Actual source code: mpisbaij.c
1: #define PETSCMAT_DLL
3: #include src/mat/impls/baij/mpi/mpibaij.h
4: #include mpisbaij.h
5: #include src/mat/impls/sbaij/seq/sbaij.h
7: EXTERN PetscErrorCode MatSetUpMultiply_MPISBAIJ(Mat);
8: EXTERN PetscErrorCode MatSetUpMultiply_MPISBAIJ_2comm(Mat);
9: EXTERN PetscErrorCode DisAssemble_MPISBAIJ(Mat);
10: EXTERN PetscErrorCode MatIncreaseOverlap_MPISBAIJ(Mat,PetscInt,IS[],PetscInt);
11: EXTERN PetscErrorCode MatGetValues_SeqSBAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],PetscScalar []);
12: EXTERN PetscErrorCode MatGetValues_SeqBAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],PetscScalar []);
13: EXTERN PetscErrorCode MatSetValues_SeqSBAIJ(Mat,PetscInt,const PetscInt [],PetscInt,const PetscInt [],const PetscScalar [],InsertMode);
14: EXTERN PetscErrorCode MatSetValuesBlocked_SeqSBAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode);
15: EXTERN PetscErrorCode MatSetValuesBlocked_SeqBAIJ(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const PetscScalar[],InsertMode);
16: EXTERN PetscErrorCode MatGetRow_SeqSBAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
17: EXTERN PetscErrorCode MatRestoreRow_SeqSBAIJ(Mat,PetscInt,PetscInt*,PetscInt**,PetscScalar**);
18: EXTERN PetscErrorCode MatZeroRows_SeqSBAIJ(Mat,IS,PetscScalar*);
19: EXTERN PetscErrorCode MatZeroRows_SeqBAIJ(Mat,IS,PetscScalar *);
20: EXTERN PetscErrorCode MatGetRowMaxAbs_MPISBAIJ(Mat,Vec,PetscInt[]);
21: EXTERN PetscErrorCode MatRelax_MPISBAIJ(Mat,Vec,PetscReal,MatSORType,PetscReal,PetscInt,PetscInt,Vec);
23: /* UGLY, ugly, ugly
24: When MatScalar == PetscScalar the function MatSetValuesBlocked_MPIBAIJ_MatScalar() does
25: not exist. Otherwise ..._MatScalar() takes matrix elements in single precision and
26: inserts them into the single precision data structure. The function MatSetValuesBlocked_MPIBAIJ()
27: converts the entries into single precision and then calls ..._MatScalar() to put them
28: into the single precision data structures.
29: */
30: #if defined(PETSC_USE_MAT_SINGLE)
31: EXTERN PetscErrorCode MatSetValuesBlocked_SeqSBAIJ_MatScalar(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const MatScalar[],InsertMode);
32: EXTERN PetscErrorCode MatSetValues_MPISBAIJ_MatScalar(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const MatScalar[],InsertMode);
33: EXTERN PetscErrorCode MatSetValuesBlocked_MPISBAIJ_MatScalar(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const MatScalar[],InsertMode);
34: EXTERN PetscErrorCode MatSetValues_MPISBAIJ_HT_MatScalar(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const MatScalar[],InsertMode);
35: EXTERN PetscErrorCode MatSetValuesBlocked_MPISBAIJ_HT_MatScalar(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const MatScalar[],InsertMode);
36: #else
37: #define MatSetValuesBlocked_SeqSBAIJ_MatScalar MatSetValuesBlocked_SeqSBAIJ
38: #define MatSetValues_MPISBAIJ_MatScalar MatSetValues_MPISBAIJ
39: #define MatSetValuesBlocked_MPISBAIJ_MatScalar MatSetValuesBlocked_MPISBAIJ
40: #define MatSetValues_MPISBAIJ_HT_MatScalar MatSetValues_MPISBAIJ_HT
41: #define MatSetValuesBlocked_MPISBAIJ_HT_MatScalar MatSetValuesBlocked_MPISBAIJ_HT
42: #endif
47: PetscErrorCode MatStoreValues_MPISBAIJ(Mat mat)
48: {
49: Mat_MPISBAIJ *aij = (Mat_MPISBAIJ *)mat->data;
53: MatStoreValues(aij->A);
54: MatStoreValues(aij->B);
55: return(0);
56: }
62: PetscErrorCode MatRetrieveValues_MPISBAIJ(Mat mat)
63: {
64: Mat_MPISBAIJ *aij = (Mat_MPISBAIJ *)mat->data;
68: MatRetrieveValues(aij->A);
69: MatRetrieveValues(aij->B);
70: return(0);
71: }
75: #define CHUNKSIZE 10
77: #define MatSetValues_SeqSBAIJ_A_Private(row,col,value,addv) \
78: { \
79: \
80: brow = row/bs; \
81: rp = aj + ai[brow]; ap = aa + bs2*ai[brow]; \
82: rmax = aimax[brow]; nrow = ailen[brow]; \
83: bcol = col/bs; \
84: ridx = row % bs; cidx = col % bs; \
85: low = 0; high = nrow; \
86: while (high-low > 3) { \
87: t = (low+high)/2; \
88: if (rp[t] > bcol) high = t; \
89: else low = t; \
90: } \
91: for (_i=low; _i<high; _i++) { \
92: if (rp[_i] > bcol) break; \
93: if (rp[_i] == bcol) { \
94: bap = ap + bs2*_i + bs*cidx + ridx; \
95: if (addv == ADD_VALUES) *bap += value; \
96: else *bap = value; \
97: goto a_noinsert; \
98: } \
99: } \
100: if (a->nonew == 1) goto a_noinsert; \
101: if (a->nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", row, col); \
102: MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,brow,bcol,rmax,aa,ai,aj,rp,ap,aimax,a->nonew,MatScalar); \
103: N = nrow++ - 1; \
104: /* shift up all the later entries in this row */ \
105: for (ii=N; ii>=_i; ii--) { \
106: rp[ii+1] = rp[ii]; \
107: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar)); \
108: } \
109: if (N>=_i) { PetscMemzero(ap+bs2*_i,bs2*sizeof(MatScalar)); } \
110: rp[_i] = bcol; \
111: ap[bs2*_i + bs*cidx + ridx] = value; \
112: a_noinsert:; \
113: ailen[brow] = nrow; \
114: }
115: #ifndef MatSetValues_SeqBAIJ_B_Private
116: #define MatSetValues_SeqSBAIJ_B_Private(row,col,value,addv) \
117: { \
118: brow = row/bs; \
119: rp = bj + bi[brow]; ap = ba + bs2*bi[brow]; \
120: rmax = bimax[brow]; nrow = bilen[brow]; \
121: bcol = col/bs; \
122: ridx = row % bs; cidx = col % bs; \
123: low = 0; high = nrow; \
124: while (high-low > 3) { \
125: t = (low+high)/2; \
126: if (rp[t] > bcol) high = t; \
127: else low = t; \
128: } \
129: for (_i=low; _i<high; _i++) { \
130: if (rp[_i] > bcol) break; \
131: if (rp[_i] == bcol) { \
132: bap = ap + bs2*_i + bs*cidx + ridx; \
133: if (addv == ADD_VALUES) *bap += value; \
134: else *bap = value; \
135: goto b_noinsert; \
136: } \
137: } \
138: if (b->nonew == 1) goto b_noinsert; \
139: if (b->nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", row, col); \
140: MatSeqXAIJReallocateAIJ(B,b->mbs,bs2,nrow,brow,bcol,rmax,ba,bi,bj,rp,ap,bimax,b->nonew,MatScalar); \
141: N = nrow++ - 1; \
142: /* shift up all the later entries in this row */ \
143: for (ii=N; ii>=_i; ii--) { \
144: rp[ii+1] = rp[ii]; \
145: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar)); \
146: } \
147: if (N>=_i) { PetscMemzero(ap+bs2*_i,bs2*sizeof(MatScalar));} \
148: rp[_i] = bcol; \
149: ap[bs2*_i + bs*cidx + ridx] = value; \
150: b_noinsert:; \
151: bilen[brow] = nrow; \
152: }
153: #endif
155: #if defined(PETSC_USE_MAT_SINGLE)
158: PetscErrorCode MatSetValues_MPISBAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
159: {
160: Mat_MPISBAIJ *b = (Mat_MPISBAIJ*)mat->data;
162: PetscInt i,N = m*n;
163: MatScalar *vsingle;
166: if (N > b->setvalueslen) {
167: PetscFree(b->setvaluescopy);
168: PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);
169: b->setvalueslen = N;
170: }
171: vsingle = b->setvaluescopy;
173: for (i=0; i<N; i++) {
174: vsingle[i] = v[i];
175: }
176: MatSetValues_MPISBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);
177: return(0);
178: }
182: PetscErrorCode MatSetValuesBlocked_MPISBAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
183: {
184: Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)mat->data;
186: PetscInt i,N = m*n*b->bs2;
187: MatScalar *vsingle;
190: if (N > b->setvalueslen) {
191: PetscFree(b->setvaluescopy);
192: PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);
193: b->setvalueslen = N;
194: }
195: vsingle = b->setvaluescopy;
196: for (i=0; i<N; i++) {
197: vsingle[i] = v[i];
198: }
199: MatSetValuesBlocked_MPISBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);
200: return(0);
201: }
202: #endif
204: /* Only add/insert a(i,j) with i<=j (blocks).
205: Any a(i,j) with i>j input by user is ingored.
206: */
209: PetscErrorCode MatSetValues_MPISBAIJ_MatScalar(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const MatScalar v[],InsertMode addv)
210: {
211: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
212: MatScalar value;
213: PetscTruth roworiented = baij->roworiented;
215: PetscInt i,j,row,col;
216: PetscInt rstart_orig=mat->rmap.rstart;
217: PetscInt rend_orig=mat->rmap.rend,cstart_orig=mat->cmap.rstart;
218: PetscInt cend_orig=mat->cmap.rend,bs=mat->rmap.bs;
220: /* Some Variables required in the macro */
221: Mat A = baij->A;
222: Mat_SeqSBAIJ *a = (Mat_SeqSBAIJ*)(A)->data;
223: PetscInt *aimax=a->imax,*ai=a->i,*ailen=a->ilen,*aj=a->j;
224: MatScalar *aa=a->a;
226: Mat B = baij->B;
227: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)(B)->data;
228: PetscInt *bimax=b->imax,*bi=b->i,*bilen=b->ilen,*bj=b->j;
229: MatScalar *ba=b->a;
231: PetscInt *rp,ii,nrow,_i,rmax,N,brow,bcol;
232: PetscInt low,high,t,ridx,cidx,bs2=a->bs2;
233: MatScalar *ap,*bap;
235: /* for stash */
236: PetscInt n_loc, *in_loc = PETSC_NULL;
237: MatScalar *v_loc = PETSC_NULL;
241: if (!baij->donotstash){
242: if (n > baij->n_loc) {
243: PetscFree(baij->in_loc);
244: PetscFree(baij->v_loc);
245: PetscMalloc(n*sizeof(PetscInt),&baij->in_loc);
246: PetscMalloc(n*sizeof(MatScalar),&baij->v_loc);
247: baij->n_loc = n;
248: }
249: in_loc = baij->in_loc;
250: v_loc = baij->v_loc;
251: }
253: for (i=0; i<m; i++) {
254: if (im[i] < 0) continue;
255: #if defined(PETSC_USE_DEBUG)
256: if (im[i] >= mat->rmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap.N-1);
257: #endif
258: if (im[i] >= rstart_orig && im[i] < rend_orig) { /* this processor entry */
259: row = im[i] - rstart_orig; /* local row index */
260: for (j=0; j<n; j++) {
261: if (im[i]/bs > in[j]/bs){
262: if (a->ignore_ltriangular){
263: continue; /* ignore lower triangular blocks */
264: } else {
265: SETERRQ(PETSC_ERR_USER,"Lower triangular value cannot be set for sbaij format. Ignoring these values, run with -mat_ignore_lower_triangular or call MatSetOption(mat,MAT_IGNORE_LOWER_TRIANGULAR,PETSC_TRUE)");
266: }
267: }
268: if (in[j] >= cstart_orig && in[j] < cend_orig){ /* diag entry (A) */
269: col = in[j] - cstart_orig; /* local col index */
270: brow = row/bs; bcol = col/bs;
271: if (brow > bcol) continue; /* ignore lower triangular blocks of A */
272: if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
273: MatSetValues_SeqSBAIJ_A_Private(row,col,value,addv);
274: /* MatSetValues_SeqBAIJ(baij->A,1,&row,1,&col,&value,addv); */
275: } else if (in[j] < 0) continue;
276: #if defined(PETSC_USE_DEBUG)
277: else if (in[j] >= mat->cmap.N) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[j],mat->cmap.N-1);}
278: #endif
279: else { /* off-diag entry (B) */
280: if (mat->was_assembled) {
281: if (!baij->colmap) {
282: CreateColmap_MPIBAIJ_Private(mat);
283: }
284: #if defined (PETSC_USE_CTABLE)
285: PetscTableFind(baij->colmap,in[j]/bs + 1,&col);
286: col = col - 1;
287: #else
288: col = baij->colmap[in[j]/bs] - 1;
289: #endif
290: if (col < 0 && !((Mat_SeqSBAIJ*)(baij->A->data))->nonew) {
291: DisAssemble_MPISBAIJ(mat);
292: col = in[j];
293: /* Reinitialize the variables required by MatSetValues_SeqBAIJ_B_Private() */
294: B = baij->B;
295: b = (Mat_SeqBAIJ*)(B)->data;
296: bimax=b->imax;bi=b->i;bilen=b->ilen;bj=b->j;
297: ba=b->a;
298: } else col += in[j]%bs;
299: } else col = in[j];
300: if (roworiented) value = v[i*n+j]; else value = v[i+j*m];
301: MatSetValues_SeqSBAIJ_B_Private(row,col,value,addv);
302: /* MatSetValues_SeqBAIJ(baij->B,1,&row,1,&col,&value,addv); */
303: }
304: }
305: } else { /* off processor entry */
306: if (!baij->donotstash) {
307: n_loc = 0;
308: for (j=0; j<n; j++){
309: if (im[i]/bs > in[j]/bs) continue; /* ignore lower triangular blocks */
310: in_loc[n_loc] = in[j];
311: if (roworiented) {
312: v_loc[n_loc] = v[i*n+j];
313: } else {
314: v_loc[n_loc] = v[j*m+i];
315: }
316: n_loc++;
317: }
318: MatStashValuesRow_Private(&mat->stash,im[i],n_loc,in_loc,v_loc);
319: }
320: }
321: }
322: return(0);
323: }
327: PetscErrorCode MatSetValuesBlocked_MPISBAIJ_MatScalar(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const MatScalar v[],InsertMode addv)
328: {
329: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
330: const MatScalar *value;
331: MatScalar *barray=baij->barray;
332: PetscTruth roworiented = baij->roworiented;
333: PetscErrorCode ierr;
334: PetscInt i,j,ii,jj,row,col,rstart=baij->rstartbs;
335: PetscInt rend=baij->rendbs,cstart=baij->rstartbs,stepval;
336: PetscInt cend=baij->rendbs,bs=mat->rmap.bs,bs2=baij->bs2;
339: if(!barray) {
340: PetscMalloc(bs2*sizeof(MatScalar),&barray);
341: baij->barray = barray;
342: }
344: if (roworiented) {
345: stepval = (n-1)*bs;
346: } else {
347: stepval = (m-1)*bs;
348: }
349: for (i=0; i<m; i++) {
350: if (im[i] < 0) continue;
351: #if defined(PETSC_USE_DEBUG)
352: if (im[i] >= baij->Mbs) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large, row %D max %D",im[i],baij->Mbs-1);
353: #endif
354: if (im[i] >= rstart && im[i] < rend) {
355: row = im[i] - rstart;
356: for (j=0; j<n; j++) {
357: /* If NumCol = 1 then a copy is not required */
358: if ((roworiented) && (n == 1)) {
359: barray = (MatScalar*) v + i*bs2;
360: } else if((!roworiented) && (m == 1)) {
361: barray = (MatScalar*) v + j*bs2;
362: } else { /* Here a copy is required */
363: if (roworiented) {
364: value = v + i*(stepval+bs)*bs + j*bs;
365: } else {
366: value = v + j*(stepval+bs)*bs + i*bs;
367: }
368: for (ii=0; ii<bs; ii++,value+=stepval) {
369: for (jj=0; jj<bs; jj++) {
370: *barray++ = *value++;
371: }
372: }
373: barray -=bs2;
374: }
375:
376: if (in[j] >= cstart && in[j] < cend){
377: col = in[j] - cstart;
378: MatSetValuesBlocked_SeqSBAIJ(baij->A,1,&row,1,&col,barray,addv);
379: }
380: else if (in[j] < 0) continue;
381: #if defined(PETSC_USE_DEBUG)
382: else if (in[j] >= baij->Nbs) {SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large, col %D max %D",in[j],baij->Nbs-1);}
383: #endif
384: else {
385: if (mat->was_assembled) {
386: if (!baij->colmap) {
387: CreateColmap_MPIBAIJ_Private(mat);
388: }
390: #if defined(PETSC_USE_DEBUG)
391: #if defined (PETSC_USE_CTABLE)
392: { PetscInt data;
393: PetscTableFind(baij->colmap,in[j]+1,&data);
394: if ((data - 1) % bs) SETERRQ(PETSC_ERR_PLIB,"Incorrect colmap");
395: }
396: #else
397: if ((baij->colmap[in[j]] - 1) % bs) SETERRQ(PETSC_ERR_PLIB,"Incorrect colmap");
398: #endif
399: #endif
400: #if defined (PETSC_USE_CTABLE)
401: PetscTableFind(baij->colmap,in[j]+1,&col);
402: col = (col - 1)/bs;
403: #else
404: col = (baij->colmap[in[j]] - 1)/bs;
405: #endif
406: if (col < 0 && !((Mat_SeqBAIJ*)(baij->A->data))->nonew) {
407: DisAssemble_MPISBAIJ(mat);
408: col = in[j];
409: }
410: }
411: else col = in[j];
412: MatSetValuesBlocked_SeqBAIJ(baij->B,1,&row,1,&col,barray,addv);
413: }
414: }
415: } else {
416: if (!baij->donotstash) {
417: if (roworiented) {
418: MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
419: } else {
420: MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
421: }
422: }
423: }
424: }
425: return(0);
426: }
430: PetscErrorCode MatGetValues_MPISBAIJ(Mat mat,PetscInt m,const PetscInt idxm[],PetscInt n,const PetscInt idxn[],PetscScalar v[])
431: {
432: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
434: PetscInt bs=mat->rmap.bs,i,j,bsrstart = mat->rmap.rstart,bsrend = mat->rmap.rend;
435: PetscInt bscstart = mat->cmap.rstart,bscend = mat->cmap.rend,row,col,data;
438: for (i=0; i<m; i++) {
439: if (idxm[i] < 0) continue; /* SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",idxm[i]); */
440: if (idxm[i] >= mat->rmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",idxm[i],mat->rmap.N-1);
441: if (idxm[i] >= bsrstart && idxm[i] < bsrend) {
442: row = idxm[i] - bsrstart;
443: for (j=0; j<n; j++) {
444: if (idxn[j] < 0) continue; /* SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative column %D",idxn[j]); */
445: if (idxn[j] >= mat->cmap.N) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",idxn[j],mat->cmap.N-1);
446: if (idxn[j] >= bscstart && idxn[j] < bscend){
447: col = idxn[j] - bscstart;
448: MatGetValues_SeqSBAIJ(baij->A,1,&row,1,&col,v+i*n+j);
449: } else {
450: if (!baij->colmap) {
451: CreateColmap_MPIBAIJ_Private(mat);
452: }
453: #if defined (PETSC_USE_CTABLE)
454: PetscTableFind(baij->colmap,idxn[j]/bs+1,&data);
455: data --;
456: #else
457: data = baij->colmap[idxn[j]/bs]-1;
458: #endif
459: if((data < 0) || (baij->garray[data/bs] != idxn[j]/bs)) *(v+i*n+j) = 0.0;
460: else {
461: col = data + idxn[j]%bs;
462: MatGetValues_SeqBAIJ(baij->B,1,&row,1,&col,v+i*n+j);
463: }
464: }
465: }
466: } else {
467: SETERRQ(PETSC_ERR_SUP,"Only local values currently supported");
468: }
469: }
470: return(0);
471: }
475: PetscErrorCode MatNorm_MPISBAIJ(Mat mat,NormType type,PetscReal *norm)
476: {
477: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
479: PetscReal sum[2],*lnorm2;
482: if (baij->size == 1) {
483: MatNorm(baij->A,type,norm);
484: } else {
485: if (type == NORM_FROBENIUS) {
486: PetscMalloc(2*sizeof(PetscReal),&lnorm2);
487: MatNorm(baij->A,type,lnorm2);
488: *lnorm2 = (*lnorm2)*(*lnorm2); lnorm2++; /* squar power of norm(A) */
489: MatNorm(baij->B,type,lnorm2);
490: *lnorm2 = (*lnorm2)*(*lnorm2); lnorm2--; /* squar power of norm(B) */
491: MPI_Allreduce(lnorm2,&sum,2,MPIU_REAL,MPI_SUM,((PetscObject)mat)->comm);
492: *norm = sqrt(sum[0] + 2*sum[1]);
493: PetscFree(lnorm2);
494: } else if (type == NORM_INFINITY || type == NORM_1) { /* max row/column sum */
495: Mat_SeqSBAIJ *amat=(Mat_SeqSBAIJ*)baij->A->data;
496: Mat_SeqBAIJ *bmat=(Mat_SeqBAIJ*)baij->B->data;
497: PetscReal *rsum,*rsum2,vabs;
498: PetscInt *jj,*garray=baij->garray,rstart=baij->rstartbs,nz;
499: PetscInt brow,bcol,col,bs=baij->A->rmap.bs,row,grow,gcol,mbs=amat->mbs;
500: MatScalar *v;
502: PetscMalloc((2*mat->cmap.N+1)*sizeof(PetscReal),&rsum);
503: rsum2 = rsum + mat->cmap.N;
504: PetscMemzero(rsum,mat->cmap.N*sizeof(PetscReal));
505: /* Amat */
506: v = amat->a; jj = amat->j;
507: for (brow=0; brow<mbs; brow++) {
508: grow = bs*(rstart + brow);
509: nz = amat->i[brow+1] - amat->i[brow];
510: for (bcol=0; bcol<nz; bcol++){
511: gcol = bs*(rstart + *jj); jj++;
512: for (col=0; col<bs; col++){
513: for (row=0; row<bs; row++){
514: vabs = PetscAbsScalar(*v); v++;
515: rsum[gcol+col] += vabs;
516: /* non-diagonal block */
517: if (bcol > 0 && vabs > 0.0) rsum[grow+row] += vabs;
518: }
519: }
520: }
521: }
522: /* Bmat */
523: v = bmat->a; jj = bmat->j;
524: for (brow=0; brow<mbs; brow++) {
525: grow = bs*(rstart + brow);
526: nz = bmat->i[brow+1] - bmat->i[brow];
527: for (bcol=0; bcol<nz; bcol++){
528: gcol = bs*garray[*jj]; jj++;
529: for (col=0; col<bs; col++){
530: for (row=0; row<bs; row++){
531: vabs = PetscAbsScalar(*v); v++;
532: rsum[gcol+col] += vabs;
533: rsum[grow+row] += vabs;
534: }
535: }
536: }
537: }
538: MPI_Allreduce(rsum,rsum2,mat->cmap.N,MPIU_REAL,MPI_SUM,((PetscObject)mat)->comm);
539: *norm = 0.0;
540: for (col=0; col<mat->cmap.N; col++) {
541: if (rsum2[col] > *norm) *norm = rsum2[col];
542: }
543: PetscFree(rsum);
544: } else {
545: SETERRQ(PETSC_ERR_SUP,"No support for this norm yet");
546: }
547: }
548: return(0);
549: }
553: PetscErrorCode MatAssemblyBegin_MPISBAIJ(Mat mat,MatAssemblyType mode)
554: {
555: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
557: PetscInt nstash,reallocs;
558: InsertMode addv;
561: if (baij->donotstash) {
562: return(0);
563: }
565: /* make sure all processors are either in INSERTMODE or ADDMODE */
566: MPI_Allreduce(&mat->insertmode,&addv,1,MPI_INT,MPI_BOR,((PetscObject)mat)->comm);
567: if (addv == (ADD_VALUES|INSERT_VALUES)) {
568: SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Some processors inserted others added");
569: }
570: mat->insertmode = addv; /* in case this processor had no cache */
572: MatStashScatterBegin_Private(mat,&mat->stash,mat->rmap.range);
573: MatStashScatterBegin_Private(mat,&mat->bstash,baij->rangebs);
574: MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);
575: PetscInfo2(mat,"Stash has %D entries,uses %D mallocs.\n",nstash,reallocs);
576: MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);
577: PetscInfo2(mat,"Block-Stash has %D entries, uses %D mallocs.\n",nstash,reallocs);
578: return(0);
579: }
583: PetscErrorCode MatAssemblyEnd_MPISBAIJ(Mat mat,MatAssemblyType mode)
584: {
585: Mat_MPISBAIJ *baij=(Mat_MPISBAIJ*)mat->data;
586: Mat_SeqSBAIJ *a=(Mat_SeqSBAIJ*)baij->A->data;
588: PetscInt i,j,rstart,ncols,flg,bs2=baij->bs2;
589: PetscInt *row,*col,other_disassembled;
590: PetscMPIInt n;
591: PetscTruth r1,r2,r3;
592: MatScalar *val;
593: InsertMode addv = mat->insertmode;
595: /* do not use 'b=(Mat_SeqBAIJ*)baij->B->data' as B can be reset in disassembly */
598: if (!baij->donotstash) {
599: while (1) {
600: MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);
601: if (!flg) break;
603: for (i=0; i<n;) {
604: /* Now identify the consecutive vals belonging to the same row */
605: for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; }
606: if (j < n) ncols = j-i;
607: else ncols = n-i;
608: /* Now assemble all these values with a single function call */
609: MatSetValues_MPISBAIJ_MatScalar(mat,1,row+i,ncols,col+i,val+i,addv);
610: i = j;
611: }
612: }
613: MatStashScatterEnd_Private(&mat->stash);
614: /* Now process the block-stash. Since the values are stashed column-oriented,
615: set the roworiented flag to column oriented, and after MatSetValues()
616: restore the original flags */
617: r1 = baij->roworiented;
618: r2 = a->roworiented;
619: r3 = ((Mat_SeqBAIJ*)baij->B->data)->roworiented;
620: baij->roworiented = PETSC_FALSE;
621: a->roworiented = PETSC_FALSE;
622: ((Mat_SeqBAIJ*)baij->B->data)->roworiented = PETSC_FALSE; /* b->roworinted */
623: while (1) {
624: MatStashScatterGetMesg_Private(&mat->bstash,&n,&row,&col,&val,&flg);
625: if (!flg) break;
626:
627: for (i=0; i<n;) {
628: /* Now identify the consecutive vals belonging to the same row */
629: for (j=i,rstart=row[j]; j<n; j++) { if (row[j] != rstart) break; }
630: if (j < n) ncols = j-i;
631: else ncols = n-i;
632: MatSetValuesBlocked_MPISBAIJ_MatScalar(mat,1,row+i,ncols,col+i,val+i*bs2,addv);
633: i = j;
634: }
635: }
636: MatStashScatterEnd_Private(&mat->bstash);
637: baij->roworiented = r1;
638: a->roworiented = r2;
639: ((Mat_SeqBAIJ*)baij->B->data)->roworiented = r3; /* b->roworinted */
640: }
642: MatAssemblyBegin(baij->A,mode);
643: MatAssemblyEnd(baij->A,mode);
645: /* determine if any processor has disassembled, if so we must
646: also disassemble ourselfs, in order that we may reassemble. */
647: /*
648: if nonzero structure of submatrix B cannot change then we know that
649: no processor disassembled thus we can skip this stuff
650: */
651: if (!((Mat_SeqBAIJ*)baij->B->data)->nonew) {
652: MPI_Allreduce(&mat->was_assembled,&other_disassembled,1,MPI_INT,MPI_PROD,((PetscObject)mat)->comm);
653: if (mat->was_assembled && !other_disassembled) {
654: DisAssemble_MPISBAIJ(mat);
655: }
656: }
658: if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) {
659: MatSetUpMultiply_MPISBAIJ(mat); /* setup Mvctx and sMvctx */
660: }
661: ((Mat_SeqBAIJ*)baij->B->data)->compressedrow.use = PETSC_TRUE; /* b->compressedrow.use */
662: MatAssemblyBegin(baij->B,mode);
663: MatAssemblyEnd(baij->B,mode);
664:
665: PetscFree(baij->rowvalues);
666: baij->rowvalues = 0;
668: return(0);
669: }
674: static PetscErrorCode MatView_MPISBAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer)
675: {
676: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
677: PetscErrorCode ierr;
678: PetscInt bs = mat->rmap.bs;
679: PetscMPIInt size = baij->size,rank = baij->rank;
680: PetscTruth iascii,isdraw;
681: PetscViewer sviewer;
682: PetscViewerFormat format;
685: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);
686: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
687: if (iascii) {
688: PetscViewerGetFormat(viewer,&format);
689: if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
690: MatInfo info;
691: MPI_Comm_rank(((PetscObject)mat)->comm,&rank);
692: MatGetInfo(mat,MAT_LOCAL,&info);
693: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D bs %D mem %D\n",
694: rank,mat->rmap.N,(PetscInt)info.nz_used*bs,(PetscInt)info.nz_allocated*bs,
695: mat->rmap.bs,(PetscInt)info.memory);
696: MatGetInfo(baij->A,MAT_LOCAL,&info);
697: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used*bs);
698: MatGetInfo(baij->B,MAT_LOCAL,&info);
699: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used*bs);
700: PetscViewerFlush(viewer);
701: PetscViewerASCIIPrintf(viewer,"Information on VecScatter used in matrix-vector product: \n");
702: VecScatterView(baij->Mvctx,viewer);
703: return(0);
704: } else if (format == PETSC_VIEWER_ASCII_INFO) {
705: PetscViewerASCIIPrintf(viewer," block size is %D\n",bs);
706: return(0);
707: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
708: return(0);
709: }
710: }
712: if (isdraw) {
713: PetscDraw draw;
714: PetscTruth isnull;
715: PetscViewerDrawGetDraw(viewer,0,&draw);
716: PetscDrawIsNull(draw,&isnull); if (isnull) return(0);
717: }
719: if (size == 1) {
720: PetscObjectSetName((PetscObject)baij->A,((PetscObject)mat)->name);
721: MatView(baij->A,viewer);
722: } else {
723: /* assemble the entire matrix onto first processor. */
724: Mat A;
725: Mat_SeqSBAIJ *Aloc;
726: Mat_SeqBAIJ *Bloc;
727: PetscInt M = mat->rmap.N,N = mat->cmap.N,*ai,*aj,col,i,j,k,*rvals,mbs = baij->mbs;
728: MatScalar *a;
730: /* Should this be the same type as mat? */
731: MatCreate(((PetscObject)mat)->comm,&A);
732: if (!rank) {
733: MatSetSizes(A,M,N,M,N);
734: } else {
735: MatSetSizes(A,0,0,M,N);
736: }
737: MatSetType(A,MATMPISBAIJ);
738: MatMPISBAIJSetPreallocation(A,mat->rmap.bs,0,PETSC_NULL,0,PETSC_NULL);
739: PetscLogObjectParent(mat,A);
741: /* copy over the A part */
742: Aloc = (Mat_SeqSBAIJ*)baij->A->data;
743: ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
744: PetscMalloc(bs*sizeof(PetscInt),&rvals);
746: for (i=0; i<mbs; i++) {
747: rvals[0] = bs*(baij->rstartbs + i);
748: for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
749: for (j=ai[i]; j<ai[i+1]; j++) {
750: col = (baij->cstartbs+aj[j])*bs;
751: for (k=0; k<bs; k++) {
752: MatSetValues_MPISBAIJ_MatScalar(A,bs,rvals,1,&col,a,INSERT_VALUES);
753: col++; a += bs;
754: }
755: }
756: }
757: /* copy over the B part */
758: Bloc = (Mat_SeqBAIJ*)baij->B->data;
759: ai = Bloc->i; aj = Bloc->j; a = Bloc->a;
760: for (i=0; i<mbs; i++) {
761:
762: rvals[0] = bs*(baij->rstartbs + i);
763: for (j=1; j<bs; j++) { rvals[j] = rvals[j-1] + 1; }
764: for (j=ai[i]; j<ai[i+1]; j++) {
765: col = baij->garray[aj[j]]*bs;
766: for (k=0; k<bs; k++) {
767: MatSetValues_MPIBAIJ(A,bs,rvals,1,&col,a,INSERT_VALUES);
768: col++; a += bs;
769: }
770: }
771: }
772: PetscFree(rvals);
773: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
774: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
775: /*
776: Everyone has to call to draw the matrix since the graphics waits are
777: synchronized across all processors that share the PetscDraw object
778: */
779: PetscViewerGetSingleton(viewer,&sviewer);
780: if (!rank) {
781: PetscObjectSetName((PetscObject)((Mat_MPISBAIJ*)(A->data))->A,((PetscObject)mat)->name);
782: MatView(((Mat_MPISBAIJ*)(A->data))->A,sviewer);
783: }
784: PetscViewerRestoreSingleton(viewer,&sviewer);
785: MatDestroy(A);
786: }
787: return(0);
788: }
792: PetscErrorCode MatView_MPISBAIJ(Mat mat,PetscViewer viewer)
793: {
795: PetscTruth iascii,isdraw,issocket,isbinary;
798: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);
799: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
800: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_SOCKET,&issocket);
801: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);
802: if (iascii || isdraw || issocket || isbinary) {
803: MatView_MPISBAIJ_ASCIIorDraworSocket(mat,viewer);
804: } else {
805: SETERRQ1(PETSC_ERR_SUP,"Viewer type %s not supported by MPISBAIJ matrices",((PetscObject)viewer)->type_name);
806: }
807: return(0);
808: }
812: PetscErrorCode MatDestroy_MPISBAIJ(Mat mat)
813: {
814: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
818: #if defined(PETSC_USE_LOG)
819: PetscLogObjectState((PetscObject)mat,"Rows=%D,Cols=%D",mat->rmap.N,mat->cmap.N);
820: #endif
821: MatStashDestroy_Private(&mat->stash);
822: MatStashDestroy_Private(&mat->bstash);
823: MatDestroy(baij->A);
824: MatDestroy(baij->B);
825: #if defined (PETSC_USE_CTABLE)
826: if (baij->colmap) {PetscTableDestroy(baij->colmap);}
827: #else
828: PetscFree(baij->colmap);
829: #endif
830: PetscFree(baij->garray);
831: if (baij->lvec) {VecDestroy(baij->lvec);}
832: if (baij->Mvctx) {VecScatterDestroy(baij->Mvctx);}
833: if (baij->slvec0) {
834: VecDestroy(baij->slvec0);
835: VecDestroy(baij->slvec0b);
836: }
837: if (baij->slvec1) {
838: VecDestroy(baij->slvec1);
839: VecDestroy(baij->slvec1a);
840: VecDestroy(baij->slvec1b);
841: }
842: if (baij->sMvctx) {VecScatterDestroy(baij->sMvctx);}
843: PetscFree(baij->rowvalues);
844: PetscFree(baij->barray);
845: PetscFree(baij->hd);
846: #if defined(PETSC_USE_MAT_SINGLE)
847: PetscFree(baij->setvaluescopy);
848: #endif
849: PetscFree(baij->in_loc);
850: PetscFree(baij->v_loc);
851: PetscFree(baij->rangebs);
852: PetscFree(baij);
854: PetscObjectChangeTypeName((PetscObject)mat,0);
855: PetscObjectComposeFunction((PetscObject)mat,"MatStoreValues_C","",PETSC_NULL);
856: PetscObjectComposeFunction((PetscObject)mat,"MatRetrieveValues_C","",PETSC_NULL);
857: PetscObjectComposeFunction((PetscObject)mat,"MatGetDiagonalBlock_C","",PETSC_NULL);
858: PetscObjectComposeFunction((PetscObject)mat,"MatMPISBAIJSetPreallocation_C","",PETSC_NULL);
859: return(0);
860: }
864: PetscErrorCode MatMult_MPISBAIJ(Mat A,Vec xx,Vec yy)
865: {
866: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
868: PetscInt nt,mbs=a->mbs,bs=A->rmap.bs;
869: PetscScalar *x,*from,zero=0.0;
870:
872: VecGetLocalSize(xx,&nt);
873: if (nt != A->cmap.n) {
874: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible partition of A and xx");
875: }
877: /* diagonal part */
878: (*a->A->ops->mult)(a->A,xx,a->slvec1a);
879: VecSet(a->slvec1b,zero);
881: /* subdiagonal part */
882: (*a->B->ops->multtranspose)(a->B,xx,a->slvec0b);
884: /* copy x into the vec slvec0 */
885: VecGetArray(a->slvec0,&from);
886: VecGetArray(xx,&x);
888: PetscMemcpy(from,x,bs*mbs*sizeof(MatScalar));
889: VecRestoreArray(a->slvec0,&from);
890: VecRestoreArray(xx,&x);
891:
892: VecScatterBegin(a->sMvctx,a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD);
893: VecScatterEnd(a->sMvctx,a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD);
894: /* supperdiagonal part */
895: (*a->B->ops->multadd)(a->B,a->slvec1b,a->slvec1a,yy);
896: return(0);
897: }
901: PetscErrorCode MatMult_MPISBAIJ_2comm(Mat A,Vec xx,Vec yy)
902: {
903: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
905: PetscInt nt;
908: VecGetLocalSize(xx,&nt);
909: if (nt != A->cmap.n) {
910: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible partition of A and xx");
911: }
912: VecGetLocalSize(yy,&nt);
913: if (nt != A->rmap.N) {
914: SETERRQ(PETSC_ERR_ARG_SIZ,"Incompatible parition of A and yy");
915: }
917: VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
918: /* do diagonal part */
919: (*a->A->ops->mult)(a->A,xx,yy);
920: /* do supperdiagonal part */
921: VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
922: (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);
923: /* do subdiagonal part */
924: (*a->B->ops->multtranspose)(a->B,xx,a->lvec);
925: VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);
926: VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);
928: return(0);
929: }
933: PetscErrorCode MatMultAdd_MPISBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
934: {
935: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
937: PetscInt mbs=a->mbs,bs=A->rmap.bs;
938: PetscScalar *x,*from,zero=0.0;
939:
941: /*
942: PetscSynchronizedPrintf(((PetscObject)A)->comm," MatMultAdd is called ...\n");
943: PetscSynchronizedFlush(((PetscObject)A)->comm);
944: */
945: /* diagonal part */
946: (*a->A->ops->multadd)(a->A,xx,yy,a->slvec1a);
947: VecSet(a->slvec1b,zero);
949: /* subdiagonal part */
950: (*a->B->ops->multtranspose)(a->B,xx,a->slvec0b);
952: /* copy x into the vec slvec0 */
953: VecGetArray(a->slvec0,&from);
954: VecGetArray(xx,&x);
955: PetscMemcpy(from,x,bs*mbs*sizeof(MatScalar));
956: VecRestoreArray(a->slvec0,&from);
957:
958: VecScatterBegin(a->sMvctx,a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD);
959: VecRestoreArray(xx,&x);
960: VecScatterEnd(a->sMvctx,a->slvec0,a->slvec1,ADD_VALUES,SCATTER_FORWARD);
961:
962: /* supperdiagonal part */
963: (*a->B->ops->multadd)(a->B,a->slvec1b,a->slvec1a,zz);
964:
965: return(0);
966: }
970: PetscErrorCode MatMultAdd_MPISBAIJ_2comm(Mat A,Vec xx,Vec yy,Vec zz)
971: {
972: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
976: VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
977: /* do diagonal part */
978: (*a->A->ops->multadd)(a->A,xx,yy,zz);
979: /* do supperdiagonal part */
980: VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
981: (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);
983: /* do subdiagonal part */
984: (*a->B->ops->multtranspose)(a->B,xx,a->lvec);
985: VecScatterBegin(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);
986: VecScatterEnd(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);
988: return(0);
989: }
991: /*
992: This only works correctly for square matrices where the subblock A->A is the
993: diagonal block
994: */
997: PetscErrorCode MatGetDiagonal_MPISBAIJ(Mat A,Vec v)
998: {
999: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1003: /* if (a->rmap.N != a->cmap.N) SETERRQ(PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block"); */
1004: MatGetDiagonal(a->A,v);
1005: return(0);
1006: }
1010: PetscErrorCode MatScale_MPISBAIJ(Mat A,PetscScalar aa)
1011: {
1012: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1016: MatScale(a->A,aa);
1017: MatScale(a->B,aa);
1018: return(0);
1019: }
1023: PetscErrorCode MatGetRow_MPISBAIJ(Mat matin,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1024: {
1025: Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)matin->data;
1026: PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p;
1028: PetscInt bs = matin->rmap.bs,bs2 = mat->bs2,i,*cworkA,*cworkB,**pcA,**pcB;
1029: PetscInt nztot,nzA,nzB,lrow,brstart = matin->rmap.rstart,brend = matin->rmap.rend;
1030: PetscInt *cmap,*idx_p,cstart = mat->rstartbs;
1033: if (mat->getrowactive == PETSC_TRUE) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Already active");
1034: mat->getrowactive = PETSC_TRUE;
1036: if (!mat->rowvalues && (idx || v)) {
1037: /*
1038: allocate enough space to hold information from the longest row.
1039: */
1040: Mat_SeqSBAIJ *Aa = (Mat_SeqSBAIJ*)mat->A->data;
1041: Mat_SeqBAIJ *Ba = (Mat_SeqBAIJ*)mat->B->data;
1042: PetscInt max = 1,mbs = mat->mbs,tmp;
1043: for (i=0; i<mbs; i++) {
1044: tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i]; /* row length */
1045: if (max < tmp) { max = tmp; }
1046: }
1047: PetscMalloc(max*bs2*(sizeof(PetscInt)+sizeof(PetscScalar)),&mat->rowvalues);
1048: mat->rowindices = (PetscInt*)(mat->rowvalues + max*bs2);
1049: }
1050:
1051: if (row < brstart || row >= brend) SETERRQ(PETSC_ERR_SUP,"Only local rows")
1052: lrow = row - brstart; /* local row index */
1054: pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB;
1055: if (!v) {pvA = 0; pvB = 0;}
1056: if (!idx) {pcA = 0; if (!v) pcB = 0;}
1057: (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);
1058: (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);
1059: nztot = nzA + nzB;
1061: cmap = mat->garray;
1062: if (v || idx) {
1063: if (nztot) {
1064: /* Sort by increasing column numbers, assuming A and B already sorted */
1065: PetscInt imark = -1;
1066: if (v) {
1067: *v = v_p = mat->rowvalues;
1068: for (i=0; i<nzB; i++) {
1069: if (cmap[cworkB[i]/bs] < cstart) v_p[i] = vworkB[i];
1070: else break;
1071: }
1072: imark = i;
1073: for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i];
1074: for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i];
1075: }
1076: if (idx) {
1077: *idx = idx_p = mat->rowindices;
1078: if (imark > -1) {
1079: for (i=0; i<imark; i++) {
1080: idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs;
1081: }
1082: } else {
1083: for (i=0; i<nzB; i++) {
1084: if (cmap[cworkB[i]/bs] < cstart)
1085: idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1086: else break;
1087: }
1088: imark = i;
1089: }
1090: for (i=0; i<nzA; i++) idx_p[imark+i] = cstart*bs + cworkA[i];
1091: for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1092: }
1093: } else {
1094: if (idx) *idx = 0;
1095: if (v) *v = 0;
1096: }
1097: }
1098: *nz = nztot;
1099: (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);
1100: (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);
1101: return(0);
1102: }
1106: PetscErrorCode MatRestoreRow_MPISBAIJ(Mat mat,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1107: {
1108: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
1111: if (!baij->getrowactive) {
1112: SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"MatGetRow() must be called first");
1113: }
1114: baij->getrowactive = PETSC_FALSE;
1115: return(0);
1116: }
1120: PetscErrorCode MatGetRowUpperTriangular_MPISBAIJ(Mat A)
1121: {
1122: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1123: Mat_SeqSBAIJ *aA = (Mat_SeqSBAIJ*)a->A->data;
1126: aA->getrow_utriangular = PETSC_TRUE;
1127: return(0);
1128: }
1131: PetscErrorCode MatRestoreRowUpperTriangular_MPISBAIJ(Mat A)
1132: {
1133: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1134: Mat_SeqSBAIJ *aA = (Mat_SeqSBAIJ*)a->A->data;
1137: aA->getrow_utriangular = PETSC_FALSE;
1138: return(0);
1139: }
1143: PetscErrorCode MatRealPart_MPISBAIJ(Mat A)
1144: {
1145: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1149: MatRealPart(a->A);
1150: MatRealPart(a->B);
1151: return(0);
1152: }
1156: PetscErrorCode MatImaginaryPart_MPISBAIJ(Mat A)
1157: {
1158: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1162: MatImaginaryPart(a->A);
1163: MatImaginaryPart(a->B);
1164: return(0);
1165: }
1169: PetscErrorCode MatZeroEntries_MPISBAIJ(Mat A)
1170: {
1171: Mat_MPISBAIJ *l = (Mat_MPISBAIJ*)A->data;
1175: MatZeroEntries(l->A);
1176: MatZeroEntries(l->B);
1177: return(0);
1178: }
1182: PetscErrorCode MatGetInfo_MPISBAIJ(Mat matin,MatInfoType flag,MatInfo *info)
1183: {
1184: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)matin->data;
1185: Mat A = a->A,B = a->B;
1187: PetscReal isend[5],irecv[5];
1190: info->block_size = (PetscReal)matin->rmap.bs;
1191: MatGetInfo(A,MAT_LOCAL,info);
1192: isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded;
1193: isend[3] = info->memory; isend[4] = info->mallocs;
1194: MatGetInfo(B,MAT_LOCAL,info);
1195: isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded;
1196: isend[3] += info->memory; isend[4] += info->mallocs;
1197: if (flag == MAT_LOCAL) {
1198: info->nz_used = isend[0];
1199: info->nz_allocated = isend[1];
1200: info->nz_unneeded = isend[2];
1201: info->memory = isend[3];
1202: info->mallocs = isend[4];
1203: } else if (flag == MAT_GLOBAL_MAX) {
1204: MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_MAX,((PetscObject)matin)->comm);
1205: info->nz_used = irecv[0];
1206: info->nz_allocated = irecv[1];
1207: info->nz_unneeded = irecv[2];
1208: info->memory = irecv[3];
1209: info->mallocs = irecv[4];
1210: } else if (flag == MAT_GLOBAL_SUM) {
1211: MPI_Allreduce(isend,irecv,5,MPIU_REAL,MPI_SUM,((PetscObject)matin)->comm);
1212: info->nz_used = irecv[0];
1213: info->nz_allocated = irecv[1];
1214: info->nz_unneeded = irecv[2];
1215: info->memory = irecv[3];
1216: info->mallocs = irecv[4];
1217: } else {
1218: SETERRQ1(PETSC_ERR_ARG_WRONG,"Unknown MatInfoType argument %d",(int)flag);
1219: }
1220: info->rows_global = (PetscReal)A->rmap.N;
1221: info->columns_global = (PetscReal)A->cmap.N;
1222: info->rows_local = (PetscReal)A->rmap.N;
1223: info->columns_local = (PetscReal)A->cmap.N;
1224: info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */
1225: info->fill_ratio_needed = 0;
1226: info->factor_mallocs = 0;
1227: return(0);
1228: }
1232: PetscErrorCode MatSetOption_MPISBAIJ(Mat A,MatOption op,PetscTruth flg)
1233: {
1234: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1235: Mat_SeqSBAIJ *aA = (Mat_SeqSBAIJ*)a->A->data;
1239: switch (op) {
1240: case MAT_NEW_NONZERO_LOCATIONS:
1241: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1242: case MAT_KEEP_ZEROED_ROWS:
1243: case MAT_NEW_NONZERO_LOCATION_ERR:
1244: MatSetOption(a->A,op,flg);
1245: MatSetOption(a->B,op,flg);
1246: break;
1247: case MAT_ROW_ORIENTED:
1248: a->roworiented = flg;
1249: MatSetOption(a->A,op,flg);
1250: MatSetOption(a->B,op,flg);
1251: break;
1252: case MAT_NEW_DIAGONALS:
1253: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1254: break;
1255: case MAT_IGNORE_OFF_PROC_ENTRIES:
1256: a->donotstash = flg;
1257: break;
1258: case MAT_USE_HASH_TABLE:
1259: a->ht_flag = flg;
1260: break;
1261: case MAT_HERMITIAN:
1262: if (flg) SETERRQ(PETSC_ERR_SUP,"Matrix must be symmetric");
1263: case MAT_SYMMETRIC:
1264: case MAT_STRUCTURALLY_SYMMETRIC:
1265: case MAT_SYMMETRY_ETERNAL:
1266: if (!flg) SETERRQ(PETSC_ERR_SUP,"Matrix must be symmetric");
1267: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1268: break;
1269: case MAT_IGNORE_LOWER_TRIANGULAR:
1270: aA->ignore_ltriangular = flg;
1271: break;
1272: case MAT_ERROR_LOWER_TRIANGULAR:
1273: aA->ignore_ltriangular = flg;
1274: break;
1275: case MAT_GETROW_UPPERTRIANGULAR:
1276: aA->getrow_utriangular = flg;
1277: break;
1278: default:
1279: SETERRQ1(PETSC_ERR_SUP,"unknown option %d",op);
1280: }
1281: return(0);
1282: }
1286: PetscErrorCode MatTranspose_MPISBAIJ(Mat A,Mat *B)
1287: {
1290: MatDuplicate(A,MAT_COPY_VALUES,B);
1291: return(0);
1292: }
1296: PetscErrorCode MatDiagonalScale_MPISBAIJ(Mat mat,Vec ll,Vec rr)
1297: {
1298: Mat_MPISBAIJ *baij = (Mat_MPISBAIJ*)mat->data;
1299: Mat a=baij->A, b=baij->B;
1301: PetscInt nv,m,n;
1302: PetscTruth flg;
1305: if (ll != rr){
1306: VecEqual(ll,rr,&flg);
1307: if (!flg)
1308: SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"For symmetric format, left and right scaling vectors must be same\n");
1309: }
1310: if (!ll) return(0);
1312: MatGetLocalSize(mat,&m,&n);
1313: if (m != n) SETERRQ2(PETSC_ERR_ARG_SIZ,"For symmetric format, local size %d %d must be same",m,n);
1314:
1315: VecGetLocalSize(rr,&nv);
1316: if (nv!=n) SETERRQ(PETSC_ERR_ARG_SIZ,"Left and right vector non-conforming local size");
1318: VecScatterBegin(baij->Mvctx,rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD);
1319:
1320: /* left diagonalscale the off-diagonal part */
1321: (*b->ops->diagonalscale)(b,ll,PETSC_NULL);
1322:
1323: /* scale the diagonal part */
1324: (*a->ops->diagonalscale)(a,ll,rr);
1326: /* right diagonalscale the off-diagonal part */
1327: VecScatterEnd(baij->Mvctx,rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD);
1328: (*b->ops->diagonalscale)(b,PETSC_NULL,baij->lvec);
1329: return(0);
1330: }
1334: PetscErrorCode MatSetUnfactored_MPISBAIJ(Mat A)
1335: {
1336: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
1340: MatSetUnfactored(a->A);
1341: return(0);
1342: }
1344: static PetscErrorCode MatDuplicate_MPISBAIJ(Mat,MatDuplicateOption,Mat *);
1348: PetscErrorCode MatEqual_MPISBAIJ(Mat A,Mat B,PetscTruth *flag)
1349: {
1350: Mat_MPISBAIJ *matB = (Mat_MPISBAIJ*)B->data,*matA = (Mat_MPISBAIJ*)A->data;
1351: Mat a,b,c,d;
1352: PetscTruth flg;
1356: a = matA->A; b = matA->B;
1357: c = matB->A; d = matB->B;
1359: MatEqual(a,c,&flg);
1360: if (flg) {
1361: MatEqual(b,d,&flg);
1362: }
1363: MPI_Allreduce(&flg,flag,1,MPI_INT,MPI_LAND,((PetscObject)A)->comm);
1364: return(0);
1365: }
1369: PetscErrorCode MatCopy_MPISBAIJ(Mat A,Mat B,MatStructure str)
1370: {
1372: Mat_MPISBAIJ *a = (Mat_MPISBAIJ *)A->data;
1373: Mat_MPISBAIJ *b = (Mat_MPISBAIJ *)B->data;
1376: /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */
1377: if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) {
1378: MatGetRowUpperTriangular(A);
1379: MatCopy_Basic(A,B,str);
1380: MatRestoreRowUpperTriangular(A);
1381: } else {
1382: MatCopy(a->A,b->A,str);
1383: MatCopy(a->B,b->B,str);
1384: }
1385: return(0);
1386: }
1390: PetscErrorCode MatSetUpPreallocation_MPISBAIJ(Mat A)
1391: {
1395: MatMPISBAIJSetPreallocation(A,PetscMax(A->rmap.bs,1),PETSC_DEFAULT,0,PETSC_DEFAULT,0);
1396: return(0);
1397: }
1399: #include petscblaslapack.h
1402: PetscErrorCode MatAXPY_MPISBAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
1403: {
1405: Mat_MPISBAIJ *xx=(Mat_MPISBAIJ *)X->data,*yy=(Mat_MPISBAIJ *)Y->data;
1406: PetscBLASInt bnz,one=1;
1407: Mat_SeqSBAIJ *xa,*ya;
1408: Mat_SeqBAIJ *xb,*yb;
1411: if (str == SAME_NONZERO_PATTERN) {
1412: PetscScalar alpha = a;
1413: xa = (Mat_SeqSBAIJ *)xx->A->data;
1414: ya = (Mat_SeqSBAIJ *)yy->A->data;
1415: bnz = (PetscBLASInt)xa->nz;
1416: BLASaxpy_(&bnz,&alpha,xa->a,&one,ya->a,&one);
1417: xb = (Mat_SeqBAIJ *)xx->B->data;
1418: yb = (Mat_SeqBAIJ *)yy->B->data;
1419: bnz = (PetscBLASInt)xb->nz;
1420: BLASaxpy_(&bnz,&alpha,xb->a,&one,yb->a,&one);
1421: } else {
1422: MatGetRowUpperTriangular(X);
1423: MatAXPY_Basic(Y,a,X,str);
1424: MatRestoreRowUpperTriangular(X);
1425: }
1426: return(0);
1427: }
1431: PetscErrorCode MatGetSubMatrices_MPISBAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
1432: {
1434: PetscInt i;
1435: PetscTruth flg;
1438: for (i=0; i<n; i++) {
1439: ISEqual(irow[i],icol[i],&flg);
1440: if (!flg) {
1441: SETERRQ(PETSC_ERR_SUP,"Can only get symmetric submatrix for MPISBAIJ matrices");
1442: }
1443: }
1444: MatGetSubMatrices_MPIBAIJ(A,n,irow,icol,scall,B);
1445: return(0);
1446: }
1447:
1449: /* -------------------------------------------------------------------*/
1450: static struct _MatOps MatOps_Values = {
1451: MatSetValues_MPISBAIJ,
1452: MatGetRow_MPISBAIJ,
1453: MatRestoreRow_MPISBAIJ,
1454: MatMult_MPISBAIJ,
1455: /* 4*/ MatMultAdd_MPISBAIJ,
1456: MatMult_MPISBAIJ, /* transpose versions are same as non-transpose */
1457: MatMultAdd_MPISBAIJ,
1458: 0,
1459: 0,
1460: 0,
1461: /*10*/ 0,
1462: 0,
1463: 0,
1464: MatRelax_MPISBAIJ,
1465: MatTranspose_MPISBAIJ,
1466: /*15*/ MatGetInfo_MPISBAIJ,
1467: MatEqual_MPISBAIJ,
1468: MatGetDiagonal_MPISBAIJ,
1469: MatDiagonalScale_MPISBAIJ,
1470: MatNorm_MPISBAIJ,
1471: /*20*/ MatAssemblyBegin_MPISBAIJ,
1472: MatAssemblyEnd_MPISBAIJ,
1473: 0,
1474: MatSetOption_MPISBAIJ,
1475: MatZeroEntries_MPISBAIJ,
1476: /*25*/ 0,
1477: 0,
1478: 0,
1479: 0,
1480: 0,
1481: /*30*/ MatSetUpPreallocation_MPISBAIJ,
1482: 0,
1483: 0,
1484: 0,
1485: 0,
1486: /*35*/ MatDuplicate_MPISBAIJ,
1487: 0,
1488: 0,
1489: 0,
1490: 0,
1491: /*40*/ MatAXPY_MPISBAIJ,
1492: MatGetSubMatrices_MPISBAIJ,
1493: MatIncreaseOverlap_MPISBAIJ,
1494: MatGetValues_MPISBAIJ,
1495: MatCopy_MPISBAIJ,
1496: /*45*/ 0,
1497: MatScale_MPISBAIJ,
1498: 0,
1499: 0,
1500: 0,
1501: /*50*/ 0,
1502: 0,
1503: 0,
1504: 0,
1505: 0,
1506: /*55*/ 0,
1507: 0,
1508: MatSetUnfactored_MPISBAIJ,
1509: 0,
1510: MatSetValuesBlocked_MPISBAIJ,
1511: /*60*/ 0,
1512: 0,
1513: 0,
1514: 0,
1515: 0,
1516: /*65*/ 0,
1517: 0,
1518: 0,
1519: 0,
1520: 0,
1521: /*70*/ MatGetRowMaxAbs_MPISBAIJ,
1522: 0,
1523: 0,
1524: 0,
1525: 0,
1526: /*75*/ 0,
1527: 0,
1528: 0,
1529: 0,
1530: 0,
1531: /*80*/ 0,
1532: 0,
1533: 0,
1534: 0,
1535: MatLoad_MPISBAIJ,
1536: /*85*/ 0,
1537: 0,
1538: 0,
1539: 0,
1540: 0,
1541: /*90*/ 0,
1542: 0,
1543: 0,
1544: 0,
1545: 0,
1546: /*95*/ 0,
1547: 0,
1548: 0,
1549: 0,
1550: 0,
1551: /*100*/0,
1552: 0,
1553: 0,
1554: 0,
1555: 0,
1556: /*105*/0,
1557: MatRealPart_MPISBAIJ,
1558: MatImaginaryPart_MPISBAIJ,
1559: MatGetRowUpperTriangular_MPISBAIJ,
1560: MatRestoreRowUpperTriangular_MPISBAIJ
1561: };
1567: PetscErrorCode MatGetDiagonalBlock_MPISBAIJ(Mat A,PetscTruth *iscopy,MatReuse reuse,Mat *a)
1568: {
1570: *a = ((Mat_MPISBAIJ *)A->data)->A;
1571: *iscopy = PETSC_FALSE;
1572: return(0);
1573: }
1579: PetscErrorCode MatMPISBAIJSetPreallocation_MPISBAIJ(Mat B,PetscInt bs,PetscInt d_nz,PetscInt *d_nnz,PetscInt o_nz,PetscInt *o_nnz)
1580: {
1581: Mat_MPISBAIJ *b;
1583: PetscInt i,mbs,Mbs;
1586: PetscOptionsBegin(((PetscObject)B)->comm,((PetscObject)B)->prefix,"Options for MPISBAIJ matrix","Mat");
1587: PetscOptionsInt("-mat_block_size","Set the blocksize used to store the matrix","MatMPIBAIJSetPreallocation",bs,&bs,PETSC_NULL);
1588: PetscOptionsEnd();
1590: if (bs < 1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Invalid block size specified, must be positive");
1591: if (d_nz == PETSC_DECIDE || d_nz == PETSC_DEFAULT) d_nz = 3;
1592: if (o_nz == PETSC_DECIDE || o_nz == PETSC_DEFAULT) o_nz = 1;
1593: if (d_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"d_nz cannot be less than 0: value %D",d_nz);
1594: if (o_nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"o_nz cannot be less than 0: value %D",o_nz);
1596: B->rmap.bs = B->cmap.bs = bs;
1597: PetscMapSetUp(&B->rmap);
1598: PetscMapSetUp(&B->cmap);
1600: if (d_nnz) {
1601: for (i=0; i<B->rmap.n/bs; i++) {
1602: 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]);
1603: }
1604: }
1605: if (o_nnz) {
1606: for (i=0; i<B->rmap.n/bs; i++) {
1607: 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]);
1608: }
1609: }
1610: B->preallocated = PETSC_TRUE;
1612: b = (Mat_MPISBAIJ*)B->data;
1613: mbs = B->rmap.n/bs;
1614: Mbs = B->rmap.N/bs;
1615: if (mbs*bs != B->rmap.n) {
1616: SETERRQ2(PETSC_ERR_ARG_SIZ,"No of local rows %D must be divisible by blocksize %D",B->rmap.N,bs);
1617: }
1619: B->rmap.bs = bs;
1620: b->bs2 = bs*bs;
1621: b->mbs = mbs;
1622: b->nbs = mbs;
1623: b->Mbs = Mbs;
1624: b->Nbs = Mbs;
1626: for (i=0; i<=b->size; i++) {
1627: b->rangebs[i] = B->rmap.range[i]/bs;
1628: }
1629: b->rstartbs = B->rmap.rstart/bs;
1630: b->rendbs = B->rmap.rend/bs;
1631:
1632: b->cstartbs = B->cmap.rstart/bs;
1633: b->cendbs = B->cmap.rend/bs;
1634:
1635: MatCreate(PETSC_COMM_SELF,&b->A);
1636: MatSetSizes(b->A,B->rmap.n,B->cmap.n,B->rmap.n,B->cmap.n);
1637: MatSetType(b->A,MATSEQSBAIJ);
1638: MatSeqSBAIJSetPreallocation(b->A,bs,d_nz,d_nnz);
1639: PetscLogObjectParent(B,b->A);
1641: MatCreate(PETSC_COMM_SELF,&b->B);
1642: MatSetSizes(b->B,B->rmap.n,B->cmap.N,B->rmap.n,B->cmap.N);
1643: MatSetType(b->B,MATSEQBAIJ);
1644: MatSeqBAIJSetPreallocation(b->B,bs,o_nz,o_nnz);
1645: PetscLogObjectParent(B,b->B);
1647: /* build cache for off array entries formed */
1648: MatStashCreate_Private(((PetscObject)B)->comm,bs,&B->bstash);
1650: return(0);
1651: }
1654: /*MC
1655: MATMPISBAIJ - MATMPISBAIJ = "mpisbaij" - A matrix type to be used for distributed symmetric sparse block matrices,
1656: based on block compressed sparse row format. Only the upper triangular portion of the matrix is stored.
1658: Options Database Keys:
1659: . -mat_type mpisbaij - sets the matrix type to "mpisbaij" during a call to MatSetFromOptions()
1661: Level: beginner
1663: .seealso: MatCreateMPISBAIJ
1664: M*/
1669: PetscErrorCode MatCreate_MPISBAIJ(Mat B)
1670: {
1671: Mat_MPISBAIJ *b;
1673: PetscTruth flg;
1677: PetscNewLog(B,Mat_MPISBAIJ,&b);
1678: B->data = (void*)b;
1679: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
1681: B->ops->destroy = MatDestroy_MPISBAIJ;
1682: B->ops->view = MatView_MPISBAIJ;
1683: B->mapping = 0;
1684: B->factor = 0;
1685: B->assembled = PETSC_FALSE;
1687: B->insertmode = NOT_SET_VALUES;
1688: MPI_Comm_rank(((PetscObject)B)->comm,&b->rank);
1689: MPI_Comm_size(((PetscObject)B)->comm,&b->size);
1691: /* build local table of row and column ownerships */
1692: PetscMalloc((b->size+2)*sizeof(PetscInt),&b->rangebs);
1694: /* build cache for off array entries formed */
1695: MatStashCreate_Private(((PetscObject)B)->comm,1,&B->stash);
1696: b->donotstash = PETSC_FALSE;
1697: b->colmap = PETSC_NULL;
1698: b->garray = PETSC_NULL;
1699: b->roworiented = PETSC_TRUE;
1701: #if defined(PETSC_USE_MAT_SINGLE)
1702: /* stuff for MatSetValues_XXX in single precision */
1703: b->setvalueslen = 0;
1704: b->setvaluescopy = PETSC_NULL;
1705: #endif
1707: /* stuff used in block assembly */
1708: b->barray = 0;
1710: /* stuff used for matrix vector multiply */
1711: b->lvec = 0;
1712: b->Mvctx = 0;
1713: b->slvec0 = 0;
1714: b->slvec0b = 0;
1715: b->slvec1 = 0;
1716: b->slvec1a = 0;
1717: b->slvec1b = 0;
1718: b->sMvctx = 0;
1720: /* stuff for MatGetRow() */
1721: b->rowindices = 0;
1722: b->rowvalues = 0;
1723: b->getrowactive = PETSC_FALSE;
1725: /* hash table stuff */
1726: b->ht = 0;
1727: b->hd = 0;
1728: b->ht_size = 0;
1729: b->ht_flag = PETSC_FALSE;
1730: b->ht_fact = 0;
1731: b->ht_total_ct = 0;
1732: b->ht_insert_ct = 0;
1734: b->in_loc = 0;
1735: b->v_loc = 0;
1736: b->n_loc = 0;
1737: PetscOptionsBegin(((PetscObject)B)->comm,PETSC_NULL,"Options for loading MPISBAIJ matrix 1","Mat");
1738: PetscOptionsTruth("-mat_use_hash_table","Use hash table to save memory in constructing matrix","MatSetOption",PETSC_FALSE,&flg,PETSC_NULL);
1739: if (flg) {
1740: PetscReal fact = 1.39;
1741: MatSetOption(B,MAT_USE_HASH_TABLE,PETSC_TRUE);
1742: PetscOptionsReal("-mat_use_hash_table","Use hash table factor","MatMPIBAIJSetHashTableFactor",fact,&fact,PETSC_NULL);
1743: if (fact <= 1.0) fact = 1.39;
1744: MatMPIBAIJSetHashTableFactor(B,fact);
1745: PetscInfo1(B,"Hash table Factor used %5.2f\n",fact);
1746: }
1747: PetscOptionsEnd();
1749: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C",
1750: "MatStoreValues_MPISBAIJ",
1751: MatStoreValues_MPISBAIJ);
1752: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C",
1753: "MatRetrieveValues_MPISBAIJ",
1754: MatRetrieveValues_MPISBAIJ);
1755: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatGetDiagonalBlock_C",
1756: "MatGetDiagonalBlock_MPISBAIJ",
1757: MatGetDiagonalBlock_MPISBAIJ);
1758: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatMPISBAIJSetPreallocation_C",
1759: "MatMPISBAIJSetPreallocation_MPISBAIJ",
1760: MatMPISBAIJSetPreallocation_MPISBAIJ);
1761: B->symmetric = PETSC_TRUE;
1762: B->structurally_symmetric = PETSC_TRUE;
1763: B->symmetric_set = PETSC_TRUE;
1764: B->structurally_symmetric_set = PETSC_TRUE;
1765: PetscObjectChangeTypeName((PetscObject)B,MATMPISBAIJ);
1766: return(0);
1767: }
1770: /*MC
1771: MATSBAIJ - MATSBAIJ = "sbaij" - A matrix type to be used for symmetric block sparse matrices.
1773: This matrix type is identical to MATSEQSBAIJ when constructed with a single process communicator,
1774: and MATMPISBAIJ otherwise.
1776: Options Database Keys:
1777: . -mat_type sbaij - sets the matrix type to "sbaij" during a call to MatSetFromOptions()
1779: Level: beginner
1781: .seealso: MatCreateMPISBAIJ,MATSEQSBAIJ,MATMPISBAIJ
1782: M*/
1787: PetscErrorCode MatCreate_SBAIJ(Mat A)
1788: {
1790: PetscMPIInt size;
1793: MPI_Comm_size(((PetscObject)A)->comm,&size);
1794: if (size == 1) {
1795: MatSetType(A,MATSEQSBAIJ);
1796: } else {
1797: MatSetType(A,MATMPISBAIJ);
1798: }
1799: return(0);
1800: }
1805: /*@C
1806: MatMPISBAIJSetPreallocation - For good matrix assembly performance
1807: the user should preallocate the matrix storage by setting the parameters
1808: d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately,
1809: performance can be increased by more than a factor of 50.
1811: Collective on Mat
1813: Input Parameters:
1814: + A - the matrix
1815: . bs - size of blockk
1816: . d_nz - number of block nonzeros per block row in diagonal portion of local
1817: submatrix (same for all local rows)
1818: . d_nnz - array containing the number of block nonzeros in the various block rows
1819: in the upper triangular and diagonal part of the in diagonal portion of the local
1820: (possibly different for each block row) or PETSC_NULL. You must leave room
1821: for the diagonal entry even if it is zero.
1822: . o_nz - number of block nonzeros per block row in the off-diagonal portion of local
1823: submatrix (same for all local rows).
1824: - o_nnz - array containing the number of nonzeros in the various block rows of the
1825: off-diagonal portion of the local submatrix (possibly different for
1826: each block row) or PETSC_NULL.
1829: Options Database Keys:
1830: . -mat_no_unroll - uses code that does not unroll the loops in the
1831: block calculations (much slower)
1832: . -mat_block_size - size of the blocks to use
1834: Notes:
1836: If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one processor
1837: than it must be used on all processors that share the object for that argument.
1839: If the *_nnz parameter is given then the *_nz parameter is ignored
1841: Storage Information:
1842: For a square global matrix we define each processor's diagonal portion
1843: to be its local rows and the corresponding columns (a square submatrix);
1844: each processor's off-diagonal portion encompasses the remainder of the
1845: local matrix (a rectangular submatrix).
1847: The user can specify preallocated storage for the diagonal part of
1848: the local submatrix with either d_nz or d_nnz (not both). Set
1849: d_nz=PETSC_DEFAULT and d_nnz=PETSC_NULL for PETSc to control dynamic
1850: memory allocation. Likewise, specify preallocated storage for the
1851: off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
1853: You can call MatGetInfo() to get information on how effective the preallocation was;
1854: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
1855: You can also run with the option -info and look for messages with the string
1856: malloc in them to see if additional memory allocation was needed.
1858: Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
1859: the figure below we depict these three local rows and all columns (0-11).
1861: .vb
1862: 0 1 2 3 4 5 6 7 8 9 10 11
1863: -------------------
1864: row 3 | o o o d d d o o o o o o
1865: row 4 | o o o d d d o o o o o o
1866: row 5 | o o o d d d o o o o o o
1867: -------------------
1868: .ve
1869:
1870: Thus, any entries in the d locations are stored in the d (diagonal)
1871: submatrix, and any entries in the o locations are stored in the
1872: o (off-diagonal) submatrix. Note that the d matrix is stored in
1873: MatSeqSBAIJ format and the o submatrix in MATSEQBAIJ format.
1875: Now d_nz should indicate the number of block nonzeros per row in the upper triangular
1876: plus the diagonal part of the d matrix,
1877: and o_nz should indicate the number of block nonzeros per row in the o matrix.
1878: In general, for PDE problems in which most nonzeros are near the diagonal,
1879: one expects d_nz >> o_nz. For large problems you MUST preallocate memory
1880: or you will get TERRIBLE performance; see the users' manual chapter on
1881: matrices.
1883: Level: intermediate
1885: .keywords: matrix, block, aij, compressed row, sparse, parallel
1887: .seealso: MatCreate(), MatCreateSeqSBAIJ(), MatSetValues(), MatCreateMPIBAIJ()
1888: @*/
1889: PetscErrorCode MatMPISBAIJSetPreallocation(Mat B,PetscInt bs,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[])
1890: {
1891: PetscErrorCode ierr,(*f)(Mat,PetscInt,PetscInt,const PetscInt[],PetscInt,const PetscInt[]);
1894: PetscObjectQueryFunction((PetscObject)B,"MatMPISBAIJSetPreallocation_C",(void (**)(void))&f);
1895: if (f) {
1896: (*f)(B,bs,d_nz,d_nnz,o_nz,o_nnz);
1897: }
1898: return(0);
1899: }
1903: /*@C
1904: MatCreateMPISBAIJ - Creates a sparse parallel matrix in symmetric block AIJ format
1905: (block compressed row). For good matrix assembly performance
1906: the user should preallocate the matrix storage by setting the parameters
1907: d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately,
1908: performance can be increased by more than a factor of 50.
1910: Collective on MPI_Comm
1912: Input Parameters:
1913: + comm - MPI communicator
1914: . bs - size of blockk
1915: . m - number of local rows (or PETSC_DECIDE to have calculated if M is given)
1916: This value should be the same as the local size used in creating the
1917: y vector for the matrix-vector product y = Ax.
1918: . n - number of local columns (or PETSC_DECIDE to have calculated if N is given)
1919: This value should be the same as the local size used in creating the
1920: x vector for the matrix-vector product y = Ax.
1921: . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given)
1922: . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given)
1923: . d_nz - number of block nonzeros per block row in diagonal portion of local
1924: submatrix (same for all local rows)
1925: . d_nnz - array containing the number of block nonzeros in the various block rows
1926: in the upper triangular portion of the in diagonal portion of the local
1927: (possibly different for each block block row) or PETSC_NULL.
1928: You must leave room for the diagonal entry even if it is zero.
1929: . o_nz - number of block nonzeros per block row in the off-diagonal portion of local
1930: submatrix (same for all local rows).
1931: - o_nnz - array containing the number of nonzeros in the various block rows of the
1932: off-diagonal portion of the local submatrix (possibly different for
1933: each block row) or PETSC_NULL.
1935: Output Parameter:
1936: . A - the matrix
1938: Options Database Keys:
1939: . -mat_no_unroll - uses code that does not unroll the loops in the
1940: block calculations (much slower)
1941: . -mat_block_size - size of the blocks to use
1942: . -mat_mpi - use the parallel matrix data structures even on one processor
1943: (defaults to using SeqBAIJ format on one processor)
1945: It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
1946: MatXXXXSetPreallocation() paradgm instead of this routine directly. This is definitely
1947: true if you plan to use the external direct solvers such as SuperLU, MUMPS or Spooles.
1948: [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
1950: Notes:
1951: The number of rows and columns must be divisible by blocksize.
1952: This matrix type does not support complex Hermitian operation.
1954: The user MUST specify either the local or global matrix dimensions
1955: (possibly both).
1957: If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one processor
1958: than it must be used on all processors that share the object for that argument.
1960: If the *_nnz parameter is given then the *_nz parameter is ignored
1962: Storage Information:
1963: For a square global matrix we define each processor's diagonal portion
1964: to be its local rows and the corresponding columns (a square submatrix);
1965: each processor's off-diagonal portion encompasses the remainder of the
1966: local matrix (a rectangular submatrix).
1968: The user can specify preallocated storage for the diagonal part of
1969: the local submatrix with either d_nz or d_nnz (not both). Set
1970: d_nz=PETSC_DEFAULT and d_nnz=PETSC_NULL for PETSc to control dynamic
1971: memory allocation. Likewise, specify preallocated storage for the
1972: off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
1974: Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
1975: the figure below we depict these three local rows and all columns (0-11).
1977: .vb
1978: 0 1 2 3 4 5 6 7 8 9 10 11
1979: -------------------
1980: row 3 | o o o d d d o o o o o o
1981: row 4 | o o o d d d o o o o o o
1982: row 5 | o o o d d d o o o o o o
1983: -------------------
1984: .ve
1985:
1986: Thus, any entries in the d locations are stored in the d (diagonal)
1987: submatrix, and any entries in the o locations are stored in the
1988: o (off-diagonal) submatrix. Note that the d matrix is stored in
1989: MatSeqSBAIJ format and the o submatrix in MATSEQBAIJ format.
1991: Now d_nz should indicate the number of block nonzeros per row in the upper triangular
1992: plus the diagonal part of the d matrix,
1993: and o_nz should indicate the number of block nonzeros per row in the o matrix.
1994: In general, for PDE problems in which most nonzeros are near the diagonal,
1995: one expects d_nz >> o_nz. For large problems you MUST preallocate memory
1996: or you will get TERRIBLE performance; see the users' manual chapter on
1997: matrices.
1999: Level: intermediate
2001: .keywords: matrix, block, aij, compressed row, sparse, parallel
2003: .seealso: MatCreate(), MatCreateSeqSBAIJ(), MatSetValues(), MatCreateMPIBAIJ()
2004: @*/
2006: PetscErrorCode MatCreateMPISBAIJ(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt M,PetscInt N,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[],Mat *A)
2007: {
2009: PetscMPIInt size;
2012: MatCreate(comm,A);
2013: MatSetSizes(*A,m,n,M,N);
2014: MPI_Comm_size(comm,&size);
2015: if (size > 1) {
2016: MatSetType(*A,MATMPISBAIJ);
2017: MatMPISBAIJSetPreallocation(*A,bs,d_nz,d_nnz,o_nz,o_nnz);
2018: } else {
2019: MatSetType(*A,MATSEQSBAIJ);
2020: MatSeqSBAIJSetPreallocation(*A,bs,d_nz,d_nnz);
2021: }
2022: return(0);
2023: }
2028: static PetscErrorCode MatDuplicate_MPISBAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat)
2029: {
2030: Mat mat;
2031: Mat_MPISBAIJ *a,*oldmat = (Mat_MPISBAIJ*)matin->data;
2033: PetscInt len=0,nt,bs=matin->rmap.bs,mbs=oldmat->mbs;
2034: PetscScalar *array;
2037: *newmat = 0;
2038: MatCreate(((PetscObject)matin)->comm,&mat);
2039: MatSetSizes(mat,matin->rmap.n,matin->cmap.n,matin->rmap.N,matin->cmap.N);
2040: MatSetType(mat,((PetscObject)matin)->type_name);
2041: PetscMemcpy(mat->ops,matin->ops,sizeof(struct _MatOps));
2042: PetscMapCopy(((PetscObject)matin)->comm,&matin->rmap,&mat->rmap);
2043: PetscMapCopy(((PetscObject)matin)->comm,&matin->cmap,&mat->cmap);
2044:
2045: mat->factor = matin->factor;
2046: mat->preallocated = PETSC_TRUE;
2047: mat->assembled = PETSC_TRUE;
2048: mat->insertmode = NOT_SET_VALUES;
2050: a = (Mat_MPISBAIJ*)mat->data;
2051: a->bs2 = oldmat->bs2;
2052: a->mbs = oldmat->mbs;
2053: a->nbs = oldmat->nbs;
2054: a->Mbs = oldmat->Mbs;
2055: a->Nbs = oldmat->Nbs;
2058: a->size = oldmat->size;
2059: a->rank = oldmat->rank;
2060: a->donotstash = oldmat->donotstash;
2061: a->roworiented = oldmat->roworiented;
2062: a->rowindices = 0;
2063: a->rowvalues = 0;
2064: a->getrowactive = PETSC_FALSE;
2065: a->barray = 0;
2066: a->rstartbs = oldmat->rstartbs;
2067: a->rendbs = oldmat->rendbs;
2068: a->cstartbs = oldmat->cstartbs;
2069: a->cendbs = oldmat->cendbs;
2071: /* hash table stuff */
2072: a->ht = 0;
2073: a->hd = 0;
2074: a->ht_size = 0;
2075: a->ht_flag = oldmat->ht_flag;
2076: a->ht_fact = oldmat->ht_fact;
2077: a->ht_total_ct = 0;
2078: a->ht_insert_ct = 0;
2079:
2080: PetscMemcpy(a->rangebs,oldmat->rangebs,(a->size+2)*sizeof(PetscInt));
2081: MatStashCreate_Private(((PetscObject)matin)->comm,1,&mat->stash);
2082: MatStashCreate_Private(((PetscObject)matin)->comm,matin->rmap.bs,&mat->bstash);
2083: if (oldmat->colmap) {
2084: #if defined (PETSC_USE_CTABLE)
2085: PetscTableCreateCopy(oldmat->colmap,&a->colmap);
2086: #else
2087: PetscMalloc((a->Nbs)*sizeof(PetscInt),&a->colmap);
2088: PetscLogObjectMemory(mat,(a->Nbs)*sizeof(PetscInt));
2089: PetscMemcpy(a->colmap,oldmat->colmap,(a->Nbs)*sizeof(PetscInt));
2090: #endif
2091: } else a->colmap = 0;
2093: if (oldmat->garray && (len = ((Mat_SeqBAIJ*)(oldmat->B->data))->nbs)) {
2094: PetscMalloc(len*sizeof(PetscInt),&a->garray);
2095: PetscLogObjectMemory(mat,len*sizeof(PetscInt));
2096: PetscMemcpy(a->garray,oldmat->garray,len*sizeof(PetscInt));
2097: } else a->garray = 0;
2098:
2099: VecDuplicate(oldmat->lvec,&a->lvec);
2100: PetscLogObjectParent(mat,a->lvec);
2101: VecScatterCopy(oldmat->Mvctx,&a->Mvctx);
2102: PetscLogObjectParent(mat,a->Mvctx);
2104: VecDuplicate(oldmat->slvec0,&a->slvec0);
2105: PetscLogObjectParent(mat,a->slvec0);
2106: VecDuplicate(oldmat->slvec1,&a->slvec1);
2107: PetscLogObjectParent(mat,a->slvec1);
2109: VecGetLocalSize(a->slvec1,&nt);
2110: VecGetArray(a->slvec1,&array);
2111: VecCreateSeqWithArray(PETSC_COMM_SELF,bs*mbs,array,&a->slvec1a);
2112: VecCreateSeqWithArray(PETSC_COMM_SELF,nt-bs*mbs,array+bs*mbs,&a->slvec1b);
2113: VecRestoreArray(a->slvec1,&array);
2114: VecGetArray(a->slvec0,&array);
2115: VecCreateSeqWithArray(PETSC_COMM_SELF,nt-bs*mbs,array+bs*mbs,&a->slvec0b);
2116: VecRestoreArray(a->slvec0,&array);
2117: PetscLogObjectParent(mat,a->slvec0);
2118: PetscLogObjectParent(mat,a->slvec1);
2119: PetscLogObjectParent(mat,a->slvec0b);
2120: PetscLogObjectParent(mat,a->slvec1a);
2121: PetscLogObjectParent(mat,a->slvec1b);
2123: /* VecScatterCopy(oldmat->sMvctx,&a->sMvctx); - not written yet, replaced by the lazy trick: */
2124: PetscObjectReference((PetscObject)oldmat->sMvctx);
2125: a->sMvctx = oldmat->sMvctx;
2126: PetscLogObjectParent(mat,a->sMvctx);
2128: MatDuplicate(oldmat->A,cpvalues,&a->A);
2129: PetscLogObjectParent(mat,a->A);
2130: MatDuplicate(oldmat->B,cpvalues,&a->B);
2131: PetscLogObjectParent(mat,a->B);
2132: PetscFListDuplicate(((PetscObject)matin)->qlist,&((PetscObject)mat)->qlist);
2133: *newmat = mat;
2134: return(0);
2135: }
2137: #include petscsys.h
2141: PetscErrorCode MatLoad_MPISBAIJ(PetscViewer viewer, MatType type,Mat *newmat)
2142: {
2143: Mat A;
2145: PetscInt i,nz,j,rstart,rend;
2146: PetscScalar *vals,*buf;
2147: MPI_Comm comm = ((PetscObject)viewer)->comm;
2148: MPI_Status status;
2149: PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag,*sndcounts = 0,*browners,maxnz,*rowners,*locrowlens;
2150: PetscInt header[4],*rowlengths = 0,M,N,m,*cols;
2151: PetscInt *procsnz = 0,jj,*mycols,*ibuf;
2152: PetscInt bs=1,Mbs,mbs,extra_rows;
2153: PetscInt *dlens,*odlens,*mask,*masked1,*masked2,rowcount,odcount;
2154: PetscInt dcount,kmax,k,nzcount,tmp;
2155: int fd;
2156:
2158: PetscOptionsBegin(comm,PETSC_NULL,"Options for loading MPISBAIJ matrix 2","Mat");
2159: PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,PETSC_NULL);
2160: PetscOptionsEnd();
2162: MPI_Comm_size(comm,&size);
2163: MPI_Comm_rank(comm,&rank);
2164: if (!rank) {
2165: PetscViewerBinaryGetDescriptor(viewer,&fd);
2166: PetscBinaryRead(fd,(char *)header,4,PETSC_INT);
2167: if (header[0] != MAT_FILE_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not matrix object");
2168: if (header[3] < 0) {
2169: SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format, cannot load as MPISBAIJ");
2170: }
2171: }
2173: MPI_Bcast(header+1,3,MPIU_INT,0,comm);
2174: M = header[1]; N = header[2];
2176: if (M != N) SETERRQ(PETSC_ERR_SUP,"Can only do square matrices");
2178: /*
2179: This code adds extra rows to make sure the number of rows is
2180: divisible by the blocksize
2181: */
2182: Mbs = M/bs;
2183: extra_rows = bs - M + bs*(Mbs);
2184: if (extra_rows == bs) extra_rows = 0;
2185: else Mbs++;
2186: if (extra_rows &&!rank) {
2187: PetscInfo(viewer,"Padding loaded matrix to match blocksize\n");
2188: }
2190: /* determine ownership of all rows */
2191: mbs = Mbs/size + ((Mbs % size) > rank);
2192: m = mbs*bs;
2193: PetscMalloc(2*(size+2)*sizeof(PetscMPIInt),&rowners);
2194: browners = rowners + size + 1;
2195: MPI_Allgather(&mbs,1,MPI_INT,rowners+1,1,MPI_INT,comm);
2196: rowners[0] = 0;
2197: for (i=2; i<=size; i++) rowners[i] += rowners[i-1];
2198: for (i=0; i<=size; i++) browners[i] = rowners[i]*bs;
2199: rstart = rowners[rank];
2200: rend = rowners[rank+1];
2201:
2202: /* distribute row lengths to all processors */
2203: PetscMalloc((rend-rstart)*bs*sizeof(PetscMPIInt),&locrowlens);
2204: if (!rank) {
2205: PetscMalloc((M+extra_rows)*sizeof(PetscInt),&rowlengths);
2206: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
2207: for (i=0; i<extra_rows; i++) rowlengths[M+i] = 1;
2208: PetscMalloc(size*sizeof(PetscMPIInt),&sndcounts);
2209: for (i=0; i<size; i++) sndcounts[i] = browners[i+1] - browners[i];
2210: MPI_Scatterv(rowlengths,sndcounts,browners,MPIU_INT,locrowlens,(rend-rstart)*bs,MPIU_INT,0,comm);
2211: PetscFree(sndcounts);
2212: } else {
2213: MPI_Scatterv(0,0,0,MPIU_INT,locrowlens,(rend-rstart)*bs,MPIU_INT,0,comm);
2214: }
2215:
2216: if (!rank) { /* procs[0] */
2217: /* calculate the number of nonzeros on each processor */
2218: PetscMalloc(size*sizeof(PetscInt),&procsnz);
2219: PetscMemzero(procsnz,size*sizeof(PetscInt));
2220: for (i=0; i<size; i++) {
2221: for (j=rowners[i]*bs; j< rowners[i+1]*bs; j++) {
2222: procsnz[i] += rowlengths[j];
2223: }
2224: }
2225: PetscFree(rowlengths);
2226:
2227: /* determine max buffer needed and allocate it */
2228: maxnz = 0;
2229: for (i=0; i<size; i++) {
2230: maxnz = PetscMax(maxnz,procsnz[i]);
2231: }
2232: PetscMalloc(maxnz*sizeof(PetscInt),&cols);
2234: /* read in my part of the matrix column indices */
2235: nz = procsnz[0];
2236: PetscMalloc(nz*sizeof(PetscInt),&ibuf);
2237: mycols = ibuf;
2238: if (size == 1) nz -= extra_rows;
2239: PetscBinaryRead(fd,mycols,nz,PETSC_INT);
2240: if (size == 1) for (i=0; i< extra_rows; i++) { mycols[nz+i] = M+i; }
2242: /* read in every ones (except the last) and ship off */
2243: for (i=1; i<size-1; i++) {
2244: nz = procsnz[i];
2245: PetscBinaryRead(fd,cols,nz,PETSC_INT);
2246: MPI_Send(cols,nz,MPIU_INT,i,tag,comm);
2247: }
2248: /* read in the stuff for the last proc */
2249: if (size != 1) {
2250: nz = procsnz[size-1] - extra_rows; /* the extra rows are not on the disk */
2251: PetscBinaryRead(fd,cols,nz,PETSC_INT);
2252: for (i=0; i<extra_rows; i++) cols[nz+i] = M+i;
2253: MPI_Send(cols,nz+extra_rows,MPIU_INT,size-1,tag,comm);
2254: }
2255: PetscFree(cols);
2256: } else { /* procs[i], i>0 */
2257: /* determine buffer space needed for message */
2258: nz = 0;
2259: for (i=0; i<m; i++) {
2260: nz += locrowlens[i];
2261: }
2262: PetscMalloc(nz*sizeof(PetscInt),&ibuf);
2263: mycols = ibuf;
2264: /* receive message of column indices*/
2265: MPI_Recv(mycols,nz,MPIU_INT,0,tag,comm,&status);
2266: MPI_Get_count(&status,MPIU_INT,&maxnz);
2267: if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
2268: }
2270: /* loop over local rows, determining number of off diagonal entries */
2271: PetscMalloc(2*(rend-rstart+1)*sizeof(PetscInt),&dlens);
2272: odlens = dlens + (rend-rstart);
2273: PetscMalloc(3*Mbs*sizeof(PetscInt),&mask);
2274: PetscMemzero(mask,3*Mbs*sizeof(PetscInt));
2275: masked1 = mask + Mbs;
2276: masked2 = masked1 + Mbs;
2277: rowcount = 0; nzcount = 0;
2278: for (i=0; i<mbs; i++) {
2279: dcount = 0;
2280: odcount = 0;
2281: for (j=0; j<bs; j++) {
2282: kmax = locrowlens[rowcount];
2283: for (k=0; k<kmax; k++) {
2284: tmp = mycols[nzcount++]/bs; /* block col. index */
2285: if (!mask[tmp]) {
2286: mask[tmp] = 1;
2287: if (tmp < rstart || tmp >= rend) masked2[odcount++] = tmp; /* entry in off-diag portion */
2288: else masked1[dcount++] = tmp; /* entry in diag portion */
2289: }
2290: }
2291: rowcount++;
2292: }
2293:
2294: dlens[i] = dcount; /* d_nzz[i] */
2295: odlens[i] = odcount; /* o_nzz[i] */
2297: /* zero out the mask elements we set */
2298: for (j=0; j<dcount; j++) mask[masked1[j]] = 0;
2299: for (j=0; j<odcount; j++) mask[masked2[j]] = 0;
2300: }
2301:
2302: /* create our matrix */
2303: MatCreate(comm,&A);
2304: MatSetSizes(A,m,m,PETSC_DETERMINE,PETSC_DETERMINE);
2305: MatSetType(A,type);
2306: MatMPISBAIJSetPreallocation(A,bs,0,dlens,0,odlens);
2307:
2308: if (!rank) {
2309: PetscMalloc(maxnz*sizeof(PetscScalar),&buf);
2310: /* read in my part of the matrix numerical values */
2311: nz = procsnz[0];
2312: vals = buf;
2313: mycols = ibuf;
2314: if (size == 1) nz -= extra_rows;
2315: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
2316: if (size == 1) for (i=0; i< extra_rows; i++) { vals[nz+i] = 1.0; }
2318: /* insert into matrix */
2319: jj = rstart*bs;
2320: for (i=0; i<m; i++) {
2321: MatSetValues(A,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);
2322: mycols += locrowlens[i];
2323: vals += locrowlens[i];
2324: jj++;
2325: }
2327: /* read in other processors (except the last one) and ship out */
2328: for (i=1; i<size-1; i++) {
2329: nz = procsnz[i];
2330: vals = buf;
2331: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
2332: MPI_Send(vals,nz,MPIU_SCALAR,i,((PetscObject)A)->tag,comm);
2333: }
2334: /* the last proc */
2335: if (size != 1){
2336: nz = procsnz[i] - extra_rows;
2337: vals = buf;
2338: PetscBinaryRead(fd,vals,nz,PETSC_SCALAR);
2339: for (i=0; i<extra_rows; i++) vals[nz+i] = 1.0;
2340: MPI_Send(vals,nz+extra_rows,MPIU_SCALAR,size-1,((PetscObject)A)->tag,comm);
2341: }
2342: PetscFree(procsnz);
2344: } else {
2345: /* receive numeric values */
2346: PetscMalloc(nz*sizeof(PetscScalar),&buf);
2348: /* receive message of values*/
2349: vals = buf;
2350: mycols = ibuf;
2351: MPI_Recv(vals,nz,MPIU_SCALAR,0,((PetscObject)A)->tag,comm,&status);
2352: MPI_Get_count(&status,MPIU_SCALAR,&maxnz);
2353: if (maxnz != nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
2355: /* insert into matrix */
2356: jj = rstart*bs;
2357: for (i=0; i<m; i++) {
2358: MatSetValues_MPISBAIJ(A,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);
2359: mycols += locrowlens[i];
2360: vals += locrowlens[i];
2361: jj++;
2362: }
2363: }
2365: PetscFree(locrowlens);
2366: PetscFree(buf);
2367: PetscFree(ibuf);
2368: PetscFree(rowners);
2369: PetscFree(dlens);
2370: PetscFree(mask);
2371: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
2372: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
2373: *newmat = A;
2374: return(0);
2375: }
2379: /*XXXXX@
2380: MatMPISBAIJSetHashTableFactor - Sets the factor required to compute the size of the HashTable.
2382: Input Parameters:
2383: . mat - the matrix
2384: . fact - factor
2386: Collective on Mat
2388: Level: advanced
2390: Notes:
2391: This can also be set by the command line option: -mat_use_hash_table fact
2393: .keywords: matrix, hashtable, factor, HT
2395: .seealso: MatSetOption()
2396: @XXXXX*/
2401: PetscErrorCode MatGetRowMaxAbs_MPISBAIJ(Mat A,Vec v,PetscInt idx[])
2402: {
2403: Mat_MPISBAIJ *a = (Mat_MPISBAIJ*)A->data;
2404: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)(a->B)->data;
2405: PetscReal atmp;
2406: PetscReal *work,*svalues,*rvalues;
2408: PetscInt i,bs,mbs,*bi,*bj,brow,j,ncols,krow,kcol,col,row,Mbs,bcol;
2409: PetscMPIInt rank,size;
2410: PetscInt *rowners_bs,dest,count,source;
2411: PetscScalar *va;
2412: MatScalar *ba;
2413: MPI_Status stat;
2416: if (idx) SETERRQ(PETSC_ERR_SUP,"Send email to petsc-maint@mcs.anl.gov");
2417: MatGetRowMaxAbs(a->A,v,PETSC_NULL);
2418: VecGetArray(v,&va);
2420: MPI_Comm_size(((PetscObject)A)->comm,&size);
2421: MPI_Comm_rank(((PetscObject)A)->comm,&rank);
2423: bs = A->rmap.bs;
2424: mbs = a->mbs;
2425: Mbs = a->Mbs;
2426: ba = b->a;
2427: bi = b->i;
2428: bj = b->j;
2430: /* find ownerships */
2431: rowners_bs = A->rmap.range;
2433: /* each proc creates an array to be distributed */
2434: PetscMalloc(bs*Mbs*sizeof(PetscReal),&work);
2435: PetscMemzero(work,bs*Mbs*sizeof(PetscReal));
2437: /* row_max for B */
2438: if (rank != size-1){
2439: for (i=0; i<mbs; i++) {
2440: ncols = bi[1] - bi[0]; bi++;
2441: brow = bs*i;
2442: for (j=0; j<ncols; j++){
2443: bcol = bs*(*bj);
2444: for (kcol=0; kcol<bs; kcol++){
2445: col = bcol + kcol; /* local col index */
2446: col += rowners_bs[rank+1]; /* global col index */
2447: for (krow=0; krow<bs; krow++){
2448: atmp = PetscAbsScalar(*ba); ba++;
2449: row = brow + krow; /* local row index */
2450: if (PetscRealPart(va[row]) < atmp) va[row] = atmp;
2451: if (work[col] < atmp) work[col] = atmp;
2452: }
2453: }
2454: bj++;
2455: }
2456: }
2458: /* send values to its owners */
2459: for (dest=rank+1; dest<size; dest++){
2460: svalues = work + rowners_bs[dest];
2461: count = rowners_bs[dest+1]-rowners_bs[dest];
2462: MPI_Send(svalues,count,MPIU_REAL,dest,rank,((PetscObject)A)->comm);
2463: }
2464: }
2465:
2466: /* receive values */
2467: if (rank){
2468: rvalues = work;
2469: count = rowners_bs[rank+1]-rowners_bs[rank];
2470: for (source=0; source<rank; source++){
2471: MPI_Recv(rvalues,count,MPIU_REAL,MPI_ANY_SOURCE,MPI_ANY_TAG,((PetscObject)A)->comm,&stat);
2472: /* process values */
2473: for (i=0; i<count; i++){
2474: if (PetscRealPart(va[i]) < rvalues[i]) va[i] = rvalues[i];
2475: }
2476: }
2477: }
2479: VecRestoreArray(v,&va);
2480: PetscFree(work);
2481: return(0);
2482: }
2486: PetscErrorCode MatRelax_MPISBAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
2487: {
2488: Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)matin->data;
2490: PetscInt mbs=mat->mbs,bs=matin->rmap.bs;
2491: PetscScalar *x,*b,*ptr,zero=0.0;
2492: Vec bb1;
2493:
2495: if (its <= 0 || lits <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
2496: if (bs > 1)
2497: SETERRQ(PETSC_ERR_SUP,"SSOR for block size > 1 is not yet implemented");
2499: if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP){
2500: if ( flag & SOR_ZERO_INITIAL_GUESS ) {
2501: (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,lits,xx);
2502: its--;
2503: }
2505: VecDuplicate(bb,&bb1);
2506: while (its--){
2507:
2508: /* lower triangular part: slvec0b = - B^T*xx */
2509: (*mat->B->ops->multtranspose)(mat->B,xx,mat->slvec0b);
2510:
2511: /* copy xx into slvec0a */
2512: VecGetArray(mat->slvec0,&ptr);
2513: VecGetArray(xx,&x);
2514: PetscMemcpy(ptr,x,bs*mbs*sizeof(MatScalar));
2515: VecRestoreArray(mat->slvec0,&ptr);
2517: VecScale(mat->slvec0,-1.0);
2519: /* copy bb into slvec1a */
2520: VecGetArray(mat->slvec1,&ptr);
2521: VecGetArray(bb,&b);
2522: PetscMemcpy(ptr,b,bs*mbs*sizeof(MatScalar));
2523: VecRestoreArray(mat->slvec1,&ptr);
2525: /* set slvec1b = 0 */
2526: VecSet(mat->slvec1b,zero);
2528: VecScatterBegin(mat->sMvctx,mat->slvec0,mat->slvec1,ADD_VALUES,SCATTER_FORWARD);
2529: VecRestoreArray(xx,&x);
2530: VecRestoreArray(bb,&b);
2531: VecScatterEnd(mat->sMvctx,mat->slvec0,mat->slvec1,ADD_VALUES,SCATTER_FORWARD);
2533: /* upper triangular part: bb1 = bb1 - B*x */
2534: (*mat->B->ops->multadd)(mat->B,mat->slvec1b,mat->slvec1a,bb1);
2535:
2536: /* local diagonal sweep */
2537: (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,lits,xx);
2538: }
2539: VecDestroy(bb1);
2540: } else {
2541: SETERRQ(PETSC_ERR_SUP,"MatSORType is not supported for SBAIJ matrix format");
2542: }
2543: return(0);
2544: }
2548: PetscErrorCode MatRelax_MPISBAIJ_2comm(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
2549: {
2550: Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)matin->data;
2552: Vec lvec1,bb1;
2553:
2555: if (its <= 0 || lits <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
2556: if (matin->rmap.bs > 1)
2557: SETERRQ(PETSC_ERR_SUP,"SSOR for block size > 1 is not yet implemented");
2559: if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP){
2560: if ( flag & SOR_ZERO_INITIAL_GUESS ) {
2561: (*mat->A->ops->relax)(mat->A,bb,omega,flag,fshift,lits,lits,xx);
2562: its--;
2563: }
2565: VecDuplicate(mat->lvec,&lvec1);
2566: VecDuplicate(bb,&bb1);
2567: while (its--){
2568: VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2569:
2570: /* lower diagonal part: bb1 = bb - B^T*xx */
2571: (*mat->B->ops->multtranspose)(mat->B,xx,lvec1);
2572: VecScale(lvec1,-1.0);
2574: VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2575: VecCopy(bb,bb1);
2576: VecScatterBegin(mat->Mvctx,lvec1,bb1,ADD_VALUES,SCATTER_REVERSE);
2578: /* upper diagonal part: bb1 = bb1 - B*x */
2579: VecScale(mat->lvec,-1.0);
2580: (*mat->B->ops->multadd)(mat->B,mat->lvec,bb1,bb1);
2582: VecScatterEnd(mat->Mvctx,lvec1,bb1,ADD_VALUES,SCATTER_REVERSE);
2583:
2584: /* diagonal sweep */
2585: (*mat->A->ops->relax)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,lits,xx);
2586: }
2587: VecDestroy(lvec1);
2588: VecDestroy(bb1);
2589: } else {
2590: SETERRQ(PETSC_ERR_SUP,"MatSORType is not supported for SBAIJ matrix format");
2591: }
2592: return(0);
2593: }