Actual source code: sbaijcholmod.c
petsc-3.8.4 2018-03-24
2: /*
3: Provides an interface to the CHOLMOD sparse solver available through SuiteSparse version 4.2.1
5: When build with PETSC_USE_64BIT_INDICES this will use Suitesparse_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 one cannot use 64BIT_INDICES on 32bit machines [as Suitesparse_long is 32bit only]
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;
21: static void CholmodErrorHandler(int status,const char *file,int line,const char *message)
22: {
26: if (status > CHOLMOD_OK) {
27: PetscInfo4(static_F,"CHOLMOD warning %d at %s:%d: %s\n",status,file,line,message);CHKERRV(ierr);
28: } else if (status == CHOLMOD_OK) { /* Documentation says this can happen, but why? */
29: PetscInfo3(static_F,"CHOLMOD OK at %s:%d: %s\n",file,line,message);CHKERRV(ierr);
30: } else {
31: PetscErrorPrintf("CHOLMOD error %d at %s:%d: %s\n",status,file,line,message);CHKERRV(ierr);
32: }
33: PetscFunctionReturnVoid();
34: }
36: PetscErrorCode CholmodStart(Mat F)
37: {
39: Mat_CHOLMOD *chol=(Mat_CHOLMOD*)F->data;
40: cholmod_common *c;
41: PetscBool flg;
44: if (chol->common) return(0);
45: PetscMalloc1(1,&chol->common);
46: !cholmod_X_start(chol->common);
48: c = chol->common;
49: c->error_handler = CholmodErrorHandler;
51: #define CHOLMOD_OPTION_DOUBLE(name,help) do { \
52: PetscReal tmp = (PetscReal)c->name; \
53: PetscOptionsReal("-mat_cholmod_" #name,help,"None",tmp,&tmp,NULL); \
54: c->name = (double)tmp; \
55: } while (0)
57: #define CHOLMOD_OPTION_INT(name,help) do { \
58: PetscInt tmp = (PetscInt)c->name; \
59: PetscOptionsInt("-mat_cholmod_" #name,help,"None",tmp,&tmp,NULL); \
60: c->name = (int)tmp; \
61: } 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,NULL); \
66: if (tmp < 0) SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_OUTOFRANGE,"value must be positive"); \
67: c->name = (size_t)tmp; \
68: } while (0)
70: #define CHOLMOD_OPTION_BOOL(name,help) do { \
71: PetscBool tmp = (PetscBool) !!c->name; \
72: PetscOptionsBool("-mat_cholmod_" #name,help,"None",tmp,&tmp,NULL); \
73: c->name = (int)tmp; \
74: } while (0)
76: PetscOptionsBegin(PetscObjectComm((PetscObject)F),((PetscObject)F)->prefix,"CHOLMOD Options","Mat");
77: /* CHOLMOD handles first-time packing and refactor-packing separately, but we usually want them to be the same. */
78: chol->pack = (PetscBool)c->final_pack;
80: #if defined(PETSC_USE_SUITESPARSE_GPU)
81: c->useGPU = 1;
82: CHOLMOD_OPTION_INT(useGPU,"Use GPU for BLAS 1, otherwise 0");
83: #endif
85: PetscOptionsBool("-mat_cholmod_pack","Pack factors after factorization [disable for frequent repeat factorization]","None",chol->pack,&chol->pack,NULL);
86: c->final_pack = (int)chol->pack;
88: CHOLMOD_OPTION_DOUBLE(dbound,"Minimum absolute value of diagonal entries of D");
89: CHOLMOD_OPTION_DOUBLE(grow0,"Global growth ratio when factors are modified");
90: CHOLMOD_OPTION_DOUBLE(grow1,"Column growth ratio when factors are modified");
91: CHOLMOD_OPTION_SIZE_T(grow2,"Affine column growth constant when factors are modified");
92: CHOLMOD_OPTION_SIZE_T(maxrank,"Max rank of update, larger values are faster but use more memory [2,4,8]");
93: {
94: static const char *const list[] = {"SIMPLICIAL","AUTO","SUPERNODAL","MatCholmodFactorType","MAT_CHOLMOD_FACTOR_",0};
95: PetscOptionsEnum("-mat_cholmod_factor","Factorization method","None",list,(PetscEnum)c->supernodal,(PetscEnum*)&c->supernodal,NULL);
96: }
97: if (c->supernodal) CHOLMOD_OPTION_DOUBLE(supernodal_switch,"flop/nnz_L threshold for switching to supernodal factorization");
98: CHOLMOD_OPTION_BOOL(final_asis,"Leave factors \"as is\"");
99: CHOLMOD_OPTION_BOOL(final_pack,"Pack the columns when finished (use FALSE if the factors will be updated later)");
100: if (!c->final_asis) {
101: CHOLMOD_OPTION_BOOL(final_super,"Leave supernodal factors instead of converting to simplicial");
102: CHOLMOD_OPTION_BOOL(final_ll,"Turn LDL' factorization into LL'");
103: CHOLMOD_OPTION_BOOL(final_monotonic,"Ensure columns are monotonic when done");
104: CHOLMOD_OPTION_BOOL(final_resymbol,"Remove numerically zero values resulting from relaxed supernodal amalgamation");
105: }
106: {
107: PetscReal tmp[] = {(PetscReal)c->zrelax[0],(PetscReal)c->zrelax[1],(PetscReal)c->zrelax[2]};
108: PetscInt n = 3;
109: PetscOptionsRealArray("-mat_cholmod_zrelax","3 real supernodal relaxed amalgamation parameters","None",tmp,&n,&flg);
110: if (flg && n != 3) SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_OUTOFRANGE,"must provide exactly 3 parameters to -mat_cholmod_zrelax");
111: if (flg) while (n--) c->zrelax[n] = (double)tmp[n];
112: }
113: {
114: PetscInt n,tmp[] = {(PetscInt)c->nrelax[0],(PetscInt)c->nrelax[1],(PetscInt)c->nrelax[2]};
115: PetscOptionsIntArray("-mat_cholmod_nrelax","3 size_t supernodal relaxed amalgamation parameters","None",tmp,&n,&flg);
116: if (flg && n != 3) SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_OUTOFRANGE,"must provide exactly 3 parameters to -mat_cholmod_nrelax");
117: if (flg) while (n--) c->nrelax[n] = (size_t)tmp[n];
118: }
119: CHOLMOD_OPTION_BOOL(prefer_upper,"Work with upper triangular form [faster when using fill-reducing ordering, slower in natural ordering]");
120: CHOLMOD_OPTION_BOOL(default_nesdis,"Use NESDIS instead of METIS for nested dissection");
121: CHOLMOD_OPTION_INT(print,"Verbosity level");
122: PetscOptionsEnd();
123: return(0);
124: }
126: static PetscErrorCode MatWrapCholmod_seqsbaij(Mat A,PetscBool values,cholmod_sparse *C,PetscBool *aijalloc)
127: {
128: Mat_SeqSBAIJ *sbaij = (Mat_SeqSBAIJ*)A->data;
132: PetscMemzero(C,sizeof(*C));
133: /* 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 */
134: C->nrow = (size_t)A->cmap->n;
135: C->ncol = (size_t)A->rmap->n;
136: C->nzmax = (size_t)sbaij->maxnz;
137: C->p = sbaij->i;
138: C->i = sbaij->j;
139: C->x = sbaij->a;
140: C->stype = -1;
141: C->itype = CHOLMOD_INT_TYPE;
142: C->xtype = CHOLMOD_SCALAR_TYPE;
143: C->dtype = CHOLMOD_DOUBLE;
144: C->sorted = 1;
145: C->packed = 1;
146: *aijalloc = PETSC_FALSE;
147: return(0);
148: }
150: static PetscErrorCode VecWrapCholmodRead(Vec X,cholmod_dense *Y)
151: {
152: PetscErrorCode ierr;
153: const PetscScalar *x;
154: PetscInt n;
157: PetscMemzero(Y,sizeof(*Y));
158: VecGetArrayRead(X,&x);
159: VecGetSize(X,&n);
161: Y->x = (double*)x;
162: Y->nrow = n;
163: Y->ncol = 1;
164: Y->nzmax = n;
165: Y->d = n;
166: Y->x = (double*)x;
167: Y->xtype = CHOLMOD_SCALAR_TYPE;
168: Y->dtype = CHOLMOD_DOUBLE;
169: return(0);
170: }
172: static PetscErrorCode VecUnWrapCholmodRead(Vec X,cholmod_dense *Y)
173: {
174: PetscErrorCode ierr;
177: VecRestoreArrayRead(X,NULL);
178: return(0);
179: }
181: PETSC_INTERN PetscErrorCode MatDestroy_CHOLMOD(Mat F)
182: {
184: Mat_CHOLMOD *chol=(Mat_CHOLMOD*)F->data;
187: !cholmod_X_free_factor(&chol->factor,chol->common);
188: !cholmod_X_finish(chol->common);
189: PetscFree(chol->common);
190: PetscFree(chol->matrix);
191: PetscFree(F->data);
192: return(0);
193: }
195: static PetscErrorCode MatSolve_CHOLMOD(Mat,Vec,Vec);
197: /*static const char *const CholmodOrderingMethods[] = {"User","AMD","METIS","NESDIS(default)","Natural","NESDIS(small=20000)","NESDIS(small=4,no constrained)","NESDIS()"};*/
199: static PetscErrorCode MatFactorInfo_CHOLMOD(Mat F,PetscViewer viewer)
200: {
201: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->data;
202: const cholmod_common *c = chol->common;
203: PetscErrorCode ierr;
204: PetscInt i;
207: if (F->ops->solve != MatSolve_CHOLMOD) return(0);
208: PetscViewerASCIIPrintf(viewer,"CHOLMOD run parameters:\n");
209: PetscViewerASCIIPushTab(viewer);
210: PetscViewerASCIIPrintf(viewer,"Pack factors after symbolic factorization: %s\n",chol->pack ? "TRUE" : "FALSE");
211: PetscViewerASCIIPrintf(viewer,"Common.dbound %g (Smallest absolute value of diagonal entries of D)\n",c->dbound);
212: PetscViewerASCIIPrintf(viewer,"Common.grow0 %g\n",c->grow0);
213: PetscViewerASCIIPrintf(viewer,"Common.grow1 %g\n",c->grow1);
214: PetscViewerASCIIPrintf(viewer,"Common.grow2 %u\n",(unsigned)c->grow2);
215: PetscViewerASCIIPrintf(viewer,"Common.maxrank %u\n",(unsigned)c->maxrank);
216: PetscViewerASCIIPrintf(viewer,"Common.supernodal_switch %g\n",c->supernodal_switch);
217: PetscViewerASCIIPrintf(viewer,"Common.supernodal %d\n",c->supernodal);
218: PetscViewerASCIIPrintf(viewer,"Common.final_asis %d\n",c->final_asis);
219: PetscViewerASCIIPrintf(viewer,"Common.final_super %d\n",c->final_super);
220: PetscViewerASCIIPrintf(viewer,"Common.final_ll %d\n",c->final_ll);
221: PetscViewerASCIIPrintf(viewer,"Common.final_pack %d\n",c->final_pack);
222: PetscViewerASCIIPrintf(viewer,"Common.final_monotonic %d\n",c->final_monotonic);
223: PetscViewerASCIIPrintf(viewer,"Common.final_resymbol %d\n",c->final_resymbol);
224: PetscViewerASCIIPrintf(viewer,"Common.zrelax [%g,%g,%g]\n",c->zrelax[0],c->zrelax[1],c->zrelax[2]);
225: PetscViewerASCIIPrintf(viewer,"Common.nrelax [%u,%u,%u]\n",(unsigned)c->nrelax[0],(unsigned)c->nrelax[1],(unsigned)c->nrelax[2]);
226: PetscViewerASCIIPrintf(viewer,"Common.prefer_upper %d\n",c->prefer_upper);
227: PetscViewerASCIIPrintf(viewer,"Common.print %d\n",c->print);
228: for (i=0; i<c->nmethods; i++) {
229: PetscViewerASCIIPrintf(viewer,"Ordering method %D%s:\n",i,i==c->selected ? " [SELECTED]" : "");
230: PetscViewerASCIIPrintf(viewer," lnz %g, fl %g, prune_dense %g, prune_dense2 %g\n",
231: c->method[i].lnz,c->method[i].fl,c->method[i].prune_dense,c->method[i].prune_dense2);
232: }
233: PetscViewerASCIIPrintf(viewer,"Common.postorder %d\n",c->postorder);
234: PetscViewerASCIIPrintf(viewer,"Common.default_nesdis %d (use NESDIS instead of METIS for nested dissection)\n",c->default_nesdis);
235: /* Statistics */
236: PetscViewerASCIIPrintf(viewer,"Common.fl %g (flop count from most recent analysis)\n",c->fl);
237: PetscViewerASCIIPrintf(viewer,"Common.lnz %g (fundamental nz in L)\n",c->lnz);
238: PetscViewerASCIIPrintf(viewer,"Common.anz %g\n",c->anz);
239: PetscViewerASCIIPrintf(viewer,"Common.modfl %g (flop count from most recent update)\n",c->modfl);
240: PetscViewerASCIIPrintf(viewer,"Common.malloc_count %g (number of live objects)\n",(double)c->malloc_count);
241: PetscViewerASCIIPrintf(viewer,"Common.memory_usage %g (peak memory usage in bytes)\n",(double)c->memory_usage);
242: PetscViewerASCIIPrintf(viewer,"Common.memory_inuse %g (current memory usage in bytes)\n",(double)c->memory_inuse);
243: PetscViewerASCIIPrintf(viewer,"Common.nrealloc_col %g (number of column reallocations)\n",c->nrealloc_col);
244: PetscViewerASCIIPrintf(viewer,"Common.nrealloc_factor %g (number of factor reallocations due to column reallocations)\n",c->nrealloc_factor);
245: PetscViewerASCIIPrintf(viewer,"Common.ndbounds_hit %g (number of times diagonal was modified by dbound)\n",c->ndbounds_hit);
246: PetscViewerASCIIPrintf(viewer,"Common.rowfacfl %g (number of flops in last call to cholmod_rowfac)\n",c->rowfacfl);
247: PetscViewerASCIIPrintf(viewer,"Common.aatfl %g (number of flops to compute A(:,f)*A(:,f)')\n",c->aatfl);
248: #if defined(PETSC_USE_SUITESPARSE_GPU)
249: PetscViewerASCIIPrintf(viewer,"Common.useGPU %d\n",c->useGPU);
250: #endif
251: PetscViewerASCIIPopTab(viewer);
252: return(0);
253: }
255: PETSC_INTERN PetscErrorCode MatView_CHOLMOD(Mat F,PetscViewer viewer)
256: {
257: PetscErrorCode ierr;
258: PetscBool iascii;
259: PetscViewerFormat format;
262: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
263: if (iascii) {
264: PetscViewerGetFormat(viewer,&format);
265: if (format == PETSC_VIEWER_ASCII_INFO) {
266: MatFactorInfo_CHOLMOD(F,viewer);
267: }
268: }
269: return(0);
270: }
272: static PetscErrorCode MatSolve_CHOLMOD(Mat F,Vec B,Vec X)
273: {
274: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->data;
275: cholmod_dense cholB,*cholX;
276: PetscScalar *x;
280: VecWrapCholmodRead(B,&cholB);
281: static_F = F;
282: cholX = cholmod_X_solve(CHOLMOD_A,chol->factor,&cholB,chol->common);
283: if (!cholX) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"CHOLMOD failed");
284: VecUnWrapCholmodRead(B,&cholB);
285: VecGetArray(X,&x);
286: PetscMemcpy(x,cholX->x,cholX->nrow*sizeof(*x));
287: !cholmod_X_free_dense(&cholX,chol->common);
288: VecRestoreArray(X,&x);
289: return(0);
290: }
292: static PetscErrorCode MatCholeskyFactorNumeric_CHOLMOD(Mat F,Mat A,const MatFactorInfo *info)
293: {
294: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->data;
295: cholmod_sparse cholA;
296: PetscBool aijalloc;
300: (*chol->Wrap)(A,PETSC_TRUE,&cholA,&aijalloc);
301: static_F = F;
302: !cholmod_X_factorize(&cholA,chol->factor,chol->common);
303: if (ierr) SETERRQ1(PetscObjectComm((PetscObject)F),PETSC_ERR_LIB,"CHOLMOD factorization failed with status %d",chol->common->status);
304: if (chol->common->status == CHOLMOD_NOT_POSDEF) SETERRQ1(PetscObjectComm((PetscObject)F),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: }
313: PETSC_INTERN PetscErrorCode MatCholeskyFactorSymbolic_CHOLMOD(Mat F,Mat A,IS perm,const MatFactorInfo *info)
314: {
315: Mat_CHOLMOD *chol = (Mat_CHOLMOD*)F->data;
317: cholmod_sparse cholA;
318: PetscBool aijalloc;
319: PetscInt *fset = 0;
320: size_t fsize = 0;
323: (*chol->Wrap)(A,PETSC_FALSE,&cholA,&aijalloc);
324: static_F = F;
325: if (chol->factor) {
326: !cholmod_X_resymbol(&cholA,fset,fsize,(int)chol->pack,chol->factor,chol->common);
327: if (ierr) SETERRQ1(PetscObjectComm((PetscObject)F),PETSC_ERR_LIB,"CHOLMOD analysis failed with status %d",chol->common->status);
328: } else if (perm) {
329: const PetscInt *ip;
330: ISGetIndices(perm,&ip);
331: chol->factor = cholmod_X_analyze_p(&cholA,(PetscInt*)ip,fset,fsize,chol->common);
332: if (!chol->factor) SETERRQ1(PetscObjectComm((PetscObject)F),PETSC_ERR_LIB,"CHOLMOD analysis failed with status %d",chol->common->status);
333: ISRestoreIndices(perm,&ip);
334: } else {
335: chol->factor = cholmod_X_analyze(&cholA,chol->common);
336: if (!chol->factor) SETERRQ1(PetscObjectComm((PetscObject)F),PETSC_ERR_LIB,"CHOLMOD analysis failed with status %d",chol->common->status);
337: }
339: if (aijalloc) {PetscFree3(cholA.p,cholA.i,cholA.x);}
341: F->ops->choleskyfactornumeric = MatCholeskyFactorNumeric_CHOLMOD;
342: return(0);
343: }
345: static PetscErrorCode MatFactorGetSolverPackage_seqsbaij_cholmod(Mat A,const MatSolverPackage *type)
346: {
348: *type = MATSOLVERCHOLMOD;
349: return(0);
350: }
352: /*MC
353: MATSOLVERCHOLMOD = "cholmod" - A matrix type providing direct solvers (Cholesky) for sequential matrices
354: via the external package CHOLMOD.
356: Use ./configure --download-suitesparse to install PETSc to use CHOLMOD
358: Use -pc_type lu -pc_factor_mat_solver_package cholmod to use this direct solver
360: Consult CHOLMOD documentation for more information about the Common parameters
361: which correspond to the options database keys below.
363: Options Database Keys:
364: + -mat_cholmod_dbound <0> - Minimum absolute value of diagonal entries of D (None)
365: . -mat_cholmod_grow0 <1.2> - Global growth ratio when factors are modified (None)
366: . -mat_cholmod_grow1 <1.2> - Column growth ratio when factors are modified (None)
367: . -mat_cholmod_grow2 <5> - Affine column growth constant when factors are modified (None)
368: . -mat_cholmod_maxrank <8> - Max rank of update, larger values are faster but use more memory [2,4,8] (None)
369: . -mat_cholmod_factor <AUTO> - (choose one of) SIMPLICIAL AUTO SUPERNODAL
370: . -mat_cholmod_supernodal_switch <40> - flop/nnz_L threshold for switching to supernodal factorization (None)
371: . -mat_cholmod_final_asis <TRUE> - Leave factors "as is" (None)
372: . -mat_cholmod_final_pack <TRUE> - Pack the columns when finished (use FALSE if the factors will be updated later) (None)
373: . -mat_cholmod_zrelax <0.8> - 3 real supernodal relaxed amalgamation parameters (None)
374: . -mat_cholmod_nrelax <4> - 3 size_t supernodal relaxed amalgamation parameters (None)
375: . -mat_cholmod_prefer_upper <TRUE> - Work with upper triangular form (faster when using fill-reducing ordering, slower in natural ordering) (None)
376: - -mat_cholmod_print <3> - Verbosity level (None)
378: Level: beginner
380: Note: CHOLMOD is part of SuiteSparse http://faculty.cse.tamu.edu/davis/suitesparse.html
382: .seealso: PCCHOLESKY, PCFactorSetMatSolverPackage(), MatSolverPackage
383: M*/
385: PETSC_INTERN PetscErrorCode MatGetFactor_seqsbaij_cholmod(Mat A,MatFactorType ftype,Mat *F)
386: {
387: Mat B;
388: Mat_CHOLMOD *chol;
390: PetscInt m=A->rmap->n,n=A->cmap->n,bs;
393: if (ftype != MAT_FACTOR_CHOLESKY) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_SUP,"CHOLMOD cannot do %s factorization with SBAIJ, only %s",
394: MatFactorTypes[ftype],MatFactorTypes[MAT_FACTOR_CHOLESKY]);
395: MatGetBlockSize(A,&bs);
396: if (bs != 1) SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"CHOLMOD only supports block size=1, given %D",bs);
397: /* Create the factorization matrix F */
398: MatCreate(PetscObjectComm((PetscObject)A),&B);
399: MatSetSizes(B,PETSC_DECIDE,PETSC_DECIDE,m,n);
400: PetscStrallocpy("cholmod",&((PetscObject)B)->type_name);
401: MatSetUp(B);
402: PetscNewLog(B,&chol);
404: chol->Wrap = MatWrapCholmod_seqsbaij;
405: B->data = chol;
407: B->ops->getinfo = MatGetInfo_External;
408: B->ops->view = MatView_CHOLMOD;
409: B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_CHOLMOD;
410: B->ops->destroy = MatDestroy_CHOLMOD;
411: PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_seqsbaij_cholmod);
412: B->factortype = MAT_FACTOR_CHOLESKY;
413: B->assembled = PETSC_TRUE; /* required by -ksp_view */
414: B->preallocated = PETSC_TRUE;
416: CholmodStart(B);
418: PetscFree(B->solvertype);
419: PetscStrallocpy(MATSOLVERCHOLMOD,&B->solvertype);
421: *F = B;
422: return(0);
423: }