Actual source code: sbaijcholmod.c
petsc-3.3-p7 2013-05-11
2: /*
3: Provides an interface to the CHOLMOD 1.7.1 sparse solver
5: When build with PETSC_USE_64BIT_INDICES this will use UF_Long as the
6: integer type in UMFPACK, otherwise it will use int. This means
7: all integers in this file as simply declared as PetscInt. Also it means
8: that UMFPACK UL_Long version MUST be built with 64 bit integers when used.
10: */
12: #include <../src/mat/impls/sbaij/seq/sbaij.h>
13: #include <../src/mat/impls/sbaij/seq/cholmod/cholmodimpl.h>
15: /*
16: This is a terrible hack, but it allows the error handler to retain a context.
17: Note that this hack really cannot be made both reentrant and concurrent.
18: */
19: static Mat static_F;
23: static void CholmodErrorHandler(int status,const char *file,int line,const char *message)
24: {
27: if (status > CHOLMOD_OK) {
28: PetscInfo4(static_F,"CHOLMOD warning %d at %s:%d: %s",status,file,line,message);
29: } else if (status == CHOLMOD_OK) { /* Documentation says this can happen, but why? */
30: PetscInfo3(static_F,"CHOLMOD OK at %s:%d: %s",file,line,message);
31: } else {
32: PetscErrorPrintf("CHOLMOD error %d at %s:%d: %s\n",status,file,line,message);
33: }
34: PetscFunctionReturnVoid();
35: }
39: PetscErrorCode CholmodStart(Mat F)
40: {
42: Mat_CHOLMOD *chol=(Mat_CHOLMOD*)F->spptr;
43: cholmod_common *c;
44: PetscBool flg;
47: if (chol->common) return(0);
48: PetscMalloc(sizeof(*chol->common),&chol->common);
49: !cholmod_X_start(chol->common);
50: c = chol->common;
51: c->error_handler = CholmodErrorHandler;
53: #define CHOLMOD_OPTION_DOUBLE(name,help) do { \
54: PetscReal tmp = (PetscReal)c->name; \
55: PetscOptionsReal("-mat_cholmod_" #name,help,"None",tmp,&tmp,0); \
56: c->name = (double)tmp; \
57: } while (0)
58: #define CHOLMOD_OPTION_INT(name,help) do { \
59: PetscInt tmp = (PetscInt)c->name; \
60: PetscOptionsInt("-mat_cholmod_" #name,help,"None",tmp,&tmp,0); \
61: c->name = (int)tmp; \
62: } while (0)
63: #define CHOLMOD_OPTION_SIZE_T(name,help) do { \
64: PetscInt tmp = (PetscInt)c->name; \
65: PetscOptionsInt("-mat_cholmod_" #name,help,"None",tmp,&tmp,0); \
66: if (tmp < 0) SETERRQ(((PetscObject)F)->comm,PETSC_ERR_ARG_OUTOFRANGE,"value must be positive"); \
67: c->name = (size_t)tmp; \
68: } while (0)
69: #define CHOLMOD_OPTION_TRUTH(name,help) do { \
70: PetscBool tmp = (PetscBool)!!c->name; \
71: PetscOptionsBool("-mat_cholmod_" #name,help,"None",tmp,&tmp,0); \
72: c->name = (int)tmp; \
73: } while (0)
75: PetscOptionsBegin(((PetscObject)F)->comm,((PetscObject)F)->prefix,"CHOLMOD Options","Mat");
76: /* CHOLMOD handles first-time packing and refactor-packing separately, but we usually want them to be the same. */
77: chol->pack = (PetscBool)c->final_pack;
78: PetscOptionsBool("-mat_cholmod_pack","Pack factors after factorization [disable for frequent repeat factorization]","None",chol->pack,&chol->pack,0);
79: c->final_pack = (int)chol->pack;
81: CHOLMOD_OPTION_DOUBLE(dbound,"Minimum absolute value of diagonal entries of D");
82: CHOLMOD_OPTION_DOUBLE(grow0,"Global growth ratio when factors are modified");
83: CHOLMOD_OPTION_DOUBLE(grow1,"Column growth ratio when factors are modified");
84: CHOLMOD_OPTION_SIZE_T(grow2,"Affine column growth constant when factors are modified");
85: CHOLMOD_OPTION_SIZE_T(maxrank,"Max rank of update, larger values are faster but use more memory [2,4,8]");
86: {
87: static const char *const list[] = {"SIMPLICIAL","AUTO","SUPERNODAL","MatCholmodFactorType","MAT_CHOLMOD_FACTOR_",0};
88: PetscEnum choice = (PetscEnum)c->supernodal;
89: PetscOptionsEnum("-mat_cholmod_factor","Factorization method","None",list,(PetscEnum)c->supernodal,&choice,0);
90: c->supernodal = (int)choice;
91: }
92: if (c->supernodal) CHOLMOD_OPTION_DOUBLE(supernodal_switch,"flop/nnz_L threshold for switching to supernodal factorization");
93: CHOLMOD_OPTION_TRUTH(final_asis,"Leave factors \"as is\"");
94: CHOLMOD_OPTION_TRUTH(final_pack,"Pack the columns when finished (use FALSE if the factors will be updated later)");
95: if (!c->final_asis) {
96: CHOLMOD_OPTION_TRUTH(final_super,"Leave supernodal factors instead of converting to simplicial");
97: CHOLMOD_OPTION_TRUTH(final_ll,"Turn LDL' factorization into LL'");
98: CHOLMOD_OPTION_TRUTH(final_monotonic,"Ensure columns are monotonic when done");
99: CHOLMOD_OPTION_TRUTH(final_resymbol,"Remove numerically zero values resulting from relaxed supernodal amalgamation");
100: }
101: {
102: PetscReal tmp[] = {(PetscReal)c->zrelax[0],(PetscReal)c->zrelax[1],(PetscReal)c->zrelax[2]};
103: PetscInt n = 3;
104: PetscOptionsRealArray("-mat_cholmod_zrelax","3 real supernodal relaxed amalgamation parameters","None",tmp,&n,&flg);
105: if (flg && n != 3) SETERRQ(((PetscObject)F)->comm,PETSC_ERR_ARG_OUTOFRANGE,"must provide exactly 3 parameters to -mat_cholmod_zrelax");
106: if (flg) while (n--) c->zrelax[n] = (double)tmp[n];
107: }
108: {
109: PetscInt n,tmp[] = {(PetscInt)c->nrelax[0],(PetscInt)c->nrelax[1],(PetscInt)c->nrelax[2]};
110: PetscOptionsIntArray("-mat_cholmod_nrelax","3 size_t supernodal relaxed amalgamation parameters","None",tmp,&n,&flg);
111: if (flg && n != 3) SETERRQ(((PetscObject)F)->comm,PETSC_ERR_ARG_OUTOFRANGE,"must provide exactly 3 parameters to -mat_cholmod_nrelax");
112: if (flg) while (n--) c->nrelax[n] = (size_t)tmp[n];
113: }
114: CHOLMOD_OPTION_TRUTH(prefer_upper,"Work with upper triangular form [faster when using fill-reducing ordering, slower in natural ordering]");
115: CHOLMOD_OPTION_TRUTH(default_nesdis,"Use NESDIS instead of METIS for nested dissection");
116: CHOLMOD_OPTION_INT(print,"Verbosity level");
117: PetscOptionsEnd();
118: return(0);
119: }
123: static PetscErrorCode MatWrapCholmod_seqsbaij(Mat A,PetscBool values,cholmod_sparse *C,PetscBool *aijalloc)
124: {
125: Mat_SeqSBAIJ *sbaij = (Mat_SeqSBAIJ*)A->data;
129: PetscMemzero(C,sizeof(*C));
130: /* CHOLMOD uses column alignment, SBAIJ stores the upper factor, so we pass it on as a lower factor, swapping the meaning of row and column */
131: C->nrow = (size_t)A->cmap->n;
132: C->ncol = (size_t)A->rmap->n;
133: C->nzmax = (size_t)sbaij->maxnz;
134: C->p = sbaij->i;
135: C->i = sbaij->j;
136: C->x = sbaij->a;
137: C->stype = -1;
138: C->itype = CHOLMOD_INT_TYPE;
139: C->xtype = CHOLMOD_SCALAR_TYPE;
140: C->dtype = CHOLMOD_DOUBLE;
141: C->sorted = 1;
142: C->packed = 1;
143: *aijalloc = PETSC_FALSE;
144: return(0);
145: }
149: static PetscErrorCode VecWrapCholmod(Vec X,cholmod_dense *Y)
150: {
152: PetscScalar *x;
153: PetscInt n;
156: PetscMemzero(Y,sizeof(*Y));
157: VecGetArray(X,&x);
158: VecGetSize(X,&n);
159: Y->x = (double*)x;
160: Y->nrow = n;
161: Y->ncol = 1;
162: Y->nzmax = n;
163: Y->d = n;
164: Y->x = (double*)x;
165: Y->xtype = CHOLMOD_SCALAR_TYPE;
166: Y->dtype = CHOLMOD_DOUBLE;
167: return(0);
168: }
172: PetscErrorCode MatDestroy_CHOLMOD(Mat F)
173: {
175: Mat_CHOLMOD *chol=(Mat_CHOLMOD*)F->spptr;
178: if (chol) {
179: !cholmod_X_free_factor(&chol->factor,chol->common);
180: !cholmod_X_finish(chol->common);
181: PetscFree(chol->common);
182: PetscFree(chol->matrix);
183: (*chol->Destroy)(F);
184: }
185: PetscFree(F->spptr);
186: return(0);
187: }
189: static PetscErrorCode MatSolve_CHOLMOD(Mat,Vec,Vec);
191: static const char *const CholmodOrderingMethods[] = {"User","AMD","METIS","NESDIS(default)","Natural","NESDIS(small=20000)","NESDIS(small=4,no constrained)","NESDIS()"};
195: static PetscErrorCode MatFactorInfo_CHOLMOD(Mat F,PetscViewer viewer)
196: {
197: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->spptr;
198: const cholmod_common *c = chol->common;
200: PetscInt i;
203: if (F->ops->solve != MatSolve_CHOLMOD) return(0);
204: PetscViewerASCIIPrintf(viewer,"CHOLMOD run parameters:\n");
205: PetscViewerASCIIPushTab(viewer);
206: PetscViewerASCIIPrintf(viewer,"Pack factors after symbolic factorization: %s\n",chol->pack?"TRUE":"FALSE");
207: PetscViewerASCIIPrintf(viewer,"Common.dbound %g (Smallest absolute value of diagonal entries of D)\n",c->dbound);
208: PetscViewerASCIIPrintf(viewer,"Common.grow0 %g\n",c->grow0);
209: PetscViewerASCIIPrintf(viewer,"Common.grow1 %g\n",c->grow1);
210: PetscViewerASCIIPrintf(viewer,"Common.grow2 %u\n",(unsigned)c->grow2);
211: PetscViewerASCIIPrintf(viewer,"Common.maxrank %u\n",(unsigned)c->maxrank);
212: PetscViewerASCIIPrintf(viewer,"Common.supernodal_switch %g\n",c->supernodal_switch);
213: PetscViewerASCIIPrintf(viewer,"Common.supernodal %d\n",c->supernodal);
214: PetscViewerASCIIPrintf(viewer,"Common.final_asis %d\n",c->final_asis);
215: PetscViewerASCIIPrintf(viewer,"Common.final_super %d\n",c->final_super);
216: PetscViewerASCIIPrintf(viewer,"Common.final_ll %d\n",c->final_ll);
217: PetscViewerASCIIPrintf(viewer,"Common.final_pack %d\n",c->final_pack);
218: PetscViewerASCIIPrintf(viewer,"Common.final_monotonic %d\n",c->final_monotonic);
219: PetscViewerASCIIPrintf(viewer,"Common.final_resymbol %d\n",c->final_resymbol);
220: PetscViewerASCIIPrintf(viewer,"Common.zrelax [%g,%g,%g]\n",c->zrelax[0],c->zrelax[1],c->zrelax[2]);
221: PetscViewerASCIIPrintf(viewer,"Common.nrelax [%u,%u,%u]\n",(unsigned)c->nrelax[0],(unsigned)c->nrelax[1],(unsigned)c->nrelax[2]);
222: PetscViewerASCIIPrintf(viewer,"Common.prefer_upper %d\n",c->prefer_upper);
223: PetscViewerASCIIPrintf(viewer,"Common.print %d\n",c->print);
224: for (i=0; i<c->nmethods; i++) {
225: PetscViewerASCIIPrintf(viewer,"Ordering method %D%s:\n",i,i==c->selected?" [SELECTED]":"");
226: PetscViewerASCIIPrintf(viewer," lnz %g, fl %g, prune_dense %g, prune_dense2 %g\n",
227: c->method[i].lnz,c->method[i].fl,c->method[i].prune_dense,c->method[i].prune_dense2);
228: }
229: PetscViewerASCIIPrintf(viewer,"Common.postorder %d\n",c->postorder);
230: PetscViewerASCIIPrintf(viewer,"Common.default_nesdis %d (use NESDIS instead of METIS for nested dissection)\n",c->default_nesdis);
231: /* Statistics */
232: PetscViewerASCIIPrintf(viewer,"Common.fl %g (flop count from most recent analysis)\n",c->fl);
233: PetscViewerASCIIPrintf(viewer,"Common.lnz %g (fundamental nz in L)\n",c->lnz);
234: PetscViewerASCIIPrintf(viewer,"Common.anz %g\n",c->anz);
235: PetscViewerASCIIPrintf(viewer,"Common.modfl %g (flop count from most recent update)\n",c->modfl);
236: PetscViewerASCIIPrintf(viewer,"Common.malloc_count %g (number of live objects)\n",(double)c->malloc_count);
237: PetscViewerASCIIPrintf(viewer,"Common.memory_usage %g (peak memory usage in bytes)\n",(double)c->memory_usage);
238: PetscViewerASCIIPrintf(viewer,"Common.memory_inuse %g (current memory usage in bytes)\n",(double)c->memory_inuse);
239: PetscViewerASCIIPrintf(viewer,"Common.nrealloc_col %g (number of column reallocations)\n",c->nrealloc_col);
240: PetscViewerASCIIPrintf(viewer,"Common.nrealloc_factor %g (number of factor reallocations due to column reallocations)\n",c->nrealloc_factor);
241: PetscViewerASCIIPrintf(viewer,"Common.ndbounds_hit %g (number of times diagonal was modified by dbound)\n",c->ndbounds_hit);
242: PetscViewerASCIIPrintf(viewer,"Common.rowfacfl %g (number of flops in last call to cholmod_rowfac)\n",c->rowfacfl);
243: PetscViewerASCIIPrintf(viewer,"Common.aatfl %g (number of flops to compute A(:,f)*A(:,f)')\n",c->aatfl);
244: PetscViewerASCIIPopTab(viewer);
245: return(0);
246: }
250: PetscErrorCode MatView_CHOLMOD(Mat F,PetscViewer viewer)
251: {
252: PetscErrorCode ierr;
253: PetscBool iascii;
254: PetscViewerFormat format;
257: MatView_SeqSBAIJ(F,viewer);
258: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
259: if (iascii) {
260: PetscViewerGetFormat(viewer,&format);
261: if (format == PETSC_VIEWER_ASCII_INFO) {
262: MatFactorInfo_CHOLMOD(F,viewer);
263: }
264: }
265: return(0);
266: }
270: static PetscErrorCode MatSolve_CHOLMOD(Mat F,Vec B,Vec X)
271: {
272: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->spptr;
273: cholmod_dense cholB,*cholX;
274: PetscScalar *x;
278: VecWrapCholmod(B,&cholB);
279: static_F = F;
280: cholX = cholmod_X_solve(CHOLMOD_A,chol->factor,&cholB,chol->common);
281: if (!cholX) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"CHOLMOD failed");
282: VecGetArray(X,&x);
283: PetscMemcpy(x,cholX->x,cholX->nrow*sizeof(*x));
284: !cholmod_X_free_dense(&cholX,chol->common);
285: VecRestoreArray(X,&x);
286: return(0);
287: }
291: static PetscErrorCode MatCholeskyFactorNumeric_CHOLMOD(Mat F,Mat A,const MatFactorInfo *info)
292: {
293: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->spptr;
294: cholmod_sparse cholA;
295: PetscBool aijalloc;
299: (*chol->Wrap)(A,PETSC_TRUE,&cholA,&aijalloc);
300: static_F = F;
301: !cholmod_X_factorize(&cholA,chol->factor,chol->common);
302: if (ierr) SETERRQ1(((PetscObject)F)->comm,PETSC_ERR_LIB,"CHOLMOD factorization failed with status %d",chol->common->status);
303: if (chol->common->status == CHOLMOD_NOT_POSDEF)
304: SETERRQ1(((PetscObject)F)->comm,PETSC_ERR_MAT_CH_ZRPVT,"CHOLMOD detected that the matrix is not positive definite, failure at column %u",(unsigned)chol->factor->minor);
306: if (aijalloc) {PetscFree3(cholA.p,cholA.i,cholA.x);}
308: F->ops->solve = MatSolve_CHOLMOD;
309: F->ops->solvetranspose = MatSolve_CHOLMOD;
310: return(0);
311: }
315: PetscErrorCode MatCholeskyFactorSymbolic_CHOLMOD(Mat F,Mat A,IS perm,const MatFactorInfo *info)
316: {
317: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->spptr;
319: cholmod_sparse cholA;
320: PetscBool aijalloc;
321: PetscInt *fset = 0;
322: size_t fsize = 0;
325: (*chol->Wrap)(A,PETSC_FALSE,&cholA,&aijalloc);
326: static_F = F;
327: if (chol->factor) {
328: !cholmod_X_resymbol(&cholA,fset,fsize,(int)chol->pack,chol->factor,chol->common);
329: if (ierr) SETERRQ1(((PetscObject)F)->comm,PETSC_ERR_LIB,"CHOLMOD analysis failed with status %d",chol->common->status);
330: } else if (perm) {
331: const PetscInt *ip;
332: ISGetIndices(perm,&ip);
333: chol->factor = cholmod_X_analyze_p(&cholA,(PetscInt*)ip,fset,fsize,chol->common);
334: if (!chol->factor) SETERRQ1(((PetscObject)F)->comm,PETSC_ERR_LIB,"CHOLMOD analysis failed with status %d",chol->common->status);
335: ISRestoreIndices(perm,&ip);
336: } else {
337: chol->factor = cholmod_X_analyze(&cholA,chol->common);
338: if (!chol->factor) SETERRQ1(((PetscObject)F)->comm,PETSC_ERR_LIB,"CHOLMOD analysis failed with status %d",chol->common->status);
339: }
341: if (aijalloc) {PetscFree3(cholA.p,cholA.i,cholA.x);}
343: F->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_CHOLMOD;
344: return(0);
345: }
347: EXTERN_C_BEGIN
350: PetscErrorCode MatFactorGetSolverPackage_seqsbaij_cholmod(Mat A,const MatSolverPackage *type)
351: {
353: *type = MATSOLVERCHOLMOD;
354: return(0);
355: }
356: EXTERN_C_END
358: /*MC
359: MATSOLVERCHOLMOD = "cholmod" - A matrix type providing direct solvers (Cholesky) for sequential matrices
360: via the external package CHOLMOD.
362: ./configure --download-cholmod to install PETSc to use CHOLMOD
364: Consult CHOLMOD documentation for more information about the Common parameters
365: which correspond to the options database keys below.
367: Options Database Keys:
368: + -mat_cholmod_dbound <0> - Minimum absolute value of diagonal entries of D (None)
369: . -mat_cholmod_grow0 <1.2> - Global growth ratio when factors are modified (None)
370: . -mat_cholmod_grow1 <1.2> - Column growth ratio when factors are modified (None)
371: . -mat_cholmod_grow2 <5> - Affine column growth constant when factors are modified (None)
372: . -mat_cholmod_maxrank <8> - Max rank of update, larger values are faster but use more memory [2,4,8] (None)
373: . -mat_cholmod_factor <AUTO> - (choose one of) SIMPLICIAL AUTO SUPERNODAL
374: . -mat_cholmod_supernodal_switch <40> - flop/nnz_L threshold for switching to supernodal factorization (None)
375: . -mat_cholmod_final_asis <TRUE> - Leave factors "as is" (None)
376: . -mat_cholmod_final_pack <TRUE> - Pack the columns when finished (use FALSE if the factors will be updated later) (None)
377: . -mat_cholmod_zrelax <0.8> - 3 real supernodal relaxed amalgamation parameters (None)
378: . -mat_cholmod_nrelax <4> - 3 size_t supernodal relaxed amalgamation parameters (None)
379: . -mat_cholmod_prefer_upper <TRUE> - Work with upper triangular form (faster when using fill-reducing ordering, slower in natural ordering) (None)
380: - -mat_cholmod_print <3> - Verbosity level (None)
382: Level: beginner
384: .seealso: PCCHOLESKY, PCFactorSetMatSolverPackage(), MatSolverPackage
385: M*/
386: EXTERN_C_BEGIN
389: PetscErrorCode MatGetFactor_seqsbaij_cholmod(Mat A,MatFactorType ftype,Mat *F)
390: {
391: Mat B;
392: Mat_CHOLMOD *chol;
394: PetscInt m=A->rmap->n,n=A->cmap->n,bs;
397: if (ftype != MAT_FACTOR_CHOLESKY) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_SUP,"CHOLMOD cannot do %s factorization with SBAIJ, only %s",
398: MatFactorTypes[ftype],MatFactorTypes[MAT_FACTOR_CHOLESKY]);
399: MatGetBlockSize(A,&bs);
400: if (bs != 1) SETERRQ1(((PetscObject)A)->comm,PETSC_ERR_SUP,"CHOLMOD only supports block size=1, given %D",bs);
401: /* Create the factorization matrix F */
402: MatCreate(((PetscObject)A)->comm,&B);
403: MatSetSizes(B,PETSC_DECIDE,PETSC_DECIDE,m,n);
404: MatSetType(B,((PetscObject)A)->type_name);
405: MatSeqSBAIJSetPreallocation(B,1,0,PETSC_NULL);
406: PetscNewLog(B,Mat_CHOLMOD,&chol);
407: chol->Wrap = MatWrapCholmod_seqsbaij;
408: chol->Destroy = MatDestroy_SeqSBAIJ;
409: B->spptr = chol;
411: B->ops->view = MatView_CHOLMOD;
412: B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_CHOLMOD;
413: B->ops->destroy = MatDestroy_CHOLMOD;
414: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatFactorGetSolverPackage_C","MatFactorGetSolverPackage_seqsbaij_cholmod",MatFactorGetSolverPackage_seqsbaij_cholmod);
415: B->factortype = MAT_FACTOR_CHOLESKY;
416: B->assembled = PETSC_TRUE; /* required by -ksp_view */
417: B->preallocated = PETSC_TRUE;
419: CholmodStart(B);
420: *F = B;
421: return(0);
422: }
423: EXTERN_C_END