xref: /petsc/src/mat/impls/baij/seq/baij.c (revision c3b5f7ba6bc5ce25a01a67bb37ba5d34b02bbbd7)
1 
2 /*
3     Defines the basic matrix operations for the BAIJ (compressed row)
4   matrix storage format.
5 */
6 #include <../src/mat/impls/baij/seq/baij.h>  /*I   "petscmat.h"  I*/
7 #include <petscblaslapack.h>
8 #include <petsc/private/kernels/blockinvert.h>
9 #include <petsc/private/kernels/blockmatmult.h>
10 
11 #if defined(PETSC_HAVE_HYPRE)
12 PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*);
13 #endif
14 
15 #if defined(PETSC_HAVE_MKL_SPARSE_OPTIMIZE)
16 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBAIJMKL(Mat,MatType,MatReuse,Mat*);
17 #endif
18 PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat,MatType,MatReuse,Mat*);
19 
20 PetscErrorCode MatGetColumnReductions_SeqBAIJ(Mat A,PetscInt type,PetscReal *reductions)
21 {
22   Mat_SeqBAIJ    *a_aij = (Mat_SeqBAIJ*) A->data;
23   PetscInt       m,n,i;
24   PetscInt       ib,jb,bs = A->rmap->bs;
25   MatScalar      *a_val = a_aij->a;
26 
27   PetscFunctionBegin;
28   PetscCall(MatGetSize(A,&m,&n));
29   for (i=0; i<n; i++) reductions[i] = 0.0;
30   if (type == NORM_2) {
31     for (i=a_aij->i[0]; i<a_aij->i[A->rmap->n/bs]; i++) {
32       for (jb=0; jb<bs; jb++) {
33         for (ib=0; ib<bs; ib++) {
34           reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val * *a_val);
35           a_val++;
36         }
37       }
38     }
39   } else if (type == NORM_1) {
40     for (i=a_aij->i[0]; i<a_aij->i[A->rmap->n/bs]; i++) {
41       for (jb=0; jb<bs; jb++) {
42         for (ib=0; ib<bs; ib++) {
43           reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val);
44           a_val++;
45         }
46       }
47     }
48   } else if (type == NORM_INFINITY) {
49     for (i=a_aij->i[0]; i<a_aij->i[A->rmap->n/bs]; i++) {
50       for (jb=0; jb<bs; jb++) {
51         for (ib=0; ib<bs; ib++) {
52           int col = A->cmap->rstart + a_aij->j[i] * bs + jb;
53           reductions[col] = PetscMax(PetscAbsScalar(*a_val), reductions[col]);
54           a_val++;
55         }
56       }
57     }
58   } else if (type == REDUCTION_SUM_REALPART || type == REDUCTION_MEAN_REALPART) {
59     for (i=a_aij->i[0]; i<a_aij->i[A->rmap->n/bs]; i++) {
60       for (jb=0; jb<bs; jb++) {
61         for (ib=0; ib<bs; ib++) {
62           reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscRealPart(*a_val);
63           a_val++;
64         }
65       }
66     }
67   } else if (type == REDUCTION_SUM_IMAGINARYPART || type == REDUCTION_MEAN_IMAGINARYPART) {
68     for (i=a_aij->i[0]; i<a_aij->i[A->rmap->n/bs]; i++) {
69       for (jb=0; jb<bs; jb++) {
70         for (ib=0; ib<bs; ib++) {
71           reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscImaginaryPart(*a_val);
72           a_val++;
73         }
74       }
75     }
76   } else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Unknown reduction type");
77   if (type == NORM_2) {
78     for (i=0; i<n; i++) reductions[i] = PetscSqrtReal(reductions[i]);
79   } else if (type == REDUCTION_MEAN_REALPART || type == REDUCTION_MEAN_IMAGINARYPART) {
80     for (i=0; i<n; i++) reductions[i] /= m;
81   }
82   PetscFunctionReturn(0);
83 }
84 
85 PetscErrorCode MatInvertBlockDiagonal_SeqBAIJ(Mat A,const PetscScalar **values)
86 {
87   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*) A->data;
88   PetscInt       *diag_offset,i,bs = A->rmap->bs,mbs = a->mbs,ipvt[5],bs2 = bs*bs,*v_pivots;
89   MatScalar      *v    = a->a,*odiag,*diag,work[25],*v_work;
90   PetscReal      shift = 0.0;
91   PetscBool      allowzeropivot,zeropivotdetected=PETSC_FALSE;
92 
93   PetscFunctionBegin;
94   allowzeropivot = PetscNot(A->erroriffailure);
95 
96   if (a->idiagvalid) {
97     if (values) *values = a->idiag;
98     PetscFunctionReturn(0);
99   }
100   PetscCall(MatMarkDiagonal_SeqBAIJ(A));
101   diag_offset = a->diag;
102   if (!a->idiag) {
103     PetscCall(PetscMalloc1(bs2*mbs,&a->idiag));
104     PetscCall(PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar)));
105   }
106   diag  = a->idiag;
107   if (values) *values = a->idiag;
108   /* factor and invert each block */
109   switch (bs) {
110   case 1:
111     for (i=0; i<mbs; i++) {
112       odiag    = v + 1*diag_offset[i];
113       diag[0]  = odiag[0];
114 
115       if (PetscAbsScalar(diag[0] + shift) < PETSC_MACHINE_EPSILON) {
116         if (allowzeropivot) {
117           A->factorerrortype             = MAT_FACTOR_NUMERIC_ZEROPIVOT;
118           A->factorerror_zeropivot_value = PetscAbsScalar(diag[0]);
119           A->factorerror_zeropivot_row   = i;
120           PetscCall(PetscInfo(A,"Zero pivot, row %" PetscInt_FMT "\n",i));
121         } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot, row %" PetscInt_FMT " pivot value %g tolerance %g",i,(double)PetscAbsScalar(diag[0]),(double)PETSC_MACHINE_EPSILON);
122       }
123 
124       diag[0]  = (PetscScalar)1.0 / (diag[0] + shift);
125       diag    += 1;
126     }
127     break;
128   case 2:
129     for (i=0; i<mbs; i++) {
130       odiag    = v + 4*diag_offset[i];
131       diag[0]  = odiag[0]; diag[1] = odiag[1]; diag[2] = odiag[2]; diag[3] = odiag[3];
132       PetscCall(PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected));
133       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
134       diag    += 4;
135     }
136     break;
137   case 3:
138     for (i=0; i<mbs; i++) {
139       odiag    = v + 9*diag_offset[i];
140       diag[0]  = odiag[0]; diag[1] = odiag[1]; diag[2] = odiag[2]; diag[3] = odiag[3];
141       diag[4]  = odiag[4]; diag[5] = odiag[5]; diag[6] = odiag[6]; diag[7] = odiag[7];
142       diag[8]  = odiag[8];
143       PetscCall(PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected));
144       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
145       diag    += 9;
146     }
147     break;
148   case 4:
149     for (i=0; i<mbs; i++) {
150       odiag  = v + 16*diag_offset[i];
151       PetscCall(PetscArraycpy(diag,odiag,16));
152       PetscCall(PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected));
153       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
154       diag  += 16;
155     }
156     break;
157   case 5:
158     for (i=0; i<mbs; i++) {
159       odiag  = v + 25*diag_offset[i];
160       PetscCall(PetscArraycpy(diag,odiag,25));
161       PetscCall(PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected));
162       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
163       diag  += 25;
164     }
165     break;
166   case 6:
167     for (i=0; i<mbs; i++) {
168       odiag  = v + 36*diag_offset[i];
169       PetscCall(PetscArraycpy(diag,odiag,36));
170       PetscCall(PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected));
171       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
172       diag  += 36;
173     }
174     break;
175   case 7:
176     for (i=0; i<mbs; i++) {
177       odiag  = v + 49*diag_offset[i];
178       PetscCall(PetscArraycpy(diag,odiag,49));
179       PetscCall(PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected));
180       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
181       diag  += 49;
182     }
183     break;
184   default:
185     PetscCall(PetscMalloc2(bs,&v_work,bs,&v_pivots));
186     for (i=0; i<mbs; i++) {
187       odiag  = v + bs2*diag_offset[i];
188       PetscCall(PetscArraycpy(diag,odiag,bs2));
189       PetscCall(PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected));
190       if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
191       diag  += bs2;
192     }
193     PetscCall(PetscFree2(v_work,v_pivots));
194   }
195   a->idiagvalid = PETSC_TRUE;
196   PetscFunctionReturn(0);
197 }
198 
199 PetscErrorCode MatSOR_SeqBAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
200 {
201   Mat_SeqBAIJ       *a = (Mat_SeqBAIJ*)A->data;
202   PetscScalar       *x,*work,*w,*workt,*t;
203   const MatScalar   *v,*aa = a->a, *idiag;
204   const PetscScalar *b,*xb;
205   PetscScalar       s[7], xw[7]={0}; /* avoid some compilers thinking xw is uninitialized */
206   PetscInt          m = a->mbs,i,i2,nz,bs = A->rmap->bs,bs2 = bs*bs,k,j,idx,it;
207   const PetscInt    *diag,*ai = a->i,*aj = a->j,*vi;
208 
209   PetscFunctionBegin;
210   its = its*lits;
211   PetscCheck(!(flag & SOR_EISENSTAT),PETSC_COMM_SELF,PETSC_ERR_SUP,"No support yet for Eisenstat");
212   PetscCheck(its > 0,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Relaxation requires global its %" PetscInt_FMT " and local its %" PetscInt_FMT " both positive",its,lits);
213   PetscCheck(!fshift,PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for diagonal shift");
214   PetscCheck(omega == 1.0,PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for non-trivial relaxation factor");
215   PetscCheck(!(flag & SOR_APPLY_UPPER) && !(flag & SOR_APPLY_LOWER),PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for applying upper or lower triangular parts");
216 
217   if (!a->idiagvalid) PetscCall(MatInvertBlockDiagonal(A,NULL));
218 
219   if (!m) PetscFunctionReturn(0);
220   diag  = a->diag;
221   idiag = a->idiag;
222   k    = PetscMax(A->rmap->n,A->cmap->n);
223   if (!a->mult_work) {
224     PetscCall(PetscMalloc1(k+1,&a->mult_work));
225   }
226   if (!a->sor_workt) {
227     PetscCall(PetscMalloc1(k,&a->sor_workt));
228   }
229   if (!a->sor_work) {
230     PetscCall(PetscMalloc1(bs,&a->sor_work));
231   }
232   work = a->mult_work;
233   t    = a->sor_workt;
234   w    = a->sor_work;
235 
236   PetscCall(VecGetArray(xx,&x));
237   PetscCall(VecGetArrayRead(bb,&b));
238 
239   if (flag & SOR_ZERO_INITIAL_GUESS) {
240     if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
241       switch (bs) {
242       case 1:
243         PetscKernel_v_gets_A_times_w_1(x,idiag,b);
244         t[0] = b[0];
245         i2     = 1;
246         idiag += 1;
247         for (i=1; i<m; i++) {
248           v  = aa + ai[i];
249           vi = aj + ai[i];
250           nz = diag[i] - ai[i];
251           s[0] = b[i2];
252           for (j=0; j<nz; j++) {
253             xw[0] = x[vi[j]];
254             PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
255           }
256           t[i2] = s[0];
257           PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
258           x[i2]  = xw[0];
259           idiag += 1;
260           i2    += 1;
261         }
262         break;
263       case 2:
264         PetscKernel_v_gets_A_times_w_2(x,idiag,b);
265         t[0] = b[0]; t[1] = b[1];
266         i2     = 2;
267         idiag += 4;
268         for (i=1; i<m; i++) {
269           v  = aa + 4*ai[i];
270           vi = aj + ai[i];
271           nz = diag[i] - ai[i];
272           s[0] = b[i2]; s[1] = b[i2+1];
273           for (j=0; j<nz; j++) {
274             idx = 2*vi[j];
275             it  = 4*j;
276             xw[0] = x[idx]; xw[1] = x[1+idx];
277             PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
278           }
279           t[i2] = s[0]; t[i2+1] = s[1];
280           PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
281           x[i2]   = xw[0]; x[i2+1] = xw[1];
282           idiag  += 4;
283           i2     += 2;
284         }
285         break;
286       case 3:
287         PetscKernel_v_gets_A_times_w_3(x,idiag,b);
288         t[0] = b[0]; t[1] = b[1]; t[2] = b[2];
289         i2     = 3;
290         idiag += 9;
291         for (i=1; i<m; i++) {
292           v  = aa + 9*ai[i];
293           vi = aj + ai[i];
294           nz = diag[i] - ai[i];
295           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
296           while (nz--) {
297             idx = 3*(*vi++);
298             xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
299             PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
300             v  += 9;
301           }
302           t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2];
303           PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
304           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
305           idiag  += 9;
306           i2     += 3;
307         }
308         break;
309       case 4:
310         PetscKernel_v_gets_A_times_w_4(x,idiag,b);
311         t[0] = b[0]; t[1] = b[1]; t[2] = b[2]; t[3] = b[3];
312         i2     = 4;
313         idiag += 16;
314         for (i=1; i<m; i++) {
315           v  = aa + 16*ai[i];
316           vi = aj + ai[i];
317           nz = diag[i] - ai[i];
318           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3];
319           while (nz--) {
320             idx = 4*(*vi++);
321             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
322             PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
323             v  += 16;
324           }
325           t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2]; t[i2 + 3] = s[3];
326           PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
327           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3];
328           idiag  += 16;
329           i2     += 4;
330         }
331         break;
332       case 5:
333         PetscKernel_v_gets_A_times_w_5(x,idiag,b);
334         t[0] = b[0]; t[1] = b[1]; t[2] = b[2]; t[3] = b[3]; t[4] = b[4];
335         i2     = 5;
336         idiag += 25;
337         for (i=1; i<m; i++) {
338           v  = aa + 25*ai[i];
339           vi = aj + ai[i];
340           nz = diag[i] - ai[i];
341           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4];
342           while (nz--) {
343             idx = 5*(*vi++);
344             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
345             PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
346             v  += 25;
347           }
348           t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2]; t[i2+3] = s[3]; t[i2+4] = s[4];
349           PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
350           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4];
351           idiag  += 25;
352           i2     += 5;
353         }
354         break;
355       case 6:
356         PetscKernel_v_gets_A_times_w_6(x,idiag,b);
357         t[0] = b[0]; t[1] = b[1]; t[2] = b[2]; t[3] = b[3]; t[4] = b[4]; t[5] = b[5];
358         i2     = 6;
359         idiag += 36;
360         for (i=1; i<m; i++) {
361           v  = aa + 36*ai[i];
362           vi = aj + ai[i];
363           nz = diag[i] - ai[i];
364           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5];
365           while (nz--) {
366             idx = 6*(*vi++);
367             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
368             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
369             PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
370             v  += 36;
371           }
372           t[i2]   = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2];
373           t[i2+3] = s[3]; t[i2+4] = s[4]; t[i2+5] = s[5];
374           PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
375           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5];
376           idiag  += 36;
377           i2     += 6;
378         }
379         break;
380       case 7:
381         PetscKernel_v_gets_A_times_w_7(x,idiag,b);
382         t[0] = b[0]; t[1] = b[1]; t[2] = b[2];
383         t[3] = b[3]; t[4] = b[4]; t[5] = b[5]; t[6] = b[6];
384         i2     = 7;
385         idiag += 49;
386         for (i=1; i<m; i++) {
387           v  = aa + 49*ai[i];
388           vi = aj + ai[i];
389           nz = diag[i] - ai[i];
390           s[0] = b[i2];   s[1] = b[i2+1]; s[2] = b[i2+2];
391           s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5]; s[6] = b[i2+6];
392           while (nz--) {
393             idx = 7*(*vi++);
394             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
395             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
396             PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
397             v  += 49;
398           }
399           t[i2]   = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2];
400           t[i2+3] = s[3]; t[i2+4] = s[4]; t[i2+5] = s[5]; t[i2+6] = s[6];
401           PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
402           x[i2] =   xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
403           x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5]; x[i2+6] = xw[6];
404           idiag  += 49;
405           i2     += 7;
406         }
407         break;
408       default:
409         PetscKernel_w_gets_Ar_times_v(bs,bs,b,idiag,x);
410         PetscCall(PetscArraycpy(t,b,bs));
411         i2     = bs;
412         idiag += bs2;
413         for (i=1; i<m; i++) {
414           v  = aa + bs2*ai[i];
415           vi = aj + ai[i];
416           nz = diag[i] - ai[i];
417 
418           PetscCall(PetscArraycpy(w,b+i2,bs));
419           /* copy all rows of x that are needed into contiguous space */
420           workt = work;
421           for (j=0; j<nz; j++) {
422             PetscCall(PetscArraycpy(workt,x + bs*(*vi++),bs));
423             workt += bs;
424           }
425           PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
426           PetscCall(PetscArraycpy(t+i2,w,bs));
427           PetscKernel_w_gets_Ar_times_v(bs,bs,w,idiag,x+i2);
428 
429           idiag += bs2;
430           i2    += bs;
431         }
432         break;
433       }
434       /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
435       PetscCall(PetscLogFlops(1.0*bs2*a->nz));
436       xb = t;
437     }
438     else xb = b;
439     if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
440       idiag = a->idiag+bs2*(a->mbs-1);
441       i2 = bs * (m-1);
442       switch (bs) {
443       case 1:
444         s[0]  = xb[i2];
445         PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
446         x[i2] = xw[0];
447         i2   -= 1;
448         for (i=m-2; i>=0; i--) {
449           v  = aa + (diag[i]+1);
450           vi = aj + diag[i] + 1;
451           nz = ai[i+1] - diag[i] - 1;
452           s[0] = xb[i2];
453           for (j=0; j<nz; j++) {
454             xw[0] = x[vi[j]];
455             PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
456           }
457           PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
458           x[i2]  = xw[0];
459           idiag -= 1;
460           i2    -= 1;
461         }
462         break;
463       case 2:
464         s[0]  = xb[i2]; s[1] = xb[i2+1];
465         PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
466         x[i2] = xw[0]; x[i2+1] = xw[1];
467         i2    -= 2;
468         idiag -= 4;
469         for (i=m-2; i>=0; i--) {
470           v  = aa + 4*(diag[i] + 1);
471           vi = aj + diag[i] + 1;
472           nz = ai[i+1] - diag[i] - 1;
473           s[0] = xb[i2]; s[1] = xb[i2+1];
474           for (j=0; j<nz; j++) {
475             idx = 2*vi[j];
476             it  = 4*j;
477             xw[0] = x[idx]; xw[1] = x[1+idx];
478             PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
479           }
480           PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
481           x[i2]   = xw[0]; x[i2+1] = xw[1];
482           idiag  -= 4;
483           i2     -= 2;
484         }
485         break;
486       case 3:
487         s[0]  = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2];
488         PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
489         x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
490         i2    -= 3;
491         idiag -= 9;
492         for (i=m-2; i>=0; i--) {
493           v  = aa + 9*(diag[i]+1);
494           vi = aj + diag[i] + 1;
495           nz = ai[i+1] - diag[i] - 1;
496           s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2];
497           while (nz--) {
498             idx = 3*(*vi++);
499             xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
500             PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
501             v  += 9;
502           }
503           PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
504           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
505           idiag  -= 9;
506           i2     -= 3;
507         }
508         break;
509       case 4:
510         s[0]  = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3];
511         PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
512         x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3];
513         i2    -= 4;
514         idiag -= 16;
515         for (i=m-2; i>=0; i--) {
516           v  = aa + 16*(diag[i]+1);
517           vi = aj + diag[i] + 1;
518           nz = ai[i+1] - diag[i] - 1;
519           s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3];
520           while (nz--) {
521             idx = 4*(*vi++);
522             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
523             PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
524             v  += 16;
525           }
526           PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
527           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3];
528           idiag  -= 16;
529           i2     -= 4;
530         }
531         break;
532       case 5:
533         s[0]  = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4];
534         PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
535         x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4];
536         i2    -= 5;
537         idiag -= 25;
538         for (i=m-2; i>=0; i--) {
539           v  = aa + 25*(diag[i]+1);
540           vi = aj + diag[i] + 1;
541           nz = ai[i+1] - diag[i] - 1;
542           s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4];
543           while (nz--) {
544             idx = 5*(*vi++);
545             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
546             PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
547             v  += 25;
548           }
549           PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
550           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4];
551           idiag  -= 25;
552           i2     -= 5;
553         }
554         break;
555       case 6:
556         s[0]  = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5];
557         PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
558         x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5];
559         i2    -= 6;
560         idiag -= 36;
561         for (i=m-2; i>=0; i--) {
562           v  = aa + 36*(diag[i]+1);
563           vi = aj + diag[i] + 1;
564           nz = ai[i+1] - diag[i] - 1;
565           s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5];
566           while (nz--) {
567             idx = 6*(*vi++);
568             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
569             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
570             PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
571             v  += 36;
572           }
573           PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
574           x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5];
575           idiag  -= 36;
576           i2     -= 6;
577         }
578         break;
579       case 7:
580         s[0] = xb[i2];   s[1] = xb[i2+1]; s[2] = xb[i2+2];
581         s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5]; s[6] = xb[i2+6];
582         PetscKernel_v_gets_A_times_w_7(x,idiag,b);
583         x[i2]   = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
584         x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5]; x[i2+6] = xw[6];
585         i2    -= 7;
586         idiag -= 49;
587         for (i=m-2; i>=0; i--) {
588           v  = aa + 49*(diag[i]+1);
589           vi = aj + diag[i] + 1;
590           nz = ai[i+1] - diag[i] - 1;
591           s[0] = xb[i2];   s[1] = xb[i2+1]; s[2] = xb[i2+2];
592           s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5]; s[6] = xb[i2+6];
593           while (nz--) {
594             idx = 7*(*vi++);
595             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
596             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
597             PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
598             v  += 49;
599           }
600           PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
601           x[i2] =   xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
602           x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5]; x[i2+6] = xw[6];
603           idiag  -= 49;
604           i2     -= 7;
605         }
606         break;
607       default:
608         PetscCall(PetscArraycpy(w,xb+i2,bs));
609         PetscKernel_w_gets_Ar_times_v(bs,bs,w,idiag,x+i2);
610         i2    -= bs;
611         idiag -= bs2;
612         for (i=m-2; i>=0; i--) {
613           v  = aa + bs2*(diag[i]+1);
614           vi = aj + diag[i] + 1;
615           nz = ai[i+1] - diag[i] - 1;
616 
617           PetscCall(PetscArraycpy(w,xb+i2,bs));
618           /* copy all rows of x that are needed into contiguous space */
619           workt = work;
620           for (j=0; j<nz; j++) {
621             PetscCall(PetscArraycpy(workt,x + bs*(*vi++),bs));
622             workt += bs;
623           }
624           PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
625           PetscKernel_w_gets_Ar_times_v(bs,bs,w,idiag,x+i2);
626 
627           idiag -= bs2;
628           i2    -= bs;
629         }
630         break;
631       }
632       PetscCall(PetscLogFlops(1.0*bs2*(a->nz)));
633     }
634     its--;
635   }
636   while (its--) {
637     if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
638       idiag = a->idiag;
639       i2 = 0;
640       switch (bs) {
641       case 1:
642         for (i=0; i<m; i++) {
643           v  = aa + ai[i];
644           vi = aj + ai[i];
645           nz = ai[i+1] - ai[i];
646           s[0] = b[i2];
647           for (j=0; j<nz; j++) {
648             xw[0] = x[vi[j]];
649             PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
650           }
651           PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
652           x[i2] += xw[0];
653           idiag += 1;
654           i2    += 1;
655         }
656         break;
657       case 2:
658         for (i=0; i<m; i++) {
659           v  = aa + 4*ai[i];
660           vi = aj + ai[i];
661           nz = ai[i+1] - ai[i];
662           s[0] = b[i2]; s[1] = b[i2+1];
663           for (j=0; j<nz; j++) {
664             idx = 2*vi[j];
665             it  = 4*j;
666             xw[0] = x[idx]; xw[1] = x[1+idx];
667             PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
668           }
669           PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
670           x[i2]  += xw[0]; x[i2+1] += xw[1];
671           idiag  += 4;
672           i2     += 2;
673         }
674         break;
675       case 3:
676         for (i=0; i<m; i++) {
677           v  = aa + 9*ai[i];
678           vi = aj + ai[i];
679           nz = ai[i+1] - ai[i];
680           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
681           while (nz--) {
682             idx = 3*(*vi++);
683             xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
684             PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
685             v  += 9;
686           }
687           PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
688           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
689           idiag  += 9;
690           i2     += 3;
691         }
692         break;
693       case 4:
694         for (i=0; i<m; i++) {
695           v  = aa + 16*ai[i];
696           vi = aj + ai[i];
697           nz = ai[i+1] - ai[i];
698           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3];
699           while (nz--) {
700             idx = 4*(*vi++);
701             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
702             PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
703             v  += 16;
704           }
705           PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
706           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3];
707           idiag  += 16;
708           i2     += 4;
709         }
710         break;
711       case 5:
712         for (i=0; i<m; i++) {
713           v  = aa + 25*ai[i];
714           vi = aj + ai[i];
715           nz = ai[i+1] - ai[i];
716           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4];
717           while (nz--) {
718             idx = 5*(*vi++);
719             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
720             PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
721             v  += 25;
722           }
723           PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
724           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3]; x[i2+4] += xw[4];
725           idiag  += 25;
726           i2     += 5;
727         }
728         break;
729       case 6:
730         for (i=0; i<m; i++) {
731           v  = aa + 36*ai[i];
732           vi = aj + ai[i];
733           nz = ai[i+1] - ai[i];
734           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5];
735           while (nz--) {
736             idx = 6*(*vi++);
737             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
738             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
739             PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
740             v  += 36;
741           }
742           PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
743           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
744           x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5];
745           idiag  += 36;
746           i2     += 6;
747         }
748         break;
749       case 7:
750         for (i=0; i<m; i++) {
751           v  = aa + 49*ai[i];
752           vi = aj + ai[i];
753           nz = ai[i+1] - ai[i];
754           s[0] = b[i2];   s[1] = b[i2+1]; s[2] = b[i2+2];
755           s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5]; s[6] = b[i2+6];
756           while (nz--) {
757             idx = 7*(*vi++);
758             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
759             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
760             PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
761             v  += 49;
762           }
763           PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
764           x[i2]   += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
765           x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5]; x[i2+6] += xw[6];
766           idiag  += 49;
767           i2     += 7;
768         }
769         break;
770       default:
771         for (i=0; i<m; i++) {
772           v  = aa + bs2*ai[i];
773           vi = aj + ai[i];
774           nz = ai[i+1] - ai[i];
775 
776           PetscCall(PetscArraycpy(w,b+i2,bs));
777           /* copy all rows of x that are needed into contiguous space */
778           workt = work;
779           for (j=0; j<nz; j++) {
780             PetscCall(PetscArraycpy(workt,x + bs*(*vi++),bs));
781             workt += bs;
782           }
783           PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
784           PetscKernel_w_gets_w_plus_Ar_times_v(bs,bs,w,idiag,x+i2);
785 
786           idiag += bs2;
787           i2    += bs;
788         }
789         break;
790       }
791       PetscCall(PetscLogFlops(2.0*bs2*a->nz));
792     }
793     if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
794       idiag = a->idiag+bs2*(a->mbs-1);
795       i2 = bs * (m-1);
796       switch (bs) {
797       case 1:
798         for (i=m-1; i>=0; i--) {
799           v  = aa + ai[i];
800           vi = aj + ai[i];
801           nz = ai[i+1] - ai[i];
802           s[0] = b[i2];
803           for (j=0; j<nz; j++) {
804             xw[0] = x[vi[j]];
805             PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
806           }
807           PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
808           x[i2] += xw[0];
809           idiag -= 1;
810           i2    -= 1;
811         }
812         break;
813       case 2:
814         for (i=m-1; i>=0; i--) {
815           v  = aa + 4*ai[i];
816           vi = aj + ai[i];
817           nz = ai[i+1] - ai[i];
818           s[0] = b[i2]; s[1] = b[i2+1];
819           for (j=0; j<nz; j++) {
820             idx = 2*vi[j];
821             it  = 4*j;
822             xw[0] = x[idx]; xw[1] = x[1+idx];
823             PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
824           }
825           PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
826           x[i2]  += xw[0]; x[i2+1] += xw[1];
827           idiag  -= 4;
828           i2     -= 2;
829         }
830         break;
831       case 3:
832         for (i=m-1; i>=0; i--) {
833           v  = aa + 9*ai[i];
834           vi = aj + ai[i];
835           nz = ai[i+1] - ai[i];
836           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
837           while (nz--) {
838             idx = 3*(*vi++);
839             xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
840             PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
841             v  += 9;
842           }
843           PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
844           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
845           idiag  -= 9;
846           i2     -= 3;
847         }
848         break;
849       case 4:
850         for (i=m-1; i>=0; i--) {
851           v  = aa + 16*ai[i];
852           vi = aj + ai[i];
853           nz = ai[i+1] - ai[i];
854           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3];
855           while (nz--) {
856             idx = 4*(*vi++);
857             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
858             PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
859             v  += 16;
860           }
861           PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
862           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3];
863           idiag  -= 16;
864           i2     -= 4;
865         }
866         break;
867       case 5:
868         for (i=m-1; i>=0; i--) {
869           v  = aa + 25*ai[i];
870           vi = aj + ai[i];
871           nz = ai[i+1] - ai[i];
872           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4];
873           while (nz--) {
874             idx = 5*(*vi++);
875             xw[0]  = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
876             PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
877             v  += 25;
878           }
879           PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
880           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3]; x[i2+4] += xw[4];
881           idiag  -= 25;
882           i2     -= 5;
883         }
884         break;
885       case 6:
886         for (i=m-1; i>=0; i--) {
887           v  = aa + 36*ai[i];
888           vi = aj + ai[i];
889           nz = ai[i+1] - ai[i];
890           s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5];
891           while (nz--) {
892             idx = 6*(*vi++);
893             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
894             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
895             PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
896             v  += 36;
897           }
898           PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
899           x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
900           x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5];
901           idiag  -= 36;
902           i2     -= 6;
903         }
904         break;
905       case 7:
906         for (i=m-1; i>=0; i--) {
907           v  = aa + 49*ai[i];
908           vi = aj + ai[i];
909           nz = ai[i+1] - ai[i];
910           s[0] = b[i2];   s[1] = b[i2+1]; s[2] = b[i2+2];
911           s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5]; s[6] = b[i2+6];
912           while (nz--) {
913             idx = 7*(*vi++);
914             xw[0] = x[idx];   xw[1] = x[1+idx]; xw[2] = x[2+idx];
915             xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
916             PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
917             v  += 49;
918           }
919           PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
920           x[i2] +=   xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
921           x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5]; x[i2+6] += xw[6];
922           idiag  -= 49;
923           i2     -= 7;
924         }
925         break;
926       default:
927         for (i=m-1; i>=0; i--) {
928           v  = aa + bs2*ai[i];
929           vi = aj + ai[i];
930           nz = ai[i+1] - ai[i];
931 
932           PetscCall(PetscArraycpy(w,b+i2,bs));
933           /* copy all rows of x that are needed into contiguous space */
934           workt = work;
935           for (j=0; j<nz; j++) {
936             PetscCall(PetscArraycpy(workt,x + bs*(*vi++),bs));
937             workt += bs;
938           }
939           PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
940           PetscKernel_w_gets_w_plus_Ar_times_v(bs,bs,w,idiag,x+i2);
941 
942           idiag -= bs2;
943           i2    -= bs;
944         }
945         break;
946       }
947       PetscCall(PetscLogFlops(2.0*bs2*(a->nz)));
948     }
949   }
950   PetscCall(VecRestoreArray(xx,&x));
951   PetscCall(VecRestoreArrayRead(bb,&b));
952   PetscFunctionReturn(0);
953 }
954 
955 /*
956     Special version for direct calls from Fortran (Used in PETSc-fun3d)
957 */
958 #if defined(PETSC_HAVE_FORTRAN_CAPS)
959 #define matsetvaluesblocked4_ MATSETVALUESBLOCKED4
960 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
961 #define matsetvaluesblocked4_ matsetvaluesblocked4
962 #endif
963 
964 PETSC_EXTERN void matsetvaluesblocked4_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[])
965 {
966   Mat               A  = *AA;
967   Mat_SeqBAIJ       *a = (Mat_SeqBAIJ*)A->data;
968   PetscInt          *rp,k,low,high,t,ii,jj,row,nrow,i,col,l,N,m = *mm,n = *nn;
969   PetscInt          *ai    =a->i,*ailen=a->ilen;
970   PetscInt          *aj    =a->j,stepval,lastcol = -1;
971   const PetscScalar *value = v;
972   MatScalar         *ap,*aa = a->a,*bap;
973 
974   PetscFunctionBegin;
975   if (A->rmap->bs != 4) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Can only be called with a block size of 4");
976   stepval = (n-1)*4;
977   for (k=0; k<m; k++) { /* loop over added rows */
978     row  = im[k];
979     rp   = aj + ai[row];
980     ap   = aa + 16*ai[row];
981     nrow = ailen[row];
982     low  = 0;
983     high = nrow;
984     for (l=0; l<n; l++) { /* loop over added columns */
985       col = in[l];
986       if (col <= lastcol)  low = 0;
987       else                high = nrow;
988       lastcol = col;
989       value   = v + k*(stepval+4 + l)*4;
990       while (high-low > 7) {
991         t = (low+high)/2;
992         if (rp[t] > col) high = t;
993         else             low  = t;
994       }
995       for (i=low; i<high; i++) {
996         if (rp[i] > col) break;
997         if (rp[i] == col) {
998           bap = ap +  16*i;
999           for (ii=0; ii<4; ii++,value+=stepval) {
1000             for (jj=ii; jj<16; jj+=4) {
1001               bap[jj] += *value++;
1002             }
1003           }
1004           goto noinsert2;
1005         }
1006       }
1007       N = nrow++ - 1;
1008       high++; /* added new column index thus must search to one higher than before */
1009       /* shift up all the later entries in this row */
1010       for (ii=N; ii>=i; ii--) {
1011         rp[ii+1] = rp[ii];
1012         PetscCallVoid(PetscArraycpy(ap+16*(ii+1),ap+16*(ii),16));
1013       }
1014       if (N >= i) {
1015         PetscCallVoid(PetscArrayzero(ap+16*i,16));
1016       }
1017       rp[i] = col;
1018       bap   = ap +  16*i;
1019       for (ii=0; ii<4; ii++,value+=stepval) {
1020         for (jj=ii; jj<16; jj+=4) {
1021           bap[jj] = *value++;
1022         }
1023       }
1024       noinsert2:;
1025       low = i;
1026     }
1027     ailen[row] = nrow;
1028   }
1029   PetscFunctionReturnVoid();
1030 }
1031 
1032 #if defined(PETSC_HAVE_FORTRAN_CAPS)
1033 #define matsetvalues4_ MATSETVALUES4
1034 #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
1035 #define matsetvalues4_ matsetvalues4
1036 #endif
1037 
1038 PETSC_EXTERN void matsetvalues4_(Mat *AA,PetscInt *mm,PetscInt *im,PetscInt *nn,PetscInt *in,PetscScalar *v)
1039 {
1040   Mat         A  = *AA;
1041   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1042   PetscInt    *rp,k,low,high,t,row,nrow,i,col,l,N,n = *nn,m = *mm;
1043   PetscInt    *ai=a->i,*ailen=a->ilen;
1044   PetscInt    *aj=a->j,brow,bcol;
1045   PetscInt    ridx,cidx,lastcol = -1;
1046   MatScalar   *ap,value,*aa=a->a,*bap;
1047 
1048   PetscFunctionBegin;
1049   for (k=0; k<m; k++) { /* loop over added rows */
1050     row  = im[k]; brow = row/4;
1051     rp   = aj + ai[brow];
1052     ap   = aa + 16*ai[brow];
1053     nrow = ailen[brow];
1054     low  = 0;
1055     high = nrow;
1056     for (l=0; l<n; l++) { /* loop over added columns */
1057       col   = in[l]; bcol = col/4;
1058       ridx  = row % 4; cidx = col % 4;
1059       value = v[l + k*n];
1060       if (col <= lastcol)  low = 0;
1061       else                high = nrow;
1062       lastcol = col;
1063       while (high-low > 7) {
1064         t = (low+high)/2;
1065         if (rp[t] > bcol) high = t;
1066         else              low  = t;
1067       }
1068       for (i=low; i<high; i++) {
1069         if (rp[i] > bcol) break;
1070         if (rp[i] == bcol) {
1071           bap   = ap +  16*i + 4*cidx + ridx;
1072           *bap += value;
1073           goto noinsert1;
1074         }
1075       }
1076       N = nrow++ - 1;
1077       high++; /* added new column thus must search to one higher than before */
1078       /* shift up all the later entries in this row */
1079       PetscCallVoid(PetscArraymove(rp+i+1,rp+i,N-i+1));
1080       PetscCallVoid(PetscArraymove(ap+16*i+16,ap+16*i,16*(N-i+1)));
1081       PetscCallVoid(PetscArrayzero(ap+16*i,16));
1082       rp[i]                    = bcol;
1083       ap[16*i + 4*cidx + ridx] = value;
1084 noinsert1:;
1085       low = i;
1086     }
1087     ailen[brow] = nrow;
1088   }
1089   PetscFunctionReturnVoid();
1090 }
1091 
1092 /*
1093      Checks for missing diagonals
1094 */
1095 PetscErrorCode MatMissingDiagonal_SeqBAIJ(Mat A,PetscBool  *missing,PetscInt *d)
1096 {
1097   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1098   PetscInt       *diag,*ii = a->i,i;
1099 
1100   PetscFunctionBegin;
1101   PetscCall(MatMarkDiagonal_SeqBAIJ(A));
1102   *missing = PETSC_FALSE;
1103   if (A->rmap->n > 0 && !ii) {
1104     *missing = PETSC_TRUE;
1105     if (d) *d = 0;
1106     PetscCall(PetscInfo(A,"Matrix has no entries therefore is missing diagonal\n"));
1107   } else {
1108     PetscInt n;
1109     n = PetscMin(a->mbs, a->nbs);
1110     diag = a->diag;
1111     for (i=0; i<n; i++) {
1112       if (diag[i] >= ii[i+1]) {
1113         *missing = PETSC_TRUE;
1114         if (d) *d = i;
1115         PetscCall(PetscInfo(A,"Matrix is missing block diagonal number %" PetscInt_FMT "\n",i));
1116         break;
1117       }
1118     }
1119   }
1120   PetscFunctionReturn(0);
1121 }
1122 
1123 PetscErrorCode MatMarkDiagonal_SeqBAIJ(Mat A)
1124 {
1125   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1126   PetscInt       i,j,m = a->mbs;
1127 
1128   PetscFunctionBegin;
1129   if (!a->diag) {
1130     PetscCall(PetscMalloc1(m,&a->diag));
1131     PetscCall(PetscLogObjectMemory((PetscObject)A,m*sizeof(PetscInt)));
1132     a->free_diag = PETSC_TRUE;
1133   }
1134   for (i=0; i<m; i++) {
1135     a->diag[i] = a->i[i+1];
1136     for (j=a->i[i]; j<a->i[i+1]; j++) {
1137       if (a->j[j] == i) {
1138         a->diag[i] = j;
1139         break;
1140       }
1141     }
1142   }
1143   PetscFunctionReturn(0);
1144 }
1145 
1146 static PetscErrorCode MatGetRowIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool blockcompressed,PetscInt *nn,const PetscInt *inia[],const PetscInt *inja[],PetscBool  *done)
1147 {
1148   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1149   PetscInt       i,j,n = a->mbs,nz = a->i[n],*tia,*tja,bs = A->rmap->bs,k,l,cnt;
1150   PetscInt       **ia = (PetscInt**)inia,**ja = (PetscInt**)inja;
1151 
1152   PetscFunctionBegin;
1153   *nn = n;
1154   if (!ia) PetscFunctionReturn(0);
1155   if (symmetric) {
1156     PetscCall(MatToSymmetricIJ_SeqAIJ(n,a->i,a->j,PETSC_TRUE,0,0,&tia,&tja));
1157     nz   = tia[n];
1158   } else {
1159     tia = a->i; tja = a->j;
1160   }
1161 
1162   if (!blockcompressed && bs > 1) {
1163     (*nn) *= bs;
1164     /* malloc & create the natural set of indices */
1165     PetscCall(PetscMalloc1((n+1)*bs,ia));
1166     if (n) {
1167       (*ia)[0] = oshift;
1168       for (j=1; j<bs; j++) {
1169         (*ia)[j] = (tia[1]-tia[0])*bs+(*ia)[j-1];
1170       }
1171     }
1172 
1173     for (i=1; i<n; i++) {
1174       (*ia)[i*bs] = (tia[i]-tia[i-1])*bs + (*ia)[i*bs-1];
1175       for (j=1; j<bs; j++) {
1176         (*ia)[i*bs+j] = (tia[i+1]-tia[i])*bs + (*ia)[i*bs+j-1];
1177       }
1178     }
1179     if (n) {
1180       (*ia)[n*bs] = (tia[n]-tia[n-1])*bs + (*ia)[n*bs-1];
1181     }
1182 
1183     if (inja) {
1184       PetscCall(PetscMalloc1(nz*bs*bs,ja));
1185       cnt = 0;
1186       for (i=0; i<n; i++) {
1187         for (j=0; j<bs; j++) {
1188           for (k=tia[i]; k<tia[i+1]; k++) {
1189             for (l=0; l<bs; l++) {
1190               (*ja)[cnt++] = bs*tja[k] + l;
1191             }
1192           }
1193         }
1194       }
1195     }
1196 
1197     if (symmetric) { /* deallocate memory allocated in MatToSymmetricIJ_SeqAIJ() */
1198       PetscCall(PetscFree(tia));
1199       PetscCall(PetscFree(tja));
1200     }
1201   } else if (oshift == 1) {
1202     if (symmetric) {
1203       nz = tia[A->rmap->n/bs];
1204       /*  add 1 to i and j indices */
1205       for (i=0; i<A->rmap->n/bs+1; i++) tia[i] = tia[i] + 1;
1206       *ia = tia;
1207       if (ja) {
1208         for (i=0; i<nz; i++) tja[i] = tja[i] + 1;
1209         *ja = tja;
1210       }
1211     } else {
1212       nz = a->i[A->rmap->n/bs];
1213       /* malloc space and  add 1 to i and j indices */
1214       PetscCall(PetscMalloc1(A->rmap->n/bs+1,ia));
1215       for (i=0; i<A->rmap->n/bs+1; i++) (*ia)[i] = a->i[i] + 1;
1216       if (ja) {
1217         PetscCall(PetscMalloc1(nz,ja));
1218         for (i=0; i<nz; i++) (*ja)[i] = a->j[i] + 1;
1219       }
1220     }
1221   } else {
1222     *ia = tia;
1223     if (ja) *ja = tja;
1224   }
1225   PetscFunctionReturn(0);
1226 }
1227 
1228 static PetscErrorCode MatRestoreRowIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool blockcompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool  *done)
1229 {
1230   PetscFunctionBegin;
1231   if (!ia) PetscFunctionReturn(0);
1232   if ((!blockcompressed && A->rmap->bs > 1) || (symmetric || oshift == 1)) {
1233     PetscCall(PetscFree(*ia));
1234     if (ja) PetscCall(PetscFree(*ja));
1235   }
1236   PetscFunctionReturn(0);
1237 }
1238 
1239 PetscErrorCode MatDestroy_SeqBAIJ(Mat A)
1240 {
1241   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1242 
1243   PetscFunctionBegin;
1244 #if defined(PETSC_USE_LOG)
1245   PetscLogObjectState((PetscObject)A,"Rows=%" PetscInt_FMT ", Cols=%" PetscInt_FMT ", NZ=%" PetscInt_FMT,A->rmap->N,A->cmap->n,a->nz);
1246 #endif
1247   PetscCall(MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i));
1248   PetscCall(ISDestroy(&a->row));
1249   PetscCall(ISDestroy(&a->col));
1250   if (a->free_diag) PetscCall(PetscFree(a->diag));
1251   PetscCall(PetscFree(a->idiag));
1252   if (a->free_imax_ilen) PetscCall(PetscFree2(a->imax,a->ilen));
1253   PetscCall(PetscFree(a->solve_work));
1254   PetscCall(PetscFree(a->mult_work));
1255   PetscCall(PetscFree(a->sor_workt));
1256   PetscCall(PetscFree(a->sor_work));
1257   PetscCall(ISDestroy(&a->icol));
1258   PetscCall(PetscFree(a->saved_values));
1259   PetscCall(PetscFree2(a->compressedrow.i,a->compressedrow.rindex));
1260 
1261   PetscCall(MatDestroy(&a->sbaijMat));
1262   PetscCall(MatDestroy(&a->parent));
1263   PetscCall(PetscFree(A->data));
1264 
1265   PetscCall(PetscObjectChangeTypeName((PetscObject)A,NULL));
1266   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJGetArray_C",NULL));
1267   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJRestoreArray_C",NULL));
1268   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL));
1269   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL));
1270   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetColumnIndices_C",NULL));
1271   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqaij_C",NULL));
1272   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqsbaij_C",NULL));
1273   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetPreallocation_C",NULL));
1274   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetPreallocationCSR_C",NULL));
1275   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqbstrm_C",NULL));
1276   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL));
1277 #if defined(PETSC_HAVE_HYPRE)
1278   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_hypre_C",NULL));
1279 #endif
1280   PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_is_C",NULL));
1281   PetscFunctionReturn(0);
1282 }
1283 
1284 PetscErrorCode MatSetOption_SeqBAIJ(Mat A,MatOption op,PetscBool flg)
1285 {
1286   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1287 
1288   PetscFunctionBegin;
1289   switch (op) {
1290   case MAT_ROW_ORIENTED:
1291     a->roworiented = flg;
1292     break;
1293   case MAT_KEEP_NONZERO_PATTERN:
1294     a->keepnonzeropattern = flg;
1295     break;
1296   case MAT_NEW_NONZERO_LOCATIONS:
1297     a->nonew = (flg ? 0 : 1);
1298     break;
1299   case MAT_NEW_NONZERO_LOCATION_ERR:
1300     a->nonew = (flg ? -1 : 0);
1301     break;
1302   case MAT_NEW_NONZERO_ALLOCATION_ERR:
1303     a->nonew = (flg ? -2 : 0);
1304     break;
1305   case MAT_UNUSED_NONZERO_LOCATION_ERR:
1306     a->nounused = (flg ? -1 : 0);
1307     break;
1308   case MAT_FORCE_DIAGONAL_ENTRIES:
1309   case MAT_IGNORE_OFF_PROC_ENTRIES:
1310   case MAT_USE_HASH_TABLE:
1311   case MAT_SORTED_FULL:
1312     PetscCall(PetscInfo(A,"Option %s ignored\n",MatOptions[op]));
1313     break;
1314   case MAT_SPD:
1315   case MAT_SYMMETRIC:
1316   case MAT_STRUCTURALLY_SYMMETRIC:
1317   case MAT_HERMITIAN:
1318   case MAT_SYMMETRY_ETERNAL:
1319   case MAT_SUBMAT_SINGLEIS:
1320   case MAT_STRUCTURE_ONLY:
1321     /* These options are handled directly by MatSetOption() */
1322     break;
1323   default:
1324     SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op);
1325   }
1326   PetscFunctionReturn(0);
1327 }
1328 
1329 /* used for both SeqBAIJ and SeqSBAIJ matrices */
1330 PetscErrorCode MatGetRow_SeqBAIJ_private(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v,PetscInt *ai,PetscInt *aj,PetscScalar *aa)
1331 {
1332   PetscInt       itmp,i,j,k,M,bn,bp,*idx_i,bs,bs2;
1333   MatScalar      *aa_i;
1334   PetscScalar    *v_i;
1335 
1336   PetscFunctionBegin;
1337   bs  = A->rmap->bs;
1338   bs2 = bs*bs;
1339   PetscCheck(row >= 0 && row < A->rmap->N,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %" PetscInt_FMT " out of range", row);
1340 
1341   bn  = row/bs;   /* Block number */
1342   bp  = row % bs; /* Block Position */
1343   M   = ai[bn+1] - ai[bn];
1344   *nz = bs*M;
1345 
1346   if (v) {
1347     *v = NULL;
1348     if (*nz) {
1349       PetscCall(PetscMalloc1(*nz,v));
1350       for (i=0; i<M; i++) { /* for each block in the block row */
1351         v_i  = *v + i*bs;
1352         aa_i = aa + bs2*(ai[bn] + i);
1353         for (j=bp,k=0; j<bs2; j+=bs,k++) v_i[k] = aa_i[j];
1354       }
1355     }
1356   }
1357 
1358   if (idx) {
1359     *idx = NULL;
1360     if (*nz) {
1361       PetscCall(PetscMalloc1(*nz,idx));
1362       for (i=0; i<M; i++) { /* for each block in the block row */
1363         idx_i = *idx + i*bs;
1364         itmp  = bs*aj[ai[bn] + i];
1365         for (j=0; j<bs; j++) idx_i[j] = itmp++;
1366       }
1367     }
1368   }
1369   PetscFunctionReturn(0);
1370 }
1371 
1372 PetscErrorCode MatGetRow_SeqBAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1373 {
1374   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1375 
1376   PetscFunctionBegin;
1377   PetscCall(MatGetRow_SeqBAIJ_private(A,row,nz,idx,v,a->i,a->j,a->a));
1378   PetscFunctionReturn(0);
1379 }
1380 
1381 PetscErrorCode MatRestoreRow_SeqBAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1382 {
1383   PetscFunctionBegin;
1384   if (nz)  *nz = 0;
1385   if (idx) PetscCall(PetscFree(*idx));
1386   if (v)   PetscCall(PetscFree(*v));
1387   PetscFunctionReturn(0);
1388 }
1389 
1390 PetscErrorCode MatTranspose_SeqBAIJ(Mat A,MatReuse reuse,Mat *B)
1391 {
1392   Mat_SeqBAIJ    *a=(Mat_SeqBAIJ*)A->data,*at;
1393   Mat            C;
1394   PetscInt       i,j,k,*aj=a->j,*ai=a->i,bs=A->rmap->bs,mbs=a->mbs,nbs=a->nbs,*atfill;
1395   PetscInt       bs2=a->bs2,*ati,*atj,anzj,kr;
1396   MatScalar      *ata,*aa=a->a;
1397 
1398   PetscFunctionBegin;
1399   PetscCall(PetscCalloc1(1+nbs,&atfill));
1400   if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_INPLACE_MATRIX) {
1401     for (i=0; i<ai[mbs]; i++) atfill[aj[i]] += 1; /* count num of non-zeros in row aj[i] */
1402 
1403     PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&C));
1404     PetscCall(MatSetSizes(C,A->cmap->n,A->rmap->N,A->cmap->n,A->rmap->N));
1405     PetscCall(MatSetType(C,((PetscObject)A)->type_name));
1406     PetscCall(MatSeqBAIJSetPreallocation(C,bs,0,atfill));
1407 
1408     at  = (Mat_SeqBAIJ*)C->data;
1409     ati = at->i;
1410     for (i=0; i<nbs; i++) at->ilen[i] = at->imax[i] = ati[i+1] - ati[i];
1411   } else {
1412     C = *B;
1413     at = (Mat_SeqBAIJ*)C->data;
1414     ati = at->i;
1415   }
1416 
1417   atj = at->j;
1418   ata = at->a;
1419 
1420   /* Copy ati into atfill so we have locations of the next free space in atj */
1421   PetscCall(PetscArraycpy(atfill,ati,nbs));
1422 
1423   /* Walk through A row-wise and mark nonzero entries of A^T. */
1424   for (i=0; i<mbs; i++) {
1425     anzj = ai[i+1] - ai[i];
1426     for (j=0; j<anzj; j++) {
1427       atj[atfill[*aj]] = i;
1428       for (kr=0; kr<bs; kr++) {
1429         for (k=0; k<bs; k++) {
1430           ata[bs2*atfill[*aj]+k*bs+kr] = *aa++;
1431         }
1432       }
1433       atfill[*aj++] += 1;
1434     }
1435   }
1436   PetscCall(MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY));
1437   PetscCall(MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY));
1438 
1439   /* Clean up temporary space and complete requests. */
1440   PetscCall(PetscFree(atfill));
1441 
1442   if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) {
1443     PetscCall(MatSetBlockSizes(C,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs)));
1444     *B = C;
1445   } else {
1446     PetscCall(MatHeaderMerge(A,&C));
1447   }
1448   PetscFunctionReturn(0);
1449 }
1450 
1451 PetscErrorCode MatIsTranspose_SeqBAIJ(Mat A,Mat B,PetscReal tol,PetscBool  *f)
1452 {
1453   Mat            Btrans;
1454 
1455   PetscFunctionBegin;
1456   *f   = PETSC_FALSE;
1457   PetscCall(MatTranspose_SeqBAIJ(A,MAT_INITIAL_MATRIX,&Btrans));
1458   PetscCall(MatEqual_SeqBAIJ(B,Btrans,f));
1459   PetscCall(MatDestroy(&Btrans));
1460   PetscFunctionReturn(0);
1461 }
1462 
1463 /* Used for both SeqBAIJ and SeqSBAIJ matrices */
1464 PetscErrorCode MatView_SeqBAIJ_Binary(Mat mat,PetscViewer viewer)
1465 {
1466   Mat_SeqBAIJ    *A = (Mat_SeqBAIJ*)mat->data;
1467   PetscInt       header[4],M,N,m,bs,nz,cnt,i,j,k,l;
1468   PetscInt       *rowlens,*colidxs;
1469   PetscScalar    *matvals;
1470 
1471   PetscFunctionBegin;
1472   PetscCall(PetscViewerSetUp(viewer));
1473 
1474   M  = mat->rmap->N;
1475   N  = mat->cmap->N;
1476   m  = mat->rmap->n;
1477   bs = mat->rmap->bs;
1478   nz = bs*bs*A->nz;
1479 
1480   /* write matrix header */
1481   header[0] = MAT_FILE_CLASSID;
1482   header[1] = M; header[2] = N; header[3] = nz;
1483   PetscCall(PetscViewerBinaryWrite(viewer,header,4,PETSC_INT));
1484 
1485   /* store row lengths */
1486   PetscCall(PetscMalloc1(m,&rowlens));
1487   for (cnt=0, i=0; i<A->mbs; i++)
1488     for (j=0; j<bs; j++)
1489       rowlens[cnt++] = bs*(A->i[i+1] - A->i[i]);
1490   PetscCall(PetscViewerBinaryWrite(viewer,rowlens,m,PETSC_INT));
1491   PetscCall(PetscFree(rowlens));
1492 
1493   /* store column indices  */
1494   PetscCall(PetscMalloc1(nz,&colidxs));
1495   for (cnt=0, i=0; i<A->mbs; i++)
1496     for (k=0; k<bs; k++)
1497       for (j=A->i[i]; j<A->i[i+1]; j++)
1498         for (l=0; l<bs; l++)
1499           colidxs[cnt++] = bs*A->j[j] + l;
1500   PetscCheck(cnt == nz,PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: cnt = %" PetscInt_FMT " nz = %" PetscInt_FMT,cnt,nz);
1501   PetscCall(PetscViewerBinaryWrite(viewer,colidxs,nz,PETSC_INT));
1502   PetscCall(PetscFree(colidxs));
1503 
1504   /* store nonzero values */
1505   PetscCall(PetscMalloc1(nz,&matvals));
1506   for (cnt=0, i=0; i<A->mbs; i++)
1507     for (k=0; k<bs; k++)
1508       for (j=A->i[i]; j<A->i[i+1]; j++)
1509         for (l=0; l<bs; l++)
1510           matvals[cnt++] = A->a[bs*(bs*j + l) + k];
1511   PetscCheck(cnt == nz,PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: cnt = %" PetscInt_FMT " nz = %" PetscInt_FMT,cnt,nz);
1512   PetscCall(PetscViewerBinaryWrite(viewer,matvals,nz,PETSC_SCALAR));
1513   PetscCall(PetscFree(matvals));
1514 
1515   /* write block size option to the viewer's .info file */
1516   PetscCall(MatView_Binary_BlockSizes(mat,viewer));
1517   PetscFunctionReturn(0);
1518 }
1519 
1520 static PetscErrorCode MatView_SeqBAIJ_ASCII_structonly(Mat A,PetscViewer viewer)
1521 {
1522   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
1523   PetscInt       i,bs = A->rmap->bs,k;
1524 
1525   PetscFunctionBegin;
1526   PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE));
1527   for (i=0; i<a->mbs; i++) {
1528     PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT "-%" PetscInt_FMT ":",i*bs,i*bs+bs-1));
1529     for (k=a->i[i]; k<a->i[i+1]; k++) {
1530       PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT "-%" PetscInt_FMT ") ",bs*a->j[k],bs*a->j[k]+bs-1));
1531     }
1532     PetscCall(PetscViewerASCIIPrintf(viewer,"\n"));
1533   }
1534   PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE));
1535   PetscFunctionReturn(0);
1536 }
1537 
1538 static PetscErrorCode MatView_SeqBAIJ_ASCII(Mat A,PetscViewer viewer)
1539 {
1540   Mat_SeqBAIJ       *a = (Mat_SeqBAIJ*)A->data;
1541   PetscInt          i,j,bs = A->rmap->bs,k,l,bs2=a->bs2;
1542   PetscViewerFormat format;
1543 
1544   PetscFunctionBegin;
1545   if (A->structure_only) {
1546     PetscCall(MatView_SeqBAIJ_ASCII_structonly(A,viewer));
1547     PetscFunctionReturn(0);
1548   }
1549 
1550   PetscCall(PetscViewerGetFormat(viewer,&format));
1551   if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1552     PetscCall(PetscViewerASCIIPrintf(viewer,"  block size is %" PetscInt_FMT "\n",bs));
1553   } else if (format == PETSC_VIEWER_ASCII_MATLAB) {
1554     const char *matname;
1555     Mat        aij;
1556     PetscCall(MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&aij));
1557     PetscCall(PetscObjectGetName((PetscObject)A,&matname));
1558     PetscCall(PetscObjectSetName((PetscObject)aij,matname));
1559     PetscCall(MatView(aij,viewer));
1560     PetscCall(MatDestroy(&aij));
1561   } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
1562       PetscFunctionReturn(0);
1563   } else if (format == PETSC_VIEWER_ASCII_COMMON) {
1564     PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE));
1565     for (i=0; i<a->mbs; i++) {
1566       for (j=0; j<bs; j++) {
1567         PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT ":",i*bs+j));
1568         for (k=a->i[i]; k<a->i[i+1]; k++) {
1569           for (l=0; l<bs; l++) {
1570 #if defined(PETSC_USE_COMPLEX)
1571             if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) > 0.0 && PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1572               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %gi) ",bs*a->j[k]+l,
1573                                              (double)PetscRealPart(a->a[bs2*k + l*bs + j]),(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j])));
1574             } else if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) < 0.0 && PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1575               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %gi) ",bs*a->j[k]+l,
1576                                              (double)PetscRealPart(a->a[bs2*k + l*bs + j]),-(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j])));
1577             } else if (PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1578               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",bs*a->j[k]+l,(double)PetscRealPart(a->a[bs2*k + l*bs + j])));
1579             }
1580 #else
1581             if (a->a[bs2*k + l*bs + j] != 0.0) {
1582               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",bs*a->j[k]+l,(double)a->a[bs2*k + l*bs + j]));
1583             }
1584 #endif
1585           }
1586         }
1587         PetscCall(PetscViewerASCIIPrintf(viewer,"\n"));
1588       }
1589     }
1590     PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE));
1591   } else {
1592     PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_FALSE));
1593     for (i=0; i<a->mbs; i++) {
1594       for (j=0; j<bs; j++) {
1595         PetscCall(PetscViewerASCIIPrintf(viewer,"row %" PetscInt_FMT ":",i*bs+j));
1596         for (k=a->i[i]; k<a->i[i+1]; k++) {
1597           for (l=0; l<bs; l++) {
1598 #if defined(PETSC_USE_COMPLEX)
1599             if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) > 0.0) {
1600               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g + %g i) ",bs*a->j[k]+l,
1601                                              (double)PetscRealPart(a->a[bs2*k + l*bs + j]),(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j])));
1602             } else if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) < 0.0) {
1603               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g - %g i) ",bs*a->j[k]+l,
1604                                              (double)PetscRealPart(a->a[bs2*k + l*bs + j]),-(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j])));
1605             } else {
1606               PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",bs*a->j[k]+l,(double)PetscRealPart(a->a[bs2*k + l*bs + j])));
1607             }
1608 #else
1609             PetscCall(PetscViewerASCIIPrintf(viewer," (%" PetscInt_FMT ", %g) ",bs*a->j[k]+l,(double)a->a[bs2*k + l*bs + j]));
1610 #endif
1611           }
1612         }
1613         PetscCall(PetscViewerASCIIPrintf(viewer,"\n"));
1614       }
1615     }
1616     PetscCall(PetscViewerASCIIUseTabs(viewer,PETSC_TRUE));
1617   }
1618   PetscCall(PetscViewerFlush(viewer));
1619   PetscFunctionReturn(0);
1620 }
1621 
1622 #include <petscdraw.h>
1623 static PetscErrorCode MatView_SeqBAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
1624 {
1625   Mat               A = (Mat) Aa;
1626   Mat_SeqBAIJ       *a=(Mat_SeqBAIJ*)A->data;
1627   PetscInt          row,i,j,k,l,mbs=a->mbs,color,bs=A->rmap->bs,bs2=a->bs2;
1628   PetscReal         xl,yl,xr,yr,x_l,x_r,y_l,y_r;
1629   MatScalar         *aa;
1630   PetscViewer       viewer;
1631   PetscViewerFormat format;
1632 
1633   PetscFunctionBegin;
1634   PetscCall(PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer));
1635   PetscCall(PetscViewerGetFormat(viewer,&format));
1636   PetscCall(PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr));
1637 
1638   /* loop over matrix elements drawing boxes */
1639 
1640   if (format != PETSC_VIEWER_DRAW_CONTOUR) {
1641     PetscErrorCode ierr;
1642     ierr = PetscDrawCollectiveBegin(draw);PetscCall(ierr);
1643     /* Blue for negative, Cyan for zero and  Red for positive */
1644     color = PETSC_DRAW_BLUE;
1645     for (i=0,row=0; i<mbs; i++,row+=bs) {
1646       for (j=a->i[i]; j<a->i[i+1]; j++) {
1647         y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1648         x_l = a->j[j]*bs; x_r = x_l + 1.0;
1649         aa  = a->a + j*bs2;
1650         for (k=0; k<bs; k++) {
1651           for (l=0; l<bs; l++) {
1652             if (PetscRealPart(*aa++) >=  0.) continue;
1653             PetscCall(PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color));
1654           }
1655         }
1656       }
1657     }
1658     color = PETSC_DRAW_CYAN;
1659     for (i=0,row=0; i<mbs; i++,row+=bs) {
1660       for (j=a->i[i]; j<a->i[i+1]; j++) {
1661         y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1662         x_l = a->j[j]*bs; x_r = x_l + 1.0;
1663         aa  = a->a + j*bs2;
1664         for (k=0; k<bs; k++) {
1665           for (l=0; l<bs; l++) {
1666             if (PetscRealPart(*aa++) != 0.) continue;
1667             PetscCall(PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color));
1668           }
1669         }
1670       }
1671     }
1672     color = PETSC_DRAW_RED;
1673     for (i=0,row=0; i<mbs; i++,row+=bs) {
1674       for (j=a->i[i]; j<a->i[i+1]; j++) {
1675         y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1676         x_l = a->j[j]*bs; x_r = x_l + 1.0;
1677         aa  = a->a + j*bs2;
1678         for (k=0; k<bs; k++) {
1679           for (l=0; l<bs; l++) {
1680             if (PetscRealPart(*aa++) <= 0.) continue;
1681             PetscCall(PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color));
1682           }
1683         }
1684       }
1685     }
1686     ierr = PetscDrawCollectiveEnd(draw);PetscCall(ierr);
1687   } else {
1688     /* use contour shading to indicate magnitude of values */
1689     /* first determine max of all nonzero values */
1690     PetscReal      minv = 0.0, maxv = 0.0;
1691     PetscDraw      popup;
1692     PetscErrorCode ierr;
1693 
1694     for (i=0; i<a->nz*a->bs2; i++) {
1695       if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
1696     }
1697     if (minv >= maxv) maxv = minv + PETSC_SMALL;
1698     PetscCall(PetscDrawGetPopup(draw,&popup));
1699     PetscCall(PetscDrawScalePopup(popup,0.0,maxv));
1700 
1701     ierr = PetscDrawCollectiveBegin(draw);PetscCall(ierr);
1702     for (i=0,row=0; i<mbs; i++,row+=bs) {
1703       for (j=a->i[i]; j<a->i[i+1]; j++) {
1704         y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1705         x_l = a->j[j]*bs; x_r = x_l + 1.0;
1706         aa  = a->a + j*bs2;
1707         for (k=0; k<bs; k++) {
1708           for (l=0; l<bs; l++) {
1709             MatScalar v = *aa++;
1710             color = PetscDrawRealToColor(PetscAbsScalar(v),minv,maxv);
1711             PetscCall(PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color));
1712           }
1713         }
1714       }
1715     }
1716     ierr = PetscDrawCollectiveEnd(draw);PetscCall(ierr);
1717   }
1718   PetscFunctionReturn(0);
1719 }
1720 
1721 static PetscErrorCode MatView_SeqBAIJ_Draw(Mat A,PetscViewer viewer)
1722 {
1723   PetscReal      xl,yl,xr,yr,w,h;
1724   PetscDraw      draw;
1725   PetscBool      isnull;
1726 
1727   PetscFunctionBegin;
1728   PetscCall(PetscViewerDrawGetDraw(viewer,0,&draw));
1729   PetscCall(PetscDrawIsNull(draw,&isnull));
1730   if (isnull) PetscFunctionReturn(0);
1731 
1732   xr   = A->cmap->n; yr = A->rmap->N; h = yr/10.0; w = xr/10.0;
1733   xr  += w;          yr += h;        xl = -w;     yl = -h;
1734   PetscCall(PetscDrawSetCoordinates(draw,xl,yl,xr,yr));
1735   PetscCall(PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer));
1736   PetscCall(PetscDrawZoom(draw,MatView_SeqBAIJ_Draw_Zoom,A));
1737   PetscCall(PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL));
1738   PetscCall(PetscDrawSave(draw));
1739   PetscFunctionReturn(0);
1740 }
1741 
1742 PetscErrorCode MatView_SeqBAIJ(Mat A,PetscViewer viewer)
1743 {
1744   PetscBool      iascii,isbinary,isdraw;
1745 
1746   PetscFunctionBegin;
1747   PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii));
1748   PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary));
1749   PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw));
1750   if (iascii) {
1751     PetscCall(MatView_SeqBAIJ_ASCII(A,viewer));
1752   } else if (isbinary) {
1753     PetscCall(MatView_SeqBAIJ_Binary(A,viewer));
1754   } else if (isdraw) {
1755     PetscCall(MatView_SeqBAIJ_Draw(A,viewer));
1756   } else {
1757     Mat B;
1758     PetscCall(MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&B));
1759     PetscCall(MatView(B,viewer));
1760     PetscCall(MatDestroy(&B));
1761   }
1762   PetscFunctionReturn(0);
1763 }
1764 
1765 PetscErrorCode MatGetValues_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
1766 {
1767   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1768   PetscInt    *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
1769   PetscInt    *ai = a->i,*ailen = a->ilen;
1770   PetscInt    brow,bcol,ridx,cidx,bs=A->rmap->bs,bs2=a->bs2;
1771   MatScalar   *ap,*aa = a->a;
1772 
1773   PetscFunctionBegin;
1774   for (k=0; k<m; k++) { /* loop over rows */
1775     row = im[k]; brow = row/bs;
1776     if (row < 0) {v += n; continue;} /* negative row */
1777     PetscCheck(row < A->rmap->N,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %" PetscInt_FMT " too large", row);
1778     rp   = aj + ai[brow]; ap = aa + bs2*ai[brow];
1779     nrow = ailen[brow];
1780     for (l=0; l<n; l++) { /* loop over columns */
1781       if (in[l] < 0) {v++; continue;} /* negative column */
1782       PetscCheck(in[l] < A->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column %" PetscInt_FMT " too large", in[l]);
1783       col  = in[l];
1784       bcol = col/bs;
1785       cidx = col%bs;
1786       ridx = row%bs;
1787       high = nrow;
1788       low  = 0; /* assume unsorted */
1789       while (high-low > 5) {
1790         t = (low+high)/2;
1791         if (rp[t] > bcol) high = t;
1792         else             low  = t;
1793       }
1794       for (i=low; i<high; i++) {
1795         if (rp[i] > bcol) break;
1796         if (rp[i] == bcol) {
1797           *v++ = ap[bs2*i+bs*cidx+ridx];
1798           goto finished;
1799         }
1800       }
1801       *v++ = 0.0;
1802 finished:;
1803     }
1804   }
1805   PetscFunctionReturn(0);
1806 }
1807 
1808 PetscErrorCode MatSetValuesBlocked_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
1809 {
1810   Mat_SeqBAIJ       *a = (Mat_SeqBAIJ*)A->data;
1811   PetscInt          *rp,k,low,high,t,ii,jj,row,nrow,i,col,l,rmax,N,lastcol = -1;
1812   PetscInt          *imax=a->imax,*ai=a->i,*ailen=a->ilen;
1813   PetscInt          *aj        =a->j,nonew=a->nonew,bs2=a->bs2,bs=A->rmap->bs,stepval;
1814   PetscBool         roworiented=a->roworiented;
1815   const PetscScalar *value     = v;
1816   MatScalar         *ap=NULL,*aa = a->a,*bap;
1817 
1818   PetscFunctionBegin;
1819   if (roworiented) {
1820     stepval = (n-1)*bs;
1821   } else {
1822     stepval = (m-1)*bs;
1823   }
1824   for (k=0; k<m; k++) { /* loop over added rows */
1825     row = im[k];
1826     if (row < 0) continue;
1827     PetscCheck(row < a->mbs,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Block row index too large %" PetscInt_FMT " max %" PetscInt_FMT,row,a->mbs-1);
1828     rp   = aj + ai[row];
1829     if (!A->structure_only) ap = aa + bs2*ai[row];
1830     rmax = imax[row];
1831     nrow = ailen[row];
1832     low  = 0;
1833     high = nrow;
1834     for (l=0; l<n; l++) { /* loop over added columns */
1835       if (in[l] < 0) continue;
1836       PetscCheck(in[l] < a->nbs,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Block column index too large %" PetscInt_FMT " max %" PetscInt_FMT,in[l],a->nbs-1);
1837       col = in[l];
1838       if (!A->structure_only) {
1839         if (roworiented) {
1840           value = v + (k*(stepval+bs) + l)*bs;
1841         } else {
1842           value = v + (l*(stepval+bs) + k)*bs;
1843         }
1844       }
1845       if (col <= lastcol) low = 0;
1846       else high = nrow;
1847       lastcol = col;
1848       while (high-low > 7) {
1849         t = (low+high)/2;
1850         if (rp[t] > col) high = t;
1851         else             low  = t;
1852       }
1853       for (i=low; i<high; i++) {
1854         if (rp[i] > col) break;
1855         if (rp[i] == col) {
1856           if (A->structure_only) goto noinsert2;
1857           bap = ap +  bs2*i;
1858           if (roworiented) {
1859             if (is == ADD_VALUES) {
1860               for (ii=0; ii<bs; ii++,value+=stepval) {
1861                 for (jj=ii; jj<bs2; jj+=bs) {
1862                   bap[jj] += *value++;
1863                 }
1864               }
1865             } else {
1866               for (ii=0; ii<bs; ii++,value+=stepval) {
1867                 for (jj=ii; jj<bs2; jj+=bs) {
1868                   bap[jj] = *value++;
1869                 }
1870               }
1871             }
1872           } else {
1873             if (is == ADD_VALUES) {
1874               for (ii=0; ii<bs; ii++,value+=bs+stepval) {
1875                 for (jj=0; jj<bs; jj++) {
1876                   bap[jj] += value[jj];
1877                 }
1878                 bap += bs;
1879               }
1880             } else {
1881               for (ii=0; ii<bs; ii++,value+=bs+stepval) {
1882                 for (jj=0; jj<bs; jj++) {
1883                   bap[jj]  = value[jj];
1884                 }
1885                 bap += bs;
1886               }
1887             }
1888           }
1889           goto noinsert2;
1890         }
1891       }
1892       if (nonew == 1) goto noinsert2;
1893       PetscCheck(nonew != -1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new blocked index new nonzero block (%" PetscInt_FMT ", %" PetscInt_FMT ") in the matrix", row, col);
1894       if (A->structure_only) {
1895         MatSeqXAIJReallocateAIJ_structure_only(A,a->mbs,bs2,nrow,row,col,rmax,ai,aj,rp,imax,nonew,MatScalar);
1896       } else {
1897         MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
1898       }
1899       N = nrow++ - 1; high++;
1900       /* shift up all the later entries in this row */
1901       PetscCall(PetscArraymove(rp+i+1,rp+i,N-i+1));
1902       rp[i] = col;
1903       if (!A->structure_only) {
1904         PetscCall(PetscArraymove(ap+bs2*(i+1),ap+bs2*i,bs2*(N-i+1)));
1905         bap   = ap +  bs2*i;
1906         if (roworiented) {
1907           for (ii=0; ii<bs; ii++,value+=stepval) {
1908             for (jj=ii; jj<bs2; jj+=bs) {
1909               bap[jj] = *value++;
1910             }
1911           }
1912         } else {
1913           for (ii=0; ii<bs; ii++,value+=stepval) {
1914             for (jj=0; jj<bs; jj++) {
1915               *bap++ = *value++;
1916             }
1917           }
1918         }
1919       }
1920 noinsert2:;
1921       low = i;
1922     }
1923     ailen[row] = nrow;
1924   }
1925   PetscFunctionReturn(0);
1926 }
1927 
1928 PetscErrorCode MatAssemblyEnd_SeqBAIJ(Mat A,MatAssemblyType mode)
1929 {
1930   Mat_SeqBAIJ    *a     = (Mat_SeqBAIJ*)A->data;
1931   PetscInt       fshift = 0,i,*ai = a->i,*aj = a->j,*imax = a->imax;
1932   PetscInt       m      = A->rmap->N,*ip,N,*ailen = a->ilen;
1933   PetscInt       mbs  = a->mbs,bs2 = a->bs2,rmax = 0;
1934   MatScalar      *aa  = a->a,*ap;
1935   PetscReal      ratio=0.6;
1936 
1937   PetscFunctionBegin;
1938   if (mode == MAT_FLUSH_ASSEMBLY) PetscFunctionReturn(0);
1939 
1940   if (m) rmax = ailen[0];
1941   for (i=1; i<mbs; i++) {
1942     /* move each row back by the amount of empty slots (fshift) before it*/
1943     fshift += imax[i-1] - ailen[i-1];
1944     rmax    = PetscMax(rmax,ailen[i]);
1945     if (fshift) {
1946       ip = aj + ai[i];
1947       ap = aa + bs2*ai[i];
1948       N  = ailen[i];
1949       PetscCall(PetscArraymove(ip-fshift,ip,N));
1950       if (!A->structure_only) {
1951         PetscCall(PetscArraymove(ap-bs2*fshift,ap,bs2*N));
1952       }
1953     }
1954     ai[i] = ai[i-1] + ailen[i-1];
1955   }
1956   if (mbs) {
1957     fshift += imax[mbs-1] - ailen[mbs-1];
1958     ai[mbs] = ai[mbs-1] + ailen[mbs-1];
1959   }
1960 
1961   /* reset ilen and imax for each row */
1962   a->nonzerorowcnt = 0;
1963   if (A->structure_only) {
1964     PetscCall(PetscFree2(a->imax,a->ilen));
1965   } else { /* !A->structure_only */
1966     for (i=0; i<mbs; i++) {
1967       ailen[i] = imax[i] = ai[i+1] - ai[i];
1968       a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0);
1969     }
1970   }
1971   a->nz = ai[mbs];
1972 
1973   /* diagonals may have moved, so kill the diagonal pointers */
1974   a->idiagvalid = PETSC_FALSE;
1975   if (fshift && a->diag) {
1976     PetscCall(PetscFree(a->diag));
1977     PetscCall(PetscLogObjectMemory((PetscObject)A,-(mbs+1)*sizeof(PetscInt)));
1978     a->diag = NULL;
1979   }
1980   if (fshift) PetscCheck(a->nounused != -1,PETSC_COMM_SELF,PETSC_ERR_PLIB, "Unused space detected in matrix: %" PetscInt_FMT " X %" PetscInt_FMT " block size %" PetscInt_FMT ", %" PetscInt_FMT " unneeded", m, A->cmap->n, A->rmap->bs, fshift*bs2);
1981   PetscCall(PetscInfo(A,"Matrix size: %" PetscInt_FMT " X %" PetscInt_FMT ", block size %" PetscInt_FMT "; storage space: %" PetscInt_FMT " unneeded, %" PetscInt_FMT " used\n",m,A->cmap->n,A->rmap->bs,fshift*bs2,a->nz*bs2));
1982   PetscCall(PetscInfo(A,"Number of mallocs during MatSetValues is %" PetscInt_FMT "\n",a->reallocs));
1983   PetscCall(PetscInfo(A,"Most nonzeros blocks in any row is %" PetscInt_FMT "\n",rmax));
1984 
1985   A->info.mallocs    += a->reallocs;
1986   a->reallocs         = 0;
1987   A->info.nz_unneeded = (PetscReal)fshift*bs2;
1988   a->rmax             = rmax;
1989 
1990   if (!A->structure_only) {
1991     PetscCall(MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,mbs,ratio));
1992   }
1993   PetscFunctionReturn(0);
1994 }
1995 
1996 /*
1997    This function returns an array of flags which indicate the locations of contiguous
1998    blocks that should be zeroed. for eg: if bs = 3  and is = [0,1,2,3,5,6,7,8,9]
1999    then the resulting sizes = [3,1,1,3,1] corresponding to sets [(0,1,2),(3),(5),(6,7,8),(9)]
2000    Assume: sizes should be long enough to hold all the values.
2001 */
2002 static PetscErrorCode MatZeroRows_SeqBAIJ_Check_Blocks(PetscInt idx[],PetscInt n,PetscInt bs,PetscInt sizes[], PetscInt *bs_max)
2003 {
2004   PetscInt  i,j,k,row;
2005   PetscBool flg;
2006 
2007   PetscFunctionBegin;
2008   for (i=0,j=0; i<n; j++) {
2009     row = idx[i];
2010     if (row%bs!=0) { /* Not the beginning of a block */
2011       sizes[j] = 1;
2012       i++;
2013     } else if (i+bs > n) { /* complete block doesn't exist (at idx end) */
2014       sizes[j] = 1;         /* Also makes sure at least 'bs' values exist for next else */
2015       i++;
2016     } else { /* Beginning of the block, so check if the complete block exists */
2017       flg = PETSC_TRUE;
2018       for (k=1; k<bs; k++) {
2019         if (row+k != idx[i+k]) { /* break in the block */
2020           flg = PETSC_FALSE;
2021           break;
2022         }
2023       }
2024       if (flg) { /* No break in the bs */
2025         sizes[j] = bs;
2026         i       += bs;
2027       } else {
2028         sizes[j] = 1;
2029         i++;
2030       }
2031     }
2032   }
2033   *bs_max = j;
2034   PetscFunctionReturn(0);
2035 }
2036 
2037 PetscErrorCode MatZeroRows_SeqBAIJ(Mat A,PetscInt is_n,const PetscInt is_idx[],PetscScalar diag,Vec x, Vec b)
2038 {
2039   Mat_SeqBAIJ       *baij=(Mat_SeqBAIJ*)A->data;
2040   PetscInt          i,j,k,count,*rows;
2041   PetscInt          bs=A->rmap->bs,bs2=baij->bs2,*sizes,row,bs_max;
2042   PetscScalar       zero = 0.0;
2043   MatScalar         *aa;
2044   const PetscScalar *xx;
2045   PetscScalar       *bb;
2046 
2047   PetscFunctionBegin;
2048   /* fix right hand side if needed */
2049   if (x && b) {
2050     PetscCall(VecGetArrayRead(x,&xx));
2051     PetscCall(VecGetArray(b,&bb));
2052     for (i=0; i<is_n; i++) {
2053       bb[is_idx[i]] = diag*xx[is_idx[i]];
2054     }
2055     PetscCall(VecRestoreArrayRead(x,&xx));
2056     PetscCall(VecRestoreArray(b,&bb));
2057   }
2058 
2059   /* Make a copy of the IS and  sort it */
2060   /* allocate memory for rows,sizes */
2061   PetscCall(PetscMalloc2(is_n,&rows,2*is_n,&sizes));
2062 
2063   /* copy IS values to rows, and sort them */
2064   for (i=0; i<is_n; i++) rows[i] = is_idx[i];
2065   PetscCall(PetscSortInt(is_n,rows));
2066 
2067   if (baij->keepnonzeropattern) {
2068     for (i=0; i<is_n; i++) sizes[i] = 1;
2069     bs_max          = is_n;
2070   } else {
2071     PetscCall(MatZeroRows_SeqBAIJ_Check_Blocks(rows,is_n,bs,sizes,&bs_max));
2072     A->nonzerostate++;
2073   }
2074 
2075   for (i=0,j=0; i<bs_max; j+=sizes[i],i++) {
2076     row = rows[j];
2077     PetscCheck(row >= 0 && row <= A->rmap->N,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range",row);
2078     count = (baij->i[row/bs +1] - baij->i[row/bs])*bs;
2079     aa    = ((MatScalar*)(baij->a)) + baij->i[row/bs]*bs2 + (row%bs);
2080     if (sizes[i] == bs && !baij->keepnonzeropattern) {
2081       if (diag != (PetscScalar)0.0) {
2082         if (baij->ilen[row/bs] > 0) {
2083           baij->ilen[row/bs]       = 1;
2084           baij->j[baij->i[row/bs]] = row/bs;
2085 
2086           PetscCall(PetscArrayzero(aa,count*bs));
2087         }
2088         /* Now insert all the diagonal values for this bs */
2089         for (k=0; k<bs; k++) {
2090           PetscCall((*A->ops->setvalues)(A,1,rows+j+k,1,rows+j+k,&diag,INSERT_VALUES));
2091         }
2092       } else { /* (diag == 0.0) */
2093         baij->ilen[row/bs] = 0;
2094       } /* end (diag == 0.0) */
2095     } else { /* (sizes[i] != bs) */
2096       PetscAssert(sizes[i] == 1,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Internal Error. Value should be 1");
2097       for (k=0; k<count; k++) {
2098         aa[0] =  zero;
2099         aa   += bs;
2100       }
2101       if (diag != (PetscScalar)0.0) {
2102         PetscCall((*A->ops->setvalues)(A,1,rows+j,1,rows+j,&diag,INSERT_VALUES));
2103       }
2104     }
2105   }
2106 
2107   PetscCall(PetscFree2(rows,sizes));
2108   PetscCall(MatAssemblyEnd_SeqBAIJ(A,MAT_FINAL_ASSEMBLY));
2109   PetscFunctionReturn(0);
2110 }
2111 
2112 PetscErrorCode MatZeroRowsColumns_SeqBAIJ(Mat A,PetscInt is_n,const PetscInt is_idx[],PetscScalar diag,Vec x, Vec b)
2113 {
2114   Mat_SeqBAIJ       *baij=(Mat_SeqBAIJ*)A->data;
2115   PetscInt          i,j,k,count;
2116   PetscInt          bs   =A->rmap->bs,bs2=baij->bs2,row,col;
2117   PetscScalar       zero = 0.0;
2118   MatScalar         *aa;
2119   const PetscScalar *xx;
2120   PetscScalar       *bb;
2121   PetscBool         *zeroed,vecs = PETSC_FALSE;
2122 
2123   PetscFunctionBegin;
2124   /* fix right hand side if needed */
2125   if (x && b) {
2126     PetscCall(VecGetArrayRead(x,&xx));
2127     PetscCall(VecGetArray(b,&bb));
2128     vecs = PETSC_TRUE;
2129   }
2130 
2131   /* zero the columns */
2132   PetscCall(PetscCalloc1(A->rmap->n,&zeroed));
2133   for (i=0; i<is_n; i++) {
2134     PetscCheck(is_idx[i] >= 0 && is_idx[i] < A->rmap->N,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %" PetscInt_FMT " out of range",is_idx[i]);
2135     zeroed[is_idx[i]] = PETSC_TRUE;
2136   }
2137   for (i=0; i<A->rmap->N; i++) {
2138     if (!zeroed[i]) {
2139       row = i/bs;
2140       for (j=baij->i[row]; j<baij->i[row+1]; j++) {
2141         for (k=0; k<bs; k++) {
2142           col = bs*baij->j[j] + k;
2143           if (zeroed[col]) {
2144             aa = ((MatScalar*)(baij->a)) + j*bs2 + (i%bs) + bs*k;
2145             if (vecs) bb[i] -= aa[0]*xx[col];
2146             aa[0] = 0.0;
2147           }
2148         }
2149       }
2150     } else if (vecs) bb[i] = diag*xx[i];
2151   }
2152   PetscCall(PetscFree(zeroed));
2153   if (vecs) {
2154     PetscCall(VecRestoreArrayRead(x,&xx));
2155     PetscCall(VecRestoreArray(b,&bb));
2156   }
2157 
2158   /* zero the rows */
2159   for (i=0; i<is_n; i++) {
2160     row   = is_idx[i];
2161     count = (baij->i[row/bs +1] - baij->i[row/bs])*bs;
2162     aa    = ((MatScalar*)(baij->a)) + baij->i[row/bs]*bs2 + (row%bs);
2163     for (k=0; k<count; k++) {
2164       aa[0] =  zero;
2165       aa   += bs;
2166     }
2167     if (diag != (PetscScalar)0.0) {
2168       PetscCall((*A->ops->setvalues)(A,1,&row,1,&row,&diag,INSERT_VALUES));
2169     }
2170   }
2171   PetscCall(MatAssemblyEnd_SeqBAIJ(A,MAT_FINAL_ASSEMBLY));
2172   PetscFunctionReturn(0);
2173 }
2174 
2175 PetscErrorCode MatSetValues_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
2176 {
2177   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
2178   PetscInt       *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N,lastcol = -1;
2179   PetscInt       *imax=a->imax,*ai=a->i,*ailen=a->ilen;
2180   PetscInt       *aj  =a->j,nonew=a->nonew,bs=A->rmap->bs,brow,bcol;
2181   PetscInt       ridx,cidx,bs2=a->bs2;
2182   PetscBool      roworiented=a->roworiented;
2183   MatScalar      *ap=NULL,value=0.0,*aa=a->a,*bap;
2184 
2185   PetscFunctionBegin;
2186   for (k=0; k<m; k++) { /* loop over added rows */
2187     row  = im[k];
2188     brow = row/bs;
2189     if (row < 0) continue;
2190     PetscCheck(row < A->rmap->N,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT,row,A->rmap->N-1);
2191     rp   = aj + ai[brow];
2192     if (!A->structure_only) ap = aa + bs2*ai[brow];
2193     rmax = imax[brow];
2194     nrow = ailen[brow];
2195     low  = 0;
2196     high = nrow;
2197     for (l=0; l<n; l++) { /* loop over added columns */
2198       if (in[l] < 0) continue;
2199       PetscCheck(in[l] < A->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT,in[l],A->cmap->n-1);
2200       col  = in[l]; bcol = col/bs;
2201       ridx = row % bs; cidx = col % bs;
2202       if (!A->structure_only) {
2203         if (roworiented) {
2204           value = v[l + k*n];
2205         } else {
2206           value = v[k + l*m];
2207         }
2208       }
2209       if (col <= lastcol) low = 0; else high = nrow;
2210       lastcol = col;
2211       while (high-low > 7) {
2212         t = (low+high)/2;
2213         if (rp[t] > bcol) high = t;
2214         else              low  = t;
2215       }
2216       for (i=low; i<high; i++) {
2217         if (rp[i] > bcol) break;
2218         if (rp[i] == bcol) {
2219           bap = ap +  bs2*i + bs*cidx + ridx;
2220           if (!A->structure_only) {
2221             if (is == ADD_VALUES) *bap += value;
2222             else                  *bap  = value;
2223           }
2224           goto noinsert1;
2225         }
2226       }
2227       if (nonew == 1) goto noinsert1;
2228       PetscCheck(nonew != -1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%" PetscInt_FMT ", %" PetscInt_FMT ") in the matrix", row, col);
2229       if (A->structure_only) {
2230         MatSeqXAIJReallocateAIJ_structure_only(A,a->mbs,bs2,nrow,brow,bcol,rmax,ai,aj,rp,imax,nonew,MatScalar);
2231       } else {
2232         MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,brow,bcol,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
2233       }
2234       N = nrow++ - 1; high++;
2235       /* shift up all the later entries in this row */
2236       PetscCall(PetscArraymove(rp+i+1,rp+i,N-i+1));
2237       rp[i] = bcol;
2238       if (!A->structure_only) {
2239         PetscCall(PetscArraymove(ap+bs2*(i+1),ap+bs2*i,bs2*(N-i+1)));
2240         PetscCall(PetscArrayzero(ap+bs2*i,bs2));
2241         ap[bs2*i + bs*cidx + ridx] = value;
2242       }
2243       a->nz++;
2244       A->nonzerostate++;
2245 noinsert1:;
2246       low = i;
2247     }
2248     ailen[brow] = nrow;
2249   }
2250   PetscFunctionReturn(0);
2251 }
2252 
2253 PetscErrorCode MatILUFactor_SeqBAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info)
2254 {
2255   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)inA->data;
2256   Mat            outA;
2257   PetscBool      row_identity,col_identity;
2258 
2259   PetscFunctionBegin;
2260   PetscCheck(info->levels == 0,PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels = 0 supported for in-place ILU");
2261   PetscCall(ISIdentity(row,&row_identity));
2262   PetscCall(ISIdentity(col,&col_identity));
2263   PetscCheck(row_identity && col_identity,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Row and column permutations must be identity for in-place ILU");
2264 
2265   outA            = inA;
2266   inA->factortype = MAT_FACTOR_LU;
2267   PetscCall(PetscFree(inA->solvertype));
2268   PetscCall(PetscStrallocpy(MATSOLVERPETSC,&inA->solvertype));
2269 
2270   PetscCall(MatMarkDiagonal_SeqBAIJ(inA));
2271 
2272   PetscCall(PetscObjectReference((PetscObject)row));
2273   PetscCall(ISDestroy(&a->row));
2274   a->row = row;
2275   PetscCall(PetscObjectReference((PetscObject)col));
2276   PetscCall(ISDestroy(&a->col));
2277   a->col = col;
2278 
2279   /* Create the invert permutation so that it can be used in MatLUFactorNumeric() */
2280   PetscCall(ISDestroy(&a->icol));
2281   PetscCall(ISInvertPermutation(col,PETSC_DECIDE,&a->icol));
2282   PetscCall(PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol));
2283 
2284   PetscCall(MatSeqBAIJSetNumericFactorization_inplace(inA,(PetscBool)(row_identity && col_identity)));
2285   if (!a->solve_work) {
2286     PetscCall(PetscMalloc1(inA->rmap->N+inA->rmap->bs,&a->solve_work));
2287     PetscCall(PetscLogObjectMemory((PetscObject)inA,(inA->rmap->N+inA->rmap->bs)*sizeof(PetscScalar)));
2288   }
2289   PetscCall(MatLUFactorNumeric(outA,inA,info));
2290   PetscFunctionReturn(0);
2291 }
2292 
2293 PetscErrorCode  MatSeqBAIJSetColumnIndices_SeqBAIJ(Mat mat,PetscInt *indices)
2294 {
2295   Mat_SeqBAIJ *baij = (Mat_SeqBAIJ*)mat->data;
2296   PetscInt    i,nz,mbs;
2297 
2298   PetscFunctionBegin;
2299   nz  = baij->maxnz;
2300   mbs = baij->mbs;
2301   for (i=0; i<nz; i++) {
2302     baij->j[i] = indices[i];
2303   }
2304   baij->nz = nz;
2305   for (i=0; i<mbs; i++) {
2306     baij->ilen[i] = baij->imax[i];
2307   }
2308   PetscFunctionReturn(0);
2309 }
2310 
2311 /*@
2312     MatSeqBAIJSetColumnIndices - Set the column indices for all the rows
2313        in the matrix.
2314 
2315   Input Parameters:
2316 +  mat - the SeqBAIJ matrix
2317 -  indices - the column indices
2318 
2319   Level: advanced
2320 
2321   Notes:
2322     This can be called if you have precomputed the nonzero structure of the
2323   matrix and want to provide it to the matrix object to improve the performance
2324   of the MatSetValues() operation.
2325 
2326     You MUST have set the correct numbers of nonzeros per row in the call to
2327   MatCreateSeqBAIJ(), and the columns indices MUST be sorted.
2328 
2329     MUST be called before any calls to MatSetValues();
2330 
2331 @*/
2332 PetscErrorCode  MatSeqBAIJSetColumnIndices(Mat mat,PetscInt *indices)
2333 {
2334   PetscFunctionBegin;
2335   PetscValidHeaderSpecific(mat,MAT_CLASSID,1);
2336   PetscValidIntPointer(indices,2);
2337   PetscUseMethod(mat,"MatSeqBAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));
2338   PetscFunctionReturn(0);
2339 }
2340 
2341 PetscErrorCode MatGetRowMaxAbs_SeqBAIJ(Mat A,Vec v,PetscInt idx[])
2342 {
2343   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
2344   PetscInt       i,j,n,row,bs,*ai,*aj,mbs;
2345   PetscReal      atmp;
2346   PetscScalar    *x,zero = 0.0;
2347   MatScalar      *aa;
2348   PetscInt       ncols,brow,krow,kcol;
2349 
2350   PetscFunctionBegin;
2351   /* why is this not a macro???????????????????????????????????????????????????????????????? */
2352   PetscCheck(!A->factortype,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2353   bs  = A->rmap->bs;
2354   aa  = a->a;
2355   ai  = a->i;
2356   aj  = a->j;
2357   mbs = a->mbs;
2358 
2359   PetscCall(VecSet(v,zero));
2360   PetscCall(VecGetArray(v,&x));
2361   PetscCall(VecGetLocalSize(v,&n));
2362   PetscCheck(n == A->rmap->N,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
2363   for (i=0; i<mbs; i++) {
2364     ncols = ai[1] - ai[0]; ai++;
2365     brow  = bs*i;
2366     for (j=0; j<ncols; j++) {
2367       for (kcol=0; kcol<bs; kcol++) {
2368         for (krow=0; krow<bs; krow++) {
2369           atmp = PetscAbsScalar(*aa);aa++;
2370           row  = brow + krow;   /* row index */
2371           if (PetscAbsScalar(x[row]) < atmp) {x[row] = atmp; if (idx) idx[row] = bs*(*aj) + kcol;}
2372         }
2373       }
2374       aj++;
2375     }
2376   }
2377   PetscCall(VecRestoreArray(v,&x));
2378   PetscFunctionReturn(0);
2379 }
2380 
2381 PetscErrorCode MatCopy_SeqBAIJ(Mat A,Mat B,MatStructure str)
2382 {
2383   PetscFunctionBegin;
2384   /* If the two matrices have the same copy implementation, use fast copy. */
2385   if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
2386     Mat_SeqBAIJ *a  = (Mat_SeqBAIJ*)A->data;
2387     Mat_SeqBAIJ *b  = (Mat_SeqBAIJ*)B->data;
2388     PetscInt    ambs=a->mbs,bmbs=b->mbs,abs=A->rmap->bs,bbs=B->rmap->bs,bs2=abs*abs;
2389 
2390     PetscCheck(a->i[ambs] == b->i[bmbs],PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzero blocks in matrices A %" PetscInt_FMT " and B %" PetscInt_FMT " are different",a->i[ambs],b->i[bmbs]);
2391     PetscCheck(abs == bbs,PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Block size A %" PetscInt_FMT " and B %" PetscInt_FMT " are different",abs,bbs);
2392     PetscCall(PetscArraycpy(b->a,a->a,bs2*a->i[ambs]));
2393     PetscCall(PetscObjectStateIncrease((PetscObject)B));
2394   } else {
2395     PetscCall(MatCopy_Basic(A,B,str));
2396   }
2397   PetscFunctionReturn(0);
2398 }
2399 
2400 PetscErrorCode MatSetUp_SeqBAIJ(Mat A)
2401 {
2402   PetscFunctionBegin;
2403   PetscCall(MatSeqBAIJSetPreallocation(A,A->rmap->bs,PETSC_DEFAULT,NULL));
2404   PetscFunctionReturn(0);
2405 }
2406 
2407 static PetscErrorCode MatSeqBAIJGetArray_SeqBAIJ(Mat A,PetscScalar *array[])
2408 {
2409   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2410 
2411   PetscFunctionBegin;
2412   *array = a->a;
2413   PetscFunctionReturn(0);
2414 }
2415 
2416 static PetscErrorCode MatSeqBAIJRestoreArray_SeqBAIJ(Mat A,PetscScalar *array[])
2417 {
2418   PetscFunctionBegin;
2419   *array = NULL;
2420   PetscFunctionReturn(0);
2421 }
2422 
2423 PetscErrorCode MatAXPYGetPreallocation_SeqBAIJ(Mat Y,Mat X,PetscInt *nnz)
2424 {
2425   PetscInt       bs = Y->rmap->bs,mbs = Y->rmap->N/bs;
2426   Mat_SeqBAIJ    *x = (Mat_SeqBAIJ*)X->data;
2427   Mat_SeqBAIJ    *y = (Mat_SeqBAIJ*)Y->data;
2428 
2429   PetscFunctionBegin;
2430   /* Set the number of nonzeros in the new matrix */
2431   PetscCall(MatAXPYGetPreallocation_SeqX_private(mbs,x->i,x->j,y->i,y->j,nnz));
2432   PetscFunctionReturn(0);
2433 }
2434 
2435 PetscErrorCode MatAXPY_SeqBAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
2436 {
2437   Mat_SeqBAIJ    *x = (Mat_SeqBAIJ*)X->data,*y = (Mat_SeqBAIJ*)Y->data;
2438   PetscInt       bs=Y->rmap->bs,bs2=bs*bs;
2439   PetscBLASInt   one=1;
2440 
2441   PetscFunctionBegin;
2442   if (str == UNKNOWN_NONZERO_PATTERN || (PetscDefined(USE_DEBUG) && str == SAME_NONZERO_PATTERN)) {
2443     PetscBool e = x->nz == y->nz && x->mbs == y->mbs && bs == X->rmap->bs ? PETSC_TRUE : PETSC_FALSE;
2444     if (e) {
2445       PetscCall(PetscArraycmp(x->i,y->i,x->mbs+1,&e));
2446       if (e) {
2447         PetscCall(PetscArraycmp(x->j,y->j,x->i[x->mbs],&e));
2448         if (e) str = SAME_NONZERO_PATTERN;
2449       }
2450     }
2451     if (!e) PetscCheck(str != SAME_NONZERO_PATTERN,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"MatStructure is not SAME_NONZERO_PATTERN");
2452   }
2453   if (str == SAME_NONZERO_PATTERN) {
2454     PetscScalar  alpha = a;
2455     PetscBLASInt bnz;
2456     PetscCall(PetscBLASIntCast(x->nz*bs2,&bnz));
2457     PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
2458     PetscCall(PetscObjectStateIncrease((PetscObject)Y));
2459   } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2460     PetscCall(MatAXPY_Basic(Y,a,X,str));
2461   } else {
2462     Mat      B;
2463     PetscInt *nnz;
2464     PetscCheck(bs == X->rmap->bs,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrices must have same block size");
2465     PetscCall(PetscMalloc1(Y->rmap->N,&nnz));
2466     PetscCall(MatCreate(PetscObjectComm((PetscObject)Y),&B));
2467     PetscCall(PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name));
2468     PetscCall(MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N));
2469     PetscCall(MatSetBlockSizesFromMats(B,Y,Y));
2470     PetscCall(MatSetType(B,(MatType) ((PetscObject)Y)->type_name));
2471     PetscCall(MatAXPYGetPreallocation_SeqBAIJ(Y,X,nnz));
2472     PetscCall(MatSeqBAIJSetPreallocation(B,bs,0,nnz));
2473     PetscCall(MatAXPY_BasicWithPreallocation(B,Y,a,X,str));
2474     PetscCall(MatHeaderMerge(Y,&B));
2475     PetscCall(PetscFree(nnz));
2476   }
2477   PetscFunctionReturn(0);
2478 }
2479 
2480 PETSC_INTERN PetscErrorCode MatConjugate_SeqBAIJ(Mat A)
2481 {
2482 #if defined(PETSC_USE_COMPLEX)
2483   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2484   PetscInt    i,nz = a->bs2*a->i[a->mbs];
2485   MatScalar   *aa = a->a;
2486 
2487   PetscFunctionBegin;
2488   for (i=0; i<nz; i++) aa[i] = PetscConj(aa[i]);
2489 #else
2490   PetscFunctionBegin;
2491 #endif
2492   PetscFunctionReturn(0);
2493 }
2494 
2495 PetscErrorCode MatRealPart_SeqBAIJ(Mat A)
2496 {
2497   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2498   PetscInt    i,nz = a->bs2*a->i[a->mbs];
2499   MatScalar   *aa = a->a;
2500 
2501   PetscFunctionBegin;
2502   for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
2503   PetscFunctionReturn(0);
2504 }
2505 
2506 PetscErrorCode MatImaginaryPart_SeqBAIJ(Mat A)
2507 {
2508   Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2509   PetscInt    i,nz = a->bs2*a->i[a->mbs];
2510   MatScalar   *aa = a->a;
2511 
2512   PetscFunctionBegin;
2513   for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
2514   PetscFunctionReturn(0);
2515 }
2516 
2517 /*
2518     Code almost identical to MatGetColumnIJ_SeqAIJ() should share common code
2519 */
2520 PetscErrorCode MatGetColumnIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool  *done)
2521 {
2522   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
2523   PetscInt       bs = A->rmap->bs,i,*collengths,*cia,*cja,n = A->cmap->n/bs,m = A->rmap->n/bs;
2524   PetscInt       nz = a->i[m],row,*jj,mr,col;
2525 
2526   PetscFunctionBegin;
2527   *nn = n;
2528   if (!ia) PetscFunctionReturn(0);
2529   PetscCheck(!symmetric,PETSC_COMM_SELF,PETSC_ERR_SUP,"Not for BAIJ matrices");
2530   else {
2531     PetscCall(PetscCalloc1(n,&collengths));
2532     PetscCall(PetscMalloc1(n+1,&cia));
2533     PetscCall(PetscMalloc1(nz,&cja));
2534     jj   = a->j;
2535     for (i=0; i<nz; i++) {
2536       collengths[jj[i]]++;
2537     }
2538     cia[0] = oshift;
2539     for (i=0; i<n; i++) {
2540       cia[i+1] = cia[i] + collengths[i];
2541     }
2542     PetscCall(PetscArrayzero(collengths,n));
2543     jj   = a->j;
2544     for (row=0; row<m; row++) {
2545       mr = a->i[row+1] - a->i[row];
2546       for (i=0; i<mr; i++) {
2547         col = *jj++;
2548 
2549         cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
2550       }
2551     }
2552     PetscCall(PetscFree(collengths));
2553     *ia  = cia; *ja = cja;
2554   }
2555   PetscFunctionReturn(0);
2556 }
2557 
2558 PetscErrorCode MatRestoreColumnIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool  *done)
2559 {
2560   PetscFunctionBegin;
2561   if (!ia) PetscFunctionReturn(0);
2562   PetscCall(PetscFree(*ia));
2563   PetscCall(PetscFree(*ja));
2564   PetscFunctionReturn(0);
2565 }
2566 
2567 /*
2568  MatGetColumnIJ_SeqBAIJ_Color() and MatRestoreColumnIJ_SeqBAIJ_Color() are customized from
2569  MatGetColumnIJ_SeqBAIJ() and MatRestoreColumnIJ_SeqBAIJ() by adding an output
2570  spidx[], index of a->a, to be used in MatTransposeColoringCreate() and MatFDColoringCreate()
2571  */
2572 PetscErrorCode MatGetColumnIJ_SeqBAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool  *done)
2573 {
2574   Mat_SeqBAIJ    *a = (Mat_SeqBAIJ*)A->data;
2575   PetscInt       i,*collengths,*cia,*cja,n=a->nbs,m=a->mbs;
2576   PetscInt       nz = a->i[m],row,*jj,mr,col;
2577   PetscInt       *cspidx;
2578 
2579   PetscFunctionBegin;
2580   *nn = n;
2581   if (!ia) PetscFunctionReturn(0);
2582 
2583   PetscCall(PetscCalloc1(n,&collengths));
2584   PetscCall(PetscMalloc1(n+1,&cia));
2585   PetscCall(PetscMalloc1(nz,&cja));
2586   PetscCall(PetscMalloc1(nz,&cspidx));
2587   jj   = a->j;
2588   for (i=0; i<nz; i++) {
2589     collengths[jj[i]]++;
2590   }
2591   cia[0] = oshift;
2592   for (i=0; i<n; i++) {
2593     cia[i+1] = cia[i] + collengths[i];
2594   }
2595   PetscCall(PetscArrayzero(collengths,n));
2596   jj   = a->j;
2597   for (row=0; row<m; row++) {
2598     mr = a->i[row+1] - a->i[row];
2599     for (i=0; i<mr; i++) {
2600       col = *jj++;
2601       cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */
2602       cja[cia[col] + collengths[col]++ - oshift]  = row + oshift;
2603     }
2604   }
2605   PetscCall(PetscFree(collengths));
2606   *ia    = cia;
2607   *ja    = cja;
2608   *spidx = cspidx;
2609   PetscFunctionReturn(0);
2610 }
2611 
2612 PetscErrorCode MatRestoreColumnIJ_SeqBAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool  *done)
2613 {
2614   PetscFunctionBegin;
2615   PetscCall(MatRestoreColumnIJ_SeqBAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done));
2616   PetscCall(PetscFree(*spidx));
2617   PetscFunctionReturn(0);
2618 }
2619 
2620 PetscErrorCode MatShift_SeqBAIJ(Mat Y,PetscScalar a)
2621 {
2622   Mat_SeqBAIJ     *aij = (Mat_SeqBAIJ*)Y->data;
2623 
2624   PetscFunctionBegin;
2625   if (!Y->preallocated || !aij->nz) {
2626     PetscCall(MatSeqBAIJSetPreallocation(Y,Y->rmap->bs,1,NULL));
2627   }
2628   PetscCall(MatShift_Basic(Y,a));
2629   PetscFunctionReturn(0);
2630 }
2631 
2632 /* -------------------------------------------------------------------*/
2633 static struct _MatOps MatOps_Values = {MatSetValues_SeqBAIJ,
2634                                        MatGetRow_SeqBAIJ,
2635                                        MatRestoreRow_SeqBAIJ,
2636                                        MatMult_SeqBAIJ_N,
2637                                /* 4*/  MatMultAdd_SeqBAIJ_N,
2638                                        MatMultTranspose_SeqBAIJ,
2639                                        MatMultTransposeAdd_SeqBAIJ,
2640                                        NULL,
2641                                        NULL,
2642                                        NULL,
2643                                /* 10*/ NULL,
2644                                        MatLUFactor_SeqBAIJ,
2645                                        NULL,
2646                                        NULL,
2647                                        MatTranspose_SeqBAIJ,
2648                                /* 15*/ MatGetInfo_SeqBAIJ,
2649                                        MatEqual_SeqBAIJ,
2650                                        MatGetDiagonal_SeqBAIJ,
2651                                        MatDiagonalScale_SeqBAIJ,
2652                                        MatNorm_SeqBAIJ,
2653                                /* 20*/ NULL,
2654                                        MatAssemblyEnd_SeqBAIJ,
2655                                        MatSetOption_SeqBAIJ,
2656                                        MatZeroEntries_SeqBAIJ,
2657                                /* 24*/ MatZeroRows_SeqBAIJ,
2658                                        NULL,
2659                                        NULL,
2660                                        NULL,
2661                                        NULL,
2662                                /* 29*/ MatSetUp_SeqBAIJ,
2663                                        NULL,
2664                                        NULL,
2665                                        NULL,
2666                                        NULL,
2667                                /* 34*/ MatDuplicate_SeqBAIJ,
2668                                        NULL,
2669                                        NULL,
2670                                        MatILUFactor_SeqBAIJ,
2671                                        NULL,
2672                                /* 39*/ MatAXPY_SeqBAIJ,
2673                                        MatCreateSubMatrices_SeqBAIJ,
2674                                        MatIncreaseOverlap_SeqBAIJ,
2675                                        MatGetValues_SeqBAIJ,
2676                                        MatCopy_SeqBAIJ,
2677                                /* 44*/ NULL,
2678                                        MatScale_SeqBAIJ,
2679                                        MatShift_SeqBAIJ,
2680                                        NULL,
2681                                        MatZeroRowsColumns_SeqBAIJ,
2682                                /* 49*/ NULL,
2683                                        MatGetRowIJ_SeqBAIJ,
2684                                        MatRestoreRowIJ_SeqBAIJ,
2685                                        MatGetColumnIJ_SeqBAIJ,
2686                                        MatRestoreColumnIJ_SeqBAIJ,
2687                                /* 54*/ MatFDColoringCreate_SeqXAIJ,
2688                                        NULL,
2689                                        NULL,
2690                                        NULL,
2691                                        MatSetValuesBlocked_SeqBAIJ,
2692                                /* 59*/ MatCreateSubMatrix_SeqBAIJ,
2693                                        MatDestroy_SeqBAIJ,
2694                                        MatView_SeqBAIJ,
2695                                        NULL,
2696                                        NULL,
2697                                /* 64*/ NULL,
2698                                        NULL,
2699                                        NULL,
2700                                        NULL,
2701                                        NULL,
2702                                /* 69*/ MatGetRowMaxAbs_SeqBAIJ,
2703                                        NULL,
2704                                        MatConvert_Basic,
2705                                        NULL,
2706                                        NULL,
2707                                /* 74*/ NULL,
2708                                        MatFDColoringApply_BAIJ,
2709                                        NULL,
2710                                        NULL,
2711                                        NULL,
2712                                /* 79*/ NULL,
2713                                        NULL,
2714                                        NULL,
2715                                        NULL,
2716                                        MatLoad_SeqBAIJ,
2717                                /* 84*/ NULL,
2718                                        NULL,
2719                                        NULL,
2720                                        NULL,
2721                                        NULL,
2722                                /* 89*/ NULL,
2723                                        NULL,
2724                                        NULL,
2725                                        NULL,
2726                                        NULL,
2727                                /* 94*/ NULL,
2728                                        NULL,
2729                                        NULL,
2730                                        NULL,
2731                                        NULL,
2732                                /* 99*/ NULL,
2733                                        NULL,
2734                                        NULL,
2735                                        MatConjugate_SeqBAIJ,
2736                                        NULL,
2737                                /*104*/ NULL,
2738                                        MatRealPart_SeqBAIJ,
2739                                        MatImaginaryPart_SeqBAIJ,
2740                                        NULL,
2741                                        NULL,
2742                                /*109*/ NULL,
2743                                        NULL,
2744                                        NULL,
2745                                        NULL,
2746                                        MatMissingDiagonal_SeqBAIJ,
2747                                /*114*/ NULL,
2748                                        NULL,
2749                                        NULL,
2750                                        NULL,
2751                                        NULL,
2752                                /*119*/ NULL,
2753                                        NULL,
2754                                        MatMultHermitianTranspose_SeqBAIJ,
2755                                        MatMultHermitianTransposeAdd_SeqBAIJ,
2756                                        NULL,
2757                                /*124*/ NULL,
2758                                        MatGetColumnReductions_SeqBAIJ,
2759                                        MatInvertBlockDiagonal_SeqBAIJ,
2760                                        NULL,
2761                                        NULL,
2762                                /*129*/ NULL,
2763                                        NULL,
2764                                        NULL,
2765                                        NULL,
2766                                        NULL,
2767                                /*134*/ NULL,
2768                                        NULL,
2769                                        NULL,
2770                                        NULL,
2771                                        NULL,
2772                                /*139*/ MatSetBlockSizes_Default,
2773                                        NULL,
2774                                        NULL,
2775                                        MatFDColoringSetUp_SeqXAIJ,
2776                                        NULL,
2777                                 /*144*/MatCreateMPIMatConcatenateSeqMat_SeqBAIJ,
2778                                        MatDestroySubMatrices_SeqBAIJ,
2779                                        NULL,
2780                                        NULL
2781 };
2782 
2783 PetscErrorCode  MatStoreValues_SeqBAIJ(Mat mat)
2784 {
2785   Mat_SeqBAIJ    *aij = (Mat_SeqBAIJ*)mat->data;
2786   PetscInt       nz   = aij->i[aij->mbs]*aij->bs2;
2787 
2788   PetscFunctionBegin;
2789   PetscCheck(aij->nonew == 1,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
2790 
2791   /* allocate space for values if not already there */
2792   if (!aij->saved_values) {
2793     PetscCall(PetscMalloc1(nz+1,&aij->saved_values));
2794     PetscCall(PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar)));
2795   }
2796 
2797   /* copy values over */
2798   PetscCall(PetscArraycpy(aij->saved_values,aij->a,nz));
2799   PetscFunctionReturn(0);
2800 }
2801 
2802 PetscErrorCode  MatRetrieveValues_SeqBAIJ(Mat mat)
2803 {
2804   Mat_SeqBAIJ    *aij = (Mat_SeqBAIJ*)mat->data;
2805   PetscInt       nz = aij->i[aij->mbs]*aij->bs2;
2806 
2807   PetscFunctionBegin;
2808   PetscCheck(aij->nonew == 1,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
2809   PetscCheck(aij->saved_values,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
2810 
2811   /* copy values over */
2812   PetscCall(PetscArraycpy(aij->a,aij->saved_values,nz));
2813   PetscFunctionReturn(0);
2814 }
2815 
2816 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqAIJ(Mat, MatType,MatReuse,Mat*);
2817 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqSBAIJ(Mat, MatType,MatReuse,Mat*);
2818 
2819 PetscErrorCode  MatSeqBAIJSetPreallocation_SeqBAIJ(Mat B,PetscInt bs,PetscInt nz,PetscInt *nnz)
2820 {
2821   Mat_SeqBAIJ    *b;
2822   PetscErrorCode ierr;
2823   PetscInt       i,mbs,nbs,bs2;
2824   PetscBool      flg = PETSC_FALSE,skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE;
2825 
2826   PetscFunctionBegin;
2827   if (nz >= 0 || nnz) realalloc = PETSC_TRUE;
2828   if (nz == MAT_SKIP_ALLOCATION) {
2829     skipallocation = PETSC_TRUE;
2830     nz             = 0;
2831   }
2832 
2833   PetscCall(MatSetBlockSize(B,PetscAbs(bs)));
2834   PetscCall(PetscLayoutSetUp(B->rmap));
2835   PetscCall(PetscLayoutSetUp(B->cmap));
2836   PetscCall(PetscLayoutGetBlockSize(B->rmap,&bs));
2837 
2838   B->preallocated = PETSC_TRUE;
2839 
2840   mbs = B->rmap->n/bs;
2841   nbs = B->cmap->n/bs;
2842   bs2 = bs*bs;
2843 
2844   PetscCheck(mbs*bs==B->rmap->n && nbs*bs==B->cmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number rows %" PetscInt_FMT ", cols %" PetscInt_FMT " must be divisible by blocksize %" PetscInt_FMT,B->rmap->N,B->cmap->n,bs);
2845 
2846   if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2847   PetscCheck(nz >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %" PetscInt_FMT,nz);
2848   if (nnz) {
2849     for (i=0; i<mbs; i++) {
2850       PetscCheck(nnz[i] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %" PetscInt_FMT " value %" PetscInt_FMT,i,nnz[i]);
2851       PetscCheck(nnz[i] <= nbs,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than block row length: local row %" PetscInt_FMT " value %" PetscInt_FMT " rowlength %" PetscInt_FMT,i,nnz[i],nbs);
2852     }
2853   }
2854 
2855   b    = (Mat_SeqBAIJ*)B->data;
2856   ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)B),NULL,"Optimize options for SEQBAIJ matrix 2 ","Mat");PetscCall(ierr);
2857   PetscCall(PetscOptionsBool("-mat_no_unroll","Do not optimize for block size (slow)",NULL,flg,&flg,NULL));
2858   ierr = PetscOptionsEnd();PetscCall(ierr);
2859 
2860   if (!flg) {
2861     switch (bs) {
2862     case 1:
2863       B->ops->mult    = MatMult_SeqBAIJ_1;
2864       B->ops->multadd = MatMultAdd_SeqBAIJ_1;
2865       break;
2866     case 2:
2867       B->ops->mult    = MatMult_SeqBAIJ_2;
2868       B->ops->multadd = MatMultAdd_SeqBAIJ_2;
2869       break;
2870     case 3:
2871       B->ops->mult    = MatMult_SeqBAIJ_3;
2872       B->ops->multadd = MatMultAdd_SeqBAIJ_3;
2873       break;
2874     case 4:
2875       B->ops->mult    = MatMult_SeqBAIJ_4;
2876       B->ops->multadd = MatMultAdd_SeqBAIJ_4;
2877       break;
2878     case 5:
2879       B->ops->mult    = MatMult_SeqBAIJ_5;
2880       B->ops->multadd = MatMultAdd_SeqBAIJ_5;
2881       break;
2882     case 6:
2883       B->ops->mult    = MatMult_SeqBAIJ_6;
2884       B->ops->multadd = MatMultAdd_SeqBAIJ_6;
2885       break;
2886     case 7:
2887       B->ops->mult    = MatMult_SeqBAIJ_7;
2888       B->ops->multadd = MatMultAdd_SeqBAIJ_7;
2889       break;
2890     case 9:
2891     {
2892       PetscInt version = 1;
2893       PetscCall(PetscOptionsGetInt(NULL,((PetscObject)B)->prefix,"-mat_baij_mult_version",&version,NULL));
2894       switch (version) {
2895 #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES)
2896       case 1:
2897         B->ops->mult    = MatMult_SeqBAIJ_9_AVX2;
2898         B->ops->multadd = MatMultAdd_SeqBAIJ_9_AVX2;
2899         PetscCall(PetscInfo((PetscObject)B,"Using AVX2 for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",bs));
2900         break;
2901 #endif
2902       default:
2903         B->ops->mult    = MatMult_SeqBAIJ_N;
2904         B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2905         PetscCall(PetscInfo((PetscObject)B,"Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",bs));
2906         break;
2907       }
2908       break;
2909     }
2910     case 11:
2911       B->ops->mult    = MatMult_SeqBAIJ_11;
2912       B->ops->multadd = MatMultAdd_SeqBAIJ_11;
2913       break;
2914     case 12:
2915     {
2916       PetscInt version = 1;
2917       PetscCall(PetscOptionsGetInt(NULL,((PetscObject)B)->prefix,"-mat_baij_mult_version",&version,NULL));
2918       switch (version) {
2919       case 1:
2920         B->ops->mult    = MatMult_SeqBAIJ_12_ver1;
2921         B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver1;
2922         PetscCall(PetscInfo((PetscObject)B,"Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",version,bs));
2923         break;
2924       case 2:
2925         B->ops->mult    = MatMult_SeqBAIJ_12_ver2;
2926         B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver2;
2927         PetscCall(PetscInfo((PetscObject)B,"Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",version,bs));
2928         break;
2929 #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES)
2930       case 3:
2931         B->ops->mult    = MatMult_SeqBAIJ_12_AVX2;
2932         B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver1;
2933         PetscCall(PetscInfo((PetscObject)B,"Using AVX2 for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",bs));
2934         break;
2935 #endif
2936       default:
2937         B->ops->mult    = MatMult_SeqBAIJ_N;
2938         B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2939         PetscCall(PetscInfo((PetscObject)B,"Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",bs));
2940         break;
2941       }
2942       break;
2943     }
2944     case 15:
2945     {
2946       PetscInt version = 1;
2947       PetscCall(PetscOptionsGetInt(NULL,((PetscObject)B)->prefix,"-mat_baij_mult_version",&version,NULL));
2948       switch (version) {
2949       case 1:
2950         B->ops->mult    = MatMult_SeqBAIJ_15_ver1;
2951         PetscCall(PetscInfo((PetscObject)B,"Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",version,bs));
2952         break;
2953       case 2:
2954         B->ops->mult    = MatMult_SeqBAIJ_15_ver2;
2955         PetscCall(PetscInfo((PetscObject)B,"Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",version,bs));
2956         break;
2957       case 3:
2958         B->ops->mult    = MatMult_SeqBAIJ_15_ver3;
2959         PetscCall(PetscInfo((PetscObject)B,"Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",version,bs));
2960         break;
2961       case 4:
2962         B->ops->mult    = MatMult_SeqBAIJ_15_ver4;
2963         PetscCall(PetscInfo((PetscObject)B,"Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",version,bs));
2964         break;
2965       default:
2966         B->ops->mult    = MatMult_SeqBAIJ_N;
2967         PetscCall(PetscInfo((PetscObject)B,"Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",bs));
2968         break;
2969       }
2970       B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2971       break;
2972     }
2973     default:
2974       B->ops->mult    = MatMult_SeqBAIJ_N;
2975       B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2976       PetscCall(PetscInfo((PetscObject)B,"Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n",bs));
2977       break;
2978     }
2979   }
2980   B->ops->sor = MatSOR_SeqBAIJ;
2981   b->mbs = mbs;
2982   b->nbs = nbs;
2983   if (!skipallocation) {
2984     if (!b->imax) {
2985       PetscCall(PetscMalloc2(mbs,&b->imax,mbs,&b->ilen));
2986       PetscCall(PetscLogObjectMemory((PetscObject)B,2*mbs*sizeof(PetscInt)));
2987 
2988       b->free_imax_ilen = PETSC_TRUE;
2989     }
2990     /* b->ilen will count nonzeros in each block row so far. */
2991     for (i=0; i<mbs; i++) b->ilen[i] = 0;
2992     if (!nnz) {
2993       if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2994       else if (nz < 0) nz = 1;
2995       nz = PetscMin(nz,nbs);
2996       for (i=0; i<mbs; i++) b->imax[i] = nz;
2997       PetscCall(PetscIntMultError(nz,mbs,&nz));
2998     } else {
2999       PetscInt64 nz64 = 0;
3000       for (i=0; i<mbs; i++) {b->imax[i] = nnz[i]; nz64 += nnz[i];}
3001       PetscCall(PetscIntCast(nz64,&nz));
3002     }
3003 
3004     /* allocate the matrix space */
3005     PetscCall(MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i));
3006     if (B->structure_only) {
3007       PetscCall(PetscMalloc1(nz,&b->j));
3008       PetscCall(PetscMalloc1(B->rmap->N+1,&b->i));
3009       PetscCall(PetscLogObjectMemory((PetscObject)B,(B->rmap->N+1)*sizeof(PetscInt)+nz*sizeof(PetscInt)));
3010     } else {
3011       PetscInt nzbs2 = 0;
3012       PetscCall(PetscIntMultError(nz,bs2,&nzbs2));
3013       PetscCall(PetscMalloc3(nzbs2,&b->a,nz,&b->j,B->rmap->N+1,&b->i));
3014       PetscCall(PetscLogObjectMemory((PetscObject)B,(B->rmap->N+1)*sizeof(PetscInt)+nz*(bs2*sizeof(PetscScalar)+sizeof(PetscInt))));
3015       PetscCall(PetscArrayzero(b->a,nz*bs2));
3016     }
3017     PetscCall(PetscArrayzero(b->j,nz));
3018 
3019     if (B->structure_only) {
3020       b->singlemalloc = PETSC_FALSE;
3021       b->free_a       = PETSC_FALSE;
3022     } else {
3023       b->singlemalloc = PETSC_TRUE;
3024       b->free_a       = PETSC_TRUE;
3025     }
3026     b->free_ij = PETSC_TRUE;
3027 
3028     b->i[0] = 0;
3029     for (i=1; i<mbs+1; i++) {
3030       b->i[i] = b->i[i-1] + b->imax[i-1];
3031     }
3032 
3033   } else {
3034     b->free_a  = PETSC_FALSE;
3035     b->free_ij = PETSC_FALSE;
3036   }
3037 
3038   b->bs2              = bs2;
3039   b->mbs              = mbs;
3040   b->nz               = 0;
3041   b->maxnz            = nz;
3042   B->info.nz_unneeded = (PetscReal)b->maxnz*bs2;
3043   B->was_assembled    = PETSC_FALSE;
3044   B->assembled        = PETSC_FALSE;
3045   if (realalloc) PetscCall(MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE));
3046   PetscFunctionReturn(0);
3047 }
3048 
3049 PetscErrorCode MatSeqBAIJSetPreallocationCSR_SeqBAIJ(Mat B,PetscInt bs,const PetscInt ii[],const PetscInt jj[],const PetscScalar V[])
3050 {
3051   PetscInt       i,m,nz,nz_max=0,*nnz;
3052   PetscScalar    *values=NULL;
3053   PetscBool      roworiented = ((Mat_SeqBAIJ*)B->data)->roworiented;
3054 
3055   PetscFunctionBegin;
3056   PetscCheck(bs >= 1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Invalid block size specified, must be positive but it is %" PetscInt_FMT,bs);
3057   PetscCall(PetscLayoutSetBlockSize(B->rmap,bs));
3058   PetscCall(PetscLayoutSetBlockSize(B->cmap,bs));
3059   PetscCall(PetscLayoutSetUp(B->rmap));
3060   PetscCall(PetscLayoutSetUp(B->cmap));
3061   PetscCall(PetscLayoutGetBlockSize(B->rmap,&bs));
3062   m    = B->rmap->n/bs;
3063 
3064   PetscCheck(ii[0] == 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "ii[0] must be 0 but it is %" PetscInt_FMT,ii[0]);
3065   PetscCall(PetscMalloc1(m+1, &nnz));
3066   for (i=0; i<m; i++) {
3067     nz = ii[i+1]- ii[i];
3068     PetscCheck(nz >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %" PetscInt_FMT " has a negative number of columns %" PetscInt_FMT,i,nz);
3069     nz_max = PetscMax(nz_max, nz);
3070     nnz[i] = nz;
3071   }
3072   PetscCall(MatSeqBAIJSetPreallocation(B,bs,0,nnz));
3073   PetscCall(PetscFree(nnz));
3074 
3075   values = (PetscScalar*)V;
3076   if (!values) {
3077     PetscCall(PetscCalloc1(bs*bs*(nz_max+1),&values));
3078   }
3079   for (i=0; i<m; i++) {
3080     PetscInt          ncols  = ii[i+1] - ii[i];
3081     const PetscInt    *icols = jj + ii[i];
3082     if (bs == 1 || !roworiented) {
3083       const PetscScalar *svals = values + (V ? (bs*bs*ii[i]) : 0);
3084       PetscCall(MatSetValuesBlocked_SeqBAIJ(B,1,&i,ncols,icols,svals,INSERT_VALUES));
3085     } else {
3086       PetscInt j;
3087       for (j=0; j<ncols; j++) {
3088         const PetscScalar *svals = values + (V ? (bs*bs*(ii[i]+j)) : 0);
3089         PetscCall(MatSetValuesBlocked_SeqBAIJ(B,1,&i,1,&icols[j],svals,INSERT_VALUES));
3090       }
3091     }
3092   }
3093   if (!V) PetscCall(PetscFree(values));
3094   PetscCall(MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY));
3095   PetscCall(MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY));
3096   PetscCall(MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE));
3097   PetscFunctionReturn(0);
3098 }
3099 
3100 /*@C
3101    MatSeqBAIJGetArray - gives access to the array where the data for a MATSEQBAIJ matrix is stored
3102 
3103    Not Collective
3104 
3105    Input Parameter:
3106 .  mat - a MATSEQBAIJ matrix
3107 
3108    Output Parameter:
3109 .   array - pointer to the data
3110 
3111    Level: intermediate
3112 
3113 .seealso: MatSeqBAIJRestoreArray(), MatSeqAIJGetArray(), MatSeqAIJRestoreArray()
3114 @*/
3115 PetscErrorCode MatSeqBAIJGetArray(Mat A,PetscScalar **array)
3116 {
3117   PetscFunctionBegin;
3118   PetscUseMethod(A,"MatSeqBAIJGetArray_C",(Mat,PetscScalar**),(A,array));
3119   PetscFunctionReturn(0);
3120 }
3121 
3122 /*@C
3123    MatSeqBAIJRestoreArray - returns access to the array where the data for a MATSEQBAIJ matrix is stored obtained by MatSeqBAIJGetArray()
3124 
3125    Not Collective
3126 
3127    Input Parameters:
3128 +  mat - a MATSEQBAIJ matrix
3129 -  array - pointer to the data
3130 
3131    Level: intermediate
3132 
3133 .seealso: MatSeqBAIJGetArray(), MatSeqAIJGetArray(), MatSeqAIJRestoreArray()
3134 @*/
3135 PetscErrorCode MatSeqBAIJRestoreArray(Mat A,PetscScalar **array)
3136 {
3137   PetscFunctionBegin;
3138   PetscUseMethod(A,"MatSeqBAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));
3139   PetscFunctionReturn(0);
3140 }
3141 
3142 /*MC
3143    MATSEQBAIJ - MATSEQBAIJ = "seqbaij" - A matrix type to be used for sequential block sparse matrices, based on
3144    block sparse compressed row format.
3145 
3146    Options Database Keys:
3147 + -mat_type seqbaij - sets the matrix type to "seqbaij" during a call to MatSetFromOptions()
3148 - -mat_baij_mult_version version - indicate the version of the matrix-vector product to use (0 often indicates using BLAS)
3149 
3150    Level: beginner
3151 
3152    Notes:
3153     MatSetOptions(,MAT_STRUCTURE_ONLY,PETSC_TRUE) may be called for this matrix type. In this no
3154     space is allocated for the nonzero entries and any entries passed with MatSetValues() are ignored
3155 
3156    Run with -info to see what version of the matrix-vector product is being used
3157 
3158 .seealso: MatCreateSeqBAIJ()
3159 M*/
3160 
3161 PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBSTRM(Mat, MatType,MatReuse,Mat*);
3162 
3163 PETSC_EXTERN PetscErrorCode MatCreate_SeqBAIJ(Mat B)
3164 {
3165   PetscMPIInt    size;
3166   Mat_SeqBAIJ    *b;
3167 
3168   PetscFunctionBegin;
3169   PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)B),&size));
3170   PetscCheck(size == 1,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Comm must be of size 1");
3171 
3172   PetscCall(PetscNewLog(B,&b));
3173   B->data = (void*)b;
3174   PetscCall(PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps)));
3175 
3176   b->row          = NULL;
3177   b->col          = NULL;
3178   b->icol         = NULL;
3179   b->reallocs     = 0;
3180   b->saved_values = NULL;
3181 
3182   b->roworiented        = PETSC_TRUE;
3183   b->nonew              = 0;
3184   b->diag               = NULL;
3185   B->spptr              = NULL;
3186   B->info.nz_unneeded   = (PetscReal)b->maxnz*b->bs2;
3187   b->keepnonzeropattern = PETSC_FALSE;
3188 
3189   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJGetArray_C",MatSeqBAIJGetArray_SeqBAIJ));
3190   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJRestoreArray_C",MatSeqBAIJRestoreArray_SeqBAIJ));
3191   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqBAIJ));
3192   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqBAIJ));
3193   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJSetColumnIndices_C",MatSeqBAIJSetColumnIndices_SeqBAIJ));
3194   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_seqaij_C",MatConvert_SeqBAIJ_SeqAIJ));
3195   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_seqsbaij_C",MatConvert_SeqBAIJ_SeqSBAIJ));
3196   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJSetPreallocation_C",MatSeqBAIJSetPreallocation_SeqBAIJ));
3197   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJSetPreallocationCSR_C",MatSeqBAIJSetPreallocationCSR_SeqBAIJ));
3198   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqBAIJ));
3199 #if defined(PETSC_HAVE_HYPRE)
3200   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_hypre_C",MatConvert_AIJ_HYPRE));
3201 #endif
3202   PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_is_C",MatConvert_XAIJ_IS));
3203   PetscCall(PetscObjectChangeTypeName((PetscObject)B,MATSEQBAIJ));
3204   PetscFunctionReturn(0);
3205 }
3206 
3207 PetscErrorCode MatDuplicateNoCreate_SeqBAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace)
3208 {
3209   Mat_SeqBAIJ    *c = (Mat_SeqBAIJ*)C->data,*a = (Mat_SeqBAIJ*)A->data;
3210   PetscInt       i,mbs = a->mbs,nz = a->nz,bs2 = a->bs2;
3211 
3212   PetscFunctionBegin;
3213   PetscCheck(a->i[mbs] == nz,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Corrupt matrix");
3214 
3215   if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3216     c->imax           = a->imax;
3217     c->ilen           = a->ilen;
3218     c->free_imax_ilen = PETSC_FALSE;
3219   } else {
3220     PetscCall(PetscMalloc2(mbs,&c->imax,mbs,&c->ilen));
3221     PetscCall(PetscLogObjectMemory((PetscObject)C,2*mbs*sizeof(PetscInt)));
3222     for (i=0; i<mbs; i++) {
3223       c->imax[i] = a->imax[i];
3224       c->ilen[i] = a->ilen[i];
3225     }
3226     c->free_imax_ilen = PETSC_TRUE;
3227   }
3228 
3229   /* allocate the matrix space */
3230   if (mallocmatspace) {
3231     if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3232       PetscCall(PetscCalloc1(bs2*nz,&c->a));
3233       PetscCall(PetscLogObjectMemory((PetscObject)C,a->i[mbs]*bs2*sizeof(PetscScalar)));
3234 
3235       c->i            = a->i;
3236       c->j            = a->j;
3237       c->singlemalloc = PETSC_FALSE;
3238       c->free_a       = PETSC_TRUE;
3239       c->free_ij      = PETSC_FALSE;
3240       c->parent       = A;
3241       C->preallocated = PETSC_TRUE;
3242       C->assembled    = PETSC_TRUE;
3243 
3244       PetscCall(PetscObjectReference((PetscObject)A));
3245       PetscCall(MatSetOption(A,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE));
3246       PetscCall(MatSetOption(C,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE));
3247     } else {
3248       PetscCall(PetscMalloc3(bs2*nz,&c->a,nz,&c->j,mbs+1,&c->i));
3249       PetscCall(PetscLogObjectMemory((PetscObject)C,a->i[mbs]*(bs2*sizeof(PetscScalar)+sizeof(PetscInt))+(mbs+1)*sizeof(PetscInt)));
3250 
3251       c->singlemalloc = PETSC_TRUE;
3252       c->free_a       = PETSC_TRUE;
3253       c->free_ij      = PETSC_TRUE;
3254 
3255       PetscCall(PetscArraycpy(c->i,a->i,mbs+1));
3256       if (mbs > 0) {
3257         PetscCall(PetscArraycpy(c->j,a->j,nz));
3258         if (cpvalues == MAT_COPY_VALUES) {
3259           PetscCall(PetscArraycpy(c->a,a->a,bs2*nz));
3260         } else {
3261           PetscCall(PetscArrayzero(c->a,bs2*nz));
3262         }
3263       }
3264       C->preallocated = PETSC_TRUE;
3265       C->assembled    = PETSC_TRUE;
3266     }
3267   }
3268 
3269   c->roworiented = a->roworiented;
3270   c->nonew       = a->nonew;
3271 
3272   PetscCall(PetscLayoutReference(A->rmap,&C->rmap));
3273   PetscCall(PetscLayoutReference(A->cmap,&C->cmap));
3274 
3275   c->bs2         = a->bs2;
3276   c->mbs         = a->mbs;
3277   c->nbs         = a->nbs;
3278 
3279   if (a->diag) {
3280     if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3281       c->diag      = a->diag;
3282       c->free_diag = PETSC_FALSE;
3283     } else {
3284       PetscCall(PetscMalloc1(mbs+1,&c->diag));
3285       PetscCall(PetscLogObjectMemory((PetscObject)C,(mbs+1)*sizeof(PetscInt)));
3286       for (i=0; i<mbs; i++) c->diag[i] = a->diag[i];
3287       c->free_diag = PETSC_TRUE;
3288     }
3289   } else c->diag = NULL;
3290 
3291   c->nz         = a->nz;
3292   c->maxnz      = a->nz;         /* Since we allocate exactly the right amount */
3293   c->solve_work = NULL;
3294   c->mult_work  = NULL;
3295   c->sor_workt  = NULL;
3296   c->sor_work   = NULL;
3297 
3298   c->compressedrow.use   = a->compressedrow.use;
3299   c->compressedrow.nrows = a->compressedrow.nrows;
3300   if (a->compressedrow.use) {
3301     i    = a->compressedrow.nrows;
3302     PetscCall(PetscMalloc2(i+1,&c->compressedrow.i,i+1,&c->compressedrow.rindex));
3303     PetscCall(PetscLogObjectMemory((PetscObject)C,(2*i+1)*sizeof(PetscInt)));
3304     PetscCall(PetscArraycpy(c->compressedrow.i,a->compressedrow.i,i+1));
3305     PetscCall(PetscArraycpy(c->compressedrow.rindex,a->compressedrow.rindex,i));
3306   } else {
3307     c->compressedrow.use    = PETSC_FALSE;
3308     c->compressedrow.i      = NULL;
3309     c->compressedrow.rindex = NULL;
3310   }
3311   C->nonzerostate = A->nonzerostate;
3312 
3313   PetscCall(PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist));
3314   PetscFunctionReturn(0);
3315 }
3316 
3317 PetscErrorCode MatDuplicate_SeqBAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
3318 {
3319   PetscFunctionBegin;
3320   PetscCall(MatCreate(PetscObjectComm((PetscObject)A),B));
3321   PetscCall(MatSetSizes(*B,A->rmap->N,A->cmap->n,A->rmap->N,A->cmap->n));
3322   PetscCall(MatSetType(*B,MATSEQBAIJ));
3323   PetscCall(MatDuplicateNoCreate_SeqBAIJ(*B,A,cpvalues,PETSC_TRUE));
3324   PetscFunctionReturn(0);
3325 }
3326 
3327 /* Used for both SeqBAIJ and SeqSBAIJ matrices */
3328 PetscErrorCode MatLoad_SeqBAIJ_Binary(Mat mat,PetscViewer viewer)
3329 {
3330   PetscInt       header[4],M,N,nz,bs,m,n,mbs,nbs,rows,cols,sum,i,j,k;
3331   PetscInt       *rowidxs,*colidxs;
3332   PetscScalar    *matvals;
3333 
3334   PetscFunctionBegin;
3335   PetscCall(PetscViewerSetUp(viewer));
3336 
3337   /* read matrix header */
3338   PetscCall(PetscViewerBinaryRead(viewer,header,4,NULL,PETSC_INT));
3339   PetscCheck(header[0] == MAT_FILE_CLASSID,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Not a matrix object in file");
3340   M = header[1]; N = header[2]; nz = header[3];
3341   PetscCheck(M >= 0,PetscObjectComm((PetscObject)viewer),PETSC_ERR_FILE_UNEXPECTED,"Matrix row size (%" PetscInt_FMT ") in file is negative",M);
3342   PetscCheck(N >= 0,PetscObjectComm((PetscObject)viewer),PETSC_ERR_FILE_UNEXPECTED,"Matrix column size (%" PetscInt_FMT ") in file is negative",N);
3343   PetscCheck(nz >= 0,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk, cannot load as SeqBAIJ");
3344 
3345   /* set block sizes from the viewer's .info file */
3346   PetscCall(MatLoad_Binary_BlockSizes(mat,viewer));
3347   /* set local and global sizes if not set already */
3348   if (mat->rmap->n < 0) mat->rmap->n = M;
3349   if (mat->cmap->n < 0) mat->cmap->n = N;
3350   if (mat->rmap->N < 0) mat->rmap->N = M;
3351   if (mat->cmap->N < 0) mat->cmap->N = N;
3352   PetscCall(PetscLayoutSetUp(mat->rmap));
3353   PetscCall(PetscLayoutSetUp(mat->cmap));
3354 
3355   /* check if the matrix sizes are correct */
3356   PetscCall(MatGetSize(mat,&rows,&cols));
3357   PetscCheck(M == rows && N == cols,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different sizes (%" PetscInt_FMT ", %" PetscInt_FMT ") than the input matrix (%" PetscInt_FMT ", %" PetscInt_FMT ")",M,N,rows,cols);
3358   PetscCall(MatGetBlockSize(mat,&bs));
3359   PetscCall(MatGetLocalSize(mat,&m,&n));
3360   mbs = m/bs; nbs = n/bs;
3361 
3362   /* read in row lengths, column indices and nonzero values */
3363   PetscCall(PetscMalloc1(m+1,&rowidxs));
3364   PetscCall(PetscViewerBinaryRead(viewer,rowidxs+1,m,NULL,PETSC_INT));
3365   rowidxs[0] = 0; for (i=0; i<m; i++) rowidxs[i+1] += rowidxs[i];
3366   sum = rowidxs[m];
3367   PetscCheck(sum == nz,PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent matrix data in file: nonzeros = %" PetscInt_FMT ", sum-row-lengths = %" PetscInt_FMT,nz,sum);
3368 
3369   /* read in column indices and nonzero values */
3370   PetscCall(PetscMalloc2(rowidxs[m],&colidxs,nz,&matvals));
3371   PetscCall(PetscViewerBinaryRead(viewer,colidxs,rowidxs[m],NULL,PETSC_INT));
3372   PetscCall(PetscViewerBinaryRead(viewer,matvals,rowidxs[m],NULL,PETSC_SCALAR));
3373 
3374   { /* preallocate matrix storage */
3375     PetscBT   bt; /* helper bit set to count nonzeros */
3376     PetscInt  *nnz;
3377     PetscBool sbaij;
3378 
3379     PetscCall(PetscBTCreate(nbs,&bt));
3380     PetscCall(PetscCalloc1(mbs,&nnz));
3381     PetscCall(PetscObjectTypeCompare((PetscObject)mat,MATSEQSBAIJ,&sbaij));
3382     for (i=0; i<mbs; i++) {
3383       PetscCall(PetscBTMemzero(nbs,bt));
3384       for (k=0; k<bs; k++) {
3385         PetscInt row = bs*i + k;
3386         for (j=rowidxs[row]; j<rowidxs[row+1]; j++) {
3387           PetscInt col = colidxs[j];
3388           if (!sbaij || col >= row)
3389             if (!PetscBTLookupSet(bt,col/bs)) nnz[i]++;
3390         }
3391       }
3392     }
3393     PetscCall(PetscBTDestroy(&bt));
3394     PetscCall(MatSeqBAIJSetPreallocation(mat,bs,0,nnz));
3395     PetscCall(MatSeqSBAIJSetPreallocation(mat,bs,0,nnz));
3396     PetscCall(PetscFree(nnz));
3397   }
3398 
3399   /* store matrix values */
3400   for (i=0; i<m; i++) {
3401     PetscInt row = i, s = rowidxs[i], e = rowidxs[i+1];
3402     PetscCall((*mat->ops->setvalues)(mat,1,&row,e-s,colidxs+s,matvals+s,INSERT_VALUES));
3403   }
3404 
3405   PetscCall(PetscFree(rowidxs));
3406   PetscCall(PetscFree2(colidxs,matvals));
3407   PetscCall(MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY));
3408   PetscCall(MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY));
3409   PetscFunctionReturn(0);
3410 }
3411 
3412 PetscErrorCode MatLoad_SeqBAIJ(Mat mat,PetscViewer viewer)
3413 {
3414   PetscBool isbinary;
3415 
3416   PetscFunctionBegin;
3417   PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary));
3418   PetscCheck(isbinary,PetscObjectComm((PetscObject)viewer),PETSC_ERR_SUP,"Viewer type %s not yet supported for reading %s matrices",((PetscObject)viewer)->type_name,((PetscObject)mat)->type_name);
3419   PetscCall(MatLoad_SeqBAIJ_Binary(mat,viewer));
3420   PetscFunctionReturn(0);
3421 }
3422 
3423 /*@C
3424    MatCreateSeqBAIJ - Creates a sparse matrix in block AIJ (block
3425    compressed row) format.  For good matrix assembly performance the
3426    user should preallocate the matrix storage by setting the parameter nz
3427    (or the array nnz).  By setting these parameters accurately, performance
3428    during matrix assembly can be increased by more than a factor of 50.
3429 
3430    Collective
3431 
3432    Input Parameters:
3433 +  comm - MPI communicator, set to PETSC_COMM_SELF
3434 .  bs - size of block, the blocks are ALWAYS square. One can use MatSetBlockSizes() to set a different row and column blocksize but the row
3435           blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with MatCreateVecs()
3436 .  m - number of rows
3437 .  n - number of columns
3438 .  nz - number of nonzero blocks  per block row (same for all rows)
3439 -  nnz - array containing the number of nonzero blocks in the various block rows
3440          (possibly different for each block row) or NULL
3441 
3442    Output Parameter:
3443 .  A - the matrix
3444 
3445    It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
3446    MatXXXXSetPreallocation() paradigm instead of this routine directly.
3447    [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
3448 
3449    Options Database Keys:
3450 +   -mat_no_unroll - uses code that does not unroll the loops in the
3451                      block calculations (much slower)
3452 -    -mat_block_size - size of the blocks to use
3453 
3454    Level: intermediate
3455 
3456    Notes:
3457    The number of rows and columns must be divisible by blocksize.
3458 
3459    If the nnz parameter is given then the nz parameter is ignored
3460 
3461    A nonzero block is any block that as 1 or more nonzeros in it
3462 
3463    The block AIJ format is fully compatible with standard Fortran 77
3464    storage.  That is, the stored row and column indices can begin at
3465    either one (as in Fortran) or zero.  See the users' manual for details.
3466 
3467    Specify the preallocated storage with either nz or nnz (not both).
3468    Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3469    allocation.  See Users-Manual: ch_mat for details.
3470    matrices.
3471 
3472 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateBAIJ()
3473 @*/
3474 PetscErrorCode  MatCreateSeqBAIJ(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
3475 {
3476   PetscFunctionBegin;
3477   PetscCall(MatCreate(comm,A));
3478   PetscCall(MatSetSizes(*A,m,n,m,n));
3479   PetscCall(MatSetType(*A,MATSEQBAIJ));
3480   PetscCall(MatSeqBAIJSetPreallocation(*A,bs,nz,(PetscInt*)nnz));
3481   PetscFunctionReturn(0);
3482 }
3483 
3484 /*@C
3485    MatSeqBAIJSetPreallocation - Sets the block size and expected nonzeros
3486    per row in the matrix. For good matrix assembly performance the
3487    user should preallocate the matrix storage by setting the parameter nz
3488    (or the array nnz).  By setting these parameters accurately, performance
3489    during matrix assembly can be increased by more than a factor of 50.
3490 
3491    Collective
3492 
3493    Input Parameters:
3494 +  B - the matrix
3495 .  bs - size of block, the blocks are ALWAYS square. One can use MatSetBlockSizes() to set a different row and column blocksize but the row
3496           blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with MatCreateVecs()
3497 .  nz - number of block nonzeros per block row (same for all rows)
3498 -  nnz - array containing the number of block nonzeros in the various block rows
3499          (possibly different for each block row) or NULL
3500 
3501    Options Database Keys:
3502 +   -mat_no_unroll - uses code that does not unroll the loops in the
3503                      block calculations (much slower)
3504 -   -mat_block_size - size of the blocks to use
3505 
3506    Level: intermediate
3507 
3508    Notes:
3509    If the nnz parameter is given then the nz parameter is ignored
3510 
3511    You can call MatGetInfo() to get information on how effective the preallocation was;
3512    for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3513    You can also run with the option -info and look for messages with the string
3514    malloc in them to see if additional memory allocation was needed.
3515 
3516    The block AIJ format is fully compatible with standard Fortran 77
3517    storage.  That is, the stored row and column indices can begin at
3518    either one (as in Fortran) or zero.  See the users' manual for details.
3519 
3520    Specify the preallocated storage with either nz or nnz (not both).
3521    Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3522    allocation.  See Users-Manual: ch_mat for details.
3523 
3524 .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateBAIJ(), MatGetInfo()
3525 @*/
3526 PetscErrorCode  MatSeqBAIJSetPreallocation(Mat B,PetscInt bs,PetscInt nz,const PetscInt nnz[])
3527 {
3528   PetscFunctionBegin;
3529   PetscValidHeaderSpecific(B,MAT_CLASSID,1);
3530   PetscValidType(B,1);
3531   PetscValidLogicalCollectiveInt(B,bs,2);
3532   PetscTryMethod(B,"MatSeqBAIJSetPreallocation_C",(Mat,PetscInt,PetscInt,const PetscInt[]),(B,bs,nz,nnz));
3533   PetscFunctionReturn(0);
3534 }
3535 
3536 /*@C
3537    MatSeqBAIJSetPreallocationCSR - Creates a sparse parallel matrix in BAIJ format using the given nonzero structure and (optional) numerical values
3538 
3539    Collective
3540 
3541    Input Parameters:
3542 +  B - the matrix
3543 .  i - the indices into j for the start of each local row (starts with zero)
3544 .  j - the column indices for each local row (starts with zero) these must be sorted for each row
3545 -  v - optional values in the matrix
3546 
3547    Level: advanced
3548 
3549    Notes:
3550    The order of the entries in values is specified by the MatOption MAT_ROW_ORIENTED.  For example, C programs
3551    may want to use the default MAT_ROW_ORIENTED=PETSC_TRUE and use an array v[nnz][bs][bs] where the second index is
3552    over rows within a block and the last index is over columns within a block row.  Fortran programs will likely set
3553    MAT_ROW_ORIENTED=PETSC_FALSE and use a Fortran array v(bs,bs,nnz) in which the first index is over rows within a
3554    block column and the second index is over columns within a block.
3555 
3556    Though this routine has Preallocation() in the name it also sets the exact nonzero locations of the matrix entries and usually the numerical values as well
3557 
3558 .seealso: MatCreate(), MatCreateSeqBAIJ(), MatSetValues(), MatSeqBAIJSetPreallocation(), MATSEQBAIJ
3559 @*/
3560 PetscErrorCode  MatSeqBAIJSetPreallocationCSR(Mat B,PetscInt bs,const PetscInt i[],const PetscInt j[], const PetscScalar v[])
3561 {
3562   PetscFunctionBegin;
3563   PetscValidHeaderSpecific(B,MAT_CLASSID,1);
3564   PetscValidType(B,1);
3565   PetscValidLogicalCollectiveInt(B,bs,2);
3566   PetscTryMethod(B,"MatSeqBAIJSetPreallocationCSR_C",(Mat,PetscInt,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,bs,i,j,v));
3567   PetscFunctionReturn(0);
3568 }
3569 
3570 /*@
3571      MatCreateSeqBAIJWithArrays - Creates an sequential BAIJ matrix using matrix elements provided by the user.
3572 
3573      Collective
3574 
3575    Input Parameters:
3576 +  comm - must be an MPI communicator of size 1
3577 .  bs - size of block
3578 .  m - number of rows
3579 .  n - number of columns
3580 .  i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of elements in that row block row of the matrix
3581 .  j - column indices
3582 -  a - matrix values
3583 
3584    Output Parameter:
3585 .  mat - the matrix
3586 
3587    Level: advanced
3588 
3589    Notes:
3590        The i, j, and a arrays are not copied by this routine, the user must free these arrays
3591     once the matrix is destroyed
3592 
3593        You cannot set new nonzero locations into this matrix, that will generate an error.
3594 
3595        The i and j indices are 0 based
3596 
3597        When block size is greater than 1 the matrix values must be stored using the BAIJ storage format (see the BAIJ code to determine this).
3598 
3599       The order of the entries in values is the same as the block compressed sparse row storage format; that is, it is
3600       the same as a three dimensional array in Fortran values(bs,bs,nnz) that contains the first column of the first
3601       block, followed by the second column of the first block etc etc.  That is, the blocks are contiguous in memory
3602       with column-major ordering within blocks.
3603 
3604 .seealso: MatCreate(), MatCreateBAIJ(), MatCreateSeqBAIJ()
3605 
3606 @*/
3607 PetscErrorCode  MatCreateSeqBAIJWithArrays(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat)
3608 {
3609   PetscInt       ii;
3610   Mat_SeqBAIJ    *baij;
3611 
3612   PetscFunctionBegin;
3613   PetscCheck(bs == 1,PETSC_COMM_SELF,PETSC_ERR_SUP,"block size %" PetscInt_FMT " > 1 is not supported yet",bs);
3614   if (m > 0) PetscCheck(i[0] == 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
3615 
3616   PetscCall(MatCreate(comm,mat));
3617   PetscCall(MatSetSizes(*mat,m,n,m,n));
3618   PetscCall(MatSetType(*mat,MATSEQBAIJ));
3619   PetscCall(MatSeqBAIJSetPreallocation(*mat,bs,MAT_SKIP_ALLOCATION,NULL));
3620   baij = (Mat_SeqBAIJ*)(*mat)->data;
3621   PetscCall(PetscMalloc2(m,&baij->imax,m,&baij->ilen));
3622   PetscCall(PetscLogObjectMemory((PetscObject)*mat,2*m*sizeof(PetscInt)));
3623 
3624   baij->i = i;
3625   baij->j = j;
3626   baij->a = a;
3627 
3628   baij->singlemalloc = PETSC_FALSE;
3629   baij->nonew        = -1;             /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
3630   baij->free_a       = PETSC_FALSE;
3631   baij->free_ij      = PETSC_FALSE;
3632 
3633   for (ii=0; ii<m; ii++) {
3634     baij->ilen[ii] = baij->imax[ii] = i[ii+1] - i[ii];
3635     PetscCheck(i[ii+1] - i[ii] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row length in i (row indices) row = %" PetscInt_FMT " length = %" PetscInt_FMT,ii,i[ii+1] - i[ii]);
3636   }
3637   if (PetscDefined(USE_DEBUG)) {
3638     for (ii=0; ii<baij->i[m]; ii++) {
3639       PetscCheck(j[ii] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %" PetscInt_FMT " index = %" PetscInt_FMT,ii,j[ii]);
3640       PetscCheck(j[ii] <= n - 1,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column index to large at location = %" PetscInt_FMT " index = %" PetscInt_FMT,ii,j[ii]);
3641     }
3642   }
3643 
3644   PetscCall(MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY));
3645   PetscCall(MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY));
3646   PetscFunctionReturn(0);
3647 }
3648 
3649 PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqBAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat)
3650 {
3651   PetscMPIInt    size;
3652 
3653   PetscFunctionBegin;
3654   PetscCallMPI(MPI_Comm_size(comm,&size));
3655   if (size == 1 && scall == MAT_REUSE_MATRIX) {
3656     PetscCall(MatCopy(inmat,*outmat,SAME_NONZERO_PATTERN));
3657   } else {
3658     PetscCall(MatCreateMPIMatConcatenateSeqMat_MPIBAIJ(comm,inmat,n,scall,outmat));
3659   }
3660   PetscFunctionReturn(0);
3661 }
3662