xref: /petsc/src/mat/matfd/fdmatrix.c (revision 573f477f51351f7b9b72c6b8eee6b38481c3e253)
1 
2 /*
3    This is where the abstract matrix operations are defined that are
4   used for finite difference computations of Jacobians using coloring.
5 */
6 
7 #include <petsc-private/matimpl.h>        /*I "petscmat.h" I*/
8 
9 #undef __FUNCT__
10 #define __FUNCT__ "MatFDColoringSetF"
11 PetscErrorCode  MatFDColoringSetF(MatFDColoring fd,Vec F)
12 {
13   PetscErrorCode ierr;
14 
15   PetscFunctionBegin;
16   if (F) {
17     ierr     = VecCopy(F,fd->w1);CHKERRQ(ierr);
18     fd->fset = PETSC_TRUE;
19   } else {
20     fd->fset = PETSC_FALSE;
21   }
22   PetscFunctionReturn(0);
23 }
24 
25 #include <petscdraw.h>
26 #undef __FUNCT__
27 #define __FUNCT__ "MatFDColoringView_Draw_Zoom"
28 static PetscErrorCode MatFDColoringView_Draw_Zoom(PetscDraw draw,void *Aa)
29 {
30   MatFDColoring  fd = (MatFDColoring)Aa;
31   PetscErrorCode ierr;
32   PetscInt       i,j;
33   PetscReal      x,y;
34 
35   PetscFunctionBegin;
36   /* loop over colors  */
37   for (i=0; i<fd->ncolors; i++) {
38     for (j=0; j<fd->nrows[i]; j++) {
39       y    = fd->M - fd->rows[i][j] - fd->rstart;
40       x    = fd->columnsforrow[i][j];
41       ierr = PetscDrawRectangle(draw,x,y,x+1,y+1,i+1,i+1,i+1,i+1);CHKERRQ(ierr);
42     }
43   }
44   PetscFunctionReturn(0);
45 }
46 
47 #undef __FUNCT__
48 #define __FUNCT__ "MatFDColoringView_Draw"
49 static PetscErrorCode MatFDColoringView_Draw(MatFDColoring fd,PetscViewer viewer)
50 {
51   PetscErrorCode ierr;
52   PetscBool      isnull;
53   PetscDraw      draw;
54   PetscReal      xr,yr,xl,yl,h,w;
55 
56   PetscFunctionBegin;
57   ierr = PetscViewerDrawGetDraw(viewer,0,&draw);CHKERRQ(ierr);
58   ierr = PetscDrawIsNull(draw,&isnull);CHKERRQ(ierr); if (isnull) PetscFunctionReturn(0);
59 
60   ierr = PetscObjectCompose((PetscObject)fd,"Zoomviewer",(PetscObject)viewer);CHKERRQ(ierr);
61 
62   xr   = fd->N; yr  = fd->M; h = yr/10.0; w = xr/10.0;
63   xr  += w;     yr += h;    xl = -w;     yl = -h;
64   ierr = PetscDrawSetCoordinates(draw,xl,yl,xr,yr);CHKERRQ(ierr);
65   ierr = PetscDrawZoom(draw,MatFDColoringView_Draw_Zoom,fd);CHKERRQ(ierr);
66   ierr = PetscObjectCompose((PetscObject)fd,"Zoomviewer",NULL);CHKERRQ(ierr);
67   PetscFunctionReturn(0);
68 }
69 
70 #undef __FUNCT__
71 #define __FUNCT__ "MatFDColoringView"
72 /*@C
73    MatFDColoringView - Views a finite difference coloring context.
74 
75    Collective on MatFDColoring
76 
77    Input  Parameters:
78 +  c - the coloring context
79 -  viewer - visualization context
80 
81    Level: intermediate
82 
83    Notes:
84    The available visualization contexts include
85 +     PETSC_VIEWER_STDOUT_SELF - standard output (default)
86 .     PETSC_VIEWER_STDOUT_WORLD - synchronized standard
87         output where only the first processor opens
88         the file.  All other processors send their
89         data to the first processor to print.
90 -     PETSC_VIEWER_DRAW_WORLD - graphical display of nonzero structure
91 
92    Notes:
93      Since PETSc uses only a small number of basic colors (currently 33), if the coloring
94    involves more than 33 then some seemingly identical colors are displayed making it look
95    like an illegal coloring. This is just a graphical artifact.
96 
97 .seealso: MatFDColoringCreate()
98 
99 .keywords: Mat, finite differences, coloring, view
100 @*/
101 PetscErrorCode  MatFDColoringView(MatFDColoring c,PetscViewer viewer)
102 {
103   PetscErrorCode    ierr;
104   PetscInt          i,j;
105   PetscBool         isdraw,iascii;
106   PetscViewerFormat format;
107 
108   PetscFunctionBegin;
109   PetscValidHeaderSpecific(c,MAT_FDCOLORING_CLASSID,1);
110   if (!viewer) {
111     ierr = PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)c),&viewer);CHKERRQ(ierr);
112   }
113   PetscValidHeaderSpecific(viewer,PETSC_VIEWER_CLASSID,2);
114   PetscCheckSameComm(c,1,viewer,2);
115 
116   ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);CHKERRQ(ierr);
117   ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr);
118   if (isdraw) {
119     ierr = MatFDColoringView_Draw(c,viewer);CHKERRQ(ierr);
120   } else if (iascii) {
121     ierr = PetscObjectPrintClassNamePrefixType((PetscObject)c,viewer);CHKERRQ(ierr);
122     ierr = PetscViewerASCIIPrintf(viewer,"  Error tolerance=%G\n",c->error_rel);CHKERRQ(ierr);
123     ierr = PetscViewerASCIIPrintf(viewer,"  Umin=%G\n",c->umin);CHKERRQ(ierr);
124     ierr = PetscViewerASCIIPrintf(viewer,"  Number of colors=%D\n",c->ncolors);CHKERRQ(ierr);
125 
126     ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr);
127     if (format != PETSC_VIEWER_ASCII_INFO) {
128       for (i=0; i<c->ncolors; i++) {
129         ierr = PetscViewerASCIIPrintf(viewer,"  Information for color %D\n",i);CHKERRQ(ierr);
130         ierr = PetscViewerASCIIPrintf(viewer,"    Number of columns %D\n",c->ncolumns[i]);CHKERRQ(ierr);
131         for (j=0; j<c->ncolumns[i]; j++) {
132           ierr = PetscViewerASCIIPrintf(viewer,"      %D\n",c->columns[i][j]);CHKERRQ(ierr);
133         }
134         ierr = PetscViewerASCIIPrintf(viewer,"    Number of rows %D\n",c->nrows[i]);CHKERRQ(ierr);
135         for (j=0; j<c->nrows[i]; j++) {
136           ierr = PetscViewerASCIIPrintf(viewer,"      %D %D \n",c->rows[i][j],c->columnsforrow[i][j]);CHKERRQ(ierr);
137         }
138       }
139     }
140     ierr = PetscViewerFlush(viewer);CHKERRQ(ierr);
141   }
142   PetscFunctionReturn(0);
143 }
144 
145 #undef __FUNCT__
146 #define __FUNCT__ "MatFDColoringSetParameters"
147 /*@
148    MatFDColoringSetParameters - Sets the parameters for the sparse approximation of
149    a Jacobian matrix using finite differences.
150 
151    Logically Collective on MatFDColoring
152 
153    The Jacobian is estimated with the differencing approximation
154 .vb
155        F'(u)_{:,i} = [F(u+h*dx_{i}) - F(u)]/h where
156        h = error_rel*u[i]                 if  abs(u[i]) > umin
157          = +/- error_rel*umin             otherwise, with +/- determined by the sign of u[i]
158        dx_{i} = (0, ... 1, .... 0)
159 .ve
160 
161    Input Parameters:
162 +  coloring - the coloring context
163 .  error_rel - relative error
164 -  umin - minimum allowable u-value magnitude
165 
166    Level: advanced
167 
168 .keywords: Mat, finite differences, coloring, set, parameters
169 
170 .seealso: MatFDColoringCreate(), MatFDColoringSetFromOptions()
171 
172 @*/
173 PetscErrorCode  MatFDColoringSetParameters(MatFDColoring matfd,PetscReal error,PetscReal umin)
174 {
175   PetscFunctionBegin;
176   PetscValidHeaderSpecific(matfd,MAT_FDCOLORING_CLASSID,1);
177   PetscValidLogicalCollectiveReal(matfd,error,2);
178   PetscValidLogicalCollectiveReal(matfd,umin,3);
179   if (error != PETSC_DEFAULT) matfd->error_rel = error;
180   if (umin != PETSC_DEFAULT)  matfd->umin      = umin;
181   PetscFunctionReturn(0);
182 }
183 
184 
185 
186 #undef __FUNCT__
187 #define __FUNCT__ "MatFDColoringGetFunction"
188 /*@C
189    MatFDColoringGetFunction - Gets the function to use for computing the Jacobian.
190 
191    Not Collective
192 
193    Input Parameters:
194 .  coloring - the coloring context
195 
196    Output Parameters:
197 +  f - the function
198 -  fctx - the optional user-defined function context
199 
200    Level: intermediate
201 
202 .keywords: Mat, Jacobian, finite differences, set, function
203 
204 .seealso: MatFDColoringCreate(), MatFDColoringSetFunction(), MatFDColoringSetFromOptions()
205 
206 @*/
207 PetscErrorCode  MatFDColoringGetFunction(MatFDColoring matfd,PetscErrorCode (**f)(void),void **fctx)
208 {
209   PetscFunctionBegin;
210   PetscValidHeaderSpecific(matfd,MAT_FDCOLORING_CLASSID,1);
211   if (f) *f = matfd->f;
212   if (fctx) *fctx = matfd->fctx;
213   PetscFunctionReturn(0);
214 }
215 
216 #undef __FUNCT__
217 #define __FUNCT__ "MatFDColoringSetFunction"
218 /*@C
219    MatFDColoringSetFunction - Sets the function to use for computing the Jacobian.
220 
221    Logically Collective on MatFDColoring
222 
223    Input Parameters:
224 +  coloring - the coloring context
225 .  f - the function
226 -  fctx - the optional user-defined function context
227 
228    Calling sequence of (*f) function:
229     For SNES:    PetscErrorCode (*f)(SNES,Vec,Vec,void*)
230     If not using SNES: PetscErrorCode (*f)(void *dummy,Vec,Vec,void*) and dummy is ignored
231 
232    Level: advanced
233 
234    Notes: This function is usually used automatically by SNES (when one uses SNESSetJacobian() with the argument
235      SNESComputeJacobianDefaultColor()) and only needs to be used by someone computing a matrix via coloring directly by
236      calling MatFDColoringApply()
237 
238    Fortran Notes:
239     In Fortran you must call MatFDColoringSetFunction() for a coloring object to
240   be used without SNES or within the SNES solvers.
241 
242 .keywords: Mat, Jacobian, finite differences, set, function
243 
244 .seealso: MatFDColoringCreate(), MatFDColoringGetFunction(), MatFDColoringSetFromOptions()
245 
246 @*/
247 PetscErrorCode  MatFDColoringSetFunction(MatFDColoring matfd,PetscErrorCode (*f)(void),void *fctx)
248 {
249   PetscFunctionBegin;
250   PetscValidHeaderSpecific(matfd,MAT_FDCOLORING_CLASSID,1);
251   matfd->f    = f;
252   matfd->fctx = fctx;
253   PetscFunctionReturn(0);
254 }
255 
256 #undef __FUNCT__
257 #define __FUNCT__ "MatFDColoringSetFromOptions"
258 /*@
259    MatFDColoringSetFromOptions - Sets coloring finite difference parameters from
260    the options database.
261 
262    Collective on MatFDColoring
263 
264    The Jacobian, F'(u), is estimated with the differencing approximation
265 .vb
266        F'(u)_{:,i} = [F(u+h*dx_{i}) - F(u)]/h where
267        h = error_rel*u[i]                 if  abs(u[i]) > umin
268          = +/- error_rel*umin             otherwise, with +/- determined by the sign of u[i]
269        dx_{i} = (0, ... 1, .... 0)
270 .ve
271 
272    Input Parameter:
273 .  coloring - the coloring context
274 
275    Options Database Keys:
276 +  -mat_fd_coloring_err <err> - Sets <err> (square root
277            of relative error in the function)
278 .  -mat_fd_coloring_umin <umin> - Sets umin, the minimum allowable u-value magnitude
279 .  -mat_fd_type - "wp" or "ds" (see MATMFFD_WP or MATMFFD_DS)
280 .  -mat_fd_coloring_view - Activates basic viewing
281 .  -mat_fd_coloring_view ::ascii_info - Activates viewing info
282 -  -mat_fd_coloring_view draw - Activates drawing
283 
284     Level: intermediate
285 
286 .keywords: Mat, finite differences, parameters
287 
288 .seealso: MatFDColoringCreate(), MatFDColoringView(), MatFDColoringSetParameters()
289 
290 @*/
291 PetscErrorCode  MatFDColoringSetFromOptions(MatFDColoring matfd)
292 {
293   PetscErrorCode ierr;
294   PetscBool      flg;
295   char           value[3];
296 
297   PetscFunctionBegin;
298   PetscValidHeaderSpecific(matfd,MAT_FDCOLORING_CLASSID,1);
299 
300   ierr = PetscObjectOptionsBegin((PetscObject)matfd);CHKERRQ(ierr);
301   ierr = PetscOptionsReal("-mat_fd_coloring_err","Square root of relative error in function","MatFDColoringSetParameters",matfd->error_rel,&matfd->error_rel,0);CHKERRQ(ierr);
302   ierr = PetscOptionsReal("-mat_fd_coloring_umin","Minimum allowable u magnitude","MatFDColoringSetParameters",matfd->umin,&matfd->umin,0);CHKERRQ(ierr);
303   ierr = PetscOptionsString("-mat_fd_type","Algorithm to compute h, wp or ds","MatFDColoringCreate",matfd->htype,value,3,&flg);CHKERRQ(ierr);
304   if (flg) {
305     if (value[0] == 'w' && value[1] == 'p') matfd->htype = "wp";
306     else if (value[0] == 'd' && value[1] == 's') matfd->htype = "ds";
307     else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Unknown finite differencing type %s",value);
308   }
309   /* process any options handlers added with PetscObjectAddOptionsHandler() */
310   ierr = PetscObjectProcessOptionsHandlers((PetscObject)matfd);CHKERRQ(ierr);
311   PetscOptionsEnd();CHKERRQ(ierr);
312   PetscFunctionReturn(0);
313 }
314 
315 #undef __FUNCT__
316 #define __FUNCT__ "MatFDColoringViewFromOptions"
317 PetscErrorCode MatFDColoringViewFromOptions(MatFDColoring fd,const char prefix[],const char optionname[])
318 {
319   PetscErrorCode    ierr;
320   PetscBool         flg;
321   PetscViewer       viewer;
322   PetscViewerFormat format;
323 
324   PetscFunctionBegin;
325   if (prefix) {
326     ierr = PetscOptionsGetViewer(PetscObjectComm((PetscObject)fd),prefix,optionname,&viewer,&format,&flg);CHKERRQ(ierr);
327   } else {
328     ierr = PetscOptionsGetViewer(PetscObjectComm((PetscObject)fd),((PetscObject)fd)->prefix,optionname,&viewer,&format,&flg);CHKERRQ(ierr);
329   }
330   if (flg) {
331     ierr = PetscViewerPushFormat(viewer,format);CHKERRQ(ierr);
332     ierr = MatFDColoringView(fd,viewer);CHKERRQ(ierr);
333     ierr = PetscViewerPopFormat(viewer);CHKERRQ(ierr);
334     ierr = PetscViewerDestroy(&viewer);CHKERRQ(ierr);
335   }
336   PetscFunctionReturn(0);
337 }
338 
339 #undef __FUNCT__
340 #define __FUNCT__ "MatFDColoringCreate"
341 /*@
342    MatFDColoringCreate - Creates a matrix coloring context for finite difference
343    computation of Jacobians.
344 
345    Collective on Mat
346 
347    Input Parameters:
348 +  mat - the matrix containing the nonzero structure of the Jacobian
349 -  iscoloring - the coloring of the matrix; usually obtained with MatGetColoring() or DMCreateColoring()
350 
351     Output Parameter:
352 .   color - the new coloring context
353 
354     Level: intermediate
355 
356 .seealso: MatFDColoringDestroy(),SNESComputeJacobianDefaultColor(), ISColoringCreate(),
357           MatFDColoringSetFunction(), MatFDColoringSetFromOptions(), MatFDColoringApply(),
358           MatFDColoringView(), MatFDColoringSetParameters(), MatGetColoring(), DMCreateColoring()
359 @*/
360 PetscErrorCode  MatFDColoringCreate(Mat mat,ISColoring iscoloring,MatFDColoring *color)
361 {
362   MatFDColoring  c;
363   MPI_Comm       comm;
364   PetscErrorCode ierr;
365   PetscInt       M,N;
366 
367   PetscFunctionBegin;
368   if (!mat->assembled) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_ARG_WRONGSTATE,"Matrix must be assembled by calls to MatAssemblyBegin/End();");
369   ierr = PetscLogEventBegin(MAT_FDColoringCreate,mat,0,0,0);CHKERRQ(ierr);
370   ierr = MatGetSize(mat,&M,&N);CHKERRQ(ierr);
371   if (M != N) SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"Only for square matrices");
372   ierr = PetscObjectGetComm((PetscObject)mat,&comm);CHKERRQ(ierr);
373   ierr = PetscHeaderCreate(c,_p_MatFDColoring,int,MAT_FDCOLORING_CLASSID,"MatFDColoring","Jacobian computation via finite differences with coloring","Mat",comm,MatFDColoringDestroy,MatFDColoringView);CHKERRQ(ierr);
374 
375   c->ctype = iscoloring->ctype;
376 
377   if (mat->ops->fdcoloringcreate) {
378     ierr = (*mat->ops->fdcoloringcreate)(mat,iscoloring,c);CHKERRQ(ierr);
379   } else SETERRQ1(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"Code not yet written for matrix type %s",((PetscObject)mat)->type_name);
380 
381   ierr = MatGetVecs(mat,NULL,&c->w1);CHKERRQ(ierr);
382   ierr = PetscLogObjectParent((PetscObject)c,(PetscObject)c->w1);CHKERRQ(ierr);
383   ierr = VecDuplicate(c->w1,&c->w2);CHKERRQ(ierr);
384   ierr = PetscLogObjectParent((PetscObject)c,(PetscObject)c->w2);CHKERRQ(ierr);
385 
386   c->error_rel    = PETSC_SQRT_MACHINE_EPSILON;
387   c->umin         = 100.0*PETSC_SQRT_MACHINE_EPSILON;
388   c->currentcolor = -1;
389   c->htype        = "wp";
390   c->fset         = PETSC_FALSE;
391 
392   *color = c;
393   ierr   = PetscObjectCompose((PetscObject)mat,"SNESMatFDColoring",(PetscObject)c);CHKERRQ(ierr);
394   ierr   = PetscLogEventEnd(MAT_FDColoringCreate,mat,0,0,0);CHKERRQ(ierr);
395   PetscFunctionReturn(0);
396 }
397 
398 #undef __FUNCT__
399 #define __FUNCT__ "MatFDColoringDestroy"
400 /*@
401     MatFDColoringDestroy - Destroys a matrix coloring context that was created
402     via MatFDColoringCreate().
403 
404     Collective on MatFDColoring
405 
406     Input Parameter:
407 .   c - coloring context
408 
409     Level: intermediate
410 
411 .seealso: MatFDColoringCreate()
412 @*/
413 PetscErrorCode  MatFDColoringDestroy(MatFDColoring *c)
414 {
415   PetscErrorCode ierr;
416   PetscInt       i;
417 
418   PetscFunctionBegin;
419   if (!*c) PetscFunctionReturn(0);
420   if (--((PetscObject)(*c))->refct > 0) {*c = 0; PetscFunctionReturn(0);}
421 
422   for (i=0; i<(*c)->ncolors; i++) {
423     ierr = PetscFree((*c)->columns[i]);CHKERRQ(ierr);
424     if ((*c)->rows) {
425       ierr = PetscFree((*c)->rows[i]);CHKERRQ(ierr);
426       ierr = PetscFree((*c)->columnsforrow[i]);CHKERRQ(ierr);
427     }
428     if ((*c)->vscaleforrow) {ierr = PetscFree((*c)->vscaleforrow[i]);CHKERRQ(ierr);}
429   }
430   ierr = PetscFree((*c)->ncolumns);CHKERRQ(ierr);
431   ierr = PetscFree((*c)->columns);CHKERRQ(ierr);
432   ierr = PetscFree((*c)->nrows);CHKERRQ(ierr);
433   if ((*c)->rows) {
434     ierr = PetscFree((*c)->rows);CHKERRQ(ierr);
435     ierr = PetscFree((*c)->columnsforrow);CHKERRQ(ierr);
436   } else {
437     ierr = PetscFree((*c)->rowcolden2sp3);CHKERRQ(ierr);
438   }
439   if ((*c)->matentry) {
440     ierr = PetscFree((*c)->matentry);CHKERRQ(ierr);
441   }
442   ierr = PetscFree((*c)->valaddr);CHKERRQ(ierr);
443   ierr = PetscFree((*c)->vscaleforrow);CHKERRQ(ierr);
444   ierr = PetscFree((*c)->dy);CHKERRQ(ierr);
445   ierr = VecDestroy(&(*c)->vscale);CHKERRQ(ierr);
446   ierr = VecDestroy(&(*c)->w1);CHKERRQ(ierr);
447   ierr = VecDestroy(&(*c)->w2);CHKERRQ(ierr);
448   ierr = VecDestroy(&(*c)->w3);CHKERRQ(ierr);
449   ierr = PetscHeaderDestroy(c);CHKERRQ(ierr);
450   PetscFunctionReturn(0);
451 }
452 
453 #undef __FUNCT__
454 #define __FUNCT__ "MatFDColoringGetPerturbedColumns"
455 /*@C
456     MatFDColoringGetPerturbedColumns - Returns the indices of the columns that
457       that are currently being perturbed.
458 
459     Not Collective
460 
461     Input Parameters:
462 .   coloring - coloring context created with MatFDColoringCreate()
463 
464     Output Parameters:
465 +   n - the number of local columns being perturbed
466 -   cols - the column indices, in global numbering
467 
468    Level: intermediate
469 
470 .seealso: MatFDColoringCreate(), MatFDColoringDestroy(), MatFDColoringView(), MatFDColoringApply()
471 
472 .keywords: coloring, Jacobian, finite differences
473 @*/
474 PetscErrorCode  MatFDColoringGetPerturbedColumns(MatFDColoring coloring,PetscInt *n,PetscInt *cols[])
475 {
476   PetscFunctionBegin;
477   if (coloring->currentcolor >= 0) {
478     *n    = coloring->ncolumns[coloring->currentcolor];
479     *cols = coloring->columns[coloring->currentcolor];
480   } else {
481     *n = 0;
482   }
483   PetscFunctionReturn(0);
484 }
485 
486 #undef __FUNCT__
487 #define __FUNCT__ "MatFDColoringApply"
488 /*@
489     MatFDColoringApply - Given a matrix for which a MatFDColoring context
490     has been created, computes the Jacobian for a function via finite differences.
491 
492     Collective on MatFDColoring
493 
494     Input Parameters:
495 +   mat - location to store Jacobian
496 .   coloring - coloring context created with MatFDColoringCreate()
497 .   x1 - location at which Jacobian is to be computed
498 -   sctx - context required by function, if this is being used with the SNES solver then it is SNES object, otherwise it is null
499 
500     Options Database Keys:
501 +    -mat_fd_type - "wp" or "ds"  (see MATMFFD_WP or MATMFFD_DS)
502 .    -mat_fd_coloring_view - Activates basic viewing or coloring
503 .    -mat_fd_coloring_view draw - Activates drawing of coloring
504 -    -mat_fd_coloring_view ::ascii_info - Activates viewing of coloring info
505 
506     Level: intermediate
507 
508 .seealso: MatFDColoringCreate(), MatFDColoringDestroy(), MatFDColoringView(), MatFDColoringSetFunction()
509 
510 .keywords: coloring, Jacobian, finite differences
511 @*/
512 PetscErrorCode  MatFDColoringApply(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx)
513 {
514   PetscErrorCode ierr;
515 
516   PetscFunctionBegin;
517   PetscValidHeaderSpecific(J,MAT_CLASSID,1);
518   PetscValidHeaderSpecific(coloring,MAT_FDCOLORING_CLASSID,2);
519   PetscValidHeaderSpecific(x1,VEC_CLASSID,3);
520   if (!coloring->f) SETERRQ(PetscObjectComm((PetscObject)J),PETSC_ERR_ARG_WRONGSTATE,"Must call MatFDColoringSetFunction()");
521   if (!J->ops->fdcoloringapply) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not supported for this matrix type %s",((PetscObject)J)->type_name);
522   ierr = PetscLogEventBegin(MAT_FDColoringApply,coloring,J,x1,0);CHKERRQ(ierr);
523   ierr = (*J->ops->fdcoloringapply)(J,coloring,x1,flag,sctx);CHKERRQ(ierr);
524   ierr = PetscLogEventEnd(MAT_FDColoringApply,coloring,J,x1,0);CHKERRQ(ierr);
525   PetscFunctionReturn(0);
526 }
527 
528 #undef __FUNCT__
529 #define __FUNCT__ "MatFDColoringApply_AIJ"
530 PetscErrorCode  MatFDColoringApply_AIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx)
531 {
532   PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void*))coloring->f;
533   PetscErrorCode ierr;
534   PetscInt       k,start,end,l,row,col,srow,**vscaleforrow;
535   PetscScalar    dx,*y,*xx,*w3_array;
536   PetscScalar    *vscale_array;
537   PetscReal      epsilon = coloring->error_rel,umin = coloring->umin,unorm;
538   Vec            w1      = coloring->w1,w2=coloring->w2,w3;
539   void           *fctx   = coloring->fctx;
540   PetscBool      flg     = PETSC_FALSE;
541   PetscInt       ctype   = coloring->ctype,N,col_start=0,col_end=0;
542   Vec            x1_tmp;
543 
544   PetscFunctionBegin;
545   ierr = MatSetUnfactored(J);CHKERRQ(ierr);
546   ierr = PetscOptionsGetBool(NULL,"-mat_fd_coloring_dont_rezero",&flg,NULL);CHKERRQ(ierr);
547   if (flg) {
548     ierr = PetscInfo(coloring,"Not calling MatZeroEntries()\n");CHKERRQ(ierr);
549   } else {
550     PetscBool assembled;
551     ierr = MatAssembled(J,&assembled);CHKERRQ(ierr);
552     if (assembled) {
553       ierr = MatZeroEntries(J);CHKERRQ(ierr);
554     }
555   }
556 
557   x1_tmp = x1;
558   if (!coloring->vscale) {
559     ierr = VecDuplicate(x1_tmp,&coloring->vscale);CHKERRQ(ierr);
560   }
561 
562   if (coloring->htype[0] == 'w') { /* tacky test; need to make systematic if we add other approaches to computing h*/
563     ierr = VecNorm(x1_tmp,NORM_2,&unorm);CHKERRQ(ierr);
564   }
565   ierr = VecGetOwnershipRange(w1,&start,&end);CHKERRQ(ierr); /* OwnershipRange is used by ghosted x! */
566 
567   /* Set w1 = F(x1) */
568   if (!coloring->fset) {
569     ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
570     ierr = (*f)(sctx,x1_tmp,w1,fctx);CHKERRQ(ierr);
571     ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
572   } else {
573     coloring->fset = PETSC_FALSE;
574   }
575 
576   if (!coloring->w3) {
577     ierr = VecDuplicate(x1_tmp,&coloring->w3);CHKERRQ(ierr);
578     ierr = PetscLogObjectParent((PetscObject)coloring,(PetscObject)coloring->w3);CHKERRQ(ierr);
579   }
580   w3 = coloring->w3;
581 
582   /* Compute all the local scale factors, including ghost points */
583   ierr = VecGetLocalSize(x1_tmp,&N);CHKERRQ(ierr);
584   ierr = VecGetArray(x1_tmp,&xx);CHKERRQ(ierr);
585   ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);
586   if (ctype == IS_COLORING_GHOSTED) {
587     col_start = 0; col_end = N;
588   } else if (ctype == IS_COLORING_GLOBAL) {
589     xx           = xx - start;
590     vscale_array = vscale_array - start;
591     col_start    = start; col_end = N + start;
592   }
593   for (col=col_start; col<col_end; col++) {
594     /* Loop over each local column, vscale[col] = 1./(epsilon*dx[col]) */
595     if (coloring->htype[0] == 'w') {
596       dx = 1.0 + unorm;
597     } else {
598       dx = xx[col];
599     }
600     if (dx == (PetscScalar)0.0) dx = 1.0;
601     if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0)     dx = umin;
602     else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin;
603     dx               *= epsilon;
604     vscale_array[col] = (PetscScalar)1.0/dx;
605   }
606   if (ctype == IS_COLORING_GLOBAL) vscale_array = vscale_array + start;
607   ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);
608   if (ctype == IS_COLORING_GLOBAL) {
609     ierr = VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
610     ierr = VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
611   }
612 
613   if (coloring->vscaleforrow) {
614     vscaleforrow = coloring->vscaleforrow;
615   } else SETERRQ(PetscObjectComm((PetscObject)J),PETSC_ERR_ARG_NULL,"Null Object: coloring->vscaleforrow");
616 
617   /*
618     Loop over each color
619   */
620   ierr = VecGetArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);
621   for (k=0; k<coloring->ncolors; k++) {
622     coloring->currentcolor = k;
623 
624     ierr = VecCopy(x1_tmp,w3);CHKERRQ(ierr);
625     ierr = VecGetArray(w3,&w3_array);CHKERRQ(ierr);
626     if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array - start;
627     /*
628       Loop over each column associated with color
629       adding the perturbation to the vector w3.
630     */
631     for (l=0; l<coloring->ncolumns[k]; l++) {
632       col = coloring->columns[k][l];    /* local column of the matrix we are probing for */
633       if (coloring->htype[0] == 'w') {
634         dx = 1.0 + unorm;
635       } else {
636         dx = xx[col];
637       }
638       if (dx == (PetscScalar)0.0) dx = 1.0;
639       if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0)     dx = umin;
640       else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin;
641       dx            *= epsilon;
642       if (!PetscAbsScalar(dx)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Computed 0 differencing parameter");
643       w3_array[col] += dx;
644     }
645     if (ctype == IS_COLORING_GLOBAL) w3_array = w3_array + start;
646     ierr = VecRestoreArray(w3,&w3_array);CHKERRQ(ierr);
647 
648     /*
649       Evaluate function at w3 = x1 + dx (here dx is a vector of perturbations)
650                            w2 = F(x1 + dx) - F(x1)
651     */
652     ierr = PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
653     ierr = (*f)(sctx,w3,w2,fctx);CHKERRQ(ierr);
654     ierr = PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);CHKERRQ(ierr);
655     ierr = VecAXPY(w2,-1.0,w1);CHKERRQ(ierr);
656 
657     /*
658       Loop over rows of vector, putting results into Jacobian matrix
659     */
660     ierr = VecGetArray(w2,&y);CHKERRQ(ierr);
661     for (l=0; l<coloring->nrows[k]; l++) {
662       row     = coloring->rows[k][l];            /* local row index */
663       col     = coloring->columnsforrow[k][l];   /* global column index */
664       y[row] *= vscale_array[vscaleforrow[k][l]];
665       srow    = row + start;
666       ierr    = MatSetValues(J,1,&srow,1,&col,y+row,INSERT_VALUES);CHKERRQ(ierr);
667     }
668     ierr = VecRestoreArray(w2,&y);CHKERRQ(ierr);
669   } /* endof for each color */
670 
671   if (ctype == IS_COLORING_GLOBAL) xx = xx + start;
672   ierr = VecRestoreArray(coloring->vscale,&vscale_array);CHKERRQ(ierr);
673   ierr = VecRestoreArray(x1_tmp,&xx);CHKERRQ(ierr);
674 
675   coloring->currentcolor = -1;
676 
677   ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
678   ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
679 
680   ierr = MatFDColoringViewFromOptions(coloring,NULL,"-mat_fd_coloring_view");CHKERRQ(ierr);
681   PetscFunctionReturn(0);
682 }
683