Actual source code: ispai.c
petsc-3.11.4 2019-09-28
2: /*
3: 3/99 Modified by Stephen Barnard to support SPAI version 3.0
4: */
6: /*
7: Provides an interface to the SPAI Sparse Approximate Inverse Preconditioner
8: Code written by Stephen Barnard.
10: Note: there is some BAD memory bleeding below!
12: This code needs work
14: 1) get rid of all memory bleeding
15: 2) fix PETSc/interface so that it gets if the matrix is symmetric from the matrix
16: rather than having the sp flag for PC_SPAI
17: 3) fix to set the block size based on the matrix block size
19: */
20: #define PETSC_SKIP_COMPLEX /* since spai uses I which conflicts with some complex implementations */
22: #include <petsc/private/pcimpl.h>
23: #include <../src/ksp/pc/impls/spai/petscspai.h>
25: /*
26: These are the SPAI include files
27: */
28: EXTERN_C_BEGIN
29: #define SPAI_USE_MPI /* required for setting SPAI_Comm correctly in basics.h */
30: #include <spai.h>
31: #include <matrix.h>
32: EXTERN_C_END
34: extern PetscErrorCode ConvertMatToMatrix(MPI_Comm,Mat,Mat,matrix**);
35: extern PetscErrorCode ConvertMatrixToMat(MPI_Comm,matrix*,Mat*);
36: extern PetscErrorCode ConvertVectorToVec(MPI_Comm,vector *v,Vec *Pv);
37: extern PetscErrorCode MM_to_PETSC(char*,char*,char*);
39: typedef struct {
41: matrix *B; /* matrix in SPAI format */
42: matrix *BT; /* transpose of matrix in SPAI format */
43: matrix *M; /* the approximate inverse in SPAI format */
45: Mat PM; /* the approximate inverse PETSc format */
47: double epsilon; /* tolerance */
48: int nbsteps; /* max number of "improvement" steps per line */
49: int max; /* max dimensions of is_I, q, etc. */
50: int maxnew; /* max number of new entries per step */
51: int block_size; /* constant block size */
52: int cache_size; /* one of (1,2,3,4,5,6) indicting size of cache */
53: int verbose; /* SPAI prints timing and statistics */
55: int sp; /* symmetric nonzero pattern */
56: MPI_Comm comm_spai; /* communicator to be used with spai */
57: } PC_SPAI;
59: /**********************************************************************/
61: static PetscErrorCode PCSetUp_SPAI(PC pc)
62: {
63: PC_SPAI *ispai = (PC_SPAI*)pc->data;
65: Mat AT;
68: init_SPAI();
70: if (ispai->sp) {
71: ConvertMatToMatrix(ispai->comm_spai,pc->pmat,pc->pmat,&ispai->B);
72: } else {
73: /* Use the transpose to get the column nonzero structure. */
74: MatTranspose(pc->pmat,MAT_INITIAL_MATRIX,&AT);
75: ConvertMatToMatrix(ispai->comm_spai,pc->pmat,AT,&ispai->B);
76: MatDestroy(&AT);
77: }
79: /* Destroy the transpose */
80: /* Don't know how to do it. PETSc developers? */
82: /* construct SPAI preconditioner */
83: /* FILE *messages */ /* file for warning messages */
84: /* double epsilon */ /* tolerance */
85: /* int nbsteps */ /* max number of "improvement" steps per line */
86: /* int max */ /* max dimensions of is_I, q, etc. */
87: /* int maxnew */ /* max number of new entries per step */
88: /* int block_size */ /* block_size == 1 specifies scalar elments
89: block_size == n specifies nxn constant-block elements
90: block_size == 0 specifies variable-block elements */
91: /* int cache_size */ /* one of (1,2,3,4,5,6) indicting size of cache. cache_size == 0 indicates no caching */
92: /* int verbose */ /* verbose == 0 specifies that SPAI is silent
93: verbose == 1 prints timing and matrix statistics */
95: bspai(ispai->B,&ispai->M,
96: stdout,
97: ispai->epsilon,
98: ispai->nbsteps,
99: ispai->max,
100: ispai->maxnew,
101: ispai->block_size,
102: ispai->cache_size,
103: ispai->verbose);
105: ConvertMatrixToMat(PetscObjectComm((PetscObject)pc),ispai->M,&ispai->PM);
107: /* free the SPAI matrices */
108: sp_free_matrix(ispai->B);
109: sp_free_matrix(ispai->M);
110: return(0);
111: }
113: /**********************************************************************/
115: static PetscErrorCode PCApply_SPAI(PC pc,Vec xx,Vec y)
116: {
117: PC_SPAI *ispai = (PC_SPAI*)pc->data;
121: /* Now using PETSc's multiply */
122: MatMult(ispai->PM,xx,y);
123: return(0);
124: }
126: /**********************************************************************/
128: static PetscErrorCode PCDestroy_SPAI(PC pc)
129: {
131: PC_SPAI *ispai = (PC_SPAI*)pc->data;
134: MatDestroy(&ispai->PM);
135: MPI_Comm_free(&(ispai->comm_spai));
136: PetscFree(pc->data);
137: return(0);
138: }
140: /**********************************************************************/
142: static PetscErrorCode PCView_SPAI(PC pc,PetscViewer viewer)
143: {
144: PC_SPAI *ispai = (PC_SPAI*)pc->data;
146: PetscBool iascii;
149: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
150: if (iascii) {
151: PetscViewerASCIIPrintf(viewer," epsilon %g\n", (double)ispai->epsilon);
152: PetscViewerASCIIPrintf(viewer," nbsteps %d\n", ispai->nbsteps);
153: PetscViewerASCIIPrintf(viewer," max %d\n", ispai->max);
154: PetscViewerASCIIPrintf(viewer," maxnew %d\n", ispai->maxnew);
155: PetscViewerASCIIPrintf(viewer," block_size %d\n",ispai->block_size);
156: PetscViewerASCIIPrintf(viewer," cache_size %d\n",ispai->cache_size);
157: PetscViewerASCIIPrintf(viewer," verbose %d\n", ispai->verbose);
158: PetscViewerASCIIPrintf(viewer," sp %d\n", ispai->sp);
159: }
160: return(0);
161: }
163: static PetscErrorCode PCSPAISetEpsilon_SPAI(PC pc,double epsilon1)
164: {
165: PC_SPAI *ispai = (PC_SPAI*)pc->data;
168: ispai->epsilon = epsilon1;
169: return(0);
170: }
172: /**********************************************************************/
174: static PetscErrorCode PCSPAISetNBSteps_SPAI(PC pc,int nbsteps1)
175: {
176: PC_SPAI *ispai = (PC_SPAI*)pc->data;
179: ispai->nbsteps = nbsteps1;
180: return(0);
181: }
183: /**********************************************************************/
185: /* added 1/7/99 g.h. */
186: static PetscErrorCode PCSPAISetMax_SPAI(PC pc,int max1)
187: {
188: PC_SPAI *ispai = (PC_SPAI*)pc->data;
191: ispai->max = max1;
192: return(0);
193: }
195: /**********************************************************************/
197: static PetscErrorCode PCSPAISetMaxNew_SPAI(PC pc,int maxnew1)
198: {
199: PC_SPAI *ispai = (PC_SPAI*)pc->data;
202: ispai->maxnew = maxnew1;
203: return(0);
204: }
206: /**********************************************************************/
208: static PetscErrorCode PCSPAISetBlockSize_SPAI(PC pc,int block_size1)
209: {
210: PC_SPAI *ispai = (PC_SPAI*)pc->data;
213: ispai->block_size = block_size1;
214: return(0);
215: }
217: /**********************************************************************/
219: static PetscErrorCode PCSPAISetCacheSize_SPAI(PC pc,int cache_size)
220: {
221: PC_SPAI *ispai = (PC_SPAI*)pc->data;
224: ispai->cache_size = cache_size;
225: return(0);
226: }
228: /**********************************************************************/
230: static PetscErrorCode PCSPAISetVerbose_SPAI(PC pc,int verbose)
231: {
232: PC_SPAI *ispai = (PC_SPAI*)pc->data;
235: ispai->verbose = verbose;
236: return(0);
237: }
239: /**********************************************************************/
241: static PetscErrorCode PCSPAISetSp_SPAI(PC pc,int sp)
242: {
243: PC_SPAI *ispai = (PC_SPAI*)pc->data;
246: ispai->sp = sp;
247: return(0);
248: }
250: /* -------------------------------------------------------------------*/
252: /*@
253: PCSPAISetEpsilon -- Set the tolerance for the SPAI preconditioner
255: Input Parameters:
256: + pc - the preconditioner
257: - eps - epsilon (default .4)
259: Notes:
260: Espilon must be between 0 and 1. It controls the
261: quality of the approximation of M to the inverse of
262: A. Higher values of epsilon lead to more work, more
263: fill, and usually better preconditioners. In many
264: cases the best choice of epsilon is the one that
265: divides the total solution time equally between the
266: preconditioner and the solver.
268: Level: intermediate
270: .seealso: PCSPAI, PCSetType()
271: @*/
272: PetscErrorCode PCSPAISetEpsilon(PC pc,double epsilon1)
273: {
277: PetscTryMethod(pc,"PCSPAISetEpsilon_C",(PC,double),(pc,epsilon1));
278: return(0);
279: }
281: /**********************************************************************/
283: /*@
284: PCSPAISetNBSteps - set maximum number of improvement steps per row in
285: the SPAI preconditioner
287: Input Parameters:
288: + pc - the preconditioner
289: - n - number of steps (default 5)
291: Notes:
292: SPAI constructs to approximation to every column of
293: the exact inverse of A in a series of improvement
294: steps. The quality of the approximation is determined
295: by epsilon. If an approximation achieving an accuracy
296: of epsilon is not obtained after ns steps, SPAI simply
297: uses the best approximation constructed so far.
299: Level: intermediate
301: .seealso: PCSPAI, PCSetType(), PCSPAISetMaxNew()
302: @*/
303: PetscErrorCode PCSPAISetNBSteps(PC pc,int nbsteps1)
304: {
308: PetscTryMethod(pc,"PCSPAISetNBSteps_C",(PC,int),(pc,nbsteps1));
309: return(0);
310: }
312: /**********************************************************************/
314: /* added 1/7/99 g.h. */
315: /*@
316: PCSPAISetMax - set the size of various working buffers in
317: the SPAI preconditioner
319: Input Parameters:
320: + pc - the preconditioner
321: - n - size (default is 5000)
323: Level: intermediate
325: .seealso: PCSPAI, PCSetType()
326: @*/
327: PetscErrorCode PCSPAISetMax(PC pc,int max1)
328: {
332: PetscTryMethod(pc,"PCSPAISetMax_C",(PC,int),(pc,max1));
333: return(0);
334: }
336: /**********************************************************************/
338: /*@
339: PCSPAISetMaxNew - set maximum number of new nonzero candidates per step
340: in SPAI preconditioner
342: Input Parameters:
343: + pc - the preconditioner
344: - n - maximum number (default 5)
346: Level: intermediate
348: .seealso: PCSPAI, PCSetType(), PCSPAISetNBSteps()
349: @*/
350: PetscErrorCode PCSPAISetMaxNew(PC pc,int maxnew1)
351: {
355: PetscTryMethod(pc,"PCSPAISetMaxNew_C",(PC,int),(pc,maxnew1));
356: return(0);
357: }
359: /**********************************************************************/
361: /*@
362: PCSPAISetBlockSize - set the block size for the SPAI preconditioner
364: Input Parameters:
365: + pc - the preconditioner
366: - n - block size (default 1)
368: Notes:
369: A block
370: size of 1 treats A as a matrix of scalar elements. A
371: block size of s > 1 treats A as a matrix of sxs
372: blocks. A block size of 0 treats A as a matrix with
373: variable sized blocks, which are determined by
374: searching for dense square diagonal blocks in A.
375: This can be very effective for finite-element
376: matrices.
378: SPAI will convert A to block form, use a block
379: version of the preconditioner algorithm, and then
380: convert the result back to scalar form.
382: In many cases the a block-size parameter other than 1
383: can lead to very significant improvement in
384: performance.
387: Level: intermediate
389: .seealso: PCSPAI, PCSetType()
390: @*/
391: PetscErrorCode PCSPAISetBlockSize(PC pc,int block_size1)
392: {
396: PetscTryMethod(pc,"PCSPAISetBlockSize_C",(PC,int),(pc,block_size1));
397: return(0);
398: }
400: /**********************************************************************/
402: /*@
403: PCSPAISetCacheSize - specify cache size in the SPAI preconditioner
405: Input Parameters:
406: + pc - the preconditioner
407: - n - cache size {0,1,2,3,4,5} (default 5)
409: Notes:
410: SPAI uses a hash table to cache messages and avoid
411: redundant communication. If suggest always using
412: 5. This parameter is irrelevant in the serial
413: version.
415: Level: intermediate
417: .seealso: PCSPAI, PCSetType()
418: @*/
419: PetscErrorCode PCSPAISetCacheSize(PC pc,int cache_size)
420: {
424: PetscTryMethod(pc,"PCSPAISetCacheSize_C",(PC,int),(pc,cache_size));
425: return(0);
426: }
428: /**********************************************************************/
430: /*@
431: PCSPAISetVerbose - verbosity level for the SPAI preconditioner
433: Input Parameters:
434: + pc - the preconditioner
435: - n - level (default 1)
437: Notes:
438: print parameters, timings and matrix statistics
440: Level: intermediate
442: .seealso: PCSPAI, PCSetType()
443: @*/
444: PetscErrorCode PCSPAISetVerbose(PC pc,int verbose)
445: {
449: PetscTryMethod(pc,"PCSPAISetVerbose_C",(PC,int),(pc,verbose));
450: return(0);
451: }
453: /**********************************************************************/
455: /*@
456: PCSPAISetSp - specify a symmetric matrix sparsity pattern in the SPAI preconditioner
458: Input Parameters:
459: + pc - the preconditioner
460: - n - 0 or 1
462: Notes:
463: If A has a symmetric nonzero pattern use -sp 1 to
464: improve performance by eliminating some communication
465: in the parallel version. Even if A does not have a
466: symmetric nonzero pattern -sp 1 may well lead to good
467: results, but the code will not follow the published
468: SPAI algorithm exactly.
471: Level: intermediate
473: .seealso: PCSPAI, PCSetType()
474: @*/
475: PetscErrorCode PCSPAISetSp(PC pc,int sp)
476: {
480: PetscTryMethod(pc,"PCSPAISetSp_C",(PC,int),(pc,sp));
481: return(0);
482: }
484: /**********************************************************************/
486: /**********************************************************************/
488: static PetscErrorCode PCSetFromOptions_SPAI(PetscOptionItems *PetscOptionsObject,PC pc)
489: {
490: PC_SPAI *ispai = (PC_SPAI*)pc->data;
492: int nbsteps1,max1,maxnew1,block_size1,cache_size,verbose,sp;
493: double epsilon1;
494: PetscBool flg;
497: PetscOptionsHead(PetscOptionsObject,"SPAI options");
498: PetscOptionsReal("-pc_spai_epsilon","","PCSPAISetEpsilon",ispai->epsilon,&epsilon1,&flg);
499: if (flg) {
500: PCSPAISetEpsilon(pc,epsilon1);
501: }
502: PetscOptionsInt("-pc_spai_nbsteps","","PCSPAISetNBSteps",ispai->nbsteps,&nbsteps1,&flg);
503: if (flg) {
504: PCSPAISetNBSteps(pc,nbsteps1);
505: }
506: /* added 1/7/99 g.h. */
507: PetscOptionsInt("-pc_spai_max","","PCSPAISetMax",ispai->max,&max1,&flg);
508: if (flg) {
509: PCSPAISetMax(pc,max1);
510: }
511: PetscOptionsInt("-pc_spai_maxnew","","PCSPAISetMaxNew",ispai->maxnew,&maxnew1,&flg);
512: if (flg) {
513: PCSPAISetMaxNew(pc,maxnew1);
514: }
515: PetscOptionsInt("-pc_spai_block_size","","PCSPAISetBlockSize",ispai->block_size,&block_size1,&flg);
516: if (flg) {
517: PCSPAISetBlockSize(pc,block_size1);
518: }
519: PetscOptionsInt("-pc_spai_cache_size","","PCSPAISetCacheSize",ispai->cache_size,&cache_size,&flg);
520: if (flg) {
521: PCSPAISetCacheSize(pc,cache_size);
522: }
523: PetscOptionsInt("-pc_spai_verbose","","PCSPAISetVerbose",ispai->verbose,&verbose,&flg);
524: if (flg) {
525: PCSPAISetVerbose(pc,verbose);
526: }
527: PetscOptionsInt("-pc_spai_sp","","PCSPAISetSp",ispai->sp,&sp,&flg);
528: if (flg) {
529: PCSPAISetSp(pc,sp);
530: }
531: PetscOptionsTail();
532: return(0);
533: }
535: /**********************************************************************/
537: /*MC
538: PCSPAI - Use the Sparse Approximate Inverse method of Grote and Barnard
539: as a preconditioner (SIAM J. Sci. Comput.; vol 18, nr 3)
541: Options Database Keys:
542: + -pc_spai_epsilon <eps> - set tolerance
543: . -pc_spai_nbstep <n> - set nbsteps
544: . -pc_spai_max <m> - set max
545: . -pc_spai_max_new <m> - set maxnew
546: . -pc_spai_block_size <n> - set block size
547: . -pc_spai_cache_size <n> - set cache size
548: . -pc_spai_sp <m> - set sp
549: - -pc_spai_set_verbose <true,false> - verbose output
551: Notes:
552: This only works with AIJ matrices.
554: Level: beginner
556: Concepts: approximate inverse
558: .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PC,
559: PCSPAISetEpsilon(), PCSPAISetMax(), PCSPAISetMaxNew(), PCSPAISetBlockSize(),
560: PCSPAISetVerbose(), PCSPAISetSp()
561: M*/
563: PETSC_EXTERN PetscErrorCode PCCreate_SPAI(PC pc)
564: {
565: PC_SPAI *ispai;
569: PetscNewLog(pc,&ispai);
570: pc->data = ispai;
572: pc->ops->destroy = PCDestroy_SPAI;
573: pc->ops->apply = PCApply_SPAI;
574: pc->ops->applyrichardson = 0;
575: pc->ops->setup = PCSetUp_SPAI;
576: pc->ops->view = PCView_SPAI;
577: pc->ops->setfromoptions = PCSetFromOptions_SPAI;
579: ispai->epsilon = .4;
580: ispai->nbsteps = 5;
581: ispai->max = 5000;
582: ispai->maxnew = 5;
583: ispai->block_size = 1;
584: ispai->cache_size = 5;
585: ispai->verbose = 0;
587: ispai->sp = 1;
588: MPI_Comm_dup(PetscObjectComm((PetscObject)pc),&(ispai->comm_spai));
590: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetEpsilon_C",PCSPAISetEpsilon_SPAI);
591: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetNBSteps_C",PCSPAISetNBSteps_SPAI);
592: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetMax_C",PCSPAISetMax_SPAI);
593: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetMaxNew_C",PCSPAISetMaxNew_SPAI);
594: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetBlockSize_C",PCSPAISetBlockSize_SPAI);
595: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetCacheSize_C",PCSPAISetCacheSize_SPAI);
596: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetVerbose_C",PCSPAISetVerbose_SPAI);
597: PetscObjectComposeFunction((PetscObject)pc,"PCSPAISetSp_C",PCSPAISetSp_SPAI);
598: return(0);
599: }
601: /**********************************************************************/
603: /*
604: Converts from a PETSc matrix to an SPAI matrix
605: */
606: PetscErrorCode ConvertMatToMatrix(MPI_Comm comm, Mat A,Mat AT,matrix **B)
607: {
608: matrix *M;
609: int i,j,col;
610: int row_indx;
611: int len,pe,local_indx,start_indx;
612: int *mapping;
613: PetscErrorCode ierr;
614: const int *cols;
615: const double *vals;
616: int n,mnl,nnl,nz,rstart,rend;
617: PetscMPIInt size,rank;
618: struct compressed_lines *rows;
621: MPI_Comm_size(comm,&size);
622: MPI_Comm_rank(comm,&rank);
623: MatGetSize(A,&n,&n);
624: MatGetLocalSize(A,&mnl,&nnl);
626: /*
627: not sure why a barrier is required. commenting out
628: MPI_Barrier(comm);
629: */
631: M = new_matrix((SPAI_Comm)comm);
633: M->n = n;
634: M->bs = 1;
635: M->max_block_size = 1;
637: M->mnls = (int*)malloc(sizeof(int)*size);
638: M->start_indices = (int*)malloc(sizeof(int)*size);
639: M->pe = (int*)malloc(sizeof(int)*n);
640: M->block_sizes = (int*)malloc(sizeof(int)*n);
641: for (i=0; i<n; i++) M->block_sizes[i] = 1;
643: MPI_Allgather(&mnl,1,MPI_INT,M->mnls,1,MPI_INT,comm);
645: M->start_indices[0] = 0;
646: for (i=1; i<size; i++) M->start_indices[i] = M->start_indices[i-1] + M->mnls[i-1];
648: M->mnl = M->mnls[M->myid];
649: M->my_start_index = M->start_indices[M->myid];
651: for (i=0; i<size; i++) {
652: start_indx = M->start_indices[i];
653: for (j=0; j<M->mnls[i]; j++) M->pe[start_indx+j] = i;
654: }
656: if (AT) {
657: M->lines = new_compressed_lines(M->mnls[rank],1);
658: } else {
659: M->lines = new_compressed_lines(M->mnls[rank],0);
660: }
662: rows = M->lines;
664: /* Determine the mapping from global indices to pointers */
665: PetscMalloc1(M->n,&mapping);
666: pe = 0;
667: local_indx = 0;
668: for (i=0; i<M->n; i++) {
669: if (local_indx >= M->mnls[pe]) {
670: pe++;
671: local_indx = 0;
672: }
673: mapping[i] = local_indx + M->start_indices[pe];
674: local_indx++;
675: }
677: /*********************************************************/
678: /************** Set up the row structure *****************/
679: /*********************************************************/
681: MatGetOwnershipRange(A,&rstart,&rend);
682: for (i=rstart; i<rend; i++) {
683: row_indx = i - rstart;
684: MatGetRow(A,i,&nz,&cols,&vals);
685: /* allocate buffers */
686: rows->ptrs[row_indx] = (int*)malloc(nz*sizeof(int));
687: rows->A[row_indx] = (double*)malloc(nz*sizeof(double));
688: /* copy the matrix */
689: for (j=0; j<nz; j++) {
690: col = cols[j];
691: len = rows->len[row_indx]++;
693: rows->ptrs[row_indx][len] = mapping[col];
694: rows->A[row_indx][len] = vals[j];
695: }
696: rows->slen[row_indx] = rows->len[row_indx];
698: MatRestoreRow(A,i,&nz,&cols,&vals);
699: }
702: /************************************************************/
703: /************** Set up the column structure *****************/
704: /*********************************************************/
706: if (AT) {
708: for (i=rstart; i<rend; i++) {
709: row_indx = i - rstart;
710: MatGetRow(AT,i,&nz,&cols,&vals);
711: /* allocate buffers */
712: rows->rptrs[row_indx] = (int*)malloc(nz*sizeof(int));
713: /* copy the matrix (i.e., the structure) */
714: for (j=0; j<nz; j++) {
715: col = cols[j];
716: len = rows->rlen[row_indx]++;
718: rows->rptrs[row_indx][len] = mapping[col];
719: }
720: MatRestoreRow(AT,i,&nz,&cols,&vals);
721: }
722: }
724: PetscFree(mapping);
726: order_pointers(M);
727: M->maxnz = calc_maxnz(M);
728: *B = M;
729: return(0);
730: }
732: /**********************************************************************/
734: /*
735: Converts from an SPAI matrix B to a PETSc matrix PB.
736: This assumes that the SPAI matrix B is stored in
737: COMPRESSED-ROW format.
738: */
739: PetscErrorCode ConvertMatrixToMat(MPI_Comm comm,matrix *B,Mat *PB)
740: {
741: PetscMPIInt size,rank;
743: int m,n,M,N;
744: int d_nz,o_nz;
745: int *d_nnz,*o_nnz;
746: int i,k,global_row,global_col,first_diag_col,last_diag_col;
747: PetscScalar val;
750: MPI_Comm_size(comm,&size);
751: MPI_Comm_rank(comm,&rank);
753: m = n = B->mnls[rank];
754: d_nz = o_nz = 0;
756: /* Determine preallocation for MatCreateMPIAIJ */
757: PetscMalloc1(m,&d_nnz);
758: PetscMalloc1(m,&o_nnz);
759: for (i=0; i<m; i++) d_nnz[i] = o_nnz[i] = 0;
760: first_diag_col = B->start_indices[rank];
761: last_diag_col = first_diag_col + B->mnls[rank];
762: for (i=0; i<B->mnls[rank]; i++) {
763: for (k=0; k<B->lines->len[i]; k++) {
764: global_col = B->lines->ptrs[i][k];
765: if ((global_col >= first_diag_col) && (global_col < last_diag_col)) d_nnz[i]++;
766: else o_nnz[i]++;
767: }
768: }
770: M = N = B->n;
771: /* Here we only know how to create AIJ format */
772: MatCreate(comm,PB);
773: MatSetSizes(*PB,m,n,M,N);
774: MatSetType(*PB,MATAIJ);
775: MatSeqAIJSetPreallocation(*PB,d_nz,d_nnz);
776: MatMPIAIJSetPreallocation(*PB,d_nz,d_nnz,o_nz,o_nnz);
778: for (i=0; i<B->mnls[rank]; i++) {
779: global_row = B->start_indices[rank]+i;
780: for (k=0; k<B->lines->len[i]; k++) {
781: global_col = B->lines->ptrs[i][k];
783: val = B->lines->A[i][k];
784: MatSetValues(*PB,1,&global_row,1,&global_col,&val,ADD_VALUES);
785: }
786: }
788: PetscFree(d_nnz);
789: PetscFree(o_nnz);
791: MatAssemblyBegin(*PB,MAT_FINAL_ASSEMBLY);
792: MatAssemblyEnd(*PB,MAT_FINAL_ASSEMBLY);
793: return(0);
794: }
796: /**********************************************************************/
798: /*
799: Converts from an SPAI vector v to a PETSc vec Pv.
800: */
801: PetscErrorCode ConvertVectorToVec(MPI_Comm comm,vector *v,Vec *Pv)
802: {
804: PetscMPIInt size,rank;
805: int m,M,i,*mnls,*start_indices,*global_indices;
808: MPI_Comm_size(comm,&size);
809: MPI_Comm_rank(comm,&rank);
811: m = v->mnl;
812: M = v->n;
815: VecCreateMPI(comm,m,M,Pv);
817: PetscMalloc1(size,&mnls);
818: MPI_Allgather(&v->mnl,1,MPI_INT,mnls,1,MPI_INT,comm);
820: PetscMalloc1(size,&start_indices);
822: start_indices[0] = 0;
823: for (i=1; i<size; i++) start_indices[i] = start_indices[i-1] +mnls[i-1];
825: PetscMalloc1(v->mnl,&global_indices);
826: for (i=0; i<v->mnl; i++) global_indices[i] = start_indices[rank] + i;
828: PetscFree(mnls);
829: PetscFree(start_indices);
831: VecSetValues(*Pv,v->mnl,global_indices,v->v,INSERT_VALUES);
832: VecAssemblyBegin(*Pv);
833: VecAssemblyEnd(*Pv);
835: PetscFree(global_indices);
836: return(0);
837: }