Actual source code: mpibaij.c
petsc-3.12.5 2020-03-29
2: #include <../src/mat/impls/baij/mpi/mpibaij.h>
4: #include <petscblaslapack.h>
5: #include <petscsf.h>
7: #if defined(PETSC_HAVE_HYPRE)
8: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*);
9: #endif
11: PetscErrorCode MatGetRowMaxAbs_MPIBAIJ(Mat A,Vec v,PetscInt idx[])
12: {
13: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
15: PetscInt i,*idxb = 0;
16: PetscScalar *va,*vb;
17: Vec vtmp;
20: MatGetRowMaxAbs(a->A,v,idx);
21: VecGetArray(v,&va);
22: if (idx) {
23: for (i=0; i<A->rmap->n; i++) {
24: if (PetscAbsScalar(va[i])) idx[i] += A->cmap->rstart;
25: }
26: }
28: VecCreateSeq(PETSC_COMM_SELF,A->rmap->n,&vtmp);
29: if (idx) {PetscMalloc1(A->rmap->n,&idxb);}
30: MatGetRowMaxAbs(a->B,vtmp,idxb);
31: VecGetArray(vtmp,&vb);
33: for (i=0; i<A->rmap->n; i++) {
34: if (PetscAbsScalar(va[i]) < PetscAbsScalar(vb[i])) {
35: va[i] = vb[i];
36: if (idx) idx[i] = A->cmap->bs*a->garray[idxb[i]/A->cmap->bs] + (idxb[i] % A->cmap->bs);
37: }
38: }
40: VecRestoreArray(v,&va);
41: VecRestoreArray(vtmp,&vb);
42: PetscFree(idxb);
43: VecDestroy(&vtmp);
44: return(0);
45: }
47: PetscErrorCode MatStoreValues_MPIBAIJ(Mat mat)
48: {
49: Mat_MPIBAIJ *aij = (Mat_MPIBAIJ*)mat->data;
53: MatStoreValues(aij->A);
54: MatStoreValues(aij->B);
55: return(0);
56: }
58: PetscErrorCode MatRetrieveValues_MPIBAIJ(Mat mat)
59: {
60: Mat_MPIBAIJ *aij = (Mat_MPIBAIJ*)mat->data;
64: MatRetrieveValues(aij->A);
65: MatRetrieveValues(aij->B);
66: return(0);
67: }
69: /*
70: Local utility routine that creates a mapping from the global column
71: number to the local number in the off-diagonal part of the local
72: storage of the matrix. This is done in a non scalable way since the
73: length of colmap equals the global matrix length.
74: */
75: PetscErrorCode MatCreateColmap_MPIBAIJ_Private(Mat mat)
76: {
77: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
78: Mat_SeqBAIJ *B = (Mat_SeqBAIJ*)baij->B->data;
80: PetscInt nbs = B->nbs,i,bs=mat->rmap->bs;
83: #if defined(PETSC_USE_CTABLE)
84: PetscTableCreate(baij->nbs,baij->Nbs+1,&baij->colmap);
85: for (i=0; i<nbs; i++) {
86: PetscTableAdd(baij->colmap,baij->garray[i]+1,i*bs+1,INSERT_VALUES);
87: }
88: #else
89: PetscCalloc1(baij->Nbs+1,&baij->colmap);
90: PetscLogObjectMemory((PetscObject)mat,baij->Nbs*sizeof(PetscInt));
91: for (i=0; i<nbs; i++) baij->colmap[baij->garray[i]] = i*bs+1;
92: #endif
93: return(0);
94: }
96: #define MatSetValues_SeqBAIJ_A_Private(row,col,value,addv,orow,ocol) \
97: { \
98: brow = row/bs; \
99: rp = aj + ai[brow]; ap = aa + bs2*ai[brow]; \
100: rmax = aimax[brow]; nrow = ailen[brow]; \
101: bcol = col/bs; \
102: ridx = row % bs; cidx = col % bs; \
103: low = 0; high = nrow; \
104: while (high-low > 3) { \
105: t = (low+high)/2; \
106: if (rp[t] > bcol) high = t; \
107: else low = t; \
108: } \
109: for (_i=low; _i<high; _i++) { \
110: if (rp[_i] > bcol) break; \
111: if (rp[_i] == bcol) { \
112: bap = ap + bs2*_i + bs*cidx + ridx; \
113: if (addv == ADD_VALUES) *bap += value; \
114: else *bap = value; \
115: goto a_noinsert; \
116: } \
117: } \
118: if (a->nonew == 1) goto a_noinsert; \
119: if (a->nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at global row/column (%D, %D) into matrix", orow, ocol); \
120: MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,brow,bcol,rmax,aa,ai,aj,rp,ap,aimax,a->nonew,MatScalar); \
121: N = nrow++ - 1; \
122: /* shift up all the later entries in this row */ \
123: PetscArraymove(rp+_i+1,rp+_i,N-_i+1);\
124: PetscArraymove(ap+bs2*(_i+1),ap+bs2*_i,bs2*(N-_i+1)); \
125: PetscArrayzero(ap+bs2*_i,bs2); \
126: rp[_i] = bcol; \
127: ap[bs2*_i + bs*cidx + ridx] = value; \
128: a_noinsert:; \
129: ailen[brow] = nrow; \
130: }
132: #define MatSetValues_SeqBAIJ_B_Private(row,col,value,addv,orow,ocol) \
133: { \
134: brow = row/bs; \
135: rp = bj + bi[brow]; ap = ba + bs2*bi[brow]; \
136: rmax = bimax[brow]; nrow = bilen[brow]; \
137: bcol = col/bs; \
138: ridx = row % bs; cidx = col % bs; \
139: low = 0; high = nrow; \
140: while (high-low > 3) { \
141: t = (low+high)/2; \
142: if (rp[t] > bcol) high = t; \
143: else low = t; \
144: } \
145: for (_i=low; _i<high; _i++) { \
146: if (rp[_i] > bcol) break; \
147: if (rp[_i] == bcol) { \
148: bap = ap + bs2*_i + bs*cidx + ridx; \
149: if (addv == ADD_VALUES) *bap += value; \
150: else *bap = value; \
151: goto b_noinsert; \
152: } \
153: } \
154: if (b->nonew == 1) goto b_noinsert; \
155: if (b->nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at global row/column (%D, %D) into matrix", orow, ocol); \
156: MatSeqXAIJReallocateAIJ(B,b->mbs,bs2,nrow,brow,bcol,rmax,ba,bi,bj,rp,ap,bimax,b->nonew,MatScalar); \
157: N = nrow++ - 1; \
158: /* shift up all the later entries in this row */ \
159: PetscArraymove(rp+_i+1,rp+_i,N-_i+1);\
160: PetscArraymove(ap+bs2*(_i+1),ap+bs2*_i,bs2*(N-_i+1));\
161: PetscArrayzero(ap+bs2*_i,bs2); \
162: rp[_i] = bcol; \
163: ap[bs2*_i + bs*cidx + ridx] = value; \
164: b_noinsert:; \
165: bilen[brow] = nrow; \
166: }
168: PetscErrorCode MatSetValues_MPIBAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
169: {
170: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
171: MatScalar value;
172: PetscBool roworiented = baij->roworiented;
174: PetscInt i,j,row,col;
175: PetscInt rstart_orig=mat->rmap->rstart;
176: PetscInt rend_orig =mat->rmap->rend,cstart_orig=mat->cmap->rstart;
177: PetscInt cend_orig =mat->cmap->rend,bs=mat->rmap->bs;
179: /* Some Variables required in the macro */
180: Mat A = baij->A;
181: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)(A)->data;
182: PetscInt *aimax=a->imax,*ai=a->i,*ailen=a->ilen,*aj=a->j;
183: MatScalar *aa =a->a;
185: Mat B = baij->B;
186: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)(B)->data;
187: PetscInt *bimax=b->imax,*bi=b->i,*bilen=b->ilen,*bj=b->j;
188: MatScalar *ba =b->a;
190: PetscInt *rp,ii,nrow,_i,rmax,N,brow,bcol;
191: PetscInt low,high,t,ridx,cidx,bs2=a->bs2;
192: MatScalar *ap,*bap;
195: for (i=0; i<m; i++) {
196: if (im[i] < 0) continue;
197: #if defined(PETSC_USE_DEBUG)
198: if (im[i] >= mat->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap->N-1);
199: #endif
200: if (im[i] >= rstart_orig && im[i] < rend_orig) {
201: row = im[i] - rstart_orig;
202: for (j=0; j<n; j++) {
203: if (in[j] >= cstart_orig && in[j] < cend_orig) {
204: col = in[j] - cstart_orig;
205: if (roworiented) value = v[i*n+j];
206: else value = v[i+j*m];
207: MatSetValues_SeqBAIJ_A_Private(row,col,value,addv,im[i],in[j]);
208: } else if (in[j] < 0) continue;
209: #if defined(PETSC_USE_DEBUG)
210: else if (in[j] >= mat->cmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[j],mat->cmap->N-1);
211: #endif
212: else {
213: if (mat->was_assembled) {
214: if (!baij->colmap) {
215: MatCreateColmap_MPIBAIJ_Private(mat);
216: }
217: #if defined(PETSC_USE_CTABLE)
218: PetscTableFind(baij->colmap,in[j]/bs + 1,&col);
219: col = col - 1;
220: #else
221: col = baij->colmap[in[j]/bs] - 1;
222: #endif
223: if (col < 0 && !((Mat_SeqBAIJ*)(baij->B->data))->nonew) {
224: MatDisAssemble_MPIBAIJ(mat);
225: col = in[j];
226: /* Reinitialize the variables required by MatSetValues_SeqBAIJ_B_Private() */
227: B = baij->B;
228: b = (Mat_SeqBAIJ*)(B)->data;
229: bimax=b->imax;bi=b->i;bilen=b->ilen;bj=b->j;
230: ba =b->a;
231: } else if (col < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) into matrix", im[i], in[j]);
232: else col += in[j]%bs;
233: } else col = in[j];
234: if (roworiented) value = v[i*n+j];
235: else value = v[i+j*m];
236: MatSetValues_SeqBAIJ_B_Private(row,col,value,addv,im[i],in[j]);
237: /* MatSetValues_SeqBAIJ(baij->B,1,&row,1,&col,&value,addv); */
238: }
239: }
240: } else {
241: if (mat->nooffprocentries) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Setting off process row %D even though MatSetOption(,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) was set",im[i]);
242: if (!baij->donotstash) {
243: mat->assembled = PETSC_FALSE;
244: if (roworiented) {
245: MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,PETSC_FALSE);
246: } else {
247: MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,PETSC_FALSE);
248: }
249: }
250: }
251: }
252: return(0);
253: }
255: PETSC_STATIC_INLINE PetscErrorCode MatSetValuesBlocked_SeqBAIJ_Inlined(Mat A,PetscInt row,PetscInt col,const PetscScalar v[],InsertMode is,PetscInt orow,PetscInt ocol)
256: {
257: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
258: PetscInt *rp,low,high,t,ii,jj,nrow,i,rmax,N;
259: PetscInt *imax=a->imax,*ai=a->i,*ailen=a->ilen;
260: PetscErrorCode ierr;
261: PetscInt *aj =a->j,nonew=a->nonew,bs2=a->bs2,bs=A->rmap->bs;
262: PetscBool roworiented=a->roworiented;
263: const PetscScalar *value = v;
264: MatScalar *ap,*aa = a->a,*bap;
267: rp = aj + ai[row];
268: ap = aa + bs2*ai[row];
269: rmax = imax[row];
270: nrow = ailen[row];
271: value = v;
272: low = 0;
273: high = nrow;
274: while (high-low > 7) {
275: t = (low+high)/2;
276: if (rp[t] > col) high = t;
277: else low = t;
278: }
279: for (i=low; i<high; i++) {
280: if (rp[i] > col) break;
281: if (rp[i] == col) {
282: bap = ap + bs2*i;
283: if (roworiented) {
284: if (is == ADD_VALUES) {
285: for (ii=0; ii<bs; ii++) {
286: for (jj=ii; jj<bs2; jj+=bs) {
287: bap[jj] += *value++;
288: }
289: }
290: } else {
291: for (ii=0; ii<bs; ii++) {
292: for (jj=ii; jj<bs2; jj+=bs) {
293: bap[jj] = *value++;
294: }
295: }
296: }
297: } else {
298: if (is == ADD_VALUES) {
299: for (ii=0; ii<bs; ii++,value+=bs) {
300: for (jj=0; jj<bs; jj++) {
301: bap[jj] += value[jj];
302: }
303: bap += bs;
304: }
305: } else {
306: for (ii=0; ii<bs; ii++,value+=bs) {
307: for (jj=0; jj<bs; jj++) {
308: bap[jj] = value[jj];
309: }
310: bap += bs;
311: }
312: }
313: }
314: goto noinsert2;
315: }
316: }
317: if (nonew == 1) goto noinsert2;
318: if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new global block indexed nonzero block (%D, %D) in the matrix", orow, ocol);
319: MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
320: N = nrow++ - 1; high++;
321: /* shift up all the later entries in this row */
322: PetscArraymove(rp+i+1,rp+i,N-i+1);
323: PetscArraymove(ap+bs2*(i+1),ap+bs2*i,bs2*(N-i+1));
324: rp[i] = col;
325: bap = ap + bs2*i;
326: if (roworiented) {
327: for (ii=0; ii<bs; ii++) {
328: for (jj=ii; jj<bs2; jj+=bs) {
329: bap[jj] = *value++;
330: }
331: }
332: } else {
333: for (ii=0; ii<bs; ii++) {
334: for (jj=0; jj<bs; jj++) {
335: *bap++ = *value++;
336: }
337: }
338: }
339: noinsert2:;
340: ailen[row] = nrow;
341: return(0);
342: }
344: /*
345: This routine should be optimized so that the block copy at ** Here a copy is required ** below is not needed
346: by passing additional stride information into the MatSetValuesBlocked_SeqBAIJ_Inlined() routine
347: */
348: PetscErrorCode MatSetValuesBlocked_MPIBAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
349: {
350: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
351: const PetscScalar *value;
352: MatScalar *barray = baij->barray;
353: PetscBool roworiented = baij->roworiented;
354: PetscErrorCode ierr;
355: PetscInt i,j,ii,jj,row,col,rstart=baij->rstartbs;
356: PetscInt rend=baij->rendbs,cstart=baij->cstartbs,stepval;
357: PetscInt cend=baij->cendbs,bs=mat->rmap->bs,bs2=baij->bs2;
360: if (!barray) {
361: PetscMalloc1(bs2,&barray);
362: baij->barray = barray;
363: }
365: if (roworiented) stepval = (n-1)*bs;
366: else stepval = (m-1)*bs;
368: for (i=0; i<m; i++) {
369: if (im[i] < 0) continue;
370: #if defined(PETSC_USE_DEBUG)
371: if (im[i] >= baij->Mbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Block indexed row too large %D max %D",im[i],baij->Mbs-1);
372: #endif
373: if (im[i] >= rstart && im[i] < rend) {
374: row = im[i] - rstart;
375: for (j=0; j<n; j++) {
376: /* If NumCol = 1 then a copy is not required */
377: if ((roworiented) && (n == 1)) {
378: barray = (MatScalar*)v + i*bs2;
379: } else if ((!roworiented) && (m == 1)) {
380: barray = (MatScalar*)v + j*bs2;
381: } else { /* Here a copy is required */
382: if (roworiented) {
383: value = v + (i*(stepval+bs) + j)*bs;
384: } else {
385: value = v + (j*(stepval+bs) + i)*bs;
386: }
387: for (ii=0; ii<bs; ii++,value+=bs+stepval) {
388: for (jj=0; jj<bs; jj++) barray[jj] = value[jj];
389: barray += bs;
390: }
391: barray -= bs2;
392: }
394: if (in[j] >= cstart && in[j] < cend) {
395: col = in[j] - cstart;
396: MatSetValuesBlocked_SeqBAIJ_Inlined(baij->A,row,col,barray,addv,im[i],in[j]);
397: } else if (in[j] < 0) continue;
398: #if defined(PETSC_USE_DEBUG)
399: else if (in[j] >= baij->Nbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Block indexed column too large %D max %D",in[j],baij->Nbs-1);
400: #endif
401: else {
402: if (mat->was_assembled) {
403: if (!baij->colmap) {
404: MatCreateColmap_MPIBAIJ_Private(mat);
405: }
407: #if defined(PETSC_USE_DEBUG)
408: #if defined(PETSC_USE_CTABLE)
409: { PetscInt data;
410: PetscTableFind(baij->colmap,in[j]+1,&data);
411: if ((data - 1) % bs) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Incorrect colmap");
412: }
413: #else
414: if ((baij->colmap[in[j]] - 1) % bs) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Incorrect colmap");
415: #endif
416: #endif
417: #if defined(PETSC_USE_CTABLE)
418: PetscTableFind(baij->colmap,in[j]+1,&col);
419: col = (col - 1)/bs;
420: #else
421: col = (baij->colmap[in[j]] - 1)/bs;
422: #endif
423: if (col < 0 && !((Mat_SeqBAIJ*)(baij->B->data))->nonew) {
424: MatDisAssemble_MPIBAIJ(mat);
425: col = in[j];
426: } else if (col < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new blocked indexed nonzero block (%D, %D) into matrix",im[i],in[j]);
427: } else col = in[j];
428: MatSetValuesBlocked_SeqBAIJ_Inlined(baij->B,row,col,barray,addv,im[i],in[j]);
429: }
430: }
431: } else {
432: if (mat->nooffprocentries) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Setting off process block indexed row %D even though MatSetOption(,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) was set",im[i]);
433: if (!baij->donotstash) {
434: if (roworiented) {
435: MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
436: } else {
437: MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
438: }
439: }
440: }
441: }
442: return(0);
443: }
445: #define HASH_KEY 0.6180339887
446: #define HASH(size,key,tmp) (tmp = (key)*HASH_KEY,(PetscInt)((size)*(tmp-(PetscInt)tmp)))
447: /* #define HASH(size,key) ((PetscInt)((size)*fmod(((key)*HASH_KEY),1))) */
448: /* #define HASH(size,key,tmp) ((PetscInt)((size)*fmod(((key)*HASH_KEY),1))) */
449: PetscErrorCode MatSetValues_MPIBAIJ_HT(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
450: {
451: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
452: PetscBool roworiented = baij->roworiented;
454: PetscInt i,j,row,col;
455: PetscInt rstart_orig=mat->rmap->rstart;
456: PetscInt rend_orig =mat->rmap->rend,Nbs=baij->Nbs;
457: PetscInt h1,key,size=baij->ht_size,bs=mat->rmap->bs,*HT=baij->ht,idx;
458: PetscReal tmp;
459: MatScalar **HD = baij->hd,value;
460: #if defined(PETSC_USE_DEBUG)
461: PetscInt total_ct=baij->ht_total_ct,insert_ct=baij->ht_insert_ct;
462: #endif
465: for (i=0; i<m; i++) {
466: #if defined(PETSC_USE_DEBUG)
467: if (im[i] < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row");
468: if (im[i] >= mat->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],mat->rmap->N-1);
469: #endif
470: row = im[i];
471: if (row >= rstart_orig && row < rend_orig) {
472: for (j=0; j<n; j++) {
473: col = in[j];
474: if (roworiented) value = v[i*n+j];
475: else value = v[i+j*m];
476: /* Look up PetscInto the Hash Table */
477: key = (row/bs)*Nbs+(col/bs)+1;
478: h1 = HASH(size,key,tmp);
481: idx = h1;
482: #if defined(PETSC_USE_DEBUG)
483: insert_ct++;
484: total_ct++;
485: if (HT[idx] != key) {
486: for (idx=h1; (idx<size) && (HT[idx]!=key); idx++,total_ct++) ;
487: if (idx == size) {
488: for (idx=0; (idx<h1) && (HT[idx]!=key); idx++,total_ct++) ;
489: if (idx == h1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"(%D,%D) has no entry in the hash table", row, col);
490: }
491: }
492: #else
493: if (HT[idx] != key) {
494: for (idx=h1; (idx<size) && (HT[idx]!=key); idx++) ;
495: if (idx == size) {
496: for (idx=0; (idx<h1) && (HT[idx]!=key); idx++) ;
497: if (idx == h1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"(%D,%D) has no entry in the hash table", row, col);
498: }
499: }
500: #endif
501: /* A HASH table entry is found, so insert the values at the correct address */
502: if (addv == ADD_VALUES) *(HD[idx]+ (col % bs)*bs + (row % bs)) += value;
503: else *(HD[idx]+ (col % bs)*bs + (row % bs)) = value;
504: }
505: } else if (!baij->donotstash) {
506: if (roworiented) {
507: MatStashValuesRow_Private(&mat->stash,im[i],n,in,v+i*n,PETSC_FALSE);
508: } else {
509: MatStashValuesCol_Private(&mat->stash,im[i],n,in,v+i,m,PETSC_FALSE);
510: }
511: }
512: }
513: #if defined(PETSC_USE_DEBUG)
514: baij->ht_total_ct += total_ct;
515: baij->ht_insert_ct += insert_ct;
516: #endif
517: return(0);
518: }
520: PetscErrorCode MatSetValuesBlocked_MPIBAIJ_HT(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
521: {
522: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
523: PetscBool roworiented = baij->roworiented;
524: PetscErrorCode ierr;
525: PetscInt i,j,ii,jj,row,col;
526: PetscInt rstart=baij->rstartbs;
527: PetscInt rend =mat->rmap->rend,stepval,bs=mat->rmap->bs,bs2=baij->bs2,nbs2=n*bs2;
528: PetscInt h1,key,size=baij->ht_size,idx,*HT=baij->ht,Nbs=baij->Nbs;
529: PetscReal tmp;
530: MatScalar **HD = baij->hd,*baij_a;
531: const PetscScalar *v_t,*value;
532: #if defined(PETSC_USE_DEBUG)
533: PetscInt total_ct=baij->ht_total_ct,insert_ct=baij->ht_insert_ct;
534: #endif
537: if (roworiented) stepval = (n-1)*bs;
538: else stepval = (m-1)*bs;
540: for (i=0; i<m; i++) {
541: #if defined(PETSC_USE_DEBUG)
542: if (im[i] < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",im[i]);
543: if (im[i] >= baij->Mbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",im[i],baij->Mbs-1);
544: #endif
545: row = im[i];
546: v_t = v + i*nbs2;
547: if (row >= rstart && row < rend) {
548: for (j=0; j<n; j++) {
549: col = in[j];
551: /* Look up into the Hash Table */
552: key = row*Nbs+col+1;
553: h1 = HASH(size,key,tmp);
555: idx = h1;
556: #if defined(PETSC_USE_DEBUG)
557: total_ct++;
558: insert_ct++;
559: if (HT[idx] != key) {
560: for (idx=h1; (idx<size) && (HT[idx]!=key); idx++,total_ct++) ;
561: if (idx == size) {
562: for (idx=0; (idx<h1) && (HT[idx]!=key); idx++,total_ct++) ;
563: if (idx == h1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"(%D,%D) has no entry in the hash table", row, col);
564: }
565: }
566: #else
567: if (HT[idx] != key) {
568: for (idx=h1; (idx<size) && (HT[idx]!=key); idx++) ;
569: if (idx == size) {
570: for (idx=0; (idx<h1) && (HT[idx]!=key); idx++) ;
571: if (idx == h1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"(%D,%D) has no entry in the hash table", row, col);
572: }
573: }
574: #endif
575: baij_a = HD[idx];
576: if (roworiented) {
577: /*value = v + i*(stepval+bs)*bs + j*bs;*/
578: /* value = v + (i*(stepval+bs)+j)*bs; */
579: value = v_t;
580: v_t += bs;
581: if (addv == ADD_VALUES) {
582: for (ii=0; ii<bs; ii++,value+=stepval) {
583: for (jj=ii; jj<bs2; jj+=bs) {
584: baij_a[jj] += *value++;
585: }
586: }
587: } else {
588: for (ii=0; ii<bs; ii++,value+=stepval) {
589: for (jj=ii; jj<bs2; jj+=bs) {
590: baij_a[jj] = *value++;
591: }
592: }
593: }
594: } else {
595: value = v + j*(stepval+bs)*bs + i*bs;
596: if (addv == ADD_VALUES) {
597: for (ii=0; ii<bs; ii++,value+=stepval,baij_a+=bs) {
598: for (jj=0; jj<bs; jj++) {
599: baij_a[jj] += *value++;
600: }
601: }
602: } else {
603: for (ii=0; ii<bs; ii++,value+=stepval,baij_a+=bs) {
604: for (jj=0; jj<bs; jj++) {
605: baij_a[jj] = *value++;
606: }
607: }
608: }
609: }
610: }
611: } else {
612: if (!baij->donotstash) {
613: if (roworiented) {
614: MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
615: } else {
616: MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
617: }
618: }
619: }
620: }
621: #if defined(PETSC_USE_DEBUG)
622: baij->ht_total_ct += total_ct;
623: baij->ht_insert_ct += insert_ct;
624: #endif
625: return(0);
626: }
628: PetscErrorCode MatGetValues_MPIBAIJ(Mat mat,PetscInt m,const PetscInt idxm[],PetscInt n,const PetscInt idxn[],PetscScalar v[])
629: {
630: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
632: PetscInt bs = mat->rmap->bs,i,j,bsrstart = mat->rmap->rstart,bsrend = mat->rmap->rend;
633: PetscInt bscstart = mat->cmap->rstart,bscend = mat->cmap->rend,row,col,data;
636: for (i=0; i<m; i++) {
637: if (idxm[i] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",idxm[i]);*/
638: if (idxm[i] >= mat->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",idxm[i],mat->rmap->N-1);
639: if (idxm[i] >= bsrstart && idxm[i] < bsrend) {
640: row = idxm[i] - bsrstart;
641: for (j=0; j<n; j++) {
642: if (idxn[j] < 0) continue; /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",idxn[j]); */
643: if (idxn[j] >= mat->cmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",idxn[j],mat->cmap->N-1);
644: if (idxn[j] >= bscstart && idxn[j] < bscend) {
645: col = idxn[j] - bscstart;
646: MatGetValues_SeqBAIJ(baij->A,1,&row,1,&col,v+i*n+j);
647: } else {
648: if (!baij->colmap) {
649: MatCreateColmap_MPIBAIJ_Private(mat);
650: }
651: #if defined(PETSC_USE_CTABLE)
652: PetscTableFind(baij->colmap,idxn[j]/bs+1,&data);
653: data--;
654: #else
655: data = baij->colmap[idxn[j]/bs]-1;
656: #endif
657: if ((data < 0) || (baij->garray[data/bs] != idxn[j]/bs)) *(v+i*n+j) = 0.0;
658: else {
659: col = data + idxn[j]%bs;
660: MatGetValues_SeqBAIJ(baij->B,1,&row,1,&col,v+i*n+j);
661: }
662: }
663: }
664: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only local values currently supported");
665: }
666: return(0);
667: }
669: PetscErrorCode MatNorm_MPIBAIJ(Mat mat,NormType type,PetscReal *nrm)
670: {
671: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
672: Mat_SeqBAIJ *amat = (Mat_SeqBAIJ*)baij->A->data,*bmat = (Mat_SeqBAIJ*)baij->B->data;
674: PetscInt i,j,bs2=baij->bs2,bs=baij->A->rmap->bs,nz,row,col;
675: PetscReal sum = 0.0;
676: MatScalar *v;
679: if (baij->size == 1) {
680: MatNorm(baij->A,type,nrm);
681: } else {
682: if (type == NORM_FROBENIUS) {
683: v = amat->a;
684: nz = amat->nz*bs2;
685: for (i=0; i<nz; i++) {
686: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
687: }
688: v = bmat->a;
689: nz = bmat->nz*bs2;
690: for (i=0; i<nz; i++) {
691: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
692: }
693: MPIU_Allreduce(&sum,nrm,1,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)mat));
694: *nrm = PetscSqrtReal(*nrm);
695: } else if (type == NORM_1) { /* max column sum */
696: PetscReal *tmp,*tmp2;
697: PetscInt *jj,*garray=baij->garray,cstart=baij->rstartbs;
698: PetscCalloc1(mat->cmap->N,&tmp);
699: PetscMalloc1(mat->cmap->N,&tmp2);
700: v = amat->a; jj = amat->j;
701: for (i=0; i<amat->nz; i++) {
702: for (j=0; j<bs; j++) {
703: col = bs*(cstart + *jj) + j; /* column index */
704: for (row=0; row<bs; row++) {
705: tmp[col] += PetscAbsScalar(*v); v++;
706: }
707: }
708: jj++;
709: }
710: v = bmat->a; jj = bmat->j;
711: for (i=0; i<bmat->nz; i++) {
712: for (j=0; j<bs; j++) {
713: col = bs*garray[*jj] + j;
714: for (row=0; row<bs; row++) {
715: tmp[col] += PetscAbsScalar(*v); v++;
716: }
717: }
718: jj++;
719: }
720: MPIU_Allreduce(tmp,tmp2,mat->cmap->N,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)mat));
721: *nrm = 0.0;
722: for (j=0; j<mat->cmap->N; j++) {
723: if (tmp2[j] > *nrm) *nrm = tmp2[j];
724: }
725: PetscFree(tmp);
726: PetscFree(tmp2);
727: } else if (type == NORM_INFINITY) { /* max row sum */
728: PetscReal *sums;
729: PetscMalloc1(bs,&sums);
730: sum = 0.0;
731: for (j=0; j<amat->mbs; j++) {
732: for (row=0; row<bs; row++) sums[row] = 0.0;
733: v = amat->a + bs2*amat->i[j];
734: nz = amat->i[j+1]-amat->i[j];
735: for (i=0; i<nz; i++) {
736: for (col=0; col<bs; col++) {
737: for (row=0; row<bs; row++) {
738: sums[row] += PetscAbsScalar(*v); v++;
739: }
740: }
741: }
742: v = bmat->a + bs2*bmat->i[j];
743: nz = bmat->i[j+1]-bmat->i[j];
744: for (i=0; i<nz; i++) {
745: for (col=0; col<bs; col++) {
746: for (row=0; row<bs; row++) {
747: sums[row] += PetscAbsScalar(*v); v++;
748: }
749: }
750: }
751: for (row=0; row<bs; row++) {
752: if (sums[row] > sum) sum = sums[row];
753: }
754: }
755: MPIU_Allreduce(&sum,nrm,1,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)mat));
756: PetscFree(sums);
757: } else SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"No support for this norm yet");
758: }
759: return(0);
760: }
762: /*
763: Creates the hash table, and sets the table
764: This table is created only once.
765: If new entried need to be added to the matrix
766: then the hash table has to be destroyed and
767: recreated.
768: */
769: PetscErrorCode MatCreateHashTable_MPIBAIJ_Private(Mat mat,PetscReal factor)
770: {
771: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
772: Mat A = baij->A,B=baij->B;
773: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data,*b=(Mat_SeqBAIJ*)B->data;
774: PetscInt i,j,k,nz=a->nz+b->nz,h1,*ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
776: PetscInt ht_size,bs2=baij->bs2,rstart=baij->rstartbs;
777: PetscInt cstart=baij->cstartbs,*garray=baij->garray,row,col,Nbs=baij->Nbs;
778: PetscInt *HT,key;
779: MatScalar **HD;
780: PetscReal tmp;
781: #if defined(PETSC_USE_INFO)
782: PetscInt ct=0,max=0;
783: #endif
786: if (baij->ht) return(0);
788: baij->ht_size = (PetscInt)(factor*nz);
789: ht_size = baij->ht_size;
791: /* Allocate Memory for Hash Table */
792: PetscCalloc2(ht_size,&baij->hd,ht_size,&baij->ht);
793: HD = baij->hd;
794: HT = baij->ht;
796: /* Loop Over A */
797: for (i=0; i<a->mbs; i++) {
798: for (j=ai[i]; j<ai[i+1]; j++) {
799: row = i+rstart;
800: col = aj[j]+cstart;
802: key = row*Nbs + col + 1;
803: h1 = HASH(ht_size,key,tmp);
804: for (k=0; k<ht_size; k++) {
805: if (!HT[(h1+k)%ht_size]) {
806: HT[(h1+k)%ht_size] = key;
807: HD[(h1+k)%ht_size] = a->a + j*bs2;
808: break;
809: #if defined(PETSC_USE_INFO)
810: } else {
811: ct++;
812: #endif
813: }
814: }
815: #if defined(PETSC_USE_INFO)
816: if (k> max) max = k;
817: #endif
818: }
819: }
820: /* Loop Over B */
821: for (i=0; i<b->mbs; i++) {
822: for (j=bi[i]; j<bi[i+1]; j++) {
823: row = i+rstart;
824: col = garray[bj[j]];
825: key = row*Nbs + col + 1;
826: h1 = HASH(ht_size,key,tmp);
827: for (k=0; k<ht_size; k++) {
828: if (!HT[(h1+k)%ht_size]) {
829: HT[(h1+k)%ht_size] = key;
830: HD[(h1+k)%ht_size] = b->a + j*bs2;
831: break;
832: #if defined(PETSC_USE_INFO)
833: } else {
834: ct++;
835: #endif
836: }
837: }
838: #if defined(PETSC_USE_INFO)
839: if (k> max) max = k;
840: #endif
841: }
842: }
844: /* Print Summary */
845: #if defined(PETSC_USE_INFO)
846: for (i=0,j=0; i<ht_size; i++) {
847: if (HT[i]) j++;
848: }
849: PetscInfo2(mat,"Average Search = %5.2f,max search = %D\n",(!j)? 0.0:((PetscReal)(ct+j))/j,max);
850: #endif
851: return(0);
852: }
854: PetscErrorCode MatAssemblyBegin_MPIBAIJ(Mat mat,MatAssemblyType mode)
855: {
856: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
858: PetscInt nstash,reallocs;
861: if (baij->donotstash || mat->nooffprocentries) return(0);
863: MatStashScatterBegin_Private(mat,&mat->stash,mat->rmap->range);
864: MatStashScatterBegin_Private(mat,&mat->bstash,baij->rangebs);
865: MatStashGetInfo_Private(&mat->stash,&nstash,&reallocs);
866: PetscInfo2(mat,"Stash has %D entries,uses %D mallocs.\n",nstash,reallocs);
867: MatStashGetInfo_Private(&mat->bstash,&nstash,&reallocs);
868: PetscInfo2(mat,"Block-Stash has %D entries, uses %D mallocs.\n",nstash,reallocs);
869: return(0);
870: }
872: PetscErrorCode MatAssemblyEnd_MPIBAIJ(Mat mat,MatAssemblyType mode)
873: {
874: Mat_MPIBAIJ *baij=(Mat_MPIBAIJ*)mat->data;
875: Mat_SeqBAIJ *a =(Mat_SeqBAIJ*)baij->A->data;
877: PetscInt i,j,rstart,ncols,flg,bs2=baij->bs2;
878: PetscInt *row,*col;
879: PetscBool r1,r2,r3,other_disassembled;
880: MatScalar *val;
881: PetscMPIInt n;
884: /* do not use 'b=(Mat_SeqBAIJ*)baij->B->data' as B can be reset in disassembly */
885: if (!baij->donotstash && !mat->nooffprocentries) {
886: while (1) {
887: MatStashScatterGetMesg_Private(&mat->stash,&n,&row,&col,&val,&flg);
888: if (!flg) break;
890: for (i=0; i<n;) {
891: /* Now identify the consecutive vals belonging to the same row */
892: for (j=i,rstart=row[j]; j<n; j++) {
893: if (row[j] != rstart) break;
894: }
895: if (j < n) ncols = j-i;
896: else ncols = n-i;
897: /* Now assemble all these values with a single function call */
898: MatSetValues_MPIBAIJ(mat,1,row+i,ncols,col+i,val+i,mat->insertmode);
899: i = j;
900: }
901: }
902: MatStashScatterEnd_Private(&mat->stash);
903: /* Now process the block-stash. Since the values are stashed column-oriented,
904: set the roworiented flag to column oriented, and after MatSetValues()
905: restore the original flags */
906: r1 = baij->roworiented;
907: r2 = a->roworiented;
908: r3 = ((Mat_SeqBAIJ*)baij->B->data)->roworiented;
910: baij->roworiented = PETSC_FALSE;
911: a->roworiented = PETSC_FALSE;
913: (((Mat_SeqBAIJ*)baij->B->data))->roworiented = PETSC_FALSE; /* b->roworiented */
914: while (1) {
915: MatStashScatterGetMesg_Private(&mat->bstash,&n,&row,&col,&val,&flg);
916: if (!flg) break;
918: for (i=0; i<n;) {
919: /* Now identify the consecutive vals belonging to the same row */
920: for (j=i,rstart=row[j]; j<n; j++) {
921: if (row[j] != rstart) break;
922: }
923: if (j < n) ncols = j-i;
924: else ncols = n-i;
925: MatSetValuesBlocked_MPIBAIJ(mat,1,row+i,ncols,col+i,val+i*bs2,mat->insertmode);
926: i = j;
927: }
928: }
929: MatStashScatterEnd_Private(&mat->bstash);
931: baij->roworiented = r1;
932: a->roworiented = r2;
934: ((Mat_SeqBAIJ*)baij->B->data)->roworiented = r3; /* b->roworiented */
935: }
937: MatAssemblyBegin(baij->A,mode);
938: MatAssemblyEnd(baij->A,mode);
940: /* determine if any processor has disassembled, if so we must
941: also disassemble ourselfs, in order that we may reassemble. */
942: /*
943: if nonzero structure of submatrix B cannot change then we know that
944: no processor disassembled thus we can skip this stuff
945: */
946: if (!((Mat_SeqBAIJ*)baij->B->data)->nonew) {
947: MPIU_Allreduce(&mat->was_assembled,&other_disassembled,1,MPIU_BOOL,MPI_PROD,PetscObjectComm((PetscObject)mat));
948: if (mat->was_assembled && !other_disassembled) {
949: MatDisAssemble_MPIBAIJ(mat);
950: }
951: }
953: if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) {
954: MatSetUpMultiply_MPIBAIJ(mat);
955: }
956: MatAssemblyBegin(baij->B,mode);
957: MatAssemblyEnd(baij->B,mode);
959: #if defined(PETSC_USE_INFO)
960: if (baij->ht && mode== MAT_FINAL_ASSEMBLY) {
961: PetscInfo1(mat,"Average Hash Table Search in MatSetValues = %5.2f\n",(double)((PetscReal)baij->ht_total_ct)/baij->ht_insert_ct);
963: baij->ht_total_ct = 0;
964: baij->ht_insert_ct = 0;
965: }
966: #endif
967: if (baij->ht_flag && !baij->ht && mode == MAT_FINAL_ASSEMBLY) {
968: MatCreateHashTable_MPIBAIJ_Private(mat,baij->ht_fact);
970: mat->ops->setvalues = MatSetValues_MPIBAIJ_HT;
971: mat->ops->setvaluesblocked = MatSetValuesBlocked_MPIBAIJ_HT;
972: }
974: PetscFree2(baij->rowvalues,baij->rowindices);
976: baij->rowvalues = 0;
978: /* if no new nonzero locations are allowed in matrix then only set the matrix state the first time through */
979: if ((!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) || !((Mat_SeqBAIJ*)(baij->A->data))->nonew) {
980: PetscObjectState state = baij->A->nonzerostate + baij->B->nonzerostate;
981: MPIU_Allreduce(&state,&mat->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)mat));
982: }
983: return(0);
984: }
986: extern PetscErrorCode MatView_SeqBAIJ(Mat,PetscViewer);
987: #include <petscdraw.h>
988: static PetscErrorCode MatView_MPIBAIJ_ASCIIorDraworSocket(Mat mat,PetscViewer viewer)
989: {
990: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
991: PetscErrorCode ierr;
992: PetscMPIInt rank = baij->rank;
993: PetscInt bs = mat->rmap->bs;
994: PetscBool iascii,isdraw;
995: PetscViewer sviewer;
996: PetscViewerFormat format;
999: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
1000: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
1001: if (iascii) {
1002: PetscViewerGetFormat(viewer,&format);
1003: if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1004: MatInfo info;
1005: MPI_Comm_rank(PetscObjectComm((PetscObject)mat),&rank);
1006: MatGetInfo(mat,MAT_LOCAL,&info);
1007: PetscViewerASCIIPushSynchronized(viewer);
1008: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] Local rows %D nz %D nz alloced %D bs %D mem %g\n",
1009: rank,mat->rmap->n,(PetscInt)info.nz_used,(PetscInt)info.nz_allocated,mat->rmap->bs,(double)info.memory);
1010: MatGetInfo(baij->A,MAT_LOCAL,&info);
1011: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] on-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);
1012: MatGetInfo(baij->B,MAT_LOCAL,&info);
1013: PetscViewerASCIISynchronizedPrintf(viewer,"[%d] off-diagonal part: nz %D \n",rank,(PetscInt)info.nz_used);
1014: PetscViewerFlush(viewer);
1015: PetscViewerASCIIPopSynchronized(viewer);
1016: PetscViewerASCIIPrintf(viewer,"Information on VecScatter used in matrix-vector product: \n");
1017: VecScatterView(baij->Mvctx,viewer);
1018: return(0);
1019: } else if (format == PETSC_VIEWER_ASCII_INFO) {
1020: PetscViewerASCIIPrintf(viewer," block size is %D\n",bs);
1021: return(0);
1022: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
1023: return(0);
1024: }
1025: }
1027: if (isdraw) {
1028: PetscDraw draw;
1029: PetscBool isnull;
1030: PetscViewerDrawGetDraw(viewer,0,&draw);
1031: PetscDrawIsNull(draw,&isnull);
1032: if (isnull) return(0);
1033: }
1035: {
1036: /* assemble the entire matrix onto first processor. */
1037: Mat A;
1038: Mat_SeqBAIJ *Aloc;
1039: PetscInt M = mat->rmap->N,N = mat->cmap->N,*ai,*aj,col,i,j,k,*rvals,mbs = baij->mbs;
1040: MatScalar *a;
1041: const char *matname;
1043: /* Here we are creating a temporary matrix, so will assume MPIBAIJ is acceptable */
1044: /* Perhaps this should be the type of mat? */
1045: MatCreate(PetscObjectComm((PetscObject)mat),&A);
1046: if (!rank) {
1047: MatSetSizes(A,M,N,M,N);
1048: } else {
1049: MatSetSizes(A,0,0,M,N);
1050: }
1051: MatSetType(A,MATMPIBAIJ);
1052: MatMPIBAIJSetPreallocation(A,mat->rmap->bs,0,NULL,0,NULL);
1053: MatSetOption(A,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_FALSE);
1054: PetscLogObjectParent((PetscObject)mat,(PetscObject)A);
1056: /* copy over the A part */
1057: Aloc = (Mat_SeqBAIJ*)baij->A->data;
1058: ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
1059: PetscMalloc1(bs,&rvals);
1061: for (i=0; i<mbs; i++) {
1062: rvals[0] = bs*(baij->rstartbs + i);
1063: for (j=1; j<bs; j++) rvals[j] = rvals[j-1] + 1;
1064: for (j=ai[i]; j<ai[i+1]; j++) {
1065: col = (baij->cstartbs+aj[j])*bs;
1066: for (k=0; k<bs; k++) {
1067: MatSetValues_MPIBAIJ(A,bs,rvals,1,&col,a,INSERT_VALUES);
1068: col++; a += bs;
1069: }
1070: }
1071: }
1072: /* copy over the B part */
1073: Aloc = (Mat_SeqBAIJ*)baij->B->data;
1074: ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
1075: for (i=0; i<mbs; i++) {
1076: rvals[0] = bs*(baij->rstartbs + i);
1077: for (j=1; j<bs; j++) rvals[j] = rvals[j-1] + 1;
1078: for (j=ai[i]; j<ai[i+1]; j++) {
1079: col = baij->garray[aj[j]]*bs;
1080: for (k=0; k<bs; k++) {
1081: MatSetValues_MPIBAIJ(A,bs,rvals,1,&col,a,INSERT_VALUES);
1082: col++; a += bs;
1083: }
1084: }
1085: }
1086: PetscFree(rvals);
1087: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
1088: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
1089: /*
1090: Everyone has to call to draw the matrix since the graphics waits are
1091: synchronized across all processors that share the PetscDraw object
1092: */
1093: PetscViewerGetSubViewer(viewer,PETSC_COMM_SELF,&sviewer);
1094: PetscObjectGetName((PetscObject)mat,&matname);
1095: if (!rank) {
1096: PetscObjectSetName((PetscObject)((Mat_MPIBAIJ*)(A->data))->A,matname);
1097: MatView_SeqBAIJ(((Mat_MPIBAIJ*)(A->data))->A,sviewer);
1098: }
1099: PetscViewerRestoreSubViewer(viewer,PETSC_COMM_SELF,&sviewer);
1100: PetscViewerFlush(viewer);
1101: MatDestroy(&A);
1102: }
1103: return(0);
1104: }
1106: static PetscErrorCode MatView_MPIBAIJ_Binary(Mat mat,PetscViewer viewer)
1107: {
1108: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)mat->data;
1109: Mat_SeqBAIJ *A = (Mat_SeqBAIJ*)a->A->data;
1110: Mat_SeqBAIJ *B = (Mat_SeqBAIJ*)a->B->data;
1112: PetscInt i,*row_lens,*crow_lens,bs = mat->rmap->bs,j,k,bs2=a->bs2,header[4],nz,rlen;
1113: PetscInt *range=0,nzmax,*column_indices,cnt,col,*garray = a->garray,cstart = mat->cmap->rstart/bs,len,pcnt,l,ll;
1114: int fd;
1115: PetscScalar *column_values;
1116: FILE *file;
1117: PetscMPIInt rank,size,tag = ((PetscObject)viewer)->tag;
1118: PetscInt message_count,flowcontrolcount;
1121: MPI_Comm_rank(PetscObjectComm((PetscObject)mat),&rank);
1122: MPI_Comm_size(PetscObjectComm((PetscObject)mat),&size);
1123: nz = bs2*(A->nz + B->nz);
1124: rlen = mat->rmap->n;
1125: PetscViewerBinaryGetDescriptor(viewer,&fd);
1126: if (!rank) {
1127: header[0] = MAT_FILE_CLASSID;
1128: header[1] = mat->rmap->N;
1129: header[2] = mat->cmap->N;
1131: MPI_Reduce(&nz,&header[3],1,MPIU_INT,MPI_SUM,0,PetscObjectComm((PetscObject)mat));
1132: PetscBinaryWrite(fd,header,4,PETSC_INT,PETSC_TRUE);
1133: /* get largest number of rows any processor has */
1134: range = mat->rmap->range;
1135: for (i=1; i<size; i++) {
1136: rlen = PetscMax(rlen,range[i+1] - range[i]);
1137: }
1138: } else {
1139: MPI_Reduce(&nz,0,1,MPIU_INT,MPI_SUM,0,PetscObjectComm((PetscObject)mat));
1140: }
1142: PetscMalloc1(rlen/bs,&crow_lens);
1143: /* compute lengths of each row */
1144: for (i=0; i<a->mbs; i++) {
1145: crow_lens[i] = A->i[i+1] - A->i[i] + B->i[i+1] - B->i[i];
1146: }
1147: /* store the row lengths to the file */
1148: PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);
1149: if (!rank) {
1150: MPI_Status status;
1151: PetscMalloc1(rlen,&row_lens);
1152: rlen = (range[1] - range[0])/bs;
1153: for (i=0; i<rlen; i++) {
1154: for (j=0; j<bs; j++) {
1155: row_lens[i*bs+j] = bs*crow_lens[i];
1156: }
1157: }
1158: PetscBinaryWrite(fd,row_lens,bs*rlen,PETSC_INT,PETSC_TRUE);
1159: for (i=1; i<size; i++) {
1160: rlen = (range[i+1] - range[i])/bs;
1161: PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);
1162: MPI_Recv(crow_lens,rlen,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);
1163: for (k=0; k<rlen; k++) {
1164: for (j=0; j<bs; j++) {
1165: row_lens[k*bs+j] = bs*crow_lens[k];
1166: }
1167: }
1168: PetscBinaryWrite(fd,row_lens,bs*rlen,PETSC_INT,PETSC_TRUE);
1169: }
1170: PetscViewerFlowControlEndMaster(viewer,&message_count);
1171: PetscFree(row_lens);
1172: } else {
1173: PetscViewerFlowControlStepWorker(viewer,rank,&message_count);
1174: MPI_Send(crow_lens,mat->rmap->n/bs,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));
1175: PetscViewerFlowControlEndWorker(viewer,&message_count);
1176: }
1177: PetscFree(crow_lens);
1179: /* load up the local column indices. Include for all rows not just one for each block row since process 0 does not have the
1180: information needed to make it for each row from a block row. This does require more communication but still not more than
1181: the communication needed for the nonzero values */
1182: nzmax = nz; /* space a largest processor needs */
1183: MPI_Reduce(&nz,&nzmax,1,MPIU_INT,MPI_MAX,0,PetscObjectComm((PetscObject)mat));
1184: PetscMalloc1(nzmax,&column_indices);
1185: cnt = 0;
1186: for (i=0; i<a->mbs; i++) {
1187: pcnt = cnt;
1188: for (j=B->i[i]; j<B->i[i+1]; j++) {
1189: if ((col = garray[B->j[j]]) > cstart) break;
1190: for (l=0; l<bs; l++) {
1191: column_indices[cnt++] = bs*col+l;
1192: }
1193: }
1194: for (k=A->i[i]; k<A->i[i+1]; k++) {
1195: for (l=0; l<bs; l++) {
1196: column_indices[cnt++] = bs*(A->j[k] + cstart)+l;
1197: }
1198: }
1199: for (; j<B->i[i+1]; j++) {
1200: for (l=0; l<bs; l++) {
1201: column_indices[cnt++] = bs*garray[B->j[j]]+l;
1202: }
1203: }
1204: len = cnt - pcnt;
1205: for (k=1; k<bs; k++) {
1206: PetscArraycpy(&column_indices[cnt],&column_indices[pcnt],len);
1207: cnt += len;
1208: }
1209: }
1210: if (cnt != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB,"Internal PETSc error: cnt = %D nz = %D",cnt,nz);
1212: /* store the columns to the file */
1213: PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);
1214: if (!rank) {
1215: MPI_Status status;
1216: PetscBinaryWrite(fd,column_indices,nz,PETSC_INT,PETSC_TRUE);
1217: for (i=1; i<size; i++) {
1218: PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);
1219: MPI_Recv(&cnt,1,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);
1220: MPI_Recv(column_indices,cnt,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);
1221: PetscBinaryWrite(fd,column_indices,cnt,PETSC_INT,PETSC_TRUE);
1222: }
1223: PetscViewerFlowControlEndMaster(viewer,&message_count);
1224: } else {
1225: PetscViewerFlowControlStepWorker(viewer,rank,&message_count);
1226: MPI_Send(&cnt,1,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));
1227: MPI_Send(column_indices,cnt,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));
1228: PetscViewerFlowControlEndWorker(viewer,&message_count);
1229: }
1230: PetscFree(column_indices);
1232: /* load up the numerical values */
1233: PetscMalloc1(nzmax,&column_values);
1234: cnt = 0;
1235: for (i=0; i<a->mbs; i++) {
1236: rlen = bs*(B->i[i+1] - B->i[i] + A->i[i+1] - A->i[i]);
1237: for (j=B->i[i]; j<B->i[i+1]; j++) {
1238: if (garray[B->j[j]] > cstart) break;
1239: for (l=0; l<bs; l++) {
1240: for (ll=0; ll<bs; ll++) {
1241: column_values[cnt + l*rlen + ll] = B->a[bs2*j+l+bs*ll];
1242: }
1243: }
1244: cnt += bs;
1245: }
1246: for (k=A->i[i]; k<A->i[i+1]; k++) {
1247: for (l=0; l<bs; l++) {
1248: for (ll=0; ll<bs; ll++) {
1249: column_values[cnt + l*rlen + ll] = A->a[bs2*k+l+bs*ll];
1250: }
1251: }
1252: cnt += bs;
1253: }
1254: for (; j<B->i[i+1]; j++) {
1255: for (l=0; l<bs; l++) {
1256: for (ll=0; ll<bs; ll++) {
1257: column_values[cnt + l*rlen + ll] = B->a[bs2*j+l+bs*ll];
1258: }
1259: }
1260: cnt += bs;
1261: }
1262: cnt += (bs-1)*rlen;
1263: }
1264: if (cnt != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Internal PETSc error: cnt = %D nz = %D",cnt,nz);
1266: /* store the column values to the file */
1267: PetscViewerFlowControlStart(viewer,&message_count,&flowcontrolcount);
1268: if (!rank) {
1269: MPI_Status status;
1270: PetscBinaryWrite(fd,column_values,nz,PETSC_SCALAR,PETSC_TRUE);
1271: for (i=1; i<size; i++) {
1272: PetscViewerFlowControlStepMaster(viewer,i,&message_count,flowcontrolcount);
1273: MPI_Recv(&cnt,1,MPIU_INT,i,tag,PetscObjectComm((PetscObject)mat),&status);
1274: MPI_Recv(column_values,cnt,MPIU_SCALAR,i,tag,PetscObjectComm((PetscObject)mat),&status);
1275: PetscBinaryWrite(fd,column_values,cnt,PETSC_SCALAR,PETSC_TRUE);
1276: }
1277: PetscViewerFlowControlEndMaster(viewer,&message_count);
1278: } else {
1279: PetscViewerFlowControlStepWorker(viewer,rank,&message_count);
1280: MPI_Send(&nz,1,MPIU_INT,0,tag,PetscObjectComm((PetscObject)mat));
1281: MPI_Send(column_values,nz,MPIU_SCALAR,0,tag,PetscObjectComm((PetscObject)mat));
1282: PetscViewerFlowControlEndWorker(viewer,&message_count);
1283: }
1284: PetscFree(column_values);
1286: PetscViewerBinaryGetInfoPointer(viewer,&file);
1287: if (file) {
1288: fprintf(file,"-matload_block_size %d\n",(int)mat->rmap->bs);
1289: }
1290: return(0);
1291: }
1293: PetscErrorCode MatView_MPIBAIJ(Mat mat,PetscViewer viewer)
1294: {
1296: PetscBool iascii,isdraw,issocket,isbinary;
1299: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
1300: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
1301: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERSOCKET,&issocket);
1302: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
1303: if (iascii || isdraw || issocket) {
1304: MatView_MPIBAIJ_ASCIIorDraworSocket(mat,viewer);
1305: } else if (isbinary) {
1306: MatView_MPIBAIJ_Binary(mat,viewer);
1307: }
1308: return(0);
1309: }
1311: PetscErrorCode MatDestroy_MPIBAIJ(Mat mat)
1312: {
1313: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1317: #if defined(PETSC_USE_LOG)
1318: PetscLogObjectState((PetscObject)mat,"Rows=%D,Cols=%D",mat->rmap->N,mat->cmap->N);
1319: #endif
1320: MatStashDestroy_Private(&mat->stash);
1321: MatStashDestroy_Private(&mat->bstash);
1322: MatDestroy(&baij->A);
1323: MatDestroy(&baij->B);
1324: #if defined(PETSC_USE_CTABLE)
1325: PetscTableDestroy(&baij->colmap);
1326: #else
1327: PetscFree(baij->colmap);
1328: #endif
1329: PetscFree(baij->garray);
1330: VecDestroy(&baij->lvec);
1331: VecScatterDestroy(&baij->Mvctx);
1332: PetscFree2(baij->rowvalues,baij->rowindices);
1333: PetscFree(baij->barray);
1334: PetscFree2(baij->hd,baij->ht);
1335: PetscFree(baij->rangebs);
1336: PetscFree(mat->data);
1338: PetscObjectChangeTypeName((PetscObject)mat,0);
1339: PetscObjectComposeFunction((PetscObject)mat,"MatStoreValues_C",NULL);
1340: PetscObjectComposeFunction((PetscObject)mat,"MatRetrieveValues_C",NULL);
1341: PetscObjectComposeFunction((PetscObject)mat,"MatMPIBAIJSetPreallocation_C",NULL);
1342: PetscObjectComposeFunction((PetscObject)mat,"MatMPIBAIJSetPreallocationCSR_C",NULL);
1343: PetscObjectComposeFunction((PetscObject)mat,"MatDiagonalScaleLocal_C",NULL);
1344: PetscObjectComposeFunction((PetscObject)mat,"MatSetHashTableFactor_C",NULL);
1345: PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpibaij_mpisbaij_C",NULL);
1346: PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpibaij_mpibstrm_C",NULL);
1347: #if defined(PETSC_HAVE_HYPRE)
1348: PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpibaij_hypre_C",NULL);
1349: #endif
1350: PetscObjectComposeFunction((PetscObject)mat,"MatConvert_mpibaij_is_C",NULL);
1351: PetscObjectComposeFunction((PetscObject)mat,"MatPtAP_is_mpibaij_C",NULL);
1352: return(0);
1353: }
1355: PetscErrorCode MatMult_MPIBAIJ(Mat A,Vec xx,Vec yy)
1356: {
1357: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1359: PetscInt nt;
1362: VecGetLocalSize(xx,&nt);
1363: if (nt != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Incompatible partition of A and xx");
1364: VecGetLocalSize(yy,&nt);
1365: if (nt != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Incompatible parition of A and yy");
1366: VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
1367: (*a->A->ops->mult)(a->A,xx,yy);
1368: VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
1369: (*a->B->ops->multadd)(a->B,a->lvec,yy,yy);
1370: return(0);
1371: }
1373: PetscErrorCode MatMultAdd_MPIBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1374: {
1375: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1379: VecScatterBegin(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
1380: (*a->A->ops->multadd)(a->A,xx,yy,zz);
1381: VecScatterEnd(a->Mvctx,xx,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
1382: (*a->B->ops->multadd)(a->B,a->lvec,zz,zz);
1383: return(0);
1384: }
1386: PetscErrorCode MatMultTranspose_MPIBAIJ(Mat A,Vec xx,Vec yy)
1387: {
1388: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1392: /* do nondiagonal part */
1393: (*a->B->ops->multtranspose)(a->B,xx,a->lvec);
1394: /* do local part */
1395: (*a->A->ops->multtranspose)(a->A,xx,yy);
1396: /* add partial results together */
1397: VecScatterBegin(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);
1398: VecScatterEnd(a->Mvctx,a->lvec,yy,ADD_VALUES,SCATTER_REVERSE);
1399: return(0);
1400: }
1402: PetscErrorCode MatMultTransposeAdd_MPIBAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1403: {
1404: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1408: /* do nondiagonal part */
1409: (*a->B->ops->multtranspose)(a->B,xx,a->lvec);
1410: /* do local part */
1411: (*a->A->ops->multtransposeadd)(a->A,xx,yy,zz);
1412: /* add partial results together */
1413: VecScatterBegin(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);
1414: VecScatterEnd(a->Mvctx,a->lvec,zz,ADD_VALUES,SCATTER_REVERSE);
1415: return(0);
1416: }
1418: /*
1419: This only works correctly for square matrices where the subblock A->A is the
1420: diagonal block
1421: */
1422: PetscErrorCode MatGetDiagonal_MPIBAIJ(Mat A,Vec v)
1423: {
1424: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1428: if (A->rmap->N != A->cmap->N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Supports only square matrix where A->A is diag block");
1429: MatGetDiagonal(a->A,v);
1430: return(0);
1431: }
1433: PetscErrorCode MatScale_MPIBAIJ(Mat A,PetscScalar aa)
1434: {
1435: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1439: MatScale(a->A,aa);
1440: MatScale(a->B,aa);
1441: return(0);
1442: }
1444: PetscErrorCode MatGetRow_MPIBAIJ(Mat matin,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1445: {
1446: Mat_MPIBAIJ *mat = (Mat_MPIBAIJ*)matin->data;
1447: PetscScalar *vworkA,*vworkB,**pvA,**pvB,*v_p;
1449: PetscInt bs = matin->rmap->bs,bs2 = mat->bs2,i,*cworkA,*cworkB,**pcA,**pcB;
1450: PetscInt nztot,nzA,nzB,lrow,brstart = matin->rmap->rstart,brend = matin->rmap->rend;
1451: PetscInt *cmap,*idx_p,cstart = mat->cstartbs;
1454: if (row < brstart || row >= brend) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only local rows");
1455: if (mat->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Already active");
1456: mat->getrowactive = PETSC_TRUE;
1458: if (!mat->rowvalues && (idx || v)) {
1459: /*
1460: allocate enough space to hold information from the longest row.
1461: */
1462: Mat_SeqBAIJ *Aa = (Mat_SeqBAIJ*)mat->A->data,*Ba = (Mat_SeqBAIJ*)mat->B->data;
1463: PetscInt max = 1,mbs = mat->mbs,tmp;
1464: for (i=0; i<mbs; i++) {
1465: tmp = Aa->i[i+1] - Aa->i[i] + Ba->i[i+1] - Ba->i[i];
1466: if (max < tmp) max = tmp;
1467: }
1468: PetscMalloc2(max*bs2,&mat->rowvalues,max*bs2,&mat->rowindices);
1469: }
1470: lrow = row - brstart;
1472: pvA = &vworkA; pcA = &cworkA; pvB = &vworkB; pcB = &cworkB;
1473: if (!v) {pvA = 0; pvB = 0;}
1474: if (!idx) {pcA = 0; if (!v) pcB = 0;}
1475: (*mat->A->ops->getrow)(mat->A,lrow,&nzA,pcA,pvA);
1476: (*mat->B->ops->getrow)(mat->B,lrow,&nzB,pcB,pvB);
1477: nztot = nzA + nzB;
1479: cmap = mat->garray;
1480: if (v || idx) {
1481: if (nztot) {
1482: /* Sort by increasing column numbers, assuming A and B already sorted */
1483: PetscInt imark = -1;
1484: if (v) {
1485: *v = v_p = mat->rowvalues;
1486: for (i=0; i<nzB; i++) {
1487: if (cmap[cworkB[i]/bs] < cstart) v_p[i] = vworkB[i];
1488: else break;
1489: }
1490: imark = i;
1491: for (i=0; i<nzA; i++) v_p[imark+i] = vworkA[i];
1492: for (i=imark; i<nzB; i++) v_p[nzA+i] = vworkB[i];
1493: }
1494: if (idx) {
1495: *idx = idx_p = mat->rowindices;
1496: if (imark > -1) {
1497: for (i=0; i<imark; i++) {
1498: idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs;
1499: }
1500: } else {
1501: for (i=0; i<nzB; i++) {
1502: if (cmap[cworkB[i]/bs] < cstart) idx_p[i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs;
1503: else break;
1504: }
1505: imark = i;
1506: }
1507: for (i=0; i<nzA; i++) idx_p[imark+i] = cstart*bs + cworkA[i];
1508: for (i=imark; i<nzB; i++) idx_p[nzA+i] = cmap[cworkB[i]/bs]*bs + cworkB[i]%bs ;
1509: }
1510: } else {
1511: if (idx) *idx = 0;
1512: if (v) *v = 0;
1513: }
1514: }
1515: *nz = nztot;
1516: (*mat->A->ops->restorerow)(mat->A,lrow,&nzA,pcA,pvA);
1517: (*mat->B->ops->restorerow)(mat->B,lrow,&nzB,pcB,pvB);
1518: return(0);
1519: }
1521: PetscErrorCode MatRestoreRow_MPIBAIJ(Mat mat,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1522: {
1523: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1526: if (!baij->getrowactive) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"MatGetRow not called");
1527: baij->getrowactive = PETSC_FALSE;
1528: return(0);
1529: }
1531: PetscErrorCode MatZeroEntries_MPIBAIJ(Mat A)
1532: {
1533: Mat_MPIBAIJ *l = (Mat_MPIBAIJ*)A->data;
1537: MatZeroEntries(l->A);
1538: MatZeroEntries(l->B);
1539: return(0);
1540: }
1542: PetscErrorCode MatGetInfo_MPIBAIJ(Mat matin,MatInfoType flag,MatInfo *info)
1543: {
1544: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)matin->data;
1545: Mat A = a->A,B = a->B;
1547: PetscLogDouble isend[5],irecv[5];
1550: info->block_size = (PetscReal)matin->rmap->bs;
1552: MatGetInfo(A,MAT_LOCAL,info);
1554: isend[0] = info->nz_used; isend[1] = info->nz_allocated; isend[2] = info->nz_unneeded;
1555: isend[3] = info->memory; isend[4] = info->mallocs;
1557: MatGetInfo(B,MAT_LOCAL,info);
1559: isend[0] += info->nz_used; isend[1] += info->nz_allocated; isend[2] += info->nz_unneeded;
1560: isend[3] += info->memory; isend[4] += info->mallocs;
1562: if (flag == MAT_LOCAL) {
1563: info->nz_used = isend[0];
1564: info->nz_allocated = isend[1];
1565: info->nz_unneeded = isend[2];
1566: info->memory = isend[3];
1567: info->mallocs = isend[4];
1568: } else if (flag == MAT_GLOBAL_MAX) {
1569: MPIU_Allreduce(isend,irecv,5,MPIU_PETSCLOGDOUBLE,MPI_MAX,PetscObjectComm((PetscObject)matin));
1571: info->nz_used = irecv[0];
1572: info->nz_allocated = irecv[1];
1573: info->nz_unneeded = irecv[2];
1574: info->memory = irecv[3];
1575: info->mallocs = irecv[4];
1576: } else if (flag == MAT_GLOBAL_SUM) {
1577: MPIU_Allreduce(isend,irecv,5,MPIU_PETSCLOGDOUBLE,MPI_SUM,PetscObjectComm((PetscObject)matin));
1579: info->nz_used = irecv[0];
1580: info->nz_allocated = irecv[1];
1581: info->nz_unneeded = irecv[2];
1582: info->memory = irecv[3];
1583: info->mallocs = irecv[4];
1584: } else SETERRQ1(PetscObjectComm((PetscObject)matin),PETSC_ERR_ARG_WRONG,"Unknown MatInfoType argument %d",(int)flag);
1585: info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */
1586: info->fill_ratio_needed = 0;
1587: info->factor_mallocs = 0;
1588: return(0);
1589: }
1591: PetscErrorCode MatSetOption_MPIBAIJ(Mat A,MatOption op,PetscBool flg)
1592: {
1593: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1597: switch (op) {
1598: case MAT_NEW_NONZERO_LOCATIONS:
1599: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1600: case MAT_UNUSED_NONZERO_LOCATION_ERR:
1601: case MAT_KEEP_NONZERO_PATTERN:
1602: case MAT_NEW_NONZERO_LOCATION_ERR:
1603: MatCheckPreallocated(A,1);
1604: MatSetOption(a->A,op,flg);
1605: MatSetOption(a->B,op,flg);
1606: break;
1607: case MAT_ROW_ORIENTED:
1608: MatCheckPreallocated(A,1);
1609: a->roworiented = flg;
1611: MatSetOption(a->A,op,flg);
1612: MatSetOption(a->B,op,flg);
1613: break;
1614: case MAT_NEW_DIAGONALS:
1615: case MAT_SORTED_FULL:
1616: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1617: break;
1618: case MAT_IGNORE_OFF_PROC_ENTRIES:
1619: a->donotstash = flg;
1620: break;
1621: case MAT_USE_HASH_TABLE:
1622: a->ht_flag = flg;
1623: a->ht_fact = 1.39;
1624: break;
1625: case MAT_SYMMETRIC:
1626: case MAT_STRUCTURALLY_SYMMETRIC:
1627: case MAT_HERMITIAN:
1628: case MAT_SUBMAT_SINGLEIS:
1629: case MAT_SYMMETRY_ETERNAL:
1630: MatCheckPreallocated(A,1);
1631: MatSetOption(a->A,op,flg);
1632: break;
1633: default:
1634: SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"unknown option %d",op);
1635: }
1636: return(0);
1637: }
1639: PetscErrorCode MatTranspose_MPIBAIJ(Mat A,MatReuse reuse,Mat *matout)
1640: {
1641: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)A->data;
1642: Mat_SeqBAIJ *Aloc;
1643: Mat B;
1645: PetscInt M =A->rmap->N,N=A->cmap->N,*ai,*aj,i,*rvals,j,k,col;
1646: PetscInt bs=A->rmap->bs,mbs=baij->mbs;
1647: MatScalar *a;
1650: if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_INPLACE_MATRIX) {
1651: MatCreate(PetscObjectComm((PetscObject)A),&B);
1652: MatSetSizes(B,A->cmap->n,A->rmap->n,N,M);
1653: MatSetType(B,((PetscObject)A)->type_name);
1654: /* Do not know preallocation information, but must set block size */
1655: MatMPIBAIJSetPreallocation(B,A->rmap->bs,PETSC_DECIDE,NULL,PETSC_DECIDE,NULL);
1656: } else {
1657: B = *matout;
1658: }
1660: /* copy over the A part */
1661: Aloc = (Mat_SeqBAIJ*)baij->A->data;
1662: ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
1663: PetscMalloc1(bs,&rvals);
1665: for (i=0; i<mbs; i++) {
1666: rvals[0] = bs*(baij->rstartbs + i);
1667: for (j=1; j<bs; j++) rvals[j] = rvals[j-1] + 1;
1668: for (j=ai[i]; j<ai[i+1]; j++) {
1669: col = (baij->cstartbs+aj[j])*bs;
1670: for (k=0; k<bs; k++) {
1671: MatSetValues_MPIBAIJ(B,1,&col,bs,rvals,a,INSERT_VALUES);
1673: col++; a += bs;
1674: }
1675: }
1676: }
1677: /* copy over the B part */
1678: Aloc = (Mat_SeqBAIJ*)baij->B->data;
1679: ai = Aloc->i; aj = Aloc->j; a = Aloc->a;
1680: for (i=0; i<mbs; i++) {
1681: rvals[0] = bs*(baij->rstartbs + i);
1682: for (j=1; j<bs; j++) rvals[j] = rvals[j-1] + 1;
1683: for (j=ai[i]; j<ai[i+1]; j++) {
1684: col = baij->garray[aj[j]]*bs;
1685: for (k=0; k<bs; k++) {
1686: MatSetValues_MPIBAIJ(B,1,&col,bs,rvals,a,INSERT_VALUES);
1687: col++;
1688: a += bs;
1689: }
1690: }
1691: }
1692: PetscFree(rvals);
1693: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
1694: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
1696: if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) *matout = B;
1697: else {
1698: MatHeaderMerge(A,&B);
1699: }
1700: return(0);
1701: }
1703: PetscErrorCode MatDiagonalScale_MPIBAIJ(Mat mat,Vec ll,Vec rr)
1704: {
1705: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
1706: Mat a = baij->A,b = baij->B;
1708: PetscInt s1,s2,s3;
1711: MatGetLocalSize(mat,&s2,&s3);
1712: if (rr) {
1713: VecGetLocalSize(rr,&s1);
1714: if (s1!=s3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"right vector non-conforming local size");
1715: /* Overlap communication with computation. */
1716: VecScatterBegin(baij->Mvctx,rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD);
1717: }
1718: if (ll) {
1719: VecGetLocalSize(ll,&s1);
1720: if (s1!=s2) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"left vector non-conforming local size");
1721: (*b->ops->diagonalscale)(b,ll,NULL);
1722: }
1723: /* scale the diagonal block */
1724: (*a->ops->diagonalscale)(a,ll,rr);
1726: if (rr) {
1727: /* Do a scatter end and then right scale the off-diagonal block */
1728: VecScatterEnd(baij->Mvctx,rr,baij->lvec,INSERT_VALUES,SCATTER_FORWARD);
1729: (*b->ops->diagonalscale)(b,NULL,baij->lvec);
1730: }
1731: return(0);
1732: }
1734: PetscErrorCode MatZeroRows_MPIBAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
1735: {
1736: Mat_MPIBAIJ *l = (Mat_MPIBAIJ *) A->data;
1737: PetscInt *lrows;
1738: PetscInt r, len;
1739: PetscBool cong;
1743: /* get locally owned rows */
1744: MatZeroRowsMapLocal_Private(A,N,rows,&len,&lrows);
1745: /* fix right hand side if needed */
1746: if (x && b) {
1747: const PetscScalar *xx;
1748: PetscScalar *bb;
1750: VecGetArrayRead(x,&xx);
1751: VecGetArray(b,&bb);
1752: for (r = 0; r < len; ++r) bb[lrows[r]] = diag*xx[lrows[r]];
1753: VecRestoreArrayRead(x,&xx);
1754: VecRestoreArray(b,&bb);
1755: }
1757: /* actually zap the local rows */
1758: /*
1759: Zero the required rows. If the "diagonal block" of the matrix
1760: is square and the user wishes to set the diagonal we use separate
1761: code so that MatSetValues() is not called for each diagonal allocating
1762: new memory, thus calling lots of mallocs and slowing things down.
1764: */
1765: /* must zero l->B before l->A because the (diag) case below may put values into l->B*/
1766: MatZeroRows_SeqBAIJ(l->B,len,lrows,0.0,NULL,NULL);
1767: MatHasCongruentLayouts(A,&cong);
1768: if ((diag != 0.0) && cong) {
1769: MatZeroRows_SeqBAIJ(l->A,len,lrows,diag,NULL,NULL);
1770: } else if (diag != 0.0) {
1771: MatZeroRows_SeqBAIJ(l->A,len,lrows,0.0,0,0);
1772: if (((Mat_SeqBAIJ*)l->A->data)->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatZeroRows() on rectangular matrices cannot be used with the Mat options \n\
1773: MAT_NEW_NONZERO_LOCATIONS,MAT_NEW_NONZERO_LOCATION_ERR,MAT_NEW_NONZERO_ALLOCATION_ERR");
1774: for (r = 0; r < len; ++r) {
1775: const PetscInt row = lrows[r] + A->rmap->rstart;
1776: MatSetValues(A,1,&row,1,&row,&diag,INSERT_VALUES);
1777: }
1778: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
1779: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
1780: } else {
1781: MatZeroRows_SeqBAIJ(l->A,len,lrows,0.0,NULL,NULL);
1782: }
1783: PetscFree(lrows);
1785: /* only change matrix nonzero state if pattern was allowed to be changed */
1786: if (!((Mat_SeqBAIJ*)(l->A->data))->keepnonzeropattern) {
1787: PetscObjectState state = l->A->nonzerostate + l->B->nonzerostate;
1788: MPIU_Allreduce(&state,&A->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)A));
1789: }
1790: return(0);
1791: }
1793: PetscErrorCode MatZeroRowsColumns_MPIBAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
1794: {
1795: Mat_MPIBAIJ *l = (Mat_MPIBAIJ*)A->data;
1796: PetscErrorCode ierr;
1797: PetscMPIInt n = A->rmap->n;
1798: PetscInt i,j,k,r,p = 0,len = 0,row,col,count;
1799: PetscInt *lrows,*owners = A->rmap->range;
1800: PetscSFNode *rrows;
1801: PetscSF sf;
1802: const PetscScalar *xx;
1803: PetscScalar *bb,*mask;
1804: Vec xmask,lmask;
1805: Mat_SeqBAIJ *baij = (Mat_SeqBAIJ*)l->B->data;
1806: PetscInt bs = A->rmap->bs, bs2 = baij->bs2;
1807: PetscScalar *aa;
1810: /* Create SF where leaves are input rows and roots are owned rows */
1811: PetscMalloc1(n, &lrows);
1812: for (r = 0; r < n; ++r) lrows[r] = -1;
1813: PetscMalloc1(N, &rrows);
1814: for (r = 0; r < N; ++r) {
1815: const PetscInt idx = rows[r];
1816: if (idx < 0 || A->rmap->N <= idx) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range [0,%D)",idx,A->rmap->N);
1817: if (idx < owners[p] || owners[p+1] <= idx) { /* short-circuit the search if the last p owns this row too */
1818: PetscLayoutFindOwner(A->rmap,idx,&p);
1819: }
1820: rrows[r].rank = p;
1821: rrows[r].index = rows[r] - owners[p];
1822: }
1823: PetscSFCreate(PetscObjectComm((PetscObject) A), &sf);
1824: PetscSFSetGraph(sf, n, N, NULL, PETSC_OWN_POINTER, rrows, PETSC_OWN_POINTER);
1825: /* Collect flags for rows to be zeroed */
1826: PetscSFReduceBegin(sf, MPIU_INT, (PetscInt *) rows, lrows, MPI_LOR);
1827: PetscSFReduceEnd(sf, MPIU_INT, (PetscInt *) rows, lrows, MPI_LOR);
1828: PetscSFDestroy(&sf);
1829: /* Compress and put in row numbers */
1830: for (r = 0; r < n; ++r) if (lrows[r] >= 0) lrows[len++] = r;
1831: /* zero diagonal part of matrix */
1832: MatZeroRowsColumns(l->A,len,lrows,diag,x,b);
1833: /* handle off diagonal part of matrix */
1834: MatCreateVecs(A,&xmask,NULL);
1835: VecDuplicate(l->lvec,&lmask);
1836: VecGetArray(xmask,&bb);
1837: for (i=0; i<len; i++) bb[lrows[i]] = 1;
1838: VecRestoreArray(xmask,&bb);
1839: VecScatterBegin(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);
1840: VecScatterEnd(l->Mvctx,xmask,lmask,ADD_VALUES,SCATTER_FORWARD);
1841: VecDestroy(&xmask);
1842: if (x) {
1843: VecScatterBegin(l->Mvctx,x,l->lvec,INSERT_VALUES,SCATTER_FORWARD);
1844: VecScatterEnd(l->Mvctx,x,l->lvec,INSERT_VALUES,SCATTER_FORWARD);
1845: VecGetArrayRead(l->lvec,&xx);
1846: VecGetArray(b,&bb);
1847: }
1848: VecGetArray(lmask,&mask);
1849: /* remove zeroed rows of off diagonal matrix */
1850: for (i = 0; i < len; ++i) {
1851: row = lrows[i];
1852: count = (baij->i[row/bs +1] - baij->i[row/bs])*bs;
1853: aa = ((MatScalar*)(baij->a)) + baij->i[row/bs]*bs2 + (row%bs);
1854: for (k = 0; k < count; ++k) {
1855: aa[0] = 0.0;
1856: aa += bs;
1857: }
1858: }
1859: /* loop over all elements of off process part of matrix zeroing removed columns*/
1860: for (i = 0; i < l->B->rmap->N; ++i) {
1861: row = i/bs;
1862: for (j = baij->i[row]; j < baij->i[row+1]; ++j) {
1863: for (k = 0; k < bs; ++k) {
1864: col = bs*baij->j[j] + k;
1865: if (PetscAbsScalar(mask[col])) {
1866: aa = ((MatScalar*)(baij->a)) + j*bs2 + (i%bs) + bs*k;
1867: if (x) bb[i] -= aa[0]*xx[col];
1868: aa[0] = 0.0;
1869: }
1870: }
1871: }
1872: }
1873: if (x) {
1874: VecRestoreArray(b,&bb);
1875: VecRestoreArrayRead(l->lvec,&xx);
1876: }
1877: VecRestoreArray(lmask,&mask);
1878: VecDestroy(&lmask);
1879: PetscFree(lrows);
1881: /* only change matrix nonzero state if pattern was allowed to be changed */
1882: if (!((Mat_SeqBAIJ*)(l->A->data))->keepnonzeropattern) {
1883: PetscObjectState state = l->A->nonzerostate + l->B->nonzerostate;
1884: MPIU_Allreduce(&state,&A->nonzerostate,1,MPIU_INT64,MPI_SUM,PetscObjectComm((PetscObject)A));
1885: }
1886: return(0);
1887: }
1889: PetscErrorCode MatSetUnfactored_MPIBAIJ(Mat A)
1890: {
1891: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1895: MatSetUnfactored(a->A);
1896: return(0);
1897: }
1899: static PetscErrorCode MatDuplicate_MPIBAIJ(Mat,MatDuplicateOption,Mat*);
1901: PetscErrorCode MatEqual_MPIBAIJ(Mat A,Mat B,PetscBool *flag)
1902: {
1903: Mat_MPIBAIJ *matB = (Mat_MPIBAIJ*)B->data,*matA = (Mat_MPIBAIJ*)A->data;
1904: Mat a,b,c,d;
1905: PetscBool flg;
1909: a = matA->A; b = matA->B;
1910: c = matB->A; d = matB->B;
1912: MatEqual(a,c,&flg);
1913: if (flg) {
1914: MatEqual(b,d,&flg);
1915: }
1916: MPIU_Allreduce(&flg,flag,1,MPIU_BOOL,MPI_LAND,PetscObjectComm((PetscObject)A));
1917: return(0);
1918: }
1920: PetscErrorCode MatCopy_MPIBAIJ(Mat A,Mat B,MatStructure str)
1921: {
1923: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
1924: Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)B->data;
1927: /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */
1928: if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) {
1929: MatCopy_Basic(A,B,str);
1930: } else {
1931: MatCopy(a->A,b->A,str);
1932: MatCopy(a->B,b->B,str);
1933: }
1934: PetscObjectStateIncrease((PetscObject)B);
1935: return(0);
1936: }
1938: PetscErrorCode MatSetUp_MPIBAIJ(Mat A)
1939: {
1943: MatMPIBAIJSetPreallocation(A,A->rmap->bs,PETSC_DEFAULT,0,PETSC_DEFAULT,0);
1944: return(0);
1945: }
1947: PetscErrorCode MatAXPYGetPreallocation_MPIBAIJ(Mat Y,const PetscInt *yltog,Mat X,const PetscInt *xltog,PetscInt *nnz)
1948: {
1950: PetscInt bs = Y->rmap->bs,m = Y->rmap->N/bs;
1951: Mat_SeqBAIJ *x = (Mat_SeqBAIJ*)X->data;
1952: Mat_SeqBAIJ *y = (Mat_SeqBAIJ*)Y->data;
1955: MatAXPYGetPreallocation_MPIX_private(m,x->i,x->j,xltog,y->i,y->j,yltog,nnz);
1956: return(0);
1957: }
1959: PetscErrorCode MatAXPY_MPIBAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
1960: {
1962: Mat_MPIBAIJ *xx=(Mat_MPIBAIJ*)X->data,*yy=(Mat_MPIBAIJ*)Y->data;
1963: PetscBLASInt bnz,one=1;
1964: Mat_SeqBAIJ *x,*y;
1965: PetscInt bs2 = Y->rmap->bs*Y->rmap->bs;
1968: if (str == SAME_NONZERO_PATTERN) {
1969: PetscScalar alpha = a;
1970: x = (Mat_SeqBAIJ*)xx->A->data;
1971: y = (Mat_SeqBAIJ*)yy->A->data;
1972: PetscBLASIntCast(x->nz*bs2,&bnz);
1973: PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
1974: x = (Mat_SeqBAIJ*)xx->B->data;
1975: y = (Mat_SeqBAIJ*)yy->B->data;
1976: PetscBLASIntCast(x->nz*bs2,&bnz);
1977: PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
1978: PetscObjectStateIncrease((PetscObject)Y);
1979: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
1980: MatAXPY_Basic(Y,a,X,str);
1981: } else {
1982: Mat B;
1983: PetscInt *nnz_d,*nnz_o,bs=Y->rmap->bs;
1984: PetscMalloc1(yy->A->rmap->N,&nnz_d);
1985: PetscMalloc1(yy->B->rmap->N,&nnz_o);
1986: MatCreate(PetscObjectComm((PetscObject)Y),&B);
1987: PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);
1988: MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);
1989: MatSetBlockSizesFromMats(B,Y,Y);
1990: MatSetType(B,MATMPIBAIJ);
1991: MatAXPYGetPreallocation_SeqBAIJ(yy->A,xx->A,nnz_d);
1992: MatAXPYGetPreallocation_MPIBAIJ(yy->B,yy->garray,xx->B,xx->garray,nnz_o);
1993: MatMPIBAIJSetPreallocation(B,bs,0,nnz_d,0,nnz_o);
1994: /* MatAXPY_BasicWithPreallocation() for BAIJ matrix is much slower than AIJ, even for bs=1 ! */
1995: MatAXPY_BasicWithPreallocation(B,Y,a,X,str);
1996: MatHeaderReplace(Y,&B);
1997: PetscFree(nnz_d);
1998: PetscFree(nnz_o);
1999: }
2000: return(0);
2001: }
2003: PetscErrorCode MatRealPart_MPIBAIJ(Mat A)
2004: {
2005: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
2009: MatRealPart(a->A);
2010: MatRealPart(a->B);
2011: return(0);
2012: }
2014: PetscErrorCode MatImaginaryPart_MPIBAIJ(Mat A)
2015: {
2016: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
2020: MatImaginaryPart(a->A);
2021: MatImaginaryPart(a->B);
2022: return(0);
2023: }
2025: PetscErrorCode MatCreateSubMatrix_MPIBAIJ(Mat mat,IS isrow,IS iscol,MatReuse call,Mat *newmat)
2026: {
2028: IS iscol_local;
2029: PetscInt csize;
2032: ISGetLocalSize(iscol,&csize);
2033: if (call == MAT_REUSE_MATRIX) {
2034: PetscObjectQuery((PetscObject)*newmat,"ISAllGather",(PetscObject*)&iscol_local);
2035: if (!iscol_local) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse");
2036: } else {
2037: ISAllGather(iscol,&iscol_local);
2038: }
2039: MatCreateSubMatrix_MPIBAIJ_Private(mat,isrow,iscol_local,csize,call,newmat);
2040: if (call == MAT_INITIAL_MATRIX) {
2041: PetscObjectCompose((PetscObject)*newmat,"ISAllGather",(PetscObject)iscol_local);
2042: ISDestroy(&iscol_local);
2043: }
2044: return(0);
2045: }
2047: /*
2048: Not great since it makes two copies of the submatrix, first an SeqBAIJ
2049: in local and then by concatenating the local matrices the end result.
2050: Writing it directly would be much like MatCreateSubMatrices_MPIBAIJ().
2051: This routine is used for BAIJ and SBAIJ matrices (unfortunate dependency).
2052: */
2053: PetscErrorCode MatCreateSubMatrix_MPIBAIJ_Private(Mat mat,IS isrow,IS iscol,PetscInt csize,MatReuse call,Mat *newmat)
2054: {
2056: PetscMPIInt rank,size;
2057: PetscInt i,m,n,rstart,row,rend,nz,*cwork,j,bs;
2058: PetscInt *ii,*jj,nlocal,*dlens,*olens,dlen,olen,jend,mglobal;
2059: Mat M,Mreuse;
2060: MatScalar *vwork,*aa;
2061: MPI_Comm comm;
2062: IS isrow_new, iscol_new;
2063: Mat_SeqBAIJ *aij;
2066: PetscObjectGetComm((PetscObject)mat,&comm);
2067: MPI_Comm_rank(comm,&rank);
2068: MPI_Comm_size(comm,&size);
2069: /* The compression and expansion should be avoided. Doesn't point
2070: out errors, might change the indices, hence buggey */
2071: ISCompressIndicesGeneral(mat->rmap->N,mat->rmap->n,mat->rmap->bs,1,&isrow,&isrow_new);
2072: ISCompressIndicesGeneral(mat->cmap->N,mat->cmap->n,mat->cmap->bs,1,&iscol,&iscol_new);
2074: if (call == MAT_REUSE_MATRIX) {
2075: PetscObjectQuery((PetscObject)*newmat,"SubMatrix",(PetscObject*)&Mreuse);
2076: if (!Mreuse) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Submatrix passed in was not used before, cannot reuse");
2077: MatCreateSubMatrices_MPIBAIJ_local(mat,1,&isrow_new,&iscol_new,MAT_REUSE_MATRIX,&Mreuse);
2078: } else {
2079: MatCreateSubMatrices_MPIBAIJ_local(mat,1,&isrow_new,&iscol_new,MAT_INITIAL_MATRIX,&Mreuse);
2080: }
2081: ISDestroy(&isrow_new);
2082: ISDestroy(&iscol_new);
2083: /*
2084: m - number of local rows
2085: n - number of columns (same on all processors)
2086: rstart - first row in new global matrix generated
2087: */
2088: MatGetBlockSize(mat,&bs);
2089: MatGetSize(Mreuse,&m,&n);
2090: m = m/bs;
2091: n = n/bs;
2093: if (call == MAT_INITIAL_MATRIX) {
2094: aij = (Mat_SeqBAIJ*)(Mreuse)->data;
2095: ii = aij->i;
2096: jj = aij->j;
2098: /*
2099: Determine the number of non-zeros in the diagonal and off-diagonal
2100: portions of the matrix in order to do correct preallocation
2101: */
2103: /* first get start and end of "diagonal" columns */
2104: if (csize == PETSC_DECIDE) {
2105: ISGetSize(isrow,&mglobal);
2106: if (mglobal == n*bs) { /* square matrix */
2107: nlocal = m;
2108: } else {
2109: nlocal = n/size + ((n % size) > rank);
2110: }
2111: } else {
2112: nlocal = csize/bs;
2113: }
2114: MPI_Scan(&nlocal,&rend,1,MPIU_INT,MPI_SUM,comm);
2115: rstart = rend - nlocal;
2116: if (rank == size - 1 && rend != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Local column sizes %D do not add up to total number of columns %D",rend,n);
2118: /* next, compute all the lengths */
2119: PetscMalloc2(m+1,&dlens,m+1,&olens);
2120: for (i=0; i<m; i++) {
2121: jend = ii[i+1] - ii[i];
2122: olen = 0;
2123: dlen = 0;
2124: for (j=0; j<jend; j++) {
2125: if (*jj < rstart || *jj >= rend) olen++;
2126: else dlen++;
2127: jj++;
2128: }
2129: olens[i] = olen;
2130: dlens[i] = dlen;
2131: }
2132: MatCreate(comm,&M);
2133: MatSetSizes(M,bs*m,bs*nlocal,PETSC_DECIDE,bs*n);
2134: MatSetType(M,((PetscObject)mat)->type_name);
2135: MatMPIBAIJSetPreallocation(M,bs,0,dlens,0,olens);
2136: MatMPISBAIJSetPreallocation(M,bs,0,dlens,0,olens);
2137: PetscFree2(dlens,olens);
2138: } else {
2139: PetscInt ml,nl;
2141: M = *newmat;
2142: MatGetLocalSize(M,&ml,&nl);
2143: if (ml != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Previous matrix must be same size/layout as request");
2144: MatZeroEntries(M);
2145: /*
2146: The next two lines are needed so we may call MatSetValues_MPIAIJ() below directly,
2147: rather than the slower MatSetValues().
2148: */
2149: M->was_assembled = PETSC_TRUE;
2150: M->assembled = PETSC_FALSE;
2151: }
2152: MatSetOption(M,MAT_ROW_ORIENTED,PETSC_FALSE);
2153: MatGetOwnershipRange(M,&rstart,&rend);
2154: aij = (Mat_SeqBAIJ*)(Mreuse)->data;
2155: ii = aij->i;
2156: jj = aij->j;
2157: aa = aij->a;
2158: for (i=0; i<m; i++) {
2159: row = rstart/bs + i;
2160: nz = ii[i+1] - ii[i];
2161: cwork = jj; jj += nz;
2162: vwork = aa; aa += nz*bs*bs;
2163: MatSetValuesBlocked_MPIBAIJ(M,1,&row,nz,cwork,vwork,INSERT_VALUES);
2164: }
2166: MatAssemblyBegin(M,MAT_FINAL_ASSEMBLY);
2167: MatAssemblyEnd(M,MAT_FINAL_ASSEMBLY);
2168: *newmat = M;
2170: /* save submatrix used in processor for next request */
2171: if (call == MAT_INITIAL_MATRIX) {
2172: PetscObjectCompose((PetscObject)M,"SubMatrix",(PetscObject)Mreuse);
2173: PetscObjectDereference((PetscObject)Mreuse);
2174: }
2175: return(0);
2176: }
2178: PetscErrorCode MatPermute_MPIBAIJ(Mat A,IS rowp,IS colp,Mat *B)
2179: {
2180: MPI_Comm comm,pcomm;
2181: PetscInt clocal_size,nrows;
2182: const PetscInt *rows;
2183: PetscMPIInt size;
2184: IS crowp,lcolp;
2188: PetscObjectGetComm((PetscObject)A,&comm);
2189: /* make a collective version of 'rowp' */
2190: PetscObjectGetComm((PetscObject)rowp,&pcomm);
2191: if (pcomm==comm) {
2192: crowp = rowp;
2193: } else {
2194: ISGetSize(rowp,&nrows);
2195: ISGetIndices(rowp,&rows);
2196: ISCreateGeneral(comm,nrows,rows,PETSC_COPY_VALUES,&crowp);
2197: ISRestoreIndices(rowp,&rows);
2198: }
2199: ISSetPermutation(crowp);
2200: /* make a local version of 'colp' */
2201: PetscObjectGetComm((PetscObject)colp,&pcomm);
2202: MPI_Comm_size(pcomm,&size);
2203: if (size==1) {
2204: lcolp = colp;
2205: } else {
2206: ISAllGather(colp,&lcolp);
2207: }
2208: ISSetPermutation(lcolp);
2209: /* now we just get the submatrix */
2210: MatGetLocalSize(A,NULL,&clocal_size);
2211: MatCreateSubMatrix_MPIBAIJ_Private(A,crowp,lcolp,clocal_size,MAT_INITIAL_MATRIX,B);
2212: /* clean up */
2213: if (pcomm!=comm) {
2214: ISDestroy(&crowp);
2215: }
2216: if (size>1) {
2217: ISDestroy(&lcolp);
2218: }
2219: return(0);
2220: }
2222: PetscErrorCode MatGetGhosts_MPIBAIJ(Mat mat,PetscInt *nghosts,const PetscInt *ghosts[])
2223: {
2224: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*) mat->data;
2225: Mat_SeqBAIJ *B = (Mat_SeqBAIJ*)baij->B->data;
2228: if (nghosts) *nghosts = B->nbs;
2229: if (ghosts) *ghosts = baij->garray;
2230: return(0);
2231: }
2233: PetscErrorCode MatGetSeqNonzeroStructure_MPIBAIJ(Mat A,Mat *newmat)
2234: {
2235: Mat B;
2236: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
2237: Mat_SeqBAIJ *ad = (Mat_SeqBAIJ*)a->A->data,*bd = (Mat_SeqBAIJ*)a->B->data;
2238: Mat_SeqAIJ *b;
2240: PetscMPIInt size,rank,*recvcounts = 0,*displs = 0;
2241: PetscInt sendcount,i,*rstarts = A->rmap->range,n,cnt,j,bs = A->rmap->bs;
2242: PetscInt m,*garray = a->garray,*lens,*jsendbuf,*a_jsendbuf,*b_jsendbuf;
2245: MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
2246: MPI_Comm_rank(PetscObjectComm((PetscObject)A),&rank);
2248: /* ----------------------------------------------------------------
2249: Tell every processor the number of nonzeros per row
2250: */
2251: PetscMalloc1(A->rmap->N/bs,&lens);
2252: for (i=A->rmap->rstart/bs; i<A->rmap->rend/bs; i++) {
2253: lens[i] = ad->i[i-A->rmap->rstart/bs+1] - ad->i[i-A->rmap->rstart/bs] + bd->i[i-A->rmap->rstart/bs+1] - bd->i[i-A->rmap->rstart/bs];
2254: }
2255: PetscMalloc1(2*size,&recvcounts);
2256: displs = recvcounts + size;
2257: for (i=0; i<size; i++) {
2258: recvcounts[i] = A->rmap->range[i+1]/bs - A->rmap->range[i]/bs;
2259: displs[i] = A->rmap->range[i]/bs;
2260: }
2261: #if defined(PETSC_HAVE_MPI_IN_PLACE)
2262: MPI_Allgatherv(MPI_IN_PLACE,0,MPI_DATATYPE_NULL,lens,recvcounts,displs,MPIU_INT,PetscObjectComm((PetscObject)A));
2263: #else
2264: sendcount = A->rmap->rend/bs - A->rmap->rstart/bs;
2265: MPI_Allgatherv(lens+A->rmap->rstart/bs,sendcount,MPIU_INT,lens,recvcounts,displs,MPIU_INT,PetscObjectComm((PetscObject)A));
2266: #endif
2267: /* ---------------------------------------------------------------
2268: Create the sequential matrix of the same type as the local block diagonal
2269: */
2270: MatCreate(PETSC_COMM_SELF,&B);
2271: MatSetSizes(B,A->rmap->N/bs,A->cmap->N/bs,PETSC_DETERMINE,PETSC_DETERMINE);
2272: MatSetType(B,MATSEQAIJ);
2273: MatSeqAIJSetPreallocation(B,0,lens);
2274: b = (Mat_SeqAIJ*)B->data;
2276: /*--------------------------------------------------------------------
2277: Copy my part of matrix column indices over
2278: */
2279: sendcount = ad->nz + bd->nz;
2280: jsendbuf = b->j + b->i[rstarts[rank]/bs];
2281: a_jsendbuf = ad->j;
2282: b_jsendbuf = bd->j;
2283: n = A->rmap->rend/bs - A->rmap->rstart/bs;
2284: cnt = 0;
2285: for (i=0; i<n; i++) {
2287: /* put in lower diagonal portion */
2288: m = bd->i[i+1] - bd->i[i];
2289: while (m > 0) {
2290: /* is it above diagonal (in bd (compressed) numbering) */
2291: if (garray[*b_jsendbuf] > A->rmap->rstart/bs + i) break;
2292: jsendbuf[cnt++] = garray[*b_jsendbuf++];
2293: m--;
2294: }
2296: /* put in diagonal portion */
2297: for (j=ad->i[i]; j<ad->i[i+1]; j++) {
2298: jsendbuf[cnt++] = A->rmap->rstart/bs + *a_jsendbuf++;
2299: }
2301: /* put in upper diagonal portion */
2302: while (m-- > 0) {
2303: jsendbuf[cnt++] = garray[*b_jsendbuf++];
2304: }
2305: }
2306: if (cnt != sendcount) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Corrupted PETSc matrix: nz given %D actual nz %D",sendcount,cnt);
2308: /*--------------------------------------------------------------------
2309: Gather all column indices to all processors
2310: */
2311: for (i=0; i<size; i++) {
2312: recvcounts[i] = 0;
2313: for (j=A->rmap->range[i]/bs; j<A->rmap->range[i+1]/bs; j++) {
2314: recvcounts[i] += lens[j];
2315: }
2316: }
2317: displs[0] = 0;
2318: for (i=1; i<size; i++) {
2319: displs[i] = displs[i-1] + recvcounts[i-1];
2320: }
2321: #if defined(PETSC_HAVE_MPI_IN_PLACE)
2322: MPI_Allgatherv(MPI_IN_PLACE,0,MPI_DATATYPE_NULL,b->j,recvcounts,displs,MPIU_INT,PetscObjectComm((PetscObject)A));
2323: #else
2324: MPI_Allgatherv(jsendbuf,sendcount,MPIU_INT,b->j,recvcounts,displs,MPIU_INT,PetscObjectComm((PetscObject)A));
2325: #endif
2326: /*--------------------------------------------------------------------
2327: Assemble the matrix into useable form (note numerical values not yet set)
2328: */
2329: /* set the b->ilen (length of each row) values */
2330: PetscArraycpy(b->ilen,lens,A->rmap->N/bs);
2331: /* set the b->i indices */
2332: b->i[0] = 0;
2333: for (i=1; i<=A->rmap->N/bs; i++) {
2334: b->i[i] = b->i[i-1] + lens[i-1];
2335: }
2336: PetscFree(lens);
2337: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2338: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2339: PetscFree(recvcounts);
2341: if (A->symmetric) {
2342: MatSetOption(B,MAT_SYMMETRIC,PETSC_TRUE);
2343: } else if (A->hermitian) {
2344: MatSetOption(B,MAT_HERMITIAN,PETSC_TRUE);
2345: } else if (A->structurally_symmetric) {
2346: MatSetOption(B,MAT_STRUCTURALLY_SYMMETRIC,PETSC_TRUE);
2347: }
2348: *newmat = B;
2349: return(0);
2350: }
2352: PetscErrorCode MatSOR_MPIBAIJ(Mat matin,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
2353: {
2354: Mat_MPIBAIJ *mat = (Mat_MPIBAIJ*)matin->data;
2356: Vec bb1 = 0;
2359: if (flag == SOR_APPLY_UPPER) {
2360: (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);
2361: return(0);
2362: }
2364: if (its > 1 || ~flag & SOR_ZERO_INITIAL_GUESS) {
2365: VecDuplicate(bb,&bb1);
2366: }
2368: if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP) {
2369: if (flag & SOR_ZERO_INITIAL_GUESS) {
2370: (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);
2371: its--;
2372: }
2374: while (its--) {
2375: VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2376: VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2378: /* update rhs: bb1 = bb - B*x */
2379: VecScale(mat->lvec,-1.0);
2380: (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);
2382: /* local sweep */
2383: (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_SYMMETRIC_SWEEP,fshift,lits,1,xx);
2384: }
2385: } else if (flag & SOR_LOCAL_FORWARD_SWEEP) {
2386: if (flag & SOR_ZERO_INITIAL_GUESS) {
2387: (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);
2388: its--;
2389: }
2390: while (its--) {
2391: VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2392: VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2394: /* update rhs: bb1 = bb - B*x */
2395: VecScale(mat->lvec,-1.0);
2396: (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);
2398: /* local sweep */
2399: (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_FORWARD_SWEEP,fshift,lits,1,xx);
2400: }
2401: } else if (flag & SOR_LOCAL_BACKWARD_SWEEP) {
2402: if (flag & SOR_ZERO_INITIAL_GUESS) {
2403: (*mat->A->ops->sor)(mat->A,bb,omega,flag,fshift,lits,1,xx);
2404: its--;
2405: }
2406: while (its--) {
2407: VecScatterBegin(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2408: VecScatterEnd(mat->Mvctx,xx,mat->lvec,INSERT_VALUES,SCATTER_FORWARD);
2410: /* update rhs: bb1 = bb - B*x */
2411: VecScale(mat->lvec,-1.0);
2412: (*mat->B->ops->multadd)(mat->B,mat->lvec,bb,bb1);
2414: /* local sweep */
2415: (*mat->A->ops->sor)(mat->A,bb1,omega,SOR_BACKWARD_SWEEP,fshift,lits,1,xx);
2416: }
2417: } else SETERRQ(PetscObjectComm((PetscObject)matin),PETSC_ERR_SUP,"Parallel version of SOR requested not supported");
2419: VecDestroy(&bb1);
2420: return(0);
2421: }
2423: PetscErrorCode MatGetColumnNorms_MPIBAIJ(Mat A,NormType type,PetscReal *norms)
2424: {
2426: Mat_MPIBAIJ *aij = (Mat_MPIBAIJ*)A->data;
2427: PetscInt N,i,*garray = aij->garray;
2428: PetscInt ib,jb,bs = A->rmap->bs;
2429: Mat_SeqBAIJ *a_aij = (Mat_SeqBAIJ*) aij->A->data;
2430: MatScalar *a_val = a_aij->a;
2431: Mat_SeqBAIJ *b_aij = (Mat_SeqBAIJ*) aij->B->data;
2432: MatScalar *b_val = b_aij->a;
2433: PetscReal *work;
2436: MatGetSize(A,NULL,&N);
2437: PetscCalloc1(N,&work);
2438: if (type == NORM_2) {
2439: for (i=a_aij->i[0]; i<a_aij->i[aij->A->rmap->n/bs]; i++) {
2440: for (jb=0; jb<bs; jb++) {
2441: for (ib=0; ib<bs; ib++) {
2442: work[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val * *a_val);
2443: a_val++;
2444: }
2445: }
2446: }
2447: for (i=b_aij->i[0]; i<b_aij->i[aij->B->rmap->n/bs]; i++) {
2448: for (jb=0; jb<bs; jb++) {
2449: for (ib=0; ib<bs; ib++) {
2450: work[garray[b_aij->j[i]] * bs + jb] += PetscAbsScalar(*b_val * *b_val);
2451: b_val++;
2452: }
2453: }
2454: }
2455: } else if (type == NORM_1) {
2456: for (i=a_aij->i[0]; i<a_aij->i[aij->A->rmap->n/bs]; i++) {
2457: for (jb=0; jb<bs; jb++) {
2458: for (ib=0; ib<bs; ib++) {
2459: work[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val);
2460: a_val++;
2461: }
2462: }
2463: }
2464: for (i=b_aij->i[0]; i<b_aij->i[aij->B->rmap->n/bs]; i++) {
2465: for (jb=0; jb<bs; jb++) {
2466: for (ib=0; ib<bs; ib++) {
2467: work[garray[b_aij->j[i]] * bs + jb] += PetscAbsScalar(*b_val);
2468: b_val++;
2469: }
2470: }
2471: }
2472: } else if (type == NORM_INFINITY) {
2473: for (i=a_aij->i[0]; i<a_aij->i[aij->A->rmap->n/bs]; i++) {
2474: for (jb=0; jb<bs; jb++) {
2475: for (ib=0; ib<bs; ib++) {
2476: int col = A->cmap->rstart + a_aij->j[i] * bs + jb;
2477: work[col] = PetscMax(PetscAbsScalar(*a_val), work[col]);
2478: a_val++;
2479: }
2480: }
2481: }
2482: for (i=b_aij->i[0]; i<b_aij->i[aij->B->rmap->n/bs]; i++) {
2483: for (jb=0; jb<bs; jb++) {
2484: for (ib=0; ib<bs; ib++) {
2485: int col = garray[b_aij->j[i]] * bs + jb;
2486: work[col] = PetscMax(PetscAbsScalar(*b_val), work[col]);
2487: b_val++;
2488: }
2489: }
2490: }
2491: } else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Unknown NormType");
2492: if (type == NORM_INFINITY) {
2493: MPIU_Allreduce(work,norms,N,MPIU_REAL,MPIU_MAX,PetscObjectComm((PetscObject)A));
2494: } else {
2495: MPIU_Allreduce(work,norms,N,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)A));
2496: }
2497: PetscFree(work);
2498: if (type == NORM_2) {
2499: for (i=0; i<N; i++) norms[i] = PetscSqrtReal(norms[i]);
2500: }
2501: return(0);
2502: }
2504: PetscErrorCode MatInvertBlockDiagonal_MPIBAIJ(Mat A,const PetscScalar **values)
2505: {
2506: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*) A->data;
2510: MatInvertBlockDiagonal(a->A,values);
2511: A->factorerrortype = a->A->factorerrortype;
2512: A->factorerror_zeropivot_value = a->A->factorerror_zeropivot_value;
2513: A->factorerror_zeropivot_row = a->A->factorerror_zeropivot_row;
2514: return(0);
2515: }
2517: PetscErrorCode MatShift_MPIBAIJ(Mat Y,PetscScalar a)
2518: {
2520: Mat_MPIBAIJ *maij = (Mat_MPIBAIJ*)Y->data;
2521: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ*)maij->A->data;
2524: if (!Y->preallocated) {
2525: MatMPIBAIJSetPreallocation(Y,Y->rmap->bs,1,NULL,0,NULL);
2526: } else if (!aij->nz) {
2527: PetscInt nonew = aij->nonew;
2528: MatSeqBAIJSetPreallocation(maij->A,Y->rmap->bs,1,NULL);
2529: aij->nonew = nonew;
2530: }
2531: MatShift_Basic(Y,a);
2532: return(0);
2533: }
2535: PetscErrorCode MatMissingDiagonal_MPIBAIJ(Mat A,PetscBool *missing,PetscInt *d)
2536: {
2537: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
2541: if (A->rmap->n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only works for square matrices");
2542: MatMissingDiagonal(a->A,missing,d);
2543: if (d) {
2544: PetscInt rstart;
2545: MatGetOwnershipRange(A,&rstart,NULL);
2546: *d += rstart/A->rmap->bs;
2548: }
2549: return(0);
2550: }
2552: PetscErrorCode MatGetDiagonalBlock_MPIBAIJ(Mat A,Mat *a)
2553: {
2555: *a = ((Mat_MPIBAIJ*)A->data)->A;
2556: return(0);
2557: }
2559: /* -------------------------------------------------------------------*/
2560: static struct _MatOps MatOps_Values = {MatSetValues_MPIBAIJ,
2561: MatGetRow_MPIBAIJ,
2562: MatRestoreRow_MPIBAIJ,
2563: MatMult_MPIBAIJ,
2564: /* 4*/ MatMultAdd_MPIBAIJ,
2565: MatMultTranspose_MPIBAIJ,
2566: MatMultTransposeAdd_MPIBAIJ,
2567: 0,
2568: 0,
2569: 0,
2570: /*10*/ 0,
2571: 0,
2572: 0,
2573: MatSOR_MPIBAIJ,
2574: MatTranspose_MPIBAIJ,
2575: /*15*/ MatGetInfo_MPIBAIJ,
2576: MatEqual_MPIBAIJ,
2577: MatGetDiagonal_MPIBAIJ,
2578: MatDiagonalScale_MPIBAIJ,
2579: MatNorm_MPIBAIJ,
2580: /*20*/ MatAssemblyBegin_MPIBAIJ,
2581: MatAssemblyEnd_MPIBAIJ,
2582: MatSetOption_MPIBAIJ,
2583: MatZeroEntries_MPIBAIJ,
2584: /*24*/ MatZeroRows_MPIBAIJ,
2585: 0,
2586: 0,
2587: 0,
2588: 0,
2589: /*29*/ MatSetUp_MPIBAIJ,
2590: 0,
2591: 0,
2592: MatGetDiagonalBlock_MPIBAIJ,
2593: 0,
2594: /*34*/ MatDuplicate_MPIBAIJ,
2595: 0,
2596: 0,
2597: 0,
2598: 0,
2599: /*39*/ MatAXPY_MPIBAIJ,
2600: MatCreateSubMatrices_MPIBAIJ,
2601: MatIncreaseOverlap_MPIBAIJ,
2602: MatGetValues_MPIBAIJ,
2603: MatCopy_MPIBAIJ,
2604: /*44*/ 0,
2605: MatScale_MPIBAIJ,
2606: MatShift_MPIBAIJ,
2607: 0,
2608: MatZeroRowsColumns_MPIBAIJ,
2609: /*49*/ 0,
2610: 0,
2611: 0,
2612: 0,
2613: 0,
2614: /*54*/ MatFDColoringCreate_MPIXAIJ,
2615: 0,
2616: MatSetUnfactored_MPIBAIJ,
2617: MatPermute_MPIBAIJ,
2618: MatSetValuesBlocked_MPIBAIJ,
2619: /*59*/ MatCreateSubMatrix_MPIBAIJ,
2620: MatDestroy_MPIBAIJ,
2621: MatView_MPIBAIJ,
2622: 0,
2623: 0,
2624: /*64*/ 0,
2625: 0,
2626: 0,
2627: 0,
2628: 0,
2629: /*69*/ MatGetRowMaxAbs_MPIBAIJ,
2630: 0,
2631: 0,
2632: 0,
2633: 0,
2634: /*74*/ 0,
2635: MatFDColoringApply_BAIJ,
2636: 0,
2637: 0,
2638: 0,
2639: /*79*/ 0,
2640: 0,
2641: 0,
2642: 0,
2643: MatLoad_MPIBAIJ,
2644: /*84*/ 0,
2645: 0,
2646: 0,
2647: 0,
2648: 0,
2649: /*89*/ 0,
2650: 0,
2651: 0,
2652: 0,
2653: 0,
2654: /*94*/ 0,
2655: 0,
2656: 0,
2657: 0,
2658: 0,
2659: /*99*/ 0,
2660: 0,
2661: 0,
2662: 0,
2663: 0,
2664: /*104*/0,
2665: MatRealPart_MPIBAIJ,
2666: MatImaginaryPart_MPIBAIJ,
2667: 0,
2668: 0,
2669: /*109*/0,
2670: 0,
2671: 0,
2672: 0,
2673: MatMissingDiagonal_MPIBAIJ,
2674: /*114*/MatGetSeqNonzeroStructure_MPIBAIJ,
2675: 0,
2676: MatGetGhosts_MPIBAIJ,
2677: 0,
2678: 0,
2679: /*119*/0,
2680: 0,
2681: 0,
2682: 0,
2683: MatGetMultiProcBlock_MPIBAIJ,
2684: /*124*/0,
2685: MatGetColumnNorms_MPIBAIJ,
2686: MatInvertBlockDiagonal_MPIBAIJ,
2687: 0,
2688: 0,
2689: /*129*/ 0,
2690: 0,
2691: 0,
2692: 0,
2693: 0,
2694: /*134*/ 0,
2695: 0,
2696: 0,
2697: 0,
2698: 0,
2699: /*139*/ MatSetBlockSizes_Default,
2700: 0,
2701: 0,
2702: MatFDColoringSetUp_MPIXAIJ,
2703: 0,
2704: /*144*/MatCreateMPIMatConcatenateSeqMat_MPIBAIJ
2705: };
2708: PETSC_INTERN PetscErrorCode MatConvert_MPIBAIJ_MPISBAIJ(Mat,MatType,MatReuse,Mat*);
2709: PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat,MatType,MatReuse,Mat*);
2711: PetscErrorCode MatMPIBAIJSetPreallocationCSR_MPIBAIJ(Mat B,PetscInt bs,const PetscInt ii[],const PetscInt jj[],const PetscScalar V[])
2712: {
2713: PetscInt m,rstart,cstart,cend;
2714: PetscInt i,j,dlen,olen,nz,nz_max=0,*d_nnz=0,*o_nnz=0;
2715: const PetscInt *JJ =0;
2716: PetscScalar *values=0;
2717: PetscBool roworiented = ((Mat_MPIBAIJ*)B->data)->roworiented;
2719: PetscBool nooffprocentries;
2722: PetscLayoutSetBlockSize(B->rmap,bs);
2723: PetscLayoutSetBlockSize(B->cmap,bs);
2724: PetscLayoutSetUp(B->rmap);
2725: PetscLayoutSetUp(B->cmap);
2726: PetscLayoutGetBlockSize(B->rmap,&bs);
2727: m = B->rmap->n/bs;
2728: rstart = B->rmap->rstart/bs;
2729: cstart = B->cmap->rstart/bs;
2730: cend = B->cmap->rend/bs;
2732: if (ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"ii[0] must be 0 but it is %D",ii[0]);
2733: PetscMalloc2(m,&d_nnz,m,&o_nnz);
2734: for (i=0; i<m; i++) {
2735: nz = ii[i+1] - ii[i];
2736: if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Local row %D has a negative number of columns %D",i,nz);
2737: nz_max = PetscMax(nz_max,nz);
2738: dlen = 0;
2739: olen = 0;
2740: JJ = jj + ii[i];
2741: for (j=0; j<nz; j++) {
2742: if (*JJ < cstart || *JJ >= cend) olen++;
2743: else dlen++;
2744: JJ++;
2745: }
2746: d_nnz[i] = dlen;
2747: o_nnz[i] = olen;
2748: }
2749: MatMPIBAIJSetPreallocation(B,bs,0,d_nnz,0,o_nnz);
2750: PetscFree2(d_nnz,o_nnz);
2752: values = (PetscScalar*)V;
2753: if (!values) {
2754: PetscCalloc1(bs*bs*nz_max,&values);
2755: }
2756: for (i=0; i<m; i++) {
2757: PetscInt row = i + rstart;
2758: PetscInt ncols = ii[i+1] - ii[i];
2759: const PetscInt *icols = jj + ii[i];
2760: if (bs == 1 || !roworiented) { /* block ordering matches the non-nested layout of MatSetValues so we can insert entire rows */
2761: const PetscScalar *svals = values + (V ? (bs*bs*ii[i]) : 0);
2762: MatSetValuesBlocked_MPIBAIJ(B,1,&row,ncols,icols,svals,INSERT_VALUES);
2763: } else { /* block ordering does not match so we can only insert one block at a time. */
2764: PetscInt j;
2765: for (j=0; j<ncols; j++) {
2766: const PetscScalar *svals = values + (V ? (bs*bs*(ii[i]+j)) : 0);
2767: MatSetValuesBlocked_MPIBAIJ(B,1,&row,1,&icols[j],svals,INSERT_VALUES);
2768: }
2769: }
2770: }
2772: if (!V) { PetscFree(values); }
2773: nooffprocentries = B->nooffprocentries;
2774: B->nooffprocentries = PETSC_TRUE;
2775: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2776: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2777: B->nooffprocentries = nooffprocentries;
2779: MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
2780: return(0);
2781: }
2783: /*@C
2784: MatMPIBAIJSetPreallocationCSR - Creates a sparse parallel matrix in BAIJ format using the given nonzero structure and (optional) numerical values
2786: Collective
2788: Input Parameters:
2789: + B - the matrix
2790: . bs - the block size
2791: . i - the indices into j for the start of each local row (starts with zero)
2792: . j - the column indices for each local row (starts with zero) these must be sorted for each row
2793: - v - optional values in the matrix
2795: Level: advanced
2797: Notes:
2798: The order of the entries in values is specified by the MatOption MAT_ROW_ORIENTED. For example, C programs
2799: may want to use the default MAT_ROW_ORIENTED=PETSC_TRUE and use an array v[nnz][bs][bs] where the second index is
2800: over rows within a block and the last index is over columns within a block row. Fortran programs will likely set
2801: MAT_ROW_ORIENTED=PETSC_FALSE and use a Fortran array v(bs,bs,nnz) in which the first index is over rows within a
2802: block column and the second index is over columns within a block.
2804: Though this routine has Preallocation() in the name it also sets the exact nonzero locations of the matrix entries and usually the numerical values as well
2806: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIBAIJSetPreallocation(), MatCreateAIJ(), MPIAIJ, MatCreateMPIBAIJWithArrays(), MPIBAIJ
2807: @*/
2808: PetscErrorCode MatMPIBAIJSetPreallocationCSR(Mat B,PetscInt bs,const PetscInt i[],const PetscInt j[], const PetscScalar v[])
2809: {
2816: PetscTryMethod(B,"MatMPIBAIJSetPreallocationCSR_C",(Mat,PetscInt,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,bs,i,j,v));
2817: return(0);
2818: }
2820: PetscErrorCode MatMPIBAIJSetPreallocation_MPIBAIJ(Mat B,PetscInt bs,PetscInt d_nz,const PetscInt *d_nnz,PetscInt o_nz,const PetscInt *o_nnz)
2821: {
2822: Mat_MPIBAIJ *b;
2824: PetscInt i;
2825: PetscMPIInt size;
2828: MatSetBlockSize(B,PetscAbs(bs));
2829: PetscLayoutSetUp(B->rmap);
2830: PetscLayoutSetUp(B->cmap);
2831: PetscLayoutGetBlockSize(B->rmap,&bs);
2833: if (d_nnz) {
2834: for (i=0; i<B->rmap->n/bs; i++) {
2835: if (d_nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"d_nnz cannot be less than -1: local row %D value %D",i,d_nnz[i]);
2836: }
2837: }
2838: if (o_nnz) {
2839: for (i=0; i<B->rmap->n/bs; i++) {
2840: if (o_nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"o_nnz cannot be less than -1: local row %D value %D",i,o_nnz[i]);
2841: }
2842: }
2844: b = (Mat_MPIBAIJ*)B->data;
2845: b->bs2 = bs*bs;
2846: b->mbs = B->rmap->n/bs;
2847: b->nbs = B->cmap->n/bs;
2848: b->Mbs = B->rmap->N/bs;
2849: b->Nbs = B->cmap->N/bs;
2851: for (i=0; i<=b->size; i++) {
2852: b->rangebs[i] = B->rmap->range[i]/bs;
2853: }
2854: b->rstartbs = B->rmap->rstart/bs;
2855: b->rendbs = B->rmap->rend/bs;
2856: b->cstartbs = B->cmap->rstart/bs;
2857: b->cendbs = B->cmap->rend/bs;
2859: #if defined(PETSC_USE_CTABLE)
2860: PetscTableDestroy(&b->colmap);
2861: #else
2862: PetscFree(b->colmap);
2863: #endif
2864: PetscFree(b->garray);
2865: VecDestroy(&b->lvec);
2866: VecScatterDestroy(&b->Mvctx);
2868: /* Because the B will have been resized we simply destroy it and create a new one each time */
2869: MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);
2870: MatDestroy(&b->B);
2871: MatCreate(PETSC_COMM_SELF,&b->B);
2872: MatSetSizes(b->B,B->rmap->n,size > 1 ? B->cmap->N : 0,B->rmap->n,size > 1 ? B->cmap->N : 0);
2873: MatSetType(b->B,MATSEQBAIJ);
2874: PetscLogObjectParent((PetscObject)B,(PetscObject)b->B);
2876: if (!B->preallocated) {
2877: MatCreate(PETSC_COMM_SELF,&b->A);
2878: MatSetSizes(b->A,B->rmap->n,B->cmap->n,B->rmap->n,B->cmap->n);
2879: MatSetType(b->A,MATSEQBAIJ);
2880: PetscLogObjectParent((PetscObject)B,(PetscObject)b->A);
2881: MatStashCreate_Private(PetscObjectComm((PetscObject)B),bs,&B->bstash);
2882: }
2884: MatSeqBAIJSetPreallocation(b->A,bs,d_nz,d_nnz);
2885: MatSeqBAIJSetPreallocation(b->B,bs,o_nz,o_nnz);
2886: B->preallocated = PETSC_TRUE;
2887: B->was_assembled = PETSC_FALSE;
2888: B->assembled = PETSC_FALSE;
2889: return(0);
2890: }
2892: extern PetscErrorCode MatDiagonalScaleLocal_MPIBAIJ(Mat,Vec);
2893: extern PetscErrorCode MatSetHashTableFactor_MPIBAIJ(Mat,PetscReal);
2895: PETSC_INTERN PetscErrorCode MatConvert_MPIBAIJ_MPIAdj(Mat B, MatType newtype,MatReuse reuse,Mat *adj)
2896: {
2897: Mat_MPIBAIJ *b = (Mat_MPIBAIJ*)B->data;
2899: Mat_SeqBAIJ *d = (Mat_SeqBAIJ*) b->A->data,*o = (Mat_SeqBAIJ*) b->B->data;
2900: PetscInt M = B->rmap->n/B->rmap->bs,i,*ii,*jj,cnt,j,k,rstart = B->rmap->rstart/B->rmap->bs;
2901: const PetscInt *id = d->i, *jd = d->j, *io = o->i, *jo = o->j, *garray = b->garray;
2904: PetscMalloc1(M+1,&ii);
2905: ii[0] = 0;
2906: for (i=0; i<M; i++) {
2907: if ((id[i+1] - id[i]) < 0) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Indices wrong %D %D %D",i,id[i],id[i+1]);
2908: if ((io[i+1] - io[i]) < 0) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Indices wrong %D %D %D",i,io[i],io[i+1]);
2909: ii[i+1] = ii[i] + id[i+1] - id[i] + io[i+1] - io[i];
2910: /* remove one from count of matrix has diagonal */
2911: for (j=id[i]; j<id[i+1]; j++) {
2912: if (jd[j] == i) {ii[i+1]--;break;}
2913: }
2914: }
2915: PetscMalloc1(ii[M],&jj);
2916: cnt = 0;
2917: for (i=0; i<M; i++) {
2918: for (j=io[i]; j<io[i+1]; j++) {
2919: if (garray[jo[j]] > rstart) break;
2920: jj[cnt++] = garray[jo[j]];
2921: }
2922: for (k=id[i]; k<id[i+1]; k++) {
2923: if (jd[k] != i) {
2924: jj[cnt++] = rstart + jd[k];
2925: }
2926: }
2927: for (; j<io[i+1]; j++) {
2928: jj[cnt++] = garray[jo[j]];
2929: }
2930: }
2931: MatCreateMPIAdj(PetscObjectComm((PetscObject)B),M,B->cmap->N/B->rmap->bs,ii,jj,NULL,adj);
2932: return(0);
2933: }
2935: #include <../src/mat/impls/aij/mpi/mpiaij.h>
2937: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqAIJ(Mat,MatType,MatReuse,Mat*);
2939: PETSC_INTERN PetscErrorCode MatConvert_MPIBAIJ_MPIAIJ(Mat A,MatType newtype,MatReuse reuse,Mat *newmat)
2940: {
2942: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
2943: Mat B;
2944: Mat_MPIAIJ *b;
2947: if (!A->assembled) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Matrix must be assembled");
2949: if (reuse == MAT_REUSE_MATRIX) {
2950: B = *newmat;
2951: } else {
2952: MatCreate(PetscObjectComm((PetscObject)A),&B);
2953: MatSetType(B,MATMPIAIJ);
2954: MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);
2955: MatSetBlockSizes(B,A->rmap->bs,A->cmap->bs);
2956: MatSeqAIJSetPreallocation(B,0,NULL);
2957: MatMPIAIJSetPreallocation(B,0,NULL,0,NULL);
2958: }
2959: b = (Mat_MPIAIJ*) B->data;
2961: if (reuse == MAT_REUSE_MATRIX) {
2962: MatConvert_SeqBAIJ_SeqAIJ(a->A, MATSEQAIJ, MAT_REUSE_MATRIX, &b->A);
2963: MatConvert_SeqBAIJ_SeqAIJ(a->B, MATSEQAIJ, MAT_REUSE_MATRIX, &b->B);
2964: } else {
2965: MatDestroy(&b->A);
2966: MatDestroy(&b->B);
2967: MatDisAssemble_MPIBAIJ(A);
2968: MatConvert_SeqBAIJ_SeqAIJ(a->A, MATSEQAIJ, MAT_INITIAL_MATRIX, &b->A);
2969: MatConvert_SeqBAIJ_SeqAIJ(a->B, MATSEQAIJ, MAT_INITIAL_MATRIX, &b->B);
2970: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
2971: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
2972: }
2973: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2974: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2976: if (reuse == MAT_INPLACE_MATRIX) {
2977: MatHeaderReplace(A,&B);
2978: } else {
2979: *newmat = B;
2980: }
2981: return(0);
2982: }
2984: /*MC
2985: MATMPIBAIJ - MATMPIBAIJ = "mpibaij" - A matrix type to be used for distributed block sparse matrices.
2987: Options Database Keys:
2988: + -mat_type mpibaij - sets the matrix type to "mpibaij" during a call to MatSetFromOptions()
2989: . -mat_block_size <bs> - set the blocksize used to store the matrix
2990: - -mat_use_hash_table <fact>
2992: Level: beginner
2994: Notes:
2995: MatSetOptions(,MAT_STRUCTURE_ONLY,PETSC_TRUE) may be called for this matrix type. In this no
2996: space is allocated for the nonzero entries and any entries passed with MatSetValues() are ignored
2998: .seealso: MatCreateMPIBAIJ
2999: M*/
3001: PETSC_INTERN PetscErrorCode MatConvert_MPIBAIJ_MPIBSTRM(Mat,MatType,MatReuse,Mat*);
3002: PETSC_INTERN PetscErrorCode MatPtAP_IS_XAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
3004: PETSC_EXTERN PetscErrorCode MatCreate_MPIBAIJ(Mat B)
3005: {
3006: Mat_MPIBAIJ *b;
3008: PetscBool flg = PETSC_FALSE;
3011: PetscNewLog(B,&b);
3012: B->data = (void*)b;
3014: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
3015: B->assembled = PETSC_FALSE;
3017: B->insertmode = NOT_SET_VALUES;
3018: MPI_Comm_rank(PetscObjectComm((PetscObject)B),&b->rank);
3019: MPI_Comm_size(PetscObjectComm((PetscObject)B),&b->size);
3021: /* build local table of row and column ownerships */
3022: PetscMalloc1(b->size+1,&b->rangebs);
3024: /* build cache for off array entries formed */
3025: MatStashCreate_Private(PetscObjectComm((PetscObject)B),1,&B->stash);
3027: b->donotstash = PETSC_FALSE;
3028: b->colmap = NULL;
3029: b->garray = NULL;
3030: b->roworiented = PETSC_TRUE;
3032: /* stuff used in block assembly */
3033: b->barray = 0;
3035: /* stuff used for matrix vector multiply */
3036: b->lvec = 0;
3037: b->Mvctx = 0;
3039: /* stuff for MatGetRow() */
3040: b->rowindices = 0;
3041: b->rowvalues = 0;
3042: b->getrowactive = PETSC_FALSE;
3044: /* hash table stuff */
3045: b->ht = 0;
3046: b->hd = 0;
3047: b->ht_size = 0;
3048: b->ht_flag = PETSC_FALSE;
3049: b->ht_fact = 0;
3050: b->ht_total_ct = 0;
3051: b->ht_insert_ct = 0;
3053: /* stuff for MatCreateSubMatrices_MPIBAIJ_local() */
3054: b->ijonly = PETSC_FALSE;
3057: PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpibaij_mpiadj_C",MatConvert_MPIBAIJ_MPIAdj);
3058: PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpibaij_mpiaij_C",MatConvert_MPIBAIJ_MPIAIJ);
3059: PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpibaij_mpisbaij_C",MatConvert_MPIBAIJ_MPISBAIJ);
3060: #if defined(PETSC_HAVE_HYPRE)
3061: PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpibaij_hypre_C",MatConvert_AIJ_HYPRE);
3062: #endif
3063: PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_MPIBAIJ);
3064: PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_MPIBAIJ);
3065: PetscObjectComposeFunction((PetscObject)B,"MatMPIBAIJSetPreallocation_C",MatMPIBAIJSetPreallocation_MPIBAIJ);
3066: PetscObjectComposeFunction((PetscObject)B,"MatMPIBAIJSetPreallocationCSR_C",MatMPIBAIJSetPreallocationCSR_MPIBAIJ);
3067: PetscObjectComposeFunction((PetscObject)B,"MatDiagonalScaleLocal_C",MatDiagonalScaleLocal_MPIBAIJ);
3068: PetscObjectComposeFunction((PetscObject)B,"MatSetHashTableFactor_C",MatSetHashTableFactor_MPIBAIJ);
3069: PetscObjectComposeFunction((PetscObject)B,"MatPtAP_is_mpibaij_C",MatPtAP_IS_XAIJ);
3070: PetscObjectComposeFunction((PetscObject)B,"MatConvert_mpibaij_is_C",MatConvert_XAIJ_IS);
3071: PetscObjectChangeTypeName((PetscObject)B,MATMPIBAIJ);
3073: PetscOptionsBegin(PetscObjectComm((PetscObject)B),NULL,"Options for loading MPIBAIJ matrix 1","Mat");
3074: PetscOptionsName("-mat_use_hash_table","Use hash table to save time in constructing matrix","MatSetOption",&flg);
3075: if (flg) {
3076: PetscReal fact = 1.39;
3077: MatSetOption(B,MAT_USE_HASH_TABLE,PETSC_TRUE);
3078: PetscOptionsReal("-mat_use_hash_table","Use hash table factor","MatMPIBAIJSetHashTableFactor",fact,&fact,NULL);
3079: if (fact <= 1.0) fact = 1.39;
3080: MatMPIBAIJSetHashTableFactor(B,fact);
3081: PetscInfo1(B,"Hash table Factor used %5.2f\n",fact);
3082: }
3083: PetscOptionsEnd();
3084: return(0);
3085: }
3087: /*MC
3088: MATBAIJ - MATBAIJ = "baij" - A matrix type to be used for block sparse matrices.
3090: This matrix type is identical to MATSEQBAIJ when constructed with a single process communicator,
3091: and MATMPIBAIJ otherwise.
3093: Options Database Keys:
3094: . -mat_type baij - sets the matrix type to "baij" during a call to MatSetFromOptions()
3096: Level: beginner
3098: .seealso: MatCreateBAIJ(),MATSEQBAIJ,MATMPIBAIJ, MatMPIBAIJSetPreallocation(), MatMPIBAIJSetPreallocationCSR()
3099: M*/
3101: /*@C
3102: MatMPIBAIJSetPreallocation - Allocates memory for a sparse parallel matrix in block AIJ format
3103: (block compressed row). For good matrix assembly performance
3104: the user should preallocate the matrix storage by setting the parameters
3105: d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately,
3106: performance can be increased by more than a factor of 50.
3108: Collective on Mat
3110: Input Parameters:
3111: + B - the matrix
3112: . bs - size of block, the blocks are ALWAYS square. One can use MatSetBlockSizes() to set a different row and column blocksize but the row
3113: blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with MatCreateVecs()
3114: . d_nz - number of block nonzeros per block row in diagonal portion of local
3115: submatrix (same for all local rows)
3116: . d_nnz - array containing the number of block nonzeros in the various block rows
3117: of the in diagonal portion of the local (possibly different for each block
3118: row) or NULL. If you plan to factor the matrix you must leave room for the diagonal entry and
3119: set it even if it is zero.
3120: . o_nz - number of block nonzeros per block row in the off-diagonal portion of local
3121: submatrix (same for all local rows).
3122: - o_nnz - array containing the number of nonzeros in the various block rows of the
3123: off-diagonal portion of the local submatrix (possibly different for
3124: each block row) or NULL.
3126: If the *_nnz parameter is given then the *_nz parameter is ignored
3128: Options Database Keys:
3129: + -mat_block_size - size of the blocks to use
3130: - -mat_use_hash_table <fact>
3132: Notes:
3133: If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one processor
3134: than it must be used on all processors that share the object for that argument.
3136: Storage Information:
3137: For a square global matrix we define each processor's diagonal portion
3138: to be its local rows and the corresponding columns (a square submatrix);
3139: each processor's off-diagonal portion encompasses the remainder of the
3140: local matrix (a rectangular submatrix).
3142: The user can specify preallocated storage for the diagonal part of
3143: the local submatrix with either d_nz or d_nnz (not both). Set
3144: d_nz=PETSC_DEFAULT and d_nnz=NULL for PETSc to control dynamic
3145: memory allocation. Likewise, specify preallocated storage for the
3146: off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
3148: Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
3149: the figure below we depict these three local rows and all columns (0-11).
3151: .vb
3152: 0 1 2 3 4 5 6 7 8 9 10 11
3153: --------------------------
3154: row 3 |o o o d d d o o o o o o
3155: row 4 |o o o d d d o o o o o o
3156: row 5 |o o o d d d o o o o o o
3157: --------------------------
3158: .ve
3160: Thus, any entries in the d locations are stored in the d (diagonal)
3161: submatrix, and any entries in the o locations are stored in the
3162: o (off-diagonal) submatrix. Note that the d and the o submatrices are
3163: stored simply in the MATSEQBAIJ format for compressed row storage.
3165: Now d_nz should indicate the number of block nonzeros per row in the d matrix,
3166: and o_nz should indicate the number of block nonzeros per row in the o matrix.
3167: In general, for PDE problems in which most nonzeros are near the diagonal,
3168: one expects d_nz >> o_nz. For large problems you MUST preallocate memory
3169: or you will get TERRIBLE performance; see the users' manual chapter on
3170: matrices.
3172: You can call MatGetInfo() to get information on how effective the preallocation was;
3173: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3174: You can also run with the option -info and look for messages with the string
3175: malloc in them to see if additional memory allocation was needed.
3177: Level: intermediate
3179: .seealso: MatCreate(), MatCreateSeqBAIJ(), MatSetValues(), MatCreateBAIJ(), MatMPIBAIJSetPreallocationCSR(), PetscSplitOwnership()
3180: @*/
3181: PetscErrorCode MatMPIBAIJSetPreallocation(Mat B,PetscInt bs,PetscInt d_nz,const PetscInt d_nnz[],PetscInt o_nz,const PetscInt o_nnz[])
3182: {
3189: PetscTryMethod(B,"MatMPIBAIJSetPreallocation_C",(Mat,PetscInt,PetscInt,const PetscInt[],PetscInt,const PetscInt[]),(B,bs,d_nz,d_nnz,o_nz,o_nnz));
3190: return(0);
3191: }
3193: /*@C
3194: MatCreateBAIJ - Creates a sparse parallel matrix in block AIJ format
3195: (block compressed row). For good matrix assembly performance
3196: the user should preallocate the matrix storage by setting the parameters
3197: d_nz (or d_nnz) and o_nz (or o_nnz). By setting these parameters accurately,
3198: performance can be increased by more than a factor of 50.
3200: Collective
3202: Input Parameters:
3203: + comm - MPI communicator
3204: . bs - size of block, the blocks are ALWAYS square. One can use MatSetBlockSizes() to set a different row and column blocksize but the row
3205: blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with MatCreateVecs()
3206: . m - number of local rows (or PETSC_DECIDE to have calculated if M is given)
3207: This value should be the same as the local size used in creating the
3208: y vector for the matrix-vector product y = Ax.
3209: . n - number of local columns (or PETSC_DECIDE to have calculated if N is given)
3210: This value should be the same as the local size used in creating the
3211: x vector for the matrix-vector product y = Ax.
3212: . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given)
3213: . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given)
3214: . d_nz - number of nonzero blocks per block row in diagonal portion of local
3215: submatrix (same for all local rows)
3216: . d_nnz - array containing the number of nonzero blocks in the various block rows
3217: of the in diagonal portion of the local (possibly different for each block
3218: row) or NULL. If you plan to factor the matrix you must leave room for the diagonal entry
3219: and set it even if it is zero.
3220: . o_nz - number of nonzero blocks per block row in the off-diagonal portion of local
3221: submatrix (same for all local rows).
3222: - o_nnz - array containing the number of nonzero blocks in the various block rows of the
3223: off-diagonal portion of the local submatrix (possibly different for
3224: each block row) or NULL.
3226: Output Parameter:
3227: . A - the matrix
3229: Options Database Keys:
3230: + -mat_block_size - size of the blocks to use
3231: - -mat_use_hash_table <fact>
3233: It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
3234: MatXXXXSetPreallocation() paradigm instead of this routine directly.
3235: [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
3237: Notes:
3238: If the *_nnz parameter is given then the *_nz parameter is ignored
3240: A nonzero block is any block that as 1 or more nonzeros in it
3242: The user MUST specify either the local or global matrix dimensions
3243: (possibly both).
3245: If PETSC_DECIDE or PETSC_DETERMINE is used for a particular argument on one processor
3246: than it must be used on all processors that share the object for that argument.
3248: Storage Information:
3249: For a square global matrix we define each processor's diagonal portion
3250: to be its local rows and the corresponding columns (a square submatrix);
3251: each processor's off-diagonal portion encompasses the remainder of the
3252: local matrix (a rectangular submatrix).
3254: The user can specify preallocated storage for the diagonal part of
3255: the local submatrix with either d_nz or d_nnz (not both). Set
3256: d_nz=PETSC_DEFAULT and d_nnz=NULL for PETSc to control dynamic
3257: memory allocation. Likewise, specify preallocated storage for the
3258: off-diagonal part of the local submatrix with o_nz or o_nnz (not both).
3260: Consider a processor that owns rows 3, 4 and 5 of a parallel matrix. In
3261: the figure below we depict these three local rows and all columns (0-11).
3263: .vb
3264: 0 1 2 3 4 5 6 7 8 9 10 11
3265: --------------------------
3266: row 3 |o o o d d d o o o o o o
3267: row 4 |o o o d d d o o o o o o
3268: row 5 |o o o d d d o o o o o o
3269: --------------------------
3270: .ve
3272: Thus, any entries in the d locations are stored in the d (diagonal)
3273: submatrix, and any entries in the o locations are stored in the
3274: o (off-diagonal) submatrix. Note that the d and the o submatrices are
3275: stored simply in the MATSEQBAIJ format for compressed row storage.
3277: Now d_nz should indicate the number of block nonzeros per row in the d matrix,
3278: and o_nz should indicate the number of block nonzeros per row in the o matrix.
3279: In general, for PDE problems in which most nonzeros are near the diagonal,
3280: one expects d_nz >> o_nz. For large problems you MUST preallocate memory
3281: or you will get TERRIBLE performance; see the users' manual chapter on
3282: matrices.
3284: Level: intermediate
3286: .seealso: MatCreate(), MatCreateSeqBAIJ(), MatSetValues(), MatCreateBAIJ(), MatMPIBAIJSetPreallocation(), MatMPIBAIJSetPreallocationCSR()
3287: @*/
3288: PetscErrorCode MatCreateBAIJ(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)
3289: {
3291: PetscMPIInt size;
3294: MatCreate(comm,A);
3295: MatSetSizes(*A,m,n,M,N);
3296: MPI_Comm_size(comm,&size);
3297: if (size > 1) {
3298: MatSetType(*A,MATMPIBAIJ);
3299: MatMPIBAIJSetPreallocation(*A,bs,d_nz,d_nnz,o_nz,o_nnz);
3300: } else {
3301: MatSetType(*A,MATSEQBAIJ);
3302: MatSeqBAIJSetPreallocation(*A,bs,d_nz,d_nnz);
3303: }
3304: return(0);
3305: }
3307: static PetscErrorCode MatDuplicate_MPIBAIJ(Mat matin,MatDuplicateOption cpvalues,Mat *newmat)
3308: {
3309: Mat mat;
3310: Mat_MPIBAIJ *a,*oldmat = (Mat_MPIBAIJ*)matin->data;
3312: PetscInt len=0;
3315: *newmat = 0;
3316: MatCreate(PetscObjectComm((PetscObject)matin),&mat);
3317: MatSetSizes(mat,matin->rmap->n,matin->cmap->n,matin->rmap->N,matin->cmap->N);
3318: MatSetType(mat,((PetscObject)matin)->type_name);
3320: mat->factortype = matin->factortype;
3321: mat->preallocated = PETSC_TRUE;
3322: mat->assembled = PETSC_TRUE;
3323: mat->insertmode = NOT_SET_VALUES;
3325: a = (Mat_MPIBAIJ*)mat->data;
3326: mat->rmap->bs = matin->rmap->bs;
3327: a->bs2 = oldmat->bs2;
3328: a->mbs = oldmat->mbs;
3329: a->nbs = oldmat->nbs;
3330: a->Mbs = oldmat->Mbs;
3331: a->Nbs = oldmat->Nbs;
3333: PetscLayoutReference(matin->rmap,&mat->rmap);
3334: PetscLayoutReference(matin->cmap,&mat->cmap);
3336: a->size = oldmat->size;
3337: a->rank = oldmat->rank;
3338: a->donotstash = oldmat->donotstash;
3339: a->roworiented = oldmat->roworiented;
3340: a->rowindices = 0;
3341: a->rowvalues = 0;
3342: a->getrowactive = PETSC_FALSE;
3343: a->barray = 0;
3344: a->rstartbs = oldmat->rstartbs;
3345: a->rendbs = oldmat->rendbs;
3346: a->cstartbs = oldmat->cstartbs;
3347: a->cendbs = oldmat->cendbs;
3349: /* hash table stuff */
3350: a->ht = 0;
3351: a->hd = 0;
3352: a->ht_size = 0;
3353: a->ht_flag = oldmat->ht_flag;
3354: a->ht_fact = oldmat->ht_fact;
3355: a->ht_total_ct = 0;
3356: a->ht_insert_ct = 0;
3358: PetscArraycpy(a->rangebs,oldmat->rangebs,a->size+1);
3359: if (oldmat->colmap) {
3360: #if defined(PETSC_USE_CTABLE)
3361: PetscTableCreateCopy(oldmat->colmap,&a->colmap);
3362: #else
3363: PetscMalloc1(a->Nbs,&a->colmap);
3364: PetscLogObjectMemory((PetscObject)mat,(a->Nbs)*sizeof(PetscInt));
3365: PetscArraycpy(a->colmap,oldmat->colmap,a->Nbs);
3366: #endif
3367: } else a->colmap = 0;
3369: if (oldmat->garray && (len = ((Mat_SeqBAIJ*)(oldmat->B->data))->nbs)) {
3370: PetscMalloc1(len,&a->garray);
3371: PetscLogObjectMemory((PetscObject)mat,len*sizeof(PetscInt));
3372: PetscArraycpy(a->garray,oldmat->garray,len);
3373: } else a->garray = 0;
3375: MatStashCreate_Private(PetscObjectComm((PetscObject)matin),matin->rmap->bs,&mat->bstash);
3376: VecDuplicate(oldmat->lvec,&a->lvec);
3377: PetscLogObjectParent((PetscObject)mat,(PetscObject)a->lvec);
3378: VecScatterCopy(oldmat->Mvctx,&a->Mvctx);
3379: PetscLogObjectParent((PetscObject)mat,(PetscObject)a->Mvctx);
3381: MatDuplicate(oldmat->A,cpvalues,&a->A);
3382: PetscLogObjectParent((PetscObject)mat,(PetscObject)a->A);
3383: MatDuplicate(oldmat->B,cpvalues,&a->B);
3384: PetscLogObjectParent((PetscObject)mat,(PetscObject)a->B);
3385: PetscFunctionListDuplicate(((PetscObject)matin)->qlist,&((PetscObject)mat)->qlist);
3386: *newmat = mat;
3387: return(0);
3388: }
3390: PetscErrorCode MatLoad_MPIBAIJ(Mat newmat,PetscViewer viewer)
3391: {
3393: int fd;
3394: PetscInt i,nz,j,rstart,rend;
3395: PetscScalar *vals,*buf;
3396: MPI_Comm comm;
3397: MPI_Status status;
3398: PetscMPIInt rank,size,maxnz;
3399: PetscInt header[4],*rowlengths = 0,M,N,m,*rowners,*cols;
3400: PetscInt *locrowlens = NULL,*procsnz = NULL,*browners = NULL;
3401: PetscInt jj,*mycols,*ibuf,bs = newmat->rmap->bs,Mbs,mbs,extra_rows,mmax;
3402: PetscMPIInt tag = ((PetscObject)viewer)->tag;
3403: PetscInt *dlens = NULL,*odlens = NULL,*mask = NULL,*masked1 = NULL,*masked2 = NULL,rowcount,odcount;
3404: PetscInt dcount,kmax,k,nzcount,tmp,mend;
3405: PetscBool isbinary;
3408: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
3409: if (!isbinary) SETERRQ2(PetscObjectComm((PetscObject)newmat),PETSC_ERR_SUP,"Viewer type %s not yet supported for reading %s matrices",((PetscObject)viewer)->type_name,((PetscObject)newmat)->type_name);
3411: /* force binary viewer to load .info file if it has not yet done so */
3412: PetscViewerSetUp(viewer);
3413: PetscObjectGetComm((PetscObject)viewer,&comm);
3414: PetscOptionsBegin(comm,NULL,"Options for loading MPIBAIJ matrix 2","Mat");
3415: PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);
3416: PetscOptionsEnd();
3417: if (bs < 0) bs = 1;
3419: MPI_Comm_size(comm,&size);
3420: MPI_Comm_rank(comm,&rank);
3421: PetscViewerBinaryGetDescriptor(viewer,&fd);
3422: if (!rank) {
3423: PetscBinaryRead(fd,(char*)header,4,NULL,PETSC_INT);
3424: if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object");
3425: if (header[3] < 0) SETERRQ(PetscObjectComm((PetscObject)newmat),PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk, cannot load as MPIAIJ");
3426: }
3427: MPI_Bcast(header+1,3,MPIU_INT,0,comm);
3428: M = header[1]; N = header[2];
3430: /* If global sizes are set, check if they are consistent with that given in the file */
3431: if (newmat->rmap->N >= 0 && newmat->rmap->N != M) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent # of rows:Matrix in file has (%D) and input matrix has (%D)",newmat->rmap->N,M);
3432: if (newmat->cmap->N >= 0 && newmat->cmap->N != N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent # of cols:Matrix in file has (%D) and input matrix has (%D)",newmat->cmap->N,N);
3434: if (M != N) SETERRQ(PetscObjectComm((PetscObject)viewer),PETSC_ERR_SUP,"Can only do square matrices");
3436: /*
3437: This code adds extra rows to make sure the number of rows is
3438: divisible by the blocksize
3439: */
3440: Mbs = M/bs;
3441: extra_rows = bs - M + bs*Mbs;
3442: if (extra_rows == bs) extra_rows = 0;
3443: else Mbs++;
3444: if (extra_rows && !rank) {
3445: PetscInfo(viewer,"Padding loaded matrix to match blocksize\n");
3446: }
3448: /* determine ownership of all rows */
3449: if (newmat->rmap->n < 0) { /* PETSC_DECIDE */
3450: mbs = Mbs/size + ((Mbs % size) > rank);
3451: m = mbs*bs;
3452: } else { /* User set */
3453: m = newmat->rmap->n;
3454: mbs = m/bs;
3455: }
3456: PetscMalloc2(size+1,&rowners,size+1,&browners);
3457: MPI_Allgather(&mbs,1,MPIU_INT,rowners+1,1,MPIU_INT,comm);
3459: /* process 0 needs enough room for process with most rows */
3460: if (!rank) {
3461: mmax = rowners[1];
3462: for (i=2; i<=size; i++) {
3463: mmax = PetscMax(mmax,rowners[i]);
3464: }
3465: mmax*=bs;
3466: } else mmax = -1; /* unused, but compiler warns anyway */
3468: rowners[0] = 0;
3469: for (i=2; i<=size; i++) rowners[i] += rowners[i-1];
3470: for (i=0; i<=size; i++) browners[i] = rowners[i]*bs;
3471: rstart = rowners[rank];
3472: rend = rowners[rank+1];
3474: /* distribute row lengths to all processors */
3475: PetscMalloc1(m,&locrowlens);
3476: if (!rank) {
3477: mend = m;
3478: if (size == 1) mend = mend - extra_rows;
3479: PetscBinaryRead(fd,locrowlens,mend,NULL,PETSC_INT);
3480: for (j=mend; j<m; j++) locrowlens[j] = 1;
3481: PetscMalloc1(mmax,&rowlengths);
3482: PetscMalloc1(size,&procsnz);
3483: procsnz[0] = 0;
3484: for (j=0; j<m; j++) {
3485: procsnz[0] += locrowlens[j];
3486: }
3487: for (i=1; i<size; i++) {
3488: mend = browners[i+1] - browners[i];
3489: if (i == size-1) mend = mend - extra_rows;
3490: PetscBinaryRead(fd,rowlengths,mend,NULL,PETSC_INT);
3491: for (j=mend; j<browners[i+1] - browners[i]; j++) rowlengths[j] = 1;
3492: /* calculate the number of nonzeros on each processor */
3493: procsnz[i] = 0;
3494: for (j=0; j<browners[i+1]-browners[i]; j++) {
3495: procsnz[i] += rowlengths[j];
3496: }
3497: MPI_Send(rowlengths,browners[i+1]-browners[i],MPIU_INT,i,tag,comm);
3498: }
3499: PetscFree(rowlengths);
3500: } else {
3501: MPI_Recv(locrowlens,m,MPIU_INT,0,tag,comm,&status);
3502: }
3504: if (!rank) {
3505: /* determine max buffer needed and allocate it */
3506: maxnz = procsnz[0];
3507: for (i=1; i<size; i++) {
3508: maxnz = PetscMax(maxnz,procsnz[i]);
3509: }
3510: PetscMalloc1(maxnz,&cols);
3512: /* read in my part of the matrix column indices */
3513: nz = procsnz[0];
3514: PetscMalloc1(nz+1,&ibuf);
3515: mycols = ibuf;
3516: if (size == 1) nz -= extra_rows;
3517: PetscBinaryRead(fd,mycols,nz,NULL,PETSC_INT);
3518: if (size == 1) {
3519: for (i=0; i< extra_rows; i++) mycols[nz+i] = M+i;
3520: }
3522: /* read in every ones (except the last) and ship off */
3523: for (i=1; i<size-1; i++) {
3524: nz = procsnz[i];
3525: PetscBinaryRead(fd,cols,nz,NULL,PETSC_INT);
3526: MPI_Send(cols,nz,MPIU_INT,i,tag,comm);
3527: }
3528: /* read in the stuff for the last proc */
3529: if (size != 1) {
3530: nz = procsnz[size-1] - extra_rows; /* the extra rows are not on the disk */
3531: PetscBinaryRead(fd,cols,nz,NULL,PETSC_INT);
3532: for (i=0; i<extra_rows; i++) cols[nz+i] = M+i;
3533: MPI_Send(cols,nz+extra_rows,MPIU_INT,size-1,tag,comm);
3534: }
3535: PetscFree(cols);
3536: } else {
3537: /* determine buffer space needed for message */
3538: nz = 0;
3539: for (i=0; i<m; i++) {
3540: nz += locrowlens[i];
3541: }
3542: PetscMalloc1(nz+1,&ibuf);
3543: mycols = ibuf;
3544: /* receive message of column indices*/
3545: MPI_Recv(mycols,nz,MPIU_INT,0,tag,comm,&status);
3546: MPI_Get_count(&status,MPIU_INT,&maxnz);
3547: if (maxnz != nz) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"something is wrong with file");
3548: }
3550: /* loop over local rows, determining number of off diagonal entries */
3551: PetscMalloc2(rend-rstart,&dlens,rend-rstart,&odlens);
3552: PetscCalloc3(Mbs,&mask,Mbs,&masked1,Mbs,&masked2);
3553: rowcount = 0; nzcount = 0;
3554: for (i=0; i<mbs; i++) {
3555: dcount = 0;
3556: odcount = 0;
3557: for (j=0; j<bs; j++) {
3558: kmax = locrowlens[rowcount];
3559: for (k=0; k<kmax; k++) {
3560: tmp = mycols[nzcount++]/bs;
3561: if (!mask[tmp]) {
3562: mask[tmp] = 1;
3563: if (tmp < rstart || tmp >= rend) masked2[odcount++] = tmp;
3564: else masked1[dcount++] = tmp;
3565: }
3566: }
3567: rowcount++;
3568: }
3570: dlens[i] = dcount;
3571: odlens[i] = odcount;
3573: /* zero out the mask elements we set */
3574: for (j=0; j<dcount; j++) mask[masked1[j]] = 0;
3575: for (j=0; j<odcount; j++) mask[masked2[j]] = 0;
3576: }
3578: MatSetSizes(newmat,m,m,M+extra_rows,N+extra_rows);
3579: MatMPIBAIJSetPreallocation(newmat,bs,0,dlens,0,odlens);
3581: if (!rank) {
3582: PetscMalloc1(maxnz+1,&buf);
3583: /* read in my part of the matrix numerical values */
3584: nz = procsnz[0];
3585: vals = buf;
3586: mycols = ibuf;
3587: if (size == 1) nz -= extra_rows;
3588: PetscBinaryRead(fd,vals,nz,NULL,PETSC_SCALAR);
3589: if (size == 1) {
3590: for (i=0; i< extra_rows; i++) vals[nz+i] = 1.0;
3591: }
3593: /* insert into matrix */
3594: jj = rstart*bs;
3595: for (i=0; i<m; i++) {
3596: MatSetValues_MPIBAIJ(newmat,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);
3597: mycols += locrowlens[i];
3598: vals += locrowlens[i];
3599: jj++;
3600: }
3601: /* read in other processors (except the last one) and ship out */
3602: for (i=1; i<size-1; i++) {
3603: nz = procsnz[i];
3604: vals = buf;
3605: PetscBinaryRead(fd,vals,nz,NULL,PETSC_SCALAR);
3606: MPIULong_Send(vals,nz,MPIU_SCALAR,i,((PetscObject)newmat)->tag,comm);
3607: }
3608: /* the last proc */
3609: if (size != 1) {
3610: nz = procsnz[i] - extra_rows;
3611: vals = buf;
3612: PetscBinaryRead(fd,vals,nz,NULL,PETSC_SCALAR);
3613: for (i=0; i<extra_rows; i++) vals[nz+i] = 1.0;
3614: MPIULong_Send(vals,nz+extra_rows,MPIU_SCALAR,size-1,((PetscObject)newmat)->tag,comm);
3615: }
3616: PetscFree(procsnz);
3617: } else {
3618: /* receive numeric values */
3619: PetscMalloc1(nz+1,&buf);
3621: /* receive message of values*/
3622: vals = buf;
3623: mycols = ibuf;
3624: MPIULong_Recv(vals,nz,MPIU_SCALAR,0,((PetscObject)newmat)->tag,comm);
3626: /* insert into matrix */
3627: jj = rstart*bs;
3628: for (i=0; i<m; i++) {
3629: MatSetValues_MPIBAIJ(newmat,1,&jj,locrowlens[i],mycols,vals,INSERT_VALUES);
3630: mycols += locrowlens[i];
3631: vals += locrowlens[i];
3632: jj++;
3633: }
3634: }
3635: PetscFree(locrowlens);
3636: PetscFree(buf);
3637: PetscFree(ibuf);
3638: PetscFree2(rowners,browners);
3639: PetscFree2(dlens,odlens);
3640: PetscFree3(mask,masked1,masked2);
3641: MatAssemblyBegin(newmat,MAT_FINAL_ASSEMBLY);
3642: MatAssemblyEnd(newmat,MAT_FINAL_ASSEMBLY);
3643: return(0);
3644: }
3646: /*@
3647: MatMPIBAIJSetHashTableFactor - Sets the factor required to compute the size of the HashTable.
3649: Input Parameters:
3650: + mat - the matrix
3651: - fact - factor
3653: Not Collective, each process can use a different factor
3655: Level: advanced
3657: Notes:
3658: This can also be set by the command line option: -mat_use_hash_table <fact>
3660: .seealso: MatSetOption()
3661: @*/
3662: PetscErrorCode MatMPIBAIJSetHashTableFactor(Mat mat,PetscReal fact)
3663: {
3667: PetscTryMethod(mat,"MatSetHashTableFactor_C",(Mat,PetscReal),(mat,fact));
3668: return(0);
3669: }
3671: PetscErrorCode MatSetHashTableFactor_MPIBAIJ(Mat mat,PetscReal fact)
3672: {
3673: Mat_MPIBAIJ *baij;
3676: baij = (Mat_MPIBAIJ*)mat->data;
3677: baij->ht_fact = fact;
3678: return(0);
3679: }
3681: PetscErrorCode MatMPIBAIJGetSeqBAIJ(Mat A,Mat *Ad,Mat *Ao,const PetscInt *colmap[])
3682: {
3683: Mat_MPIBAIJ *a = (Mat_MPIBAIJ*)A->data;
3684: PetscBool flg;
3688: PetscObjectTypeCompare((PetscObject)A,MATMPIBAIJ,&flg);
3689: if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"This function requires a MATMPIBAIJ matrix as input");
3690: if (Ad) *Ad = a->A;
3691: if (Ao) *Ao = a->B;
3692: if (colmap) *colmap = a->garray;
3693: return(0);
3694: }
3696: /*
3697: Special version for direct calls from Fortran (to eliminate two function call overheads
3698: */
3699: #if defined(PETSC_HAVE_FORTRAN_CAPS)
3700: #define matmpibaijsetvaluesblocked_ MATMPIBAIJSETVALUESBLOCKED
3701: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
3702: #define matmpibaijsetvaluesblocked_ matmpibaijsetvaluesblocked
3703: #endif
3705: /*@C
3706: MatMPIBAIJSetValuesBlocked - Direct Fortran call to replace call to MatSetValuesBlocked()
3708: Collective on Mat
3710: Input Parameters:
3711: + mat - the matrix
3712: . min - number of input rows
3713: . im - input rows
3714: . nin - number of input columns
3715: . in - input columns
3716: . v - numerical values input
3717: - addvin - INSERT_VALUES or ADD_VALUES
3719: Notes:
3720: This has a complete copy of MatSetValuesBlocked_MPIBAIJ() which is terrible code un-reuse.
3722: Level: advanced
3724: .seealso: MatSetValuesBlocked()
3725: @*/
3726: PetscErrorCode matmpibaijsetvaluesblocked_(Mat *matin,PetscInt *min,const PetscInt im[],PetscInt *nin,const PetscInt in[],const MatScalar v[],InsertMode *addvin)
3727: {
3728: /* convert input arguments to C version */
3729: Mat mat = *matin;
3730: PetscInt m = *min, n = *nin;
3731: InsertMode addv = *addvin;
3733: Mat_MPIBAIJ *baij = (Mat_MPIBAIJ*)mat->data;
3734: const MatScalar *value;
3735: MatScalar *barray = baij->barray;
3736: PetscBool roworiented = baij->roworiented;
3737: PetscErrorCode ierr;
3738: PetscInt i,j,ii,jj,row,col,rstart=baij->rstartbs;
3739: PetscInt rend=baij->rendbs,cstart=baij->cstartbs,stepval;
3740: PetscInt cend=baij->cendbs,bs=mat->rmap->bs,bs2=baij->bs2;
3743: /* tasks normally handled by MatSetValuesBlocked() */
3744: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
3745: #if defined(PETSC_USE_DEBUG)
3746: else if (mat->insertmode != addv) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot mix add values and insert values");
3747: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3748: #endif
3749: if (mat->assembled) {
3750: mat->was_assembled = PETSC_TRUE;
3751: mat->assembled = PETSC_FALSE;
3752: }
3753: PetscLogEventBegin(MAT_SetValues,mat,0,0,0);
3756: if (!barray) {
3757: PetscMalloc1(bs2,&barray);
3758: baij->barray = barray;
3759: }
3761: if (roworiented) stepval = (n-1)*bs;
3762: else stepval = (m-1)*bs;
3764: for (i=0; i<m; i++) {
3765: if (im[i] < 0) continue;
3766: #if defined(PETSC_USE_DEBUG)
3767: if (im[i] >= baij->Mbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large, row %D max %D",im[i],baij->Mbs-1);
3768: #endif
3769: if (im[i] >= rstart && im[i] < rend) {
3770: row = im[i] - rstart;
3771: for (j=0; j<n; j++) {
3772: /* If NumCol = 1 then a copy is not required */
3773: if ((roworiented) && (n == 1)) {
3774: barray = (MatScalar*)v + i*bs2;
3775: } else if ((!roworiented) && (m == 1)) {
3776: barray = (MatScalar*)v + j*bs2;
3777: } else { /* Here a copy is required */
3778: if (roworiented) {
3779: value = v + i*(stepval+bs)*bs + j*bs;
3780: } else {
3781: value = v + j*(stepval+bs)*bs + i*bs;
3782: }
3783: for (ii=0; ii<bs; ii++,value+=stepval) {
3784: for (jj=0; jj<bs; jj++) {
3785: *barray++ = *value++;
3786: }
3787: }
3788: barray -=bs2;
3789: }
3791: if (in[j] >= cstart && in[j] < cend) {
3792: col = in[j] - cstart;
3793: MatSetValuesBlocked_SeqBAIJ_Inlined(baij->A,row,col,barray,addv,im[i],in[j]);
3794: } else if (in[j] < 0) continue;
3795: #if defined(PETSC_USE_DEBUG)
3796: else if (in[j] >= baij->Nbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large, col %D max %D",in[j],baij->Nbs-1);
3797: #endif
3798: else {
3799: if (mat->was_assembled) {
3800: if (!baij->colmap) {
3801: MatCreateColmap_MPIBAIJ_Private(mat);
3802: }
3804: #if defined(PETSC_USE_DEBUG)
3805: #if defined(PETSC_USE_CTABLE)
3806: { PetscInt data;
3807: PetscTableFind(baij->colmap,in[j]+1,&data);
3808: if ((data - 1) % bs) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Incorrect colmap");
3809: }
3810: #else
3811: if ((baij->colmap[in[j]] - 1) % bs) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Incorrect colmap");
3812: #endif
3813: #endif
3814: #if defined(PETSC_USE_CTABLE)
3815: PetscTableFind(baij->colmap,in[j]+1,&col);
3816: col = (col - 1)/bs;
3817: #else
3818: col = (baij->colmap[in[j]] - 1)/bs;
3819: #endif
3820: if (col < 0 && !((Mat_SeqBAIJ*)(baij->A->data))->nonew) {
3821: MatDisAssemble_MPIBAIJ(mat);
3822: col = in[j];
3823: }
3824: } else col = in[j];
3825: MatSetValuesBlocked_SeqBAIJ_Inlined(baij->B,row,col,barray,addv,im[i],in[j]);
3826: }
3827: }
3828: } else {
3829: if (!baij->donotstash) {
3830: if (roworiented) {
3831: MatStashValuesRowBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
3832: } else {
3833: MatStashValuesColBlocked_Private(&mat->bstash,im[i],n,in,v,m,n,i);
3834: }
3835: }
3836: }
3837: }
3839: /* task normally handled by MatSetValuesBlocked() */
3840: PetscLogEventEnd(MAT_SetValues,mat,0,0,0);
3841: return(0);
3842: }
3844: /*@
3845: MatCreateMPIBAIJWithArrays - creates a MPI BAIJ matrix using arrays that contain in standard block
3846: CSR format the local rows.
3848: Collective
3850: Input Parameters:
3851: + comm - MPI communicator
3852: . bs - the block size, only a block size of 1 is supported
3853: . m - number of local rows (Cannot be PETSC_DECIDE)
3854: . n - This value should be the same as the local size used in creating the
3855: x vector for the matrix-vector product y = Ax. (or PETSC_DECIDE to have
3856: calculated if N is given) For square matrices n is almost always m.
3857: . M - number of global rows (or PETSC_DETERMINE to have calculated if m is given)
3858: . N - number of global columns (or PETSC_DETERMINE to have calculated if n is given)
3859: . i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of block elements in that rowth block row of the matrix
3860: . j - column indices
3861: - a - matrix values
3863: Output Parameter:
3864: . mat - the matrix
3866: Level: intermediate
3868: Notes:
3869: The i, j, and a arrays ARE copied by this routine into the internal format used by PETSc;
3870: thus you CANNOT change the matrix entries by changing the values of a[] after you have
3871: called this routine. Use MatCreateMPIAIJWithSplitArrays() to avoid needing to copy the arrays.
3873: The order of the entries in values is the same as the block compressed sparse row storage format; that is, it is
3874: the same as a three dimensional array in Fortran values(bs,bs,nnz) that contains the first column of the first
3875: block, followed by the second column of the first block etc etc. That is, the blocks are contiguous in memory
3876: with column-major ordering within blocks.
3878: The i and j indices are 0 based, and i indices are indices corresponding to the local j array.
3880: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatMPIAIJSetPreallocation(), MatMPIAIJSetPreallocationCSR(),
3881: MPIAIJ, MatCreateAIJ(), MatCreateMPIAIJWithSplitArrays()
3882: @*/
3883: PetscErrorCode MatCreateMPIBAIJWithArrays(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt M,PetscInt N,const PetscInt i[],const PetscInt j[],const PetscScalar a[],Mat *mat)
3884: {
3888: if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
3889: if (m < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"local number of rows (m) cannot be PETSC_DECIDE, or negative");
3890: MatCreate(comm,mat);
3891: MatSetSizes(*mat,m,n,M,N);
3892: MatSetType(*mat,MATMPIBAIJ);
3893: MatSetBlockSize(*mat,bs);
3894: MatSetUp(*mat);
3895: MatSetOption(*mat,MAT_ROW_ORIENTED,PETSC_FALSE);
3896: MatMPIBAIJSetPreallocationCSR(*mat,bs,i,j,a);
3897: MatSetOption(*mat,MAT_ROW_ORIENTED,PETSC_TRUE);
3898: return(0);
3899: }
3901: PetscErrorCode MatCreateMPIMatConcatenateSeqMat_MPIBAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat)
3902: {
3904: PetscInt m,N,i,rstart,nnz,Ii,bs,cbs;
3905: PetscInt *indx;
3906: PetscScalar *values;
3909: MatGetSize(inmat,&m,&N);
3910: if (scall == MAT_INITIAL_MATRIX) { /* symbolic phase */
3911: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)inmat->data;
3912: PetscInt *dnz,*onz,mbs,Nbs,nbs;
3913: PetscInt *bindx,rmax=a->rmax,j;
3914: PetscMPIInt rank,size;
3916: MatGetBlockSizes(inmat,&bs,&cbs);
3917: mbs = m/bs; Nbs = N/cbs;
3918: if (n == PETSC_DECIDE) {
3919: PetscSplitOwnershipBlock(comm,cbs,&n,&N);
3920: }
3921: nbs = n/cbs;
3923: PetscMalloc1(rmax,&bindx);
3924: MatPreallocateInitialize(comm,mbs,nbs,dnz,onz); /* inline function, output __end and __rstart are used below */
3926: MPI_Comm_rank(comm,&rank);
3927: MPI_Comm_rank(comm,&size);
3928: if (rank == size-1) {
3929: /* Check sum(nbs) = Nbs */
3930: if (__end != Nbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Sum of local block columns %D != global block columns %D",__end,Nbs);
3931: }
3933: rstart = __rstart; /* block rstart of *outmat; see inline function MatPreallocateInitialize */
3934: for (i=0; i<mbs; i++) {
3935: MatGetRow_SeqBAIJ(inmat,i*bs,&nnz,&indx,NULL); /* non-blocked nnz and indx */
3936: nnz = nnz/bs;
3937: for (j=0; j<nnz; j++) bindx[j] = indx[j*bs]/bs;
3938: MatPreallocateSet(i+rstart,nnz,bindx,dnz,onz);
3939: MatRestoreRow_SeqBAIJ(inmat,i*bs,&nnz,&indx,NULL);
3940: }
3941: PetscFree(bindx);
3943: MatCreate(comm,outmat);
3944: MatSetSizes(*outmat,m,n,PETSC_DETERMINE,PETSC_DETERMINE);
3945: MatSetBlockSizes(*outmat,bs,cbs);
3946: MatSetType(*outmat,MATBAIJ);
3947: MatSeqBAIJSetPreallocation(*outmat,bs,0,dnz);
3948: MatMPIBAIJSetPreallocation(*outmat,bs,0,dnz,0,onz);
3949: MatPreallocateFinalize(dnz,onz);
3950: }
3952: /* numeric phase */
3953: MatGetBlockSizes(inmat,&bs,&cbs);
3954: MatGetOwnershipRange(*outmat,&rstart,NULL);
3956: for (i=0; i<m; i++) {
3957: MatGetRow_SeqBAIJ(inmat,i,&nnz,&indx,&values);
3958: Ii = i + rstart;
3959: MatSetValues(*outmat,1,&Ii,nnz,indx,values,INSERT_VALUES);
3960: MatRestoreRow_SeqBAIJ(inmat,i,&nnz,&indx,&values);
3961: }
3962: MatAssemblyBegin(*outmat,MAT_FINAL_ASSEMBLY);
3963: MatAssemblyEnd(*outmat,MAT_FINAL_ASSEMBLY);
3964: return(0);
3965: }