xref: /petsc/src/mat/impls/aij/seq/fdaij.c (revision cf1aed2ce99d23e50336629af3ca8cf096900abb)
1 
2 #include <../src/mat/impls/aij/seq/aij.h>
3 #include <../src/mat/impls/baij/seq/baij.h>
4 
5 /*
6     This routine is shared by SeqAIJ and SeqBAIJ matrices,
7     since it operators only on the nonzero structure of the elements or blocks.
8 */
9 #undef __FUNCT__
10 #define __FUNCT__ "MatFDColoringCreate_SeqXAIJ"
11 PetscErrorCode MatFDColoringCreate_SeqXAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c)
12 {
13   PetscErrorCode ierr;
14   PetscInt       bs,nis=iscoloring->n,m=mat->rmap->n;
15   PetscBool      isBAIJ;
16 
17   PetscFunctionBegin;
18   /* set default brows and bcols for speedup inserting the dense matrix into sparse Jacobian */
19   ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr);
20   ierr = PetscObjectTypeCompare((PetscObject)mat,MATSEQBAIJ,&isBAIJ);CHKERRQ(ierr);
21   if (isBAIJ) {
22     c->brows = m;
23     c->bcols = 1;
24   } else { /* seqaij matrix */
25     /* bcols is chosen s.t. dy-array takes 50% of memory space as mat */
26     Mat_SeqAIJ *spA = (Mat_SeqAIJ*)mat->data;
27     PetscReal  mem;
28     PetscInt   nz,brows,bcols;
29 
30     bs    = 1; /* only bs=1 is supported for SeqAIJ matrix */
31 
32     nz    = spA->nz;
33     mem   = nz*(sizeof(PetscScalar) + sizeof(PetscInt)) + 3*m*sizeof(PetscInt);
34     bcols = (PetscInt)(0.5*mem /(m*sizeof(PetscScalar)));
35     brows = 1000/bcols;
36     if (bcols > nis) bcols = nis;
37     if (brows == 0 || brows > m) brows = m;
38     c->brows = brows;
39     c->bcols = bcols;
40   }
41 
42   c->M       = mat->rmap->N/bs;   /* set total rows, columns and local rows */
43   c->N       = mat->cmap->N/bs;
44   c->m       = mat->rmap->N/bs;
45   c->rstart  = 0;
46   c->ncolors = nis;
47   c->ctype   = IS_COLORING_GHOSTED;
48   PetscFunctionReturn(0);
49 }
50 
51 /*
52  Reorder Jentry such that blocked brows*bols of entries from dense matrix are inserted into Jacobian for improved cache performance
53    Input Parameters:
54 +  mat - the matrix containing the nonzero structure of the Jacobian
55 .  color - the coloring context
56 -  nz - number of local non-zeros in mat
57 */
58 #undef __FUNCT__
59 #define __FUNCT__ "MatFDColoringSetUpBlocked_AIJ_Private"
60 PetscErrorCode MatFDColoringSetUpBlocked_AIJ_Private(Mat mat,MatFDColoring c,PetscInt nz)
61 {
62   PetscErrorCode ierr;
63   PetscInt       i,j,nrows,nbcols,brows=c->brows,bcols=c->bcols,mbs=c->m,nis=c->ncolors;
64   PetscInt       *color_start,*row_start,*nrows_new,nz_new,row_end;
65 
66   PetscFunctionBegin;
67   if (brows < 1 || brows > mbs) brows = mbs;
68   ierr = PetscMalloc2(bcols+1,&color_start,bcols,&row_start);CHKERRQ(ierr);
69   ierr = PetscMalloc1(nis,&nrows_new);CHKERRQ(ierr);
70   ierr = PetscMalloc1(bcols*mat->rmap->n,&c->dy);CHKERRQ(ierr);
71   ierr = PetscLogObjectMemory((PetscObject)c,bcols*mat->rmap->n*sizeof(PetscScalar));CHKERRQ(ierr);
72 
73   nz_new = 0;
74   nbcols = 0;
75   color_start[bcols] = 0;
76 
77   if (c->htype[0] == 'd') { /* ----  c->htype == 'ds', use MatEntry --------*/
78     MatEntry       *Jentry_new,*Jentry=c->matentry;
79     ierr = PetscMalloc1(nz,&Jentry_new);CHKERRQ(ierr);
80     for (i=0; i<nis; i+=bcols) { /* loop over colors */
81       if (i + bcols > nis) {
82         color_start[nis - i] = color_start[bcols];
83         bcols                = nis - i;
84       }
85 
86       color_start[0] = color_start[bcols];
87       for (j=0; j<bcols; j++) {
88         color_start[j+1] = c->nrows[i+j] + color_start[j];
89         row_start[j]     = 0;
90       }
91 
92       row_end = brows;
93       if (row_end > mbs) row_end = mbs;
94 
95       while (row_end <= mbs) {   /* loop over block rows */
96         for (j=0; j<bcols; j++) {       /* loop over block columns */
97           nrows = c->nrows[i+j];
98           nz    = color_start[j];
99           while (row_start[j] < nrows) {
100             if (Jentry[nz].row >= row_end) {
101               color_start[j] = nz;
102               break;
103             } else { /* copy Jentry[nz] to Jentry_new[nz_new] */
104               Jentry_new[nz_new].row     = Jentry[nz].row + j*mbs; /* index in dy-array */
105               Jentry_new[nz_new].col     = Jentry[nz].col;
106               Jentry_new[nz_new].valaddr = Jentry[nz].valaddr;
107               nz_new++; nz++; row_start[j]++;
108             }
109           }
110         }
111         if (row_end == mbs) break;
112         row_end += brows;
113         if (row_end > mbs) row_end = mbs;
114       }
115       nrows_new[nbcols++] = nz_new;
116     }
117     ierr = PetscFree(Jentry);CHKERRQ(ierr);
118     c->matentry = Jentry_new;
119   } else { /* ---------  c->htype == 'wp', use MatEntry2 ------------------*/
120     MatEntry2      *Jentry2_new,*Jentry2=c->matentry2;
121     ierr = PetscMalloc1(nz,&Jentry2_new);CHKERRQ(ierr);
122     for (i=0; i<nis; i+=bcols) { /* loop over colors */
123       if (i + bcols > nis) {
124         color_start[nis - i] = color_start[bcols];
125         bcols                = nis - i;
126       }
127 
128       color_start[0] = color_start[bcols];
129       for (j=0; j<bcols; j++) {
130         color_start[j+1] = c->nrows[i+j] + color_start[j];
131         row_start[j]     = 0;
132       }
133 
134       row_end = brows;
135       if (row_end > mbs) row_end = mbs;
136 
137       while (row_end <= mbs) {   /* loop over block rows */
138         for (j=0; j<bcols; j++) {       /* loop over block columns */
139           nrows = c->nrows[i+j];
140           nz    = color_start[j];
141           while (row_start[j] < nrows) {
142             if (Jentry2[nz].row >= row_end) {
143               color_start[j] = nz;
144               break;
145             } else { /* copy Jentry2[nz] to Jentry2_new[nz_new] */
146               Jentry2_new[nz_new].row     = Jentry2[nz].row + j*mbs; /* index in dy-array */
147               Jentry2_new[nz_new].valaddr = Jentry2[nz].valaddr;
148               nz_new++; nz++; row_start[j]++;
149             }
150           }
151         }
152         if (row_end == mbs) break;
153         row_end += brows;
154         if (row_end > mbs) row_end = mbs;
155       }
156       nrows_new[nbcols++] = nz_new;
157     }
158     ierr = PetscFree(Jentry2);CHKERRQ(ierr);
159     c->matentry2 = Jentry2_new;
160   } /* ---------------------------------------------*/
161 
162   ierr = PetscFree2(color_start,row_start);CHKERRQ(ierr);
163 
164   for (i=nbcols-1; i>0; i--) nrows_new[i] -= nrows_new[i-1];
165   ierr = PetscFree(c->nrows);CHKERRQ(ierr);
166   c->nrows = nrows_new;
167   PetscFunctionReturn(0);
168 }
169 
170 #undef __FUNCT__
171 #define __FUNCT__ "MatFDColoringSetUp_SeqXAIJ"
172 PetscErrorCode MatFDColoringSetUp_SeqXAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c)
173 {
174   PetscErrorCode ierr;
175   PetscInt       i,n,nrows,mbs=c->m,j,k,m,ncols,col,nis=iscoloring->n,*rowhit,bs,bs2,*spidx,nz;
176   const PetscInt *is,*row,*ci,*cj;
177   IS             *isa;
178   PetscBool      isBAIJ;
179   PetscScalar    *A_val,**valaddrhit;
180   MatEntry       *Jentry;
181   MatEntry2      *Jentry2;
182 
183   PetscFunctionBegin;
184   ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr);
185 
186   ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr);
187   ierr = PetscObjectTypeCompare((PetscObject)mat,MATSEQBAIJ,&isBAIJ);CHKERRQ(ierr);
188   if (isBAIJ) {
189     Mat_SeqBAIJ *spA = (Mat_SeqBAIJ*)mat->data;
190     A_val = spA->a;
191     nz    = spA->nz;
192   } else {
193     Mat_SeqAIJ  *spA = (Mat_SeqAIJ*)mat->data;
194     A_val = spA->a;
195     nz    = spA->nz;
196     bs    = 1; /* only bs=1 is supported for SeqAIJ matrix */
197   }
198 
199   ierr       = PetscMalloc1(nis,&c->ncolumns);CHKERRQ(ierr);
200   ierr       = PetscMalloc1(nis,&c->columns);CHKERRQ(ierr);
201   ierr       = PetscMalloc1(nis,&c->nrows);CHKERRQ(ierr);
202   ierr       = PetscLogObjectMemory((PetscObject)c,3*nis*sizeof(PetscInt));CHKERRQ(ierr);
203 
204   if (c->htype[0] == 'd') {
205     ierr       = PetscMalloc1(nz,&Jentry);CHKERRQ(ierr);
206     ierr       = PetscLogObjectMemory((PetscObject)c,nz*sizeof(MatEntry));CHKERRQ(ierr);
207     c->matentry = Jentry;
208   } else if (c->htype[0] == 'w') {
209     ierr       = PetscMalloc1(nz,&Jentry2);CHKERRQ(ierr);
210     ierr       = PetscLogObjectMemory((PetscObject)c,nz*sizeof(MatEntry2));CHKERRQ(ierr);
211     c->matentry2 = Jentry2;
212   } else SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"htype is not supported");
213 
214   if (isBAIJ) {
215     ierr = MatGetColumnIJ_SeqBAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr);
216   } else {
217     ierr = MatGetColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr);
218   }
219 
220   ierr = PetscMalloc2(c->m,&rowhit,c->m,&valaddrhit);CHKERRQ(ierr);
221   ierr = PetscMemzero(rowhit,c->m*sizeof(PetscInt));CHKERRQ(ierr);
222 
223   nz = 0;
224   for (i=0; i<nis; i++) { /* loop over colors */
225     ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr);
226     ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr);
227 
228     c->ncolumns[i] = n;
229     if (n) {
230       ierr = PetscMalloc1(n,&c->columns[i]);CHKERRQ(ierr);
231       ierr = PetscLogObjectMemory((PetscObject)c,n*sizeof(PetscInt));CHKERRQ(ierr);
232       ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr);
233     } else {
234       c->columns[i] = 0;
235     }
236 
237     /* fast, crude version requires O(N*N) work */
238     bs2   = bs*bs;
239     nrows = 0;
240     for (j=0; j<n; j++) {  /* loop over columns */
241       col    = is[j];
242       row    = cj + ci[col];
243       m      = ci[col+1] - ci[col];
244       nrows += m;
245       for (k=0; k<m; k++) {  /* loop over columns marking them in rowhit */
246         rowhit[*row]       = col + 1;
247         valaddrhit[*row++] = &A_val[bs2*spidx[ci[col] + k]];
248       }
249     }
250     c->nrows[i] = nrows; /* total num of rows for this color */
251 
252     if (c->htype[0] == 'd') {
253       for (j=0; j<mbs; j++) { /* loop over rows */
254         if (rowhit[j]) {
255           Jentry[nz].row     = j;              /* local row index */
256           Jentry[nz].col     = rowhit[j] - 1;  /* local column index */
257           Jentry[nz].valaddr = valaddrhit[j];  /* address of mat value for this entry */
258           nz++;
259           rowhit[j] = 0.0;                     /* zero rowhit for reuse */
260         }
261       }
262     }  else { /* c->htype == 'wp' */
263       for (j=0; j<mbs; j++) { /* loop over rows */
264         if (rowhit[j]) {
265           Jentry2[nz].row     = j;              /* local row index */
266           Jentry2[nz].valaddr = valaddrhit[j];  /* address of mat value for this entry */
267           nz++;
268           rowhit[j] = 0.0;                     /* zero rowhit for reuse */
269         }
270       }
271     }
272     ierr = ISRestoreIndices(isa[i],&is);CHKERRQ(ierr);
273   }
274 
275   if (c->bcols > 1) {  /* reorder Jentry for faster MatFDColoringApply() */
276     ierr = MatFDColoringSetUpBlocked_AIJ_Private(mat,c,nz);CHKERRQ(ierr);
277   }
278 
279   if (isBAIJ) {
280     ierr = MatRestoreColumnIJ_SeqBAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr);
281     ierr = PetscMalloc1(bs*mat->rmap->n,&c->dy);CHKERRQ(ierr);
282     ierr = PetscLogObjectMemory((PetscObject)c,bs*mat->rmap->n*sizeof(PetscScalar));CHKERRQ(ierr);
283   } else {
284     ierr = MatRestoreColumnIJ_SeqAIJ_Color(mat,0,PETSC_FALSE,PETSC_FALSE,&ncols,&ci,&cj,&spidx,NULL);CHKERRQ(ierr);
285   }
286   ierr = PetscFree2(rowhit,valaddrhit);CHKERRQ(ierr);
287   ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr);
288 
289   ierr = VecCreateGhost(PetscObjectComm((PetscObject)mat),mat->rmap->n,PETSC_DETERMINE,0,NULL,&c->vscale);CHKERRQ(ierr);
290   ierr = PetscInfo3(c,"ncolors %D, brows %D and bcols %D are used.\n",c->ncolors,c->brows,c->bcols);CHKERRQ(ierr);
291   PetscFunctionReturn(0);
292 }
293