xref: /petsc/src/mat/impls/aij/mpi/fdmpiaij.c (revision 6c145f346ebf4a90e7652727dc2cc306baa3c07d) 
1 
2 #include <../src/mat/impls/aij/mpi/mpiaij.h>
3 #include <../src/mat/impls/baij/mpi/mpibaij.h>
4 
5 #undef __FUNCT__
6 #define __FUNCT__ "MatFDColoringApply_BAIJ_new"
7 PetscErrorCode  MatFDColoringApply_BAIJ_new(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx)
8 {
9   PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f;
10   PetscErrorCode ierr;
11   PetscInt       k,cstart,cend,l,row,col,nz,spidx,i,j;
12   PetscScalar    dx=0.0,*y,*xx,*w3_array,*dy_i,*dy=coloring->dy;
13   PetscScalar    *vscale_array;
14   PetscReal      epsilon=coloring->error_rel,umin=coloring->umin,unorm;
15   Vec            w1=coloring->w1,w2=coloring->w2,w3,vscale=coloring->vscale;
16   void           *fctx=coloring->fctx;
17   PetscBool      flg=PETSC_FALSE;
18   PetscInt       ctype=coloring->ctype,nxloc,nrows_k;
19   PetscScalar    *valaddr;
20   MatEntry       *Jentry=coloring->matentry;
21   const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows;
22   PetscInt       bs=J->rmap->bs;
23 
24   PetscFunctionBegin;
25   ierr = MatSetUnfactored(J);CHKERRQ(ierr);
26   ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr);
27   if (flg) {
28     ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr);
29   } else {
30     PetscBool assembled;
31     ierr = MatAssembled(J,&assembled);CHKERRQ(ierr);
32     if (assembled) {
33       ierr = MatZeroEntries(J);CHKERRQ(ierr);
34     }
35   }
36 
37   /* create vscale for storing dx */
38   if (!vscale) {
39     if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] == 'd') {
40       /* contain the "diagonal" on processor scalings followed by the off processor* - garray must be non-bloced! */
41       Mat_MPIBAIJ    *baij=(Mat_MPIBAIJ*)J->data;
42       PetscInt       *garray;
43       ierr = PetscMalloc(baij->B->cmap->n*sizeof(PetscInt),&garray);CHKERRQ(ierr);
44       for (i=0; i<baij->B->cmap->n/bs; i++) {
45         for (j=0; j<bs; j++) {
46           garray[i*bs+j] = bs*baij->garray[i]+j;
47         }
48       }
49       ierr = VecCreateGhost(PetscObjectComm((PetscObject)J),J->cmap->n,PETSC_DETERMINE,baij->B->cmap->n,garray,&vscale);CHKERRQ(ierr);
50       ierr = PetscFree(garray);CHKERRQ(ierr);
51     } else if (ctype == IS_COLORING_GHOSTED) {
52       ierr = VecDuplicate(x1,&vscale);CHKERRQ(ierr);
53     }
54     coloring->vscale = vscale;
55   }
56 
57   /* (1) Set w1 = F(x1) */
58   if (!coloring->fset) {
59     ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
60     ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr);
61     ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
62   } else {
63     coloring->fset = PETSC_FALSE;
64   }
65 
66   /* (2) Compute vscale = 1./dx - the local scale factors, including ghost points */
67   ierr = VecGetLocalSize(x1,&nxloc);CHKERRQ(ierr);
68   if (coloring->htype[0] == 'w') {
69     /* vscale = dx is a constant scalar */
70     ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr);
71     dx = 1.0/(PetscSqrtReal(1.0 + unorm)*epsilon);
72   } else {
73     ierr = VecGetArray(x1,&xx);CHKERRQ(ierr);
74     ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr);
75     for (col=0; col<nxloc; col++) {
76       dx = xx[col];
77       if (PetscAbsScalar(dx) < umin) {
78         if (PetscRealPart(dx) >= 0.0)      dx = umin;
79         else if (PetscRealPart(dx) < 0.0 ) dx = -umin;
80       }
81       dx               *= epsilon;
82       vscale_array[col] = 1.0/dx;
83     }
84     ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr);
85     ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr);
86   }
87   if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] != 'w') {
88     ierr = VecGhostUpdateBegin(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
89     ierr = VecGhostUpdateEnd(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
90   }
91 
92   /* (3) Loop over each color */
93   if (!coloring->w3) {
94     ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr);
95     ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr);
96   }
97   w3 = coloring->w3;
98 
99   ierr = VecGetOwnershipRange(x1,&cstart,&cend);CHKERRQ(ierr); /* used by ghosted vscale */
100   if (vscale) {
101     ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr);
102   }
103   nz   = 0;
104   for (k=0; k<ncolors; k++) {
105     coloring->currentcolor = k;
106 
107     /*
108       (3-1) Loop over each column associated with color
109       adding the perturbation to the vector w3 = x1 + dx.
110     */
111     ierr = VecCopy(x1,w3);CHKERRQ(ierr);
112     dy_i = dy;
113     for (i=0; i<bs; i++) {     /* Loop over a block of columns */
114       //------------------------------------------------
115       ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr);
116       if (ctype == IS_COLORING_GLOBAL) w3_array -= cstart; /* shift pointer so global index can be used */
117       if (coloring->htype[0] == 'w') {
118         for (l=0; l<ncolumns[k]; l++) {
119           col            = i + bs*coloring->columns[k][l];  /* local column (in global index!) of the matrix we are probing for */
120           w3_array[col] += 1.0/dx;
121 
122           if (i) {
123             w3_array[col-1] -= 1.0/dx; /* resume original w3[col-1] */
124           }
125 
126         }
127       } else { /* htype == 'ds' */
128         vscale_array -= cstart; /* shift pointer so global index can be used */
129         for (l=0; l<ncolumns[k]; l++) {
130           col = i + bs*coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */
131           w3_array[col] += 1.0/vscale_array[col];
132 
133           if (i) {
134             w3_array[col-1] -=  1.0/vscale_array[col-1]; /* resume original w3[col-1] */
135           }
136 
137         }
138         vscale_array += cstart;
139       }
140       if (ctype == IS_COLORING_GLOBAL) w3_array += cstart;
141       ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr);
142 
143       /*
144        (3-2) Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations)
145                            w2 = F(x1 + dx) - F(x1)
146        */
147       ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
148       ierr = VecPlaceArray(w2,dy_i);CHKERRQ(ierr); /* place w2 to the array dy_i */
149       ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr);
150       ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
151       ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr);
152       //---------------------------------------------
153       ierr = VecResetArray(w2);CHKERRQ(ierr);
154       dy_i += nxloc; /* points to dy+i*nxloc */
155     }
156 
157     /*
158      (3-3) Loop over rows of vector, putting results into Jacobian matrix
159     */
160     nrows_k = nrows[k];
161     ierr = VecGetArray(w2,&y);CHKERRQ(ierr);
162     if (coloring->htype[0] == 'w') {
163       for (l=0; l<nrows_k; l++) {
164         row     = bs*Jentry[nz].row;   /* local row index */
165         valaddr = Jentry[nz].valaddr;
166         nz++;
167         spidx = 0;
168         dy_i  = dy;
169         for (i=0; i<bs; i++) {   /* column of the block */
170           for (j=0; j<bs; j++) { /* row of the block */
171             valaddr[spidx++] = dy_i[row+j]*dx;
172           }
173           dy_i += nxloc; /* points to dy+i*nxloc */
174         }
175       }
176     } else { /* htype == 'ds' */
177       for (l=0; l<nrows_k; l++) {
178         row     = bs*Jentry[nz].row;   /* local row index */
179         col     = bs*Jentry[nz].col;   /* local column index */
180         valaddr = Jentry[nz].valaddr;
181         nz++;
182         spidx = 0;
183         dy_i  = dy;
184         for (i=0; i<bs; i++) {   /* column of the block */
185           for (j=0; j<bs; j++) { /* row of the block */
186             valaddr[spidx++] = dy_i[row+j]*vscale_array[col+i];
187           }
188           dy_i += nxloc; /* points to dy+i*nxloc */
189         }
190       }
191     }
192     ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr);
193   }
194   ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
195   ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
196   if (vscale) {
197     ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr);
198   }
199 
200   coloring->currentcolor = -1;
201   PetscFunctionReturn(0);
202 }
203 
204 extern PetscErrorCode MatCreateColmap_MPIAIJ_Private(Mat);
205 #undef __FUNCT__
206 #define __FUNCT__ "MatFDColoringApply_AIJ_new"
207 PetscErrorCode  MatFDColoringApply_AIJ_new(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx)
208 {
209   PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f;
210   PetscErrorCode ierr;
211   PetscInt       k,cstart,cend,l,row,col,nz;
212   PetscScalar    dx=0.0,*y,*xx,*w3_array;
213   PetscScalar    *vscale_array;
214   PetscReal      epsilon=coloring->error_rel,umin=coloring->umin,unorm;
215   Vec            w1=coloring->w1,w2=coloring->w2,w3,vscale=coloring->vscale;
216   void           *fctx=coloring->fctx;
217   PetscBool      flg=PETSC_FALSE;
218   PetscInt       ctype=coloring->ctype,nxloc,nrows_k;
219   MatEntry       *Jentry=coloring->matentry;
220   const PetscInt ncolors=coloring->ncolors,*ncolumns=coloring->ncolumns,*nrows=coloring->nrows;
221 
222   PetscFunctionBegin;
223   ierr = MatSetUnfactored(J);CHKERRQ(ierr);
224   ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr);
225   if (flg) {
226     ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr);
227   } else {
228     PetscBool assembled;
229     ierr = MatAssembled(J,&assembled);CHKERRQ(ierr);
230     if (assembled) {
231       ierr = MatZeroEntries(J);CHKERRQ(ierr);
232     }
233   }
234 
235   /* create vscale for storing dx */
236   if (!vscale) {
237     if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] == 'd') {
238       Mat_MPIAIJ     *aij=(Mat_MPIAIJ*)J->data;
239       ierr = VecCreateGhost(PetscObjectComm((PetscObject)J),J->cmap->n,PETSC_DETERMINE,aij->B->cmap->n,aij->garray,&vscale);CHKERRQ(ierr);
240     } else if (ctype == IS_COLORING_GHOSTED) {
241       ierr = VecDuplicate(x1,&vscale);CHKERRQ(ierr);
242     }
243     coloring->vscale = vscale;
244   }
245 
246   /* (1) Set w1 = F(x1) */
247   if (!coloring->fset) {
248     ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
249     ierr = (*f)(sctx,x1,w1,fctx);CHKERRQ(ierr);
250     ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
251   } else {
252     coloring->fset = PETSC_FALSE;
253   }
254 
255   /* (2) Compute vscale = 1./dx - the local scale factors, including ghost points */
256   if (coloring->htype[0] == 'w') {
257     /* vscale = dx is a constant scalar */
258     ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr);
259     dx = 1.0/(PetscSqrtReal(1.0 + unorm)*epsilon);
260   } else {
261     ierr = VecGetLocalSize(x1,&nxloc);CHKERRQ(ierr);
262     ierr = VecGetArray(x1,&xx);CHKERRQ(ierr);
263     ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr);
264     for (col=0; col<nxloc; col++) {
265       dx = xx[col];
266       if (PetscAbsScalar(dx) < umin) {
267         if (PetscRealPart(dx) >= 0.0)      dx = umin;
268         else if (PetscRealPart(dx) < 0.0 ) dx = -umin;
269       }
270       dx               *= epsilon;
271       vscale_array[col] = 1.0/dx;
272     }
273     ierr = VecRestoreArray(x1,&xx);CHKERRQ(ierr);
274     ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr);
275   }
276   if (ctype == IS_COLORING_GLOBAL && coloring->htype[0] != 'w') {
277     ierr = VecGhostUpdateBegin(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
278     ierr = VecGhostUpdateEnd(vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
279   }
280 
281   /* (3) Loop over each color */
282   if (!coloring->w3) {
283     ierr = VecDuplicate(x1,&coloring->w3);CHKERRQ(ierr);
284     ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr);
285   }
286   w3 = coloring->w3;
287 
288   ierr = VecGetOwnershipRange(x1,&cstart,&cend);CHKERRQ(ierr); /* used by ghosted vscale */
289   if (vscale) {
290     ierr = VecGetArray(vscale,&vscale_array);CHKERRQ(ierr);
291   }
292   nz   = 0;
293   for (k=0; k<ncolors; k++) {
294     coloring->currentcolor = k;
295 
296     /*
297       (3-1) Loop over each column associated with color
298       adding the perturbation to the vector w3 = x1 + dx.
299     */
300     ierr = VecCopy(x1,w3);CHKERRQ(ierr);
301     ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr);
302     if (ctype == IS_COLORING_GLOBAL) w3_array -= cstart; /* shift pointer so global index can be used */
303     if (coloring->htype[0] == 'w') {
304       for (l=0; l<ncolumns[k]; l++) {
305         col = coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */
306         w3_array[col] += 1.0/dx;
307       }
308     } else { /* htype == 'ds' */
309       vscale_array -= cstart; /* shift pointer so global index can be used */
310       for (l=0; l<ncolumns[k]; l++) {
311         col = coloring->columns[k][l]; /* local column (in global index!) of the matrix we are probing for */
312         w3_array[col] += 1.0/vscale_array[col];
313       }
314       vscale_array += cstart;
315     }
316     if (ctype == IS_COLORING_GLOBAL) w3_array += cstart;
317     ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr);
318 
319     /*
320       (3-2) Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations)
321                            w2 = F(x1 + dx) - F(x1)
322     */
323     ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
324     ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr);
325     ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
326     ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr);
327 
328     /*
329      (3-3) Loop over rows of vector, putting results into Jacobian matrix
330     */
331     nrows_k = nrows[k];
332     ierr = VecGetArray(w2,&y);CHKERRQ(ierr);
333     if (coloring->htype[0] == 'w') {
334       for (l=0; l<nrows_k; l++) {
335         row                     = Jentry[nz].row;   /* local row index */
336         *(Jentry[nz++].valaddr) = y[row]*dx;
337       }
338     } else { /* htype == 'ds' */
339       for (l=0; l<nrows_k; l++) {
340         row                   = Jentry[nz].row;   /* local row index */
341         *(Jentry[nz].valaddr) = y[row]*vscale_array[Jentry[nz].col];
342         nz++;
343       }
344     }
345     ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr);
346   }
347   ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
348   ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
349   if (vscale) {
350     ierr = VecRestoreArray(vscale,&vscale_array);CHKERRQ(ierr);
351   }
352 
353   coloring->currentcolor = -1;
354   PetscFunctionReturn(0);
355 }
356 
357 #undef __FUNCT__
358 #define __FUNCT__ "MatFDColoringCreate_MPIAIJ_new"
359 PetscErrorCode MatFDColoringCreate_MPIAIJ_new(Mat mat,ISColoring iscoloring,MatFDColoring c)
360 {
361   PetscErrorCode         ierr;
362   PetscMPIInt            size,*ncolsonproc,*disp,nn;
363   PetscInt               i,n,nrows,nrows_i,j,k,m,ncols,col;
364   const PetscInt         *is,*A_ci,*A_cj,*B_ci,*B_cj,*row = NULL,*ltog=NULL;
365   PetscInt               nis=iscoloring->n,nctot,*cols;
366   PetscInt               *rowhit,cstart,cend,colb;
367   IS                     *isa;
368   ISLocalToGlobalMapping map=mat->cmap->mapping;
369   PetscInt               ctype=c->ctype;
370   Mat                    A,B;
371   PetscScalar            *A_val,*B_val;
372   PetscInt               spidx;
373   PetscInt               *spidxA,*spidxB,nz,bs,bs2;
374   PetscScalar            **valaddrhit;
375   MatEntry               *Jentry;
376   PetscBool              isBAIJ;
377 #if defined(PETSC_USE_CTABLE)
378   PetscTable             colmap=NULL;
379 #else
380   PetscInt               *colmap=NULL;     /* local col number of off-diag col */
381 #endif
382 
383   PetscFunctionBegin;
384   if (ctype == IS_COLORING_GHOSTED) {
385     if (!map) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_ARG_INCOMP,"When using ghosted differencing matrix must have local to global mapping provided with MatSetLocalToGlobalMapping");
386     ierr = ISLocalToGlobalMappingGetIndices(map,&ltog);CHKERRQ(ierr);
387   }
388 
389   ierr = MatGetBlockSize(mat,&bs);CHKERRQ(ierr);
390   ierr = PetscObjectTypeCompare((PetscObject)mat,MATMPIBAIJ,&isBAIJ);CHKERRQ(ierr);
391   if (!isBAIJ) {
392     Mat_MPIAIJ             *aij=(Mat_MPIAIJ*)mat->data;
393     Mat_SeqAIJ             *spA,*spB;
394     bs = 1; /* only bs=1 is supported for non MPIBAIJ matrix */
395     A=aij->A;  spA=(Mat_SeqAIJ*)A->data; A_val=spA->a;
396     B=aij->B;  spB=(Mat_SeqAIJ*)B->data; B_val=spB->a;
397     nz = spA->nz + spB->nz; /* total nonzero entries of mat */
398     if (!aij->colmap) {
399       /* Allow access to data structures of local part of matrix
400        - creates aij->colmap which maps global column number to local number in part B */
401       ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr);
402       colmap = aij->colmap;
403     }
404     ierr = MatGetColumnIJ_SeqAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr);
405     ierr = MatGetColumnIJ_SeqAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr);
406   } else {
407     Mat_MPIBAIJ             *aij=(Mat_MPIBAIJ*)mat->data;
408     Mat_SeqBAIJ             *spA,*spB;
409     A=aij->A;  spA=(Mat_SeqBAIJ*)A->data; A_val=spA->a;
410     B=aij->B;  spB=(Mat_SeqBAIJ*)B->data; B_val=spB->a;
411     nz = spA->nz + spB->nz; /* total nonzero entries of mat */
412     if (!aij->colmap) {
413       ierr = MatCreateColmap_MPIBAIJ_Private(mat);CHKERRQ(ierr);
414       colmap = aij->colmap;
415     }
416     ierr = MatGetColumnIJ_SeqBAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr);
417     ierr = MatGetColumnIJ_SeqBAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr);
418   }
419 
420   m         = mat->rmap->n/bs;
421   cstart    = mat->cmap->rstart/bs;
422   cend      = mat->cmap->rend/bs;
423   c->M      = mat->rmap->N/bs;         /* set the global rows and columns and local rows */
424   c->N      = mat->cmap->N/bs;
425   c->m      = m;
426   c->rstart = mat->rmap->rstart/bs;
427 
428   c->ncolors = nis;
429   ierr       = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr);
430   ierr       = PetscMalloc(nis*sizeof(PetscInt*),&c->columns);CHKERRQ(ierr);
431   ierr       = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr);
432   ierr       = PetscLogObjectMemory((PetscObject)c,3*nis*sizeof(PetscInt));CHKERRQ(ierr);
433 
434   ierr       = PetscMalloc(nz*sizeof(MatEntry),&Jentry);CHKERRQ(ierr);
435   ierr       = PetscLogObjectMemory((PetscObject)c,nz*sizeof(MatEntry));CHKERRQ(ierr);
436   c->matentry = Jentry;
437 
438   ierr = PetscMalloc2(m+1,PetscInt,&rowhit,m+1,PetscScalar*,&valaddrhit);CHKERRQ(ierr);
439   nz = 0;
440   ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr);
441   for (i=0; i<nis; i++) { /* for each local color */
442     ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr);
443     ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr);
444 
445     c->ncolumns[i] = n; /* local number of columns of this color on this process */
446     if (n) {
447       ierr = PetscMalloc(n*sizeof(PetscInt),&c->columns[i]);CHKERRQ(ierr);
448       ierr = PetscLogObjectMemory((PetscObject)c,n*sizeof(PetscInt));CHKERRQ(ierr);
449       ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr);
450       /* convert global column indices to local ones! */
451 
452     } else {
453       c->columns[i] = 0;
454     }
455 
456     if (ctype == IS_COLORING_GLOBAL) {
457       /* Determine nctot, the total (parallel) number of columns of this color */
458       ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr);
459       ierr = PetscMalloc2(size,PetscMPIInt,&ncolsonproc,size,PetscMPIInt,&disp);CHKERRQ(ierr);
460 
461       /* ncolsonproc[j]: local ncolumns on proc[j] of this color */
462       ierr  = PetscMPIIntCast(n,&nn);CHKERRQ(ierr);
463       ierr  = MPI_Allgather(&nn,1,MPI_INT,ncolsonproc,1,MPI_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr);
464       nctot = 0; for (j=0; j<size; j++) nctot += ncolsonproc[j];
465       if (!nctot) {
466         ierr = PetscInfo(mat,"Coloring of matrix has some unneeded colors with no corresponding rows\n");CHKERRQ(ierr);
467       }
468 
469       disp[0] = 0;
470       for (j=1; j<size; j++) {
471         disp[j] = disp[j-1] + ncolsonproc[j-1];
472       }
473 
474       /* Get cols, the complete list of columns for this color on each process */
475       ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr);
476       ierr = MPI_Allgatherv((void*)is,n,MPIU_INT,cols,ncolsonproc,disp,MPIU_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr);
477       ierr = PetscFree2(ncolsonproc,disp);CHKERRQ(ierr);
478     } else if (ctype == IS_COLORING_GHOSTED) {
479       /* Determine local number of columns of this color on this process, including ghost points */
480       nctot = n;
481       ierr  = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr);
482       ierr  = PetscMemcpy(cols,is,n*sizeof(PetscInt));CHKERRQ(ierr);
483     } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not provided for this MatFDColoring type");
484 
485     /* Mark all rows affect by these columns */
486     ierr    = PetscMemzero(rowhit,m*sizeof(PetscInt));CHKERRQ(ierr);
487     bs2     = bs*bs;
488     nrows_i = 0;
489     for (j=0; j<nctot; j++) { /* loop over columns*/
490       if (ctype == IS_COLORING_GHOSTED) {
491         col = ltog[cols[j]];
492       } else {
493         col = cols[j];
494       }
495       if (col >= cstart && col < cend) { /* column is in diagonal block of matrix A */
496         row      = A_cj + A_ci[col-cstart];
497         nrows    = A_ci[col-cstart+1] - A_ci[col-cstart];
498         nrows_i += nrows;
499         /* loop over columns of A marking them in rowhit */
500         for (k=0; k<nrows; k++) {
501           /* set valaddrhit for part A */
502           spidx            = spidxA[A_ci[col-cstart] + k];
503           valaddrhit[*row] = &A_val[bs2*spidx];
504           rowhit[*row++]   = col - cstart + 1; /* local column index */
505         }
506       } else { /* column is in off-diagonal block of matrix B */
507 #if defined(PETSC_USE_CTABLE)
508         ierr = PetscTableFind(colmap,col+1,&colb);CHKERRQ(ierr);
509         colb--;
510 #else
511         colb = colmap[col] - 1; /* local column index */
512 #endif
513         if (colb == -1) {
514           nrows = 0;
515         } else {
516           colb  = colb/bs;
517           row   = B_cj + B_ci[colb];
518           nrows = B_ci[colb+1] - B_ci[colb];
519         }
520         nrows_i += nrows;
521         /* loop over columns of B marking them in rowhit */
522         for (k=0; k<nrows; k++) {
523           /* set valaddrhit for part B */
524           spidx            = spidxB[B_ci[colb] + k];
525           valaddrhit[*row] = &B_val[bs2*spidx];
526           rowhit[*row++]   = colb + 1 + cend - cstart; /* local column index */
527         }
528       }
529     }
530     c->nrows[i] = nrows_i;
531 
532     for (j=0; j<m; j++) {
533       if (rowhit[j]) {
534         Jentry[nz].row     = j;              /* local row index */
535         Jentry[nz].col     = rowhit[j] - 1;  /* local column index */
536         Jentry[nz].valaddr = valaddrhit[j];  /* address of mat value for this entry */
537         nz++;
538       }
539     }
540     ierr = PetscFree(cols);CHKERRQ(ierr);
541   }
542   ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr);
543 
544   ierr = PetscFree2(rowhit,valaddrhit);CHKERRQ(ierr);
545   if (isBAIJ) {
546     ierr = MatRestoreColumnIJ_SeqBAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr);
547     ierr = MatRestoreColumnIJ_SeqBAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr);
548     ierr = PetscMalloc(bs*mat->cmap->n*sizeof(PetscScalar),&c->dy);CHKERRQ(ierr);
549     mat->ops->fdcoloringapply = MatFDColoringApply_BAIJ_new;
550   } else {
551     ierr = MatRestoreColumnIJ_SeqAIJ_Color(A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&spidxA,NULL);CHKERRQ(ierr);
552     ierr = MatRestoreColumnIJ_SeqAIJ_Color(B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&spidxB,NULL);CHKERRQ(ierr);
553     mat->ops->fdcoloringapply = MatFDColoringApply_AIJ_new;
554   }
555 
556   if (ctype == IS_COLORING_GHOSTED) {
557     ierr = ISLocalToGlobalMappingRestoreIndices(map,&ltog);CHKERRQ(ierr);
558   }
559   PetscFunctionReturn(0);
560 }
561 
562 /*------------------------------------------------------*/
563 #undef __FUNCT__
564 #define __FUNCT__ "MatFDColoringCreate_MPIAIJ"
565 PetscErrorCode MatFDColoringCreate_MPIAIJ(Mat mat,ISColoring iscoloring,MatFDColoring c)
566 {
567   Mat_MPIAIJ             *aij = (Mat_MPIAIJ*)mat->data;
568   PetscErrorCode         ierr;
569   PetscMPIInt            size,*ncolsonproc,*disp,nn;
570   PetscInt               i,n,nrows,j,k,m,ncols,col;
571   const PetscInt         *is,*A_ci,*A_cj,*B_ci,*B_cj,*rows = 0,*ltog;
572   PetscInt               nis = iscoloring->n,nctot,*cols;
573   PetscInt               *rowhit,M,cstart,cend,colb;
574   PetscInt               *columnsforrow,l;
575   IS                     *isa;
576   PetscBool              done,flg;
577   ISLocalToGlobalMapping map   = mat->cmap->mapping;
578   PetscInt               ctype=c->ctype;
579   PetscBool              new_impl=PETSC_FALSE;
580 
581   PetscFunctionBegin;
582   ierr = PetscOptionsName("-new","using new impls","",&new_impl);CHKERRQ(ierr);
583   if (new_impl){
584     ierr =  MatFDColoringCreate_MPIAIJ_new(mat,iscoloring,c);CHKERRQ(ierr);
585     PetscFunctionReturn(0);
586   }
587   if (ctype == IS_COLORING_GHOSTED && !map) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_ARG_INCOMP,"When using ghosted differencing matrix must have local to global mapping provided with MatSetLocalToGlobalMapping");
588 
589   if (map) {ierr = ISLocalToGlobalMappingGetIndices(map,&ltog);CHKERRQ(ierr);}
590   else     ltog = NULL;
591   ierr = ISColoringGetIS(iscoloring,PETSC_IGNORE,&isa);CHKERRQ(ierr);
592 
593   M         = mat->rmap->n;
594   cstart    = mat->cmap->rstart;
595   cend      = mat->cmap->rend;
596   c->M      = mat->rmap->N;         /* set the global rows and columns and local rows */
597   c->N      = mat->cmap->N;
598   c->m      = mat->rmap->n;
599   c->rstart = mat->rmap->rstart;
600 
601   c->ncolors = nis;
602   ierr       = PetscMalloc(nis*sizeof(PetscInt),&c->ncolumns);CHKERRQ(ierr);
603   ierr       = PetscMalloc(nis*sizeof(PetscInt*),&c->columns);CHKERRQ(ierr);
604   ierr       = PetscMalloc(nis*sizeof(PetscInt),&c->nrows);CHKERRQ(ierr);
605   ierr       = PetscMalloc(nis*sizeof(PetscInt*),&c->rows);CHKERRQ(ierr);
606   ierr       = PetscMalloc(nis*sizeof(PetscInt*),&c->columnsforrow);CHKERRQ(ierr);
607   ierr       = PetscLogObjectMemory((PetscObject)c,5*nis*sizeof(PetscInt));CHKERRQ(ierr);
608 
609   /* Allow access to data structures of local part of matrix */
610   if (!aij->colmap) {
611     ierr = MatCreateColmap_MPIAIJ_Private(mat);CHKERRQ(ierr);
612   }
613   ierr = MatGetColumnIJ(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&done);CHKERRQ(ierr);
614   ierr = MatGetColumnIJ(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&done);CHKERRQ(ierr);
615 
616   ierr = PetscMalloc((M+1)*sizeof(PetscInt),&rowhit);CHKERRQ(ierr);
617   ierr = PetscMalloc((M+1)*sizeof(PetscInt),&columnsforrow);CHKERRQ(ierr);
618 
619   for (i=0; i<nis; i++) {
620     ierr = ISGetLocalSize(isa[i],&n);CHKERRQ(ierr);
621     ierr = ISGetIndices(isa[i],&is);CHKERRQ(ierr);
622 
623     c->ncolumns[i] = n; /* local number of columns of this color on this process */
624     if (n) {
625       ierr = PetscMalloc(n*sizeof(PetscInt),&c->columns[i]);CHKERRQ(ierr);
626       ierr = PetscLogObjectMemory((PetscObject)c,n*sizeof(PetscInt));CHKERRQ(ierr);
627       ierr = PetscMemcpy(c->columns[i],is,n*sizeof(PetscInt));CHKERRQ(ierr);
628     } else {
629       c->columns[i] = 0;
630     }
631 
632     if (ctype == IS_COLORING_GLOBAL) {
633       /* Determine the total (parallel) number of columns of this color */
634       ierr = MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);CHKERRQ(ierr);
635       ierr = PetscMalloc2(size,PetscMPIInt,&ncolsonproc,size,PetscMPIInt,&disp);CHKERRQ(ierr);
636 
637       ierr  = PetscMPIIntCast(n,&nn);CHKERRQ(ierr);
638       ierr  = MPI_Allgather(&nn,1,MPI_INT,ncolsonproc,1,MPI_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr);
639       nctot = 0; for (j=0; j<size; j++) nctot += ncolsonproc[j];
640       if (!nctot) {
641         ierr = PetscInfo(mat,"Coloring of matrix has some unneeded colors with no corresponding rows\n");CHKERRQ(ierr);
642       }
643 
644       disp[0] = 0;
645       for (j=1; j<size; j++) {
646         disp[j] = disp[j-1] + ncolsonproc[j-1];
647       }
648 
649       /* Get complete list of columns for color on each processor */
650       ierr = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr);
651       ierr = MPI_Allgatherv((void*)is,n,MPIU_INT,cols,ncolsonproc,disp,MPIU_INT,PetscObjectComm((PetscObject)mat));CHKERRQ(ierr);
652       ierr = PetscFree2(ncolsonproc,disp);CHKERRQ(ierr);
653     } else if (ctype == IS_COLORING_GHOSTED) {
654       /* Determine local number of columns of this color on this process, including ghost points */
655       nctot = n;
656       ierr  = PetscMalloc((nctot+1)*sizeof(PetscInt),&cols);CHKERRQ(ierr);
657       ierr  = PetscMemcpy(cols,is,n*sizeof(PetscInt));CHKERRQ(ierr);
658     } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not provided for this MatFDColoring type");
659 
660     /*
661        Mark all rows affect by these columns
662     */
663     /* Temporary option to allow for debugging/testing */
664     flg  = PETSC_FALSE;
665     ierr = PetscOptionsGetBool(NULL,"-matfdcoloring_slow",&flg,NULL);CHKERRQ(ierr);
666     if (!flg) { /*-----------------------------------------------------------------------------*/
667       /* crude, fast version */
668       ierr = PetscMemzero(rowhit,M*sizeof(PetscInt));CHKERRQ(ierr);
669       /* loop over columns*/
670       for (j=0; j<nctot; j++) {
671         if (ctype == IS_COLORING_GHOSTED) {
672           col = ltog[cols[j]];
673         } else {
674           col = cols[j];
675         }
676         if (col >= cstart && col < cend) {
677           /* column is in diagonal block of matrix */
678           rows = A_cj + A_ci[col-cstart];
679           m    = A_ci[col-cstart+1] - A_ci[col-cstart];
680         } else {
681 #if defined(PETSC_USE_CTABLE)
682           ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr);
683           colb--;
684 #else
685           colb = aij->colmap[col] - 1;
686 #endif
687           if (colb == -1) {
688             m = 0;
689           } else {
690             rows = B_cj + B_ci[colb];
691             m    = B_ci[colb+1] - B_ci[colb];
692           }
693         }
694         /* loop over columns marking them in rowhit */
695         for (k=0; k<m; k++) {
696           rowhit[*rows++] = col + 1;
697         }
698       }
699 
700       /* count the number of hits */
701       nrows = 0;
702       for (j=0; j<M; j++) {
703         if (rowhit[j]) nrows++;
704       }
705       c->nrows[i] = nrows;
706       ierr        = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->rows[i]);CHKERRQ(ierr);
707       ierr        = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->columnsforrow[i]);CHKERRQ(ierr);
708       ierr        = PetscLogObjectMemory((PetscObject)c,2*(nrows+1)*sizeof(PetscInt));CHKERRQ(ierr);
709       nrows       = 0;
710       for (j=0; j<M; j++) {
711         if (rowhit[j]) {
712           c->rows[i][nrows]          = j;              /* local row index */
713           c->columnsforrow[i][nrows] = rowhit[j] - 1;  /* global column index */
714           nrows++;
715         }
716       }
717     } else { /*-------------------------------------------------------------------------------*/
718       /* slow version, using rowhit as a linked list */
719       PetscInt currentcol,fm,mfm;
720       rowhit[M] = M;
721       nrows     = 0;
722       /* loop over columns*/
723       for (j=0; j<nctot; j++) {
724         if (ctype == IS_COLORING_GHOSTED) {
725           col = ltog[cols[j]];
726         } else {
727           col = cols[j];
728         }
729         if (col >= cstart && col < cend) {
730           /* column is in diagonal block of matrix */
731           rows = A_cj + A_ci[col-cstart];
732           m    = A_ci[col-cstart+1] - A_ci[col-cstart];
733         } else {
734 #if defined(PETSC_USE_CTABLE)
735           ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr);
736           colb--;
737 #else
738           colb = aij->colmap[col] - 1;
739 #endif
740           if (colb == -1) {
741             m = 0;
742           } else {
743             rows = B_cj + B_ci[colb];
744             m    = B_ci[colb+1] - B_ci[colb];
745           }
746         }
747 
748         /* loop over columns marking them in rowhit */
749         fm = M;    /* fm points to first entry in linked list */
750         for (k=0; k<m; k++) {
751           currentcol = *rows++;
752           /* is it already in the list? */
753           do {
754             mfm = fm;
755             fm  = rowhit[fm];
756           } while (fm < currentcol);
757           /* not in list so add it */
758           if (fm != currentcol) {
759             nrows++;
760             columnsforrow[currentcol] = col;
761             /* next three lines insert new entry into linked list */
762             rowhit[mfm]        = currentcol;
763             rowhit[currentcol] = fm;
764             fm                 = currentcol;
765             /* fm points to present position in list since we know the columns are sorted */
766           } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Invalid coloring of matrix detected");
767         }
768       }
769       c->nrows[i] = nrows;
770 
771       ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->rows[i]);CHKERRQ(ierr);
772       ierr = PetscMalloc((nrows+1)*sizeof(PetscInt),&c->columnsforrow[i]);CHKERRQ(ierr);
773       ierr = PetscLogObjectMemory((PetscObject)c,(nrows+1)*sizeof(PetscInt));CHKERRQ(ierr);
774       /* now store the linked list of rows into c->rows[i] */
775       nrows = 0;
776       fm    = rowhit[M];
777       do {
778         c->rows[i][nrows]            = fm;
779         c->columnsforrow[i][nrows++] = columnsforrow[fm];
780         fm                           = rowhit[fm];
781       } while (fm < M);
782     } /* ---------------------------------------------------------------------------------------*/
783     ierr = PetscFree(cols);CHKERRQ(ierr);
784   }
785 
786   /* Optimize by adding the vscale, and scaleforrow[][] fields */
787   /*
788        vscale will contain the "diagonal" on processor scalings followed by the off processor
789   */
790   if (ctype == IS_COLORING_GLOBAL) {
791     ierr = VecCreateGhost(PetscObjectComm((PetscObject)mat),aij->A->rmap->n,PETSC_DETERMINE,aij->B->cmap->n,aij->garray,&c->vscale);CHKERRQ(ierr);
792     ierr = PetscMalloc(c->ncolors*sizeof(PetscInt*),&c->vscaleforrow);CHKERRQ(ierr);
793     for (k=0; k<c->ncolors; k++) {
794       ierr = PetscMalloc((c->nrows[k]+1)*sizeof(PetscInt),&c->vscaleforrow[k]);CHKERRQ(ierr);
795       for (l=0; l<c->nrows[k]; l++) {
796         col = c->columnsforrow[k][l];
797         if (col >= cstart && col < cend) {
798           /* column is in diagonal block of matrix */
799           colb = col - cstart;
800         } else {
801           /* column  is in "off-processor" part */
802 #if defined(PETSC_USE_CTABLE)
803           ierr = PetscTableFind(aij->colmap,col+1,&colb);CHKERRQ(ierr);
804           colb--;
805 #else
806           colb = aij->colmap[col] - 1;
807 #endif
808           colb += cend - cstart;
809         }
810         c->vscaleforrow[k][l] = colb;
811       }
812     }
813   } else if (ctype == IS_COLORING_GHOSTED) {
814     /* Get gtol mapping */
815     PetscInt N = mat->cmap->N,nlocal,*gtol;
816     ierr = PetscMalloc((N+1)*sizeof(PetscInt),&gtol);CHKERRQ(ierr);
817     for (i=0; i<N; i++) gtol[i] = -1;
818     ierr = ISLocalToGlobalMappingGetSize(map,&nlocal);CHKERRQ(ierr);
819     for (i=0; i<nlocal; i++) gtol[ltog[i]] = i;
820 
821     c->vscale = 0; /* will be created in MatFDColoringApply() */
822     ierr      = PetscMalloc(c->ncolors*sizeof(PetscInt*),&c->vscaleforrow);CHKERRQ(ierr);
823     for (k=0; k<c->ncolors; k++) {
824       ierr = PetscMalloc((c->nrows[k]+1)*sizeof(PetscInt),&c->vscaleforrow[k]);CHKERRQ(ierr);
825       for (l=0; l<c->nrows[k]; l++) {
826         col = c->columnsforrow[k][l];      /* global column index */
827         c->vscaleforrow[k][l] = gtol[col]; /* local column index */
828       }
829     }
830     ierr = PetscFree(gtol);CHKERRQ(ierr);
831   }
832   ierr = ISColoringRestoreIS(iscoloring,&isa);CHKERRQ(ierr);
833 
834   ierr = PetscFree(rowhit);CHKERRQ(ierr);
835   ierr = PetscFree(columnsforrow);CHKERRQ(ierr);
836   ierr = MatRestoreColumnIJ(aij->A,0,PETSC_FALSE,PETSC_FALSE,&ncols,&A_ci,&A_cj,&done);CHKERRQ(ierr);
837   ierr = MatRestoreColumnIJ(aij->B,0,PETSC_FALSE,PETSC_FALSE,&ncols,&B_ci,&B_cj,&done);CHKERRQ(ierr);
838   if (map) {ierr = ISLocalToGlobalMappingRestoreIndices(map,&ltog);CHKERRQ(ierr);}
839   PetscFunctionReturn(0);
840 }
841 
842 
843 
844 
845 
846 
847