[i] + 1;
1177     vi  = aj + a->diag[i] + 1;
1178     nz  = ai[i+1] - a->diag[i] - 1;
1179     sum = tmp[row];
1180     PetscSparseDenseMinusDot(sum,tmps,v,vi,nz);
1181     x[*c--] = tmp[row] = sum*aa[a->diag[i]];
1182   }
1183 
1184   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1185   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1186   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1187   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
1188   ierr = PetscLogFlops(2.0*a->nz - A->cmap->n);CHKERRQ(ierr);
1189   PetscFunctionReturn(0);
1190 }
1191 
1192 /* ----------------------------------------------------------- */
1193 #include <../src/mat/impls/aij/seq/ftn-kernels/fsolve.h>
1194 PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering_inplace(Mat A,Vec bb,Vec xx)
1195 {
1196   Mat_SeqAIJ        *a = (Mat_SeqAIJ*)A->data;
1197   PetscErrorCode    ierr;
1198   PetscInt          n   = A->rmap->n;
1199   const PetscInt    *ai = a->i,*aj = a->j,*adiag = a->diag;
1200   PetscScalar       *x;
1201   const PetscScalar *b;
1202   const MatScalar   *aa = a->a;
1203 #if !defined(PETSC_USE_FORTRAN_KERNEL_SOLVEAIJ)
1204   PetscInt        adiag_i,i,nz,ai_i;
1205   const PetscInt  *vi;
1206   const MatScalar *v;
1207   PetscScalar     sum;
1208 #endif
1209 
1210   PetscFunctionBegin;
1211   if (!n) PetscFunctionReturn(0);
1212 
1213   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1214   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
1215 
1216 #if defined(PETSC_USE_FORTRAN_KERNEL_SOLVEAIJ)
1217   fortransolveaij_(&n,x,ai,aj,adiag,aa,b);
1218 #else
1219   /* forward solve the lower triangular */
1220   x[0] = b[0];
1221   for (i=1; i<n; i++) {
1222     ai_i = ai[i];
1223     v    = aa + ai_i;
1224     vi   = aj + ai_i;
1225     nz   = adiag[i] - ai_i;
1226     sum  = b[i];
1227     PetscSparseDenseMinusDot(sum,x,v,vi,nz);
1228     x[i] = sum;
1229   }
1230 
1231   /* backward solve the upper triangular */
1232   for (i=n-1; i>=0; i--) {
1233     adiag_i = adiag[i];
1234     v       = aa + adiag_i + 1;
1235     vi      = aj + adiag_i + 1;
1236     nz      = ai[i+1] - adiag_i - 1;
1237     sum     = x[i];
1238     PetscSparseDenseMinusDot(sum,x,v,vi,nz);
1239     x[i] = sum*aa[adiag_i];
1240   }
1241 #endif
1242   ierr = PetscLogFlops(2.0*a->nz - A->cmap->n);CHKERRQ(ierr);
1243   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1244   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
1245   PetscFunctionReturn(0);
1246 }
1247 
1248 PetscErrorCode MatSolveAdd_SeqAIJ_inplace(Mat A,Vec bb,Vec yy,Vec xx)
1249 {
1250   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1251   IS                iscol = a->col,isrow = a->row;
1252   PetscErrorCode    ierr;
1253   PetscInt          i, n = A->rmap->n,j;
1254   PetscInt          nz;
1255   const PetscInt    *rout,*cout,*r,*c,*vi,*ai = a->i,*aj = a->j;
1256   PetscScalar       *x,*tmp,sum;
1257   const PetscScalar *b;
1258   const MatScalar   *aa = a->a,*v;
1259 
1260   PetscFunctionBegin;
1261   if (yy != xx) {ierr = VecCopy(yy,xx);CHKERRQ(ierr);}
1262 
1263   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1264   ierr = VecGetArray(xx,&x);CHKERRQ(ierr);
1265   tmp  = a->solve_work;
1266 
1267   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1268   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout + (n-1);
1269 
1270   /* forward solve the lower triangular */
1271   tmp[0] = b[*r++];
1272   for (i=1; i<n; i++) {
1273     v   = aa + ai[i];
1274     vi  = aj + ai[i];
1275     nz  = a->diag[i] - ai[i];
1276     sum = b[*r++];
1277     for (j=0; j<nz; j++) sum -= v[j]*tmp[vi[j]];
1278     tmp[i] = sum;
1279   }
1280 
1281   /* backward solve the upper triangular */
1282   for (i=n-1; i>=0; i--) {
1283     v   = aa + a->diag[i] + 1;
1284     vi  = aj + a->diag[i] + 1;
1285     nz  = ai[i+1] - a->diag[i] - 1;
1286     sum = tmp[i];
1287     for (j=0; j<nz; j++) sum -= v[j]*tmp[vi[j]];
1288     tmp[i]   = sum*aa[a->diag[i]];
1289     x[*c--] += tmp[i];
1290   }
1291 
1292   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1293   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1294   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1295   ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr);
1296   ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr);
1297   PetscFunctionReturn(0);
1298 }
1299 
1300 PetscErrorCode MatSolveAdd_SeqAIJ(Mat A,Vec bb,Vec yy,Vec xx)
1301 {
1302   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1303   IS                iscol = a->col,isrow = a->row;
1304   PetscErrorCode    ierr;
1305   PetscInt          i, n = A->rmap->n,j;
1306   PetscInt          nz;
1307   const PetscInt    *rout,*cout,*r,*c,*vi,*ai = a->i,*aj = a->j,*adiag = a->diag;
1308   PetscScalar       *x,*tmp,sum;
1309   const PetscScalar *b;
1310   const MatScalar   *aa = a->a,*v;
1311 
1312   PetscFunctionBegin;
1313   if (yy != xx) {ierr = VecCopy(yy,xx);CHKERRQ(ierr);}
1314 
1315   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1316   ierr = VecGetArray(xx,&x);CHKERRQ(ierr);
1317   tmp  = a->solve_work;
1318 
1319   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1320   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1321 
1322   /* forward solve the lower triangular */
1323   tmp[0] = b[r[0]];
1324   v      = aa;
1325   vi     = aj;
1326   for (i=1; i<n; i++) {
1327     nz  = ai[i+1] - ai[i];
1328     sum = b[r[i]];
1329     for (j=0; j<nz; j++) sum -= v[j]*tmp[vi[j]];
1330     tmp[i] = sum;
1331     v     += nz;
1332     vi    += nz;
1333   }
1334 
1335   /* backward solve the upper triangular */
1336   v  = aa + adiag[n-1];
1337   vi = aj + adiag[n-1];
1338   for (i=n-1; i>=0; i--) {
1339     nz  = adiag[i] - adiag[i+1] - 1;
1340     sum = tmp[i];
1341     for (j=0; j<nz; j++) sum -= v[j]*tmp[vi[j]];
1342     tmp[i]   = sum*v[nz];
1343     x[c[i]] += tmp[i];
1344     v       += nz+1; vi += nz+1;
1345   }
1346 
1347   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1348   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1349   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1350   ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr);
1351   ierr = PetscLogFlops(2.0*a->nz);CHKERRQ(ierr);
1352   PetscFunctionReturn(0);
1353 }
1354 
1355 PetscErrorCode MatSolveTranspose_SeqAIJ_inplace(Mat A,Vec bb,Vec xx)
1356 {
1357   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1358   IS                iscol = a->col,isrow = a->row;
1359   PetscErrorCode    ierr;
1360   const PetscInt    *rout,*cout,*r,*c,*diag = a->diag,*ai = a->i,*aj = a->j,*vi;
1361   PetscInt          i,n = A->rmap->n,j;
1362   PetscInt          nz;
1363   PetscScalar       *x,*tmp,s1;
1364   const MatScalar   *aa = a->a,*v;
1365   const PetscScalar *b;
1366 
1367   PetscFunctionBegin;
1368   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1369   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
1370   tmp  = a->solve_work;
1371 
1372   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1373   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1374 
1375   /* copy the b into temp work space according to permutation */
1376   for (i=0; i<n; i++) tmp[i] = b[c[i]];
1377 
1378   /* forward solve the U^T */
1379   for (i=0; i<n; i++) {
1380     v   = aa + diag[i];
1381     vi  = aj + diag[i] + 1;
1382     nz  = ai[i+1] - diag[i] - 1;
1383     s1  = tmp[i];
1384     s1 *= (*v++);  /* multiply by inverse of diagonal entry */
1385     for (j=0; j<nz; j++) tmp[vi[j]] -= s1*v[j];
1386     tmp[i] = s1;
1387   }
1388 
1389   /* backward solve the L^T */
1390   for (i=n-1; i>=0; i--) {
1391     v  = aa + diag[i] - 1;
1392     vi = aj + diag[i] - 1;
1393     nz = diag[i] - ai[i];
1394     s1 = tmp[i];
1395     for (j=0; j>-nz; j--) tmp[vi[j]] -= s1*v[j];
1396   }
1397 
1398   /* copy tmp into x according to permutation */
1399   for (i=0; i<n; i++) x[r[i]] = tmp[i];
1400 
1401   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1402   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1403   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1404   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
1405 
1406   ierr = PetscLogFlops(2.0*a->nz-A->cmap->n);CHKERRQ(ierr);
1407   PetscFunctionReturn(0);
1408 }
1409 
1410 PetscErrorCode MatSolveTranspose_SeqAIJ(Mat A,Vec bb,Vec xx)
1411 {
1412   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1413   IS                iscol = a->col,isrow = a->row;
1414   PetscErrorCode    ierr;
1415   const PetscInt    *rout,*cout,*r,*c,*adiag = a->diag,*ai = a->i,*aj = a->j,*vi;
1416   PetscInt          i,n = A->rmap->n,j;
1417   PetscInt          nz;
1418   PetscScalar       *x,*tmp,s1;
1419   const MatScalar   *aa = a->a,*v;
1420   const PetscScalar *b;
1421 
1422   PetscFunctionBegin;
1423   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1424   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
1425   tmp  = a->solve_work;
1426 
1427   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1428   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1429 
1430   /* copy the b into temp work space according to permutation */
1431   for (i=0; i<n; i++) tmp[i] = b[c[i]];
1432 
1433   /* forward solve the U^T */
1434   for (i=0; i<n; i++) {
1435     v   = aa + adiag[i+1] + 1;
1436     vi  = aj + adiag[i+1] + 1;
1437     nz  = adiag[i] - adiag[i+1] - 1;
1438     s1  = tmp[i];
1439     s1 *= v[nz];  /* multiply by inverse of diagonal entry */
1440     for (j=0; j<nz; j++) tmp[vi[j]] -= s1*v[j];
1441     tmp[i] = s1;
1442   }
1443 
1444   /* backward solve the L^T */
1445   for (i=n-1; i>=0; i--) {
1446     v  = aa + ai[i];
1447     vi = aj + ai[i];
1448     nz = ai[i+1] - ai[i];
1449     s1 = tmp[i];
1450     for (j=0; j<nz; j++) tmp[vi[j]] -= s1*v[j];
1451   }
1452 
1453   /* copy tmp into x according to permutation */
1454   for (i=0; i<n; i++) x[r[i]] = tmp[i];
1455 
1456   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1457   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1458   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1459   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
1460 
1461   ierr = PetscLogFlops(2.0*a->nz-A->cmap->n);CHKERRQ(ierr);
1462   PetscFunctionReturn(0);
1463 }
1464 
1465 PetscErrorCode MatSolveTransposeAdd_SeqAIJ_inplace(Mat A,Vec bb,Vec zz,Vec xx)
1466 {
1467   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1468   IS                iscol = a->col,isrow = a->row;
1469   PetscErrorCode    ierr;
1470   const PetscInt    *rout,*cout,*r,*c,*diag = a->diag,*ai = a->i,*aj = a->j,*vi;
1471   PetscInt          i,n = A->rmap->n,j;
1472   PetscInt          nz;
1473   PetscScalar       *x,*tmp,s1;
1474   const MatScalar   *aa = a->a,*v;
1475   const PetscScalar *b;
1476 
1477   PetscFunctionBegin;
1478   if (zz != xx) {ierr = VecCopy(zz,xx);CHKERRQ(ierr);}
1479   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1480   ierr = VecGetArray(xx,&x);CHKERRQ(ierr);
1481   tmp  = a->solve_work;
1482 
1483   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1484   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1485 
1486   /* copy the b into temp work space according to permutation */
1487   for (i=0; i<n; i++) tmp[i] = b[c[i]];
1488 
1489   /* forward solve the U^T */
1490   for (i=0; i<n; i++) {
1491     v   = aa + diag[i];
1492     vi  = aj + diag[i] + 1;
1493     nz  = ai[i+1] - diag[i] - 1;
1494     s1  = tmp[i];
1495     s1 *= (*v++);  /* multiply by inverse of diagonal entry */
1496     for (j=0; j<nz; j++) tmp[vi[j]] -= s1*v[j];
1497     tmp[i] = s1;
1498   }
1499 
1500   /* backward solve the L^T */
1501   for (i=n-1; i>=0; i--) {
1502     v  = aa + diag[i] - 1;
1503     vi = aj + diag[i] - 1;
1504     nz = diag[i] - ai[i];
1505     s1 = tmp[i];
1506     for (j=0; j>-nz; j--) tmp[vi[j]] -= s1*v[j];
1507   }
1508 
1509   /* copy tmp into x according to permutation */
1510   for (i=0; i<n; i++) x[r[i]] += tmp[i];
1511 
1512   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1513   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1514   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1515   ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr);
1516 
1517   ierr = PetscLogFlops(2.0*a->nz-A->cmap->n);CHKERRQ(ierr);
1518   PetscFunctionReturn(0);
1519 }
1520 
1521 PetscErrorCode MatSolveTransposeAdd_SeqAIJ(Mat A,Vec bb,Vec zz,Vec xx)
1522 {
1523   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
1524   IS                iscol = a->col,isrow = a->row;
1525   PetscErrorCode    ierr;
1526   const PetscInt    *rout,*cout,*r,*c,*adiag = a->diag,*ai = a->i,*aj = a->j,*vi;
1527   PetscInt          i,n = A->rmap->n,j;
1528   PetscInt          nz;
1529   PetscScalar       *x,*tmp,s1;
1530   const MatScalar   *aa = a->a,*v;
1531   const PetscScalar *b;
1532 
1533   PetscFunctionBegin;
1534   if (zz != xx) {ierr = VecCopy(zz,xx);CHKERRQ(ierr);}
1535   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
1536   ierr = VecGetArray(xx,&x);CHKERRQ(ierr);
1537   tmp  = a->solve_work;
1538 
1539   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
1540   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
1541 
1542   /* copy the b into temp work space according to permutation */
1543   for (i=0; i<n; i++) tmp[i] = b[c[i]];
1544 
1545   /* forward solve the U^T */
1546   for (i=0; i<n; i++) {
1547     v   = aa + adiag[i+1] + 1;
1548     vi  = aj + adiag[i+1] + 1;
1549     nz  = adiag[i] - adiag[i+1] - 1;
1550     s1  = tmp[i];
1551     s1 *= v[nz];  /* multiply by inverse of diagonal entry */
1552     for (j=0; j<nz; j++) tmp[vi[j]] -= s1*v[j];
1553     tmp[i] = s1;
1554   }
1555 
1556 
1557   /* backward solve the L^T */
1558   for (i=n-1; i>=0; i--) {
1559     v  = aa + ai[i];
1560     vi = aj + ai[i];
1561     nz = ai[i+1] - ai[i];
1562     s1 = tmp[i];
1563     for (j=0; j<nz; j++) tmp[vi[j]] -= s1*v[j];
1564   }
1565 
1566   /* copy tmp into x according to permutation */
1567   for (i=0; i<n; i++) x[r[i]] += tmp[i];
1568 
1569   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
1570   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
1571   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
1572   ierr = VecRestoreArray(xx,&x);CHKERRQ(ierr);
1573 
1574   ierr = PetscLogFlops(2.0*a->nz-A->cmap->n);CHKERRQ(ierr);
1575   PetscFunctionReturn(0);
1576 }
1577 
1578 /* ----------------------------------------------------------------*/
1579 
1580 /*
1581    ilu() under revised new data structure.
1582    Factored arrays bj and ba are stored as
1583      L(0,:), L(1,:), ...,L(n-1,:),  U(n-1,:),...,U(i,:),U(i-1,:),...,U(0,:)
1584 
1585    bi=fact->i is an array of size n+1, in which
1586    bi+
1587      bi[i]:  points to 1st entry of L(i,:),i=0,...,n-1
1588      bi[n]:  points to L(n-1,n-1)+1
1589 
1590   bdiag=fact->diag is an array of size n+1,in which
1591      bdiag[i]: points to diagonal of U(i,:), i=0,...,n-1
1592      bdiag[n]: points to entry of U(n-1,0)-1
1593 
1594    U(i,:) contains bdiag[i] as its last entry, i.e.,
1595     U(i,:) = (u[i,i+1],...,u[i,n-1],diag[i])
1596 */
1597 PetscErrorCode MatILUFactorSymbolic_SeqAIJ_ilu0(Mat fact,Mat A,IS isrow,IS iscol,const MatFactorInfo *info)
1598 {
1599   Mat_SeqAIJ     *a = (Mat_SeqAIJ*)A->data,*b;
1600   PetscErrorCode ierr;
1601   const PetscInt n=A->rmap->n,*ai=a->i,*aj,*adiag=a->diag;
1602   PetscInt       i,j,k=0,nz,*bi,*bj,*bdiag;
1603   IS             isicol;
1604 
1605   PetscFunctionBegin;
1606   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
1607   ierr = MatDuplicateNoCreate_SeqAIJ(fact,A,MAT_DO_NOT_COPY_VALUES,PETSC_FALSE);CHKERRQ(ierr);
1608   b    = (Mat_SeqAIJ*)(fact)->data;
1609 
1610   /* allocate matrix arrays for new data structure */
1611   ierr = PetscMalloc3(ai[n]+1,&b->a,ai[n]+1,&b->j,n+1,&b->i);CHKERRQ(ierr);
1612   ierr = PetscLogObjectMemory((PetscObject)fact,ai[n]*(sizeof(PetscScalar)+sizeof(PetscInt))+(n+1)*sizeof(PetscInt));CHKERRQ(ierr);
1613 
1614   b->singlemalloc = PETSC_TRUE;
1615   if (!b->diag) {
1616     ierr = PetscMalloc1(n+1,&b->diag);CHKERRQ(ierr);
1617     ierr = PetscLogObjectMemory((PetscObject)fact,(n+1)*sizeof(PetscInt));CHKERRQ(ierr);
1618   }
1619   bdiag = b->diag;
1620 
1621   if (n > 0) {
1622     ierr = PetscArrayzero(b->a,ai[n]);CHKERRQ(ierr);
1623   }
1624 
1625   /* set bi and bj with new data structure */
1626   bi = b->i;
1627   bj = b->j;
1628 
1629   /* L part */
1630   bi[0] = 0;
1631   for (i=0; i<n; i++) {
1632     nz      = adiag[i] - ai[i];
1633     bi[i+1] = bi[i] + nz;
1634     aj      = a->j + ai[i];
1635     for (j=0; j<nz; j++) {
1636       /*   *bj = aj[j]; bj++; */
1637       bj[k++] = aj[j];
1638     }
1639   }
1640 
1641   /* U part */
1642   bdiag[n] = bi[n]-1;
1643   for (i=n-1; i>=0; i--) {
1644     nz = ai[i+1] - adiag[i] - 1;
1645     aj = a->j + adiag[i] + 1;
1646     for (j=0; j<nz; j++) {
1647       /*      *bj = aj[j]; bj++; */
1648       bj[k++] = aj[j];
1649     }
1650     /* diag[i] */
1651     /*    *bj = i; bj++; */
1652     bj[k++]  = i;
1653     bdiag[i] = bdiag[i+1] + nz + 1;
1654   }
1655 
1656   fact->factortype             = MAT_FACTOR_ILU;
1657   fact->info.factor_mallocs    = 0;
1658   fact->info.fill_ratio_given  = info->fill;
1659   fact->info.fill_ratio_needed = 1.0;
1660   fact->ops->lufactornumeric   = MatLUFactorNumeric_SeqAIJ;
1661   ierr = MatSeqAIJCheckInode_FactorLU(fact);CHKERRQ(ierr);
1662 
1663   b       = (Mat_SeqAIJ*)(fact)->data;
1664   b->row  = isrow;
1665   b->col  = iscol;
1666   b->icol = isicol;
1667   ierr    = PetscMalloc1(fact->rmap->n+1,&b->solve_work);CHKERRQ(ierr);
1668   ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
1669   ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
1670   PetscFunctionReturn(0);
1671 }
1672 
1673 PetscErrorCode MatILUFactorSymbolic_SeqAIJ(Mat fact,Mat A,IS isrow,IS iscol,const MatFactorInfo *info)
1674 {
1675   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data,*b;
1676   IS                 isicol;
1677   PetscErrorCode     ierr;
1678   const PetscInt     *r,*ic;
1679   PetscInt           n=A->rmap->n,*ai=a->i,*aj=a->j;
1680   PetscInt           *bi,*cols,nnz,*cols_lvl;
1681   PetscInt           *bdiag,prow,fm,nzbd,reallocs=0,dcount=0;
1682   PetscInt           i,levels,diagonal_fill;
1683   PetscBool          col_identity,row_identity,missing;
1684   PetscReal          f;
1685   PetscInt           nlnk,*lnk,*lnk_lvl=NULL;
1686   PetscBT            lnkbt;
1687   PetscInt           nzi,*bj,**bj_ptr,**bjlvl_ptr;
1688   PetscFreeSpaceList free_space    =NULL,current_space=NULL;
1689   PetscFreeSpaceList free_space_lvl=NULL,current_space_lvl=NULL;
1690 
1691   PetscFunctionBegin;
1692   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
1693   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
1694   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
1695 
1696   levels = (PetscInt)info->levels;
1697   ierr   = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
1698   ierr   = ISIdentity(iscol,&col_identity);CHKERRQ(ierr);
1699   if (!levels && row_identity && col_identity) {
1700     /* special case: ilu(0) with natural ordering */
1701     ierr = MatILUFactorSymbolic_SeqAIJ_ilu0(fact,A,isrow,iscol,info);CHKERRQ(ierr);
1702     if (a->inode.size) {
1703       fact->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_Inode;
1704     }
1705     PetscFunctionReturn(0);
1706   }
1707 
1708   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
1709   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
1710   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
1711 
1712   /* get new row and diagonal pointers, must be allocated separately because they will be given to the Mat_SeqAIJ and freed separately */
1713   ierr  = PetscMalloc1(n+1,&bi);CHKERRQ(ierr);
1714   ierr  = PetscMalloc1(n+1,&bdiag);CHKERRQ(ierr);
1715   bi[0] = bdiag[0] = 0;
1716   ierr  = PetscMalloc2(n,&bj_ptr,n,&bjlvl_ptr);CHKERRQ(ierr);
1717 
1718   /* create a linked list for storing column indices of the active row */
1719   nlnk = n + 1;
1720   ierr = PetscIncompleteLLCreate(n,n,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
1721 
1722   /* initial FreeSpace size is f*(ai[n]+1) */
1723   f                 = info->fill;
1724   diagonal_fill     = (PetscInt)info->diagonal_fill;
1725   ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(f,ai[n]+1),&free_space);CHKERRQ(ierr);
1726   current_space     = free_space;
1727   ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(f,ai[n]+1),&free_space_lvl);CHKERRQ(ierr);
1728   current_space_lvl = free_space_lvl;
1729   for (i=0; i<n; i++) {
1730     nzi = 0;
1731     /* copy current row into linked list */
1732     nnz = ai[r[i]+1] - ai[r[i]];
1733     if (!nnz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",r[i],i);
1734     cols   = aj + ai[r[i]];
1735     lnk[i] = -1; /* marker to indicate if diagonal exists */
1736     ierr   = PetscIncompleteLLInit(nnz,cols,n,ic,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
1737     nzi   += nlnk;
1738 
1739     /* make sure diagonal entry is included */
1740     if (diagonal_fill && lnk[i] == -1) {
1741       fm = n;
1742       while (lnk[fm] < i) fm = lnk[fm];
1743       lnk[i]     = lnk[fm]; /* insert diagonal into linked list */
1744       lnk[fm]    = i;
1745       lnk_lvl[i] = 0;
1746       nzi++; dcount++;
1747     }
1748 
1749     /* add pivot rows into the active row */
1750     nzbd = 0;
1751     prow = lnk[n];
1752     while (prow < i) {
1753       nnz      = bdiag[prow];
1754       cols     = bj_ptr[prow] + nnz + 1;
1755       cols_lvl = bjlvl_ptr[prow] + nnz + 1;
1756       nnz      = bi[prow+1] - bi[prow] - nnz - 1;
1757       ierr     = PetscILULLAddSorted(nnz,cols,levels,cols_lvl,prow,nlnk,lnk,lnk_lvl,lnkbt,prow);CHKERRQ(ierr);
1758       nzi     += nlnk;
1759       prow     = lnk[prow];
1760       nzbd++;
1761     }
1762     bdiag[i] = nzbd;
1763     bi[i+1]  = bi[i] + nzi;
1764     /* if free space is not available, make more free space */
1765     if (current_space->local_remaining<nzi) {
1766       nnz  = PetscIntMultTruncate(2,PetscIntMultTruncate(nzi,n - i)); /* estimated and max additional space needed */
1767       ierr = PetscFreeSpaceGet(nnz,&current_space);CHKERRQ(ierr);
1768       ierr = PetscFreeSpaceGet(nnz,&current_space_lvl);CHKERRQ(ierr);
1769       reallocs++;
1770     }
1771 
1772     /* copy data into free_space and free_space_lvl, then initialize lnk */
1773     ierr         = PetscIncompleteLLClean(n,n,nzi,lnk,lnk_lvl,current_space->array,current_space_lvl->array,lnkbt);CHKERRQ(ierr);
1774     bj_ptr[i]    = current_space->array;
1775     bjlvl_ptr[i] = current_space_lvl->array;
1776 
1777     /* make sure the active row i has diagonal entry */
1778     if (*(bj_ptr[i]+bdiag[i]) != i) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Row %D has missing diagonal in factored matrix\ntry running with -pc_factor_nonzeros_along_diagonal or -pc_factor_diagonal_fill",i);
1779 
1780     current_space->array               += nzi;
1781     current_space->local_used          += nzi;
1782     current_space->local_remaining     -= nzi;
1783     current_space_lvl->array           += nzi;
1784     current_space_lvl->local_used      += nzi;
1785     current_space_lvl->local_remaining -= nzi;
1786   }
1787 
1788   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
1789   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
1790   /* copy free_space into bj and free free_space; set bi, bj, bdiag in new datastructure; */
1791   ierr = PetscMalloc1(bi[n]+1,&bj);CHKERRQ(ierr);
1792   ierr = PetscFreeSpaceContiguous_LU(&free_space,bj,n,bi,bdiag);CHKERRQ(ierr);
1793 
1794   ierr = PetscIncompleteLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
1795   ierr = PetscFreeSpaceDestroy(free_space_lvl);CHKERRQ(ierr);
1796   ierr = PetscFree2(bj_ptr,bjlvl_ptr);CHKERRQ(ierr);
1797 
1798 #if defined(PETSC_USE_INFO)
1799   {
1800     PetscReal af = ((PetscReal)(bdiag[0]+1))/((PetscReal)ai[n]);
1801     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)f,(double)af);CHKERRQ(ierr);
1802     ierr = PetscInfo1(A,"Run with -[sub_]pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
1803     ierr = PetscInfo1(A,"PCFactorSetFill([sub]pc,%g);\n",(double)af);CHKERRQ(ierr);
1804     ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr);
1805     if (diagonal_fill) {
1806       ierr = PetscInfo1(A,"Detected and replaced %D missing diagonals\n",dcount);CHKERRQ(ierr);
1807     }
1808   }
1809 #endif
1810   /* put together the new matrix */
1811   ierr = MatSeqAIJSetPreallocation_SeqAIJ(fact,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
1812   ierr = PetscLogObjectParent((PetscObject)fact,(PetscObject)isicol);CHKERRQ(ierr);
1813   b    = (Mat_SeqAIJ*)(fact)->data;
1814 
1815   b->free_a       = PETSC_TRUE;
1816   b->free_ij      = PETSC_TRUE;
1817   b->singlemalloc = PETSC_FALSE;
1818 
1819   ierr = PetscMalloc1(bdiag[0]+1,&b->a);CHKERRQ(ierr);
1820 
1821   b->j    = bj;
1822   b->i    = bi;
1823   b->diag = bdiag;
1824   b->ilen = 0;
1825   b->imax = 0;
1826   b->row  = isrow;
1827   b->col  = iscol;
1828   ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
1829   ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
1830   b->icol = isicol;
1831 
1832   ierr = PetscMalloc1(n+1,&b->solve_work);CHKERRQ(ierr);
1833   /* In b structure:  Free imax, ilen, old a, old j.
1834      Allocate bdiag, solve_work, new a, new j */
1835   ierr     = PetscLogObjectMemory((PetscObject)fact,(bdiag[0]+1)*(sizeof(PetscInt)+sizeof(PetscScalar)));CHKERRQ(ierr);
1836   b->maxnz = b->nz = bdiag[0]+1;
1837 
1838   (fact)->info.factor_mallocs    = reallocs;
1839   (fact)->info.fill_ratio_given  = f;
1840   (fact)->info.fill_ratio_needed = ((PetscReal)(bdiag[0]+1))/((PetscReal)ai[n]);
1841   (fact)->ops->lufactornumeric   = MatLUFactorNumeric_SeqAIJ;
1842   if (a->inode.size) {
1843     (fact)->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_Inode;
1844   }
1845   ierr = MatSeqAIJCheckInode_FactorLU(fact);CHKERRQ(ierr);
1846   PetscFunctionReturn(0);
1847 }
1848 
1849 PetscErrorCode MatILUFactorSymbolic_SeqAIJ_inplace(Mat fact,Mat A,IS isrow,IS iscol,const MatFactorInfo *info)
1850 {
1851   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data,*b;
1852   IS                 isicol;
1853   PetscErrorCode     ierr;
1854   const PetscInt     *r,*ic;
1855   PetscInt           n=A->rmap->n,*ai=a->i,*aj=a->j;
1856   PetscInt           *bi,*cols,nnz,*cols_lvl;
1857   PetscInt           *bdiag,prow,fm,nzbd,reallocs=0,dcount=0;
1858   PetscInt           i,levels,diagonal_fill;
1859   PetscBool          col_identity,row_identity;
1860   PetscReal          f;
1861   PetscInt           nlnk,*lnk,*lnk_lvl=NULL;
1862   PetscBT            lnkbt;
1863   PetscInt           nzi,*bj,**bj_ptr,**bjlvl_ptr;
1864   PetscFreeSpaceList free_space    =NULL,current_space=NULL;
1865   PetscFreeSpaceList free_space_lvl=NULL,current_space_lvl=NULL;
1866   PetscBool          missing;
1867 
1868   PetscFunctionBegin;
1869   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
1870   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
1871   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
1872 
1873   f             = info->fill;
1874   levels        = (PetscInt)info->levels;
1875   diagonal_fill = (PetscInt)info->diagonal_fill;
1876 
1877   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
1878 
1879   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
1880   ierr = ISIdentity(iscol,&col_identity);CHKERRQ(ierr);
1881   if (!levels && row_identity && col_identity) { /* special case: ilu(0) with natural ordering */
1882     ierr = MatDuplicateNoCreate_SeqAIJ(fact,A,MAT_DO_NOT_COPY_VALUES,PETSC_TRUE);CHKERRQ(ierr);
1883 
1884     (fact)->ops->lufactornumeric =  MatLUFactorNumeric_SeqAIJ_inplace;
1885     if (a->inode.size) {
1886       (fact)->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_Inode_inplace;
1887     }
1888     fact->factortype               = MAT_FACTOR_ILU;
1889     (fact)->info.factor_mallocs    = 0;
1890     (fact)->info.fill_ratio_given  = info->fill;
1891     (fact)->info.fill_ratio_needed = 1.0;
1892 
1893     b    = (Mat_SeqAIJ*)(fact)->data;
1894     b->row  = isrow;
1895     b->col  = iscol;
1896     b->icol = isicol;
1897     ierr    = PetscMalloc1((fact)->rmap->n+1,&b->solve_work);CHKERRQ(ierr);
1898     ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
1899     ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
1900     PetscFunctionReturn(0);
1901   }
1902 
1903   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
1904   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
1905 
1906   /* get new row and diagonal pointers, must be allocated separately because they will be given to the Mat_SeqAIJ and freed separately */
1907   ierr  = PetscMalloc1(n+1,&bi);CHKERRQ(ierr);
1908   ierr  = PetscMalloc1(n+1,&bdiag);CHKERRQ(ierr);
1909   bi[0] = bdiag[0] = 0;
1910 
1911   ierr = PetscMalloc2(n,&bj_ptr,n,&bjlvl_ptr);CHKERRQ(ierr);
1912 
1913   /* create a linked list for storing column indices of the active row */
1914   nlnk = n + 1;
1915   ierr = PetscIncompleteLLCreate(n,n,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
1916 
1917   /* initial FreeSpace size is f*(ai[n]+1) */
1918   ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(f,ai[n]+1),&free_space);CHKERRQ(ierr);
1919   current_space     = free_space;
1920   ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(f,ai[n]+1),&free_space_lvl);CHKERRQ(ierr);
1921   current_space_lvl = free_space_lvl;
1922 
1923   for (i=0; i<n; i++) {
1924     nzi = 0;
1925     /* copy current row into linked list */
1926     nnz = ai[r[i]+1] - ai[r[i]];
1927     if (!nnz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",r[i],i);
1928     cols   = aj + ai[r[i]];
1929     lnk[i] = -1; /* marker to indicate if diagonal exists */
1930     ierr   = PetscIncompleteLLInit(nnz,cols,n,ic,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
1931     nzi   += nlnk;
1932 
1933     /* make sure diagonal entry is included */
1934     if (diagonal_fill && lnk[i] == -1) {
1935       fm = n;
1936       while (lnk[fm] < i) fm = lnk[fm];
1937       lnk[i]     = lnk[fm]; /* insert diagonal into linked list */
1938       lnk[fm]    = i;
1939       lnk_lvl[i] = 0;
1940       nzi++; dcount++;
1941     }
1942 
1943     /* add pivot rows into the active row */
1944     nzbd = 0;
1945     prow = lnk[n];
1946     while (prow < i) {
1947       nnz      = bdiag[prow];
1948       cols     = bj_ptr[prow] + nnz + 1;
1949       cols_lvl = bjlvl_ptr[prow] + nnz + 1;
1950       nnz      = bi[prow+1] - bi[prow] - nnz - 1;
1951       ierr     = PetscILULLAddSorted(nnz,cols,levels,cols_lvl,prow,nlnk,lnk,lnk_lvl,lnkbt,prow);CHKERRQ(ierr);
1952       nzi     += nlnk;
1953       prow     = lnk[prow];
1954       nzbd++;
1955     }
1956     bdiag[i] = nzbd;
1957     bi[i+1]  = bi[i] + nzi;
1958 
1959     /* if free space is not available, make more free space */
1960     if (current_space->local_remaining<nzi) {
1961       nnz  = PetscIntMultTruncate(nzi,n - i); /* estimated and max additional space needed */
1962       ierr = PetscFreeSpaceGet(nnz,&current_space);CHKERRQ(ierr);
1963       ierr = PetscFreeSpaceGet(nnz,&current_space_lvl);CHKERRQ(ierr);
1964       reallocs++;
1965     }
1966 
1967     /* copy data into free_space and free_space_lvl, then initialize lnk */
1968     ierr         = PetscIncompleteLLClean(n,n,nzi,lnk,lnk_lvl,current_space->array,current_space_lvl->array,lnkbt);CHKERRQ(ierr);
1969     bj_ptr[i]    = current_space->array;
1970     bjlvl_ptr[i] = current_space_lvl->array;
1971 
1972     /* make sure the active row i has diagonal entry */
1973     if (*(bj_ptr[i]+bdiag[i]) != i) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Row %D has missing diagonal in factored matrix\ntry running with -pc_factor_nonzeros_along_diagonal or -pc_factor_diagonal_fill",i);
1974 
1975     current_space->array               += nzi;
1976     current_space->local_used          += nzi;
1977     current_space->local_remaining     -= nzi;
1978     current_space_lvl->array           += nzi;
1979     current_space_lvl->local_used      += nzi;
1980     current_space_lvl->local_remaining -= nzi;
1981   }
1982 
1983   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
1984   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
1985 
1986   /* destroy list of free space and other temporary arrays */
1987   ierr = PetscMalloc1(bi[n]+1,&bj);CHKERRQ(ierr);
1988   ierr = PetscFreeSpaceContiguous(&free_space,bj);CHKERRQ(ierr); /* copy free_space -> bj */
1989   ierr = PetscIncompleteLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
1990   ierr = PetscFreeSpaceDestroy(free_space_lvl);CHKERRQ(ierr);
1991   ierr = PetscFree2(bj_ptr,bjlvl_ptr);CHKERRQ(ierr);
1992 
1993 #if defined(PETSC_USE_INFO)
1994   {
1995     PetscReal af = ((PetscReal)bi[n])/((PetscReal)ai[n]);
1996     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)f,(double)af);CHKERRQ(ierr);
1997     ierr = PetscInfo1(A,"Run with -[sub_]pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
1998     ierr = PetscInfo1(A,"PCFactorSetFill([sub]pc,%g);\n",(double)af);CHKERRQ(ierr);
1999     ierr = PetscInfo(A,"for best performance.\n");CHKERRQ(ierr);
2000     if (diagonal_fill) {
2001       ierr = PetscInfo1(A,"Detected and replaced %D missing diagonals\n",dcount);CHKERRQ(ierr);
2002     }
2003   }
2004 #endif
2005 
2006   /* put together the new matrix */
2007   ierr = MatSeqAIJSetPreallocation_SeqAIJ(fact,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
2008   ierr = PetscLogObjectParent((PetscObject)fact,(PetscObject)isicol);CHKERRQ(ierr);
2009   b    = (Mat_SeqAIJ*)(fact)->data;
2010 
2011   b->free_a       = PETSC_TRUE;
2012   b->free_ij      = PETSC_TRUE;
2013   b->singlemalloc = PETSC_FALSE;
2014 
2015   ierr = PetscMalloc1(bi[n],&b->a);CHKERRQ(ierr);
2016   b->j = bj;
2017   b->i = bi;
2018   for (i=0; i<n; i++) bdiag[i] += bi[i];
2019   b->diag = bdiag;
2020   b->ilen = 0;
2021   b->imax = 0;
2022   b->row  = isrow;
2023   b->col  = iscol;
2024   ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
2025   ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
2026   b->icol = isicol;
2027   ierr    = PetscMalloc1(n+1,&b->solve_work);CHKERRQ(ierr);
2028   /* In b structure:  Free imax, ilen, old a, old j.
2029      Allocate bdiag, solve_work, new a, new j */
2030   ierr     = PetscLogObjectMemory((PetscObject)fact,(bi[n]-n) * (sizeof(PetscInt)+sizeof(PetscScalar)));CHKERRQ(ierr);
2031   b->maxnz = b->nz = bi[n];
2032 
2033   (fact)->info.factor_mallocs    = reallocs;
2034   (fact)->info.fill_ratio_given  = f;
2035   (fact)->info.fill_ratio_needed = ((PetscReal)bi[n])/((PetscReal)ai[n]);
2036   (fact)->ops->lufactornumeric   =  MatLUFactorNumeric_SeqAIJ_inplace;
2037   if (a->inode.size) {
2038     (fact)->ops->lufactornumeric = MatLUFactorNumeric_SeqAIJ_Inode_inplace;
2039   }
2040   PetscFunctionReturn(0);
2041 }
2042 
2043 PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ(Mat B,Mat A,const MatFactorInfo *info)
2044 {
2045   Mat            C = B;
2046   Mat_SeqAIJ     *a=(Mat_SeqAIJ*)A->data;
2047   Mat_SeqSBAIJ   *b=(Mat_SeqSBAIJ*)C->data;
2048   IS             ip=b->row,iip = b->icol;
2049   PetscErrorCode ierr;
2050   const PetscInt *rip,*riip;
2051   PetscInt       i,j,mbs=A->rmap->n,*bi=b->i,*bj=b->j,*bdiag=b->diag,*bjtmp;
2052   PetscInt       *ai=a->i,*aj=a->j;
2053   PetscInt       k,jmin,jmax,*c2r,*il,col,nexti,ili,nz;
2054   MatScalar      *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi;
2055   PetscBool      perm_identity;
2056   FactorShiftCtx sctx;
2057   PetscReal      rs;
2058   MatScalar      d,*v;
2059 
2060   PetscFunctionBegin;
2061   /* MatPivotSetUp(): initialize shift context sctx */
2062   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
2063 
2064   if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
2065     sctx.shift_top = info->zeropivot;
2066     for (i=0; i<mbs; i++) {
2067       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
2068       d  = (aa)[a->diag[i]];
2069       rs = -PetscAbsScalar(d) - PetscRealPart(d);
2070       v  = aa+ai[i];
2071       nz = ai[i+1] - ai[i];
2072       for (j=0; j<nz; j++) rs += PetscAbsScalar(v[j]);
2073       if (rs>sctx.shift_top) sctx.shift_top = rs;
2074     }
2075     sctx.shift_top *= 1.1;
2076     sctx.nshift_max = 5;
2077     sctx.shift_lo   = 0.;
2078     sctx.shift_hi   = 1.;
2079   }
2080 
2081   ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr);
2082   ierr = ISGetIndices(iip,&riip);CHKERRQ(ierr);
2083 
2084   /* allocate working arrays
2085      c2r: linked list, keep track of pivot rows for a given column. c2r[col]: head of the list for a given col
2086      il:  for active k row, il[i] gives the index of the 1st nonzero entry in U[i,k:n-1] in bj and ba arrays
2087   */
2088   ierr = PetscMalloc3(mbs,&rtmp,mbs,&il,mbs,&c2r);CHKERRQ(ierr);
2089 
2090   do {
2091     sctx.newshift = PETSC_FALSE;
2092 
2093     for (i=0; i<mbs; i++) c2r[i] = mbs;
2094     if (mbs) il[0] = 0;
2095 
2096     for (k = 0; k<mbs; k++) {
2097       /* zero rtmp */
2098       nz    = bi[k+1] - bi[k];
2099       bjtmp = bj + bi[k];
2100       for (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0;
2101 
2102       /* load in initial unfactored row */
2103       bval = ba + bi[k];
2104       jmin = ai[rip[k]]; jmax = ai[rip[k]+1];
2105       for (j = jmin; j < jmax; j++) {
2106         col = riip[aj[j]];
2107         if (col >= k) { /* only take upper triangular entry */
2108           rtmp[col] = aa[j];
2109           *bval++   = 0.0; /* for in-place factorization */
2110         }
2111       }
2112       /* shift the diagonal of the matrix: ZeropivotApply() */
2113       rtmp[k] += sctx.shift_amount;  /* shift the diagonal of the matrix */
2114 
2115       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
2116       dk = rtmp[k];
2117       i  = c2r[k]; /* first row to be added to k_th row  */
2118 
2119       while (i < k) {
2120         nexti = c2r[i]; /* next row to be added to k_th row */
2121 
2122         /* compute multiplier, update diag(k) and U(i,k) */
2123         ili     = il[i]; /* index of first nonzero element in U(i,k:bms-1) */
2124         uikdi   = -ba[ili]*ba[bdiag[i]]; /* diagonal(k) */
2125         dk     += uikdi*ba[ili]; /* update diag[k] */
2126         ba[ili] = uikdi; /* -U(i,k) */
2127 
2128         /* add multiple of row i to k-th row */
2129         jmin = ili + 1; jmax = bi[i+1];
2130         if (jmin < jmax) {
2131           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
2132           /* update il and c2r for row i */
2133           il[i] = jmin;
2134           j     = bj[jmin]; c2r[i] = c2r[j]; c2r[j] = i;
2135         }
2136         i = nexti;
2137       }
2138 
2139       /* copy data into U(k,:) */
2140       rs   = 0.0;
2141       jmin = bi[k]; jmax = bi[k+1]-1;
2142       if (jmin < jmax) {
2143         for (j=jmin; j<jmax; j++) {
2144           col = bj[j]; ba[j] = rtmp[col]; rs += PetscAbsScalar(ba[j]);
2145         }
2146         /* add the k-th row into il and c2r */
2147         il[k] = jmin;
2148         i     = bj[jmin]; c2r[k] = c2r[i]; c2r[i] = k;
2149       }
2150 
2151       /* MatPivotCheck() */
2152       sctx.rs = rs;
2153       sctx.pv = dk;
2154       ierr    = MatPivotCheck(B,A,info,&sctx,i);CHKERRQ(ierr);
2155       if (sctx.newshift) break;
2156       dk = sctx.pv;
2157 
2158       ba[bdiag[k]] = 1.0/dk; /* U(k,k) */
2159     }
2160   } while (sctx.newshift);
2161 
2162   ierr = PetscFree3(rtmp,il,c2r);CHKERRQ(ierr);
2163   ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr);
2164   ierr = ISRestoreIndices(iip,&riip);CHKERRQ(ierr);
2165 
2166   ierr = ISIdentity(ip,&perm_identity);CHKERRQ(ierr);
2167   if (perm_identity) {
2168     B->ops->solve          = MatSolve_SeqSBAIJ_1_NaturalOrdering;
2169     B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering;
2170     B->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering;
2171     B->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering;
2172   } else {
2173     B->ops->solve          = MatSolve_SeqSBAIJ_1;
2174     B->ops->solvetranspose = MatSolve_SeqSBAIJ_1;
2175     B->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1;
2176     B->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1;
2177   }
2178 
2179   C->assembled    = PETSC_TRUE;
2180   C->preallocated = PETSC_TRUE;
2181 
2182   ierr = PetscLogFlops(C->rmap->n);CHKERRQ(ierr);
2183 
2184   /* MatPivotView() */
2185   if (sctx.nshift) {
2186     if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
2187       ierr = PetscInfo4(A,"number of shift_pd tries %D, shift_amount %g, diagonal shifted up by %e fraction top_value %e\n",sctx.nshift,(double)sctx.shift_amount,(double)sctx.shift_fraction,(double)sctx.shift_top);CHKERRQ(ierr);
2188     } else if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
2189       ierr = PetscInfo2(A,"number of shift_nz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
2190     } else if (info->shifttype == (PetscReal)MAT_SHIFT_INBLOCKS) {
2191       ierr = PetscInfo2(A,"number of shift_inblocks applied %D, each shift_amount %g\n",sctx.nshift,(double)info->shiftamount);CHKERRQ(ierr);
2192     }
2193   }
2194   PetscFunctionReturn(0);
2195 }
2196 
2197 PetscErrorCode MatCholeskyFactorNumeric_SeqAIJ_inplace(Mat B,Mat A,const MatFactorInfo *info)
2198 {
2199   Mat            C = B;
2200   Mat_SeqAIJ     *a=(Mat_SeqAIJ*)A->data;
2201   Mat_SeqSBAIJ   *b=(Mat_SeqSBAIJ*)C->data;
2202   IS             ip=b->row,iip = b->icol;
2203   PetscErrorCode ierr;
2204   const PetscInt *rip,*riip;
2205   PetscInt       i,j,mbs=A->rmap->n,*bi=b->i,*bj=b->j,*bcol,*bjtmp;
2206   PetscInt       *ai=a->i,*aj=a->j;
2207   PetscInt       k,jmin,jmax,*jl,*il,col,nexti,ili,nz;
2208   MatScalar      *rtmp,*ba=b->a,*bval,*aa=a->a,dk,uikdi;
2209   PetscBool      perm_identity;
2210   FactorShiftCtx sctx;
2211   PetscReal      rs;
2212   MatScalar      d,*v;
2213 
2214   PetscFunctionBegin;
2215   /* MatPivotSetUp(): initialize shift context sctx */
2216   ierr = PetscMemzero(&sctx,sizeof(FactorShiftCtx));CHKERRQ(ierr);
2217 
2218   if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) { /* set sctx.shift_top=max{rs} */
2219     sctx.shift_top = info->zeropivot;
2220     for (i=0; i<mbs; i++) {
2221       /* calculate sum(|aij|)-RealPart(aii), amt of shift needed for this row */
2222       d  = (aa)[a->diag[i]];
2223       rs = -PetscAbsScalar(d) - PetscRealPart(d);
2224       v  = aa+ai[i];
2225       nz = ai[i+1] - ai[i];
2226       for (j=0; j<nz; j++) rs += PetscAbsScalar(v[j]);
2227       if (rs>sctx.shift_top) sctx.shift_top = rs;
2228     }
2229     sctx.shift_top *= 1.1;
2230     sctx.nshift_max = 5;
2231     sctx.shift_lo   = 0.;
2232     sctx.shift_hi   = 1.;
2233   }
2234 
2235   ierr = ISGetIndices(ip,&rip);CHKERRQ(ierr);
2236   ierr = ISGetIndices(iip,&riip);CHKERRQ(ierr);
2237 
2238   /* initialization */
2239   ierr = PetscMalloc3(mbs,&rtmp,mbs,&il,mbs,&jl);CHKERRQ(ierr);
2240 
2241   do {
2242     sctx.newshift = PETSC_FALSE;
2243 
2244     for (i=0; i<mbs; i++) jl[i] = mbs;
2245     il[0] = 0;
2246 
2247     for (k = 0; k<mbs; k++) {
2248       /* zero rtmp */
2249       nz    = bi[k+1] - bi[k];
2250       bjtmp = bj + bi[k];
2251       for (j=0; j<nz; j++) rtmp[bjtmp[j]] = 0.0;
2252 
2253       bval = ba + bi[k];
2254       /* initialize k-th row by the perm[k]-th row of A */
2255       jmin = ai[rip[k]]; jmax = ai[rip[k]+1];
2256       for (j = jmin; j < jmax; j++) {
2257         col = riip[aj[j]];
2258         if (col >= k) { /* only take upper triangular entry */
2259           rtmp[col] = aa[j];
2260           *bval++   = 0.0; /* for in-place factorization */
2261         }
2262       }
2263       /* shift the diagonal of the matrix */
2264       if (sctx.nshift) rtmp[k] += sctx.shift_amount;
2265 
2266       /* modify k-th row by adding in those rows i with U(i,k)!=0 */
2267       dk = rtmp[k];
2268       i  = jl[k]; /* first row to be added to k_th row  */
2269 
2270       while (i < k) {
2271         nexti = jl[i]; /* next row to be added to k_th row */
2272 
2273         /* compute multiplier, update diag(k) and U(i,k) */
2274         ili     = il[i]; /* index of first nonzero element in U(i,k:bms-1) */
2275         uikdi   = -ba[ili]*ba[bi[i]]; /* diagonal(k) */
2276         dk     += uikdi*ba[ili];
2277         ba[ili] = uikdi; /* -U(i,k) */
2278 
2279         /* add multiple of row i to k-th row */
2280         jmin = ili + 1; jmax = bi[i+1];
2281         if (jmin < jmax) {
2282           for (j=jmin; j<jmax; j++) rtmp[bj[j]] += uikdi*ba[j];
2283           /* update il and jl for row i */
2284           il[i] = jmin;
2285           j     = bj[jmin]; jl[i] = jl[j]; jl[j] = i;
2286         }
2287         i = nexti;
2288       }
2289 
2290       /* shift the diagonals when zero pivot is detected */
2291       /* compute rs=sum of abs(off-diagonal) */
2292       rs   = 0.0;
2293       jmin = bi[k]+1;
2294       nz   = bi[k+1] - jmin;
2295       bcol = bj + jmin;
2296       for (j=0; j<nz; j++) {
2297         rs += PetscAbsScalar(rtmp[bcol[j]]);
2298       }
2299 
2300       sctx.rs = rs;
2301       sctx.pv = dk;
2302       ierr    = MatPivotCheck(B,A,info,&sctx,k);CHKERRQ(ierr);
2303       if (sctx.newshift) break;
2304       dk = sctx.pv;
2305 
2306       /* copy data into U(k,:) */
2307       ba[bi[k]] = 1.0/dk; /* U(k,k) */
2308       jmin      = bi[k]+1; jmax = bi[k+1];
2309       if (jmin < jmax) {
2310         for (j=jmin; j<jmax; j++) {
2311           col = bj[j]; ba[j] = rtmp[col];
2312         }
2313         /* add the k-th row into il and jl */
2314         il[k] = jmin;
2315         i     = bj[jmin]; jl[k] = jl[i]; jl[i] = k;
2316       }
2317     }
2318   } while (sctx.newshift);
2319 
2320   ierr = PetscFree3(rtmp,il,jl);CHKERRQ(ierr);
2321   ierr = ISRestoreIndices(ip,&rip);CHKERRQ(ierr);
2322   ierr = ISRestoreIndices(iip,&riip);CHKERRQ(ierr);
2323 
2324   ierr = ISIdentity(ip,&perm_identity);CHKERRQ(ierr);
2325   if (perm_identity) {
2326     B->ops->solve          = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
2327     B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
2328     B->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
2329     B->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1_NaturalOrdering_inplace;
2330   } else {
2331     B->ops->solve          = MatSolve_SeqSBAIJ_1_inplace;
2332     B->ops->solvetranspose = MatSolve_SeqSBAIJ_1_inplace;
2333     B->ops->forwardsolve   = MatForwardSolve_SeqSBAIJ_1_inplace;
2334     B->ops->backwardsolve  = MatBackwardSolve_SeqSBAIJ_1_inplace;
2335   }
2336 
2337   C->assembled    = PETSC_TRUE;
2338   C->preallocated = PETSC_TRUE;
2339 
2340   ierr = PetscLogFlops(C->rmap->n);CHKERRQ(ierr);
2341   if (sctx.nshift) {
2342     if (info->shifttype == (PetscReal)MAT_SHIFT_NONZERO) {
2343       ierr = PetscInfo2(A,"number of shiftnz tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
2344     } else if (info->shifttype == (PetscReal)MAT_SHIFT_POSITIVE_DEFINITE) {
2345       ierr = PetscInfo2(A,"number of shiftpd tries %D, shift_amount %g\n",sctx.nshift,(double)sctx.shift_amount);CHKERRQ(ierr);
2346     }
2347   }
2348   PetscFunctionReturn(0);
2349 }
2350 
2351 /*
2352    icc() under revised new data structure.
2353    Factored arrays bj and ba are stored as
2354      U(0,:),...,U(i,:),U(n-1,:)
2355 
2356    ui=fact->i is an array of size n+1, in which
2357    ui+
2358      ui[i]:  points to 1st entry of U(i,:),i=0,...,n-1
2359      ui[n]:  points to U(n-1,n-1)+1
2360 
2361   udiag=fact->diag is an array of size n,in which
2362      udiag[i]: points to diagonal of U(i,:), i=0,...,n-1
2363 
2364    U(i,:) contains udiag[i] as its last entry, i.e.,
2365     U(i,:) = (u[i,i+1],...,u[i,n-1],diag[i])
2366 */
2367 
2368 PetscErrorCode MatICCFactorSymbolic_SeqAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
2369 {
2370   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data;
2371   Mat_SeqSBAIJ       *b;
2372   PetscErrorCode     ierr;
2373   PetscBool          perm_identity,missing;
2374   PetscInt           reallocs=0,i,*ai=a->i,*aj=a->j,am=A->rmap->n,*ui,*udiag;
2375   const PetscInt     *rip,*riip;
2376   PetscInt           jmin,jmax,nzk,k,j,*jl,prow,*il,nextprow;
2377   PetscInt           nlnk,*lnk,*lnk_lvl=NULL,d;
2378   PetscInt           ncols,ncols_upper,*cols,*ajtmp,*uj,**uj_ptr,**uj_lvl_ptr;
2379   PetscReal          fill          =info->fill,levels=info->levels;
2380   PetscFreeSpaceList free_space    =NULL,current_space=NULL;
2381   PetscFreeSpaceList free_space_lvl=NULL,current_space_lvl=NULL;
2382   PetscBT            lnkbt;
2383   IS                 iperm;
2384 
2385   PetscFunctionBegin;
2386   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
2387   ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr);
2388   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d);
2389   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
2390   ierr = ISInvertPermutation(perm,PETSC_DECIDE,&iperm);CHKERRQ(ierr);
2391 
2392   ierr  = PetscMalloc1(am+1,&ui);CHKERRQ(ierr);
2393   ierr  = PetscMalloc1(am+1,&udiag);CHKERRQ(ierr);
2394   ui[0] = 0;
2395 
2396   /* ICC(0) without matrix ordering: simply rearrange column indices */
2397   if (!levels && perm_identity) {
2398     for (i=0; i<am; i++) {
2399       ncols    = ai[i+1] - a->diag[i];
2400       ui[i+1]  = ui[i] + ncols;
2401       udiag[i] = ui[i+1] - 1; /* points to the last entry of U(i,:) */
2402     }
2403     ierr = PetscMalloc1(ui[am]+1,&uj);CHKERRQ(ierr);
2404     cols = uj;
2405     for (i=0; i<am; i++) {
2406       aj    = a->j + a->diag[i] + 1; /* 1st entry of U(i,:) without diagonal */
2407       ncols = ai[i+1] - a->diag[i] -1;
2408       for (j=0; j<ncols; j++) *cols++ = aj[j];
2409       *cols++ = i; /* diagoanl is located as the last entry of U(i,:) */
2410     }
2411   } else { /* case: levels>0 || (levels=0 && !perm_identity) */
2412     ierr = ISGetIndices(iperm,&riip);CHKERRQ(ierr);
2413     ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
2414 
2415     /* initialization */
2416     ierr = PetscMalloc1(am+1,&ajtmp);CHKERRQ(ierr);
2417 
2418     /* jl: linked list for storing indices of the pivot rows
2419        il: il[i] points to the 1st nonzero entry of U(i,k:am-1) */
2420     ierr = PetscMalloc4(am,&uj_ptr,am,&uj_lvl_ptr,am,&jl,am,&il);CHKERRQ(ierr);
2421     for (i=0; i<am; i++) {
2422       jl[i] = am; il[i] = 0;
2423     }
2424 
2425     /* create and initialize a linked list for storing column indices of the active row k */
2426     nlnk = am + 1;
2427     ierr = PetscIncompleteLLCreate(am,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
2428 
2429     /* initial FreeSpace size is fill*(ai[am]+am)/2 */
2430     ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,(ai[am]+am)/2),&free_space);CHKERRQ(ierr);
2431     current_space     = free_space;
2432     ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,(ai[am]+am)/2),&free_space_lvl);CHKERRQ(ierr);
2433     current_space_lvl = free_space_lvl;
2434 
2435     for (k=0; k<am; k++) {  /* for each active row k */
2436       /* initialize lnk by the column indices of row rip[k] of A */
2437       nzk   = 0;
2438       ncols = ai[rip[k]+1] - ai[rip[k]];
2439       if (!ncols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",rip[k],k);
2440       ncols_upper = 0;
2441       for (j=0; j<ncols; j++) {
2442         i = *(aj + ai[rip[k]] + j); /* unpermuted column index */
2443         if (riip[i] >= k) { /* only take upper triangular entry */
2444           ajtmp[ncols_upper] = i;
2445           ncols_upper++;
2446         }
2447       }
2448       ierr = PetscIncompleteLLInit(ncols_upper,ajtmp,am,riip,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
2449       nzk += nlnk;
2450 
2451       /* update lnk by computing fill-in for each pivot row to be merged in */
2452       prow = jl[k]; /* 1st pivot row */
2453 
2454       while (prow < k) {
2455         nextprow = jl[prow];
2456 
2457         /* merge prow into k-th row */
2458         jmin  = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:am-1) */
2459         jmax  = ui[prow+1];
2460         ncols = jmax-jmin;
2461         i     = jmin - ui[prow];
2462         cols  = uj_ptr[prow] + i; /* points to the 2nd nzero entry in U(prow,k:am-1) */
2463         uj    = uj_lvl_ptr[prow] + i; /* levels of cols */
2464         j     = *(uj - 1);
2465         ierr  = PetscICCLLAddSorted(ncols,cols,levels,uj,am,nlnk,lnk,lnk_lvl,lnkbt,j);CHKERRQ(ierr);
2466         nzk  += nlnk;
2467 
2468         /* update il and jl for prow */
2469         if (jmin < jmax) {
2470           il[prow] = jmin;
2471           j        = *cols; jl[prow] = jl[j]; jl[j] = prow;
2472         }
2473         prow = nextprow;
2474       }
2475 
2476       /* if free space is not available, make more free space */
2477       if (current_space->local_remaining<nzk) {
2478         i    = am - k + 1; /* num of unfactored rows */
2479         i    = PetscIntMultTruncate(i,PetscMin(nzk, i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
2480         ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
2481         ierr = PetscFreeSpaceGet(i,&current_space_lvl);CHKERRQ(ierr);
2482         reallocs++;
2483       }
2484 
2485       /* copy data into free_space and free_space_lvl, then initialize lnk */
2486       if (nzk == 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Empty row %D in ICC matrix factor",k);
2487       ierr = PetscIncompleteLLClean(am,am,nzk,lnk,lnk_lvl,current_space->array,current_space_lvl->array,lnkbt);CHKERRQ(ierr);
2488 
2489       /* add the k-th row into il and jl */
2490       if (nzk > 1) {
2491         i     = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:am-1) */
2492         jl[k] = jl[i]; jl[i] = k;
2493         il[k] = ui[k] + 1;
2494       }
2495       uj_ptr[k]     = current_space->array;
2496       uj_lvl_ptr[k] = current_space_lvl->array;
2497 
2498       current_space->array           += nzk;
2499       current_space->local_used      += nzk;
2500       current_space->local_remaining -= nzk;
2501 
2502       current_space_lvl->array           += nzk;
2503       current_space_lvl->local_used      += nzk;
2504       current_space_lvl->local_remaining -= nzk;
2505 
2506       ui[k+1] = ui[k] + nzk;
2507     }
2508 
2509     ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
2510     ierr = ISRestoreIndices(iperm,&riip);CHKERRQ(ierr);
2511     ierr = PetscFree4(uj_ptr,uj_lvl_ptr,jl,il);CHKERRQ(ierr);
2512     ierr = PetscFree(ajtmp);CHKERRQ(ierr);
2513 
2514     /* copy free_space into uj and free free_space; set ui, uj, udiag in new datastructure; */
2515     ierr = PetscMalloc1(ui[am]+1,&uj);CHKERRQ(ierr);
2516     ierr = PetscFreeSpaceContiguous_Cholesky(&free_space,uj,am,ui,udiag);CHKERRQ(ierr); /* store matrix factor  */
2517     ierr = PetscIncompleteLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
2518     ierr = PetscFreeSpaceDestroy(free_space_lvl);CHKERRQ(ierr);
2519 
2520   } /* end of case: levels>0 || (levels=0 && !perm_identity) */
2521 
2522   /* put together the new matrix in MATSEQSBAIJ format */
2523   b               = (Mat_SeqSBAIJ*)(fact)->data;
2524   b->singlemalloc = PETSC_FALSE;
2525 
2526   ierr = PetscMalloc1(ui[am]+1,&b->a);CHKERRQ(ierr);
2527 
2528   b->j             = uj;
2529   b->i             = ui;
2530   b->diag          = udiag;
2531   b->free_diag     = PETSC_TRUE;
2532   b->ilen          = 0;
2533   b->imax          = 0;
2534   b->row           = perm;
2535   b->col           = perm;
2536   ierr             = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
2537   ierr             = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
2538   b->icol          = iperm;
2539   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
2540 
2541   ierr = PetscMalloc1(am+1,&b->solve_work);CHKERRQ(ierr);
2542   ierr = PetscLogObjectMemory((PetscObject)fact,ui[am]*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
2543 
2544   b->maxnz   = b->nz = ui[am];
2545   b->free_a  = PETSC_TRUE;
2546   b->free_ij = PETSC_TRUE;
2547 
2548   fact->info.factor_mallocs   = reallocs;
2549   fact->info.fill_ratio_given = fill;
2550   if (ai[am] != 0) {
2551     /* nonzeros in lower triangular part of A (including diagonals) = (ai[am]+am)/2 */
2552     fact->info.fill_ratio_needed = ((PetscReal)2*ui[am])/(ai[am]+am);
2553   } else {
2554     fact->info.fill_ratio_needed = 0.0;
2555   }
2556 #if defined(PETSC_USE_INFO)
2557   if (ai[am] != 0) {
2558     PetscReal af = fact->info.fill_ratio_needed;
2559     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)fill,(double)af);CHKERRQ(ierr);
2560     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
2561     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g) for best performance.\n",(double)af);CHKERRQ(ierr);
2562   } else {
2563     ierr = PetscInfo(A,"Empty matrix\n");CHKERRQ(ierr);
2564   }
2565 #endif
2566   fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqAIJ;
2567   PetscFunctionReturn(0);
2568 }
2569 
2570 PetscErrorCode MatICCFactorSymbolic_SeqAIJ_inplace(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
2571 {
2572   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data;
2573   Mat_SeqSBAIJ       *b;
2574   PetscErrorCode     ierr;
2575   PetscBool          perm_identity,missing;
2576   PetscInt           reallocs=0,i,*ai=a->i,*aj=a->j,am=A->rmap->n,*ui,*udiag;
2577   const PetscInt     *rip,*riip;
2578   PetscInt           jmin,jmax,nzk,k,j,*jl,prow,*il,nextprow;
2579   PetscInt           nlnk,*lnk,*lnk_lvl=NULL,d;
2580   PetscInt           ncols,ncols_upper,*cols,*ajtmp,*uj,**uj_ptr,**uj_lvl_ptr;
2581   PetscReal          fill          =info->fill,levels=info->levels;
2582   PetscFreeSpaceList free_space    =NULL,current_space=NULL;
2583   PetscFreeSpaceList free_space_lvl=NULL,current_space_lvl=NULL;
2584   PetscBT            lnkbt;
2585   IS                 iperm;
2586 
2587   PetscFunctionBegin;
2588   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
2589   ierr = MatMissingDiagonal(A,&missing,&d);CHKERRQ(ierr);
2590   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",d);
2591   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
2592   ierr = ISInvertPermutation(perm,PETSC_DECIDE,&iperm);CHKERRQ(ierr);
2593 
2594   ierr  = PetscMalloc1(am+1,&ui);CHKERRQ(ierr);
2595   ierr  = PetscMalloc1(am+1,&udiag);CHKERRQ(ierr);
2596   ui[0] = 0;
2597 
2598   /* ICC(0) without matrix ordering: simply copies fill pattern */
2599   if (!levels && perm_identity) {
2600 
2601     for (i=0; i<am; i++) {
2602       ui[i+1]  = ui[i] + ai[i+1] - a->diag[i];
2603       udiag[i] = ui[i];
2604     }
2605     ierr = PetscMalloc1(ui[am]+1,&uj);CHKERRQ(ierr);
2606     cols = uj;
2607     for (i=0; i<am; i++) {
2608       aj    = a->j + a->diag[i];
2609       ncols = ui[i+1] - ui[i];
2610       for (j=0; j<ncols; j++) *cols++ = *aj++;
2611     }
2612   } else { /* case: levels>0 || (levels=0 && !perm_identity) */
2613     ierr = ISGetIndices(iperm,&riip);CHKERRQ(ierr);
2614     ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
2615 
2616     /* initialization */
2617     ierr = PetscMalloc1(am+1,&ajtmp);CHKERRQ(ierr);
2618 
2619     /* jl: linked list for storing indices of the pivot rows
2620        il: il[i] points to the 1st nonzero entry of U(i,k:am-1) */
2621     ierr = PetscMalloc4(am,&uj_ptr,am,&uj_lvl_ptr,am,&jl,am,&il);CHKERRQ(ierr);
2622     for (i=0; i<am; i++) {
2623       jl[i] = am; il[i] = 0;
2624     }
2625 
2626     /* create and initialize a linked list for storing column indices of the active row k */
2627     nlnk = am + 1;
2628     ierr = PetscIncompleteLLCreate(am,am,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
2629 
2630     /* initial FreeSpace size is fill*(ai[am]+1) */
2631     ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,ai[am]+1),&free_space);CHKERRQ(ierr);
2632     current_space     = free_space;
2633     ierr              = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,ai[am]+1),&free_space_lvl);CHKERRQ(ierr);
2634     current_space_lvl = free_space_lvl;
2635 
2636     for (k=0; k<am; k++) {  /* for each active row k */
2637       /* initialize lnk by the column indices of row rip[k] of A */
2638       nzk   = 0;
2639       ncols = ai[rip[k]+1] - ai[rip[k]];
2640       if (!ncols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",rip[k],k);
2641       ncols_upper = 0;
2642       for (j=0; j<ncols; j++) {
2643         i = *(aj + ai[rip[k]] + j); /* unpermuted column index */
2644         if (riip[i] >= k) { /* only take upper triangular entry */
2645           ajtmp[ncols_upper] = i;
2646           ncols_upper++;
2647         }
2648       }
2649       ierr = PetscIncompleteLLInit(ncols_upper,ajtmp,am,riip,nlnk,lnk,lnk_lvl,lnkbt);CHKERRQ(ierr);
2650       nzk += nlnk;
2651 
2652       /* update lnk by computing fill-in for each pivot row to be merged in */
2653       prow = jl[k]; /* 1st pivot row */
2654 
2655       while (prow < k) {
2656         nextprow = jl[prow];
2657 
2658         /* merge prow into k-th row */
2659         jmin  = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:am-1) */
2660         jmax  = ui[prow+1];
2661         ncols = jmax-jmin;
2662         i     = jmin - ui[prow];
2663         cols  = uj_ptr[prow] + i; /* points to the 2nd nzero entry in U(prow,k:am-1) */
2664         uj    = uj_lvl_ptr[prow] + i; /* levels of cols */
2665         j     = *(uj - 1);
2666         ierr  = PetscICCLLAddSorted(ncols,cols,levels,uj,am,nlnk,lnk,lnk_lvl,lnkbt,j);CHKERRQ(ierr);
2667         nzk  += nlnk;
2668 
2669         /* update il and jl for prow */
2670         if (jmin < jmax) {
2671           il[prow] = jmin;
2672           j        = *cols; jl[prow] = jl[j]; jl[j] = prow;
2673         }
2674         prow = nextprow;
2675       }
2676 
2677       /* if free space is not available, make more free space */
2678       if (current_space->local_remaining<nzk) {
2679         i    = am - k + 1; /* num of unfactored rows */
2680         i    = PetscIntMultTruncate(i,PetscMin(nzk, i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
2681         ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
2682         ierr = PetscFreeSpaceGet(i,&current_space_lvl);CHKERRQ(ierr);
2683         reallocs++;
2684       }
2685 
2686       /* copy data into free_space and free_space_lvl, then initialize lnk */
2687       if (!nzk) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Empty row %D in ICC matrix factor",k);
2688       ierr = PetscIncompleteLLClean(am,am,nzk,lnk,lnk_lvl,current_space->array,current_space_lvl->array,lnkbt);CHKERRQ(ierr);
2689 
2690       /* add the k-th row into il and jl */
2691       if (nzk > 1) {
2692         i     = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:am-1) */
2693         jl[k] = jl[i]; jl[i] = k;
2694         il[k] = ui[k] + 1;
2695       }
2696       uj_ptr[k]     = current_space->array;
2697       uj_lvl_ptr[k] = current_space_lvl->array;
2698 
2699       current_space->array           += nzk;
2700       current_space->local_used      += nzk;
2701       current_space->local_remaining -= nzk;
2702 
2703       current_space_lvl->array           += nzk;
2704       current_space_lvl->local_used      += nzk;
2705       current_space_lvl->local_remaining -= nzk;
2706 
2707       ui[k+1] = ui[k] + nzk;
2708     }
2709 
2710 #if defined(PETSC_USE_INFO)
2711     if (ai[am] != 0) {
2712       PetscReal af = (PetscReal)ui[am]/((PetscReal)ai[am]);
2713       ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)fill,(double)af);CHKERRQ(ierr);
2714       ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
2715       ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g) for best performance.\n",(double)af);CHKERRQ(ierr);
2716     } else {
2717       ierr = PetscInfo(A,"Empty matrix\n");CHKERRQ(ierr);
2718     }
2719 #endif
2720 
2721     ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
2722     ierr = ISRestoreIndices(iperm,&riip);CHKERRQ(ierr);
2723     ierr = PetscFree4(uj_ptr,uj_lvl_ptr,jl,il);CHKERRQ(ierr);
2724     ierr = PetscFree(ajtmp);CHKERRQ(ierr);
2725 
2726     /* destroy list of free space and other temporary array(s) */
2727     ierr = PetscMalloc1(ui[am]+1,&uj);CHKERRQ(ierr);
2728     ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
2729     ierr = PetscIncompleteLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
2730     ierr = PetscFreeSpaceDestroy(free_space_lvl);CHKERRQ(ierr);
2731 
2732   } /* end of case: levels>0 || (levels=0 && !perm_identity) */
2733 
2734   /* put together the new matrix in MATSEQSBAIJ format */
2735 
2736   b               = (Mat_SeqSBAIJ*)fact->data;
2737   b->singlemalloc = PETSC_FALSE;
2738 
2739   ierr = PetscMalloc1(ui[am]+1,&b->a);CHKERRQ(ierr);
2740 
2741   b->j         = uj;
2742   b->i         = ui;
2743   b->diag      = udiag;
2744   b->free_diag = PETSC_TRUE;
2745   b->ilen      = 0;
2746   b->imax      = 0;
2747   b->row       = perm;
2748   b->col       = perm;
2749 
2750   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
2751   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
2752 
2753   b->icol          = iperm;
2754   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
2755   ierr             = PetscMalloc1(am+1,&b->solve_work);CHKERRQ(ierr);
2756   ierr             = PetscLogObjectMemory((PetscObject)fact,(ui[am]-am)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
2757   b->maxnz         = b->nz = ui[am];
2758   b->free_a        = PETSC_TRUE;
2759   b->free_ij       = PETSC_TRUE;
2760 
2761   fact->info.factor_mallocs   = reallocs;
2762   fact->info.fill_ratio_given = fill;
2763   if (ai[am] != 0) {
2764     fact->info.fill_ratio_needed = ((PetscReal)ui[am])/((PetscReal)ai[am]);
2765   } else {
2766     fact->info.fill_ratio_needed = 0.0;
2767   }
2768   fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqAIJ_inplace;
2769   PetscFunctionReturn(0);
2770 }
2771 
2772 PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
2773 {
2774   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data;
2775   Mat_SeqSBAIJ       *b;
2776   PetscErrorCode     ierr;
2777   PetscBool          perm_identity,missing;
2778   PetscReal          fill = info->fill;
2779   const PetscInt     *rip,*riip;
2780   PetscInt           i,am=A->rmap->n,*ai=a->i,*aj=a->j,reallocs=0,prow;
2781   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
2782   PetscInt           nlnk,*lnk,ncols,ncols_upper,*cols,*uj,**ui_ptr,*uj_ptr,*udiag;
2783   PetscFreeSpaceList free_space=NULL,current_space=NULL;
2784   PetscBT            lnkbt;
2785   IS                 iperm;
2786 
2787   PetscFunctionBegin;
2788   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
2789   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
2790   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
2791 
2792   /* check whether perm is the identity mapping */
2793   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
2794   ierr = ISInvertPermutation(perm,PETSC_DECIDE,&iperm);CHKERRQ(ierr);
2795   ierr = ISGetIndices(iperm,&riip);CHKERRQ(ierr);
2796   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
2797 
2798   /* initialization */
2799   ierr  = PetscMalloc1(am+1,&ui);CHKERRQ(ierr);
2800   ierr  = PetscMalloc1(am+1,&udiag);CHKERRQ(ierr);
2801   ui[0] = 0;
2802 
2803   /* jl: linked list for storing indices of the pivot rows
2804      il: il[i] points to the 1st nonzero entry of U(i,k:am-1) */
2805   ierr = PetscMalloc4(am,&ui_ptr,am,&jl,am,&il,am,&cols);CHKERRQ(ierr);
2806   for (i=0; i<am; i++) {
2807     jl[i] = am; il[i] = 0;
2808   }
2809 
2810   /* create and initialize a linked list for storing column indices of the active row k */
2811   nlnk = am + 1;
2812   ierr = PetscLLCreate(am,am,nlnk,lnk,lnkbt);CHKERRQ(ierr);
2813 
2814   /* initial FreeSpace size is fill*(ai[am]+am)/2 */
2815   ierr          = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,(ai[am]+am)/2),&free_space);CHKERRQ(ierr);
2816   current_space = free_space;
2817 
2818   for (k=0; k<am; k++) {  /* for each active row k */
2819     /* initialize lnk by the column indices of row rip[k] of A */
2820     nzk   = 0;
2821     ncols = ai[rip[k]+1] - ai[rip[k]];
2822     if (!ncols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",rip[k],k);
2823     ncols_upper = 0;
2824     for (j=0; j<ncols; j++) {
2825       i = riip[*(aj + ai[rip[k]] + j)];
2826       if (i >= k) { /* only take upper triangular entry */
2827         cols[ncols_upper] = i;
2828         ncols_upper++;
2829       }
2830     }
2831     ierr = PetscLLAdd(ncols_upper,cols,am,nlnk,lnk,lnkbt);CHKERRQ(ierr);
2832     nzk += nlnk;
2833 
2834     /* update lnk by computing fill-in for each pivot row to be merged in */
2835     prow = jl[k]; /* 1st pivot row */
2836 
2837     while (prow < k) {
2838       nextprow = jl[prow];
2839       /* merge prow into k-th row */
2840       jmin   = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:am-1) */
2841       jmax   = ui[prow+1];
2842       ncols  = jmax-jmin;
2843       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:am-1) */
2844       ierr   = PetscLLAddSorted(ncols,uj_ptr,am,nlnk,lnk,lnkbt);CHKERRQ(ierr);
2845       nzk   += nlnk;
2846 
2847       /* update il and jl for prow */
2848       if (jmin < jmax) {
2849         il[prow] = jmin;
2850         j        = *uj_ptr;
2851         jl[prow] = jl[j];
2852         jl[j]    = prow;
2853       }
2854       prow = nextprow;
2855     }
2856 
2857     /* if free space is not available, make more free space */
2858     if (current_space->local_remaining<nzk) {
2859       i    = am - k + 1; /* num of unfactored rows */
2860       i    = PetscIntMultTruncate(i,PetscMin(nzk,i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
2861       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
2862       reallocs++;
2863     }
2864 
2865     /* copy data into free space, then initialize lnk */
2866     ierr = PetscLLClean(am,am,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
2867 
2868     /* add the k-th row into il and jl */
2869     if (nzk > 1) {
2870       i     = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:am-1) */
2871       jl[k] = jl[i]; jl[i] = k;
2872       il[k] = ui[k] + 1;
2873     }
2874     ui_ptr[k] = current_space->array;
2875 
2876     current_space->array           += nzk;
2877     current_space->local_used      += nzk;
2878     current_space->local_remaining -= nzk;
2879 
2880     ui[k+1] = ui[k] + nzk;
2881   }
2882 
2883   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
2884   ierr = ISRestoreIndices(iperm,&riip);CHKERRQ(ierr);
2885   ierr = PetscFree4(ui_ptr,jl,il,cols);CHKERRQ(ierr);
2886 
2887   /* copy free_space into uj and free free_space; set ui, uj, udiag in new datastructure; */
2888   ierr = PetscMalloc1(ui[am]+1,&uj);CHKERRQ(ierr);
2889   ierr = PetscFreeSpaceContiguous_Cholesky(&free_space,uj,am,ui,udiag);CHKERRQ(ierr); /* store matrix factor */
2890   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
2891 
2892   /* put together the new matrix in MATSEQSBAIJ format */
2893 
2894   b               = (Mat_SeqSBAIJ*)fact->data;
2895   b->singlemalloc = PETSC_FALSE;
2896   b->free_a       = PETSC_TRUE;
2897   b->free_ij      = PETSC_TRUE;
2898 
2899   ierr = PetscMalloc1(ui[am]+1,&b->a);CHKERRQ(ierr);
2900 
2901   b->j         = uj;
2902   b->i         = ui;
2903   b->diag      = udiag;
2904   b->free_diag = PETSC_TRUE;
2905   b->ilen      = 0;
2906   b->imax      = 0;
2907   b->row       = perm;
2908   b->col       = perm;
2909 
2910   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
2911   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
2912 
2913   b->icol          = iperm;
2914   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
2915 
2916   ierr = PetscMalloc1(am+1,&b->solve_work);CHKERRQ(ierr);
2917   ierr = PetscLogObjectMemory((PetscObject)fact,ui[am]*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
2918 
2919   b->maxnz = b->nz = ui[am];
2920 
2921   fact->info.factor_mallocs   = reallocs;
2922   fact->info.fill_ratio_given = fill;
2923   if (ai[am] != 0) {
2924     /* nonzeros in lower triangular part of A (including diagonals) = (ai[am]+am)/2 */
2925     fact->info.fill_ratio_needed = ((PetscReal)2*ui[am])/(ai[am]+am);
2926   } else {
2927     fact->info.fill_ratio_needed = 0.0;
2928   }
2929 #if defined(PETSC_USE_INFO)
2930   if (ai[am] != 0) {
2931     PetscReal af = fact->info.fill_ratio_needed;
2932     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)fill,(double)af);CHKERRQ(ierr);
2933     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
2934     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g) for best performance.\n",(double)af);CHKERRQ(ierr);
2935   } else {
2936     ierr = PetscInfo(A,"Empty matrix\n");CHKERRQ(ierr);
2937   }
2938 #endif
2939   fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqAIJ;
2940   PetscFunctionReturn(0);
2941 }
2942 
2943 PetscErrorCode MatCholeskyFactorSymbolic_SeqAIJ_inplace(Mat fact,Mat A,IS perm,const MatFactorInfo *info)
2944 {
2945   Mat_SeqAIJ         *a = (Mat_SeqAIJ*)A->data;
2946   Mat_SeqSBAIJ       *b;
2947   PetscErrorCode     ierr;
2948   PetscBool          perm_identity,missing;
2949   PetscReal          fill = info->fill;
2950   const PetscInt     *rip,*riip;
2951   PetscInt           i,am=A->rmap->n,*ai=a->i,*aj=a->j,reallocs=0,prow;
2952   PetscInt           *jl,jmin,jmax,nzk,*ui,k,j,*il,nextprow;
2953   PetscInt           nlnk,*lnk,ncols,ncols_upper,*cols,*uj,**ui_ptr,*uj_ptr;
2954   PetscFreeSpaceList free_space=NULL,current_space=NULL;
2955   PetscBT            lnkbt;
2956   IS                 iperm;
2957 
2958   PetscFunctionBegin;
2959   if (A->rmap->n != A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must be square matrix, rows %D columns %D",A->rmap->n,A->cmap->n);
2960   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
2961   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
2962 
2963   /* check whether perm is the identity mapping */
2964   ierr = ISIdentity(perm,&perm_identity);CHKERRQ(ierr);
2965   ierr = ISInvertPermutation(perm,PETSC_DECIDE,&iperm);CHKERRQ(ierr);
2966   ierr = ISGetIndices(iperm,&riip);CHKERRQ(ierr);
2967   ierr = ISGetIndices(perm,&rip);CHKERRQ(ierr);
2968 
2969   /* initialization */
2970   ierr  = PetscMalloc1(am+1,&ui);CHKERRQ(ierr);
2971   ui[0] = 0;
2972 
2973   /* jl: linked list for storing indices of the pivot rows
2974      il: il[i] points to the 1st nonzero entry of U(i,k:am-1) */
2975   ierr = PetscMalloc4(am,&ui_ptr,am,&jl,am,&il,am,&cols);CHKERRQ(ierr);
2976   for (i=0; i<am; i++) {
2977     jl[i] = am; il[i] = 0;
2978   }
2979 
2980   /* create and initialize a linked list for storing column indices of the active row k */
2981   nlnk = am + 1;
2982   ierr = PetscLLCreate(am,am,nlnk,lnk,lnkbt);CHKERRQ(ierr);
2983 
2984   /* initial FreeSpace size is fill*(ai[am]+1) */
2985   ierr          = PetscFreeSpaceGet(PetscRealIntMultTruncate(fill,ai[am]+1),&free_space);CHKERRQ(ierr);
2986   current_space = free_space;
2987 
2988   for (k=0; k<am; k++) {  /* for each active row k */
2989     /* initialize lnk by the column indices of row rip[k] of A */
2990     nzk   = 0;
2991     ncols = ai[rip[k]+1] - ai[rip[k]];
2992     if (!ncols) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_CH_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",rip[k],k);
2993     ncols_upper = 0;
2994     for (j=0; j<ncols; j++) {
2995       i = riip[*(aj + ai[rip[k]] + j)];
2996       if (i >= k) { /* only take upper triangular entry */
2997         cols[ncols_upper] = i;
2998         ncols_upper++;
2999       }
3000     }
3001     ierr = PetscLLAdd(ncols_upper,cols,am,nlnk,lnk,lnkbt);CHKERRQ(ierr);
3002     nzk += nlnk;
3003 
3004     /* update lnk by computing fill-in for each pivot row to be merged in */
3005     prow = jl[k]; /* 1st pivot row */
3006 
3007     while (prow < k) {
3008       nextprow = jl[prow];
3009       /* merge prow into k-th row */
3010       jmin   = il[prow] + 1; /* index of the 2nd nzero entry in U(prow,k:am-1) */
3011       jmax   = ui[prow+1];
3012       ncols  = jmax-jmin;
3013       uj_ptr = ui_ptr[prow] + jmin - ui[prow]; /* points to the 2nd nzero entry in U(prow,k:am-1) */
3014       ierr   = PetscLLAddSorted(ncols,uj_ptr,am,nlnk,lnk,lnkbt);CHKERRQ(ierr);
3015       nzk   += nlnk;
3016 
3017       /* update il and jl for prow */
3018       if (jmin < jmax) {
3019         il[prow] = jmin;
3020         j        = *uj_ptr; jl[prow] = jl[j]; jl[j] = prow;
3021       }
3022       prow = nextprow;
3023     }
3024 
3025     /* if free space is not available, make more free space */
3026     if (current_space->local_remaining<nzk) {
3027       i    = am - k + 1; /* num of unfactored rows */
3028       i    = PetscMin(i*nzk, i*(i-1)); /* i*nzk, i*(i-1): estimated and max additional space needed */
3029       ierr = PetscFreeSpaceGet(i,&current_space);CHKERRQ(ierr);
3030       reallocs++;
3031     }
3032 
3033     /* copy data into free space, then initialize lnk */
3034     ierr = PetscLLClean(am,am,nzk,lnk,current_space->array,lnkbt);CHKERRQ(ierr);
3035 
3036     /* add the k-th row into il and jl */
3037     if (nzk-1 > 0) {
3038       i     = current_space->array[1]; /* col value of the first nonzero element in U(k, k+1:am-1) */
3039       jl[k] = jl[i]; jl[i] = k;
3040       il[k] = ui[k] + 1;
3041     }
3042     ui_ptr[k] = current_space->array;
3043 
3044     current_space->array           += nzk;
3045     current_space->local_used      += nzk;
3046     current_space->local_remaining -= nzk;
3047 
3048     ui[k+1] = ui[k] + nzk;
3049   }
3050 
3051 #if defined(PETSC_USE_INFO)
3052   if (ai[am] != 0) {
3053     PetscReal af = (PetscReal)(ui[am])/((PetscReal)ai[am]);
3054     ierr = PetscInfo3(A,"Reallocs %D Fill ratio:given %g needed %g\n",reallocs,(double)fill,(double)af);CHKERRQ(ierr);
3055     ierr = PetscInfo1(A,"Run with -pc_factor_fill %g or use \n",(double)af);CHKERRQ(ierr);
3056     ierr = PetscInfo1(A,"PCFactorSetFill(pc,%g) for best performance.\n",(double)af);CHKERRQ(ierr);
3057   } else {
3058     ierr = PetscInfo(A,"Empty matrix\n");CHKERRQ(ierr);
3059   }
3060 #endif
3061 
3062   ierr = ISRestoreIndices(perm,&rip);CHKERRQ(ierr);
3063   ierr = ISRestoreIndices(iperm,&riip);CHKERRQ(ierr);
3064   ierr = PetscFree4(ui_ptr,jl,il,cols);CHKERRQ(ierr);
3065 
3066   /* destroy list of free space and other temporary array(s) */
3067   ierr = PetscMalloc1(ui[am]+1,&uj);CHKERRQ(ierr);
3068   ierr = PetscFreeSpaceContiguous(&free_space,uj);CHKERRQ(ierr);
3069   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
3070 
3071   /* put together the new matrix in MATSEQSBAIJ format */
3072 
3073   b               = (Mat_SeqSBAIJ*)fact->data;
3074   b->singlemalloc = PETSC_FALSE;
3075   b->free_a       = PETSC_TRUE;
3076   b->free_ij      = PETSC_TRUE;
3077 
3078   ierr = PetscMalloc1(ui[am]+1,&b->a);CHKERRQ(ierr);
3079 
3080   b->j    = uj;
3081   b->i    = ui;
3082   b->diag = 0;
3083   b->ilen = 0;
3084   b->imax = 0;
3085   b->row  = perm;
3086   b->col  = perm;
3087 
3088   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
3089   ierr = PetscObjectReference((PetscObject)perm);CHKERRQ(ierr);
3090 
3091   b->icol          = iperm;
3092   b->pivotinblocks = PETSC_FALSE; /* need to get from MatFactorInfo */
3093 
3094   ierr     = PetscMalloc1(am+1,&b->solve_work);CHKERRQ(ierr);
3095   ierr     = PetscLogObjectMemory((PetscObject)fact,(ui[am]-am)*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
3096   b->maxnz = b->nz = ui[am];
3097 
3098   fact->info.factor_mallocs   = reallocs;
3099   fact->info.fill_ratio_given = fill;
3100   if (ai[am] != 0) {
3101     fact->info.fill_ratio_needed = ((PetscReal)ui[am])/((PetscReal)ai[am]);
3102   } else {
3103     fact->info.fill_ratio_needed = 0.0;
3104   }
3105   fact->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_SeqAIJ_inplace;
3106   PetscFunctionReturn(0);
3107 }
3108 
3109 PetscErrorCode MatSolve_SeqAIJ_NaturalOrdering(Mat A,Vec bb,Vec xx)
3110 {
3111   Mat_SeqAIJ        *a = (Mat_SeqAIJ*)A->data;
3112   PetscErrorCode    ierr;
3113   PetscInt          n   = A->rmap->n;
3114   const PetscInt    *ai = a->i,*aj = a->j,*adiag = a->diag,*vi;
3115   PetscScalar       *x,sum;
3116   const PetscScalar *b;
3117   const MatScalar   *aa = a->a,*v;
3118   PetscInt          i,nz;
3119 
3120   PetscFunctionBegin;
3121   if (!n) PetscFunctionReturn(0);
3122 
3123   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
3124   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
3125 
3126   /* forward solve the lower triangular */
3127   x[0] = b[0];
3128   v    = aa;
3129   vi   = aj;
3130   for (i=1; i<n; i++) {
3131     nz  = ai[i+1] - ai[i];
3132     sum = b[i];
3133     PetscSparseDenseMinusDot(sum,x,v,vi,nz);
3134     v   += nz;
3135     vi  += nz;
3136     x[i] = sum;
3137   }
3138 
3139   /* backward solve the upper triangular */
3140   for (i=n-1; i>=0; i--) {
3141     v   = aa + adiag[i+1] + 1;
3142     vi  = aj + adiag[i+1] + 1;
3143     nz  = adiag[i] - adiag[i+1]-1;
3144     sum = x[i];
3145     PetscSparseDenseMinusDot(sum,x,v,vi,nz);
3146     x[i] = sum*v[nz]; /* x[i]=aa[adiag[i]]*sum; v++; */
3147   }
3148 
3149   ierr = PetscLogFlops(2.0*a->nz - A->cmap->n);CHKERRQ(ierr);
3150   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
3151   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
3152   PetscFunctionReturn(0);
3153 }
3154 
3155 PetscErrorCode MatSolve_SeqAIJ(Mat A,Vec bb,Vec xx)
3156 {
3157   Mat_SeqAIJ        *a    = (Mat_SeqAIJ*)A->data;
3158   IS                iscol = a->col,isrow = a->row;
3159   PetscErrorCode    ierr;
3160   PetscInt          i,n=A->rmap->n,*vi,*ai=a->i,*aj=a->j,*adiag = a->diag,nz;
3161   const PetscInt    *rout,*cout,*r,*c;
3162   PetscScalar       *x,*tmp,sum;
3163   const PetscScalar *b;
3164   const MatScalar   *aa = a->a,*v;
3165 
3166   PetscFunctionBegin;
3167   if (!n) PetscFunctionReturn(0);
3168 
3169   ierr = VecGetArrayRead(bb,&b);CHKERRQ(ierr);
3170   ierr = VecGetArrayWrite(xx,&x);CHKERRQ(ierr);
3171   tmp  = a->solve_work;
3172 
3173   ierr = ISGetIndices(isrow,&rout);CHKERRQ(ierr); r = rout;
3174   ierr = ISGetIndices(iscol,&cout);CHKERRQ(ierr); c = cout;
3175 
3176   /* forward solve the lower triangular */
3177   tmp[0] = b[r[0]];
3178   v      = aa;
3179   vi     = aj;
3180   for (i=1; i<n; i++) {
3181     nz  = ai[i+1] - ai[i];
3182     sum = b[r[i]];
3183     PetscSparseDenseMinusDot(sum,tmp,v,vi,nz);
3184     tmp[i] = sum;
3185     v     += nz; vi += nz;
3186   }
3187 
3188   /* backward solve the upper triangular */
3189   for (i=n-1; i>=0; i--) {
3190     v   = aa + adiag[i+1]+1;
3191     vi  = aj + adiag[i+1]+1;
3192     nz  = adiag[i]-adiag[i+1]-1;
3193     sum = tmp[i];
3194     PetscSparseDenseMinusDot(sum,tmp,v,vi,nz);
3195     x[c[i]] = tmp[i] = sum*v[nz]; /* v[nz] = aa[adiag[i]] */
3196   }
3197 
3198   ierr = ISRestoreIndices(isrow,&rout);CHKERRQ(ierr);
3199   ierr = ISRestoreIndices(iscol,&cout);CHKERRQ(ierr);
3200   ierr = VecRestoreArrayRead(bb,&b);CHKERRQ(ierr);
3201   ierr = VecRestoreArrayWrite(xx,&x);CHKERRQ(ierr);
3202   ierr = PetscLogFlops(2*a->nz - A->cmap->n);CHKERRQ(ierr);
3203   PetscFunctionReturn(0);
3204 }
3205 
3206 /*
3207     This will get a new name and become a varient of MatILUFactor_SeqAIJ() there is no longer separate functions in the matrix function table for dt factors
3208 */
3209 PetscErrorCode MatILUDTFactor_SeqAIJ(Mat A,IS isrow,IS iscol,const MatFactorInfo *info,Mat *fact)
3210 {
3211   Mat            B = *fact;
3212   Mat_SeqAIJ     *a=(Mat_SeqAIJ*)A->data,*b;
3213   IS             isicol;
3214   PetscErrorCode ierr;
3215   const PetscInt *r,*ic;
3216   PetscInt       i,n=A->rmap->n,*ai=a->i,*aj=a->j,*ajtmp,*adiag;
3217   PetscInt       *bi,*bj,*bdiag,*bdiag_rev;
3218   PetscInt       row,nzi,nzi_bl,nzi_bu,*im,nzi_al,nzi_au;
3219   PetscInt       nlnk,*lnk;
3220   PetscBT        lnkbt;
3221   PetscBool      row_identity,icol_identity;
3222   MatScalar      *aatmp,*pv,*batmp,*ba,*rtmp,*pc,multiplier,*vtmp,diag_tmp;
3223   const PetscInt *ics;
3224   PetscInt       j,nz,*pj,*bjtmp,k,ncut,*jtmp;
3225   PetscReal      dt     =info->dt,shift=info->shiftamount;
3226   PetscInt       dtcount=(PetscInt)info->dtcount,nnz_max;
3227   PetscBool      missing;
3228 
3229   PetscFunctionBegin;
3230   if (dt      == PETSC_DEFAULT) dt = 0.005;
3231   if (dtcount == PETSC_DEFAULT) dtcount = (PetscInt)(1.5*a->rmax);
3232 
3233   /* ------- symbolic factorization, can be reused ---------*/
3234   ierr = MatMissingDiagonal(A,&missing,&i);CHKERRQ(ierr);
3235   if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry %D",i);
3236   adiag=a->diag;
3237 
3238   ierr = ISInvertPermutation(iscol,PETSC_DECIDE,&isicol);CHKERRQ(ierr);
3239 
3240   /* bdiag is location of diagonal in factor */
3241   ierr = PetscMalloc1(n+1,&bdiag);CHKERRQ(ierr);     /* becomes b->diag */
3242   ierr = PetscMalloc1(n+1,&bdiag_rev);CHKERRQ(ierr); /* temporary */
3243 
3244   /* allocate row pointers bi */
3245   ierr = PetscMalloc1(2*n+2,&bi);CHKERRQ(ierr);
3246 
3247   /* allocate bj and ba; max num of nonzero entries is (ai[n]+2*n*dtcount+2) */
3248   if (dtcount > n-1) dtcount = n-1; /* diagonal is excluded */
3249   nnz_max = ai[n]+2*n*dtcount+2;
3250 
3251   ierr = PetscMalloc1(nnz_max+1,&bj);CHKERRQ(ierr);
3252   ierr = PetscMalloc1(nnz_max+1,&ba);CHKERRQ(ierr);
3253 
3254   /* put together the new matrix */
3255   ierr = MatSeqAIJSetPreallocation_SeqAIJ(B,MAT_SKIP_ALLOCATION,NULL);CHKERRQ(ierr);
3256   ierr = PetscLogObjectParent((PetscObject)B,(PetscObject)isicol);CHKERRQ(ierr);
3257   b    = (Mat_SeqAIJ*)B->data;
3258 
3259   b->free_a       = PETSC_TRUE;
3260   b->free_ij      = PETSC_TRUE;
3261   b->singlemalloc = PETSC_FALSE;
3262 
3263   b->a    = ba;
3264   b->j    = bj;
3265   b->i    = bi;
3266   b->diag = bdiag;
3267   b->ilen = 0;
3268   b->imax = 0;
3269   b->row  = isrow;
3270   b->col  = iscol;
3271   ierr    = PetscObjectReference((PetscObject)isrow);CHKERRQ(ierr);
3272   ierr    = PetscObjectReference((PetscObject)iscol);CHKERRQ(ierr);
3273   b->icol = isicol;
3274 
3275   ierr     = PetscMalloc1(n+1,&b->solve_work);CHKERRQ(ierr);
3276   ierr     = PetscLogObjectMemory((PetscObject)B,nnz_max*(sizeof(PetscInt)+sizeof(MatScalar)));CHKERRQ(ierr);
3277   b->maxnz = nnz_max;
3278 
3279   B->factortype            = MAT_FACTOR_ILUDT;
3280   B->info.factor_mallocs   = 0;
3281   B->info.fill_ratio_given = ((PetscReal)nnz_max)/((PetscReal)ai[n]);
3282   /* ------- end of symbolic factorization ---------*/
3283 
3284   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
3285   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
3286   ics  = ic;
3287 
3288   /* linked list for storing column indices of the active row */
3289   nlnk = n + 1;
3290   ierr = PetscLLCreate(n,n,nlnk,lnk,lnkbt);CHKERRQ(ierr);
3291 
3292   /* im: used by PetscLLAddSortedLU(); jtmp: working array for column indices of active row */
3293   ierr = PetscMalloc2(n,&im,n,&jtmp);CHKERRQ(ierr);
3294   /* rtmp, vtmp: working arrays for sparse and contiguous row entries of active row */
3295   ierr = PetscMalloc2(n,&rtmp,n,&vtmp);CHKERRQ(ierr);
3296   ierr = PetscArrayzero(rtmp,n);CHKERRQ(ierr);
3297 
3298   bi[0]        = 0;
3299   bdiag[0]     = nnz_max-1; /* location of diag[0] in factor B */
3300   bdiag_rev[n] = bdiag[0];
3301   bi[2*n+1]    = bdiag[0]+1; /* endof bj and ba array */
3302   for (i=0; i<n; i++) {
3303     /* copy initial fill into linked list */
3304     nzi = ai[r[i]+1] - ai[r[i]];
3305     if (!nzi) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Empty row in matrix: row in original ordering %D in permuted ordering %D",r[i],i);
3306     nzi_al = adiag[r[i]] - ai[r[i]];
3307     nzi_au = ai[r[i]+1] - adiag[r[i]] -1;
3308     ajtmp  = aj + ai[r[i]];
3309     ierr   = PetscLLAddPerm(nzi,ajtmp,ic,n,nlnk,lnk,lnkbt);CHKERRQ(ierr);
3310 
3311     /* load in initial (unfactored row) */
3312     aatmp = a->a + ai[r[i]];
3313     for (j=0; j<nzi; j++) {
3314       rtmp[ics[*ajtmp++]] = *aatmp++;
3315     }
3316 
3317     /* add pivot rows into linked list */
3318     row = lnk[n];
3319     while (row < i) {
3320       nzi_bl = bi[row+1] - bi[row] + 1;
3321       bjtmp  = bj + bdiag[row+1]+1; /* points to 1st column next to the diagonal in U */
3322       ierr   = PetscLLAddSortedLU(bjtmp,row,nlnk,lnk,lnkbt,i,nzi_bl,im);CHKERRQ(ierr);
3323       nzi   += nlnk;
3324       row    = lnk[row];
3325     }
3326 
3327     /* copy data from lnk into jtmp, then initialize lnk */
3328     ierr = PetscLLClean(n,n,nzi,lnk,jtmp,lnkbt);CHKERRQ(ierr);
3329 
3330     /* numerical factorization */
3331     bjtmp = jtmp;
3332     row   = *bjtmp++; /* 1st pivot row */
3333     while (row < i) {
3334       pc         = rtmp + row;
3335       pv         = ba + bdiag[row]; /* 1./(diag of the pivot row) */
3336       multiplier = (*pc) * (*pv);
3337       *pc        = multiplier;
3338       if (PetscAbsScalar(*pc) > dt) { /* apply tolerance dropping rule */
3339         pj = bj + bdiag[row+1] + 1;         /* point to 1st entry of U(row,:) */
3340         pv = ba + bdiag[row+1] + 1;
3341         nz = bdiag[row] - bdiag[row+1] - 1;         /* num of entries in U(row,:), excluding diagonal */
3342         for (j=0; j<nz; j++) rtmp[*pj++] -= multiplier * (*pv++);
3343         ierr = PetscLogFlops(1+2*nz);CHKERRQ(ierr);
3344       }
3345       row = *bjtmp++;
3346     }
3347 
3348     /* copy sparse rtmp into contiguous vtmp; separate L and U part */
3349     diag_tmp = rtmp[i];  /* save diagonal value - may not needed?? */
3350     nzi_bl   = 0; j = 0;
3351     while (jtmp[j] < i) { /* Note: jtmp is sorted */
3352       vtmp[j] = rtmp[jtmp[j]]; rtmp[jtmp[j]]=0.0;
3353       nzi_bl++; j++;
3354     }
3355     nzi_bu = nzi - nzi_bl -1;
3356     while (j < nzi) {
3357       vtmp[j] = rtmp[jtmp[j]]; rtmp[jtmp[j]]=0.0;
3358       j++;
3359     }
3360 
3361     bjtmp = bj + bi[i];
3362     batmp = ba + bi[i];
3363     /* apply level dropping rule to L part */
3364     ncut = nzi_al + dtcount;
3365     if (ncut < nzi_bl) {
3366       ierr = PetscSortSplit(ncut,nzi_bl,vtmp,jtmp);CHKERRQ(ierr);
3367       ierr = PetscSortIntWithScalarArray(ncut,jtmp,vtmp);CHKERRQ(ierr);
3368     } else {
3369       ncut = nzi_bl;
3370     }
3371     for (j=0; j<ncut; j++) {
3372       bjtmp[j] = jtmp[j];
3373       batmp[j] = vtmp[j];
3374     }
3375     bi[i+1] = bi[i] + ncut;
3376     nzi     = ncut + 1;
3377 
3378     /* apply level dropping rule to U part */
3379     ncut = nzi_au + dtcount;
3380     if (ncut < nzi_bu) {
3381       ierr = PetscSortSplit(ncut,nzi_bu,vtmp+nzi_bl+1,jtmp+nzi_bl+1);CHKERRQ(ierr);
3382       ierr = PetscSortIntWithScalarArray(ncut,jtmp+nzi_bl+1,vtmp+nzi_bl+1);CHKERRQ(ierr);
3383     } else {
3384       ncut = nzi_bu;
3385     }
3386     nzi += ncut;
3387 
3388     /* mark bdiagonal */
3389     bdiag[i+1]       = bdiag[i] - (ncut + 1);
3390     bdiag_rev[n-i-1] = bdiag[i+1];
3391     bi[2*n - i]      = bi[2*n - i +1] - (ncut + 1);
3392     bjtmp            = bj + bdiag[i];
3393     batmp            = ba + bdiag[i];
3394     *bjtmp           = i;
3395     *batmp           = diag_tmp; /* rtmp[i]; */
3396     if (*batmp == 0.0) {
3397       *batmp = dt+shift;
3398     }
3399     *batmp = 1.0/(*batmp); /* invert diagonal entries for simplier triangular solves */
3400 
3401     bjtmp = bj + bdiag[i+1]+1;
3402     batmp = ba + bdiag[i+1]+1;
3403     for (k=0; k<ncut; k++) {
3404       bjtmp[k] = jtmp[nzi_bl+1+k];
3405       batmp[k] = vtmp[nzi_bl+1+k];
3406     }
3407 
3408     im[i] = nzi;   /* used by PetscLLAddSortedLU() */
3409   } /* for (i=0; i<n; i++) */
3410   if (bi[n] >= bdiag[n]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"end of L array %d cannot >= the beginning of U array %d",bi[n],bdiag[n]);
3411 
3412   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
3413   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
3414 
3415   ierr = PetscLLDestroy(lnk,lnkbt);CHKERRQ(ierr);
3416   ierr = PetscFree2(im,jtmp);CHKERRQ(ierr);
3417   ierr = PetscFree2(rtmp,vtmp);CHKERRQ(ierr);
3418   ierr = PetscFree(bdiag_rev);CHKERRQ(ierr);
3419 
3420   ierr     = PetscLogFlops(B->cmap->n);CHKERRQ(ierr);
3421   b->maxnz = b->nz = bi[n] + bdiag[0] - bdiag[n];
3422 
3423   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
3424   ierr = ISIdentity(isicol,&icol_identity);CHKERRQ(ierr);
3425   if (row_identity && icol_identity) {
3426     B->ops->solve = MatSolve_SeqAIJ_NaturalOrdering;
3427   } else {
3428     B->ops->solve = MatSolve_SeqAIJ;
3429   }
3430 
3431   B->ops->solveadd          = 0;
3432   B->ops->solvetranspose    = 0;
3433   B->ops->solvetransposeadd = 0;
3434   B->ops->matsolve          = 0;
3435   B->assembled              = PETSC_TRUE;
3436   B->preallocated           = PETSC_TRUE;
3437   PetscFunctionReturn(0);
3438 }
3439 
3440 /* a wraper of MatILUDTFactor_SeqAIJ() */
3441 /*
3442     This will get a new name and become a varient of MatILUFactor_SeqAIJ() there is no longer separate functions in the matrix function table for dt factors
3443 */
3444 
3445 PetscErrorCode  MatILUDTFactorSymbolic_SeqAIJ(Mat fact,Mat A,IS row,IS col,const MatFactorInfo *info)
3446 {
3447   PetscErrorCode ierr;
3448 
3449   PetscFunctionBegin;
3450   ierr = MatILUDTFactor_SeqAIJ(A,row,col,info,&fact);CHKERRQ(ierr);
3451   PetscFunctionReturn(0);
3452 }
3453 
3454 /*
3455    same as MatLUFactorNumeric_SeqAIJ(), except using contiguous array matrix factors
3456    - intend to replace existing MatLUFactorNumeric_SeqAIJ()
3457 */
3458 /*
3459     This will get a new name and become a varient of MatILUFactor_SeqAIJ() there is no longer separate functions in the matrix function table for dt factors
3460 */
3461 
3462 PetscErrorCode  MatILUDTFactorNumeric_SeqAIJ(Mat fact,Mat A,const MatFactorInfo *info)
3463 {
3464   Mat            C     =fact;
3465   Mat_SeqAIJ     *a    =(Mat_SeqAIJ*)A->data,*b=(Mat_SeqAIJ*)C->data;
3466   IS             isrow = b->row,isicol = b->icol;
3467   PetscErrorCode ierr;
3468   const PetscInt *r,*ic,*ics;
3469   PetscInt       i,j,k,n=A->rmap->n,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
3470   PetscInt       *ajtmp,*bjtmp,nz,nzl,nzu,row,*bdiag = b->diag,*pj;
3471   MatScalar      *rtmp,*pc,multiplier,*v,*pv,*aa=a->a;
3472   PetscReal      dt=info->dt,shift=info->shiftamount;
3473   PetscBool      row_identity, col_identity;
3474 
3475   PetscFunctionBegin;
3476   ierr = ISGetIndices(isrow,&r);CHKERRQ(ierr);
3477   ierr = ISGetIndices(isicol,&ic);CHKERRQ(ierr);
3478   ierr = PetscMalloc1(n+1,&rtmp);CHKERRQ(ierr);
3479   ics  = ic;
3480 
3481   for (i=0; i<n; i++) {
3482     /* initialize rtmp array */
3483     nzl   = bi[i+1] - bi[i];       /* num of nozeros in L(i,:) */
3484     bjtmp = bj + bi[i];
3485     for  (j=0; j<nzl; j++) rtmp[*bjtmp++] = 0.0;
3486     rtmp[i] = 0.0;
3487     nzu     = bdiag[i] - bdiag[i+1]; /* num of nozeros in U(i,:) */
3488     bjtmp   = bj + bdiag[i+1] + 1;
3489     for  (j=0; j<nzu; j++) rtmp[*bjtmp++] = 0.0;
3490 
3491     /* load in initial unfactored row of A */
3492     nz    = ai[r[i]+1] - ai[r[i]];
3493     ajtmp = aj + ai[r[i]];
3494     v     = aa + ai[r[i]];
3495     for (j=0; j<nz; j++) {
3496       rtmp[ics[*ajtmp++]] = v[j];
3497     }
3498 
3499     /* numerical factorization */
3500     bjtmp = bj + bi[i]; /* point to 1st entry of L(i,:) */
3501     nzl   = bi[i+1] - bi[i]; /* num of entries in L(i,:) */
3502     k     = 0;
3503     while (k < nzl) {
3504       row        = *bjtmp++;
3505       pc         = rtmp + row;
3506       pv         = b->a + bdiag[row]; /* 1./(diag of the pivot row) */
3507       multiplier = (*pc) * (*pv);
3508       *pc        = multiplier;
3509       if (PetscAbsScalar(multiplier) > dt) {
3510         pj = bj + bdiag[row+1] + 1;         /* point to 1st entry of U(row,:) */
3511         pv = b->a + bdiag[row+1] + 1;
3512         nz = bdiag[row] - bdiag[row+1] - 1;         /* num of entries in U(row,:), excluding diagonal */
3513         for (j=0; j<nz; j++) rtmp[*pj++] -= multiplier * (*pv++);
3514         ierr = PetscLogFlops(1+2*nz);CHKERRQ(ierr);
3515       }
3516       k++;
3517     }
3518 
3519     /* finished row so stick it into b->a */
3520     /* L-part */
3521     pv  = b->a + bi[i];
3522     pj  = bj + bi[i];
3523     nzl = bi[i+1] - bi[i];
3524     for (j=0; j<nzl; j++) {
3525       pv[j] = rtmp[pj[j]];
3526     }
3527 
3528     /* diagonal: invert diagonal entries for simplier triangular solves */
3529     if (rtmp[i] == 0.0) rtmp[i] = dt+shift;
3530     b->a[bdiag[i]] = 1.0/rtmp[i];
3531 
3532     /* U-part */
3533     pv  = b->a + bdiag[i+1] + 1;
3534     pj  = bj + bdiag[i+1] + 1;
3535     nzu = bdiag[i] - bdiag[i+1] - 1;
3536     for (j=0; j<nzu; j++) {
3537       pv[j] = rtmp[pj[j]];
3538     }
3539   }
3540 
3541   ierr = PetscFree(rtmp);CHKERRQ(ierr);
3542   ierr = ISRestoreIndices(isicol,&ic);CHKERRQ(ierr);
3543   ierr = ISRestoreIndices(isrow,&r);CHKERRQ(ierr);
3544 
3545   ierr = ISIdentity(isrow,&row_identity);CHKERRQ(ierr);
3546   ierr = ISIdentity(isicol,&col_identity);CHKERRQ(ierr);
3547   if (row_identity && col_identity) {
3548     C->ops->solve = MatSolve_SeqAIJ_NaturalOrdering;
3549   } else {
3550     C->ops->solve = MatSolve_SeqAIJ;
3551   }
3552   C->ops->solveadd          = 0;
3553   C->ops->solvetranspose    = 0;
3554   C->ops->solvetransposeadd = 0;
3555   C->ops->matsolve          = 0;
3556   C->assembled              = PETSC_TRUE;
3557   C->preallocated           = PETSC_TRUE;
3558 
3559   ierr = PetscLogFlops(C->cmap->n);CHKERRQ(ierr);
3560   PetscFunctionReturn(0);
3561 }
3562