xref: /petsc/src/mat/impls/baij/mpi/mpibaij.c (revision d03495bd6a90c318a635ff2f45c05071e91761ec)
1 /*$Id: mpibaij.c,v 1.223 2001/06/21 22:08:18 buschelm Exp buschelm $*/
2 
3 #include "src/mat/impls/baij/mpi/mpibaij.h"   /*I  "petscmat.h"  I*/
4 #include "src/vec/vecimpl.h"
5 
6 EXTERN int MatSetUpMultiply_MPIBAIJ(Mat);
7 EXTERN int DisAssemble_MPIBAIJ(Mat);
8 EXTERN int MatIncreaseOverlap_MPIBAIJ(Mat,int,IS *,int);
9 EXTERN int MatGetSubMatrices_MPIBAIJ(Mat,int,IS *,IS *,MatReuse,Mat **);
10 EXTERN int MatGetValues_SeqBAIJ(Mat,int,int *,int,int *,Scalar *);
11 EXTERN int MatSetValues_SeqBAIJ(Mat,int,int *,int,int *,Scalar *,InsertMode);
12 EXTERN int MatSetValuesBlocked_SeqBAIJ(Mat,int,int*,int,int*,Scalar*,InsertMode);
13 EXTERN int MatGetRow_SeqBAIJ(Mat,int,int*,int**,Scalar**);
14 EXTERN int MatRestoreRow_SeqBAIJ(Mat,int,int*,int**,Scalar**);
15 EXTERN int MatPrintHelp_SeqBAIJ(Mat);
16 EXTERN int MatZeroRows_SeqBAIJ(Mat,IS,Scalar*);
17 
18 /*  UGLY, ugly, ugly
19    When MatScalar == Scalar the function MatSetValuesBlocked_MPIBAIJ_MatScalar() does
20    not exist. Otherwise ..._MatScalar() takes matrix elements in single precision and
21    inserts them into the single precision data structure. The function MatSetValuesBlocked_MPIBAIJ()
22    converts the entries into single precision and then calls ..._MatScalar() to put them
23    into the single precision data structures.
24 */
25 #if defined(PETSC_USE_MAT_SINGLE)
26 EXTERN int MatSetValuesBlocked_SeqBAIJ_MatScalar(Mat,int,int*,int,int*,MatScalar*,InsertMode);
27 EXTERN int MatSetValues_MPIBAIJ_MatScalar(Mat,int,int*,int,int*,MatScalar*,InsertMode);
28 EXTERN int MatSetValuesBlocked_MPIBAIJ_MatScalar(Mat,int,int*,int,int*,MatScalar*,InsertMode);
29 EXTERN int MatSetValues_MPIBAIJ_HT_MatScalar(Mat,int,int*,int,int*,MatScalar*,InsertMode);
30 EXTERN int MatSetValuesBlocked_MPIBAIJ_HT_MatScalar(Mat,int,int*,int,int*,MatScalar*,InsertMode);
31 #else
32 #define MatSetValuesBlocked_SeqBAIJ_MatScalar      MatSetValuesBlocked_SeqBAIJ
33 #define MatSetValues_MPIBAIJ_MatScalar             MatSetValues_MPIBAIJ
34 #define MatSetValuesBlocked_MPIBAIJ_MatScalar      MatSetValuesBlocked_MPIBAIJ
35 #define MatSetValues_MPIBAIJ_HT_MatScalar          MatSetValues_MPIBAIJ_HT
36 #define MatSetValuesBlocked_MPIBAIJ_HT_MatScalar   MatSetValuesBlocked_MPIBAIJ_HT
37 #endif
38 
39 #undef __FUNCT__
40 #define __FUNCT__ "MatGetRowMax_MPIBAIJ"
41 int MatGetRowMax_MPIBAIJ(Mat A,Vec v)
42 {
43   Mat_MPIBAIJ  *a = (Mat_MPIBAIJ*)A->data;
44   int          ierr,i;
45   Scalar       *va,*vb;
46   Vec          vtmp;
47 
48   PetscFunctionBegin;
49 
50   ierr = MatGetRowMax(a->A,v);CHKERRQ(ierr);
51   ierr = VecGetArray(v,&va);CHKERRQ(ierr);
52 
53   ierr = VecCreateSeq(PETSC_COMM_SELF,A->m,&vtmp);CHKERRQ(ierr);
54   ierr = MatGetRowMax(a->B,vtmp);CHKERRQ(ierr);
55   ierr = VecGetArray(vtmp,&vb);CHKERRQ(ierr);
56 
57   for (i=0; i<A->m; i++){
58     if (PetscAbsScalar(va[i]) < PetscAbsScalar(vb[i])) va[i] = vb[i];
59   }
60 
61   ierr = VecRestoreArray(v,&va);CHKERRQ(ierr);
62   ierr = VecRestoreArray(vtmp,&vb);CHKERRQ(ierr);
63   ierr = VecDestroy(vtmp);CHKERRQ(ierr);
64 
65   PetscFunctionReturn(0);
66 }
67 
68 EXTERN_C_BEGIN
69 #undef __FUNCT__
70 #define __FUNCT__ "MatStoreValues_MPIBAIJ"
71 int MatStoreValues_MPIBAIJ(Mat mat)
72 {
73   Mat_MPIBAIJ *aij = (Mat_MPIBAIJ *)mat->data;
74   int         ierr;
75 
76   PetscFunctionBegin;
77   ierr = MatStoreValues(aij->A);CHKERRQ(ierr);
78   ierr = MatStoreValues(aij->B);CHKERRQ(ierr);
79   PetscFunctionReturn(0);
80 }
81 EXTERN_C_END
82 
83 EXTERN_C_BEGIN
84 #undef __FUNCT__
85 #define __FUNCT__ "MatRetrieveValues_MPIBAIJ"
86 int MatRetrieveValues_MPIBAIJ(Mat mat)
87 {
88   Mat_MPIBAIJ *aij = (Mat_MPIBAIJ *)mat->data;
89   int         ierr;
90 
91   PetscFunctionBegin;
92   ierr = MatRetrieveValues(aij->A);CHKERRQ(ierr);
93   ierr = MatRetrieveValues(aij->B);CHKERRQ(ierr);
94   PetscFunctionReturn(0);
95 }
96 EXTERN_C_END
97 
98 /*
99      Local utility routine that creates a mapping from the global column
100    number to the local number in the off-diagonal part of the local
101    storage of the matrix.  This is done in a non scable way since the
102    length of colmap equals the global matrix length.
103 */
104 #undef __FUNCT__
105 #define __FUNCT__ "CreateColmap_MPIBAIJ_Private"
106 static int CreateColmap_MPIBAIJ_Private(Mat mat)
107 {
108   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
109   Mat_SeqBAIJ *B = (Mat_SeqBAIJ*)baij->B->data;
110   int         nbs = B->nbs,i,bs=B->bs,ierr;
111 
112   PetscFunctionBegin;
113 #if defined (PETSC_USE_CTABLE)
114   ierr = PetscTableCreate(baij->nbs,&baij->colmap);CHKERRQ(ierr);
115   for (i=0; i<nbs; i++){
116     ierr = PetscTableAdd(baij->colmap,baij->garray[i]+1,i*bs+1);CHKERRQ(ierr);
117   }
118 #else
119   ierr = PetscMalloc((baij->Nbs+1)*sizeof(int),&baij->colmap);CHKERRQ(ierr);
120   PetscLogObjectMemory(mat,baij->Nbs*sizeof(int));
121   ierr = PetscMemzero(baij->colmap,baij->Nbs*sizeof(int));CHKERRQ(ierr);
122   for (i=0; i<nbs; i++) baij->colmap[baij->garray[i]] = i*bs+1;
123 #endif
124   PetscFunctionReturn(0);
125 }
126 
127 #define CHUNKSIZE  10
128 
129 #define  MatSetValues_SeqBAIJ_A_Private(row,col,value,addv) \
130 { \
131  \
132     brow = row/bs;  \
133     rp   = aj + ai[brow]; ap = aa + bs2*ai[brow]; \
134     rmax = aimax[brow]; nrow = ailen[brow]; \
135       bcol = col/bs; \
136       ridx = row % bs; cidx = col % bs; \
137       low = 0; high = nrow; \
138       while (high-low > 3) { \
139         t = (low+high)/2; \
140         if (rp[t] > bcol) high = t; \
141         else              low  = t; \
142       } \
143       for (_i=low; _i<high; _i++) { \
144         if (rp[_i] > bcol) break; \
145         if (rp[_i] == bcol) { \
146           bap  = ap +  bs2*_i + bs*cidx + ridx; \
147           if (addv == ADD_VALUES) *bap += value;  \
148           else                    *bap  = value;  \
149           goto a_noinsert; \
150         } \
151       } \
152       if (a->nonew == 1) goto a_noinsert; \
153       else if (a->nonew == -1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero into matrix"); \
154       if (nrow >= rmax) { \
155         /* there is no extra room in row, therefore enlarge */ \
156         int       new_nz = ai[a->mbs] + CHUNKSIZE,len,*new_i,*new_j; \
157         MatScalar *new_a; \
158  \
159         if (a->nonew == -2) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); \
160  \
161         /* malloc new storage space */ \
162         len     = new_nz*(sizeof(int)+bs2*sizeof(MatScalar))+(a->mbs+1)*sizeof(int); \
163         ierr = PetscMalloc(len,&new_a);CHKERRQ(ierr); \
164         new_j   = (int*)(new_a + bs2*new_nz); \
165         new_i   = new_j + new_nz; \
166  \
167         /* copy over old data into new slots */ \
168         for (ii=0; ii<brow+1; ii++) {new_i[ii] = ai[ii];} \
169         for (ii=brow+1; ii<a->mbs+1; ii++) {new_i[ii] = ai[ii]+CHUNKSIZE;} \
170         ierr = PetscMemcpy(new_j,aj,(ai[brow]+nrow)*sizeof(int));CHKERRQ(ierr); \
171         len = (new_nz - CHUNKSIZE - ai[brow] - nrow); \
172         ierr = PetscMemcpy(new_j+ai[brow]+nrow+CHUNKSIZE,aj+ai[brow]+nrow,len*sizeof(int));CHKERRQ(ierr); \
173         ierr = PetscMemcpy(new_a,aa,(ai[brow]+nrow)*bs2*sizeof(MatScalar));CHKERRQ(ierr); \
174         ierr = PetscMemzero(new_a+bs2*(ai[brow]+nrow),bs2*CHUNKSIZE*sizeof(Scalar));CHKERRQ(ierr); \
175         ierr = PetscMemcpy(new_a+bs2*(ai[brow]+nrow+CHUNKSIZE), \
176                     aa+bs2*(ai[brow]+nrow),bs2*len*sizeof(MatScalar));CHKERRQ(ierr);  \
177         /* free up old matrix storage */ \
178         ierr = PetscFree(a->a);CHKERRQ(ierr);  \
179         if (!a->singlemalloc) { \
180           ierr = PetscFree(a->i);CHKERRQ(ierr); \
181           ierr = PetscFree(a->j);CHKERRQ(ierr);\
182         } \
183         aa = a->a = new_a; ai = a->i = new_i; aj = a->j = new_j;  \
184         a->singlemalloc = PETSC_TRUE; \
185  \
186         rp   = aj + ai[brow]; ap = aa + bs2*ai[brow]; \
187         rmax = aimax[brow] = aimax[brow] + CHUNKSIZE; \
188         PetscLogObjectMemory(A,CHUNKSIZE*(sizeof(int) + bs2*sizeof(MatScalar))); \
189         a->maxnz += bs2*CHUNKSIZE; \
190         a->reallocs++; \
191         a->nz++; \
192       } \
193       N = nrow++ - 1;  \
194       /* shift up all the later entries in this row */ \
195       for (ii=N; ii>=_i; ii--) { \
196         rp[ii+1] = rp[ii]; \
197         ierr = PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar));CHKERRQ(ierr); \
198       } \
199       if (N>=_i) { ierr = PetscMemzero(ap+bs2*_i,bs2*sizeof(MatScalar));CHKERRQ(ierr); }  \
200       rp[_i]                      = bcol;  \
201       ap[bs2*_i + bs*cidx + ridx] = value;  \
202       a_noinsert:; \
203     ailen[brow] = nrow; \
204 }
205 
206 #define  MatSetValues_SeqBAIJ_B_Private(row,col,value,addv) \
207 { \
208     brow = row/bs;  \
209     rp   = bj + bi[brow]; ap = ba + bs2*bi[brow]; \
210     rmax = bimax[brow]; nrow = bilen[brow]; \
211       bcol = col/bs; \
212       ridx = row % bs; cidx = col % bs; \
213       low = 0; high = nrow; \
214       while (high-low > 3) { \
215         t = (low+high)/2; \
216         if (rp[t] > bcol) high = t; \
217         else              low  = t; \
218       } \
219       for (_i=low; _i<high; _i++) { \
220         if (rp[_i] > bcol) break; \
221         if (rp[_i] == bcol) { \
222           bap  = ap +  bs2*_i + bs*cidx + ridx; \
223           if (addv == ADD_VALUES) *bap += value;  \
224           else                    *bap  = value;  \
225           goto b_noinsert; \
226         } \
227       } \
228       if (b->nonew == 1) goto b_noinsert; \
229       else if (b->nonew == -1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero into matrix"); \
230       if (nrow >= rmax) { \
231         /* there is no extra room in row, therefore enlarge */ \
232         int       new_nz = bi[b->mbs] + CHUNKSIZE,len,*new_i,*new_j; \
233         MatScalar *new_a; \
234  \
235         if (b->nonew == -2) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix"); \
236  \
237         /* malloc new storage space */ \
238         len     = new_nz*(sizeof(int)+bs2*sizeof(MatScalar))+(b->mbs+1)*sizeof(int); \
239         ierr    = PetscMalloc(len,&new_a);CHKERRQ(ierr); \
240         new_j   = (int*)(new_a + bs2*new_nz); \
241         new_i   = new_j + new_nz; \
242  \
243         /* copy over old data into new slots */ \
244         for (ii=0; ii<brow+1; ii++) {new_i[ii] = bi[ii];} \
245         for (ii=brow+1; ii<b->mbs+1; ii++) {new_i[ii] = bi[ii]+CHUNKSIZE;} \
246         ierr = PetscMemcpy(new_j,bj,(bi[brow]+nrow)*sizeof(int));CHKERRQ(ierr); \
247         len  = (new_nz - CHUNKSIZE - bi[brow] - nrow); \
248         ierr = PetscMemcpy(new_j+bi[brow]+nrow+CHUNKSIZE,bj+bi[brow]+nrow,len*sizeof(int));CHKERRQ(ierr); \
249         ierr = PetscMemcpy(new_a,ba,(bi[brow]+nrow)*bs2*sizeof(MatScalar));CHKERRQ(ierr); \
250         ierr = PetscMemzero(new_a+bs2*(bi[brow]+nrow),bs2*CHUNKSIZE*sizeof(MatScalar));CHKERRQ(ierr); \
251         ierr = PetscMemcpy(new_a+bs2*(bi[brow]+nrow+CHUNKSIZE), \
252                     ba+bs2*(bi[brow]+nrow),bs2*len*sizeof(MatScalar));CHKERRQ(ierr);  \
253         /* free up old matrix storage */ \
254         ierr = PetscFree(b->a);CHKERRQ(ierr);  \
255         if (!b->singlemalloc) { \
256           ierr = PetscFree(b->i);CHKERRQ(ierr); \
257           ierr = PetscFree(b->j);CHKERRQ(ierr); \
258         } \
259         ba = b->a = new_a; bi = b->i = new_i; bj = b->j = new_j;  \
260         b->singlemalloc = PETSC_TRUE; \
261  \
262         rp   = bj + bi[brow]; ap = ba + bs2*bi[brow]; \
263         rmax = bimax[brow] = bimax[brow] + CHUNKSIZE; \
264         PetscLogObjectMemory(B,CHUNKSIZE*(sizeof(int) + bs2*sizeof(MatScalar))); \
265         b->maxnz += bs2*CHUNKSIZE; \
266         b->reallocs++; \
267         b->nz++; \
268       } \
269       N = nrow++ - 1;  \
270       /* shift up all the later entries in this row */ \
271       for (ii=N; ii>=_i; ii--) { \
272         rp[ii+1] = rp[ii]; \
273         ierr = PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar));CHKERRQ(ierr); \
274       } \
275       if (N>=_i) { ierr = PetscMemzero(ap+bs2*_i,bs2*sizeof(MatScalar));CHKERRQ(ierr);}  \
276       rp[_i]                      = bcol;  \
277       ap[bs2*_i + bs*cidx + ridx] = value;  \
278       b_noinsert:; \
279     bilen[brow] = nrow; \
280 }
281 
282 #if defined(PETSC_USE_MAT_SINGLE)
283 #undef __FUNCT__
284 #define __FUNCT__ "MatSetValues_MPIBAIJ"
285 int MatSetValues_MPIBAIJ(Mat mat,int m,int *im,int n,int *in,Scalar *v,InsertMode addv)
286 {
287   Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
288   int         ierr,i,N = m*n;
289   MatScalar   *vsingle;
290 
291   PetscFunctionBegin;
292   if (N > b->setvalueslen) {
293     if (b->setvaluescopy) {ierr = PetscFree(b->setvaluescopy);CHKERRQ(ierr);}
294     ierr = PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);CHKERRQ(ierr);
295     b->setvalueslen  = N;
296   }
297   vsingle = b->setvaluescopy;
298 
299   for (i=0; i<N; i++) {
300     vsingle[i] = v[i];
301   }
302   ierr = MatSetValues_MPIBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);CHKERRQ(ierr);
303   PetscFunctionReturn(0);
304 }
305 
306 #undef __FUNCT__
307 #define __FUNCT__ "MatSetValuesBlocked_MPIBAIJ"
308 int MatSetValuesBlocked_MPIBAIJ(Mat mat,int m,int *im,int n,int *in,Scalar *v,InsertMode addv)
309 {
310   Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
311   int         ierr,i,N = m*n*b->bs2;
312   MatScalar   *vsingle;
313 
314   PetscFunctionBegin;
315   if (N > b->setvalueslen) {
316     if (b->setvaluescopy) {ierr = PetscFree(b->setvaluescopy);CHKERRQ(ierr);}
317     ierr = PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);CHKERRQ(ierr);
318     b->setvalueslen  = N;
319   }
320   vsingle = b->setvaluescopy;
321   for (i=0; i<N; i++) {
322     vsingle[i] = v[i];
323   }
324   ierr = MatSetValuesBlocked_MPIBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);CHKERRQ(ierr);
325   PetscFunctionReturn(0);
326 }
327 
328 #undef __FUNCT__
329 #define __FUNCT__ "MatSetValues_MPIBAIJ_HT"
330 int MatSetValues_MPIBAIJ_HT(Mat mat,int m,int *im,int n,int *in,Scalar *v,InsertMode addv)
331 {
332   Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
333   int         ierr,i,N = m*n;
334   MatScalar   *vsingle;
335 
336   PetscFunctionBegin;
337   if (N > b->setvalueslen) {
338     if (b->setvaluescopy) {ierr = PetscFree(b->setvaluescopy);CHKERRQ(ierr);}
339     ierr = PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);CHKERRQ(ierr);
340     b->setvalueslen  = N;
341   }
342   vsingle = b->setvaluescopy;
343   for (i=0; i<N; i++) {
344     vsingle[i] = v[i];
345   }
346   ierr = MatSetValues_MPIBAIJ_HT_MatScalar(mat,m,im,n,in,vsingle,addv);CHKERRQ(ierr);
347   PetscFunctionReturn(0);
348 }
349 
350 #undef __FUNCT__
351 #define __FUNCT__ "MatSetValuesBlocked_MPIBAIJ_HT"
352 int MatSetValuesBlocked_MPIBAIJ_HT(Mat mat,int m,int *im,int n,int *in,Scalar *v,InsertMode addv)
353 {
354   Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
355   int         ierr,i,N = m*n*b->bs2;
356   MatScalar   *vsingle;
357 
358   PetscFunctionBegin;
359   if (N > b->setvalueslen) {
360     if (b->setvaluescopy) {ierr = PetscFree(b->setvaluescopy);CHKERRQ(ierr);}
361     ierr = PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);CHKERRQ(ierr);
362     b->setvalueslen  = N;
363   }
364   vsingle = b->setvaluescopy;
365   for (i=0; i<N; i++) {
366     vsingle[i] = v[i];
367   }
368   ierr = MatSetValuesBlocked_MPIBAIJ_HT_MatScalar(mat,m,im,n,in,vsingle,addv);CHKERRQ(ierr);
369   PetscFunctionReturn(0);
370 }
371 #endif
372 
373 #undef __FUNCT__
374 #define __FUNCT__ "MatSetValues_MPIBAIJ"
375 int MatSetValues_MPIBAIJ_MatScalar(Mat mat,int m,int *im,int n,int *in,MatScalar *v,InsertMode addv)
376 {
377   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
378   MatScalar   value;
379   PetscTruth  roworiented = baij->roworiented;
380   int         ierr,i,j,row,col;
381   int         rstart_orig=baij->rstart_bs;
382   int         rend_orig=baij->rend_bs,cstart_orig=baij->cstart_bs;
383   int         cend_orig=baij->cend_bs,bs=baij->bs;
384 
385   /* Some Variables required in the macro */
386   Mat         A = baij->A;
387   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)(A)->data;
388   int         *aimax=a->imax,*ai=a->i,*ailen=a->ilen,*aj=a->j;
389   MatScalar   *aa=a->a;
390 
391   Mat         B = baij->B;
392   Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)(B)->data;
393   int         *bimax=b->imax,*bi=b->i,*bilen=b->ilen,*bj=b->j;
394   MatScalar   *ba=b->a;
395 
396   int         *rp,ii,nrow,_i,rmax,N,brow,bcol;
397   int         low,high,t,ridx,cidx,bs2=a->bs2;
398   MatScalar   *ap,*bap;
399 
400   PetscFunctionBegin;
401   for (i=0; i<m; i++) {
402     if (im[i] < 0) continue;
403 #if defined(PETSC_USE_BOPT_g)
404     if (im[i] >= mat->M) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
405 #endif
406     if (im[i] >= rstart_orig && im[i] < rend_orig) {
407       row = im[i] - rstart_orig;
408       for (j=0; j<n; j++) {
409         if (in[j] >= cstart_orig && in[j] < cend_orig){
410           col = in[j] - cstart_orig;
411           if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
412           MatSetValues_SeqBAIJ_A_Private(row,col,value,addv);
413           /* ierr = MatSetValues_SeqBAIJ(baij->A,1,&row,1,&col,&value,addv);CHKERRQ(ierr); */
414         } else if (in[j] < 0) continue;
415 #if defined(PETSC_USE_BOPT_g)
416         else if (in[j] >= mat->N) {SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Column too large");}
417 #endif
418         else {
419           if (mat->was_assembled) {
420             if (!baij->colmap) {
421               ierr = CreateColmap_MPIBAIJ_Private(mat);CHKERRQ(ierr);
422             }
423 #if defined (PETSC_USE_CTABLE)
424             ierr = PetscTableFind(baij->colmap,in[j]/bs + 1,&col);CHKERRQ(ierr);
425             col  = col - 1 + in[j]%bs;
426 #else
427             col = baij->colmap[in[j]/bs] - 1 + in[j]%bs;
428 #endif
429             if (col < 0 && !((Mat_SeqBAIJ*)(baij->A->data))->nonew) {
430               ierr = DisAssemble_MPIBAIJ(mat);CHKERRQ(ierr);
431               col =  in[j];
432               /* Reinitialize the variables required by MatSetValues_SeqBAIJ_B_Private() */
433               B = baij->B;
434               b = (Mat_SeqBAIJ*)(B)->data;
435               bimax=b->imax;bi=b->i;bilen=b->ilen;bj=b->j;
436               ba=b->a;
437             }
438           } else col = in[j];
439           if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
440           MatSetValues_SeqBAIJ_B_Private(row,col,value,addv);
441           /* ierr = MatSetValues_SeqBAIJ(baij->B,1,&row,1,&col,&value,addv);CHKERRQ(ierr); */
442         }
443       }
444     } else {
445       if (!baij->donotstash) {
446         if (roworiented) {
447           ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n);CHKERRQ(ierr);
448         } else {
449           ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m);CHKERRQ(ierr);
450         }
451       }
452     }
453   }
454   PetscFunctionReturn(0);
455 }
456 
457 #undef __FUNCT__
458 #define __FUNCT__ "MatSetValuesBlocked_MPIBAIJ_MatScalar"
459 int MatSetValuesBlocked_MPIBAIJ_MatScalar(Mat mat,int m,int *im,int n,int *in,MatScalar *v,InsertMode addv)
460 {
461   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
462   MatScalar   *value,*barray=baij->barray;
463   PetscTruth  roworiented = baij->roworiented;
464   int         ierr,i,j,ii,jj,row,col,rstart=baij->rstart;
465   int         rend=baij->rend,cstart=baij->cstart,stepval;
466   int         cend=baij->cend,bs=baij->bs,bs2=baij->bs2;
467 
468   PetscFunctionBegin;
469   if(!barray) {
470     ierr         = PetscMalloc(bs2*sizeof(MatScalar),&barray);CHKERRQ(ierr);
471     baij->barray = barray;
472   }
473 
474   if (roworiented) {
475     stepval = (n-1)*bs;
476   } else {
477     stepval = (m-1)*bs;
478   }
479   for (i=0; i<m; i++) {
480     if (im[i] < 0) continue;
481 #if defined(PETSC_USE_BOPT_g)
482     if (im[i] >= baij->Mbs) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large, row %d max %d",im[i],baij->Mbs);
483 #endif
484     if (im[i] >= rstart && im[i] < rend) {
485       row = im[i] - rstart;
486       for (j=0; j<n; j++) {
487         /* If NumCol = 1 then a copy is not required */
488         if ((roworiented) && (n == 1)) {
489           barray = v + i*bs2;
490         } else if((!roworiented) && (m == 1)) {
491           barray = v + j*bs2;
492         } else { /* Here a copy is required */
493           if (roworiented) {
494             value = v + i*(stepval+bs)*bs + j*bs;
495           } else {
496             value = v + j*(stepval+bs)*bs + i*bs;
497           }
498           for (ii=0; ii<bs; ii++,value+=stepval) {
499             for (jj=0; jj<bs; jj++) {
500               *barray++  = *value++;
501             }
502           }
503           barray -=bs2;
504         }
505 
506         if (in[j] >= cstart && in[j] < cend){
507           col  = in[j] - cstart;
508           ierr = MatSetValuesBlocked_SeqBAIJ_MatScalar(baij->A,1,&row,1,&col,barray,addv);CHKERRQ(ierr);
509         }
510         else if (in[j] < 0) continue;
511 #if defined(PETSC_USE_BOPT_g)
512         else if (in[j] >= baij->Nbs) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large, col %d max %d",in[j],baij->Nbs);}
513 #endif
514         else {
515           if (mat->was_assembled) {
516             if (!baij->colmap) {
517               ierr = CreateColmap_MPIBAIJ_Private(mat);CHKERRQ(ierr);
518             }
519 
520 #if defined(PETSC_USE_BOPT_g)
521 #if defined (PETSC_USE_CTABLE)
522             { int data;
523               ierr = PetscTableFind(baij->colmap,in[j]+1,&data);CHKERRQ(ierr);
524               if ((data - 1) % bs) SETERRQ(PETSC_ERR_PLIB,"Incorrect colmap");
525             }
526 #else
527             if ((baij->colmap[in[j]] - 1) % bs) SETERRQ(PETSC_ERR_PLIB,"Incorrect colmap");
528 #endif
529 #endif
530 #if defined (PETSC_USE_CTABLE)
531 	    ierr = PetscTableFind(baij->colmap,in[j]+1,&col);CHKERRQ(ierr);
532             col  = (col - 1)/bs;
533 #else
534             col = (baij->colmap[in[j]] - 1)/bs;
535 #endif
536             if (col < 0 && !((Mat_SeqBAIJ*)(baij->A->data))->nonew) {
537               ierr = DisAssemble_MPIBAIJ(mat);CHKERRQ(ierr);
538               col =  in[j];
539             }
540           }
541           else col = in[j];
542           ierr = MatSetValuesBlocked_SeqBAIJ_MatScalar(baij->B,1,&row,1,&col,barray,addv);CHKERRQ(ierr);
543         }
544       }
545     } else {
546       if (!baij->donotstash) {
547         if (roworiented) {
548           ierr = MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);CHKERRQ(ierr);
549         } else {
550           ierr = MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);CHKERRQ(ierr);
551         }
552       }
553     }
554   }
555   PetscFunctionReturn(0);
556 }
557 
558 #define HASH_KEY 0.6180339887
559 #define HASH(size,key,tmp) (tmp = (key)*HASH_KEY,(int)((size)*(tmp-(int)tmp)))
560 /* #define HASH(size,key) ((int)((size)*fmod(((key)*HASH_KEY),1))) */
561 /* #define HASH(size,key,tmp) ((int)((size)*fmod(((key)*HASH_KEY),1))) */
562 #undef __FUNCT__
563 #define __FUNCT__ "MatSetValues_MPIBAIJ_HT_MatScalar"
564 int MatSetValues_MPIBAIJ_HT_MatScalar(Mat mat,int m,int *im,int n,int *in,MatScalar *v,InsertMode addv)
565 {
566   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
567   PetscTruth  roworiented = baij->roworiented;
568   int         ierr,i,j,row,col;
569   int         rstart_orig=baij->rstart_bs;
570   int         rend_orig=baij->rend_bs,Nbs=baij->Nbs;
571   int         h1,key,size=baij->ht_size,bs=baij->bs,*HT=baij->ht,idx;
572   PetscReal   tmp;
573   MatScalar   **HD = baij->hd,value;
574 #if defined(PETSC_USE_BOPT_g)
575   int         total_ct=baij->ht_total_ct,insert_ct=baij->ht_insert_ct;
576 #endif
577 
578   PetscFunctionBegin;
579 
580   for (i=0; i<m; i++) {
581 #if defined(PETSC_USE_BOPT_g)
582     if (im[i] < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Negative row");
583     if (im[i] >= mat->M) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
584 #endif
585       row = im[i];
586     if (row >= rstart_orig && row < rend_orig) {
587       for (j=0; j<n; j++) {
588         col = in[j];
589         if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
590         /* Look up into the Hash Table */
591         key = (row/bs)*Nbs+(col/bs)+1;
592         h1  = HASH(size,key,tmp);
593 
594 
595         idx = h1;
596 #if defined(PETSC_USE_BOPT_g)
597         insert_ct++;
598         total_ct++;
599         if (HT[idx] != key) {
600           for (idx=h1; (idx<size) && (HT[idx]!=key); idx++,total_ct++);
601           if (idx == size) {
602             for (idx=0; (idx<h1) && (HT[idx]!=key); idx++,total_ct++);
603             if (idx == h1) {
604               SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"(row,col) has no entry in the hash table");
605             }
606           }
607         }
608 #else
609         if (HT[idx] != key) {
610           for (idx=h1; (idx<size) && (HT[idx]!=key); idx++);
611           if (idx == size) {
612             for (idx=0; (idx<h1) && (HT[idx]!=key); idx++);
613             if (idx == h1) {
614               SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"(row,col) has no entry in the hash table");
615             }
616           }
617         }
618 #endif
619         /* A HASH table entry is found, so insert the values at the correct address */
620         if (addv == ADD_VALUES) *(HD[idx]+ (col % bs)*bs + (row % bs)) += value;
621         else                    *(HD[idx]+ (col % bs)*bs + (row % bs))  = value;
622       }
623     } else {
624       if (!baij->donotstash) {
625         if (roworiented) {
626           ierr = MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n);CHKERRQ(ierr);
627         } else {
628           ierr = MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m);CHKERRQ(ierr);
629         }
630       }
631     }
632   }
633 #if defined(PETSC_USE_BOPT_g)
634   baij->ht_total_ct = total_ct;
635   baij->ht_insert_ct = insert_ct;
636 #endif
637   PetscFunctionReturn(0);
638 }
639 
640 #undef __FUNCT__
641 #define __FUNCT__ "MatSetValuesBlocked_MPIBAIJ_HT_MatScalar"
642 int MatSetValuesBlocked_MPIBAIJ_HT_MatScalar(Mat mat,int m,int *im,int n,int *in,MatScalar *v,InsertMode addv)
643 {
644   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
645   PetscTruth  roworiented = baij->roworiented;
646   int         ierr,i,j,ii,jj,row,col;
647   int         rstart=baij->rstart ;
648   int         rend=baij->rend,stepval,bs=baij->bs,bs2=baij->bs2;
649   int         h1,key,size=baij->ht_size,idx,*HT=baij->ht,Nbs=baij->Nbs;
650   PetscReal   tmp;
651   MatScalar   **HD = baij->hd,*baij_a;
652   MatScalar   *v_t,*value;
653 #if defined(PETSC_USE_BOPT_g)
654   int         total_ct=baij->ht_total_ct,insert_ct=baij->ht_insert_ct;
655 #endif
656 
657   PetscFunctionBegin;
658 
659   if (roworiented) {
660     stepval = (n-1)*bs;
661   } else {
662     stepval = (m-1)*bs;
663   }
664   for (i=0; i<m; i++) {
665 #if defined(PETSC_USE_BOPT_g)
666     if (im[i] < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Negative row");
667     if (im[i] >= baij->Mbs) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
668 #endif
669     row   = im[i];
670     v_t   = v + i*bs2;
671     if (row >= rstart && row < rend) {
672       for (j=0; j<n; j++) {
673         col = in[j];
674 
675         /* Look up into the Hash Table */
676         key = row*Nbs+col+1;
677         h1  = HASH(size,key,tmp);
678 
679         idx = h1;
680 #if defined(PETSC_USE_BOPT_g)
681         total_ct++;
682         insert_ct++;
683        if (HT[idx] != key) {
684           for (idx=h1; (idx<size) && (HT[idx]!=key); idx++,total_ct++);
685           if (idx == size) {
686             for (idx=0; (idx<h1) && (HT[idx]!=key); idx++,total_ct++);
687             if (idx == h1) {
688               SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"(row,col) has no entry in the hash table");
689             }
690           }
691         }
692 #else
693         if (HT[idx] != key) {
694           for (idx=h1; (idx<size) && (HT[idx]!=key); idx++);
695           if (idx == size) {
696             for (idx=0; (idx<h1) && (HT[idx]!=key); idx++);
697             if (idx == h1) {
698               SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"(row,col) has no entry in the hash table");
699             }
700           }
701         }
702 #endif
703         baij_a = HD[idx];
704         if (roworiented) {
705           /*value = v + i*(stepval+bs)*bs + j*bs;*/
706           /* value = v + (i*(stepval+bs)+j)*bs; */
707           value = v_t;
708           v_t  += bs;
709           if (addv == ADD_VALUES) {
710             for (ii=0; ii<bs; ii++,value+=stepval) {
711               for (jj=ii; jj<bs2; jj+=bs) {
712                 baij_a[jj]  += *value++;
713               }
714             }
715           } else {
716             for (ii=0; ii<bs; ii++,value+=stepval) {
717               for (jj=ii; jj<bs2; jj+=bs) {
718                 baij_a[jj]  = *value++;
719               }
720             }
721           }
722         } else {
723           value = v + j*(stepval+bs)*bs + i*bs;
724           if (addv == ADD_VALUES) {
725             for (ii=0; ii<bs; ii++,value+=stepval,baij_a+=bs) {
726               for (jj=0; jj<bs; jj++) {
727                 baij_a[jj]  += *value++;
728               }
729             }
730           } else {
731             for (ii=0; ii<bs; ii++,value+=stepval,baij_a+=bs) {
732               for (jj=0; jj<bs; jj++) {
733                 baij_a[jj]  = *value++;
734               }
735             }
736           }
737         }
738       }
739     } else {
740       if (!baij->donotstash) {
741         if (roworiented) {
742           ierr = MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);CHKERRQ(ierr);
743         } else {
744           ierr = MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);CHKERRQ(ierr);
745         }
746       }
747     }
748   }
749 #if defined(PETSC_USE_BOPT_g)
750   baij->ht_total_ct = total_ct;
751   baij->ht_insert_ct = insert_ct;
752 #endif
753   PetscFunctionReturn(0);
754 }
755 
756 #undef __FUNCT__
757 #define __FUNCT__ "MatGetValues_MPIBAIJ"
758 int MatGetValues_MPIBAIJ(Mat mat,int m,int *idxm,int n,int *idxn,Scalar *v)
759 {
760   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
761   int        bs=baij->bs,ierr,i,j,bsrstart = baij->rstart*bs,bsrend = baij->rend*bs;
762   int        bscstart = baij->cstart*bs,bscend = baij->cend*bs,row,col,data;
763 
764   PetscFunctionBegin;
765   for (i=0; i<m; i++) {
766     if (idxm[i] < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Negative row");
767     if (idxm[i] >= mat->M) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
768     if (idxm[i] >= bsrstart && idxm[i] < bsrend) {
769       row = idxm[i] - bsrstart;
770       for (j=0; j<n; j++) {
771         if (idxn[j] < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Negative column");
772         if (idxn[j] >= mat->N) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Column too large");
773         if (idxn[j] >= bscstart && idxn[j] < bscend){
774           col = idxn[j] - bscstart;
775           ierr = MatGetValues_SeqBAIJ(baij->A,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr);
776         } else {
777           if (!baij->colmap) {
778             ierr = CreateColmap_MPIBAIJ_Private(mat);CHKERRQ(ierr);
779           }
780 #if defined (PETSC_USE_CTABLE)
781           ierr = PetscTableFind(baij->colmap,idxn[j]/bs+1,&data);CHKERRQ(ierr);
782           data --;
783 #else
784           data = baij->colmap[idxn[j]/bs]-1;
785 #endif
786           if((data < 0) || (baij->garray[data/bs] != idxn[j]/bs)) *(v+i*n+j) = 0.0;
787           else {
788             col  = data + idxn[j]%bs;
789             ierr = MatGetValues_SeqBAIJ(baij->B,1,&row,1,&col,v+i*n+j);CHKERRQ(ierr);
790           }
791         }
792       }
793     } else {
794       SETERRQ(PETSC_ERR_SUP,"Only local values currently supported");
795     }
796   }
797  PetscFunctionReturn(0);
798 }
799 
800 #undef __FUNCT__
801 #define __FUNCT__ "MatNorm_MPIBAIJ"
802 int MatNorm_MPIBAIJ(Mat mat,NormType type,PetscReal *norm)
803 {
804   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
805   Mat_SeqBAIJ *amat = (Mat_SeqBAIJ*)baij->A->data,*bmat = (Mat_SeqBAIJ*)baij->B->data;
806   int        ierr,i,bs2=baij->bs2;
807   PetscReal  sum = 0.0;
808   MatScalar  *v;
809 
810   PetscFunctionBegin;
811   if (baij->size == 1) {
812     ierr =  MatNorm(baij->A,type,norm);CHKERRQ(ierr);
813   } else {
814     if (type == NORM_FROBENIUS) {
815       v = amat->a;
816       for (i=0; i<amat->nz*bs2; i++) {
817 #if defined(PETSC_USE_COMPLEX)
818         sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
819 #else
820         sum += (*v)*(*v); v++;
821 #endif
822       }
823       v = bmat->a;
824       for (i=0; i<bmat->nz*bs2; i++) {
825 #if defined(PETSC_USE_COMPLEX)
826         sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
827 #else
828         sum += (*v)*(*v); v++;
829 #endif
830       }
831       ierr = MPI_Allreduce(&sum,norm,1,MPI_DOUBLE,MPI_SUM,mat->comm);CHKERRQ(ierr);
832       *norm = sqrt(*norm);
833     } else {
834       SETERRQ(PETSC_ERR_SUP,"No support for this norm yet");
835     }
836   }
837   PetscFunctionReturn(0);
838 }
839 
840 
841 /*
842   Creates the hash table, and sets the table
843   This table is created only once.
844   If new entried need to be added to the matrix
845   then the hash table has to be destroyed and
846   recreated.
847 */
848 #undef __FUNCT__
849 #define __FUNCT__ "MatCreateHashTable_MPIBAIJ_Private"
850 int MatCreateHashTable_MPIBAIJ_Private(Mat mat,PetscReal factor)
851 {
852   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
853   Mat         A = baij->A,B=baij->B;
854   Mat_SeqBAIJ *a=(Mat_SeqBAIJ *)A->data,*b=(Mat_SeqBAIJ *)B->data;
855   int         i,j,k,nz=a->nz+b->nz,h1,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
856   int         size,bs2=baij->bs2,rstart=baij->rstart,ierr;
857   int         cstart=baij->cstart,*garray=baij->garray,row,col,Nbs=baij->Nbs;
858   int         *HT,key;
859   MatScalar   **HD;
860   PetscReal   tmp;
861 #if defined(PETSC_USE_BOPT_g)
862   int         ct=0,max=0;
863 #endif
864 
865   PetscFunctionBegin;
866   baij->ht_size=(int)(factor*nz);
867   size = baij->ht_size;
868 
869   if (baij->ht) {
870     PetscFunctionReturn(0);
871   }
872 
873   /* Allocate Memory for Hash Table */
874   ierr     = PetscMalloc((size)*(sizeof(int)+sizeof(MatScalar*))+1,&baij->hd);CHKERRQ(ierr);
875   baij->ht = (int*)(baij->hd + size);
876   HD       = baij->hd;
877   HT       = baij->ht;
878 
879 
880   ierr = PetscMemzero(HD,size*(sizeof(int)+sizeof(Scalar*)));CHKERRQ(ierr);
881 
882 
883   /* Loop Over A */
884   for (i=0; i<a->mbs; i++) {
885     for (j=ai[i]; j<ai[i+1]; j++) {
886       row = i+rstart;
887       col = aj[j]+cstart;
888 
889       key = row*Nbs + col + 1;
890       h1  = HASH(size,key,tmp);
891       for (k=0; k<size; k++){
892         if (HT[(h1+k)%size] == 0.0) {
893           HT[(h1+k)%size] = key;
894           HD[(h1+k)%size] = a->a + j*bs2;
895           break;
896 #if defined(PETSC_USE_BOPT_g)
897         } else {
898           ct++;
899 #endif
900         }
901       }
902 #if defined(PETSC_USE_BOPT_g)
903       if (k> max) max = k;
904 #endif
905     }
906   }
907   /* Loop Over B */
908   for (i=0; i<b->mbs; i++) {
909     for (j=bi[i]; j<bi[i+1]; j++) {
910       row = i+rstart;
911       col = garray[bj[j]];
912       key = row*Nbs + col + 1;
913       h1  = HASH(size,key,tmp);
914       for (k=0; k<size; k++){
915         if (HT[(h1+k)%size] == 0.0) {
916           HT[(h1+k)%size] = key;
917           HD[(h1+k)%size] = b->a + j*bs2;
918           break;
919 #if defined(PETSC_USE_BOPT_g)
920         } else {
921           ct++;
922 #endif
923         }
924       }
925 #if defined(PETSC_USE_BOPT_g)
926       if (k> max) max = k;
927 #endif
928     }
929   }
930 
931   /* Print Summary */
932 #if defined(PETSC_USE_BOPT_g)
933   for (i=0,j=0; i<size; i++) {
934     if (HT[i]) {j++;}
935   }
936   PetscLogInfo(0,"MatCreateHashTable_MPIBAIJ_Private: Average Search = %5.2f,max search = %d\n",(j== 0)? 0.0:((PetscReal)(ct+j))/j,max);
937 #endif
938   PetscFunctionReturn(0);
939 }
940 
941 #undef __FUNCT__
942 #define __FUNCT__ "MatAssemblyBegin_MPIBAIJ"
943 int MatAssemblyBegin_MPIBAIJ(Mat mat,MatAssemblyType mode)
944 {
945   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
946   int         ierr,nstash,reallocs;
947   InsertMode  addv;
948 
949   PetscFunctionBegin;
950   if (baij->donotstash) {
951     PetscFunctionReturn(0);
952   }
953 
954   /* make sure all processors are either in INSERTMODE or ADDMODE */
955   ierr = MPI_Allreduce(&mat->insertmode,&addv,1,MPI_INT,MPI_BOR,mat->comm);CHKERRQ(ierr);
956   if (addv == (ADD_VALUES|INSERT_VALUES)) {
957     SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Some processors inserted others added");
958   }
959   mat->insertmode = addv; /* in case this processor had no cache */
960 
961   ierr = MatStashScatterBegin_Private(&mat->stash,baij->rowners_bs);CHKERRQ(ierr);
962   ierr = MatStashScatterBegin_Private(&mat->bstash,baij->rowners);CHKERRQ(ierr);
963   ierr = MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);CHKERRQ(ierr);
964   PetscLogInfo(0,"MatAssemblyBegin_MPIBAIJ:Stash has %d entries,uses %d mallocs.\n",nstash,reallocs);
965   ierr = MatStashGetInfo_Private(&mat->bstash,&nstash,&reallocs);CHKERRQ(ierr);
966   PetscLogInfo(0,"MatAssemblyBegin_MPIBAIJ:Block-Stash has %d entries, uses %d mallocs.\n",nstash,reallocs);
967   PetscFunctionReturn(0);
968 }
969 
970 #undef __FUNCT__
971 #define __FUNCT__ "MatAssemblyEnd_MPIBAIJ"
972 int MatAssemblyEnd_MPIBAIJ(Mat mat,MatAssemblyType mode)
973 {
974   Mat_MPIBAIJ *baij=(Mat_MPIBAIJ*)mat->data;
975   Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)baij->A->data,*b=(Mat_SeqBAIJ*)baij->B->data;
976   int         i,j,rstart,ncols,n,ierr,flg,bs2=baij->bs2;
977   int         *row,*col,other_disassembled;
978   PetscTruth  r1,r2,r3;
979   MatScalar   *val;
980   InsertMode  addv = mat->insertmode;
981 
982   PetscFunctionBegin;
983   if (!baij->donotstash) {
984     while (1) {
985       ierr = MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);CHKERRQ(ierr);
986       if (!flg) break;
987 
988       for (i=0; i<n;) {
989         /* Now identify the consecutive vals belonging to the same row */
990         for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; }
991         if (j < n) ncols = j-i;
992         else       ncols = n-i;
993         /* Now assemble all these values with a single function call */
994         ierr = MatSetValues_MPIBAIJ_MatScalar(mat,1,row+i,ncols,col+i,val+i,addv);CHKERRQ(ierr);
995         i = j;
996       }
997     }
998     ierr = MatStashScatterEnd_Private(&mat->stash);CHKERRQ(ierr);
999     /* Now process the block-stash. Since the values are stashed column-oriented,
1000        set the roworiented flag to column oriented, and after MatSetValues()
1001        restore the original flags */
1002     r1 = baij->roworiented;
1003     r2 = a->roworiented;
1004     r3 = b->roworiented;
1005     baij->roworiented = PETSC_FALSE;
1006     a->roworiented    = PETSC_FALSE;
1007     b->roworiented    = PETSC_FALSE;
1008     while (1) {
1009       ierr = MatStashScatterGetMesg_Private(&mat->bstash,&n,&row,&col,&val,&flg);CHKERRQ(ierr);
1010       if (!flg) break;
1011 
1012       for (i=0; i<n;) {
1013         /* Now identify the consecutive vals belonging to the same row */
1014         for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; }
1015         if (j < n) ncols = j-i;
1016         else       ncols = n-i;
1017         ierr = MatSetValuesBlocked_MPIBAIJ_MatScalar(mat,1,row+i,ncols,col+i,val+i*bs2,addv);CHKERRQ(ierr);
1018         i = j;
1019       }
1020     }
1021     ierr = MatStashScatterEnd_Private(&mat->bstash);CHKERRQ(ierr);
1022     baij->roworiented = r1;
1023     a->roworiented    = r2;
1024     b->roworiented    = r3;
1025   }
1026 
1027   ierr = MatAssemblyBegin(baij->A,mode);CHKERRQ(ierr);
1028   ierr = MatAssemblyEnd(baij->A,mode);CHKERRQ(ierr);
1029 
1030   /* determine if any processor has disassembled, if so we must
1031      also disassemble ourselfs, in order that we may reassemble. */
1032   /*
1033      if nonzero structure of submatrix B cannot change then we know that
1034      no processor disassembled thus we can skip this stuff
1035   */
1036   if (!((Mat_SeqBAIJ*)baij->B->data)->nonew)  {
1037     ierr = MPI_Allreduce(&mat->was_assembled,&other_disassembled,1,MPI_INT,MPI_PROD,mat->comm);CHKERRQ(ierr);
1038     if (mat->was_assembled && !other_disassembled) {
1039       ierr = DisAssemble_MPIBAIJ(mat);CHKERRQ(ierr);
1040     }
1041   }
1042 
1043   if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) {
1044     ierr = MatSetUpMultiply_MPIBAIJ(mat);CHKERRQ(ierr);
1045   }
1046   ierr = MatAssemblyBegin(baij->B,mode);CHKERRQ(ierr);
1047   ierr = MatAssemblyEnd(baij->B,mode);CHKERRQ(ierr);
1048 
1049 #if defined(PETSC_USE_BOPT_g)
1050   if (baij->ht && mode== MAT_FINAL_ASSEMBLY) {
1051     PetscLogInfo(0,"MatAssemblyEnd_MPIBAIJ:Average Hash Table Search in MatSetValues = %5.2f\n",((double)baij->ht_total_ct)/baij->ht_insert_ct);
1052     baij->ht_total_ct  = 0;
1053     baij->ht_insert_ct = 0;
1054   }
1055 #endif
1056   if (baij->ht_flag && !baij->ht && mode == MAT_FINAL_ASSEMBLY) {
1057     ierr = MatCreateHashTable_MPIBAIJ_Private(mat,baij->ht_fact);CHKERRQ(ierr);
1058     mat->ops->setvalues        = MatSetValues_MPIBAIJ_HT;
1059     mat->ops->setvaluesblocked = MatSetValuesBlocked_MPIBAIJ_HT;
1060   }
1061 
1062   if (baij->rowvalues) {
1063     ierr = PetscFree(baij->rowvalues);CHKERRQ(ierr);
1064     baij->rowvalues = 0;
1065   }
1066   PetscFunctionReturn(0);
1067 }
1068 
1069 #undef __FUNCT__
1070 #define __FUNCT__ "MatView_MPIBAIJ_ASCIIorDraworSocket"
1071 static int MatView_MPIBAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer)
1072 {
1073   Mat_MPIBAIJ       *baij = (Mat_MPIBAIJ*)mat->data;
1074   int               ierr,bs = baij->bs,size = baij->size,rank = baij->rank;
1075   PetscTruth        isascii,isdraw;
1076   PetscViewer       sviewer;
1077   PetscViewerFormat format;
1078 
1079   PetscFunctionBegin;
1080   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&isascii);CHKERRQ(ierr);
1081   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);CHKERRQ(ierr);
1082   if (isascii) {
1083     ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr);
1084     if (format == PETSC_VIEWER_ASCII_INFO_LONG) {
1085       MatInfo info;
1086       ierr = MPI_Comm_rank(mat->comm,&rank);CHKERRQ(ierr);
1087       ierr = MatGetInfo(mat,MAT_LOCAL,&info);CHKERRQ(ierr);
1088       ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %d nz %d nz alloced %d bs %d mem %d\n",
1089               rank,mat->m,(int)info.nz_used*bs,(int)info.nz_allocated*bs,
1090               baij->bs,(int)info.memory);CHKERRQ(ierr);
1091       ierr = MatGetInfo(baij->A,MAT_LOCAL,&info);CHKERRQ(ierr);
1092       ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %d \n",rank,(int)info.nz_used*bs);CHKERRQ(ierr);
1093       ierr = MatGetInfo(baij->B,MAT_LOCAL,&info);CHKERRQ(ierr);
1094       ierr = PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %d \n",rank,(int)info.nz_used*bs);CHKERRQ(ierr);
1095       ierr = PetscViewerFlush(viewer);CHKERRQ(ierr);
1096       ierr = VecScatterView(baij->Mvctx,viewer);CHKERRQ(ierr);
1097       PetscFunctionReturn(0);
1098     } else if (format == PETSC_VIEWER_ASCII_INFO) {
1099       ierr = PetscViewerASCIIPrintf(viewer,"  block size is %d\n",bs);CHKERRQ(ierr);
1100       PetscFunctionReturn(0);
1101     }
1102   }
1103 
1104   if (isdraw) {
1105     PetscDraw       draw;
1106     PetscTruth isnull;
1107     ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr);
1108     ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); if (isnull) PetscFunctionReturn(0);
1109   }
1110 
1111   if (size == 1) {
1112     ierr = PetscObjectSetName((PetscObject)baij->A,mat->name);CHKERRQ(ierr);
1113     ierr = MatView(baij->A,viewer);CHKERRQ(ierr);
1114   } else {
1115     /* assemble the entire matrix onto first processor. */
1116     Mat         A;
1117     Mat_SeqBAIJ *Aloc;
1118     int         M = mat->M,N = mat->N,*ai,*aj,col,i,j,k,*rvals,mbs = baij->mbs;
1119     MatScalar   *a;
1120 
1121     if (!rank) {
1122       ierr = MatCreateMPIBAIJ(mat->comm,baij->bs,M,N,M,N,0,PETSC_NULL,0,PETSC_NULL,&A);CHKERRQ(ierr);
1123     } else {
1124       ierr = MatCreateMPIBAIJ(mat->comm,baij->bs,0,0,M,N,0,PETSC_NULL,0,PETSC_NULL,&A);CHKERRQ(ierr);
1125     }
1126     PetscLogObjectParent(mat,A);
1127 
1128     /* copy over the A part */
1129     Aloc = (Mat_SeqBAIJ*)baij->A->data;
1130     ai   = Aloc->i; aj = Aloc->j; a = Aloc->a;
1131     ierr = PetscMalloc(bs*sizeof(int),&rvals);CHKERRQ(ierr);
1132 
1133     for (i=0; i<mbs; i++) {
1134       rvals[0] = bs*(baij->rstart + i);
1135       for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
1136       for (j=ai[i]; j<ai[i+1]; j++) {
1137         col = (baij->cstart+aj[j])*bs;
1138         for (k=0; k<bs; k++) {
1139           ierr = MatSetValues_MPIBAIJ_MatScalar(A,bs,rvals,1,&col,a,INSERT_VALUES);CHKERRQ(ierr);
1140           col++; a += bs;
1141         }
1142       }
1143     }
1144     /* copy over the B part */
1145     Aloc = (Mat_SeqBAIJ*)baij->B->data;
1146     ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
1147     for (i=0; i<mbs; i++) {
1148       rvals[0] = bs*(baij->rstart + i);
1149       for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
1150       for (j=ai[i]; j<ai[i+1]; j++) {
1151         col = baij->garray[aj[j]]*bs;
1152         for (k=0; k<bs; k++) {
1153           ierr = MatSetValues_MPIBAIJ_MatScalar(A,bs,rvals,1,&col,a,INSERT_VALUES);CHKERRQ(ierr);
1154           col++; a += bs;
1155         }
1156       }
1157     }
1158     ierr = PetscFree(rvals);CHKERRQ(ierr);
1159     ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
1160     ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
1161     /*
1162        Everyone has to call to draw the matrix since the graphics waits are
1163        synchronized across all processors that share the PetscDraw object
1164     */
1165     ierr = PetscViewerGetSingleton(viewer,&sviewer);CHKERRQ(ierr);
1166     if (!rank) {
1167       ierr = PetscObjectSetName((PetscObject)((Mat_MPIBAIJ*)(A->data))->A,mat->name);CHKERRQ(ierr);
1168       ierr = MatView(((Mat_MPIBAIJ*)(A->data))->A,sviewer);CHKERRQ(ierr);
1169     }
1170     ierr = PetscViewerRestoreSingleton(viewer,&sviewer);CHKERRQ(ierr);
1171     ierr = MatDestroy(A);CHKERRQ(ierr);
1172   }
1173   PetscFunctionReturn(0);
1174 }
1175 
1176 #undef __FUNCT__
1177 #define __FUNCT__ "MatView_MPIBAIJ"
1178 int MatView_MPIBAIJ(Mat mat,PetscViewer viewer)
1179 {
1180   int        ierr;
1181   PetscTruth isascii,isdraw,issocket,isbinary;
1182 
1183   PetscFunctionBegin;
1184   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&isascii);CHKERRQ(ierr);
1185   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);CHKERRQ(ierr);
1186   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_SOCKET,&issocket);CHKERRQ(ierr);
1187   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);CHKERRQ(ierr);
1188   if (isascii || isdraw || issocket || isbinary) {
1189     ierr = MatView_MPIBAIJ_ASCIIorDraworSocket(mat,viewer);CHKERRQ(ierr);
1190   } else {
1191     SETERRQ1(1,"Viewer type %s not supported by MPIBAIJ matrices",((PetscObject)viewer)->type_name);
1192   }
1193   PetscFunctionReturn(0);
1194 }
1195 
1196 #undef __FUNCT__
1197 #define __FUNCT__ "MatDestroy_MPIBAIJ"
1198 int MatDestroy_MPIBAIJ(Mat mat)
1199 {
1200   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1201   int         ierr;
1202 
1203   PetscFunctionBegin;
1204 #if defined(PETSC_USE_LOG)
1205   PetscLogObjectState((PetscObject)mat,"Rows=%d,Cols=%d",mat->M,mat->N);
1206 #endif
1207   ierr = MatStashDestroy_Private(&mat->stash);CHKERRQ(ierr);
1208   ierr = MatStashDestroy_Private(&mat->bstash);CHKERRQ(ierr);
1209   ierr = PetscFree(baij->rowners);CHKERRQ(ierr);
1210   ierr = MatDestroy(baij->A);CHKERRQ(ierr);
1211   ierr = MatDestroy(baij->B);CHKERRQ(ierr);
1212 #if defined (PETSC_USE_CTABLE)
1213   if (baij->colmap) {ierr = PetscTableDelete(baij->colmap);CHKERRQ(ierr);}
1214 #else
1215   if (baij->colmap) {ierr = PetscFree(baij->colmap);CHKERRQ(ierr);}
1216 #endif
1217   if (baij->garray) {ierr = PetscFree(baij->garray);CHKERRQ(ierr);}
1218   if (baij->lvec)   {ierr = VecDestroy(baij->lvec);CHKERRQ(ierr);}
1219   if (baij->Mvctx)  {ierr = VecScatterDestroy(baij->Mvctx);CHKERRQ(ierr);}
1220   if (baij->rowvalues) {ierr = PetscFree(baij->rowvalues);CHKERRQ(ierr);}
1221   if (baij->barray) {ierr = PetscFree(baij->barray);CHKERRQ(ierr);}
1222   if (baij->hd) {ierr = PetscFree(baij->hd);CHKERRQ(ierr);}
1223 #if defined(PETSC_USE_MAT_SINGLE)
1224   if (baij->setvaluescopy) {ierr = PetscFree(baij->setvaluescopy);CHKERRQ(ierr);}
1225 #endif
1226   ierr = PetscFree(baij);CHKERRQ(ierr);
1227   PetscFunctionReturn(0);
1228 }
1229 
1230 #undef __FUNCT__
1231 #define __FUNCT__ "MatMult_MPIBAIJ"
1232 int MatMult_MPIBAIJ(Mat A,Vec xx,Vec yy)
1233 {
1234   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1235   int         ierr,nt;
1236 
1237   PetscFunctionBegin;
1238   ierr = VecGetLocalSize(xx,&nt);CHKERRQ(ierr);
1239   if (nt != A->n) {
1240     SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible partition of A and xx");
1241   }
1242   ierr = VecGetLocalSize(yy,&nt);CHKERRQ(ierr);
1243   if (nt != A->m) {
1244     SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible parition of A and yy");
1245   }
1246   ierr = VecScatterBegin(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);CHKERRQ(ierr);
1247   ierr = (*a->A->ops->mult)(a->A,xx,yy);CHKERRQ(ierr);
1248   ierr = VecScatterEnd(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);CHKERRQ(ierr);
1249   ierr = (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);CHKERRQ(ierr);
1250   ierr = VecScatterPostRecvs(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);CHKERRQ(ierr);
1251   PetscFunctionReturn(0);
1252 }
1253 
1254 #undef __FUNCT__
1255 #define __FUNCT__ "MatMultAdd_MPIBAIJ"
1256 int MatMultAdd_MPIBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1257 {
1258   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1259   int        ierr;
1260 
1261   PetscFunctionBegin;
1262   ierr = VecScatterBegin(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);CHKERRQ(ierr);
1263   ierr = (*a->A->ops->multadd)(a->A,xx,yy,zz);CHKERRQ(ierr);
1264   ierr = VecScatterEnd(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);CHKERRQ(ierr);
1265   ierr = (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);CHKERRQ(ierr);
1266   PetscFunctionReturn(0);
1267 }
1268 
1269 #undef __FUNCT__
1270 #define __FUNCT__ "MatMultTranspose_MPIBAIJ"
1271 int MatMultTranspose_MPIBAIJ(Mat A,Vec xx,Vec yy)
1272 {
1273   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1274   int         ierr;
1275 
1276   PetscFunctionBegin;
1277   /* do nondiagonal part */
1278   ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr);
1279   /* send it on its way */
1280   ierr = VecScatterBegin(a->lvec,yy,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);CHKERRQ(ierr);
1281   /* do local part */
1282   ierr = (*a->A->ops->multtranspose)(a->A,xx,yy);CHKERRQ(ierr);
1283   /* receive remote parts: note this assumes the values are not actually */
1284   /* inserted in yy until the next line, which is true for my implementation*/
1285   /* but is not perhaps always true. */
1286   ierr = VecScatterEnd(a->lvec,yy,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);CHKERRQ(ierr);
1287   PetscFunctionReturn(0);
1288 }
1289 
1290 #undef __FUNCT__
1291 #define __FUNCT__ "MatMultTransposeAdd_MPIBAIJ"
1292 int MatMultTransposeAdd_MPIBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1293 {
1294   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1295   int         ierr;
1296 
1297   PetscFunctionBegin;
1298   /* do nondiagonal part */
1299   ierr = (*a->B->ops->multtranspose)(a->B,xx,a->lvec);CHKERRQ(ierr);
1300   /* send it on its way */
1301   ierr = VecScatterBegin(a->lvec,zz,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);CHKERRQ(ierr);
1302   /* do local part */
1303   ierr = (*a->A->ops->multtransposeadd)(a->A,xx,yy,zz);CHKERRQ(ierr);
1304   /* receive remote parts: note this assumes the values are not actually */
1305   /* inserted in yy until the next line, which is true for my implementation*/
1306   /* but is not perhaps always true. */
1307   ierr = VecScatterEnd(a->lvec,zz,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);CHKERRQ(ierr);
1308   PetscFunctionReturn(0);
1309 }
1310 
1311 /*
1312   This only works correctly for square matrices where the subblock A->A is the
1313    diagonal block
1314 */
1315 #undef __FUNCT__
1316 #define __FUNCT__ "MatGetDiagonal_MPIBAIJ"
1317 int MatGetDiagonal_MPIBAIJ(Mat A,Vec v)
1318 {
1319   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1320   int         ierr;
1321 
1322   PetscFunctionBegin;
1323   if (A->M != A->N) SETERRQ(PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block");
1324   ierr = MatGetDiagonal(a->A,v);CHKERRQ(ierr);
1325   PetscFunctionReturn(0);
1326 }
1327 
1328 #undef __FUNCT__
1329 #define __FUNCT__ "MatScale_MPIBAIJ"
1330 int MatScale_MPIBAIJ(Scalar *aa,Mat A)
1331 {
1332   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1333   int         ierr;
1334 
1335   PetscFunctionBegin;
1336   ierr = MatScale(aa,a->A);CHKERRQ(ierr);
1337   ierr = MatScale(aa,a->B);CHKERRQ(ierr);
1338   PetscFunctionReturn(0);
1339 }
1340 
1341 #undef __FUNCT__
1342 #define __FUNCT__ "MatGetOwnershipRange_MPIBAIJ"
1343 int MatGetOwnershipRange_MPIBAIJ(Mat matin,int *m,int *n)
1344 {
1345   Mat_MPIBAIJ *mat = (Mat_MPIBAIJ*)matin->data;
1346 
1347   PetscFunctionBegin;
1348   if (m) *m = mat->rstart_bs;
1349   if (n) *n = mat->rend_bs;
1350   PetscFunctionReturn(0);
1351 }
1352 
1353 #undef __FUNCT__
1354 #define __FUNCT__ "MatGetRow_MPIBAIJ"
1355 int MatGetRow_MPIBAIJ(Mat matin,int row,int *nz,int **idx,Scalar **v)
1356 {
1357   Mat_MPIBAIJ *mat = (Mat_MPIBAIJ*)matin->data;
1358   Scalar     *vworkA,*vworkB,**pvA,**pvB,*v_p;
1359   int        bs = mat->bs,bs2 = mat->bs2,i,ierr,*cworkA,*cworkB,**pcA,**pcB;
1360   int        nztot,nzA,nzB,lrow,brstart = mat->rstart*bs,brend = mat->rend*bs;
1361   int        *cmap,*idx_p,cstart = mat->cstart;
1362 
1363   PetscFunctionBegin;
1364   if (mat->getrowactive == PETSC_TRUE) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Already active");
1365   mat->getrowactive = PETSC_TRUE;
1366 
1367   if (!mat->rowvalues && (idx || v)) {
1368     /*
1369         allocate enough space to hold information from the longest row.
1370     */
1371     Mat_SeqBAIJ *Aa = (Mat_SeqBAIJ*)mat->A->data,*Ba = (Mat_SeqBAIJ*)mat->B->data;
1372     int     max = 1,mbs = mat->mbs,tmp;
1373     for (i=0; i<mbs; i++) {
1374       tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i];
1375       if (max < tmp) { max = tmp; }
1376     }
1377     ierr = PetscMalloc(max*bs2*(sizeof(int)+sizeof(Scalar)),&mat->rowvalues);CHKERRQ(ierr);
1378     mat->rowindices = (int*)(mat->rowvalues + max*bs2);
1379   }
1380 
1381   if (row < brstart || row >= brend) SETERRQ(PETSC_ERR_SUP,"Only local rows")
1382   lrow = row - brstart;
1383 
1384   pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB;
1385   if (!v)   {pvA = 0; pvB = 0;}
1386   if (!idx) {pcA = 0; if (!v) pcB = 0;}
1387   ierr = (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr);
1388   ierr = (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr);
1389   nztot = nzA + nzB;
1390 
1391   cmap  = mat->garray;
1392   if (v  || idx) {
1393     if (nztot) {
1394       /* Sort by increasing column numbers, assuming A and B already sorted */
1395       int imark = -1;
1396       if (v) {
1397         *v = v_p = mat->rowvalues;
1398         for (i=0; i<nzB; i++) {
1399           if (cmap[cworkB[i]/bs] < cstart)   v_p[i] = vworkB[i];
1400           else break;
1401         }
1402         imark = i;
1403         for (i=0; i<nzA; i++)     v_p[imark+i] = vworkA[i];
1404         for (i=imark; i<nzB; i++) v_p[nzA+i]   = vworkB[i];
1405       }
1406       if (idx) {
1407         *idx = idx_p = mat->rowindices;
1408         if (imark > -1) {
1409           for (i=0; i<imark; i++) {
1410             idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs;
1411           }
1412         } else {
1413           for (i=0; i<nzB; i++) {
1414             if (cmap[cworkB[i]/bs] < cstart)
1415               idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1416             else break;
1417           }
1418           imark = i;
1419         }
1420         for (i=0; i<nzA; i++)     idx_p[imark+i] = cstart*bs + cworkA[i];
1421         for (i=imark; i<nzB; i++) idx_p[nzA+i]   = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1422       }
1423     } else {
1424       if (idx) *idx = 0;
1425       if (v)   *v   = 0;
1426     }
1427   }
1428   *nz = nztot;
1429   ierr = (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);CHKERRQ(ierr);
1430   ierr = (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);CHKERRQ(ierr);
1431   PetscFunctionReturn(0);
1432 }
1433 
1434 #undef __FUNCT__
1435 #define __FUNCT__ "MatRestoreRow_MPIBAIJ"
1436 int MatRestoreRow_MPIBAIJ(Mat mat,int row,int *nz,int **idx,Scalar **v)
1437 {
1438   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1439 
1440   PetscFunctionBegin;
1441   if (baij->getrowactive == PETSC_FALSE) {
1442     SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"MatGetRow not called");
1443   }
1444   baij->getrowactive = PETSC_FALSE;
1445   PetscFunctionReturn(0);
1446 }
1447 
1448 #undef __FUNCT__
1449 #define __FUNCT__ "MatGetBlockSize_MPIBAIJ"
1450 int MatGetBlockSize_MPIBAIJ(Mat mat,int *bs)
1451 {
1452   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1453 
1454   PetscFunctionBegin;
1455   *bs = baij->bs;
1456   PetscFunctionReturn(0);
1457 }
1458 
1459 #undef __FUNCT__
1460 #define __FUNCT__ "MatZeroEntries_MPIBAIJ"
1461 int MatZeroEntries_MPIBAIJ(Mat A)
1462 {
1463   Mat_MPIBAIJ *l = (Mat_MPIBAIJ*)A->data;
1464   int         ierr;
1465 
1466   PetscFunctionBegin;
1467   ierr = MatZeroEntries(l->A);CHKERRQ(ierr);
1468   ierr = MatZeroEntries(l->B);CHKERRQ(ierr);
1469   PetscFunctionReturn(0);
1470 }
1471 
1472 #undef __FUNCT__
1473 #define __FUNCT__ "MatGetInfo_MPIBAIJ"
1474 int MatGetInfo_MPIBAIJ(Mat matin,MatInfoType flag,MatInfo *info)
1475 {
1476   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)matin->data;
1477   Mat         A = a->A,B = a->B;
1478   int         ierr;
1479   PetscReal   isend[5],irecv[5];
1480 
1481   PetscFunctionBegin;
1482   info->block_size     = (double)a->bs;
1483   ierr = MatGetInfo(A,MAT_LOCAL,info);CHKERRQ(ierr);
1484   isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded;
1485   isend[3] = info->memory;  isend[4] = info->mallocs;
1486   ierr = MatGetInfo(B,MAT_LOCAL,info);CHKERRQ(ierr);
1487   isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded;
1488   isend[3] += info->memory;  isend[4] += info->mallocs;
1489   if (flag == MAT_LOCAL) {
1490     info->nz_used      = isend[0];
1491     info->nz_allocated = isend[1];
1492     info->nz_unneeded  = isend[2];
1493     info->memory       = isend[3];
1494     info->mallocs      = isend[4];
1495   } else if (flag == MAT_GLOBAL_MAX) {
1496     ierr = MPI_Allreduce(isend,irecv,5,MPI_DOUBLE,MPI_MAX,matin->comm);CHKERRQ(ierr);
1497     info->nz_used      = irecv[0];
1498     info->nz_allocated = irecv[1];
1499     info->nz_unneeded  = irecv[2];
1500     info->memory       = irecv[3];
1501     info->mallocs      = irecv[4];
1502   } else if (flag == MAT_GLOBAL_SUM) {
1503     ierr = MPI_Allreduce(isend,irecv,5,MPI_DOUBLE,MPI_SUM,matin->comm);CHKERRQ(ierr);
1504     info->nz_used      = irecv[0];
1505     info->nz_allocated = irecv[1];
1506     info->nz_unneeded  = irecv[2];
1507     info->memory       = irecv[3];
1508     info->mallocs      = irecv[4];
1509   } else {
1510     SETERRQ1(1,"Unknown MatInfoType argument %d",flag);
1511   }
1512   info->rows_global       = (double)A->M;
1513   info->columns_global    = (double)A->N;
1514   info->rows_local        = (double)A->m;
1515   info->columns_local     = (double)A->N;
1516   info->fill_ratio_given  = 0; /* no parallel LU/ILU/Cholesky */
1517   info->fill_ratio_needed = 0;
1518   info->factor_mallocs    = 0;
1519   PetscFunctionReturn(0);
1520 }
1521 
1522 #undef __FUNCT__
1523 #define __FUNCT__ "MatSetOption_MPIBAIJ"
1524 int MatSetOption_MPIBAIJ(Mat A,MatOption op)
1525 {
1526   Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1527   int         ierr;
1528 
1529   PetscFunctionBegin;
1530   switch (op) {
1531   case MAT_NO_NEW_NONZERO_LOCATIONS:
1532   case MAT_YES_NEW_NONZERO_LOCATIONS:
1533   case MAT_COLUMNS_UNSORTED:
1534   case MAT_COLUMNS_SORTED:
1535   case MAT_NEW_NONZERO_ALLOCATION_ERR:
1536   case MAT_KEEP_ZEROED_ROWS:
1537   case MAT_NEW_NONZERO_LOCATION_ERR:
1538     ierr = MatSetOption(a->A,op);CHKERRQ(ierr);
1539     ierr = MatSetOption(a->B,op);CHKERRQ(ierr);
1540     break;
1541   case MAT_ROW_ORIENTED:
1542     a->roworiented = PETSC_TRUE;
1543     ierr = MatSetOption(a->A,op);CHKERRQ(ierr);
1544     ierr = MatSetOption(a->B,op);CHKERRQ(ierr);
1545     break;
1546   case MAT_ROWS_SORTED:
1547   case MAT_ROWS_UNSORTED:
1548   case MAT_YES_NEW_DIAGONALS:
1549   case MAT_USE_HASH_TABLE:
1550   case MAT_USE_SINGLE_PRECISION_SOLVES:
1551     PetscLogInfo(A,"Info:MatSetOption_MPIBAIJ:Option ignored\n");
1552     break;
1553   case MAT_COLUMN_ORIENTED:
1554     a->roworiented = PETSC_FALSE;
1555     ierr = MatSetOption(a->A,op);CHKERRQ(ierr);
1556     ierr = MatSetOption(a->B,op);CHKERRQ(ierr);
1557     break;
1558   case MAT_IGNORE_OFF_PROC_ENTRIES:
1559     a->donotstash = PETSC_TRUE;
1560     break;
1561   case MAT_NO_NEW_DIAGONALS:
1562     SETERRQ(PETSC_ERR_SUP,"MAT_NO_NEW_DIAGONALS");
1563     break;
1564   case MAT_USE_HASH_TABLE:
1565     a->ht_flag = PETSC_TRUE;
1566     break;
1567   default:
1568     SETERRQ(PETSC_ERR_SUP,"unknown option");
1569     break;
1570   }
1571   PetscFunctionReturn(0);
1572 }
1573 
1574 #undef __FUNCT__
1575 #define __FUNCT__ "MatTranspose_MPIBAIJ("
1576 int MatTranspose_MPIBAIJ(Mat A,Mat *matout)
1577 {
1578   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)A->data;
1579   Mat_SeqBAIJ *Aloc;
1580   Mat         B;
1581   int         ierr,M=A->M,N=A->N,*ai,*aj,i,*rvals,j,k,col;
1582   int         bs=baij->bs,mbs=baij->mbs;
1583   MatScalar   *a;
1584 
1585   PetscFunctionBegin;
1586   if (!matout && M != N) SETERRQ(PETSC_ERR_ARG_SIZ,"Square matrix only for in-place");
1587   ierr = MatCreateMPIBAIJ(A->comm,baij->bs,A->n,A->m,N,M,0,PETSC_NULL,0,PETSC_NULL,&B);CHKERRQ(ierr);
1588 
1589   /* copy over the A part */
1590   Aloc = (Mat_SeqBAIJ*)baij->A->data;
1591   ai   = Aloc->i; aj = Aloc->j; a = Aloc->a;
1592   ierr = PetscMalloc(bs*sizeof(int),&rvals);CHKERRQ(ierr);
1593 
1594   for (i=0; i<mbs; i++) {
1595     rvals[0] = bs*(baij->rstart + i);
1596     for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
1597     for (j=ai[i]; j<ai[i+1]; j++) {
1598       col = (baij->cstart+aj[j])*bs;
1599       for (k=0; k<bs; k++) {
1600         ierr = MatSetValues_MPIBAIJ_MatScalar(B,1,&col,bs,rvals,a,INSERT_VALUES);CHKERRQ(ierr);
1601         col++; a += bs;
1602       }
1603     }
1604   }
1605   /* copy over the B part */
1606   Aloc = (Mat_SeqBAIJ*)baij->B->data;
1607   ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
1608   for (i=0; i<mbs; i++) {
1609     rvals[0] = bs*(baij->rstart + i);
1610     for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
1611     for (j=ai[i]; j<ai[i+1]; j++) {
1612       col = baij->garray[aj[j]]*bs;
1613       for (k=0; k<bs; k++) {
1614         ierr = MatSetValues_MPIBAIJ_MatScalar(B,1,&col,bs,rvals,a,INSERT_VALUES);CHKERRQ(ierr);
1615         col++; a += bs;
1616       }
1617     }
1618   }
1619   ierr = PetscFree(rvals);CHKERRQ(ierr);
1620   ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
1621   ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
1622 
1623   if (matout) {
1624     *matout = B;
1625   } else {
1626     ierr = MatHeaderCopy(A,B);CHKERRQ(ierr);
1627   }
1628   PetscFunctionReturn(0);
1629 }
1630 
1631 #undef __FUNCT__
1632 #define __FUNCT__ "MatDiagonalScale_MPIBAIJ"
1633 int MatDiagonalScale_MPIBAIJ(Mat mat,Vec ll,Vec rr)
1634 {
1635   Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1636   Mat         a = baij->A,b = baij->B;
1637   int         ierr,s1,s2,s3;
1638 
1639   PetscFunctionBegin;
1640   ierr = MatGetLocalSize(mat,&s2,&s3);CHKERRQ(ierr);
1641   if (rr) {
1642     ierr = VecGetLocalSize(rr,&s1);CHKERRQ(ierr);
1643     if (s1!=s3) SETERRQ(PETSC_ERR_ARG_SIZ,"right vector non-conforming local size");
1644     /* Overlap communication with computation. */
1645     ierr = VecScatterBegin(rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD,baij->Mvctx);CHKERRQ(ierr);
1646   }
1647   if (ll) {
1648     ierr = VecGetLocalSize(ll,&s1);CHKERRQ(ierr);
1649     if (s1!=s2) SETERRQ(PETSC_ERR_ARG_SIZ,"left vector non-conforming local size");
1650     ierr = (*b->ops->diagonalscale)(b,ll,PETSC_NULL);CHKERRQ(ierr);
1651   }
1652   /* scale  the diagonal block */
1653   ierr = (*a->ops->diagonalscale)(a,ll,rr);CHKERRQ(ierr);
1654 
1655   if (rr) {
1656     /* Do a scatter end and then right scale the off-diagonal block */
1657     ierr = VecScatterEnd(rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD,baij->Mvctx);CHKERRQ(ierr);
1658     ierr = (*b->ops->diagonalscale)(b,PETSC_NULL,baij->lvec);CHKERRQ(ierr);
1659   }
1660 
1661   PetscFunctionReturn(0);
1662 }
1663 
1664 #undef __FUNCT__
1665 #define __FUNCT__ "MatZeroRows_MPIBAIJ"
1666 int MatZeroRows_MPIBAIJ(Mat A,IS is,Scalar *diag)
1667 {
1668   Mat_MPIBAIJ    *l = (Mat_MPIBAIJ*)A->data;
1669   int            i,ierr,N,*rows,*owners = l->rowners,size = l->size;
1670   int            *procs,*nprocs,j,idx,nsends,*work,row;
1671   int            nmax,*svalues,*starts,*owner,nrecvs,rank = l->rank;
1672   int            *rvalues,tag = A->tag,count,base,slen,n,*source;
1673   int            *lens,imdex,*lrows,*values,bs=l->bs,rstart_bs=l->rstart_bs;
1674   MPI_Comm       comm = A->comm;
1675   MPI_Request    *send_waits,*recv_waits;
1676   MPI_Status     recv_status,*send_status;
1677   IS             istmp;
1678   PetscTruth     found;
1679 
1680   PetscFunctionBegin;
1681   ierr = ISGetLocalSize(is,&N);CHKERRQ(ierr);
1682   ierr = ISGetIndices(is,&rows);CHKERRQ(ierr);
1683 
1684   /*  first count number of contributors to each processor */
1685   ierr  = PetscMalloc(2*size*sizeof(int),&nprocs);CHKERRQ(ierr);
1686   ierr  = PetscMemzero(nprocs,2*size*sizeof(int));CHKERRQ(ierr);
1687   procs = nprocs + size;
1688   ierr  = PetscMalloc((N+1)*sizeof(int),&owner);CHKERRQ(ierr); /* see note*/
1689   for (i=0; i<N; i++) {
1690     idx   = rows[i];
1691     found = PETSC_FALSE;
1692     for (j=0; j<size; j++) {
1693       if (idx >= owners[j]*bs && idx < owners[j+1]*bs) {
1694         nprocs[j]++; procs[j] = 1; owner[i] = j; found = PETSC_TRUE; break;
1695       }
1696     }
1697     if (!found) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Index out of range");
1698   }
1699   nsends = 0;  for (i=0; i<size; i++) { nsends += procs[i];}
1700 
1701   /* inform other processors of number of messages and max length*/
1702   ierr   = PetscMalloc(2*size*sizeof(int),&work);CHKERRQ(ierr);
1703   ierr   = MPI_Allreduce(nprocs,work,2*size,MPI_INT,PetscMaxSum_Op,comm);CHKERRQ(ierr);
1704   nmax   = work[rank];
1705   nrecvs = work[size+rank];
1706   ierr   = PetscFree(work);CHKERRQ(ierr);
1707 
1708   /* post receives:   */
1709   ierr = PetscMalloc((nrecvs+1)*(nmax+1)*sizeof(int),&rvalues);CHKERRQ(ierr);
1710   ierr = PetscMalloc((nrecvs+1)*sizeof(MPI_Request),&recv_waits);CHKERRQ(ierr);
1711   for (i=0; i<nrecvs; i++) {
1712     ierr = MPI_Irecv(rvalues+nmax*i,nmax,MPI_INT,MPI_ANY_SOURCE,tag,comm,recv_waits+i);CHKERRQ(ierr);
1713   }
1714 
1715   /* do sends:
1716      1) starts[i] gives the starting index in svalues for stuff going to
1717      the ith processor
1718   */
1719   ierr = PetscMalloc((N+1)*sizeof(int),&svalues);CHKERRQ(ierr);
1720   ierr = PetscMalloc((nsends+1)*sizeof(MPI_Request),&send_waits);CHKERRQ(ierr);
1721   ierr = PetscMalloc((size+1)*sizeof(int),&starts);CHKERRQ(ierr);
1722   starts[0]  = 0;
1723   for (i=1; i<size; i++) { starts[i] = starts[i-1] + nprocs[i-1];}
1724   for (i=0; i<N; i++) {
1725     svalues[starts[owner[i]]++] = rows[i];
1726   }
1727   ierr = ISRestoreIndices(is,&rows);CHKERRQ(ierr);
1728 
1729   starts[0] = 0;
1730   for (i=1; i<size+1; i++) { starts[i] = starts[i-1] + nprocs[i-1];}
1731   count = 0;
1732   for (i=0; i<size; i++) {
1733     if (procs[i]) {
1734       ierr = MPI_Isend(svalues+starts[i],nprocs[i],MPI_INT,i,tag,comm,send_waits+count++);CHKERRQ(ierr);
1735     }
1736   }
1737   ierr = PetscFree(starts);CHKERRQ(ierr);
1738 
1739   base = owners[rank]*bs;
1740 
1741   /*  wait on receives */
1742   ierr   = PetscMalloc(2*(nrecvs+1)*sizeof(int),&lens);CHKERRQ(ierr);
1743   source = lens + nrecvs;
1744   count  = nrecvs; slen = 0;
1745   while (count) {
1746     ierr = MPI_Waitany(nrecvs,recv_waits,&imdex,&recv_status);CHKERRQ(ierr);
1747     /* unpack receives into our local space */
1748     ierr = MPI_Get_count(&recv_status,MPI_INT,&n);CHKERRQ(ierr);
1749     source[imdex]  = recv_status.MPI_SOURCE;
1750     lens[imdex]    = n;
1751     slen          += n;
1752     count--;
1753   }
1754   ierr = PetscFree(recv_waits);CHKERRQ(ierr);
1755 
1756   /* move the data into the send scatter */
1757   ierr = PetscMalloc((slen+1)*sizeof(int),&lrows);CHKERRQ(ierr);
1758   count = 0;
1759   for (i=0; i<nrecvs; i++) {
1760     values = rvalues + i*nmax;
1761     for (j=0; j<lens[i]; j++) {
1762       lrows[count++] = values[j] - base;
1763     }
1764   }
1765   ierr = PetscFree(rvalues);CHKERRQ(ierr);
1766   ierr = PetscFree(lens);CHKERRQ(ierr);
1767   ierr = PetscFree(owner);CHKERRQ(ierr);
1768   ierr = PetscFree(nprocs);CHKERRQ(ierr);
1769 
1770   /* actually zap the local rows */
1771   ierr = ISCreateGeneral(PETSC_COMM_SELF,slen,lrows,&istmp);CHKERRQ(ierr);
1772   PetscLogObjectParent(A,istmp);
1773 
1774   /*
1775         Zero the required rows. If the "diagonal block" of the matrix
1776      is square and the user wishes to set the diagonal we use seperate
1777      code so that MatSetValues() is not called for each diagonal allocating
1778      new memory, thus calling lots of mallocs and slowing things down.
1779 
1780        Contributed by: Mathew Knepley
1781   */
1782   /* must zero l->B before l->A because the (diag) case below may put values into l->B*/
1783   ierr = MatZeroRows_SeqBAIJ(l->B,istmp,0);CHKERRQ(ierr);
1784   if (diag && (l->A->M == l->A->N)) {
1785     ierr = MatZeroRows_SeqBAIJ(l->A,istmp,diag);CHKERRQ(ierr);
1786   } else if (diag) {
1787     ierr = MatZeroRows_SeqBAIJ(l->A,istmp,0);CHKERRQ(ierr);
1788     if (((Mat_SeqBAIJ*)l->A->data)->nonew) {
1789       SETERRQ(PETSC_ERR_SUP,"MatZeroRows() on rectangular matrices cannot be used with the Mat options \n\
1790 MAT_NO_NEW_NONZERO_LOCATIONS,MAT_NEW_NONZERO_LOCATION_ERR,MAT_NEW_NONZERO_ALLOCATION_ERR");
1791     }
1792     for (i=0; i<slen; i++) {
1793       row  = lrows[i] + rstart_bs;
1794       ierr = MatSetValues(A,1,&row,1,&row,diag,INSERT_VALUES);CHKERRQ(ierr);
1795     }
1796     ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
1797     ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
1798   } else {
1799     ierr = MatZeroRows_SeqBAIJ(l->A,istmp,0);CHKERRQ(ierr);
1800   }
1801 
1802   ierr = ISDestroy(istmp);CHKERRQ(ierr);
1803   ierr = PetscFree(lrows);CHKERRQ(ierr);
1804 
1805   /* wait on sends */
1806   if (nsends) {
1807     ierr = PetscMalloc(nsends*sizeof(MPI_Status),&send_status);CHKERRQ(ierr);
1808     ierr = MPI_Waitall(nsends,send_waits,send_status);CHKERRQ(ierr);
1809     ierr = PetscFree(send_status);CHKERRQ(ierr);
1810   }
1811   ierr = PetscFree(send_waits);CHKERRQ(ierr);
1812   ierr = PetscFree(svalues);CHKERRQ(ierr);
1813 
1814   PetscFunctionReturn(0);
1815 }
1816 
1817 #undef __FUNCT__
1818 #define __FUNCT__ "MatPrintHelp_MPIBAIJ"
1819 int MatPrintHelp_MPIBAIJ(Mat A)
1820 {
1821   Mat_MPIBAIJ *a   = (Mat_MPIBAIJ*)A->data;
1822   MPI_Comm    comm = A->comm;
1823   static int  called = 0;
1824   int         ierr;
1825 
1826   PetscFunctionBegin;
1827   if (!a->rank) {
1828     ierr = MatPrintHelp_SeqBAIJ(a->A);CHKERRQ(ierr);
1829   }
1830   if (called) {PetscFunctionReturn(0);} else called = 1;
1831   ierr = (*PetscHelpPrintf)(comm," Options for MATMPIBAIJ matrix format (the defaults):\n");CHKERRQ(ierr);
1832   ierr = (*PetscHelpPrintf)(comm,"  -mat_use_hash_table <factor>: Use hashtable for efficient matrix assembly\n");CHKERRQ(ierr);
1833   PetscFunctionReturn(0);
1834 }
1835 
1836 #undef __FUNCT__
1837 #define __FUNCT__ "MatSetUnfactored_MPIBAIJ"
1838 int MatSetUnfactored_MPIBAIJ(Mat A)
1839 {
1840   Mat_MPIBAIJ *a   = (Mat_MPIBAIJ*)A->data;
1841   int         ierr;
1842 
1843   PetscFunctionBegin;
1844   ierr = MatSetUnfactored(a->A);CHKERRQ(ierr);
1845   PetscFunctionReturn(0);
1846 }
1847 
1848 static int MatDuplicate_MPIBAIJ(Mat,MatDuplicateOption,Mat *);
1849 
1850 #undef __FUNCT__
1851 #define __FUNCT__ "MatEqual_MPIBAIJ"
1852 int MatEqual_MPIBAIJ(Mat A,Mat B,PetscTruth *flag)
1853 {
1854   Mat_MPIBAIJ *matB = (Mat_MPIBAIJ*)B->data,*matA = (Mat_MPIBAIJ*)A->data;
1855   Mat         a,b,c,d;
1856   PetscTruth  flg;
1857   int         ierr;
1858 
1859   PetscFunctionBegin;
1860   ierr = PetscTypeCompare((PetscObject)B,MATMPIBAIJ,&flg);CHKERRQ(ierr);
1861   if (!flg) SETERRQ(PETSC_ERR_ARG_INCOMP,"Matrices must be same type");
1862   a = matA->A; b = matA->B;
1863   c = matB->A; d = matB->B;
1864 
1865   ierr = MatEqual(a,c,&flg);CHKERRQ(ierr);
1866   if (flg == PETSC_TRUE) {
1867     ierr = MatEqual(b,d,&flg);CHKERRQ(ierr);
1868   }
1869   ierr = MPI_Allreduce(&flg,flag,1,MPI_INT,MPI_LAND,A->comm);CHKERRQ(ierr);
1870   PetscFunctionReturn(0);
1871 }
1872 
1873 
1874 #undef __FUNCT__
1875 #define __FUNCT__ "MatSetUpPreallocation_MPIBAIJ"
1876 int MatSetUpPreallocation_MPIBAIJ(Mat A)
1877 {
1878   int        ierr;
1879 
1880   PetscFunctionBegin;
1881   ierr =  MatMPIBAIJSetPreallocation(A,1,PETSC_DEFAULT,0,PETSC_DEFAULT,0);CHKERRQ(ierr);
1882   PetscFunctionReturn(0);
1883 }
1884 
1885 /* -------------------------------------------------------------------*/
1886 static struct _MatOps MatOps_Values = {
1887   MatSetValues_MPIBAIJ,
1888   MatGetRow_MPIBAIJ,
1889   MatRestoreRow_MPIBAIJ,
1890   MatMult_MPIBAIJ,
1891   MatMultAdd_MPIBAIJ,
1892   MatMultTranspose_MPIBAIJ,
1893   MatMultTransposeAdd_MPIBAIJ,
1894   0,
1895   0,
1896   0,
1897   0,
1898   0,
1899   0,
1900   0,
1901   MatTranspose_MPIBAIJ,
1902   MatGetInfo_MPIBAIJ,
1903   MatEqual_MPIBAIJ,
1904   MatGetDiagonal_MPIBAIJ,
1905   MatDiagonalScale_MPIBAIJ,
1906   MatNorm_MPIBAIJ,
1907   MatAssemblyBegin_MPIBAIJ,
1908   MatAssemblyEnd_MPIBAIJ,
1909   0,
1910   MatSetOption_MPIBAIJ,
1911   MatZeroEntries_MPIBAIJ,
1912   MatZeroRows_MPIBAIJ,
1913   0,
1914   0,
1915   0,
1916   0,
1917   MatSetUpPreallocation_MPIBAIJ,
1918   0,
1919   MatGetOwnershipRange_MPIBAIJ,
1920   0,
1921   0,
1922   0,
1923   0,
1924   MatDuplicate_MPIBAIJ,
1925   0,
1926   0,
1927   0,
1928   0,
1929   0,
1930   MatGetSubMatrices_MPIBAIJ,
1931   MatIncreaseOverlap_MPIBAIJ,
1932   MatGetValues_MPIBAIJ,
1933   0,
1934   MatPrintHelp_MPIBAIJ,
1935   MatScale_MPIBAIJ,
1936   0,
1937   0,
1938   0,
1939   MatGetBlockSize_MPIBAIJ,
1940   0,
1941   0,
1942   0,
1943   0,
1944   0,
1945   0,
1946   MatSetUnfactored_MPIBAIJ,
1947   0,
1948   MatSetValuesBlocked_MPIBAIJ,
1949   0,
1950   MatDestroy_MPIBAIJ,
1951   MatView_MPIBAIJ,
1952   MatGetMaps_Petsc,
1953   0,
1954   0,
1955   0,
1956   0,
1957   0,
1958   0,
1959   MatGetRowMax_MPIBAIJ};
1960 
1961 
1962 EXTERN_C_BEGIN
1963 #undef __FUNCT__
1964 #define __FUNCT__ "MatGetDiagonalBlock_MPIBAIJ"
1965 int MatGetDiagonalBlock_MPIBAIJ(Mat A,PetscTruth *iscopy,MatReuse reuse,Mat *a)
1966 {
1967   PetscFunctionBegin;
1968   *a      = ((Mat_MPIBAIJ *)A->data)->A;
1969   *iscopy = PETSC_FALSE;
1970   PetscFunctionReturn(0);
1971 }
1972 EXTERN_C_END
1973 
1974 EXTERN_C_BEGIN
1975 #undef __FUNCT__
1976 #define __FUNCT__ "MatCreate_MPIBAIJ"
1977 int MatCreate_MPIBAIJ(Mat B)
1978 {
1979   Mat_MPIBAIJ  *b;
1980   int          ierr;
1981   PetscTruth   flg;
1982 
1983   PetscFunctionBegin;
1984 
1985   ierr = PetscNew(Mat_MPIBAIJ,&b);CHKERRQ(ierr);
1986   B->data = (void*)b;
1987 
1988   ierr    = PetscMemzero(b,sizeof(Mat_MPIBAIJ));CHKERRQ(ierr);
1989   ierr    = PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));CHKERRQ(ierr);
1990   B->mapping    = 0;
1991   B->factor     = 0;
1992   B->assembled  = PETSC_FALSE;
1993 
1994   B->insertmode = NOT_SET_VALUES;
1995   ierr = MPI_Comm_rank(B->comm,&b->rank);CHKERRQ(ierr);
1996   ierr = MPI_Comm_size(B->comm,&b->size);CHKERRQ(ierr);
1997 
1998   /* build local table of row and column ownerships */
1999   ierr          = PetscMalloc(3*(b->size+2)*sizeof(int),&b->rowners);CHKERRQ(ierr);
2000   PetscLogObjectMemory(B,3*(b->size+2)*sizeof(int)+sizeof(struct _p_Mat)+sizeof(Mat_MPIBAIJ));
2001   b->cowners    = b->rowners + b->size + 2;
2002   b->rowners_bs = b->cowners + b->size + 2;
2003 
2004   /* build cache for off array entries formed */
2005   ierr = MatStashCreate_Private(B->comm,1,&B->stash);CHKERRQ(ierr);
2006   b->donotstash  = PETSC_FALSE;
2007   b->colmap      = PETSC_NULL;
2008   b->garray      = PETSC_NULL;
2009   b->roworiented = PETSC_TRUE;
2010 
2011 #if defined(PETSC_USE_MAT_SINGLE)
2012   /* stuff for MatSetValues_XXX in single precision */
2013   b->setvalueslen     = 0;
2014   b->setvaluescopy    = PETSC_NULL;
2015 #endif
2016 
2017   /* stuff used in block assembly */
2018   b->barray       = 0;
2019 
2020   /* stuff used for matrix vector multiply */
2021   b->lvec         = 0;
2022   b->Mvctx        = 0;
2023 
2024   /* stuff for MatGetRow() */
2025   b->rowindices   = 0;
2026   b->rowvalues    = 0;
2027   b->getrowactive = PETSC_FALSE;
2028 
2029   /* hash table stuff */
2030   b->ht           = 0;
2031   b->hd           = 0;
2032   b->ht_size      = 0;
2033   b->ht_flag      = PETSC_FALSE;
2034   b->ht_fact      = 0;
2035   b->ht_total_ct  = 0;
2036   b->ht_insert_ct = 0;
2037 
2038   ierr = PetscOptionsHasName(PETSC_NULL,"-mat_use_hash_table",&flg);CHKERRQ(ierr);
2039   if (flg) {
2040     double fact = 1.39;
2041     ierr = MatSetOption(B,MAT_USE_HASH_TABLE);CHKERRQ(ierr);
2042     ierr = PetscOptionsGetDouble(PETSC_NULL,"-mat_use_hash_table",&fact,PETSC_NULL);CHKERRQ(ierr);
2043     if (fact <= 1.0) fact = 1.39;
2044     ierr = MatMPIBAIJSetHashTableFactor(B,fact);CHKERRQ(ierr);
2045     PetscLogInfo(0,"MatCreateMPIBAIJ:Hash table Factor used %5.2f\n",fact);
2046   }
2047   ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C",
2048                                      "MatStoreValues_MPIBAIJ",
2049                                      MatStoreValues_MPIBAIJ);CHKERRQ(ierr);
2050   ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C",
2051                                      "MatRetrieveValues_MPIBAIJ",
2052                                      MatRetrieveValues_MPIBAIJ);CHKERRQ(ierr);
2053   ierr = PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetDiagonalBlock_C",
2054                                      "MatGetDiagonalBlock_MPIBAIJ",
2055                                      MatGetDiagonalBlock_MPIBAIJ);CHKERRQ(ierr);
2056   PetscFunctionReturn(0);
2057 }
2058 EXTERN_C_END
2059 
2060 #undef __FUNCT__
2061 #define __FUNCT__ "MatMPIBAIJSetPreallocation"
2062 /*@C
2063    MatMPIBAIJSetPreallocation - Creates a sparse parallel matrix in block AIJ format
2064    (block compressed row).  For good matrix assembly performance
2065    the user should preallocate the matrix storage by setting the parameters
2066    d_nz (or d_nnz) and o_nz (or o_nnz).  By setting these parameters accurately,
2067    performance can be increased by more than a factor of 50.
2068 
2069    Collective on Mat
2070 
2071    Input Parameters:
2072 +  A - the matrix
2073 .  bs   - size of blockk
2074 .  d_nz  - number of block nonzeros per block row in diagonal portion of local
2075            submatrix  (same for all local rows)
2076 .  d_nnz - array containing the number of block nonzeros in the various block rows
2077            of the in diagonal portion of the local (possibly different for each block
2078            row) or PETSC_NULL.  You must leave room for the diagonal entry even if it is zero.
2079 .  o_nz  - number of block nonzeros per block row in the off-diagonal portion of local
2080            submatrix (same for all local rows).
2081 -  o_nnz - array containing the number of nonzeros in the various block rows of the
2082            off-diagonal portion of the local submatrix (possibly different for
2083            each block row) or PETSC_NULL.
2084 
2085    Output Parameter:
2086 
2087 
2088    Options Database Keys:
2089 .   -mat_no_unroll - uses code that does not unroll the loops in the
2090                      block calculations (much slower)
2091 .   -mat_block_size - size of the blocks to use
2092 
2093    Notes:
2094    If PETSC_DECIDE or  PETSC_DETERMINE is used for a particular argument on one processor
2095    than it must be used on all processors that share the object for that argument.
2096 
2097    Storage Information:
2098    For a square global matrix we define each processor's diagonal portion
2099    to be its local rows and the corresponding columns (a square submatrix);
2100    each processor's off-diagonal portion encompasses the remainder of the
2101    local matrix (a rectangular submatrix).
2102 
2103    The user can specify preallocated storage for the diagonal part of
2104    the local submatrix with either d_nz or d_nnz (not both).  Set
2105    d_nz=PETSC_DEFAULT and d_nnz=PETSC_NULL for PETSc to control dynamic
2106    memory allocation.  Likewise, specify preallocated storage for the
2107    off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
2108 
2109    Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
2110    the figure below we depict these three local rows and all columns (0-11).
2111 
2112 .vb
2113            0 1 2 3 4 5 6 7 8 9 10 11
2114           -------------------
2115    row 3  |  o o o d d d o o o o o o
2116    row 4  |  o o o d d d o o o o o o
2117    row 5  |  o o o d d d o o o o o o
2118           -------------------
2119 .ve
2120 
2121    Thus, any entries in the d locations are stored in the d (diagonal)
2122    submatrix, and any entries in the o locations are stored in the
2123    o (off-diagonal) submatrix.  Note that the d and the o submatrices are
2124    stored simply in the MATSEQBAIJ format for compressed row storage.
2125 
2126    Now d_nz should indicate the number of block nonzeros per row in the d matrix,
2127    and o_nz should indicate the number of block nonzeros per row in the o matrix.
2128    In general, for PDE problems in which most nonzeros are near the diagonal,
2129    one expects d_nz >> o_nz.   For large problems you MUST preallocate memory
2130    or you will get TERRIBLE performance; see the users' manual chapter on
2131    matrices.
2132 
2133    Level: intermediate
2134 
2135 .keywords: matrix, block, aij, compressed row, sparse, parallel
2136 
2137 .seealso: MatCreate(), MatCreateSeqBAIJ(), MatSetValues(), MatCreateMPIBAIJ()
2138 @*/
2139 int MatMPIBAIJSetPreallocation(Mat B,int bs,int d_nz,int *d_nnz,int o_nz,int *o_nnz)
2140 {
2141   Mat_MPIBAIJ  *b;
2142   int          ierr,i;
2143   PetscTruth   flg2;
2144 
2145   PetscFunctionBegin;
2146   ierr = PetscTypeCompare((PetscObject)B,MATMPIBAIJ,&flg2);CHKERRQ(ierr);
2147   if (!flg2) PetscFunctionReturn(0);
2148 
2149   B->preallocated = PETSC_TRUE;
2150   ierr = PetscOptionsGetInt(PETSC_NULL,"-mat_block_size",&bs,PETSC_NULL);CHKERRQ(ierr);
2151 
2152   if (bs < 1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Invalid block size specified, must be positive");
2153   if (d_nz == PETSC_DEFAULT || d_nz == PETSC_DECIDE) d_nz = 5;
2154   if (o_nz == PETSC_DEFAULT || o_nz == PETSC_DECIDE) o_nz = 2;
2155   if (d_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"d_nz cannot be less than 0: value %d",d_nz);
2156   if (o_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"o_nz cannot be less than 0: value %d",o_nz);
2157   if (d_nnz) {
2158   for (i=0; i<B->m/bs; i++) {
2159       if (d_nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"d_nnz cannot be less than -1: local row %d value %d",i,d_nnz[i]);
2160     }
2161   }
2162   if (o_nnz) {
2163     for (i=0; i<B->m/bs; i++) {
2164       if (o_nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"o_nnz cannot be less than -1: local row %d value %d",i,o_nnz[i]);
2165     }
2166   }
2167 
2168   ierr = PetscSplitOwnershipBlock(B->comm,bs,&B->m,&B->M);CHKERRQ(ierr);
2169   ierr = PetscSplitOwnershipBlock(B->comm,bs,&B->n,&B->N);CHKERRQ(ierr);
2170   ierr = MapCreateMPI(B->comm,B->m,B->M,&B->rmap);CHKERRQ(ierr);
2171   ierr = MapCreateMPI(B->comm,B->n,B->N,&B->cmap);CHKERRQ(ierr);
2172 
2173   b = (Mat_MPIBAIJ*)B->data;
2174   b->bs  = bs;
2175   b->bs2 = bs*bs;
2176   b->mbs = B->m/bs;
2177   b->nbs = B->n/bs;
2178   b->Mbs = B->M/bs;
2179   b->Nbs = B->N/bs;
2180 
2181   ierr = MPI_Allgather(&b->mbs,1,MPI_INT,b->rowners+1,1,MPI_INT,B->comm);CHKERRQ(ierr);
2182   b->rowners[0]    = 0;
2183   for (i=2; i<=b->size; i++) {
2184     b->rowners[i] += b->rowners[i-1];
2185   }
2186   b->rstart    = b->rowners[b->rank];
2187   b->rend      = b->rowners[b->rank+1];
2188 
2189   ierr = MPI_Allgather(&b->nbs,1,MPI_INT,b->cowners+1,1,MPI_INT,B->comm);CHKERRQ(ierr);
2190   b->cowners[0] = 0;
2191   for (i=2; i<=b->size; i++) {
2192     b->cowners[i] += b->cowners[i-1];
2193   }
2194   b->cstart    = b->cowners[b->rank];
2195   b->cend      = b->cowners[b->rank+1];
2196 
2197   for (i=0; i<=b->size; i++) {
2198     b->rowners_bs[i] = b->rowners[i]*bs;
2199   }
2200   b->rstart_bs = b->rstart*bs;
2201   b->rend_bs   = b->rend*bs;
2202   b->cstart_bs = b->cstart*bs;
2203   b->cend_bs   = b->cend*bs;
2204 
2205   ierr = MatCreateSeqBAIJ(PETSC_COMM_SELF,bs,B->m,B->n,d_nz,d_nnz,&b->A);CHKERRQ(ierr);
2206   PetscLogObjectParent(B,b->A);
2207   ierr = MatCreateSeqBAIJ(PETSC_COMM_SELF,bs,B->m,B->N,o_nz,o_nnz,&b->B);CHKERRQ(ierr);
2208   PetscLogObjectParent(B,b->B);
2209   ierr = MatStashCreate_Private(B->comm,bs,&B->bstash);CHKERRQ(ierr);
2210 
2211   PetscFunctionReturn(0);
2212 }
2213 
2214 #undef __FUNCT__
2215 #define __FUNCT__ "MatCreateMPIBAIJ"
2216 /*@C
2217    MatCreateMPIBAIJ - Creates a sparse parallel matrix in block AIJ format
2218    (block compressed row).  For good matrix assembly performance
2219    the user should preallocate the matrix storage by setting the parameters
2220    d_nz (or d_nnz) and o_nz (or o_nnz).  By setting these parameters accurately,
2221    performance can be increased by more than a factor of 50.
2222 
2223    Collective on MPI_Comm
2224 
2225    Input Parameters:
2226 +  comm - MPI communicator
2227 .  bs   - size of blockk
2228 .  m - number of local rows (or PETSC_DECIDE to have calculated if M is given)
2229            This value should be the same as the local size used in creating the
2230            y vector for the matrix-vector product y = Ax.
2231 .  n - number of local columns (or PETSC_DECIDE to have calculated if N is given)
2232            This value should be the same as the local size used in creating the
2233            x vector for the matrix-vector product y = Ax.
2234 .  M - number of global rows (or PETSC_DETERMINE to have calculated if m is given)
2235 .  N - number of global columns (or PETSC_DETERMINE to have calculated if n is given)
2236 .  d_nz  - number of nonzero blocks per block row in diagonal portion of local
2237            submatrix  (same for all local rows)
2238 .  d_nnz - array containing the number of nonzero blocks in the various block rows
2239            of the in diagonal portion of the local (possibly different for each block
2240            row) or PETSC_NULL.  You must leave room for the diagonal entry even if it is zero.
2241 .  o_nz  - number of nonzero blocks per block row in the off-diagonal portion of local
2242            submatrix (same for all local rows).
2243 -  o_nnz - array containing the number of nonzero blocks in the various block rows of the
2244            off-diagonal portion of the local submatrix (possibly different for
2245            each block row) or PETSC_NULL.
2246 
2247    Output Parameter:
2248 .  A - the matrix
2249 
2250    Options Database Keys:
2251 .   -mat_no_unroll - uses code that does not unroll the loops in the
2252                      block calculations (much slower)
2253 .   -mat_block_size - size of the blocks to use
2254 
2255    Notes:
2256    A nonzero block is any block that as 1 or more nonzeros in it
2257 
2258    The user MUST specify either the local or global matrix dimensions
2259    (possibly both).
2260 
2261    If PETSC_DECIDE or  PETSC_DETERMINE is used for a particular argument on one processor
2262    than it must be used on all processors that share the object for that argument.
2263 
2264    Storage Information:
2265    For a square global matrix we define each processor's diagonal portion
2266    to be its local rows and the corresponding columns (a square submatrix);
2267    each processor's off-diagonal portion encompasses the remainder of the
2268    local matrix (a rectangular submatrix).
2269 
2270    The user can specify preallocated storage for the diagonal part of
2271    the local submatrix with either d_nz or d_nnz (not both).  Set
2272    d_nz=PETSC_DEFAULT and d_nnz=PETSC_NULL for PETSc to control dynamic
2273    memory allocation.  Likewise, specify preallocated storage for the
2274    off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
2275 
2276    Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
2277    the figure below we depict these three local rows and all columns (0-11).
2278 
2279 .vb
2280            0 1 2 3 4 5 6 7 8 9 10 11
2281           -------------------
2282    row 3  |  o o o d d d o o o o o o
2283    row 4  |  o o o d d d o o o o o o
2284    row 5  |  o o o d d d o o o o o o
2285           -------------------
2286 .ve
2287 
2288    Thus, any entries in the d locations are stored in the d (diagonal)
2289    submatrix, and any entries in the o locations are stored in the
2290    o (off-diagonal) submatrix.  Note that the d and the o submatrices are
2291    stored simply in the MATSEQBAIJ format for compressed row storage.
2292 
2293    Now d_nz should indicate the number of block nonzeros per row in the d matrix,
2294    and o_nz should indicate the number of block nonzeros per row in the o matrix.
2295    In general, for PDE problems in which most nonzeros are near the diagonal,
2296    one expects d_nz >> o_nz.   For large problems you MUST preallocate memory
2297    or you will get TERRIBLE performance; see the users' manual chapter on
2298    matrices.
2299 
2300    Level: intermediate
2301 
2302 .keywords: matrix, block, aij, compressed row, sparse, parallel
2303 
2304 .seealso: MatCreate(), MatCreateSeqBAIJ(), MatSetValues(), MatCreateMPIBAIJ()
2305 @*/
2306 int MatCreateMPIBAIJ(MPI_Comm comm,int bs,int m,int n,int M,int N,int d_nz,int *d_nnz,int o_nz,int *o_nnz,Mat *A)
2307 {
2308   int ierr,size;
2309 
2310   PetscFunctionBegin;
2311   ierr = MatCreate(comm,m,n,M,N,A);CHKERRQ(ierr);
2312   ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
2313   if (size > 1) {
2314     ierr = MatSetType(*A,MATMPIBAIJ);CHKERRQ(ierr);
2315     ierr = MatMPIBAIJSetPreallocation(*A,bs,d_nz,d_nnz,o_nz,o_nnz);CHKERRQ(ierr);
2316   } else {
2317     ierr = MatSetType(*A,MATSEQBAIJ);CHKERRQ(ierr);
2318     ierr = MatSeqBAIJSetPreallocation(*A,bs,d_nz,d_nnz);CHKERRQ(ierr);
2319   }
2320   PetscFunctionReturn(0);
2321 }
2322 
2323 #undef __FUNCT__
2324 #define __FUNCT__ "MatDuplicate_MPIBAIJ"
2325 static int MatDuplicate_MPIBAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat)
2326 {
2327   Mat         mat;
2328   Mat_MPIBAIJ *a,*oldmat = (Mat_MPIBAIJ*)matin->data;
2329   int         ierr,len=0;
2330 
2331   PetscFunctionBegin;
2332   *newmat       = 0;
2333   ierr = MatCreate(matin->comm,matin->m,matin->n,matin->M,matin->N,&mat);CHKERRQ(ierr);
2334   ierr = MatSetType(mat,MATMPIBAIJ);CHKERRQ(ierr);
2335   mat->preallocated = PETSC_TRUE;
2336   mat->assembled    = PETSC_TRUE;
2337   a      = (Mat_MPIBAIJ*)mat->data;
2338   a->bs  = oldmat->bs;
2339   a->bs2 = oldmat->bs2;
2340   a->mbs = oldmat->mbs;
2341   a->nbs = oldmat->nbs;
2342   a->Mbs = oldmat->Mbs;
2343   a->Nbs = oldmat->Nbs;
2344 
2345   a->rstart       = oldmat->rstart;
2346   a->rend         = oldmat->rend;
2347   a->cstart       = oldmat->cstart;
2348   a->cend         = oldmat->cend;
2349   a->size         = oldmat->size;
2350   a->rank         = oldmat->rank;
2351   a->donotstash   = oldmat->donotstash;
2352   a->roworiented  = oldmat->roworiented;
2353   a->rowindices   = 0;
2354   a->rowvalues    = 0;
2355   a->getrowactive = PETSC_FALSE;
2356   a->barray       = 0;
2357   a->rstart_bs    = oldmat->rstart_bs;
2358   a->rend_bs      = oldmat->rend_bs;
2359   a->cstart_bs    = oldmat->cstart_bs;
2360   a->cend_bs      = oldmat->cend_bs;
2361 
2362   /* hash table stuff */
2363   a->ht           = 0;
2364   a->hd           = 0;
2365   a->ht_size      = 0;
2366   a->ht_flag      = oldmat->ht_flag;
2367   a->ht_fact      = oldmat->ht_fact;
2368   a->ht_total_ct  = 0;
2369   a->ht_insert_ct = 0;
2370 
2371   ierr = PetscMemcpy(a->rowners,oldmat->rowners,3*(a->size+2)*sizeof(int));CHKERRQ(ierr);
2372   ierr = MatStashCreate_Private(matin->comm,1,&mat->stash);CHKERRQ(ierr);
2373   ierr = MatStashCreate_Private(matin->comm,oldmat->bs,&mat->bstash);CHKERRQ(ierr);
2374   if (oldmat->colmap) {
2375 #if defined (PETSC_USE_CTABLE)
2376   ierr = PetscTableCreateCopy(oldmat->colmap,&a->colmap);CHKERRQ(ierr);
2377 #else
2378   ierr = PetscMalloc((a->Nbs)*sizeof(int),&a->colmap);CHKERRQ(ierr);
2379   PetscLogObjectMemory(mat,(a->Nbs)*sizeof(int));
2380   ierr = PetscMemcpy(a->colmap,oldmat->colmap,(a->Nbs)*sizeof(int));CHKERRQ(ierr);
2381 #endif
2382   } else a->colmap = 0;
2383   if (oldmat->garray && (len = ((Mat_SeqBAIJ*)(oldmat->B->data))->nbs)) {
2384     ierr = PetscMalloc(len*sizeof(int),&a->garray);CHKERRQ(ierr);
2385     PetscLogObjectMemory(mat,len*sizeof(int));
2386     ierr = PetscMemcpy(a->garray,oldmat->garray,len*sizeof(int));CHKERRQ(ierr);
2387   } else a->garray = 0;
2388 
2389   ierr =  VecDuplicate(oldmat->lvec,&a->lvec);CHKERRQ(ierr);
2390   PetscLogObjectParent(mat,a->lvec);
2391   ierr =  VecScatterCopy(oldmat->Mvctx,&a->Mvctx);CHKERRQ(ierr);
2392 
2393   PetscLogObjectParent(mat,a->Mvctx);
2394   ierr =  MatDuplicate(oldmat->A,cpvalues,&a->A);CHKERRQ(ierr);
2395   PetscLogObjectParent(mat,a->A);
2396   ierr =  MatDuplicate(oldmat->B,cpvalues,&a->B);CHKERRQ(ierr);
2397   PetscLogObjectParent(mat,a->B);
2398   ierr = PetscFListDuplicate(matin->qlist,&mat->qlist);CHKERRQ(ierr);
2399   *newmat = mat;
2400   PetscFunctionReturn(0);
2401 }
2402 
2403 #include "petscsys.h"
2404 
2405 EXTERN_C_BEGIN
2406 #undef __FUNCT__
2407 #define __FUNCT__ "MatLoad_MPIBAIJ"
2408 int MatLoad_MPIBAIJ(PetscViewer viewer,MatType type,Mat *newmat)
2409 {
2410   Mat          A;
2411   int          i,nz,ierr,j,rstart,rend,fd;
2412   Scalar       *vals,*buf;
2413   MPI_Comm     comm = ((PetscObject)viewer)->comm;
2414   MPI_Status   status;
2415   int          header[4],rank,size,*rowlengths = 0,M,N,m,*rowners,*browners,maxnz,*cols;
2416   int          *locrowlens,*sndcounts = 0,*procsnz = 0,jj,*mycols,*ibuf;
2417   int          tag = ((PetscObject)viewer)->tag,bs=1,Mbs,mbs,extra_rows;
2418   int          *dlens,*odlens,*mask,*masked1,*masked2,rowcount,odcount;
2419   int          dcount,kmax,k,nzcount,tmp;
2420 
2421   PetscFunctionBegin;
2422   ierr = PetscOptionsGetInt(PETSC_NULL,"-matload_block_size",&bs,PETSC_NULL);CHKERRQ(ierr);
2423 
2424   ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
2425   ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr);
2426   if (!rank) {
2427     ierr = PetscViewerBinaryGetDescriptor(viewer,&fd);CHKERRQ(ierr);
2428     ierr = PetscBinaryRead(fd,(char *)header,4,PETSC_INT);CHKERRQ(ierr);
2429     if (header[0] != MAT_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not matrix object");
2430     if (header[3] < 0) {
2431       SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format, cannot load as MPIBAIJ");
2432     }
2433   }
2434 
2435   ierr = MPI_Bcast(header+1,3,MPI_INT,0,comm);CHKERRQ(ierr);
2436   M = header[1]; N = header[2];
2437 
2438   if (M != N) SETERRQ(PETSC_ERR_SUP,"Can only do square matrices");
2439 
2440   /*
2441      This code adds extra rows to make sure the number of rows is
2442      divisible by the blocksize
2443   */
2444   Mbs        = M/bs;
2445   extra_rows = bs - M + bs*(Mbs);
2446   if (extra_rows == bs) extra_rows = 0;
2447   else                  Mbs++;
2448   if (extra_rows &&!rank) {
2449     PetscLogInfo(0,"MatLoad_MPIBAIJ:Padding loaded matrix to match blocksize\n");
2450   }
2451 
2452   /* determine ownership of all rows */
2453   mbs        = Mbs/size + ((Mbs % size) > rank);
2454   m          = mbs*bs;
2455   ierr       = PetscMalloc(2*(size+2)*sizeof(int),&rowners);CHKERRQ(ierr);
2456   browners   = rowners + size + 1;
2457   ierr       = MPI_Allgather(&mbs,1,MPI_INT,rowners+1,1,MPI_INT,comm);CHKERRQ(ierr);
2458   rowners[0] = 0;
2459   for (i=2; i<=size; i++) rowners[i] += rowners[i-1];
2460   for (i=0; i<=size;  i++) browners[i] = rowners[i]*bs;
2461   rstart = rowners[rank];
2462   rend   = rowners[rank+1];
2463 
2464   /* distribute row lengths to all processors */
2465   ierr = PetscMalloc((rend-rstart)*bs*sizeof(int),&locrowlens);CHKERRQ(ierr);
2466   if (!rank) {
2467     ierr = PetscMalloc((M+extra_rows)*sizeof(int),&rowlengths);CHKERRQ(ierr);
2468     ierr = PetscBinaryRead(fd,rowlengths,M,PETSC_INT);CHKERRQ(ierr);
2469     for (i=0; i<extra_rows; i++) rowlengths[M+i] = 1;
2470     ierr = PetscMalloc(size*sizeof(int),&sndcounts);CHKERRQ(ierr);
2471     for (i=0; i<size; i++) sndcounts[i] = browners[i+1] - browners[i];
2472     ierr = MPI_Scatterv(rowlengths,sndcounts,browners,MPI_INT,locrowlens,(rend-rstart)*bs,MPI_INT,0,comm);CHKERRQ(ierr);
2473     ierr = PetscFree(sndcounts);CHKERRQ(ierr);
2474   } else {
2475     ierr = MPI_Scatterv(0,0,0,MPI_INT,locrowlens,(rend-rstart)*bs,MPI_INT,0,comm);CHKERRQ(ierr);
2476   }
2477 
2478   if (!rank) {
2479     /* calculate the number of nonzeros on each processor */
2480     ierr = PetscMalloc(size*sizeof(int),&procsnz);CHKERRQ(ierr);
2481     ierr = PetscMemzero(procsnz,size*sizeof(int));CHKERRQ(ierr);
2482     for (i=0; i<size; i++) {
2483       for (j=rowners[i]*bs; j< rowners[i+1]*bs; j++) {
2484         procsnz[i] += rowlengths[j];
2485       }
2486     }
2487     ierr = PetscFree(rowlengths);CHKERRQ(ierr);
2488 
2489     /* determine max buffer needed and allocate it */
2490     maxnz = 0;
2491     for (i=0; i<size; i++) {
2492       maxnz = PetscMax(maxnz,procsnz[i]);
2493     }
2494     ierr = PetscMalloc(maxnz*sizeof(int),&cols);CHKERRQ(ierr);
2495 
2496     /* read in my part of the matrix column indices  */
2497     nz     = procsnz[0];
2498     ierr   = PetscMalloc(nz*sizeof(int),&ibuf);CHKERRQ(ierr);
2499     mycols = ibuf;
2500     if (size == 1)  nz -= extra_rows;
2501     ierr = PetscBinaryRead(fd,mycols,nz,PETSC_INT);CHKERRQ(ierr);
2502     if (size == 1)  for (i=0; i< extra_rows; i++) { mycols[nz+i] = M+i; }
2503 
2504     /* read in every ones (except the last) and ship off */
2505     for (i=1; i<size-1; i++) {
2506       nz   = procsnz[i];
2507       ierr = PetscBinaryRead(fd,cols,nz,PETSC_INT);CHKERRQ(ierr);
2508       ierr = MPI_Send(cols,nz,MPI_INT,i,tag,comm);CHKERRQ(ierr);
2509     }
2510     /* read in the stuff for the last proc */
2511     if (size != 1) {
2512       nz   = procsnz[size-1] - extra_rows;  /* the extra rows are not on the disk */
2513       ierr = PetscBinaryRead(fd,cols,nz,PETSC_INT);CHKERRQ(ierr);
2514       for (i=0; i<extra_rows; i++) cols[nz+i] = M+i;
2515       ierr = MPI_Send(cols,nz+extra_rows,MPI_INT,size-1,tag,comm);CHKERRQ(ierr);
2516     }
2517     ierr = PetscFree(cols);CHKERRQ(ierr);
2518   } else {
2519     /* determine buffer space needed for message */
2520     nz = 0;
2521     for (i=0; i<m; i++) {
2522       nz += locrowlens[i];
2523     }
2524     ierr   = PetscMalloc(nz*sizeof(int),&ibuf);CHKERRQ(ierr);
2525     mycols = ibuf;
2526     /* receive message of column indices*/
2527     ierr = MPI_Recv(mycols,nz,MPI_INT,0,tag,comm,&status);CHKERRQ(ierr);
2528     ierr = MPI_Get_count(&status,MPI_INT,&maxnz);CHKERRQ(ierr);
2529     if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
2530   }
2531 
2532   /* loop over local rows, determining number of off diagonal entries */
2533   ierr     = PetscMalloc(2*(rend-rstart+1)*sizeof(int),&dlens);CHKERRQ(ierr);
2534   odlens   = dlens + (rend-rstart);
2535   ierr     = PetscMalloc(3*Mbs*sizeof(int),&mask);CHKERRQ(ierr);
2536   ierr     = PetscMemzero(mask,3*Mbs*sizeof(int));CHKERRQ(ierr);
2537   masked1  = mask    + Mbs;
2538   masked2  = masked1 + Mbs;
2539   rowcount = 0; nzcount = 0;
2540   for (i=0; i<mbs; i++) {
2541     dcount  = 0;
2542     odcount = 0;
2543     for (j=0; j<bs; j++) {
2544       kmax = locrowlens[rowcount];
2545       for (k=0; k<kmax; k++) {
2546         tmp = mycols[nzcount++]/bs;
2547         if (!mask[tmp]) {
2548           mask[tmp] = 1;
2549           if (tmp < rstart || tmp >= rend) masked2[odcount++] = tmp;
2550           else masked1[dcount++] = tmp;
2551         }
2552       }
2553       rowcount++;
2554     }
2555 
2556     dlens[i]  = dcount;
2557     odlens[i] = odcount;
2558 
2559     /* zero out the mask elements we set */
2560     for (j=0; j<dcount; j++) mask[masked1[j]] = 0;
2561     for (j=0; j<odcount; j++) mask[masked2[j]] = 0;
2562   }
2563 
2564   /* create our matrix */
2565   ierr = MatCreateMPIBAIJ(comm,bs,m,m,M+extra_rows,N+extra_rows,0,dlens,0,odlens,newmat);CHKERRQ(ierr);
2566   A = *newmat;
2567   MatSetOption(A,MAT_COLUMNS_SORTED);
2568 
2569   if (!rank) {
2570     ierr = PetscMalloc(maxnz*sizeof(Scalar),&buf);CHKERRQ(ierr);
2571     /* read in my part of the matrix numerical values  */
2572     nz = procsnz[0];
2573     vals = buf;
2574     mycols = ibuf;
2575     if (size == 1)  nz -= extra_rows;
2576     ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr);
2577     if (size == 1)  for (i=0; i< extra_rows; i++) { vals[nz+i] = 1.0; }
2578 
2579     /* insert into matrix */
2580     jj      = rstart*bs;
2581     for (i=0; i<m; i++) {
2582       ierr = MatSetValues(A,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);CHKERRQ(ierr);
2583       mycols += locrowlens[i];
2584       vals   += locrowlens[i];
2585       jj++;
2586     }
2587     /* read in other processors (except the last one) and ship out */
2588     for (i=1; i<size-1; i++) {
2589       nz   = procsnz[i];
2590       vals = buf;
2591       ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr);
2592       ierr = MPI_Send(vals,nz,MPIU_SCALAR,i,A->tag,comm);CHKERRQ(ierr);
2593     }
2594     /* the last proc */
2595     if (size != 1){
2596       nz   = procsnz[i] - extra_rows;
2597       vals = buf;
2598       ierr = PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);CHKERRQ(ierr);
2599       for (i=0; i<extra_rows; i++) vals[nz+i] = 1.0;
2600       ierr = MPI_Send(vals,nz+extra_rows,MPIU_SCALAR,size-1,A->tag,comm);CHKERRQ(ierr);
2601     }
2602     ierr = PetscFree(procsnz);CHKERRQ(ierr);
2603   } else {
2604     /* receive numeric values */
2605     ierr = PetscMalloc(nz*sizeof(Scalar),&buf);CHKERRQ(ierr);
2606 
2607     /* receive message of values*/
2608     vals   = buf;
2609     mycols = ibuf;
2610     ierr   = MPI_Recv(vals,nz,MPIU_SCALAR,0,A->tag,comm,&status);CHKERRQ(ierr);
2611     ierr   = MPI_Get_count(&status,MPIU_SCALAR,&maxnz);CHKERRQ(ierr);
2612     if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
2613 
2614     /* insert into matrix */
2615     jj      = rstart*bs;
2616     for (i=0; i<m; i++) {
2617       ierr    = MatSetValues(A,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);CHKERRQ(ierr);
2618       mycols += locrowlens[i];
2619       vals   += locrowlens[i];
2620       jj++;
2621     }
2622   }
2623   ierr = PetscFree(locrowlens);CHKERRQ(ierr);
2624   ierr = PetscFree(buf);CHKERRQ(ierr);
2625   ierr = PetscFree(ibuf);CHKERRQ(ierr);
2626   ierr = PetscFree(rowners);CHKERRQ(ierr);
2627   ierr = PetscFree(dlens);CHKERRQ(ierr);
2628   ierr = PetscFree(mask);CHKERRQ(ierr);
2629   ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
2630   ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
2631   PetscFunctionReturn(0);
2632 }
2633 EXTERN_C_END
2634 
2635 #undef __FUNCT__
2636 #define __FUNCT__ "MatMPIBAIJSetHashTableFactor"
2637 /*@
2638    MatMPIBAIJSetHashTableFactor - Sets the factor required to compute the size of the HashTable.
2639 
2640    Input Parameters:
2641 .  mat  - the matrix
2642 .  fact - factor
2643 
2644    Collective on Mat
2645 
2646    Level: advanced
2647 
2648   Notes:
2649    This can also be set by the command line option: -mat_use_hash_table fact
2650 
2651 .keywords: matrix, hashtable, factor, HT
2652 
2653 .seealso: MatSetOption()
2654 @*/
2655 int MatMPIBAIJSetHashTableFactor(Mat mat,PetscReal fact)
2656 {
2657   Mat_MPIBAIJ *baij;
2658   int         ierr;
2659   PetscTruth  flg;
2660 
2661   PetscFunctionBegin;
2662   PetscValidHeaderSpecific(mat,MAT_COOKIE);
2663   ierr = PetscTypeCompare((PetscObject)mat,MATMPIBAIJ,&flg);CHKERRQ(ierr);
2664   if (!flg) {
2665     SETERRQ(PETSC_ERR_ARG_WRONG,"Incorrect matrix type. Use MPIBAIJ only.");
2666   }
2667   baij = (Mat_MPIBAIJ*)mat->data;
2668   baij->ht_fact = fact;
2669   PetscFunctionReturn(0);
2670 }
2671 
2672 #undef __FUNCT__
2673 #define __FUNCT__ "MatMPIBAIJGetSeqBAIJ"
2674 int MatMPIBAIJGetSeqBAIJ(Mat A,Mat *Ad,Mat *Ao,int **colmap)
2675 {
2676   Mat_MPIBAIJ *a = (Mat_MPIBAIJ *)A->data;
2677   PetscFunctionBegin;
2678   *Ad     = a->A;
2679   *Ao     = a->B;
2680   *colmap = a->garray;
2681   PetscFunctionReturn(0);
2682 }
2683