Actual source code: mpisbaij.c
1: /*$Id: mpisbaij.c,v 1.61 2001/08/10 03:31:37 bsmith Exp $*/
3: #include src/mat/impls/baij/mpi/mpibaij.h
4: #include src/vec/vecimpl.h
5: #include mpisbaij.h
6: #include src/mat/impls/sbaij/seq/sbaij.h
8: extern int MatSetUpMultiply_MPISBAIJ(Mat);
9: extern int MatSetUpMultiply_MPISBAIJ_2comm(Mat);
10: extern int DisAssemble_MPISBAIJ(Mat);
11: extern int MatGetValues_SeqSBAIJ(Mat,int,const int[],int,const int[],PetscScalar []);
12: extern int MatGetValues_SeqBAIJ(Mat,int,const int[],int,const int[],PetscScalar []);
13: extern int MatSetValues_SeqSBAIJ(Mat,int,const int [],int,const int [],const PetscScalar [],InsertMode);
14: extern int MatSetValuesBlocked_SeqSBAIJ(Mat,int,const int[],int,const int[],const PetscScalar[],InsertMode);
15: extern int MatSetValuesBlocked_SeqBAIJ(Mat,int,const int[],int,const int[],const PetscScalar[],InsertMode);
16: extern int MatGetRow_SeqSBAIJ(Mat,int,int*,int**,PetscScalar**);
17: extern int MatRestoreRow_SeqSBAIJ(Mat,int,int*,int**,PetscScalar**);
18: extern int MatPrintHelp_SeqSBAIJ(Mat);
19: extern int MatZeroRows_SeqSBAIJ(Mat,IS,PetscScalar*);
20: extern int MatZeroRows_SeqBAIJ(Mat,IS,PetscScalar *);
21: extern int MatGetRowMax_MPISBAIJ(Mat,Vec);
22: extern int MatRelax_MPISBAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,int,int,Vec);
24: /* UGLY, ugly, ugly
25: When MatScalar == PetscScalar the function MatSetValuesBlocked_MPIBAIJ_MatScalar() does
26: not exist. Otherwise ..._MatScalar() takes matrix elements in single precision and
27: inserts them into the single precision data structure. The function MatSetValuesBlocked_MPIBAIJ()
28: converts the entries into single precision and then calls ..._MatScalar() to put them
29: into the single precision data structures.
30: */
31: #if defined(PETSC_USE_MAT_SINGLE)
32: extern int MatSetValuesBlocked_SeqSBAIJ_MatScalar(Mat,int,const int[],int,const int[],const MatScalar[],InsertMode);
33: extern int MatSetValues_MPISBAIJ_MatScalar(Mat,int,const int[],int,const int[],const MatScalar[],InsertMode);
34: extern int MatSetValuesBlocked_MPISBAIJ_MatScalar(Mat,int,const int[],int,const int[],const MatScalar[],InsertMode);
35: extern int MatSetValues_MPISBAIJ_HT_MatScalar(Mat,int,const int[],int,const int[],const MatScalar[],InsertMode);
36: extern int MatSetValuesBlocked_MPISBAIJ_HT_MatScalar(Mat,int,const int[],int,const int[],const MatScalar[],InsertMode);
37: #else
38: #define MatSetValuesBlocked_SeqSBAIJ_MatScalar MatSetValuesBlocked_SeqSBAIJ
39: #define MatSetValues_MPISBAIJ_MatScalar MatSetValues_MPISBAIJ
40: #define MatSetValuesBlocked_MPISBAIJ_MatScalar MatSetValuesBlocked_MPISBAIJ
41: #define MatSetValues_MPISBAIJ_HT_MatScalar MatSetValues_MPISBAIJ_HT
42: #define MatSetValuesBlocked_MPISBAIJ_HT_MatScalar MatSetValuesBlocked_MPISBAIJ_HT
43: #endif
45: EXTERN_C_BEGIN
48: int MatStoreValues_MPISBAIJ(Mat mat)
49: {
50: Mat_MPISBAIJ *aij = (Mat_MPISBAIJ *)mat->data;
51: int ierr;
54: MatStoreValues(aij->A);
55: MatStoreValues(aij->B);
56: return(0);
57: }
58: EXTERN_C_END
60: EXTERN_C_BEGIN
63: int MatRetrieveValues_MPISBAIJ(Mat mat)
64: {
65: Mat_MPISBAIJ *aij = (Mat_MPISBAIJ *)mat->data;
66: int ierr;
69: MatRetrieveValues(aij->A);
70: MatRetrieveValues(aij->B);
71: return(0);
72: }
73: EXTERN_C_END
75: /*
76: Local utility routine that creates a mapping from the global column
77: number to the local number in the off-diagonal part of the local
78: storage of the matrix. This is done in a non scable way since the
79: length of colmap equals the global matrix length.
80: */
83: static int CreateColmap_MPISBAIJ_Private(Mat mat)
84: {
86: SETERRQ(1,"Function not yet written for SBAIJ format");
87: /* return(0); */
88: }
90: #define CHUNKSIZE 10
92: #define MatSetValues_SeqSBAIJ_A_Private(row,col,value,addv) \
93: { \
94: \
95: brow = row/bs; \
96: rp = aj + ai[brow]; ap = aa + bs2*ai[brow]; \
97: rmax = aimax[brow]; nrow = ailen[brow]; \
98: bcol = col/bs; \
99: ridx = row % bs; cidx = col % bs; \
100: low = 0; high = nrow; \
101: while (high-low > 3) { \
102: t = (low+high)/2; \
103: if (rp[t] > bcol) high = t; \
104: else low = t; \
105: } \
106: for (_i=low; _i<high; _i++) { \
107: if (rp[_i] > bcol) break; \
108: if (rp[_i] == bcol) { \
109: bap = ap + bs2*_i + bs*cidx + ridx; \
110: if (addv == ADD_VALUES) *bap += value; \
111: else *bap = value; \
112: goto a_noinsert; \
113: } \
114: } \
115: if (a->nonew == 1) goto a_noinsert; \
116: else if (a->nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%d, %d) into matrix", row, col); \
117: if (nrow >= rmax) { \
118: /* there is no extra room in row, therefore enlarge */ \
119: int new_nz = ai[a->mbs] + CHUNKSIZE,len,*new_i,*new_j; \
120: MatScalar *new_a; \
121: \
122: if (a->nonew == -2) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%d, %d) in the matrix", row, col); \
123: \
124: /* malloc new storage space */ \
125: len = new_nz*(sizeof(int)+bs2*sizeof(MatScalar))+(a->mbs+1)*sizeof(int); \
126: PetscMalloc(len,&new_a); \
127: new_j = (int*)(new_a + bs2*new_nz); \
128: new_i = new_j + new_nz; \
129: \
130: /* copy over old data into new slots */ \
131: for (ii=0; ii<brow+1; ii++) {new_i[ii] = ai[ii];} \
132: for (ii=brow+1; ii<a->mbs+1; ii++) {new_i[ii] = ai[ii]+CHUNKSIZE;} \
133: PetscMemcpy(new_j,aj,(ai[brow]+nrow)*sizeof(int)); \
134: len = (new_nz - CHUNKSIZE - ai[brow] - nrow); \
135: PetscMemcpy(new_j+ai[brow]+nrow+CHUNKSIZE,aj+ai[brow]+nrow,len*sizeof(int)); \
136: PetscMemcpy(new_a,aa,(ai[brow]+nrow)*bs2*sizeof(MatScalar)); \
137: PetscMemzero(new_a+bs2*(ai[brow]+nrow),bs2*CHUNKSIZE*sizeof(PetscScalar)); \
138: PetscMemcpy(new_a+bs2*(ai[brow]+nrow+CHUNKSIZE), \
139: aa+bs2*(ai[brow]+nrow),bs2*len*sizeof(MatScalar)); \
140: /* free up old matrix storage */ \
141: PetscFree(a->a); \
142: if (!a->singlemalloc) { \
143: PetscFree(a->i); \
144: PetscFree(a->j);\
145: } \
146: aa = a->a = new_a; ai = a->i = new_i; aj = a->j = new_j; \
147: a->singlemalloc = PETSC_TRUE; \
148: \
149: rp = aj + ai[brow]; ap = aa + bs2*ai[brow]; \
150: rmax = aimax[brow] = aimax[brow] + CHUNKSIZE; \
151: PetscLogObjectMemory(A,CHUNKSIZE*(sizeof(int) + bs2*sizeof(MatScalar))); \
152: a->s_maxnz += bs2*CHUNKSIZE; \
153: a->reallocs++; \
154: a->s_nz++; \
155: } \
156: N = nrow++ - 1; \
157: /* shift up all the later entries in this row */ \
158: for (ii=N; ii>=_i; ii--) { \
159: rp[ii+1] = rp[ii]; \
160: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar)); \
161: } \
162: if (N>=_i) { PetscMemzero(ap+bs2*_i,bs2*sizeof(MatScalar)); } \
163: rp[_i] = bcol; \
164: ap[bs2*_i + bs*cidx + ridx] = value; \
165: a_noinsert:; \
166: ailen[brow] = nrow; \
167: }
168: #ifndef MatSetValues_SeqBAIJ_B_Private
169: #define MatSetValues_SeqSBAIJ_B_Private(row,col,value,addv) \
170: { \
171: brow = row/bs; \
172: rp = bj + bi[brow]; ap = ba + bs2*bi[brow]; \
173: rmax = bimax[brow]; nrow = bilen[brow]; \
174: bcol = col/bs; \
175: ridx = row % bs; cidx = col % bs; \
176: low = 0; high = nrow; \
177: while (high-low > 3) { \
178: t = (low+high)/2; \
179: if (rp[t] > bcol) high = t; \
180: else low = t; \
181: } \
182: for (_i=low; _i<high; _i++) { \
183: if (rp[_i] > bcol) break; \
184: if (rp[_i] == bcol) { \
185: bap = ap + bs2*_i + bs*cidx + ridx; \
186: if (addv == ADD_VALUES) *bap += value; \
187: else *bap = value; \
188: goto b_noinsert; \
189: } \
190: } \
191: if (b->nonew == 1) goto b_noinsert; \
192: else if (b->nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%d, %d) into matrix", row, col); \
193: if (nrow >= rmax) { \
194: /* there is no extra room in row, therefore enlarge */ \
195: int new_nz = bi[b->mbs] + CHUNKSIZE,len,*new_i,*new_j; \
196: MatScalar *new_a; \
197: \
198: if (b->nonew == -2) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%d, %d) in the matrix", row, col); \
199: \
200: /* malloc new storage space */ \
201: len = new_nz*(sizeof(int)+bs2*sizeof(MatScalar))+(b->mbs+1)*sizeof(int); \
202: PetscMalloc(len,&new_a); \
203: new_j = (int*)(new_a + bs2*new_nz); \
204: new_i = new_j + new_nz; \
205: \
206: /* copy over old data into new slots */ \
207: for (ii=0; ii<brow+1; ii++) {new_i[ii] = bi[ii];} \
208: for (ii=brow+1; ii<b->mbs+1; ii++) {new_i[ii] = bi[ii]+CHUNKSIZE;} \
209: PetscMemcpy(new_j,bj,(bi[brow]+nrow)*sizeof(int)); \
210: len = (new_nz - CHUNKSIZE - bi[brow] - nrow); \
211: PetscMemcpy(new_j+bi[brow]+nrow+CHUNKSIZE,bj+bi[brow]+nrow,len*sizeof(int)); \
212: PetscMemcpy(new_a,ba,(bi[brow]+nrow)*bs2*sizeof(MatScalar)); \
213: PetscMemzero(new_a+bs2*(bi[brow]+nrow),bs2*CHUNKSIZE*sizeof(MatScalar)); \
214: PetscMemcpy(new_a+bs2*(bi[brow]+nrow+CHUNKSIZE), \
215: ba+bs2*(bi[brow]+nrow),bs2*len*sizeof(MatScalar)); \
216: /* free up old matrix storage */ \
217: PetscFree(b->a); \
218: if (!b->singlemalloc) { \
219: PetscFree(b->i); \
220: PetscFree(b->j); \
221: } \
222: ba = b->a = new_a; bi = b->i = new_i; bj = b->j = new_j; \
223: b->singlemalloc = PETSC_TRUE; \
224: \
225: rp = bj + bi[brow]; ap = ba + bs2*bi[brow]; \
226: rmax = bimax[brow] = bimax[brow] + CHUNKSIZE; \
227: PetscLogObjectMemory(B,CHUNKSIZE*(sizeof(int) + bs2*sizeof(MatScalar))); \
228: b->maxnz += bs2*CHUNKSIZE; \
229: b->reallocs++; \
230: b->nz++; \
231: } \
232: N = nrow++ - 1; \
233: /* shift up all the later entries in this row */ \
234: for (ii=N; ii>=_i; ii--) { \
235: rp[ii+1] = rp[ii]; \
236: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar)); \
237: } \
238: if (N>=_i) { PetscMemzero(ap+bs2*_i,bs2*sizeof(MatScalar));} \
239: rp[_i] = bcol; \
240: ap[bs2*_i + bs*cidx + ridx] = value; \
241: b_noinsert:; \
242: bilen[brow] = nrow; \
243: }
244: #endif
246: #if defined(PETSC_USE_MAT_SINGLE)
249: int MatSetValues_MPISBAIJ(Mat mat,int m,const int im[],int n,const int in[],const PetscScalar v[],InsertMode addv)
250: {
251: Mat_MPISBAIJ *b = (Mat_MPISBAIJ*)mat->data;
252: int ierr,i,N = m*n;
253: MatScalar *vsingle;
256: if (N > b->setvalueslen) {
257: if (b->setvaluescopy) {PetscFree(b->setvaluescopy);}
258: PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);
259: b->setvalueslen = N;
260: }
261: vsingle = b->setvaluescopy;
263: for (i=0; i<N; i++) {
264: vsingle[i] = v[i];
265: }
266: MatSetValues_MPISBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);
267: return(0);
268: }
272: int MatSetValuesBlocked_MPISBAIJ(Mat mat,int m,const int im[],int n,const int in[],const PetscScalar v[],InsertMode addv)
273: {
274: Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
275: int ierr,i,N = m*n*b->bs2;
276: MatScalar *vsingle;
279: if (N > b->setvalueslen) {
280: if (b->setvaluescopy) {PetscFree(b->setvaluescopy);}
281: PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);
282: b->setvalueslen = N;
283: }
284: vsingle = b->setvaluescopy;
285: for (i=0; i<N; i++) {
286: vsingle[i] = v[i];
287: }
288: MatSetValuesBlocked_MPISBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);
289: return(0);
290: }
294: int MatSetValues_MPISBAIJ_HT(Mat mat,int m,const int im[],int n,const int in[],const PetscScalar v[],InsertMode addv)
295: {
296: Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
297: int ierr,i,N = m*n;
298: MatScalar *vsingle;
301: SETERRQ(1,"Function not yet written for SBAIJ format");
302: /* return(0); */
303: }
307: int MatSetValuesBlocked_MPISBAIJ_HT(Mat mat,int m,const int im[],int n,const int in[],const PetscScalar v[],InsertMode addv)
308: {
309: Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
310: int ierr,i,N = m*n*b->bs2;
311: MatScalar *vsingle;
314: SETERRQ(1,"Function not yet written for SBAIJ format");
315: /* return(0); */
316: }
317: #endif
319: /* Only add/insert a(i,j) with i<=j (blocks).
320: Any a(i,j) with i>j input by user is ingored.
321: */
324: int MatSetValues_MPISBAIJ_MatScalar(Mat mat,int m,const int im[],int n,const int in[],const MatScalar v[],InsertMode addv)
325: {
326: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
327: MatScalar value;
328: PetscTruth roworiented = baij->roworiented;
329: int ierr,i,j,row,col;
330: int rstart_orig=baij->rstart_bs;
331: int rend_orig=baij->rend_bs,cstart_orig=baij->cstart_bs;
332: int cend_orig=baij->cend_bs,bs=baij->bs;
334: /* Some Variables required in the macro */
335: Mat A = baij->A;
336: Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)(A)->data;
337: int *aimax=a->imax,*ai=a->i,*ailen=a->ilen,*aj=a->j;
338: MatScalar *aa=a->a;
340: Mat B = baij->B;
341: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)(B)->data;
342: int *bimax=b->imax,*bi=b->i,*bilen=b->ilen,*bj=b->j;
343: MatScalar *ba=b->a;
345: int *rp,ii,nrow,_i,rmax,N,brow,bcol;
346: int low,high,t,ridx,cidx,bs2=a->bs2;
347: MatScalar *ap,*bap;
349: /* for stash */
350: int n_loc, *in_loc=0;
351: MatScalar *v_loc=0;
355: if(!baij->donotstash){
356: PetscMalloc(n*sizeof(int),&in_loc);
357: PetscMalloc(n*sizeof(MatScalar),&v_loc);
358: }
360: for (i=0; i<m; i++) {
361: if (im[i] < 0) continue;
362: #if defined(PETSC_USE_BOPT_g)
363: if (im[i] >= mat->M) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %d max %d",im[i],mat->M-1);
364: #endif
365: if (im[i] >= rstart_orig && im[i] < rend_orig) { /* this processor entry */
366: row = im[i] - rstart_orig; /* local row index */
367: for (j=0; j<n; j++) {
368: if (im[i]/bs > in[j]/bs) continue; /* ignore lower triangular blocks */
369: if (in[j] >= cstart_orig && in[j] < cend_orig){ /* diag entry (A) */
370: col = in[j] - cstart_orig; /* local col index */
371: brow = row/bs; bcol = col/bs;
372: if (brow > bcol) continue; /* ignore lower triangular blocks of A */
373: if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
374: MatSetValues_SeqSBAIJ_A_Private(row,col,value,addv);
375: /* MatSetValues_SeqBAIJ(baij->A,1,&row,1,&col,&value,addv); */
376: } else if (in[j] < 0) continue;
377: #if defined(PETSC_USE_BOPT_g)
378: else if (in[j] >= mat->N) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %d max %d",in[j],mat->N-1);}
379: #endif
380: else { /* off-diag entry (B) */
381: if (mat->was_assembled) {
382: if (!baij->colmap) {
383: CreateColmap_MPISBAIJ_Private(mat);
384: }
385: #if defined (PETSC_USE_CTABLE)
386: PetscTableFind(baij->colmap,in[j]/bs + 1,&col);
387: col = col - 1;
388: #else
389: col = baij->colmap[in[j]/bs] - 1;
390: #endif
391: if (col < 0 && !((Mat_SeqSBAIJ*)(baij->A->data))->nonew) {
392: DisAssemble_MPISBAIJ(mat);
393: col = in[j];
394: /* Reinitialize the variables required by MatSetValues_SeqBAIJ_B_Private() */
395: B = baij->B;
396: b = (Mat_SeqBAIJ*)(B)->data;
397: bimax=b->imax;bi=b->i;bilen=b->ilen;bj=b->j;
398: ba=b->a;
399: } else col += in[j]%bs;
400: } else col = in[j];
401: if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
402: MatSetValues_SeqSBAIJ_B_Private(row,col,value,addv);
403: /* MatSetValues_SeqBAIJ(baij->B,1,&row,1,&col,&value,addv); */
404: }
405: }
406: } else { /* off processor entry */
407: if (!baij->donotstash) {
408: n_loc = 0;
409: for (j=0; j<n; j++){
410: if (im[i]/bs > in[j]/bs) continue; /* ignore lower triangular blocks */
411: in_loc[n_loc] = in[j];
412: if (roworiented) {
413: v_loc[n_loc] = v[i*n+j];
414: } else {
415: v_loc[n_loc] = v[j*m+i];
416: }
417: n_loc++;
418: }
419: MatStashValuesRow_Private(&mat->stash,im[i],n_loc,in_loc,v_loc);
420: }
421: }
422: }
424: if(!baij->donotstash){
425: PetscFree(in_loc);
426: PetscFree(v_loc);
427: }
428: return(0);
429: }
433: int MatSetValuesBlocked_MPISBAIJ_MatScalar(Mat mat,int m,const int im[],int n,const int in[],const MatScalar v[],InsertMode addv)
434: {
435: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
436: const MatScalar *value;
437: MatScalar *barray=baij->barray;
438: PetscTruth roworiented = baij->roworiented;
439: int ierr,i,j,ii,jj,row,col,rstart=baij->rstart;
440: int rend=baij->rend,cstart=baij->cstart,stepval;
441: int cend=baij->cend,bs=baij->bs,bs2=baij->bs2;
444: if(!barray) {
445: PetscMalloc(bs2*sizeof(MatScalar),&barray);
446: baij->barray = barray;
447: }
449: if (roworiented) {
450: stepval = (n-1)*bs;
451: } else {
452: stepval = (m-1)*bs;
453: }
454: for (i=0; i<m; i++) {
455: if (im[i] < 0) continue;
456: #if defined(PETSC_USE_BOPT_g)
457: if (im[i] >= baij->Mbs) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large, row %d max %d",im[i],baij->Mbs-1);
458: #endif
459: if (im[i] >= rstart && im[i] < rend) {
460: row = im[i] - rstart;
461: for (j=0; j<n; j++) {
462: /* If NumCol = 1 then a copy is not required */
463: if ((roworiented) && (n == 1)) {
464: barray = (MatScalar*) v + i*bs2;
465: } else if((!roworiented) && (m == 1)) {
466: barray = (MatScalar*) v + j*bs2;
467: } else { /* Here a copy is required */
468: if (roworiented) {
469: value = v + i*(stepval+bs)*bs + j*bs;
470: } else {
471: value = v + j*(stepval+bs)*bs + i*bs;
472: }
473: for (ii=0; ii<bs; ii++,value+=stepval) {
474: for (jj=0; jj<bs; jj++) {
475: *barray++ = *value++;
476: }
477: }
478: barray -=bs2;
479: }
480:
481: if (in[j] >= cstart && in[j] < cend){
482: col = in[j] - cstart;
483: MatSetValuesBlocked_SeqSBAIJ(baij->A,1,&row,1,&col,barray,addv);
484: }
485: else if (in[j] < 0) continue;
486: #if defined(PETSC_USE_BOPT_g)
487: else if (in[j] >= baij->Nbs) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large, col %d max %d",in[j],baij->Nbs-1);}
488: #endif
489: else {
490: if (mat->was_assembled) {
491: if (!baij->colmap) {
492: CreateColmap_MPISBAIJ_Private(mat);
493: }
495: #if defined(PETSC_USE_BOPT_g)
496: #if defined (PETSC_USE_CTABLE)
497: { int data;
498: PetscTableFind(baij->colmap,in[j]+1,&data);
499: if ((data - 1) % bs) SETERRQ(PETSC_ERR_PLIB,"Incorrect colmap");
500: }
501: #else
502: if ((baij->colmap[in[j]] - 1) % bs) SETERRQ(PETSC_ERR_PLIB,"Incorrect colmap");
503: #endif
504: #endif
505: #if defined (PETSC_USE_CTABLE)
506: PetscTableFind(baij->colmap,in[j]+1,&col);
507: col = (col - 1)/bs;
508: #else
509: col = (baij->colmap[in[j]] - 1)/bs;
510: #endif
511: if (col < 0 && !((Mat_SeqBAIJ*)(baij->A->data))->nonew) {
512: DisAssemble_MPISBAIJ(mat);
513: col = in[j];
514: }
515: }
516: else col = in[j];
517: MatSetValuesBlocked_SeqBAIJ(baij->B,1,&row,1,&col,barray,addv);
518: }
519: }
520: } else {
521: if (!baij->donotstash) {
522: if (roworiented) {
523: MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
524: } else {
525: MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
526: }
527: }
528: }
529: }
530: return(0);
531: }
533: #define HASH_KEY 0.6180339887
534: #define HASH(size,key,tmp) (tmp = (key)*HASH_KEY,(int)((size)*(tmp-(int)tmp)))
535: /* #define HASH(size,key) ((int)((size)*fmod(((key)*HASH_KEY),1))) */
536: /* #define HASH(size,key,tmp) ((int)((size)*fmod(((key)*HASH_KEY),1))) */
539: int MatSetValues_MPISBAIJ_HT_MatScalar(Mat mat,int m,const int im[],int n,const int in[],const MatScalar v[],InsertMode addv)
540: {
542: SETERRQ(1,"Function not yet written for SBAIJ format");
543: /* return(0); */
544: }
548: int MatSetValuesBlocked_MPISBAIJ_HT_MatScalar(Mat mat,int m,const int im[],int n,const int in[],const MatScalar v[],InsertMode addv)
549: {
551: SETERRQ(1,"Function not yet written for SBAIJ format");
552: /* return(0); */
553: }
557: int MatGetValues_MPISBAIJ(Mat mat,int m,const int idxm[],int n,const int idxn[],PetscScalar v[])
558: {
559: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
560: int bs=baij->bs,ierr,i,j,bsrstart = baij->rstart*bs,bsrend = baij->rend*bs;
561: int bscstart = baij->cstart*bs,bscend = baij->cend*bs,row,col,data;
564: for (i=0; i<m; i++) {
565: if (idxm[i] < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %d",idxm[i]);
566: if (idxm[i] >= mat->M) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %d max %d",idxm[i],mat->M-1);
567: if (idxm[i] >= bsrstart && idxm[i] < bsrend) {
568: row = idxm[i] - bsrstart;
569: for (j=0; j<n; j++) {
570: if (idxn[j] < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative column %d",idxn[j]);
571: if (idxn[j] >= mat->N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %d max %d",idxn[j],mat->N-1);
572: if (idxn[j] >= bscstart && idxn[j] < bscend){
573: col = idxn[j] - bscstart;
574: MatGetValues_SeqSBAIJ(baij->A,1,&row,1,&col,v+i*n+j);
575: } else {
576: if (!baij->colmap) {
577: CreateColmap_MPISBAIJ_Private(mat);
578: }
579: #if defined (PETSC_USE_CTABLE)
580: PetscTableFind(baij->colmap,idxn[j]/bs+1,&data);
581: data --;
582: #else
583: data = baij->colmap[idxn[j]/bs]-1;
584: #endif
585: if((data < 0) || (baij->garray[data/bs] != idxn[j]/bs)) *(v+i*n+j) = 0.0;
586: else {
587: col = data + idxn[j]%bs;
588: MatGetValues_SeqBAIJ(baij->B,1,&row,1,&col,v+i*n+j);
589: }
590: }
591: }
592: } else {
593: SETERRQ(PETSC_ERR_SUP,"Only local values currently supported");
594: }
595: }
596: return(0);
597: }
601: int MatNorm_MPISBAIJ(Mat mat,NormType type,PetscReal *norm)
602: {
603: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
604: /* Mat_SeqSBAIJ *amat = (Mat_SeqSBAIJ*)baij->A->data; */
605: /* Mat_SeqBAIJ *bmat = (Mat_SeqBAIJ*)baij->B->data; */
606: int ierr;
607: PetscReal sum[2],*lnorm2;
610: if (baij->size == 1) {
611: MatNorm(baij->A,type,norm);
612: } else {
613: if (type == NORM_FROBENIUS) {
614: PetscMalloc(2*sizeof(PetscReal),&lnorm2);
615: MatNorm(baij->A,type,lnorm2);
616: *lnorm2 = (*lnorm2)*(*lnorm2); lnorm2++; /* squar power of norm(A) */
617: MatNorm(baij->B,type,lnorm2);
618: *lnorm2 = (*lnorm2)*(*lnorm2); lnorm2--; /* squar power of norm(B) */
619: /*
620: MPI_Comm_rank(mat->comm,&rank);
621: PetscSynchronizedPrintf(mat->comm,"[%d], lnorm2=%g, %g\n",rank,lnorm2[0],lnorm2[1]);
622: */
623: MPI_Allreduce(lnorm2,&sum,2,MPIU_REAL,MPI_SUM,mat->comm);
624: /*
625: PetscSynchronizedPrintf(mat->comm,"[%d], sum=%g, %g\n",rank,sum[0],sum[1]);
626: PetscSynchronizedFlush(mat->comm); */
627:
628: *norm = sqrt(sum[0] + 2*sum[1]);
629: PetscFree(lnorm2);
630: } else {
631: SETERRQ(PETSC_ERR_SUP,"No support for this norm yet");
632: }
633: }
634: return(0);
635: }
637: /*
638: Creates the hash table, and sets the table
639: This table is created only once.
640: If new entried need to be added to the matrix
641: then the hash table has to be destroyed and
642: recreated.
643: */
646: int MatCreateHashTable_MPISBAIJ_Private(Mat mat,PetscReal factor)
647: {
649: SETERRQ(1,"Function not yet written for SBAIJ format");
650: /* return(0); */
651: }
655: int MatAssemblyBegin_MPISBAIJ(Mat mat,MatAssemblyType mode)
656: {
657: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
658: int ierr,nstash,reallocs;
659: InsertMode addv;
662: if (baij->donotstash) {
663: return(0);
664: }
666: /* make sure all processors are either in INSERTMODE or ADDMODE */
667: MPI_Allreduce(&mat->insertmode,&addv,1,MPI_INT,MPI_BOR,mat->comm);
668: if (addv == (ADD_VALUES|INSERT_VALUES)) {
669: SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Some processors inserted others added");
670: }
671: mat->insertmode = addv; /* in case this processor had no cache */
673: MatStashScatterBegin_Private(&mat->stash,baij->rowners_bs);
674: MatStashScatterBegin_Private(&mat->bstash,baij->rowners);
675: MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);
676: PetscLogInfo(0,"MatAssemblyBegin_MPISBAIJ:Stash has %d entries,uses %d mallocs.\n",nstash,reallocs);
677: MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);
678: PetscLogInfo(0,"MatAssemblyBegin_MPISBAIJ:Block-Stash has %d entries, uses %d mallocs.\n",nstash,reallocs);
679: return(0);
680: }
684: int MatAssemblyEnd_MPISBAIJ(Mat mat,MatAssemblyType mode)
685: {
686: Mat_MPISBAIJ *baij=(Mat_MPISBAIJ*)mat->data;
687: Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)baij->A->data;
688: Mat_SeqBAIJ *b=(Mat_SeqBAIJ*)baij->B->data;
689: int i,j,rstart,ncols,n,ierr,flg,bs2=baij->bs2;
690: int *row,*col,other_disassembled;
691: PetscTruth r1,r2,r3;
692: MatScalar *val;
693: InsertMode addv = mat->insertmode;
697: if (!baij->donotstash) {
698: while (1) {
699: MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);
700: /*
701: PetscSynchronizedPrintf(mat->comm,"[%d]: in AssemblyEnd, stash, flg=%d\n",rank,flg);
702: PetscSynchronizedFlush(mat->comm);
703: */
704: if (!flg) break;
706: for (i=0; i<n;) {
707: /* Now identify the consecutive vals belonging to the same row */
708: for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; }
709: if (j < n) ncols = j-i;
710: else ncols = n-i;
711: /* Now assemble all these values with a single function call */
712: MatSetValues_MPISBAIJ_MatScalar(mat,1,row+i,ncols,col+i,val+i,addv);
713: i = j;
714: }
715: }
716: MatStashScatterEnd_Private(&mat->stash);
717: /* Now process the block-stash. Since the values are stashed column-oriented,
718: set the roworiented flag to column oriented, and after MatSetValues()
719: restore the original flags */
720: r1 = baij->roworiented;
721: r2 = a->roworiented;
722: r3 = b->roworiented;
723: baij->roworiented = PETSC_FALSE;
724: a->roworiented = PETSC_FALSE;
725: b->roworiented = PETSC_FALSE;
726: while (1) {
727: MatStashScatterGetMesg_Private(&mat->bstash,&n,&row,&col,&val,&flg);
728: if (!flg) break;
729:
730: for (i=0; i<n;) {
731: /* Now identify the consecutive vals belonging to the same row */
732: for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; }
733: if (j < n) ncols = j-i;
734: else ncols = n-i;
735: MatSetValuesBlocked_MPISBAIJ_MatScalar(mat,1,row+i,ncols,col+i,val+i*bs2,addv);
736: i = j;
737: }
738: }
739: MatStashScatterEnd_Private(&mat->bstash);
740: baij->roworiented = r1;
741: a->roworiented = r2;
742: b->roworiented = r3;
743: }
745: MatAssemblyBegin(baij->A,mode);
746: MatAssemblyEnd(baij->A,mode);
748: /* determine if any processor has disassembled, if so we must
749: also disassemble ourselfs, in order that we may reassemble. */
750: /*
751: if nonzero structure of submatrix B cannot change then we know that
752: no processor disassembled thus we can skip this stuff
753: */
754: if (!((Mat_SeqBAIJ*)baij->B->data)->nonew) {
755: MPI_Allreduce(&mat->was_assembled,&other_disassembled,1,MPI_INT,MPI_PROD,mat->comm);
756: if (mat->was_assembled && !other_disassembled) {
757: DisAssemble_MPISBAIJ(mat);
758: }
759: }
761: if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) {
762: MatSetUpMultiply_MPISBAIJ(mat); /* setup Mvctx and sMvctx */
763: }
764: MatAssemblyBegin(baij->B,mode);
765: MatAssemblyEnd(baij->B,mode);
766:
767: #if defined(PETSC_USE_BOPT_g)
768: if (baij->ht && mode== MAT_FINAL_ASSEMBLY) {
769: PetscLogInfo(0,"MatAssemblyEnd_MPISBAIJ:Average Hash Table Search in MatSetValues = %5.2f\n",((PetscReal)baij->ht_total_ct)/baij->ht_insert_ct);
770: baij->ht_total_ct = 0;
771: baij->ht_insert_ct = 0;
772: }
773: #endif
774: if (baij->ht_flag && !baij->ht && mode == MAT_FINAL_ASSEMBLY) {
775: MatCreateHashTable_MPISBAIJ_Private(mat,baij->ht_fact);
776: mat->ops->setvalues = MatSetValues_MPISBAIJ_HT;
777: mat->ops->setvaluesblocked = MatSetValuesBlocked_MPISBAIJ_HT;
778: }
780: if (baij->rowvalues) {
781: PetscFree(baij->rowvalues);
782: baij->rowvalues = 0;
783: }
785: return(0);
786: }
790: static int MatView_MPISBAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer)
791: {
792: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
793: int ierr,bs = baij->bs,size = baij->size,rank = baij->rank;
794: PetscTruth isascii,isdraw;
795: PetscViewer sviewer;
796: PetscViewerFormat format;
799: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&isascii);
800: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
801: if (isascii) {
802: PetscViewerGetFormat(viewer,&format);
803: if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
804: MatInfo info;
805: MPI_Comm_rank(mat->comm,&rank);
806: MatGetInfo(mat,MAT_LOCAL,&info);
807: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %d nz %d nz alloced %d bs %d mem %d\n",
808: rank,mat->m,(int)info.nz_used*bs,(int)info.nz_allocated*bs,
809: baij->bs,(int)info.memory);
810: MatGetInfo(baij->A,MAT_LOCAL,&info);
811: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %d \n",rank,(int)info.nz_used*bs);
812: MatGetInfo(baij->B,MAT_LOCAL,&info);
813: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %d \n",rank,(int)info.nz_used*bs);
814: PetscViewerFlush(viewer);
815: VecScatterView(baij->Mvctx,viewer);
816: return(0);
817: } else if (format == PETSC_VIEWER_ASCII_INFO) {
818: PetscViewerASCIIPrintf(viewer," block size is %d\n",bs);
819: return(0);
820: }
821: }
823: if (isdraw) {
824: PetscDraw draw;
825: PetscTruth isnull;
826: PetscViewerDrawGetDraw(viewer,0,&draw);
827: PetscDrawIsNull(draw,&isnull); if (isnull) return(0);
828: }
830: if (size == 1) {
831: PetscObjectSetName((PetscObject)baij->A,mat->name);
832: MatView(baij->A,viewer);
833: } else {
834: /* assemble the entire matrix onto first processor. */
835: Mat A;
836: Mat_SeqSBAIJ *Aloc;
837: Mat_SeqBAIJ *Bloc;
838: int M = mat->M,N = mat->N,*ai,*aj,col,i,j,k,*rvals,mbs = baij->mbs;
839: MatScalar *a;
841: if (!rank) {
842: MatCreateMPISBAIJ(mat->comm,baij->bs,M,N,M,N,0,PETSC_NULL,0,PETSC_NULL,&A);
843: } else {
844: MatCreateMPISBAIJ(mat->comm,baij->bs,0,0,M,N,0,PETSC_NULL,0,PETSC_NULL,&A);
845: }
846: PetscLogObjectParent(mat,A);
848: /* copy over the A part */
849: Aloc = (Mat_SeqSBAIJ*)baij->A->data;
850: ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
851: PetscMalloc(bs*sizeof(int),&rvals);
853: for (i=0; i<mbs; i++) {
854: rvals[0] = bs*(baij->rstart + i);
855: for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
856: for (j=ai[i]; j<ai[i+1]; j++) {
857: col = (baij->cstart+aj[j])*bs;
858: for (k=0; k<bs; k++) {
859: MatSetValues_MPISBAIJ_MatScalar(A,bs,rvals,1,&col,a,INSERT_VALUES);
860: col++; a += bs;
861: }
862: }
863: }
864: /* copy over the B part */
865: Bloc = (Mat_SeqBAIJ*)baij->B->data;
866: ai = Bloc->i; aj = Bloc->j; a = Bloc->a;
867: for (i=0; i<mbs; i++) {
868: rvals[0] = bs*(baij->rstart + i);
869: for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
870: for (j=ai[i]; j<ai[i+1]; j++) {
871: col = baij->garray[aj[j]]*bs;
872: for (k=0; k<bs; k++) {
873: MatSetValues_MPISBAIJ_MatScalar(A,bs,rvals,1,&col,a,INSERT_VALUES);
874: col++; a += bs;
875: }
876: }
877: }
878: PetscFree(rvals);
879: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
880: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
881: /*
882: Everyone has to call to draw the matrix since the graphics waits are
883: synchronized across all processors that share the PetscDraw object
884: */
885: PetscViewerGetSingleton(viewer,&sviewer);
886: if (!rank) {
887: PetscObjectSetName((PetscObject)((Mat_MPISBAIJ*)(A->data))->A,mat->name);
888: MatView(((Mat_MPISBAIJ*)(A->data))->A,sviewer);
889: }
890: PetscViewerRestoreSingleton(viewer,&sviewer);
891: MatDestroy(A);
892: }
893: return(0);
894: }
898: int MatView_MPISBAIJ(Mat mat,PetscViewer viewer)
899: {
900: int ierr;
901: PetscTruth isascii,isdraw,issocket,isbinary;
904: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&isascii);
905: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
906: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_SOCKET,&issocket);
907: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);
908: if (isascii || isdraw || issocket || isbinary) {
909: MatView_MPISBAIJ_ASCIIorDraworSocket(mat,viewer);
910: } else {
911: SETERRQ1(1,"Viewer type %s not supported by MPISBAIJ matrices",((PetscObject)viewer)->type_name);
912: }
913: return(0);
914: }
918: int MatDestroy_MPISBAIJ(Mat mat)
919: {
920: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
921: int ierr;
924: #if defined(PETSC_USE_LOG)
925: PetscLogObjectState((PetscObject)mat,"Rows=%d,Cols=%d",mat->M,mat->N);
926: #endif
927: MatStashDestroy_Private(&mat->stash);
928: MatStashDestroy_Private(&mat->bstash);
929: PetscFree(baij->rowners);
930: MatDestroy(baij->A);
931: MatDestroy(baij->B);
932: #if defined (PETSC_USE_CTABLE)
933: if (baij->colmap) {PetscTableDelete(baij->colmap);}
934: #else
935: if (baij->colmap) {PetscFree(baij->colmap);}
936: #endif
937: if (baij->garray) {PetscFree(baij->garray);}
938: if (baij->lvec) {VecDestroy(baij->lvec);}
939: if (baij->Mvctx) {VecScatterDestroy(baij->Mvctx);}
940: if (baij->slvec0) {
941: VecDestroy(baij->slvec0);
942: VecDestroy(baij->slvec0b);
943: }
944: if (baij->slvec1) {
945: VecDestroy(baij->slvec1);
946: VecDestroy(baij->slvec1a);
947: VecDestroy(baij->slvec1b);
948: }
949: if (baij->sMvctx) {VecScatterDestroy(baij->sMvctx);}
950: if (baij->rowvalues) {PetscFree(baij->rowvalues);}
951: if (baij->barray) {PetscFree(baij->barray);}
952: if (baij->hd) {PetscFree(baij->hd);}
953: #if defined(PETSC_USE_MAT_SINGLE)
954: if (baij->setvaluescopy) {PetscFree(baij->setvaluescopy);}
955: #endif
956: PetscFree(baij);
957: return(0);
958: }
962: int MatMult_MPISBAIJ(Mat A,Vec xx,Vec yy)
963: {
964: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
965: int ierr,nt,mbs=a->mbs,bs=a->bs;
966: PetscScalar *x,*from,zero=0.0;
967:
969: /*
970: PetscSynchronizedPrintf(A->comm," _1comm is called ...\n");
971: PetscSynchronizedFlush(A->comm);
972: */
973: VecGetLocalSize(xx,&nt);
974: if (nt != A->n) {
975: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible partition of A and xx");
976: }
977: VecGetLocalSize(yy,&nt);
978: if (nt != A->m) {
979: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible parition of A and yy");
980: }
982: /* diagonal part */
983: (*a->A->ops->mult)(a->A,xx,a->slvec1a);
984: VecSet(&zero,a->slvec1b);
986: /* subdiagonal part */
987: (*a->B->ops->multtranspose)(a->B,xx,a->slvec0b);
989: /* copy x into the vec slvec0 */
990: VecGetArray(a->slvec0,&from);
991: VecGetArray(xx,&x);
992: PetscMemcpy(from,x,bs*mbs*sizeof(MatScalar));
993: VecRestoreArray(a->slvec0,&from);
994:
995: VecScatterBegin(a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD,a->sMvctx);
996: VecRestoreArray(xx,&x);
997: VecScatterEnd(a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD,a->sMvctx);
998:
999: /* supperdiagonal part */
1000: (*a->B->ops->multadd)(a->B,a->slvec1b,a->slvec1a,yy);
1001:
1002: return(0);
1003: }
1007: int MatMult_MPISBAIJ_2comm(Mat A,Vec xx,Vec yy)
1008: {
1009: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1010: int ierr,nt;
1013: VecGetLocalSize(xx,&nt);
1014: if (nt != A->n) {
1015: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible partition of A and xx");
1016: }
1017: VecGetLocalSize(yy,&nt);
1018: if (nt != A->m) {
1019: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible parition of A and yy");
1020: }
1022: VecScatterBegin(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);
1023: /* do diagonal part */
1024: (*a->A->ops->mult)(a->A,xx,yy);
1025: /* do supperdiagonal part */
1026: VecScatterEnd(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);
1027: (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);
1028: /* do subdiagonal part */
1029: (*a->B->ops->multtranspose)(a->B,xx,a->lvec);
1030: VecScatterBegin(a->lvec,yy,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);
1031: VecScatterEnd(a->lvec,yy,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);
1033: return(0);
1034: }
1038: int MatMultAdd_MPISBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1039: {
1040: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1041: int ierr,mbs=a->mbs,bs=a->bs;
1042: PetscScalar *x,*from,zero=0.0;
1043:
1045: /*
1046: PetscSynchronizedPrintf(A->comm," MatMultAdd is called ...\n");
1047: PetscSynchronizedFlush(A->comm);
1048: */
1049: /* diagonal part */
1050: (*a->A->ops->multadd)(a->A,xx,yy,a->slvec1a);
1051: VecSet(&zero,a->slvec1b);
1053: /* subdiagonal part */
1054: (*a->B->ops->multtranspose)(a->B,xx,a->slvec0b);
1056: /* copy x into the vec slvec0 */
1057: VecGetArray(a->slvec0,&from);
1058: VecGetArray(xx,&x);
1059: PetscMemcpy(from,x,bs*mbs*sizeof(MatScalar));
1060: VecRestoreArray(a->slvec0,&from);
1061:
1062: VecScatterBegin(a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD,a->sMvctx);
1063: VecRestoreArray(xx,&x);
1064: VecScatterEnd(a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD,a->sMvctx);
1065:
1066: /* supperdiagonal part */
1067: (*a->B->ops->multadd)(a->B,a->slvec1b,a->slvec1a,zz);
1068:
1069: return(0);
1070: }
1074: int MatMultAdd_MPISBAIJ_2comm(Mat A,Vec xx,Vec yy,Vec zz)
1075: {
1076: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1077: int ierr;
1080: VecScatterBegin(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);
1081: /* do diagonal part */
1082: (*a->A->ops->multadd)(a->A,xx,yy,zz);
1083: /* do supperdiagonal part */
1084: VecScatterEnd(xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD,a->Mvctx);
1085: (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);
1087: /* do subdiagonal part */
1088: (*a->B->ops->multtranspose)(a->B,xx,a->lvec);
1089: VecScatterBegin(a->lvec,zz,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);
1090: VecScatterEnd(a->lvec,zz,ADD_VALUES,SCATTER_REVERSE,a->Mvctx);
1092: return(0);
1093: }
1097: int MatMultTranspose_MPISBAIJ(Mat A,Vec xx,Vec yy)
1098: {
1100: SETERRQ(1,"Matrix is symmetric. Call MatMult().");
1101: /* return(0); */
1102: }
1106: int MatMultTransposeAdd_MPISBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1107: {
1109: SETERRQ(1,"Matrix is symmetric. Call MatMultAdd().");
1110: /* return(0); */
1111: }
1113: /*
1114: This only works correctly for square matrices where the subblock A->A is the
1115: diagonal block
1116: */
1119: int MatGetDiagonal_MPISBAIJ(Mat A,Vec v)
1120: {
1121: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1122: int ierr;
1125: /* if (a->M != a->N) SETERRQ(PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block"); */
1126: MatGetDiagonal(a->A,v);
1127: return(0);
1128: }
1132: int MatScale_MPISBAIJ(const PetscScalar *aa,Mat A)
1133: {
1134: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1135: int ierr;
1138: MatScale(aa,a->A);
1139: MatScale(aa,a->B);
1140: return(0);
1141: }
1145: int MatGetRow_MPISBAIJ(Mat matin,int row,int *nz,int **idx,PetscScalar **v)
1146: {
1147: Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)matin->data;
1148: PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p;
1149: int bs = mat->bs,bs2 = mat->bs2,i,ierr,*cworkA,*cworkB,**pcA,**pcB;
1150: int nztot,nzA,nzB,lrow,brstart = mat->rstart*bs,brend = mat->rend*bs;
1151: int *cmap,*idx_p,cstart = mat->cstart;
1154: if (mat->getrowactive == PETSC_TRUE) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Already active");
1155: mat->getrowactive = PETSC_TRUE;
1157: if (!mat->rowvalues && (idx || v)) {
1158: /*
1159: allocate enough space to hold information from the longest row.
1160: */
1161: Mat_SeqSBAIJ *Aa = (Mat_SeqSBAIJ*)mat->A->data;
1162: Mat_SeqBAIJ *Ba = (Mat_SeqBAIJ*)mat->B->data;
1163: int max = 1,mbs = mat->mbs,tmp;
1164: for (i=0; i<mbs; i++) {
1165: tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i]; /* row length */
1166: if (max < tmp) { max = tmp; }
1167: }
1168: PetscMalloc(max*bs2*(sizeof(int)+sizeof(PetscScalar)),&mat->rowvalues);
1169: mat->rowindices = (int*)(mat->rowvalues + max*bs2);
1170: }
1171:
1172: if (row < brstart || row >= brend) SETERRQ(PETSC_ERR_SUP,"Only local rows")
1173: lrow = row - brstart; /* local row index */
1175: pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB;
1176: if (!v) {pvA = 0; pvB = 0;}
1177: if (!idx) {pcA = 0; if (!v) pcB = 0;}
1178: (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);
1179: (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);
1180: nztot = nzA + nzB;
1182: cmap = mat->garray;
1183: if (v || idx) {
1184: if (nztot) {
1185: /* Sort by increasing column numbers, assuming A and B already sorted */
1186: int imark = -1;
1187: if (v) {
1188: *v = v_p = mat->rowvalues;
1189: for (i=0; i<nzB; i++) {
1190: if (cmap[cworkB[i]/bs] < cstart) v_p[i] = vworkB[i];
1191: else break;
1192: }
1193: imark = i;
1194: for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i];
1195: for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i];
1196: }
1197: if (idx) {
1198: *idx = idx_p = mat->rowindices;
1199: if (imark > -1) {
1200: for (i=0; i<imark; i++) {
1201: idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs;
1202: }
1203: } else {
1204: for (i=0; i<nzB; i++) {
1205: if (cmap[cworkB[i]/bs] < cstart)
1206: idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1207: else break;
1208: }
1209: imark = i;
1210: }
1211: for (i=0; i<nzA; i++) idx_p[imark+i] = cstart*bs + cworkA[i];
1212: for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1213: }
1214: } else {
1215: if (idx) *idx = 0;
1216: if (v) *v = 0;
1217: }
1218: }
1219: *nz = nztot;
1220: (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);
1221: (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);
1222: return(0);
1223: }
1227: int MatRestoreRow_MPISBAIJ(Mat mat,int row,int *nz,int **idx,PetscScalar **v)
1228: {
1229: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
1232: if (baij->getrowactive == PETSC_FALSE) {
1233: SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"MatGetRow not called");
1234: }
1235: baij->getrowactive = PETSC_FALSE;
1236: return(0);
1237: }
1241: int MatGetBlockSize_MPISBAIJ(Mat mat,int *bs)
1242: {
1243: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
1246: *bs = baij->bs;
1247: return(0);
1248: }
1252: int MatZeroEntries_MPISBAIJ(Mat A)
1253: {
1254: Mat_MPISBAIJ *l = (Mat_MPISBAIJ*)A->data;
1255: int ierr;
1258: MatZeroEntries(l->A);
1259: MatZeroEntries(l->B);
1260: return(0);
1261: }
1265: int MatGetInfo_MPISBAIJ(Mat matin,MatInfoType flag,MatInfo *info)
1266: {
1267: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)matin->data;
1268: Mat A = a->A,B = a->B;
1269: int ierr;
1270: PetscReal isend[5],irecv[5];
1273: info->block_size = (PetscReal)a->bs;
1274: MatGetInfo(A,MAT_LOCAL,info);
1275: isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded;
1276: isend[3] = info->memory; isend[4] = info->mallocs;
1277: MatGetInfo(B,MAT_LOCAL,info);
1278: isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded;
1279: isend[3] += info->memory; isend[4] += info->mallocs;
1280: if (flag == MAT_LOCAL) {
1281: info->nz_used = isend[0];
1282: info->nz_allocated = isend[1];
1283: info->nz_unneeded = isend[2];
1284: info->memory = isend[3];
1285: info->mallocs = isend[4];
1286: } else if (flag == MAT_GLOBAL_MAX) {
1287: MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_MAX,matin->comm);
1288: info->nz_used = irecv[0];
1289: info->nz_allocated = irecv[1];
1290: info->nz_unneeded = irecv[2];
1291: info->memory = irecv[3];
1292: info->mallocs = irecv[4];
1293: } else if (flag == MAT_GLOBAL_SUM) {
1294: MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_SUM,matin->comm);
1295: info->nz_used = irecv[0];
1296: info->nz_allocated = irecv[1];
1297: info->nz_unneeded = irecv[2];
1298: info->memory = irecv[3];
1299: info->mallocs = irecv[4];
1300: } else {
1301: SETERRQ1(1,"Unknown MatInfoType argument %d",flag);
1302: }
1303: info->rows_global = (PetscReal)A->M;
1304: info->columns_global = (PetscReal)A->N;
1305: info->rows_local = (PetscReal)A->m;
1306: info->columns_local = (PetscReal)A->N;
1307: info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */
1308: info->fill_ratio_needed = 0;
1309: info->factor_mallocs = 0;
1310: return(0);
1311: }
1315: int MatSetOption_MPISBAIJ(Mat A,MatOption op)
1316: {
1317: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1318: int ierr;
1321: switch (op) {
1322: case MAT_NO_NEW_NONZERO_LOCATIONS:
1323: case MAT_YES_NEW_NONZERO_LOCATIONS:
1324: case MAT_COLUMNS_UNSORTED:
1325: case MAT_COLUMNS_SORTED:
1326: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1327: case MAT_KEEP_ZEROED_ROWS:
1328: case MAT_NEW_NONZERO_LOCATION_ERR:
1329: MatSetOption(a->A,op);
1330: MatSetOption(a->B,op);
1331: break;
1332: case MAT_ROW_ORIENTED:
1333: a->roworiented = PETSC_TRUE;
1334: MatSetOption(a->A,op);
1335: MatSetOption(a->B,op);
1336: break;
1337: case MAT_ROWS_SORTED:
1338: case MAT_ROWS_UNSORTED:
1339: case MAT_YES_NEW_DIAGONALS:
1340: PetscLogInfo(A,"Info:MatSetOption_MPIBAIJ:Option ignored\n");
1341: break;
1342: case MAT_COLUMN_ORIENTED:
1343: a->roworiented = PETSC_FALSE;
1344: MatSetOption(a->A,op);
1345: MatSetOption(a->B,op);
1346: break;
1347: case MAT_IGNORE_OFF_PROC_ENTRIES:
1348: a->donotstash = PETSC_TRUE;
1349: break;
1350: case MAT_NO_NEW_DIAGONALS:
1351: SETERRQ(PETSC_ERR_SUP,"MAT_NO_NEW_DIAGONALS");
1352: case MAT_USE_HASH_TABLE:
1353: a->ht_flag = PETSC_TRUE;
1354: break;
1355: default:
1356: SETERRQ(PETSC_ERR_SUP,"unknown option");
1357: }
1358: return(0);
1359: }
1363: int MatTranspose_MPISBAIJ(Mat A,Mat *B)
1364: {
1367: MatDuplicate(A,MAT_COPY_VALUES,B);
1368: return(0);
1369: }
1373: int MatDiagonalScale_MPISBAIJ(Mat mat,Vec ll,Vec rr)
1374: {
1375: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
1376: Mat a = baij->A,b = baij->B;
1377: int ierr,s1,s2,s3;
1380: if (ll != rr) {
1381: SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"For symmetric format, left and right scaling vectors must be same\n");
1382: }
1383: MatGetLocalSize(mat,&s2,&s3);
1384: if (rr) {
1385: VecGetLocalSize(rr,&s1);
1386: if (s1!=s3) SETERRQ(PETSC_ERR_ARG_SIZ,"right vector non-conforming local size");
1387: /* Overlap communication with computation. */
1388: VecScatterBegin(rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD,baij->Mvctx);
1389: /*} if (ll) { */
1390: VecGetLocalSize(ll,&s1);
1391: if (s1!=s2) SETERRQ(PETSC_ERR_ARG_SIZ,"left vector non-conforming local size");
1392: (*b->ops->diagonalscale)(b,ll,PETSC_NULL);
1393: /* } */
1394: /* scale the diagonal block */
1395: (*a->ops->diagonalscale)(a,ll,rr);
1397: /* if (rr) { */
1398: /* Do a scatter end and then right scale the off-diagonal block */
1399: VecScatterEnd(rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD,baij->Mvctx);
1400: (*b->ops->diagonalscale)(b,PETSC_NULL,baij->lvec);
1401: }
1402:
1403: return(0);
1404: }
1408: int MatZeroRows_MPISBAIJ(Mat A,IS is,const PetscScalar *diag)
1409: {
1411: SETERRQ(PETSC_ERR_SUP,"No support for this function yet");
1412: }
1416: int MatPrintHelp_MPISBAIJ(Mat A)
1417: {
1418: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1419: MPI_Comm comm = A->comm;
1420: static int called = 0;
1421: int ierr;
1424: if (!a->rank) {
1425: MatPrintHelp_SeqSBAIJ(a->A);
1426: }
1427: if (called) {return(0);} else called = 1;
1428: (*PetscHelpPrintf)(comm," Options for MATMPISBAIJ matrix format (the defaults):\n");
1429: (*PetscHelpPrintf)(comm," -mat_use_hash_table <factor>: Use hashtable for efficient matrix assembly\n");
1430: return(0);
1431: }
1435: int MatSetUnfactored_MPISBAIJ(Mat A)
1436: {
1437: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1438: int ierr;
1441: MatSetUnfactored(a->A);
1442: return(0);
1443: }
1445: static int MatDuplicate_MPISBAIJ(Mat,MatDuplicateOption,Mat *);
1449: int MatEqual_MPISBAIJ(Mat A,Mat B,PetscTruth *flag)
1450: {
1451: Mat_MPISBAIJ *matB = (Mat_MPISBAIJ*)B->data,*matA = (Mat_MPISBAIJ*)A->data;
1452: Mat a,b,c,d;
1453: PetscTruth flg;
1454: int ierr;
1457: a = matA->A; b = matA->B;
1458: c = matB->A; d = matB->B;
1460: MatEqual(a,c,&flg);
1461: if (flg == PETSC_TRUE) {
1462: MatEqual(b,d,&flg);
1463: }
1464: MPI_Allreduce(&flg,flag,1,MPI_INT,MPI_LAND,A->comm);
1465: return(0);
1466: }
1470: int MatSetUpPreallocation_MPISBAIJ(Mat A)
1471: {
1472: int ierr;
1475: MatMPISBAIJSetPreallocation(A,1,PETSC_DEFAULT,0,PETSC_DEFAULT,0);
1476: return(0);
1477: }
1478: /* -------------------------------------------------------------------*/
1479: static struct _MatOps MatOps_Values = {
1480: MatSetValues_MPISBAIJ,
1481: MatGetRow_MPISBAIJ,
1482: MatRestoreRow_MPISBAIJ,
1483: MatMult_MPISBAIJ,
1484: /* 4*/ MatMultAdd_MPISBAIJ,
1485: MatMultTranspose_MPISBAIJ,
1486: MatMultTransposeAdd_MPISBAIJ,
1487: 0,
1488: 0,
1489: 0,
1490: /*10*/ 0,
1491: 0,
1492: 0,
1493: MatRelax_MPISBAIJ,
1494: MatTranspose_MPISBAIJ,
1495: /*15*/ MatGetInfo_MPISBAIJ,
1496: MatEqual_MPISBAIJ,
1497: MatGetDiagonal_MPISBAIJ,
1498: MatDiagonalScale_MPISBAIJ,
1499: MatNorm_MPISBAIJ,
1500: /*20*/ MatAssemblyBegin_MPISBAIJ,
1501: MatAssemblyEnd_MPISBAIJ,
1502: 0,
1503: MatSetOption_MPISBAIJ,
1504: MatZeroEntries_MPISBAIJ,
1505: /*25*/ MatZeroRows_MPISBAIJ,
1506: 0,
1507: 0,
1508: 0,
1509: 0,
1510: /*30*/ MatSetUpPreallocation_MPISBAIJ,
1511: 0,
1512: 0,
1513: 0,
1514: 0,
1515: /*35*/ MatDuplicate_MPISBAIJ,
1516: 0,
1517: 0,
1518: 0,
1519: 0,
1520: /*40*/ 0,
1521: 0,
1522: 0,
1523: MatGetValues_MPISBAIJ,
1524: 0,
1525: /*45*/ MatPrintHelp_MPISBAIJ,
1526: MatScale_MPISBAIJ,
1527: 0,
1528: 0,
1529: 0,
1530: /*50*/ MatGetBlockSize_MPISBAIJ,
1531: 0,
1532: 0,
1533: 0,
1534: 0,
1535: /*55*/ 0,
1536: 0,
1537: MatSetUnfactored_MPISBAIJ,
1538: 0,
1539: MatSetValuesBlocked_MPISBAIJ,
1540: /*60*/ 0,
1541: 0,
1542: 0,
1543: MatGetPetscMaps_Petsc,
1544: 0,
1545: /*65*/ 0,
1546: 0,
1547: 0,
1548: 0,
1549: 0,
1550: /*70*/ MatGetRowMax_MPISBAIJ,
1551: 0,
1552: 0,
1553: 0,
1554: 0,
1555: /*75*/ 0,
1556: 0,
1557: 0,
1558: 0,
1559: 0,
1560: /*80*/ 0,
1561: 0,
1562: 0,
1563: 0,
1564: 0,
1565: /*85*/ MatLoad_MPISBAIJ
1566: };
1569: EXTERN_C_BEGIN
1572: int MatGetDiagonalBlock_MPISBAIJ(Mat A,PetscTruth *iscopy,MatReuse reuse,Mat *a)
1573: {
1575: *a = ((Mat_MPISBAIJ *)A->data)->A;
1576: *iscopy = PETSC_FALSE;
1577: return(0);
1578: }
1579: EXTERN_C_END
1581: EXTERN_C_BEGIN
1584: int MatMPISBAIJSetPreallocation_MPISBAIJ(Mat B,int bs,int d_nz,int *d_nnz,int o_nz,int *o_nnz)
1585: {
1586: Mat_MPISBAIJ *b;
1587: int ierr,i,mbs,Mbs;
1590: PetscOptionsGetInt(B->prefix,"-mat_block_size",&bs,PETSC_NULL);
1592: if (bs < 1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Invalid block size specified, must be positive");
1593: if (d_nz == PETSC_DECIDE || d_nz == PETSC_DEFAULT) d_nz = 3;
1594: if (o_nz == PETSC_DECIDE || o_nz == PETSC_DEFAULT) o_nz = 1;
1595: if (d_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"d_nz cannot be less than 0: value %d",d_nz);
1596: if (o_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"o_nz cannot be less than 0: value %d",o_nz);
1597: if (d_nnz) {
1598: for (i=0; i<B->m/bs; i++) {
1599: if (d_nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"d_nnz cannot be less than -1: local row %d value %d",i,d_nnz[i]);
1600: }
1601: }
1602: if (o_nnz) {
1603: for (i=0; i<B->m/bs; i++) {
1604: if (o_nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"o_nnz cannot be less than -1: local row %d value %d",i,o_nnz[i]);
1605: }
1606: }
1607: B->preallocated = PETSC_TRUE;
1608: PetscSplitOwnershipBlock(B->comm,bs,&B->m,&B->M);
1609: PetscSplitOwnershipBlock(B->comm,bs,&B->n,&B->N);
1610: PetscMapCreateMPI(B->comm,B->m,B->M,&B->rmap);
1611: PetscMapCreateMPI(B->comm,B->m,B->M,&B->cmap);
1613: b = (Mat_MPISBAIJ*)B->data;
1614: mbs = B->m/bs;
1615: Mbs = B->M/bs;
1616: if (mbs*bs != B->m) {
1617: SETERRQ2(PETSC_ERR_ARG_SIZ,"No of local rows %d must be divisible by blocksize %d",B->m,bs);
1618: }
1620: b->bs = bs;
1621: b->bs2 = bs*bs;
1622: b->mbs = mbs;
1623: b->nbs = mbs;
1624: b->Mbs = Mbs;
1625: b->Nbs = Mbs;
1627: MPI_Allgather(&b->mbs,1,MPI_INT,b->rowners+1,1,MPI_INT,B->comm);
1628: b->rowners[0] = 0;
1629: for (i=2; i<=b->size; i++) {
1630: b->rowners[i] += b->rowners[i-1];
1631: }
1632: b->rstart = b->rowners[b->rank];
1633: b->rend = b->rowners[b->rank+1];
1634: b->cstart = b->rstart;
1635: b->cend = b->rend;
1636: for (i=0; i<=b->size; i++) {
1637: b->rowners_bs[i] = b->rowners[i]*bs;
1638: }
1639: b->rstart_bs = b-> rstart*bs;
1640: b->rend_bs = b->rend*bs;
1641:
1642: b->cstart_bs = b->cstart*bs;
1643: b->cend_bs = b->cend*bs;
1644:
1646: MatCreateSeqSBAIJ(PETSC_COMM_SELF,bs,B->m,B->m,d_nz,d_nnz,&b->A);
1647: PetscLogObjectParent(B,b->A);
1648: MatCreateSeqBAIJ(PETSC_COMM_SELF,bs,B->m,B->M,o_nz,o_nnz,&b->B);
1649: PetscLogObjectParent(B,b->B);
1651: /* build cache for off array entries formed */
1652: MatStashCreate_Private(B->comm,bs,&B->bstash);
1654: return(0);
1655: }
1656: EXTERN_C_END
1658: /*MC
1659: MATMPISBAIJ - MATMPISBAIJ = "mpisbaij" - A matrix type to be used for distributed symmetric sparse block matrices,
1660: based on block compressed sparse row format. Only the upper triangular portion of the matrix is stored.
1662: Options Database Keys:
1663: . -mat_type mpisbaij - sets the matrix type to "mpisbaij" during a call to MatSetFromOptions()
1665: Level: beginner
1667: .seealso: MatCreateMPISBAIJ
1668: M*/
1670: EXTERN_C_BEGIN
1673: int MatCreate_MPISBAIJ(Mat B)
1674: {
1675: Mat_MPISBAIJ *b;
1676: int ierr;
1677: PetscTruth flg;
1681: PetscNew(Mat_MPISBAIJ,&b);
1682: B->data = (void*)b;
1683: PetscMemzero(b,sizeof(Mat_MPISBAIJ));
1684: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
1686: B->ops->destroy = MatDestroy_MPISBAIJ;
1687: B->ops->view = MatView_MPISBAIJ;
1688: B->mapping = 0;
1689: B->factor = 0;
1690: B->assembled = PETSC_FALSE;
1692: B->insertmode = NOT_SET_VALUES;
1693: MPI_Comm_rank(B->comm,&b->rank);
1694: MPI_Comm_size(B->comm,&b->size);
1696: /* build local table of row and column ownerships */
1697: PetscMalloc(3*(b->size+2)*sizeof(int),&b->rowners);
1698: b->cowners = b->rowners + b->size + 2;
1699: b->rowners_bs = b->cowners + b->size + 2;
1700: PetscLogObjectMemory(B,3*(b->size+2)*sizeof(int)+sizeof(struct _p_Mat)+sizeof(Mat_MPISBAIJ));
1702: /* build cache for off array entries formed */
1703: MatStashCreate_Private(B->comm,1,&B->stash);
1704: b->donotstash = PETSC_FALSE;
1705: b->colmap = PETSC_NULL;
1706: b->garray = PETSC_NULL;
1707: b->roworiented = PETSC_TRUE;
1709: #if defined(PETSC_USE_MAT_SINGLE)
1710: /* stuff for MatSetValues_XXX in single precision */
1711: b->setvalueslen = 0;
1712: b->setvaluescopy = PETSC_NULL;
1713: #endif
1715: /* stuff used in block assembly */
1716: b->barray = 0;
1718: /* stuff used for matrix vector multiply */
1719: b->lvec = 0;
1720: b->Mvctx = 0;
1721: b->slvec0 = 0;
1722: b->slvec0b = 0;
1723: b->slvec1 = 0;
1724: b->slvec1a = 0;
1725: b->slvec1b = 0;
1726: b->sMvctx = 0;
1728: /* stuff for MatGetRow() */
1729: b->rowindices = 0;
1730: b->rowvalues = 0;
1731: b->getrowactive = PETSC_FALSE;
1733: /* hash table stuff */
1734: b->ht = 0;
1735: b->hd = 0;
1736: b->ht_size = 0;
1737: b->ht_flag = PETSC_FALSE;
1738: b->ht_fact = 0;
1739: b->ht_total_ct = 0;
1740: b->ht_insert_ct = 0;
1742: PetscOptionsHasName(B->prefix,"-mat_use_hash_table",&flg);
1743: if (flg) {
1744: PetscReal fact = 1.39;
1745: MatSetOption(B,MAT_USE_HASH_TABLE);
1746: PetscOptionsGetReal(B->prefix,"-mat_use_hash_table",&fact,PETSC_NULL);
1747: if (fact <= 1.0) fact = 1.39;
1748: MatMPIBAIJSetHashTableFactor(B,fact);
1749: PetscLogInfo(0,"MatCreateMPISBAIJ:Hash table Factor used %5.2f\n",fact);
1750: }
1751: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C",
1752: "MatStoreValues_MPISBAIJ",
1753: MatStoreValues_MPISBAIJ);
1754: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C",
1755: "MatRetrieveValues_MPISBAIJ",
1756: MatRetrieveValues_MPISBAIJ);
1757: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetDiagonalBlock_C",
1758: "MatGetDiagonalBlock_MPISBAIJ",
1759: MatGetDiagonalBlock_MPISBAIJ);
1760: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMPISBAIJSetPreallocation_C",
1761: "MatMPISBAIJSetPreallocation_MPISBAIJ",
1762: MatMPISBAIJSetPreallocation_MPISBAIJ);
1763: return(0);
1764: }
1765: EXTERN_C_END
1767: /*MC
1768: MATSBAIJ - MATSBAIJ = "sbaij" - A matrix type to be used for symmetric block sparse matrices.
1770: This matrix type is identical to MATSEQSBAIJ when constructed with a single process communicator,
1771: and MATMPISBAIJ otherwise.
1773: Options Database Keys:
1774: . -mat_type sbaij - sets the matrix type to "sbaij" during a call to MatSetFromOptions()
1776: Level: beginner
1778: .seealso: MatCreateMPISBAIJ,MATSEQSBAIJ,MATMPISBAIJ
1779: M*/
1781: EXTERN_C_BEGIN
1784: int MatCreate_SBAIJ(Mat A) {
1785: int ierr,size;
1788: PetscObjectChangeTypeName((PetscObject)A,MATSBAIJ);
1789: MPI_Comm_size(A->comm,&size);
1790: if (size == 1) {
1791: MatSetType(A,MATSEQSBAIJ);
1792: } else {
1793: MatSetType(A,MATMPISBAIJ);
1794: }
1795: return(0);
1796: }
1797: EXTERN_C_END
1801: /*@C
1802: MatMPISBAIJSetPreallocation - For good matrix assembly performance
1803: the user should preallocate the matrix storage by setting the parameters
1804: d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately,
1805: performance can be increased by more than a factor of 50.
1807: Collective on Mat
1809: Input Parameters:
1810: + A - the matrix
1811: . bs - size of blockk
1812: . d_nz - number of block nonzeros per block row in diagonal portion of local
1813: submatrix (same for all local rows)
1814: . d_nnz - array containing the number of block nonzeros in the various block rows
1815: in the upper triangular and diagonal part of the in diagonal portion of the local
1816: (possibly different for each block row) or PETSC_NULL. You must leave room
1817: for the diagonal entry even if it is zero.
1818: . o_nz - number of block nonzeros per block row in the off-diagonal portion of local
1819: submatrix (same for all local rows).
1820: - o_nnz - array containing the number of nonzeros in the various block rows of the
1821: off-diagonal portion of the local submatrix (possibly different for
1822: each block row) or PETSC_NULL.
1825: Options Database Keys:
1826: . -mat_no_unroll - uses code that does not unroll the loops in the
1827: block calculations (much slower)
1828: . -mat_block_size - size of the blocks to use
1830: Notes:
1832: If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one processor
1833: than it must be used on all processors that share the object for that argument.
1835: Storage Information:
1836: For a square global matrix we define each processor's diagonal portion
1837: to be its local rows and the corresponding columns (a square submatrix);
1838: each processor's off-diagonal portion encompasses the remainder of the
1839: local matrix (a rectangular submatrix).
1841: The user can specify preallocated storage for the diagonal part of
1842: the local submatrix with either d_nz or d_nnz (not both). Set
1843: d_nz=PETSC_DEFAULT and d_nnz=PETSC_NULL for PETSc to control dynamic
1844: memory allocation. Likewise, specify preallocated storage for the
1845: off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
1847: Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
1848: the figure below we depict these three local rows and all columns (0-11).
1850: .vb
1851: 0 1 2 3 4 5 6 7 8 9 10 11
1852: -------------------
1853: row 3 | o o o d d d o o o o o o
1854: row 4 | o o o d d d o o o o o o
1855: row 5 | o o o d d d o o o o o o
1856: -------------------
1857: .ve
1858:
1859: Thus, any entries in the d locations are stored in the d (diagonal)
1860: submatrix, and any entries in the o locations are stored in the
1861: o (off-diagonal) submatrix. Note that the d matrix is stored in
1862: MatSeqSBAIJ format and the o submatrix in MATSEQBAIJ format.
1864: Now d_nz should indicate the number of block nonzeros per row in the upper triangular
1865: plus the diagonal part of the d matrix,
1866: and o_nz should indicate the number of block nonzeros per row in the o matrix.
1867: In general, for PDE problems in which most nonzeros are near the diagonal,
1868: one expects d_nz >> o_nz. For large problems you MUST preallocate memory
1869: or you will get TERRIBLE performance; see the users' manual chapter on
1870: matrices.
1872: Level: intermediate
1874: .keywords: matrix, block, aij, compressed row, sparse, parallel
1876: .seealso: MatCreate(), MatCreateSeqSBAIJ(), MatSetValues(), MatCreateMPIBAIJ()
1877: @*/
1878: int MatMPISBAIJSetPreallocation(Mat B,int bs,int d_nz,const int d_nnz[],int o_nz,const int o_nnz[])
1879: {
1880: int ierr,(*f)(Mat,int,int,const int[],int,const int[]);
1883: PetscObjectQueryFunction((PetscObject)B,"MatMPISBAIJSetPreallocation_C",(void (**)(void))&f);
1884: if (f) {
1885: (*f)(B,bs,d_nz,d_nnz,o_nz,o_nnz);
1886: }
1887: return(0);
1888: }
1892: /*@C
1893: MatCreateMPISBAIJ - Creates a sparse parallel matrix in symmetric block AIJ format
1894: (block compressed row). For good matrix assembly performance
1895: the user should preallocate the matrix storage by setting the parameters
1896: d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately,
1897: performance can be increased by more than a factor of 50.
1899: Collective on MPI_Comm
1901: Input Parameters:
1902: + comm - MPI communicator
1903: . bs - size of blockk
1904: . m - number of local rows (or PETSC_DECIDE to have calculated if M is given)
1905: This value should be the same as the local size used in creating the
1906: y vector for the matrix-vector product y = Ax.
1907: . n - number of local columns (or PETSC_DECIDE to have calculated if N is given)
1908: This value should be the same as the local size used in creating the
1909: x vector for the matrix-vector product y = Ax.
1910: . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given)
1911: . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given)
1912: . d_nz - number of block nonzeros per block row in diagonal portion of local
1913: submatrix (same for all local rows)
1914: . d_nnz - array containing the number of block nonzeros in the various block rows
1915: in the upper triangular portion of the in diagonal portion of the local
1916: (possibly different for each block block row) or PETSC_NULL.
1917: You must leave room for the diagonal entry even if it is zero.
1918: . o_nz - number of block nonzeros per block row in the off-diagonal portion of local
1919: submatrix (same for all local rows).
1920: - o_nnz - array containing the number of nonzeros in the various block rows of the
1921: off-diagonal portion of the local submatrix (possibly different for
1922: each block row) or PETSC_NULL.
1924: Output Parameter:
1925: . A - the matrix
1927: Options Database Keys:
1928: . -mat_no_unroll - uses code that does not unroll the loops in the
1929: block calculations (much slower)
1930: . -mat_block_size - size of the blocks to use
1931: . -mat_mpi - use the parallel matrix data structures even on one processor
1932: (defaults to using SeqBAIJ format on one processor)
1934: Notes:
1935: The user MUST specify either the local or global matrix dimensions
1936: (possibly both).
1938: If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one processor
1939: than it must be used on all processors that share the object for that argument.
1941: Storage Information:
1942: For a square global matrix we define each processor's diagonal portion
1943: to be its local rows and the corresponding columns (a square submatrix);
1944: each processor's off-diagonal portion encompasses the remainder of the
1945: local matrix (a rectangular submatrix).
1947: The user can specify preallocated storage for the diagonal part of
1948: the local submatrix with either d_nz or d_nnz (not both). Set
1949: d_nz=PETSC_DEFAULT and d_nnz=PETSC_NULL for PETSc to control dynamic
1950: memory allocation. Likewise, specify preallocated storage for the
1951: off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
1953: Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
1954: the figure below we depict these three local rows and all columns (0-11).
1956: .vb
1957: 0 1 2 3 4 5 6 7 8 9 10 11
1958: -------------------
1959: row 3 | o o o d d d o o o o o o
1960: row 4 | o o o d d d o o o o o o
1961: row 5 | o o o d d d o o o o o o
1962: -------------------
1963: .ve
1964:
1965: Thus, any entries in the d locations are stored in the d (diagonal)
1966: submatrix, and any entries in the o locations are stored in the
1967: o (off-diagonal) submatrix. Note that the d matrix is stored in
1968: MatSeqSBAIJ format and the o submatrix in MATSEQBAIJ format.
1970: Now d_nz should indicate the number of block nonzeros per row in the upper triangular
1971: plus the diagonal part of the d matrix,
1972: and o_nz should indicate the number of block nonzeros per row in the o matrix.
1973: In general, for PDE problems in which most nonzeros are near the diagonal,
1974: one expects d_nz >> o_nz. For large problems you MUST preallocate memory
1975: or you will get TERRIBLE performance; see the users' manual chapter on
1976: matrices.
1978: Level: intermediate
1980: .keywords: matrix, block, aij, compressed row, sparse, parallel
1982: .seealso: MatCreate(), MatCreateSeqSBAIJ(), MatSetValues(), MatCreateMPIBAIJ()
1983: @*/
1985: int MatCreateMPISBAIJ(MPI_Comm comm,int bs,int m,int n,int M,int N,int d_nz,const int d_nnz[],int o_nz,const int o_nnz[],Mat *A)
1986: {
1987: int ierr,size;
1990: MatCreate(comm,m,n,M,N,A);
1991: MPI_Comm_size(comm,&size);
1992: if (size > 1) {
1993: MatSetType(*A,MATMPISBAIJ);
1994: MatMPISBAIJSetPreallocation(*A,bs,d_nz,d_nnz,o_nz,o_nnz);
1995: } else {
1996: MatSetType(*A,MATSEQSBAIJ);
1997: MatSeqSBAIJSetPreallocation(*A,bs,d_nz,d_nnz);
1998: }
1999: return(0);
2000: }
2005: static int MatDuplicate_MPISBAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat)
2006: {
2007: Mat mat;
2008: Mat_MPISBAIJ *a,*oldmat = (Mat_MPISBAIJ*)matin->data;
2009: int ierr,len=0;
2012: *newmat = 0;
2013: MatCreate(matin->comm,matin->m,matin->n,matin->M,matin->N,&mat);
2014: MatSetType(mat,MATMPISBAIJ);
2015: mat->preallocated = PETSC_TRUE;
2016: a = (Mat_MPISBAIJ*)mat->data;
2017: a->bs = oldmat->bs;
2018: a->bs2 = oldmat->bs2;
2019: a->mbs = oldmat->mbs;
2020: a->nbs = oldmat->nbs;
2021: a->Mbs = oldmat->Mbs;
2022: a->Nbs = oldmat->Nbs;
2023:
2024: a->rstart = oldmat->rstart;
2025: a->rend = oldmat->rend;
2026: a->cstart = oldmat->cstart;
2027: a->cend = oldmat->cend;
2028: a->size = oldmat->size;
2029: a->rank = oldmat->rank;
2030: a->donotstash = oldmat->donotstash;
2031: a->roworiented = oldmat->roworiented;
2032: a->rowindices = 0;
2033: a->rowvalues = 0;
2034: a->getrowactive = PETSC_FALSE;
2035: a->barray = 0;
2036: a->rstart_bs = oldmat->rstart_bs;
2037: a->rend_bs = oldmat->rend_bs;
2038: a->cstart_bs = oldmat->cstart_bs;
2039: a->cend_bs = oldmat->cend_bs;
2041: /* hash table stuff */
2042: a->ht = 0;
2043: a->hd = 0;
2044: a->ht_size = 0;
2045: a->ht_flag = oldmat->ht_flag;
2046: a->ht_fact = oldmat->ht_fact;
2047: a->ht_total_ct = 0;
2048: a->ht_insert_ct = 0;
2050: PetscMalloc(3*(a->size+2)*sizeof(int),&a->rowners);
2051: PetscLogObjectMemory(mat,3*(a->size+2)*sizeof(int)+sizeof(struct _p_Mat)+sizeof(Mat_MPISBAIJ));
2052: a->cowners = a->rowners + a->size + 2;
2053: a->rowners_bs = a->cowners + a->size + 2;
2054: PetscMemcpy(a->rowners,oldmat->rowners,3*(a->size+2)*sizeof(int));
2055: MatStashCreate_Private(matin->comm,1,&mat->stash);
2056: MatStashCreate_Private(matin->comm,oldmat->bs,&mat->bstash);
2057: if (oldmat->colmap) {
2058: #if defined (PETSC_USE_CTABLE)
2059: PetscTableCreateCopy(oldmat->colmap,&a->colmap);
2060: #else
2061: PetscMalloc((a->Nbs)*sizeof(int),&a->colmap);
2062: PetscLogObjectMemory(mat,(a->Nbs)*sizeof(int));
2063: PetscMemcpy(a->colmap,oldmat->colmap,(a->Nbs)*sizeof(int));
2064: #endif
2065: } else a->colmap = 0;
2066: if (oldmat->garray && (len = ((Mat_SeqBAIJ*)(oldmat->B->data))->nbs)) {
2067: PetscMalloc(len*sizeof(int),&a->garray);
2068: PetscLogObjectMemory(mat,len*sizeof(int));
2069: PetscMemcpy(a->garray,oldmat->garray,len*sizeof(int));
2070: } else a->garray = 0;
2071:
2072: VecDuplicate(oldmat->lvec,&a->lvec);
2073: PetscLogObjectParent(mat,a->lvec);
2074: VecScatterCopy(oldmat->Mvctx,&a->Mvctx);
2076: PetscLogObjectParent(mat,a->Mvctx);
2077: MatDuplicate(oldmat->A,cpvalues,&a->A);
2078: PetscLogObjectParent(mat,a->A);
2079: MatDuplicate(oldmat->B,cpvalues,&a->B);
2080: PetscLogObjectParent(mat,a->B);
2081: PetscFListDuplicate(mat->qlist,&matin->qlist);
2082: *newmat = mat;
2083: return(0);
2084: }
2086: #include petscsys.h
2090: int MatLoad_MPISBAIJ(PetscViewer viewer,MatType type,Mat *newmat)
2091: {
2092: Mat A;
2093: int i,nz,ierr,j,rstart,rend,fd;
2094: PetscScalar *vals,*buf;
2095: MPI_Comm comm = ((PetscObject)viewer)->comm;
2096: MPI_Status status;
2097: int header[4],rank,size,*rowlengths = 0,M,N,m,*rowners,*browners,maxnz,*cols;
2098: int *locrowlens,*sndcounts = 0,*procsnz = 0,jj,*mycols,*ibuf;
2099: int tag = ((PetscObject)viewer)->tag,bs=1,Mbs,mbs,extra_rows;
2100: int *dlens,*odlens,*mask,*masked1,*masked2,rowcount,odcount;
2101: int dcount,kmax,k,nzcount,tmp;
2102:
2104: PetscOptionsGetInt(PETSC_NULL,"-matload_block_size",&bs,PETSC_NULL);
2106: MPI_Comm_size(comm,&size);
2107: MPI_Comm_rank(comm,&rank);
2108: if (!rank) {
2109: PetscViewerBinaryGetDescriptor(viewer,&fd);
2110: PetscBinaryRead(fd,(char *)header,4,PETSC_INT);
2111: if (header[0] != MAT_FILE_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not matrix object");
2112: if (header[3] < 0) {
2113: SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format, cannot load as MPISBAIJ");
2114: }
2115: }
2117: MPI_Bcast(header+1,3,MPI_INT,0,comm);
2118: M = header[1]; N = header[2];
2120: if (M != N) SETERRQ(PETSC_ERR_SUP,"Can only do square matrices");
2122: /*
2123: This code adds extra rows to make sure the number of rows is
2124: divisible by the blocksize
2125: */
2126: Mbs = M/bs;
2127: extra_rows = bs - M + bs*(Mbs);
2128: if (extra_rows == bs) extra_rows = 0;
2129: else Mbs++;
2130: if (extra_rows &&!rank) {
2131: PetscLogInfo(0,"MatLoad_MPISBAIJ:Padding loaded matrix to match blocksize\n");
2132: }
2134: /* determine ownership of all rows */
2135: mbs = Mbs/size + ((Mbs % size) > rank);
2136: m = mbs*bs;
2137: PetscMalloc(2*(size+2)*sizeof(int),&rowners);
2138: browners = rowners + size + 1;
2139: MPI_Allgather(&mbs,1,MPI_INT,rowners+1,1,MPI_INT,comm);
2140: rowners[0] = 0;
2141: for (i=2; i<=size; i++) rowners[i] += rowners[i-1];
2142: for (i=0; i<=size; i++) browners[i] = rowners[i]*bs;
2143: rstart = rowners[rank];
2144: rend = rowners[rank+1];
2145:
2146: /* distribute row lengths to all processors */
2147: PetscMalloc((rend-rstart)*bs*sizeof(int),&locrowlens);
2148: if (!rank) {
2149: PetscMalloc((M+extra_rows)*sizeof(int),&rowlengths);
2150: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
2151: for (i=0; i<extra_rows; i++) rowlengths[M+i] = 1;
2152: PetscMalloc(size*sizeof(int),&sndcounts);
2153: for (i=0; i<size; i++) sndcounts[i] = browners[i+1] - browners[i];
2154: MPI_Scatterv(rowlengths,sndcounts,browners,MPI_INT,locrowlens,(rend-rstart)*bs,MPI_INT,0,comm);
2155: PetscFree(sndcounts);
2156: } else {
2157: MPI_Scatterv(0,0,0,MPI_INT,locrowlens,(rend-rstart)*bs,MPI_INT,0,comm);
2158: }
2159:
2160: if (!rank) { /* procs[0] */
2161: /* calculate the number of nonzeros on each processor */
2162: PetscMalloc(size*sizeof(int),&procsnz);
2163: PetscMemzero(procsnz,size*sizeof(int));
2164: for (i=0; i<size; i++) {
2165: for (j=rowners[i]*bs; j< rowners[i+1]*bs; j++) {
2166: procsnz[i] += rowlengths[j];
2167: }
2168: }
2169: PetscFree(rowlengths);
2170:
2171: /* determine max buffer needed and allocate it */
2172: maxnz = 0;
2173: for (i=0; i<size; i++) {
2174: maxnz = PetscMax(maxnz,procsnz[i]);
2175: }
2176: PetscMalloc(maxnz*sizeof(int),&cols);
2178: /* read in my part of the matrix column indices */
2179: nz = procsnz[0];
2180: PetscMalloc(nz*sizeof(int),&ibuf);
2181: mycols = ibuf;
2182: if (size == 1) nz -= extra_rows;
2183: PetscBinaryRead(fd,mycols,nz,PETSC_INT);
2184: if (size == 1) for (i=0; i< extra_rows; i++) { mycols[nz+i] = M+i; }
2186: /* read in every ones (except the last) and ship off */
2187: for (i=1; i<size-1; i++) {
2188: nz = procsnz[i];
2189: PetscBinaryRead(fd,cols,nz,PETSC_INT);
2190: MPI_Send(cols,nz,MPI_INT,i,tag,comm);
2191: }
2192: /* read in the stuff for the last proc */
2193: if (size != 1) {
2194: nz = procsnz[size-1] - extra_rows; /* the extra rows are not on the disk */
2195: PetscBinaryRead(fd,cols,nz,PETSC_INT);
2196: for (i=0; i<extra_rows; i++) cols[nz+i] = M+i;
2197: MPI_Send(cols,nz+extra_rows,MPI_INT,size-1,tag,comm);
2198: }
2199: PetscFree(cols);
2200: } else { /* procs[i], i>0 */
2201: /* determine buffer space needed for message */
2202: nz = 0;
2203: for (i=0; i<m; i++) {
2204: nz += locrowlens[i];
2205: }
2206: PetscMalloc(nz*sizeof(int),&ibuf);
2207: mycols = ibuf;
2208: /* receive message of column indices*/
2209: MPI_Recv(mycols,nz,MPI_INT,0,tag,comm,&status);
2210: MPI_Get_count(&status,MPI_INT,&maxnz);
2211: if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
2212: }
2214: /* loop over local rows, determining number of off diagonal entries */
2215: PetscMalloc(2*(rend-rstart+1)*sizeof(int),&dlens);
2216: odlens = dlens + (rend-rstart);
2217: PetscMalloc(3*Mbs*sizeof(int),&mask);
2218: PetscMemzero(mask,3*Mbs*sizeof(int));
2219: masked1 = mask + Mbs;
2220: masked2 = masked1 + Mbs;
2221: rowcount = 0; nzcount = 0;
2222: for (i=0; i<mbs; i++) {
2223: dcount = 0;
2224: odcount = 0;
2225: for (j=0; j<bs; j++) {
2226: kmax = locrowlens[rowcount];
2227: for (k=0; k<kmax; k++) {
2228: tmp = mycols[nzcount++]/bs; /* block col. index */
2229: if (!mask[tmp]) {
2230: mask[tmp] = 1;
2231: if (tmp < rstart || tmp >= rend) masked2[odcount++] = tmp; /* entry in off-diag portion */
2232: else masked1[dcount++] = tmp; /* entry in diag portion */
2233: }
2234: }
2235: rowcount++;
2236: }
2237:
2238: dlens[i] = dcount; /* d_nzz[i] */
2239: odlens[i] = odcount; /* o_nzz[i] */
2241: /* zero out the mask elements we set */
2242: for (j=0; j<dcount; j++) mask[masked1[j]] = 0;
2243: for (j=0; j<odcount; j++) mask[masked2[j]] = 0;
2244: }
2245:
2246: /* create our matrix */
2247: MatCreate(comm,m,m,PETSC_DETERMINE,PETSC_DETERMINE,&A);
2248: MatSetType(A,type);
2249: MatMPISBAIJSetPreallocation(A,bs,0,dlens,0,odlens);
2250: MatSetOption(A,MAT_COLUMNS_SORTED);
2251:
2252: if (!rank) {
2253: PetscMalloc(maxnz*sizeof(PetscScalar),&buf);
2254: /* read in my part of the matrix numerical values */
2255: nz = procsnz[0];
2256: vals = buf;
2257: mycols = ibuf;
2258: if (size == 1) nz -= extra_rows;
2259: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
2260: if (size == 1) for (i=0; i< extra_rows; i++) { vals[nz+i] = 1.0; }
2262: /* insert into matrix */
2263: jj = rstart*bs;
2264: for (i=0; i<m; i++) {
2265: MatSetValues(A,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);
2266: mycols += locrowlens[i];
2267: vals += locrowlens[i];
2268: jj++;
2269: }
2271: /* read in other processors (except the last one) and ship out */
2272: for (i=1; i<size-1; i++) {
2273: nz = procsnz[i];
2274: vals = buf;
2275: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
2276: MPI_Send(vals,nz,MPIU_SCALAR,i,A->tag,comm);
2277: }
2278: /* the last proc */
2279: if (size != 1){
2280: nz = procsnz[i] - extra_rows;
2281: vals = buf;
2282: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
2283: for (i=0; i<extra_rows; i++) vals[nz+i] = 1.0;
2284: MPI_Send(vals,nz+extra_rows,MPIU_SCALAR,size-1,A->tag,comm);
2285: }
2286: PetscFree(procsnz);
2288: } else {
2289: /* receive numeric values */
2290: PetscMalloc(nz*sizeof(PetscScalar),&buf);
2292: /* receive message of values*/
2293: vals = buf;
2294: mycols = ibuf;
2295: MPI_Recv(vals,nz,MPIU_SCALAR,0,A->tag,comm,&status);
2296: MPI_Get_count(&status,MPIU_SCALAR,&maxnz);
2297: if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
2299: /* insert into matrix */
2300: jj = rstart*bs;
2301: for (i=0; i<m; i++) {
2302: MatSetValues_MPISBAIJ(A,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);
2303: mycols += locrowlens[i];
2304: vals += locrowlens[i];
2305: jj++;
2306: }
2307: }
2309: PetscFree(locrowlens);
2310: PetscFree(buf);
2311: PetscFree(ibuf);
2312: PetscFree(rowners);
2313: PetscFree(dlens);
2314: PetscFree(mask);
2315: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
2316: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
2317: *newmat = A;
2318: return(0);
2319: }
2323: /*@
2324: MatMPISBAIJSetHashTableFactor - Sets the factor required to compute the size of the HashTable.
2326: Input Parameters:
2327: . mat - the matrix
2328: . fact - factor
2330: Collective on Mat
2332: Level: advanced
2334: Notes:
2335: This can also be set by the command line option: -mat_use_hash_table fact
2337: .keywords: matrix, hashtable, factor, HT
2339: .seealso: MatSetOption()
2340: @*/
2341: int MatMPISBAIJSetHashTableFactor(Mat mat,PetscReal fact)
2342: {
2344: SETERRQ(1,"Function not yet written for SBAIJ format");
2345: /* return(0); */
2346: }
2350: int MatGetRowMax_MPISBAIJ(Mat A,Vec v)
2351: {
2352: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
2353: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)(a->B)->data;
2354: PetscReal atmp;
2355: PetscReal *work,*svalues,*rvalues;
2356: int ierr,i,bs,mbs,*bi,*bj,brow,j,ncols,krow,kcol,col,row,Mbs,bcol;
2357: int rank,size,*rowners_bs,dest,count,source;
2358: PetscScalar *va;
2359: MatScalar *ba;
2360: MPI_Status stat;
2363: MatGetRowMax(a->A,v);
2364: VecGetArray(v,&va);
2366: MPI_Comm_size(A->comm,&size);
2367: MPI_Comm_rank(A->comm,&rank);
2369: bs = a->bs;
2370: mbs = a->mbs;
2371: Mbs = a->Mbs;
2372: ba = b->a;
2373: bi = b->i;
2374: bj = b->j;
2375: /*
2376: PetscSynchronizedPrintf(A->comm,"[%d] M: %d, bs: %d, mbs: %d \n",rank,bs*Mbs,bs,mbs);
2377: PetscSynchronizedFlush(A->comm);
2378: */
2380: /* find ownerships */
2381: rowners_bs = a->rowners_bs;
2382: /*
2383: if (!rank){
2384: for (i=0; i<size+1; i++) PetscPrintf(PETSC_COMM_SELF," rowners_bs[%d]: %d\n",i,rowners_bs[i]);
2385: }
2386: */
2388: /* each proc creates an array to be distributed */
2389: PetscMalloc(bs*Mbs*sizeof(PetscReal),&work);
2390: PetscMemzero(work,bs*Mbs*sizeof(PetscReal));
2392: /* row_max for B */
2393: if (rank != size-1){
2394: for (i=0; i<mbs; i++) {
2395: ncols = bi[1] - bi[0]; bi++;
2396: brow = bs*i;
2397: for (j=0; j<ncols; j++){
2398: bcol = bs*(*bj);
2399: for (kcol=0; kcol<bs; kcol++){
2400: col = bcol + kcol; /* local col index */
2401: col += rowners_bs[rank+1]; /* global col index */
2402: /* PetscPrintf(PETSC_COMM_SELF,"[%d], col: %d\n",rank,col); */
2403: for (krow=0; krow<bs; krow++){
2404: atmp = PetscAbsScalar(*ba); ba++;
2405: row = brow + krow; /* local row index */
2406: /* printf("val[%d,%d]: %g\n",row,col,atmp); */
2407: if (PetscRealPart(va[row]) < atmp) va[row] = atmp;
2408: if (work[col] < atmp) work[col] = atmp;
2409: }
2410: }
2411: bj++;
2412: }
2413: }
2414: /*
2415: PetscPrintf(PETSC_COMM_SELF,"[%d], work: ",rank);
2416: for (i=0; i<bs*Mbs; i++) PetscPrintf(PETSC_COMM_SELF,"%g ",work[i]);
2417: PetscPrintf(PETSC_COMM_SELF,"[%d]: \n");
2418: */
2420: /* send values to its owners */
2421: for (dest=rank+1; dest<size; dest++){
2422: svalues = work + rowners_bs[dest];
2423: count = rowners_bs[dest+1]-rowners_bs[dest];
2424: MPI_Send(svalues,count,MPIU_REAL,dest,rank,A->comm);
2425: /*
2426: PetscSynchronizedPrintf(A->comm,"[%d] sends %d values to [%d]: %g, %g, %g, %g\n",rank,count,dest,svalues[0],svalues[1],svalues[2],svalues[3]);
2427: PetscSynchronizedFlush(A->comm);
2428: */
2429: }
2430: }
2431:
2432: /* receive values */
2433: if (rank){
2434: rvalues = work;
2435: count = rowners_bs[rank+1]-rowners_bs[rank];
2436: for (source=0; source<rank; source++){
2437: MPI_Recv(rvalues,count,MPIU_REAL,MPI_ANY_SOURCE,MPI_ANY_TAG,A->comm,&stat);
2438: /* process values */
2439: for (i=0; i<count; i++){
2440: if (PetscRealPart(va[i]) < rvalues[i]) va[i] = rvalues[i];
2441: }
2442: /*
2443: PetscSynchronizedPrintf(A->comm,"[%d] received %d values from [%d]: %g, %g, %g, %g \n",rank,count,stat.MPI_SOURCE,rvalues[0],rvalues[1],rvalues[2],rvalues[3]);
2444: PetscSynchronizedFlush(A->comm);
2445: */
2446: }
2447: }
2449: VecRestoreArray(v,&va);
2450: PetscFree(work);
2451: return(0);
2452: }
2456: int MatRelax_MPISBAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,int its,int lits,Vec xx)
2457: {
2458: Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)matin->data;
2459: int ierr,mbs=mat->mbs,bs=mat->bs;
2460: PetscScalar mone=-1.0,*x,*b,*ptr,zero=0.0;
2461: Vec bb1;
2462:
2464: if (its <= 0 || lits <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %d and local its %d both positive",its,lits);
2465: if (bs > 1)
2466: SETERRQ(PETSC_ERR_SUP,"SSOR for block size > 1 is not yet implemented");
2468: if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP){
2469: if ( flag & SOR_ZERO_INITIAL_GUESS ) {
2470: (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,lits,xx);
2471: its--;
2472: }
2474: VecDuplicate(bb,&bb1);
2475: while (its--){
2476:
2477: /* lower triangular part: slvec0b = - B^T*xx */
2478: (*mat->B->ops->multtranspose)(mat->B,xx,mat->slvec0b);
2479:
2480: /* copy xx into slvec0a */
2481: VecGetArray(mat->slvec0,&ptr);
2482: VecGetArray(xx,&x);
2483: PetscMemcpy(ptr,x,bs*mbs*sizeof(MatScalar));
2484: VecRestoreArray(mat->slvec0,&ptr);
2486: VecScale(&mone,mat->slvec0);
2488: /* copy bb into slvec1a */
2489: VecGetArray(mat->slvec1,&ptr);
2490: VecGetArray(bb,&b);
2491: PetscMemcpy(ptr,b,bs*mbs*sizeof(MatScalar));
2492: VecRestoreArray(mat->slvec1,&ptr);
2494: /* set slvec1b = 0 */
2495: VecSet(&zero,mat->slvec1b);
2497: VecScatterBegin(mat->slvec0,mat->slvec1,ADD_VALUES,SCATTER_FORWARD,mat->sMvctx);
2498: VecRestoreArray(xx,&x);
2499: VecRestoreArray(bb,&b);
2500: VecScatterEnd(mat->slvec0,mat->slvec1,ADD_VALUES,SCATTER_FORWARD,mat->sMvctx);
2502: /* upper triangular part: bb1 = bb1 - B*x */
2503: (*mat->B->ops->multadd)(mat->B,mat->slvec1b,mat->slvec1a,bb1);
2504:
2505: /* local diagonal sweep */
2506: (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,lits,xx);
2507: }
2508: VecDestroy(bb1);
2509: } else {
2510: SETERRQ(PETSC_ERR_SUP,"MatSORType is not supported for SBAIJ matrix format");
2511: }
2512: return(0);
2513: }
2517: int MatRelax_MPISBAIJ_2comm(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,int its,int lits,Vec xx)
2518: {
2519: Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)matin->data;
2520: int ierr;
2521: PetscScalar mone=-1.0;
2522: Vec lvec1,bb1;
2523:
2525: if (its <= 0 || lits <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %d and local its %d both positive",its,lits);
2526: if (mat->bs > 1)
2527: SETERRQ(PETSC_ERR_SUP,"SSOR for block size > 1 is not yet implemented");
2529: if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP){
2530: if ( flag & SOR_ZERO_INITIAL_GUESS ) {
2531: (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,lits,xx);
2532: its--;
2533: }
2535: VecDuplicate(mat->lvec,&lvec1);
2536: VecDuplicate(bb,&bb1);
2537: while (its--){
2538: VecScatterBegin(xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD,mat->Mvctx);
2539:
2540: /* lower diagonal part: bb1 = bb - B^T*xx */
2541: (*mat->B->ops->multtranspose)(mat->B,xx,lvec1);
2542: VecScale(&mone,lvec1);
2544: VecScatterEnd(xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD,mat->Mvctx);
2545: VecCopy(bb,bb1);
2546: VecScatterBegin(lvec1,bb1,ADD_VALUES,SCATTER_REVERSE,mat->Mvctx);
2548: /* upper diagonal part: bb1 = bb1 - B*x */
2549: VecScale(&mone,mat->lvec);
2550: (*mat->B->ops->multadd)(mat->B,mat->lvec,bb1,bb1);
2552: VecScatterEnd(lvec1,bb1,ADD_VALUES,SCATTER_REVERSE,mat->Mvctx);
2553:
2554: /* diagonal sweep */
2555: (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,lits,xx);
2556: }
2557: VecDestroy(lvec1);
2558: VecDestroy(bb1);
2559: } else {
2560: SETERRQ(PETSC_ERR_SUP,"MatSORType is not supported for SBAIJ matrix format");
2561: }
2562: return(0);
2563: }