Actual source code: aij.c
petsc-master 2020-01-15
2: /*
3: Defines the basic matrix operations for the AIJ (compressed row)
4: matrix storage format.
5: */
8: #include <../src/mat/impls/aij/seq/aij.h>
9: #include <petscblaslapack.h>
10: #include <petscbt.h>
11: #include <petsc/private/kernels/blocktranspose.h>
13: PetscErrorCode MatSeqAIJSetTypeFromOptions(Mat A)
14: {
15: PetscErrorCode ierr;
16: PetscBool flg;
17: char type[256];
20: PetscObjectOptionsBegin((PetscObject)A);
21: PetscOptionsFList("-mat_seqaij_type","Matrix SeqAIJ type","MatSeqAIJSetType",MatSeqAIJList,"seqaij",type,256,&flg);
22: if (flg) {
23: MatSeqAIJSetType(A,type);
24: }
25: PetscOptionsEnd();
26: return(0);
27: }
29: PetscErrorCode MatGetColumnNorms_SeqAIJ(Mat A,NormType type,PetscReal *norms)
30: {
32: PetscInt i,m,n;
33: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data;
36: MatGetSize(A,&m,&n);
37: PetscArrayzero(norms,n);
38: if (type == NORM_2) {
39: for (i=0; i<aij->i[m]; i++) {
40: norms[aij->j[i]] += PetscAbsScalar(aij->a[i]*aij->a[i]);
41: }
42: } else if (type == NORM_1) {
43: for (i=0; i<aij->i[m]; i++) {
44: norms[aij->j[i]] += PetscAbsScalar(aij->a[i]);
45: }
46: } else if (type == NORM_INFINITY) {
47: for (i=0; i<aij->i[m]; i++) {
48: norms[aij->j[i]] = PetscMax(PetscAbsScalar(aij->a[i]),norms[aij->j[i]]);
49: }
50: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown NormType");
52: if (type == NORM_2) {
53: for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]);
54: }
55: return(0);
56: }
58: PetscErrorCode MatFindOffBlockDiagonalEntries_SeqAIJ(Mat A,IS *is)
59: {
60: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
61: PetscInt i,m=A->rmap->n,cnt = 0, bs = A->rmap->bs;
62: const PetscInt *jj = a->j,*ii = a->i;
63: PetscInt *rows;
64: PetscErrorCode ierr;
67: for (i=0; i<m; i++) {
68: if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) {
69: cnt++;
70: }
71: }
72: PetscMalloc1(cnt,&rows);
73: cnt = 0;
74: for (i=0; i<m; i++) {
75: if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) {
76: rows[cnt] = i;
77: cnt++;
78: }
79: }
80: ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,is);
81: return(0);
82: }
84: PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat A,PetscInt *nrows,PetscInt **zrows)
85: {
86: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
87: const MatScalar *aa = a->a;
88: PetscInt i,m=A->rmap->n,cnt = 0;
89: const PetscInt *ii = a->i,*jj = a->j,*diag;
90: PetscInt *rows;
91: PetscErrorCode ierr;
94: MatMarkDiagonal_SeqAIJ(A);
95: diag = a->diag;
96: for (i=0; i<m; i++) {
97: if ((diag[i] >= ii[i+1]) || (jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) {
98: cnt++;
99: }
100: }
101: PetscMalloc1(cnt,&rows);
102: cnt = 0;
103: for (i=0; i<m; i++) {
104: if ((diag[i] >= ii[i+1]) || (jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) {
105: rows[cnt++] = i;
106: }
107: }
108: *nrows = cnt;
109: *zrows = rows;
110: return(0);
111: }
113: PetscErrorCode MatFindZeroDiagonals_SeqAIJ(Mat A,IS *zrows)
114: {
115: PetscInt nrows,*rows;
119: *zrows = NULL;
120: MatFindZeroDiagonals_SeqAIJ_Private(A,&nrows,&rows);
121: ISCreateGeneral(PetscObjectComm((PetscObject)A),nrows,rows,PETSC_OWN_POINTER,zrows);
122: return(0);
123: }
125: PetscErrorCode MatFindNonzeroRows_SeqAIJ(Mat A,IS *keptrows)
126: {
127: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
128: const MatScalar *aa;
129: PetscInt m=A->rmap->n,cnt = 0;
130: const PetscInt *ii;
131: PetscInt n,i,j,*rows;
132: PetscErrorCode ierr;
135: *keptrows = 0;
136: ii = a->i;
137: for (i=0; i<m; i++) {
138: n = ii[i+1] - ii[i];
139: if (!n) {
140: cnt++;
141: goto ok1;
142: }
143: aa = a->a + ii[i];
144: for (j=0; j<n; j++) {
145: if (aa[j] != 0.0) goto ok1;
146: }
147: cnt++;
148: ok1:;
149: }
150: if (!cnt) return(0);
151: PetscMalloc1(A->rmap->n-cnt,&rows);
152: cnt = 0;
153: for (i=0; i<m; i++) {
154: n = ii[i+1] - ii[i];
155: if (!n) continue;
156: aa = a->a + ii[i];
157: for (j=0; j<n; j++) {
158: if (aa[j] != 0.0) {
159: rows[cnt++] = i;
160: break;
161: }
162: }
163: }
164: ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,keptrows);
165: return(0);
166: }
168: PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is)
169: {
170: PetscErrorCode ierr;
171: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data;
172: PetscInt i,m = Y->rmap->n;
173: const PetscInt *diag;
174: MatScalar *aa = aij->a;
175: const PetscScalar *v;
176: PetscBool missing;
177: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
178: PetscBool inserted = PETSC_FALSE;
179: #endif
182: if (Y->assembled) {
183: MatMissingDiagonal_SeqAIJ(Y,&missing,NULL);
184: if (!missing) {
185: diag = aij->diag;
186: VecGetArrayRead(D,&v);
187: if (is == INSERT_VALUES) {
188: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
189: inserted = PETSC_TRUE;
190: #endif
191: for (i=0; i<m; i++) {
192: aa[diag[i]] = v[i];
193: }
194: } else {
195: for (i=0; i<m; i++) {
196: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
197: if (v[i] != 0.0) inserted = PETSC_TRUE;
198: #endif
199: aa[diag[i]] += v[i];
200: }
201: }
202: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
203: if (inserted) Y->offloadmask = PETSC_OFFLOAD_CPU;
204: #endif
205: VecRestoreArrayRead(D,&v);
206: return(0);
207: }
208: MatSeqAIJInvalidateDiagonal(Y);
209: }
210: MatDiagonalSet_Default(Y,D,is);
211: return(0);
212: }
214: PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *m,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
215: {
216: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
218: PetscInt i,ishift;
221: *m = A->rmap->n;
222: if (!ia) return(0);
223: ishift = 0;
224: if (symmetric && !A->structurally_symmetric) {
225: MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,PETSC_TRUE,ishift,oshift,(PetscInt**)ia,(PetscInt**)ja);
226: } else if (oshift == 1) {
227: PetscInt *tia;
228: PetscInt nz = a->i[A->rmap->n];
229: /* malloc space and add 1 to i and j indices */
230: PetscMalloc1(A->rmap->n+1,&tia);
231: for (i=0; i<A->rmap->n+1; i++) tia[i] = a->i[i] + 1;
232: *ia = tia;
233: if (ja) {
234: PetscInt *tja;
235: PetscMalloc1(nz+1,&tja);
236: for (i=0; i<nz; i++) tja[i] = a->j[i] + 1;
237: *ja = tja;
238: }
239: } else {
240: *ia = a->i;
241: if (ja) *ja = a->j;
242: }
243: return(0);
244: }
246: PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
247: {
251: if (!ia) return(0);
252: if ((symmetric && !A->structurally_symmetric) || oshift == 1) {
253: PetscFree(*ia);
254: if (ja) {PetscFree(*ja);}
255: }
256: return(0);
257: }
259: PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
260: {
261: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
263: PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n;
264: PetscInt nz = a->i[m],row,*jj,mr,col;
267: *nn = n;
268: if (!ia) return(0);
269: if (symmetric) {
270: MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,PETSC_TRUE,0,oshift,(PetscInt**)ia,(PetscInt**)ja);
271: } else {
272: PetscCalloc1(n,&collengths);
273: PetscMalloc1(n+1,&cia);
274: PetscMalloc1(nz,&cja);
275: jj = a->j;
276: for (i=0; i<nz; i++) {
277: collengths[jj[i]]++;
278: }
279: cia[0] = oshift;
280: for (i=0; i<n; i++) {
281: cia[i+1] = cia[i] + collengths[i];
282: }
283: PetscArrayzero(collengths,n);
284: jj = a->j;
285: for (row=0; row<m; row++) {
286: mr = a->i[row+1] - a->i[row];
287: for (i=0; i<mr; i++) {
288: col = *jj++;
290: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
291: }
292: }
293: PetscFree(collengths);
294: *ia = cia; *ja = cja;
295: }
296: return(0);
297: }
299: PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
300: {
304: if (!ia) return(0);
306: PetscFree(*ia);
307: PetscFree(*ja);
308: return(0);
309: }
311: /*
312: MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from
313: MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output
314: spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ() and MatFDColoringCreate_SeqXAIJ()
315: */
316: PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
317: {
318: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
320: PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n;
321: PetscInt nz = a->i[m],row,mr,col,tmp;
322: PetscInt *cspidx;
323: const PetscInt *jj;
326: *nn = n;
327: if (!ia) return(0);
329: PetscCalloc1(n,&collengths);
330: PetscMalloc1(n+1,&cia);
331: PetscMalloc1(nz,&cja);
332: PetscMalloc1(nz,&cspidx);
333: jj = a->j;
334: for (i=0; i<nz; i++) {
335: collengths[jj[i]]++;
336: }
337: cia[0] = oshift;
338: for (i=0; i<n; i++) {
339: cia[i+1] = cia[i] + collengths[i];
340: }
341: PetscArrayzero(collengths,n);
342: jj = a->j;
343: for (row=0; row<m; row++) {
344: mr = a->i[row+1] - a->i[row];
345: for (i=0; i<mr; i++) {
346: col = *jj++;
347: tmp = cia[col] + collengths[col]++ - oshift;
348: cspidx[tmp] = a->i[row] + i; /* index of a->j */
349: cja[tmp] = row + oshift;
350: }
351: }
352: PetscFree(collengths);
353: *ia = cia;
354: *ja = cja;
355: *spidx = cspidx;
356: return(0);
357: }
359: PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
360: {
364: MatRestoreColumnIJ_SeqAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done);
365: PetscFree(*spidx);
366: return(0);
367: }
369: PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[])
370: {
371: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
372: PetscInt *ai = a->i;
376: PetscArraycpy(a->a+ai[row],v,ai[row+1]-ai[row]);
377: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
378: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && ai[row+1]-ai[row]) A->offloadmask = PETSC_OFFLOAD_CPU;
379: #endif
380: return(0);
381: }
383: /*
384: MatSeqAIJSetValuesLocalFast - An optimized version of MatSetValuesLocal() for SeqAIJ matrices with several assumptions
386: - a single row of values is set with each call
387: - no row or column indices are negative or (in error) larger than the number of rows or columns
388: - the values are always added to the matrix, not set
389: - no new locations are introduced in the nonzero structure of the matrix
391: This does NOT assume the global column indices are sorted
393: */
395: #include <petsc/private/isimpl.h>
396: PetscErrorCode MatSeqAIJSetValuesLocalFast(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
397: {
398: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
399: PetscInt low,high,t,row,nrow,i,col,l;
400: const PetscInt *rp,*ai = a->i,*ailen = a->ilen,*aj = a->j;
401: PetscInt lastcol = -1;
402: MatScalar *ap,value,*aa = a->a;
403: const PetscInt *ridx = A->rmap->mapping->indices,*cidx = A->cmap->mapping->indices;
405: row = ridx[im[0]];
406: rp = aj + ai[row];
407: ap = aa + ai[row];
408: nrow = ailen[row];
409: low = 0;
410: high = nrow;
411: for (l=0; l<n; l++) { /* loop over added columns */
412: col = cidx[in[l]];
413: value = v[l];
415: if (col <= lastcol) low = 0;
416: else high = nrow;
417: lastcol = col;
418: while (high-low > 5) {
419: t = (low+high)/2;
420: if (rp[t] > col) high = t;
421: else low = t;
422: }
423: for (i=low; i<high; i++) {
424: if (rp[i] == col) {
425: ap[i] += value;
426: low = i + 1;
427: break;
428: }
429: }
430: }
431: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
432: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && m && n) A->offloadmask = PETSC_OFFLOAD_CPU;
433: #endif
434: return 0;
435: }
437: PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
438: {
439: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
440: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
441: PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen;
443: PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1;
444: MatScalar *ap=NULL,value=0.0,*aa = a->a;
445: PetscBool ignorezeroentries = a->ignorezeroentries;
446: PetscBool roworiented = a->roworiented;
447: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
448: PetscBool inserted = PETSC_FALSE;
449: #endif
452: for (k=0; k<m; k++) { /* loop over added rows */
453: row = im[k];
454: if (row < 0) continue;
455: #if defined(PETSC_USE_DEBUG)
456: if (row >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap->n-1);
457: #endif
458: rp = aj + ai[row];
459: if (!A->structure_only) ap = aa + ai[row];
460: rmax = imax[row]; nrow = ailen[row];
461: low = 0;
462: high = nrow;
463: for (l=0; l<n; l++) { /* loop over added columns */
464: if (in[l] < 0) continue;
465: #if defined(PETSC_USE_DEBUG)
466: if (in[l] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap->n-1);
467: #endif
468: col = in[l];
469: if (v && !A->structure_only) value = roworiented ? v[l + k*n] : v[k + l*m];
470: if (!A->structure_only && value == 0.0 && ignorezeroentries && is == ADD_VALUES && row != col) continue;
472: if (col <= lastcol) low = 0;
473: else high = nrow;
474: lastcol = col;
475: while (high-low > 5) {
476: t = (low+high)/2;
477: if (rp[t] > col) high = t;
478: else low = t;
479: }
480: for (i=low; i<high; i++) {
481: if (rp[i] > col) break;
482: if (rp[i] == col) {
483: if (!A->structure_only) {
484: if (is == ADD_VALUES) {
485: ap[i] += value;
486: (void)PetscLogFlops(1.0);
487: }
488: else ap[i] = value;
489: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
490: inserted = PETSC_TRUE;
491: #endif
492: }
493: low = i + 1;
494: goto noinsert;
495: }
496: }
497: if (value == 0.0 && ignorezeroentries && row != col) goto noinsert;
498: if (nonew == 1) goto noinsert;
499: if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%D,%D) in the matrix",row,col);
500: if (A->structure_only) {
501: MatSeqXAIJReallocateAIJ_structure_only(A,A->rmap->n,1,nrow,row,col,rmax,ai,aj,rp,imax,nonew,MatScalar);
502: } else {
503: MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
504: }
505: N = nrow++ - 1; a->nz++; high++;
506: /* shift up all the later entries in this row */
507: PetscArraymove(rp+i+1,rp+i,N-i+1);
508: rp[i] = col;
509: if (!A->structure_only){
510: PetscArraymove(ap+i+1,ap+i,N-i+1);
511: ap[i] = value;
512: }
513: low = i + 1;
514: A->nonzerostate++;
515: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
516: inserted = PETSC_TRUE;
517: #endif
518: noinsert:;
519: }
520: ailen[row] = nrow;
521: }
522: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
523: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && inserted) A->offloadmask = PETSC_OFFLOAD_CPU;
524: #endif
525: return(0);
526: }
528: PetscErrorCode MatSetValues_SeqAIJ_SortedFull(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
529: {
530: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
531: PetscInt *rp,k,row;
532: PetscInt *ai = a->i,*ailen = a->ilen;
534: PetscInt *aj = a->j;
535: MatScalar *aa = a->a,*ap;
538: for (k=0; k<m; k++) { /* loop over added rows */
539: row = im[k];
540: rp = aj + ai[row];
541: ap = aa + ai[row];
542: if (!A->was_assembled) {
543: PetscMemcpy(rp,in,n*sizeof(PetscInt));
544: }
545: if (!A->structure_only) {
546: if (v) {
547: PetscMemcpy(ap,v,n*sizeof(PetscScalar));
548: v += n;
549: } else {
550: PetscMemzero(ap,n*sizeof(PetscScalar));
551: }
552: }
553: ailen[row] = n;
554: a->nz += n;
555: }
556: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
557: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && m && n) A->offloadmask = PETSC_OFFLOAD_CPU;
558: #endif
559: return(0);
560: }
563: PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
564: {
565: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
566: PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
567: PetscInt *ai = a->i,*ailen = a->ilen;
568: MatScalar *ap,*aa = a->a;
571: for (k=0; k<m; k++) { /* loop over rows */
572: row = im[k];
573: if (row < 0) {v += n; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",row); */
574: if (row >= A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap->n-1);
575: rp = aj + ai[row]; ap = aa + ai[row];
576: nrow = ailen[row];
577: for (l=0; l<n; l++) { /* loop over columns */
578: if (in[l] < 0) {v++; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",in[l]); */
579: if (in[l] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap->n-1);
580: col = in[l];
581: high = nrow; low = 0; /* assume unsorted */
582: while (high-low > 5) {
583: t = (low+high)/2;
584: if (rp[t] > col) high = t;
585: else low = t;
586: }
587: for (i=low; i<high; i++) {
588: if (rp[i] > col) break;
589: if (rp[i] == col) {
590: *v++ = ap[i];
591: goto finished;
592: }
593: }
594: *v++ = 0.0;
595: finished:;
596: }
597: }
598: return(0);
599: }
602: PetscErrorCode MatView_SeqAIJ_Binary(Mat A,PetscViewer viewer)
603: {
604: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
606: PetscInt i,*col_lens;
607: int fd;
608: FILE *file;
611: PetscViewerBinaryGetDescriptor(viewer,&fd);
612: PetscMalloc1(4+A->rmap->n,&col_lens);
614: col_lens[0] = MAT_FILE_CLASSID;
615: col_lens[1] = A->rmap->n;
616: col_lens[2] = A->cmap->n;
617: col_lens[3] = a->nz;
619: /* store lengths of each row and write (including header) to file */
620: for (i=0; i<A->rmap->n; i++) {
621: col_lens[4+i] = a->i[i+1] - a->i[i];
622: }
623: PetscBinaryWrite(fd,col_lens,4+A->rmap->n,PETSC_INT,PETSC_TRUE);
624: PetscFree(col_lens);
626: /* store column indices (zero start index) */
627: PetscBinaryWrite(fd,a->j,a->nz,PETSC_INT,PETSC_FALSE);
629: /* store nonzero values */
630: PetscBinaryWrite(fd,a->a,a->nz,PETSC_SCALAR,PETSC_FALSE);
632: PetscViewerBinaryGetInfoPointer(viewer,&file);
633: if (file) {
634: fprintf(file,"-matload_block_size %d\n",(int)PetscAbs(A->rmap->bs));
635: }
636: return(0);
637: }
639: static PetscErrorCode MatView_SeqAIJ_ASCII_structonly(Mat A,PetscViewer viewer)
640: {
642: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
643: PetscInt i,k,m=A->rmap->N;
646: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
647: for (i=0; i<m; i++) {
648: PetscViewerASCIIPrintf(viewer,"row %D:",i);
649: for (k=a->i[i]; k<a->i[i+1]; k++) {
650: PetscViewerASCIIPrintf(viewer," (%D) ",a->j[k]);
651: }
652: PetscViewerASCIIPrintf(viewer,"\n");
653: }
654: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
655: return(0);
656: }
658: extern PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer);
660: PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer)
661: {
662: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
663: PetscErrorCode ierr;
664: PetscInt i,j,m = A->rmap->n;
665: const char *name;
666: PetscViewerFormat format;
669: if (A->structure_only) {
670: MatView_SeqAIJ_ASCII_structonly(A,viewer);
671: return(0);
672: }
674: PetscViewerGetFormat(viewer,&format);
675: if (format == PETSC_VIEWER_ASCII_MATLAB) {
676: PetscInt nofinalvalue = 0;
677: if (m && ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap->n-1))) {
678: /* Need a dummy value to ensure the dimension of the matrix. */
679: nofinalvalue = 1;
680: }
681: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
682: PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",m,A->cmap->n);
683: PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %D \n",a->nz);
684: #if defined(PETSC_USE_COMPLEX)
685: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,4);\n",a->nz+nofinalvalue);
686: #else
687: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,3);\n",a->nz+nofinalvalue);
688: #endif
689: PetscViewerASCIIPrintf(viewer,"zzz = [\n");
691: for (i=0; i<m; i++) {
692: for (j=a->i[i]; j<a->i[i+1]; j++) {
693: #if defined(PETSC_USE_COMPLEX)
694: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e %18.16e\n",i+1,a->j[j]+1,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
695: #else
696: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",i+1,a->j[j]+1,(double)a->a[j]);
697: #endif
698: }
699: }
700: if (nofinalvalue) {
701: #if defined(PETSC_USE_COMPLEX)
702: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e %18.16e\n",m,A->cmap->n,0.,0.);
703: #else
704: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",m,A->cmap->n,0.0);
705: #endif
706: }
707: PetscObjectGetName((PetscObject)A,&name);
708: PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);
709: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
710: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
711: return(0);
712: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
713: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
714: for (i=0; i<m; i++) {
715: PetscViewerASCIIPrintf(viewer,"row %D:",i);
716: for (j=a->i[i]; j<a->i[i+1]; j++) {
717: #if defined(PETSC_USE_COMPLEX)
718: if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) {
719: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
720: } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) {
721: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
722: } else if (PetscRealPart(a->a[j]) != 0.0) {
723: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
724: }
725: #else
726: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);}
727: #endif
728: }
729: PetscViewerASCIIPrintf(viewer,"\n");
730: }
731: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
732: } else if (format == PETSC_VIEWER_ASCII_SYMMODU) {
733: PetscInt nzd=0,fshift=1,*sptr;
734: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
735: PetscMalloc1(m+1,&sptr);
736: for (i=0; i<m; i++) {
737: sptr[i] = nzd+1;
738: for (j=a->i[i]; j<a->i[i+1]; j++) {
739: if (a->j[j] >= i) {
740: #if defined(PETSC_USE_COMPLEX)
741: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++;
742: #else
743: if (a->a[j] != 0.0) nzd++;
744: #endif
745: }
746: }
747: }
748: sptr[m] = nzd+1;
749: PetscViewerASCIIPrintf(viewer," %D %D\n\n",m,nzd);
750: for (i=0; i<m+1; i+=6) {
751: if (i+4<m) {
752: PetscViewerASCIIPrintf(viewer," %D %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4],sptr[i+5]);
753: } else if (i+3<m) {
754: PetscViewerASCIIPrintf(viewer," %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4]);
755: } else if (i+2<m) {
756: PetscViewerASCIIPrintf(viewer," %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);
757: } else if (i+1<m) {
758: PetscViewerASCIIPrintf(viewer," %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2]);
759: } else if (i<m) {
760: PetscViewerASCIIPrintf(viewer," %D %D\n",sptr[i],sptr[i+1]);
761: } else {
762: PetscViewerASCIIPrintf(viewer," %D\n",sptr[i]);
763: }
764: }
765: PetscViewerASCIIPrintf(viewer,"\n");
766: PetscFree(sptr);
767: for (i=0; i<m; i++) {
768: for (j=a->i[i]; j<a->i[i+1]; j++) {
769: if (a->j[j] >= i) {PetscViewerASCIIPrintf(viewer," %D ",a->j[j]+fshift);}
770: }
771: PetscViewerASCIIPrintf(viewer,"\n");
772: }
773: PetscViewerASCIIPrintf(viewer,"\n");
774: for (i=0; i<m; i++) {
775: for (j=a->i[i]; j<a->i[i+1]; j++) {
776: if (a->j[j] >= i) {
777: #if defined(PETSC_USE_COMPLEX)
778: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) {
779: PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
780: }
781: #else
782: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," %18.16e ",(double)a->a[j]);}
783: #endif
784: }
785: }
786: PetscViewerASCIIPrintf(viewer,"\n");
787: }
788: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
789: } else if (format == PETSC_VIEWER_ASCII_DENSE) {
790: PetscInt cnt = 0,jcnt;
791: PetscScalar value;
792: #if defined(PETSC_USE_COMPLEX)
793: PetscBool realonly = PETSC_TRUE;
795: for (i=0; i<a->i[m]; i++) {
796: if (PetscImaginaryPart(a->a[i]) != 0.0) {
797: realonly = PETSC_FALSE;
798: break;
799: }
800: }
801: #endif
803: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
804: for (i=0; i<m; i++) {
805: jcnt = 0;
806: for (j=0; j<A->cmap->n; j++) {
807: if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) {
808: value = a->a[cnt++];
809: jcnt++;
810: } else {
811: value = 0.0;
812: }
813: #if defined(PETSC_USE_COMPLEX)
814: if (realonly) {
815: PetscViewerASCIIPrintf(viewer," %7.5e ",(double)PetscRealPart(value));
816: } else {
817: PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",(double)PetscRealPart(value),(double)PetscImaginaryPart(value));
818: }
819: #else
820: PetscViewerASCIIPrintf(viewer," %7.5e ",(double)value);
821: #endif
822: }
823: PetscViewerASCIIPrintf(viewer,"\n");
824: }
825: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
826: } else if (format == PETSC_VIEWER_ASCII_MATRIXMARKET) {
827: PetscInt fshift=1;
828: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
829: #if defined(PETSC_USE_COMPLEX)
830: PetscViewerASCIIPrintf(viewer,"%%%%MatrixMarket matrix coordinate complex general\n");
831: #else
832: PetscViewerASCIIPrintf(viewer,"%%%%MatrixMarket matrix coordinate real general\n");
833: #endif
834: PetscViewerASCIIPrintf(viewer,"%D %D %D\n", m, A->cmap->n, a->nz);
835: for (i=0; i<m; i++) {
836: for (j=a->i[i]; j<a->i[i+1]; j++) {
837: #if defined(PETSC_USE_COMPLEX)
838: PetscViewerASCIIPrintf(viewer,"%D %D %g %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
839: #else
840: PetscViewerASCIIPrintf(viewer,"%D %D %g\n", i+fshift, a->j[j]+fshift, (double)a->a[j]);
841: #endif
842: }
843: }
844: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
845: } else {
846: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
847: if (A->factortype) {
848: for (i=0; i<m; i++) {
849: PetscViewerASCIIPrintf(viewer,"row %D:",i);
850: /* L part */
851: for (j=a->i[i]; j<a->i[i+1]; j++) {
852: #if defined(PETSC_USE_COMPLEX)
853: if (PetscImaginaryPart(a->a[j]) > 0.0) {
854: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
855: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
856: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));
857: } else {
858: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
859: }
860: #else
861: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
862: #endif
863: }
864: /* diagonal */
865: j = a->diag[i];
866: #if defined(PETSC_USE_COMPLEX)
867: if (PetscImaginaryPart(a->a[j]) > 0.0) {
868: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)PetscImaginaryPart(1.0/a->a[j]));
869: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
870: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)(-PetscImaginaryPart(1.0/a->a[j])));
871: } else {
872: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(1.0/a->a[j]));
873: }
874: #else
875: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)(1.0/a->a[j]));
876: #endif
878: /* U part */
879: for (j=a->diag[i+1]+1; j<a->diag[i]; j++) {
880: #if defined(PETSC_USE_COMPLEX)
881: if (PetscImaginaryPart(a->a[j]) > 0.0) {
882: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
883: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
884: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));
885: } else {
886: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
887: }
888: #else
889: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
890: #endif
891: }
892: PetscViewerASCIIPrintf(viewer,"\n");
893: }
894: } else {
895: for (i=0; i<m; i++) {
896: PetscViewerASCIIPrintf(viewer,"row %D:",i);
897: for (j=a->i[i]; j<a->i[i+1]; j++) {
898: #if defined(PETSC_USE_COMPLEX)
899: if (PetscImaginaryPart(a->a[j]) > 0.0) {
900: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
901: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
902: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
903: } else {
904: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
905: }
906: #else
907: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
908: #endif
909: }
910: PetscViewerASCIIPrintf(viewer,"\n");
911: }
912: }
913: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
914: }
915: PetscViewerFlush(viewer);
916: return(0);
917: }
919: #include <petscdraw.h>
920: PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
921: {
922: Mat A = (Mat) Aa;
923: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
924: PetscErrorCode ierr;
925: PetscInt i,j,m = A->rmap->n;
926: int color;
927: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r;
928: PetscViewer viewer;
929: PetscViewerFormat format;
932: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
933: PetscViewerGetFormat(viewer,&format);
934: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
936: /* loop over matrix elements drawing boxes */
938: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
939: PetscDrawCollectiveBegin(draw);
940: /* Blue for negative, Cyan for zero and Red for positive */
941: color = PETSC_DRAW_BLUE;
942: for (i=0; i<m; i++) {
943: y_l = m - i - 1.0; y_r = y_l + 1.0;
944: for (j=a->i[i]; j<a->i[i+1]; j++) {
945: x_l = a->j[j]; x_r = x_l + 1.0;
946: if (PetscRealPart(a->a[j]) >= 0.) continue;
947: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
948: }
949: }
950: color = PETSC_DRAW_CYAN;
951: for (i=0; i<m; i++) {
952: y_l = m - i - 1.0; y_r = y_l + 1.0;
953: for (j=a->i[i]; j<a->i[i+1]; j++) {
954: x_l = a->j[j]; x_r = x_l + 1.0;
955: if (a->a[j] != 0.) continue;
956: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
957: }
958: }
959: color = PETSC_DRAW_RED;
960: for (i=0; i<m; i++) {
961: y_l = m - i - 1.0; y_r = y_l + 1.0;
962: for (j=a->i[i]; j<a->i[i+1]; j++) {
963: x_l = a->j[j]; x_r = x_l + 1.0;
964: if (PetscRealPart(a->a[j]) <= 0.) continue;
965: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
966: }
967: }
968: PetscDrawCollectiveEnd(draw);
969: } else {
970: /* use contour shading to indicate magnitude of values */
971: /* first determine max of all nonzero values */
972: PetscReal minv = 0.0, maxv = 0.0;
973: PetscInt nz = a->nz, count = 0;
974: PetscDraw popup;
976: for (i=0; i<nz; i++) {
977: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
978: }
979: if (minv >= maxv) maxv = minv + PETSC_SMALL;
980: PetscDrawGetPopup(draw,&popup);
981: PetscDrawScalePopup(popup,minv,maxv);
983: PetscDrawCollectiveBegin(draw);
984: for (i=0; i<m; i++) {
985: y_l = m - i - 1.0;
986: y_r = y_l + 1.0;
987: for (j=a->i[i]; j<a->i[i+1]; j++) {
988: x_l = a->j[j];
989: x_r = x_l + 1.0;
990: color = PetscDrawRealToColor(PetscAbsScalar(a->a[count]),minv,maxv);
991: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
992: count++;
993: }
994: }
995: PetscDrawCollectiveEnd(draw);
996: }
997: return(0);
998: }
1000: #include <petscdraw.h>
1001: PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer)
1002: {
1004: PetscDraw draw;
1005: PetscReal xr,yr,xl,yl,h,w;
1006: PetscBool isnull;
1009: PetscViewerDrawGetDraw(viewer,0,&draw);
1010: PetscDrawIsNull(draw,&isnull);
1011: if (isnull) return(0);
1013: xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0;
1014: xr += w; yr += h; xl = -w; yl = -h;
1015: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
1016: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
1017: PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);
1018: PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL);
1019: PetscDrawSave(draw);
1020: return(0);
1021: }
1023: PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer)
1024: {
1026: PetscBool iascii,isbinary,isdraw;
1029: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
1030: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
1031: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
1032: if (iascii) {
1033: MatView_SeqAIJ_ASCII(A,viewer);
1034: } else if (isbinary) {
1035: MatView_SeqAIJ_Binary(A,viewer);
1036: } else if (isdraw) {
1037: MatView_SeqAIJ_Draw(A,viewer);
1038: }
1039: MatView_SeqAIJ_Inode(A,viewer);
1040: return(0);
1041: }
1043: PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode)
1044: {
1045: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1047: PetscInt fshift = 0,i,*ai = a->i,*aj = a->j,*imax = a->imax;
1048: PetscInt m = A->rmap->n,*ip,N,*ailen = a->ilen,rmax = 0;
1049: MatScalar *aa = a->a,*ap;
1050: PetscReal ratio = 0.6;
1053: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
1054: MatSeqAIJInvalidateDiagonal(A);
1055: if (A->was_assembled && A->ass_nonzerostate == A->nonzerostate) return(0);
1057: if (m) rmax = ailen[0]; /* determine row with most nonzeros */
1058: for (i=1; i<m; i++) {
1059: /* move each row back by the amount of empty slots (fshift) before it*/
1060: fshift += imax[i-1] - ailen[i-1];
1061: rmax = PetscMax(rmax,ailen[i]);
1062: if (fshift) {
1063: ip = aj + ai[i];
1064: ap = aa + ai[i];
1065: N = ailen[i];
1066: PetscArraymove(ip-fshift,ip,N);
1067: if (!A->structure_only) {
1068: PetscArraymove(ap-fshift,ap,N);
1069: }
1070: }
1071: ai[i] = ai[i-1] + ailen[i-1];
1072: }
1073: if (m) {
1074: fshift += imax[m-1] - ailen[m-1];
1075: ai[m] = ai[m-1] + ailen[m-1];
1076: }
1078: /* reset ilen and imax for each row */
1079: a->nonzerorowcnt = 0;
1080: if (A->structure_only) {
1081: PetscFree(a->imax);
1082: PetscFree(a->ilen);
1083: } else { /* !A->structure_only */
1084: for (i=0; i<m; i++) {
1085: ailen[i] = imax[i] = ai[i+1] - ai[i];
1086: a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0);
1087: }
1088: }
1089: a->nz = ai[m];
1090: if (fshift && a->nounused == -1) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_PLIB, "Unused space detected in matrix: %D X %D, %D unneeded", m, A->cmap->n, fshift);
1092: MatMarkDiagonal_SeqAIJ(A);
1093: PetscInfo4(A,"Matrix size: %D X %D; storage space: %D unneeded,%D used\n",m,A->cmap->n,fshift,a->nz);
1094: PetscInfo1(A,"Number of mallocs during MatSetValues() is %D\n",a->reallocs);
1095: PetscInfo1(A,"Maximum nonzeros in any row is %D\n",rmax);
1097: A->info.mallocs += a->reallocs;
1098: a->reallocs = 0;
1099: A->info.nz_unneeded = (PetscReal)fshift;
1100: a->rmax = rmax;
1102: if (!A->structure_only) {
1103: MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,m,ratio);
1104: }
1105: MatAssemblyEnd_SeqAIJ_Inode(A,mode);
1106: return(0);
1107: }
1109: PetscErrorCode MatRealPart_SeqAIJ(Mat A)
1110: {
1111: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1112: PetscInt i,nz = a->nz;
1113: MatScalar *aa = a->a;
1117: for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
1118: MatSeqAIJInvalidateDiagonal(A);
1119: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
1120: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
1121: #endif
1122: return(0);
1123: }
1125: PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A)
1126: {
1127: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1128: PetscInt i,nz = a->nz;
1129: MatScalar *aa = a->a;
1133: for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
1134: MatSeqAIJInvalidateDiagonal(A);
1135: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
1136: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
1137: #endif
1138: return(0);
1139: }
1141: PetscErrorCode MatZeroEntries_SeqAIJ(Mat A)
1142: {
1143: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1147: PetscArrayzero(a->a,a->i[A->rmap->n]);
1148: MatSeqAIJInvalidateDiagonal(A);
1149: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
1150: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
1151: #endif
1152: return(0);
1153: }
1155: PetscErrorCode MatDestroy_SeqAIJ(Mat A)
1156: {
1157: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1161: #if defined(PETSC_USE_LOG)
1162: PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap->n,A->cmap->n,a->nz);
1163: #endif
1164: MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);
1165: ISDestroy(&a->row);
1166: ISDestroy(&a->col);
1167: PetscFree(a->diag);
1168: PetscFree(a->ibdiag);
1169: PetscFree(a->imax);
1170: PetscFree(a->ilen);
1171: PetscFree(a->ipre);
1172: PetscFree3(a->idiag,a->mdiag,a->ssor_work);
1173: PetscFree(a->solve_work);
1174: ISDestroy(&a->icol);
1175: PetscFree(a->saved_values);
1176: ISColoringDestroy(&a->coloring);
1177: PetscFree2(a->compressedrow.i,a->compressedrow.rindex);
1178: PetscFree(a->matmult_abdense);
1180: MatDestroy_SeqAIJ_Inode(A);
1181: PetscFree(A->data);
1183: PetscObjectChangeTypeName((PetscObject)A,0);
1184: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetColumnIndices_C",NULL);
1185: PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL);
1186: PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL);
1187: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsbaij_C",NULL);
1188: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqbaij_C",NULL);
1189: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijperm_C",NULL);
1190: #if defined(PETSC_HAVE_ELEMENTAL)
1191: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_elemental_C",NULL);
1192: #endif
1193: #if defined(PETSC_HAVE_HYPRE)
1194: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_hypre_C",NULL);
1195: PetscObjectComposeFunction((PetscObject)A,"MatMatMatMult_transpose_seqaij_seqaij_C",NULL);
1196: #endif
1197: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqdense_C",NULL);
1198: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsell_C",NULL);
1199: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_is_C",NULL);
1200: PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL);
1201: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocation_C",NULL);
1202: PetscObjectComposeFunction((PetscObject)A,"MatResetPreallocation_C",NULL);
1203: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C",NULL);
1204: PetscObjectComposeFunction((PetscObject)A,"MatReorderForNonzeroDiagonal_C",NULL);
1205: PetscObjectComposeFunction((PetscObject)A,"MatPtAP_is_seqaij_C",NULL);
1206: return(0);
1207: }
1209: PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op,PetscBool flg)
1210: {
1211: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1215: switch (op) {
1216: case MAT_ROW_ORIENTED:
1217: a->roworiented = flg;
1218: break;
1219: case MAT_KEEP_NONZERO_PATTERN:
1220: a->keepnonzeropattern = flg;
1221: break;
1222: case MAT_NEW_NONZERO_LOCATIONS:
1223: a->nonew = (flg ? 0 : 1);
1224: break;
1225: case MAT_NEW_NONZERO_LOCATION_ERR:
1226: a->nonew = (flg ? -1 : 0);
1227: break;
1228: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1229: a->nonew = (flg ? -2 : 0);
1230: break;
1231: case MAT_UNUSED_NONZERO_LOCATION_ERR:
1232: a->nounused = (flg ? -1 : 0);
1233: break;
1234: case MAT_IGNORE_ZERO_ENTRIES:
1235: a->ignorezeroentries = flg;
1236: break;
1237: case MAT_SPD:
1238: case MAT_SYMMETRIC:
1239: case MAT_STRUCTURALLY_SYMMETRIC:
1240: case MAT_HERMITIAN:
1241: case MAT_SYMMETRY_ETERNAL:
1242: case MAT_STRUCTURE_ONLY:
1243: /* These options are handled directly by MatSetOption() */
1244: break;
1245: case MAT_NEW_DIAGONALS:
1246: case MAT_IGNORE_OFF_PROC_ENTRIES:
1247: case MAT_USE_HASH_TABLE:
1248: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1249: break;
1250: case MAT_USE_INODES:
1251: /* Not an error because MatSetOption_SeqAIJ_Inode handles this one */
1252: break;
1253: case MAT_SUBMAT_SINGLEIS:
1254: A->submat_singleis = flg;
1255: break;
1256: case MAT_SORTED_FULL:
1257: if (flg) A->ops->setvalues = MatSetValues_SeqAIJ_SortedFull;
1258: else A->ops->setvalues = MatSetValues_SeqAIJ;
1259: break;
1260: default:
1261: SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op);
1262: }
1263: MatSetOption_SeqAIJ_Inode(A,op,flg);
1264: return(0);
1265: }
1267: PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v)
1268: {
1269: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1271: PetscInt i,j,n,*ai=a->i,*aj=a->j;
1272: PetscScalar *aa=a->a,*x;
1275: VecGetLocalSize(v,&n);
1276: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1278: if (A->factortype == MAT_FACTOR_ILU || A->factortype == MAT_FACTOR_LU) {
1279: PetscInt *diag=a->diag;
1280: VecGetArrayWrite(v,&x);
1281: for (i=0; i<n; i++) x[i] = 1.0/aa[diag[i]];
1282: VecRestoreArrayWrite(v,&x);
1283: return(0);
1284: }
1286: VecGetArrayWrite(v,&x);
1287: for (i=0; i<n; i++) {
1288: x[i] = 0.0;
1289: for (j=ai[i]; j<ai[i+1]; j++) {
1290: if (aj[j] == i) {
1291: x[i] = aa[j];
1292: break;
1293: }
1294: }
1295: }
1296: VecRestoreArrayWrite(v,&x);
1297: return(0);
1298: }
1300: #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h>
1301: PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy)
1302: {
1303: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1304: PetscScalar *y;
1305: const PetscScalar *x;
1306: PetscErrorCode ierr;
1307: PetscInt m = A->rmap->n;
1308: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
1309: const MatScalar *v;
1310: PetscScalar alpha;
1311: PetscInt n,i,j;
1312: const PetscInt *idx,*ii,*ridx=NULL;
1313: Mat_CompressedRow cprow = a->compressedrow;
1314: PetscBool usecprow = cprow.use;
1315: #endif
1318: if (zz != yy) {VecCopy(zz,yy);}
1319: VecGetArrayRead(xx,&x);
1320: VecGetArray(yy,&y);
1322: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
1323: fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y);
1324: #else
1325: if (usecprow) {
1326: m = cprow.nrows;
1327: ii = cprow.i;
1328: ridx = cprow.rindex;
1329: } else {
1330: ii = a->i;
1331: }
1332: for (i=0; i<m; i++) {
1333: idx = a->j + ii[i];
1334: v = a->a + ii[i];
1335: n = ii[i+1] - ii[i];
1336: if (usecprow) {
1337: alpha = x[ridx[i]];
1338: } else {
1339: alpha = x[i];
1340: }
1341: for (j=0; j<n; j++) y[idx[j]] += alpha*v[j];
1342: }
1343: #endif
1344: PetscLogFlops(2.0*a->nz);
1345: VecRestoreArrayRead(xx,&x);
1346: VecRestoreArray(yy,&y);
1347: return(0);
1348: }
1350: PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy)
1351: {
1355: VecSet(yy,0.0);
1356: MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);
1357: return(0);
1358: }
1360: #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h>
1362: PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy)
1363: {
1364: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1365: PetscScalar *y;
1366: const PetscScalar *x;
1367: const MatScalar *aa;
1368: PetscErrorCode ierr;
1369: PetscInt m=A->rmap->n;
1370: const PetscInt *aj,*ii,*ridx=NULL;
1371: PetscInt n,i;
1372: PetscScalar sum;
1373: PetscBool usecprow=a->compressedrow.use;
1375: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1376: #pragma disjoint(*x,*y,*aa)
1377: #endif
1380: VecGetArrayRead(xx,&x);
1381: VecGetArray(yy,&y);
1382: ii = a->i;
1383: if (usecprow) { /* use compressed row format */
1384: PetscArrayzero(y,m);
1385: m = a->compressedrow.nrows;
1386: ii = a->compressedrow.i;
1387: ridx = a->compressedrow.rindex;
1388: for (i=0; i<m; i++) {
1389: n = ii[i+1] - ii[i];
1390: aj = a->j + ii[i];
1391: aa = a->a + ii[i];
1392: sum = 0.0;
1393: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1394: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1395: y[*ridx++] = sum;
1396: }
1397: } else { /* do not use compressed row format */
1398: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
1399: aj = a->j;
1400: aa = a->a;
1401: fortranmultaij_(&m,x,ii,aj,aa,y);
1402: #else
1403: for (i=0; i<m; i++) {
1404: n = ii[i+1] - ii[i];
1405: aj = a->j + ii[i];
1406: aa = a->a + ii[i];
1407: sum = 0.0;
1408: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1409: y[i] = sum;
1410: }
1411: #endif
1412: }
1413: PetscLogFlops(2.0*a->nz - a->nonzerorowcnt);
1414: VecRestoreArrayRead(xx,&x);
1415: VecRestoreArray(yy,&y);
1416: return(0);
1417: }
1419: PetscErrorCode MatMultMax_SeqAIJ(Mat A,Vec xx,Vec yy)
1420: {
1421: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1422: PetscScalar *y;
1423: const PetscScalar *x;
1424: const MatScalar *aa;
1425: PetscErrorCode ierr;
1426: PetscInt m=A->rmap->n;
1427: const PetscInt *aj,*ii,*ridx=NULL;
1428: PetscInt n,i,nonzerorow=0;
1429: PetscScalar sum;
1430: PetscBool usecprow=a->compressedrow.use;
1432: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1433: #pragma disjoint(*x,*y,*aa)
1434: #endif
1437: VecGetArrayRead(xx,&x);
1438: VecGetArray(yy,&y);
1439: if (usecprow) { /* use compressed row format */
1440: m = a->compressedrow.nrows;
1441: ii = a->compressedrow.i;
1442: ridx = a->compressedrow.rindex;
1443: for (i=0; i<m; i++) {
1444: n = ii[i+1] - ii[i];
1445: aj = a->j + ii[i];
1446: aa = a->a + ii[i];
1447: sum = 0.0;
1448: nonzerorow += (n>0);
1449: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1450: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1451: y[*ridx++] = sum;
1452: }
1453: } else { /* do not use compressed row format */
1454: ii = a->i;
1455: for (i=0; i<m; i++) {
1456: n = ii[i+1] - ii[i];
1457: aj = a->j + ii[i];
1458: aa = a->a + ii[i];
1459: sum = 0.0;
1460: nonzerorow += (n>0);
1461: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1462: y[i] = sum;
1463: }
1464: }
1465: PetscLogFlops(2.0*a->nz - nonzerorow);
1466: VecRestoreArrayRead(xx,&x);
1467: VecRestoreArray(yy,&y);
1468: return(0);
1469: }
1471: PetscErrorCode MatMultAddMax_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1472: {
1473: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1474: PetscScalar *y,*z;
1475: const PetscScalar *x;
1476: const MatScalar *aa;
1477: PetscErrorCode ierr;
1478: PetscInt m = A->rmap->n,*aj,*ii;
1479: PetscInt n,i,*ridx=NULL;
1480: PetscScalar sum;
1481: PetscBool usecprow=a->compressedrow.use;
1484: VecGetArrayRead(xx,&x);
1485: VecGetArrayPair(yy,zz,&y,&z);
1486: if (usecprow) { /* use compressed row format */
1487: if (zz != yy) {
1488: PetscArraycpy(z,y,m);
1489: }
1490: m = a->compressedrow.nrows;
1491: ii = a->compressedrow.i;
1492: ridx = a->compressedrow.rindex;
1493: for (i=0; i<m; i++) {
1494: n = ii[i+1] - ii[i];
1495: aj = a->j + ii[i];
1496: aa = a->a + ii[i];
1497: sum = y[*ridx];
1498: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1499: z[*ridx++] = sum;
1500: }
1501: } else { /* do not use compressed row format */
1502: ii = a->i;
1503: for (i=0; i<m; i++) {
1504: n = ii[i+1] - ii[i];
1505: aj = a->j + ii[i];
1506: aa = a->a + ii[i];
1507: sum = y[i];
1508: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1509: z[i] = sum;
1510: }
1511: }
1512: PetscLogFlops(2.0*a->nz);
1513: VecRestoreArrayRead(xx,&x);
1514: VecRestoreArrayPair(yy,zz,&y,&z);
1515: return(0);
1516: }
1518: #include <../src/mat/impls/aij/seq/ftn-kernels/fmultadd.h>
1519: PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1520: {
1521: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1522: PetscScalar *y,*z;
1523: const PetscScalar *x;
1524: const MatScalar *aa;
1525: PetscErrorCode ierr;
1526: const PetscInt *aj,*ii,*ridx=NULL;
1527: PetscInt m = A->rmap->n,n,i;
1528: PetscScalar sum;
1529: PetscBool usecprow=a->compressedrow.use;
1532: VecGetArrayRead(xx,&x);
1533: VecGetArrayPair(yy,zz,&y,&z);
1534: if (usecprow) { /* use compressed row format */
1535: if (zz != yy) {
1536: PetscArraycpy(z,y,m);
1537: }
1538: m = a->compressedrow.nrows;
1539: ii = a->compressedrow.i;
1540: ridx = a->compressedrow.rindex;
1541: for (i=0; i<m; i++) {
1542: n = ii[i+1] - ii[i];
1543: aj = a->j + ii[i];
1544: aa = a->a + ii[i];
1545: sum = y[*ridx];
1546: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1547: z[*ridx++] = sum;
1548: }
1549: } else { /* do not use compressed row format */
1550: ii = a->i;
1551: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
1552: aj = a->j;
1553: aa = a->a;
1554: fortranmultaddaij_(&m,x,ii,aj,aa,y,z);
1555: #else
1556: for (i=0; i<m; i++) {
1557: n = ii[i+1] - ii[i];
1558: aj = a->j + ii[i];
1559: aa = a->a + ii[i];
1560: sum = y[i];
1561: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1562: z[i] = sum;
1563: }
1564: #endif
1565: }
1566: PetscLogFlops(2.0*a->nz);
1567: VecRestoreArrayRead(xx,&x);
1568: VecRestoreArrayPair(yy,zz,&y,&z);
1569: return(0);
1570: }
1572: /*
1573: Adds diagonal pointers to sparse matrix structure.
1574: */
1575: PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A)
1576: {
1577: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1579: PetscInt i,j,m = A->rmap->n;
1582: if (!a->diag) {
1583: PetscMalloc1(m,&a->diag);
1584: PetscLogObjectMemory((PetscObject)A, m*sizeof(PetscInt));
1585: }
1586: for (i=0; i<A->rmap->n; i++) {
1587: a->diag[i] = a->i[i+1];
1588: for (j=a->i[i]; j<a->i[i+1]; j++) {
1589: if (a->j[j] == i) {
1590: a->diag[i] = j;
1591: break;
1592: }
1593: }
1594: }
1595: return(0);
1596: }
1598: PetscErrorCode MatShift_SeqAIJ(Mat A,PetscScalar v)
1599: {
1600: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1601: const PetscInt *diag = (const PetscInt*)a->diag;
1602: const PetscInt *ii = (const PetscInt*) a->i;
1603: PetscInt i,*mdiag = NULL;
1604: PetscErrorCode ierr;
1605: PetscInt cnt = 0; /* how many diagonals are missing */
1608: if (!A->preallocated || !a->nz) {
1609: MatSeqAIJSetPreallocation(A,1,NULL);
1610: MatShift_Basic(A,v);
1611: return(0);
1612: }
1614: if (a->diagonaldense) {
1615: cnt = 0;
1616: } else {
1617: PetscCalloc1(A->rmap->n,&mdiag);
1618: for (i=0; i<A->rmap->n; i++) {
1619: if (diag[i] >= ii[i+1]) {
1620: cnt++;
1621: mdiag[i] = 1;
1622: }
1623: }
1624: }
1625: if (!cnt) {
1626: MatShift_Basic(A,v);
1627: } else {
1628: PetscScalar *olda = a->a; /* preserve pointers to current matrix nonzeros structure and values */
1629: PetscInt *oldj = a->j, *oldi = a->i;
1630: PetscBool singlemalloc = a->singlemalloc,free_a = a->free_a,free_ij = a->free_ij;
1632: a->a = NULL;
1633: a->j = NULL;
1634: a->i = NULL;
1635: /* increase the values in imax for each row where a diagonal is being inserted then reallocate the matrix data structures */
1636: for (i=0; i<A->rmap->n; i++) {
1637: a->imax[i] += mdiag[i];
1638: a->imax[i] = PetscMin(a->imax[i],A->cmap->n);
1639: }
1640: MatSeqAIJSetPreallocation_SeqAIJ(A,0,a->imax);
1642: /* copy old values into new matrix data structure */
1643: for (i=0; i<A->rmap->n; i++) {
1644: MatSetValues(A,1,&i,a->imax[i] - mdiag[i],&oldj[oldi[i]],&olda[oldi[i]],ADD_VALUES);
1645: if (i < A->cmap->n) {
1646: MatSetValue(A,i,i,v,ADD_VALUES);
1647: }
1648: }
1649: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
1650: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
1651: if (singlemalloc) {
1652: PetscFree3(olda,oldj,oldi);
1653: } else {
1654: if (free_a) {PetscFree(olda);}
1655: if (free_ij) {PetscFree(oldj);}
1656: if (free_ij) {PetscFree(oldi);}
1657: }
1658: }
1659: PetscFree(mdiag);
1660: a->diagonaldense = PETSC_TRUE;
1661: return(0);
1662: }
1664: /*
1665: Checks for missing diagonals
1666: */
1667: PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscBool *missing,PetscInt *d)
1668: {
1669: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1670: PetscInt *diag,*ii = a->i,i;
1674: *missing = PETSC_FALSE;
1675: if (A->rmap->n > 0 && !ii) {
1676: *missing = PETSC_TRUE;
1677: if (d) *d = 0;
1678: PetscInfo(A,"Matrix has no entries therefore is missing diagonal\n");
1679: } else {
1680: PetscInt n;
1681: n = PetscMin(A->rmap->n, A->cmap->n);
1682: diag = a->diag;
1683: for (i=0; i<n; i++) {
1684: if (diag[i] >= ii[i+1]) {
1685: *missing = PETSC_TRUE;
1686: if (d) *d = i;
1687: PetscInfo1(A,"Matrix is missing diagonal number %D\n",i);
1688: break;
1689: }
1690: }
1691: }
1692: return(0);
1693: }
1695: #include <petscblaslapack.h>
1696: #include <petsc/private/kernels/blockinvert.h>
1698: /*
1699: Note that values is allocated externally by the PC and then passed into this routine
1700: */
1701: PetscErrorCode MatInvertVariableBlockDiagonal_SeqAIJ(Mat A,PetscInt nblocks,const PetscInt *bsizes,PetscScalar *diag)
1702: {
1703: PetscErrorCode ierr;
1704: PetscInt n = A->rmap->n, i, ncnt = 0, *indx,j,bsizemax = 0,*v_pivots;
1705: PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE;
1706: const PetscReal shift = 0.0;
1707: PetscInt ipvt[5];
1708: PetscScalar work[25],*v_work;
1711: allowzeropivot = PetscNot(A->erroriffailure);
1712: for (i=0; i<nblocks; i++) ncnt += bsizes[i];
1713: if (ncnt != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Total blocksizes %D doesn't match number matrix rows %D",ncnt,n);
1714: for (i=0; i<nblocks; i++) {
1715: bsizemax = PetscMax(bsizemax,bsizes[i]);
1716: }
1717: PetscMalloc1(bsizemax,&indx);
1718: if (bsizemax > 7) {
1719: PetscMalloc2(bsizemax,&v_work,bsizemax,&v_pivots);
1720: }
1721: ncnt = 0;
1722: for (i=0; i<nblocks; i++) {
1723: for (j=0; j<bsizes[i]; j++) indx[j] = ncnt+j;
1724: MatGetValues(A,bsizes[i],indx,bsizes[i],indx,diag);
1725: switch (bsizes[i]) {
1726: case 1:
1727: *diag = 1.0/(*diag);
1728: break;
1729: case 2:
1730: PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected);
1731: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1732: PetscKernel_A_gets_transpose_A_2(diag);
1733: break;
1734: case 3:
1735: PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected);
1736: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1737: PetscKernel_A_gets_transpose_A_3(diag);
1738: break;
1739: case 4:
1740: PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected);
1741: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1742: PetscKernel_A_gets_transpose_A_4(diag);
1743: break;
1744: case 5:
1745: PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected);
1746: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1747: PetscKernel_A_gets_transpose_A_5(diag);
1748: break;
1749: case 6:
1750: PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected);
1751: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1752: PetscKernel_A_gets_transpose_A_6(diag);
1753: break;
1754: case 7:
1755: PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected);
1756: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1757: PetscKernel_A_gets_transpose_A_7(diag);
1758: break;
1759: default:
1760: PetscKernel_A_gets_inverse_A(bsizes[i],diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected);
1761: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1762: PetscKernel_A_gets_transpose_A_N(diag,bsizes[i]);
1763: }
1764: ncnt += bsizes[i];
1765: diag += bsizes[i]*bsizes[i];
1766: }
1767: if (bsizemax > 7) {
1768: PetscFree2(v_work,v_pivots);
1769: }
1770: PetscFree(indx);
1771: return(0);
1772: }
1774: /*
1775: Negative shift indicates do not generate an error if there is a zero diagonal, just invert it anyways
1776: */
1777: PetscErrorCode MatInvertDiagonal_SeqAIJ(Mat A,PetscScalar omega,PetscScalar fshift)
1778: {
1779: Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data;
1781: PetscInt i,*diag,m = A->rmap->n;
1782: MatScalar *v = a->a;
1783: PetscScalar *idiag,*mdiag;
1786: if (a->idiagvalid) return(0);
1787: MatMarkDiagonal_SeqAIJ(A);
1788: diag = a->diag;
1789: if (!a->idiag) {
1790: PetscMalloc3(m,&a->idiag,m,&a->mdiag,m,&a->ssor_work);
1791: PetscLogObjectMemory((PetscObject)A, 3*m*sizeof(PetscScalar));
1792: v = a->a;
1793: }
1794: mdiag = a->mdiag;
1795: idiag = a->idiag;
1797: if (omega == 1.0 && PetscRealPart(fshift) <= 0.0) {
1798: for (i=0; i<m; i++) {
1799: mdiag[i] = v[diag[i]];
1800: if (!PetscAbsScalar(mdiag[i])) { /* zero diagonal */
1801: if (PetscRealPart(fshift)) {
1802: PetscInfo1(A,"Zero diagonal on row %D\n",i);
1803: A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1804: A->factorerror_zeropivot_value = 0.0;
1805: A->factorerror_zeropivot_row = i;
1806: } else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %D",i);
1807: }
1808: idiag[i] = 1.0/v[diag[i]];
1809: }
1810: PetscLogFlops(m);
1811: } else {
1812: for (i=0; i<m; i++) {
1813: mdiag[i] = v[diag[i]];
1814: idiag[i] = omega/(fshift + v[diag[i]]);
1815: }
1816: PetscLogFlops(2.0*m);
1817: }
1818: a->idiagvalid = PETSC_TRUE;
1819: return(0);
1820: }
1822: #include <../src/mat/impls/aij/seq/ftn-kernels/frelax.h>
1823: PetscErrorCode MatSOR_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
1824: {
1825: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1826: PetscScalar *x,d,sum,*t,scale;
1827: const MatScalar *v,*idiag=0,*mdiag;
1828: const PetscScalar *b, *bs,*xb, *ts;
1829: PetscErrorCode ierr;
1830: PetscInt n,m = A->rmap->n,i;
1831: const PetscInt *idx,*diag;
1834: its = its*lits;
1836: if (fshift != a->fshift || omega != a->omega) a->idiagvalid = PETSC_FALSE; /* must recompute idiag[] */
1837: if (!a->idiagvalid) {MatInvertDiagonal_SeqAIJ(A,omega,fshift);}
1838: a->fshift = fshift;
1839: a->omega = omega;
1841: diag = a->diag;
1842: t = a->ssor_work;
1843: idiag = a->idiag;
1844: mdiag = a->mdiag;
1846: VecGetArray(xx,&x);
1847: VecGetArrayRead(bb,&b);
1848: /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */
1849: if (flag == SOR_APPLY_UPPER) {
1850: /* apply (U + D/omega) to the vector */
1851: bs = b;
1852: for (i=0; i<m; i++) {
1853: d = fshift + mdiag[i];
1854: n = a->i[i+1] - diag[i] - 1;
1855: idx = a->j + diag[i] + 1;
1856: v = a->a + diag[i] + 1;
1857: sum = b[i]*d/omega;
1858: PetscSparseDensePlusDot(sum,bs,v,idx,n);
1859: x[i] = sum;
1860: }
1861: VecRestoreArray(xx,&x);
1862: VecRestoreArrayRead(bb,&b);
1863: PetscLogFlops(a->nz);
1864: return(0);
1865: }
1867: if (flag == SOR_APPLY_LOWER) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented");
1868: else if (flag & SOR_EISENSTAT) {
1869: /* Let A = L + U + D; where L is lower triangular,
1870: U is upper triangular, E = D/omega; This routine applies
1872: (L + E)^{-1} A (U + E)^{-1}
1874: to a vector efficiently using Eisenstat's trick.
1875: */
1876: scale = (2.0/omega) - 1.0;
1878: /* x = (E + U)^{-1} b */
1879: for (i=m-1; i>=0; i--) {
1880: n = a->i[i+1] - diag[i] - 1;
1881: idx = a->j + diag[i] + 1;
1882: v = a->a + diag[i] + 1;
1883: sum = b[i];
1884: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1885: x[i] = sum*idiag[i];
1886: }
1888: /* t = b - (2*E - D)x */
1889: v = a->a;
1890: for (i=0; i<m; i++) t[i] = b[i] - scale*(v[*diag++])*x[i];
1892: /* t = (E + L)^{-1}t */
1893: ts = t;
1894: diag = a->diag;
1895: for (i=0; i<m; i++) {
1896: n = diag[i] - a->i[i];
1897: idx = a->j + a->i[i];
1898: v = a->a + a->i[i];
1899: sum = t[i];
1900: PetscSparseDenseMinusDot(sum,ts,v,idx,n);
1901: t[i] = sum*idiag[i];
1902: /* x = x + t */
1903: x[i] += t[i];
1904: }
1906: PetscLogFlops(6.0*m-1 + 2.0*a->nz);
1907: VecRestoreArray(xx,&x);
1908: VecRestoreArrayRead(bb,&b);
1909: return(0);
1910: }
1911: if (flag & SOR_ZERO_INITIAL_GUESS) {
1912: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
1913: for (i=0; i<m; i++) {
1914: n = diag[i] - a->i[i];
1915: idx = a->j + a->i[i];
1916: v = a->a + a->i[i];
1917: sum = b[i];
1918: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1919: t[i] = sum;
1920: x[i] = sum*idiag[i];
1921: }
1922: xb = t;
1923: PetscLogFlops(a->nz);
1924: } else xb = b;
1925: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
1926: for (i=m-1; i>=0; i--) {
1927: n = a->i[i+1] - diag[i] - 1;
1928: idx = a->j + diag[i] + 1;
1929: v = a->a + diag[i] + 1;
1930: sum = xb[i];
1931: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1932: if (xb == b) {
1933: x[i] = sum*idiag[i];
1934: } else {
1935: x[i] = (1-omega)*x[i] + sum*idiag[i]; /* omega in idiag */
1936: }
1937: }
1938: PetscLogFlops(a->nz); /* assumes 1/2 in upper */
1939: }
1940: its--;
1941: }
1942: while (its--) {
1943: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
1944: for (i=0; i<m; i++) {
1945: /* lower */
1946: n = diag[i] - a->i[i];
1947: idx = a->j + a->i[i];
1948: v = a->a + a->i[i];
1949: sum = b[i];
1950: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1951: t[i] = sum; /* save application of the lower-triangular part */
1952: /* upper */
1953: n = a->i[i+1] - diag[i] - 1;
1954: idx = a->j + diag[i] + 1;
1955: v = a->a + diag[i] + 1;
1956: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1957: x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */
1958: }
1959: xb = t;
1960: PetscLogFlops(2.0*a->nz);
1961: } else xb = b;
1962: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
1963: for (i=m-1; i>=0; i--) {
1964: sum = xb[i];
1965: if (xb == b) {
1966: /* whole matrix (no checkpointing available) */
1967: n = a->i[i+1] - a->i[i];
1968: idx = a->j + a->i[i];
1969: v = a->a + a->i[i];
1970: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1971: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
1972: } else { /* lower-triangular part has been saved, so only apply upper-triangular */
1973: n = a->i[i+1] - diag[i] - 1;
1974: idx = a->j + diag[i] + 1;
1975: v = a->a + diag[i] + 1;
1976: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1977: x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */
1978: }
1979: }
1980: if (xb == b) {
1981: PetscLogFlops(2.0*a->nz);
1982: } else {
1983: PetscLogFlops(a->nz); /* assumes 1/2 in upper */
1984: }
1985: }
1986: }
1987: VecRestoreArray(xx,&x);
1988: VecRestoreArrayRead(bb,&b);
1989: return(0);
1990: }
1993: PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info)
1994: {
1995: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1998: info->block_size = 1.0;
1999: info->nz_allocated = a->maxnz;
2000: info->nz_used = a->nz;
2001: info->nz_unneeded = (a->maxnz - a->nz);
2002: info->assemblies = A->num_ass;
2003: info->mallocs = A->info.mallocs;
2004: info->memory = ((PetscObject)A)->mem;
2005: if (A->factortype) {
2006: info->fill_ratio_given = A->info.fill_ratio_given;
2007: info->fill_ratio_needed = A->info.fill_ratio_needed;
2008: info->factor_mallocs = A->info.factor_mallocs;
2009: } else {
2010: info->fill_ratio_given = 0;
2011: info->fill_ratio_needed = 0;
2012: info->factor_mallocs = 0;
2013: }
2014: return(0);
2015: }
2017: PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
2018: {
2019: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2020: PetscInt i,m = A->rmap->n - 1;
2021: PetscErrorCode ierr;
2022: const PetscScalar *xx;
2023: PetscScalar *bb;
2024: PetscInt d = 0;
2027: if (x && b) {
2028: VecGetArrayRead(x,&xx);
2029: VecGetArray(b,&bb);
2030: for (i=0; i<N; i++) {
2031: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2032: if (rows[i] >= A->cmap->n) continue;
2033: bb[rows[i]] = diag*xx[rows[i]];
2034: }
2035: VecRestoreArrayRead(x,&xx);
2036: VecRestoreArray(b,&bb);
2037: }
2039: if (a->keepnonzeropattern) {
2040: for (i=0; i<N; i++) {
2041: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2042: PetscArrayzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]);
2043: }
2044: if (diag != 0.0) {
2045: for (i=0; i<N; i++) {
2046: d = rows[i];
2047: if (rows[i] >= A->cmap->n) continue;
2048: if (a->diag[d] >= a->i[d+1]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in the zeroed row %D",d);
2049: }
2050: for (i=0; i<N; i++) {
2051: if (rows[i] >= A->cmap->n) continue;
2052: a->a[a->diag[rows[i]]] = diag;
2053: }
2054: }
2055: } else {
2056: if (diag != 0.0) {
2057: for (i=0; i<N; i++) {
2058: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2059: if (a->ilen[rows[i]] > 0) {
2060: if (rows[i] >= A->cmap->n) {
2061: a->ilen[rows[i]] = 0;
2062: } else {
2063: a->ilen[rows[i]] = 1;
2064: a->a[a->i[rows[i]]] = diag;
2065: a->j[a->i[rows[i]]] = rows[i];
2066: }
2067: } else if (rows[i] < A->cmap->n) { /* in case row was completely empty */
2068: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);
2069: }
2070: }
2071: } else {
2072: for (i=0; i<N; i++) {
2073: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2074: a->ilen[rows[i]] = 0;
2075: }
2076: }
2077: A->nonzerostate++;
2078: }
2079: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2080: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
2081: #endif
2082: (*A->ops->assemblyend)(A,MAT_FINAL_ASSEMBLY);
2083: return(0);
2084: }
2086: PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
2087: {
2088: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2089: PetscInt i,j,m = A->rmap->n - 1,d = 0;
2090: PetscErrorCode ierr;
2091: PetscBool missing,*zeroed,vecs = PETSC_FALSE;
2092: const PetscScalar *xx;
2093: PetscScalar *bb;
2096: if (x && b) {
2097: VecGetArrayRead(x,&xx);
2098: VecGetArray(b,&bb);
2099: vecs = PETSC_TRUE;
2100: }
2101: PetscCalloc1(A->rmap->n,&zeroed);
2102: for (i=0; i<N; i++) {
2103: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2104: PetscArrayzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]);
2106: zeroed[rows[i]] = PETSC_TRUE;
2107: }
2108: for (i=0; i<A->rmap->n; i++) {
2109: if (!zeroed[i]) {
2110: for (j=a->i[i]; j<a->i[i+1]; j++) {
2111: if (a->j[j] < A->rmap->n && zeroed[a->j[j]]) {
2112: if (vecs) bb[i] -= a->a[j]*xx[a->j[j]];
2113: a->a[j] = 0.0;
2114: }
2115: }
2116: } else if (vecs && i < A->cmap->N) bb[i] = diag*xx[i];
2117: }
2118: if (x && b) {
2119: VecRestoreArrayRead(x,&xx);
2120: VecRestoreArray(b,&bb);
2121: }
2122: PetscFree(zeroed);
2123: if (diag != 0.0) {
2124: MatMissingDiagonal_SeqAIJ(A,&missing,&d);
2125: if (missing) {
2126: for (i=0; i<N; i++) {
2127: if (rows[i] >= A->cmap->N) continue;
2128: if (a->nonew && rows[i] >= d) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D (%D)",d,rows[i]);
2129: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);
2130: }
2131: } else {
2132: for (i=0; i<N; i++) {
2133: a->a[a->diag[rows[i]]] = diag;
2134: }
2135: }
2136: }
2137: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2138: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
2139: #endif
2140: (*A->ops->assemblyend)(A,MAT_FINAL_ASSEMBLY);
2141: return(0);
2142: }
2144: PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
2145: {
2146: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2147: PetscInt *itmp;
2150: if (row < 0 || row >= A->rmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row);
2152: *nz = a->i[row+1] - a->i[row];
2153: if (v) *v = a->a + a->i[row];
2154: if (idx) {
2155: itmp = a->j + a->i[row];
2156: if (*nz) *idx = itmp;
2157: else *idx = 0;
2158: }
2159: return(0);
2160: }
2162: /* remove this function? */
2163: PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
2164: {
2166: return(0);
2167: }
2169: PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm)
2170: {
2171: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2172: MatScalar *v = a->a;
2173: PetscReal sum = 0.0;
2175: PetscInt i,j;
2178: if (type == NORM_FROBENIUS) {
2179: #if defined(PETSC_USE_REAL___FP16)
2180: PetscBLASInt one = 1,nz = a->nz;
2181: *nrm = BLASnrm2_(&nz,v,&one);
2182: #else
2183: for (i=0; i<a->nz; i++) {
2184: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
2185: }
2186: *nrm = PetscSqrtReal(sum);
2187: #endif
2188: PetscLogFlops(2*a->nz);
2189: } else if (type == NORM_1) {
2190: PetscReal *tmp;
2191: PetscInt *jj = a->j;
2192: PetscCalloc1(A->cmap->n+1,&tmp);
2193: *nrm = 0.0;
2194: for (j=0; j<a->nz; j++) {
2195: tmp[*jj++] += PetscAbsScalar(*v); v++;
2196: }
2197: for (j=0; j<A->cmap->n; j++) {
2198: if (tmp[j] > *nrm) *nrm = tmp[j];
2199: }
2200: PetscFree(tmp);
2201: PetscLogFlops(PetscMax(a->nz-1,0));
2202: } else if (type == NORM_INFINITY) {
2203: *nrm = 0.0;
2204: for (j=0; j<A->rmap->n; j++) {
2205: v = a->a + a->i[j];
2206: sum = 0.0;
2207: for (i=0; i<a->i[j+1]-a->i[j]; i++) {
2208: sum += PetscAbsScalar(*v); v++;
2209: }
2210: if (sum > *nrm) *nrm = sum;
2211: }
2212: PetscLogFlops(PetscMax(a->nz-1,0));
2213: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm");
2214: return(0);
2215: }
2217: /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */
2218: PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B)
2219: {
2221: PetscInt i,j,anzj;
2222: Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b;
2223: PetscInt an=A->cmap->N,am=A->rmap->N;
2224: PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j;
2227: /* Allocate space for symbolic transpose info and work array */
2228: PetscCalloc1(an+1,&ati);
2229: PetscMalloc1(ai[am],&atj);
2230: PetscMalloc1(an,&atfill);
2232: /* Walk through aj and count ## of non-zeros in each row of A^T. */
2233: /* Note: offset by 1 for fast conversion into csr format. */
2234: for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1;
2235: /* Form ati for csr format of A^T. */
2236: for (i=0;i<an;i++) ati[i+1] += ati[i];
2238: /* Copy ati into atfill so we have locations of the next free space in atj */
2239: PetscArraycpy(atfill,ati,an);
2241: /* Walk through A row-wise and mark nonzero entries of A^T. */
2242: for (i=0;i<am;i++) {
2243: anzj = ai[i+1] - ai[i];
2244: for (j=0;j<anzj;j++) {
2245: atj[atfill[*aj]] = i;
2246: atfill[*aj++] += 1;
2247: }
2248: }
2250: /* Clean up temporary space and complete requests. */
2251: PetscFree(atfill);
2252: MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B);
2253: MatSetBlockSizes(*B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));
2254: MatSetType(*B,((PetscObject)A)->type_name);
2256: b = (Mat_SeqAIJ*)((*B)->data);
2257: b->free_a = PETSC_FALSE;
2258: b->free_ij = PETSC_TRUE;
2259: b->nonew = 0;
2260: return(0);
2261: }
2263: PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
2264: {
2265: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) B->data;
2266: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr;
2267: MatScalar *va,*vb;
2269: PetscInt ma,na,mb,nb, i;
2272: MatGetSize(A,&ma,&na);
2273: MatGetSize(B,&mb,&nb);
2274: if (ma!=nb || na!=mb) {
2275: *f = PETSC_FALSE;
2276: return(0);
2277: }
2278: aii = aij->i; bii = bij->i;
2279: adx = aij->j; bdx = bij->j;
2280: va = aij->a; vb = bij->a;
2281: PetscMalloc1(ma,&aptr);
2282: PetscMalloc1(mb,&bptr);
2283: for (i=0; i<ma; i++) aptr[i] = aii[i];
2284: for (i=0; i<mb; i++) bptr[i] = bii[i];
2286: *f = PETSC_TRUE;
2287: for (i=0; i<ma; i++) {
2288: while (aptr[i]<aii[i+1]) {
2289: PetscInt idc,idr;
2290: PetscScalar vc,vr;
2291: /* column/row index/value */
2292: idc = adx[aptr[i]];
2293: idr = bdx[bptr[idc]];
2294: vc = va[aptr[i]];
2295: vr = vb[bptr[idc]];
2296: if (i!=idr || PetscAbsScalar(vc-vr) > tol) {
2297: *f = PETSC_FALSE;
2298: goto done;
2299: } else {
2300: aptr[i]++;
2301: if (B || i!=idc) bptr[idc]++;
2302: }
2303: }
2304: }
2305: done:
2306: PetscFree(aptr);
2307: PetscFree(bptr);
2308: return(0);
2309: }
2311: PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
2312: {
2313: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) B->data;
2314: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr;
2315: MatScalar *va,*vb;
2317: PetscInt ma,na,mb,nb, i;
2320: MatGetSize(A,&ma,&na);
2321: MatGetSize(B,&mb,&nb);
2322: if (ma!=nb || na!=mb) {
2323: *f = PETSC_FALSE;
2324: return(0);
2325: }
2326: aii = aij->i; bii = bij->i;
2327: adx = aij->j; bdx = bij->j;
2328: va = aij->a; vb = bij->a;
2329: PetscMalloc1(ma,&aptr);
2330: PetscMalloc1(mb,&bptr);
2331: for (i=0; i<ma; i++) aptr[i] = aii[i];
2332: for (i=0; i<mb; i++) bptr[i] = bii[i];
2334: *f = PETSC_TRUE;
2335: for (i=0; i<ma; i++) {
2336: while (aptr[i]<aii[i+1]) {
2337: PetscInt idc,idr;
2338: PetscScalar vc,vr;
2339: /* column/row index/value */
2340: idc = adx[aptr[i]];
2341: idr = bdx[bptr[idc]];
2342: vc = va[aptr[i]];
2343: vr = vb[bptr[idc]];
2344: if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) {
2345: *f = PETSC_FALSE;
2346: goto done;
2347: } else {
2348: aptr[i]++;
2349: if (B || i!=idc) bptr[idc]++;
2350: }
2351: }
2352: }
2353: done:
2354: PetscFree(aptr);
2355: PetscFree(bptr);
2356: return(0);
2357: }
2359: PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f)
2360: {
2364: MatIsTranspose_SeqAIJ(A,A,tol,f);
2365: return(0);
2366: }
2368: PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f)
2369: {
2373: MatIsHermitianTranspose_SeqAIJ(A,A,tol,f);
2374: return(0);
2375: }
2377: PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr)
2378: {
2379: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2380: const PetscScalar *l,*r;
2381: PetscScalar x;
2382: MatScalar *v;
2383: PetscErrorCode ierr;
2384: PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz;
2385: const PetscInt *jj;
2388: if (ll) {
2389: /* The local size is used so that VecMPI can be passed to this routine
2390: by MatDiagonalScale_MPIAIJ */
2391: VecGetLocalSize(ll,&m);
2392: if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length");
2393: VecGetArrayRead(ll,&l);
2394: v = a->a;
2395: for (i=0; i<m; i++) {
2396: x = l[i];
2397: M = a->i[i+1] - a->i[i];
2398: for (j=0; j<M; j++) (*v++) *= x;
2399: }
2400: VecRestoreArrayRead(ll,&l);
2401: PetscLogFlops(nz);
2402: }
2403: if (rr) {
2404: VecGetLocalSize(rr,&n);
2405: if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length");
2406: VecGetArrayRead(rr,&r);
2407: v = a->a; jj = a->j;
2408: for (i=0; i<nz; i++) (*v++) *= r[*jj++];
2409: VecRestoreArrayRead(rr,&r);
2410: PetscLogFlops(nz);
2411: }
2412: MatSeqAIJInvalidateDiagonal(A);
2413: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2414: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
2415: #endif
2416: return(0);
2417: }
2419: PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B)
2420: {
2421: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c;
2423: PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens;
2424: PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi;
2425: const PetscInt *irow,*icol;
2426: PetscInt nrows,ncols;
2427: PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen;
2428: MatScalar *a_new,*mat_a;
2429: Mat C;
2430: PetscBool stride;
2434: ISGetIndices(isrow,&irow);
2435: ISGetLocalSize(isrow,&nrows);
2436: ISGetLocalSize(iscol,&ncols);
2438: PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride);
2439: if (stride) {
2440: ISStrideGetInfo(iscol,&first,&step);
2441: } else {
2442: first = 0;
2443: step = 0;
2444: }
2445: if (stride && step == 1) {
2446: /* special case of contiguous rows */
2447: PetscMalloc2(nrows,&lens,nrows,&starts);
2448: /* loop over new rows determining lens and starting points */
2449: for (i=0; i<nrows; i++) {
2450: kstart = ai[irow[i]];
2451: kend = kstart + ailen[irow[i]];
2452: starts[i] = kstart;
2453: for (k=kstart; k<kend; k++) {
2454: if (aj[k] >= first) {
2455: starts[i] = k;
2456: break;
2457: }
2458: }
2459: sum = 0;
2460: while (k < kend) {
2461: if (aj[k++] >= first+ncols) break;
2462: sum++;
2463: }
2464: lens[i] = sum;
2465: }
2466: /* create submatrix */
2467: if (scall == MAT_REUSE_MATRIX) {
2468: PetscInt n_cols,n_rows;
2469: MatGetSize(*B,&n_rows,&n_cols);
2470: if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size");
2471: MatZeroEntries(*B);
2472: C = *B;
2473: } else {
2474: PetscInt rbs,cbs;
2475: MatCreate(PetscObjectComm((PetscObject)A),&C);
2476: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
2477: ISGetBlockSize(isrow,&rbs);
2478: ISGetBlockSize(iscol,&cbs);
2479: MatSetBlockSizes(C,rbs,cbs);
2480: MatSetType(C,((PetscObject)A)->type_name);
2481: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
2482: }
2483: c = (Mat_SeqAIJ*)C->data;
2485: /* loop over rows inserting into submatrix */
2486: a_new = c->a;
2487: j_new = c->j;
2488: i_new = c->i;
2490: for (i=0; i<nrows; i++) {
2491: ii = starts[i];
2492: lensi = lens[i];
2493: for (k=0; k<lensi; k++) {
2494: *j_new++ = aj[ii+k] - first;
2495: }
2496: PetscArraycpy(a_new,a->a + starts[i],lensi);
2497: a_new += lensi;
2498: i_new[i+1] = i_new[i] + lensi;
2499: c->ilen[i] = lensi;
2500: }
2501: PetscFree2(lens,starts);
2502: } else {
2503: ISGetIndices(iscol,&icol);
2504: PetscCalloc1(oldcols,&smap);
2505: PetscMalloc1(1+nrows,&lens);
2506: for (i=0; i<ncols; i++) {
2507: #if defined(PETSC_USE_DEBUG)
2508: if (icol[i] >= oldcols) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Requesting column beyond largest column icol[%D] %D <= A->cmap->n %D",i,icol[i],oldcols);
2509: #endif
2510: smap[icol[i]] = i+1;
2511: }
2513: /* determine lens of each row */
2514: for (i=0; i<nrows; i++) {
2515: kstart = ai[irow[i]];
2516: kend = kstart + a->ilen[irow[i]];
2517: lens[i] = 0;
2518: for (k=kstart; k<kend; k++) {
2519: if (smap[aj[k]]) {
2520: lens[i]++;
2521: }
2522: }
2523: }
2524: /* Create and fill new matrix */
2525: if (scall == MAT_REUSE_MATRIX) {
2526: PetscBool equal;
2528: c = (Mat_SeqAIJ*)((*B)->data);
2529: if ((*B)->rmap->n != nrows || (*B)->cmap->n != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size");
2530: PetscArraycmp(c->ilen,lens,(*B)->rmap->n,&equal);
2531: if (!equal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros");
2532: PetscArrayzero(c->ilen,(*B)->rmap->n);
2533: C = *B;
2534: } else {
2535: PetscInt rbs,cbs;
2536: MatCreate(PetscObjectComm((PetscObject)A),&C);
2537: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
2538: ISGetBlockSize(isrow,&rbs);
2539: ISGetBlockSize(iscol,&cbs);
2540: MatSetBlockSizes(C,rbs,cbs);
2541: MatSetType(C,((PetscObject)A)->type_name);
2542: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
2543: }
2544: c = (Mat_SeqAIJ*)(C->data);
2545: for (i=0; i<nrows; i++) {
2546: row = irow[i];
2547: kstart = ai[row];
2548: kend = kstart + a->ilen[row];
2549: mat_i = c->i[i];
2550: mat_j = c->j + mat_i;
2551: mat_a = c->a + mat_i;
2552: mat_ilen = c->ilen + i;
2553: for (k=kstart; k<kend; k++) {
2554: if ((tcol=smap[a->j[k]])) {
2555: *mat_j++ = tcol - 1;
2556: *mat_a++ = a->a[k];
2557: (*mat_ilen)++;
2559: }
2560: }
2561: }
2562: /* Free work space */
2563: ISRestoreIndices(iscol,&icol);
2564: PetscFree(smap);
2565: PetscFree(lens);
2566: /* sort */
2567: for (i = 0; i < nrows; i++) {
2568: PetscInt ilen;
2570: mat_i = c->i[i];
2571: mat_j = c->j + mat_i;
2572: mat_a = c->a + mat_i;
2573: ilen = c->ilen[i];
2574: PetscSortIntWithScalarArray(ilen,mat_j,mat_a);
2575: }
2576: }
2577: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2578: MatPinToCPU(C,A->pinnedtocpu);
2579: #endif
2580: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
2581: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
2583: ISRestoreIndices(isrow,&irow);
2584: *B = C;
2585: return(0);
2586: }
2588: PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat)
2589: {
2591: Mat B;
2594: if (scall == MAT_INITIAL_MATRIX) {
2595: MatCreate(subComm,&B);
2596: MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n);
2597: MatSetBlockSizesFromMats(B,mat,mat);
2598: MatSetType(B,MATSEQAIJ);
2599: MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE);
2600: *subMat = B;
2601: } else {
2602: MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN);
2603: }
2604: return(0);
2605: }
2607: PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info)
2608: {
2609: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
2611: Mat outA;
2612: PetscBool row_identity,col_identity;
2615: if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu");
2617: ISIdentity(row,&row_identity);
2618: ISIdentity(col,&col_identity);
2620: outA = inA;
2621: outA->factortype = MAT_FACTOR_LU;
2622: PetscFree(inA->solvertype);
2623: PetscStrallocpy(MATSOLVERPETSC,&inA->solvertype);
2625: PetscObjectReference((PetscObject)row);
2626: ISDestroy(&a->row);
2628: a->row = row;
2630: PetscObjectReference((PetscObject)col);
2631: ISDestroy(&a->col);
2633: a->col = col;
2635: /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */
2636: ISDestroy(&a->icol);
2637: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
2638: PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);
2640: if (!a->solve_work) { /* this matrix may have been factored before */
2641: PetscMalloc1(inA->rmap->n+1,&a->solve_work);
2642: PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar));
2643: }
2645: MatMarkDiagonal_SeqAIJ(inA);
2646: if (row_identity && col_identity) {
2647: MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info);
2648: } else {
2649: MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info);
2650: }
2651: return(0);
2652: }
2654: PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha)
2655: {
2656: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
2657: PetscScalar oalpha = alpha;
2659: PetscBLASInt one = 1,bnz;
2662: PetscBLASIntCast(a->nz,&bnz);
2663: PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&oalpha,a->a,&one));
2664: PetscLogFlops(a->nz);
2665: MatSeqAIJInvalidateDiagonal(inA);
2666: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2667: if (inA->offloadmask != PETSC_OFFLOAD_UNALLOCATED) inA->offloadmask = PETSC_OFFLOAD_CPU;
2668: #endif
2669: return(0);
2670: }
2672: PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt *submatj)
2673: {
2675: PetscInt i;
2678: if (!submatj->id) { /* delete data that are linked only to submats[id=0] */
2679: PetscFree4(submatj->sbuf1,submatj->ptr,submatj->tmp,submatj->ctr);
2681: for (i=0; i<submatj->nrqr; ++i) {
2682: PetscFree(submatj->sbuf2[i]);
2683: }
2684: PetscFree3(submatj->sbuf2,submatj->req_size,submatj->req_source1);
2686: if (submatj->rbuf1) {
2687: PetscFree(submatj->rbuf1[0]);
2688: PetscFree(submatj->rbuf1);
2689: }
2691: for (i=0; i<submatj->nrqs; ++i) {
2692: PetscFree(submatj->rbuf3[i]);
2693: }
2694: PetscFree3(submatj->req_source2,submatj->rbuf2,submatj->rbuf3);
2695: PetscFree(submatj->pa);
2696: }
2698: #if defined(PETSC_USE_CTABLE)
2699: PetscTableDestroy((PetscTable*)&submatj->rmap);
2700: if (submatj->cmap_loc) {PetscFree(submatj->cmap_loc);}
2701: PetscFree(submatj->rmap_loc);
2702: #else
2703: PetscFree(submatj->rmap);
2704: #endif
2706: if (!submatj->allcolumns) {
2707: #if defined(PETSC_USE_CTABLE)
2708: PetscTableDestroy((PetscTable*)&submatj->cmap);
2709: #else
2710: PetscFree(submatj->cmap);
2711: #endif
2712: }
2713: PetscFree(submatj->row2proc);
2715: PetscFree(submatj);
2716: return(0);
2717: }
2719: PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat C)
2720: {
2722: Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data;
2723: Mat_SubSppt *submatj = c->submatis1;
2726: (*submatj->destroy)(C);
2727: MatDestroySubMatrix_Private(submatj);
2728: return(0);
2729: }
2731: PetscErrorCode MatDestroySubMatrices_SeqAIJ(PetscInt n,Mat *mat[])
2732: {
2734: PetscInt i;
2735: Mat C;
2736: Mat_SeqAIJ *c;
2737: Mat_SubSppt *submatj;
2740: for (i=0; i<n; i++) {
2741: C = (*mat)[i];
2742: c = (Mat_SeqAIJ*)C->data;
2743: submatj = c->submatis1;
2744: if (submatj) {
2745: if (--((PetscObject)C)->refct <= 0) {
2746: (*submatj->destroy)(C);
2747: MatDestroySubMatrix_Private(submatj);
2748: PetscFree(C->defaultvectype);
2749: PetscLayoutDestroy(&C->rmap);
2750: PetscLayoutDestroy(&C->cmap);
2751: PetscHeaderDestroy(&C);
2752: }
2753: } else {
2754: MatDestroy(&C);
2755: }
2756: }
2758: /* Destroy Dummy submatrices created for reuse */
2759: MatDestroySubMatrices_Dummy(n,mat);
2761: PetscFree(*mat);
2762: return(0);
2763: }
2765: PetscErrorCode MatCreateSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
2766: {
2768: PetscInt i;
2771: if (scall == MAT_INITIAL_MATRIX) {
2772: PetscCalloc1(n+1,B);
2773: }
2775: for (i=0; i<n; i++) {
2776: MatCreateSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);
2777: }
2778: return(0);
2779: }
2781: PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov)
2782: {
2783: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2785: PetscInt row,i,j,k,l,m,n,*nidx,isz,val;
2786: const PetscInt *idx;
2787: PetscInt start,end,*ai,*aj;
2788: PetscBT table;
2791: m = A->rmap->n;
2792: ai = a->i;
2793: aj = a->j;
2795: if (ov < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used");
2797: PetscMalloc1(m+1,&nidx);
2798: PetscBTCreate(m,&table);
2800: for (i=0; i<is_max; i++) {
2801: /* Initialize the two local arrays */
2802: isz = 0;
2803: PetscBTMemzero(m,table);
2805: /* Extract the indices, assume there can be duplicate entries */
2806: ISGetIndices(is[i],&idx);
2807: ISGetLocalSize(is[i],&n);
2809: /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */
2810: for (j=0; j<n; ++j) {
2811: if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j];
2812: }
2813: ISRestoreIndices(is[i],&idx);
2814: ISDestroy(&is[i]);
2816: k = 0;
2817: for (j=0; j<ov; j++) { /* for each overlap */
2818: n = isz;
2819: for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */
2820: row = nidx[k];
2821: start = ai[row];
2822: end = ai[row+1];
2823: for (l = start; l<end; l++) {
2824: val = aj[l];
2825: if (!PetscBTLookupSet(table,val)) nidx[isz++] = val;
2826: }
2827: }
2828: }
2829: ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i));
2830: }
2831: PetscBTDestroy(&table);
2832: PetscFree(nidx);
2833: return(0);
2834: }
2836: /* -------------------------------------------------------------- */
2837: PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B)
2838: {
2839: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2841: PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n;
2842: const PetscInt *row,*col;
2843: PetscInt *cnew,j,*lens;
2844: IS icolp,irowp;
2845: PetscInt *cwork = NULL;
2846: PetscScalar *vwork = NULL;
2849: ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);
2850: ISGetIndices(irowp,&row);
2851: ISInvertPermutation(colp,PETSC_DECIDE,&icolp);
2852: ISGetIndices(icolp,&col);
2854: /* determine lengths of permuted rows */
2855: PetscMalloc1(m+1,&lens);
2856: for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i];
2857: MatCreate(PetscObjectComm((PetscObject)A),B);
2858: MatSetSizes(*B,m,n,m,n);
2859: MatSetBlockSizesFromMats(*B,A,A);
2860: MatSetType(*B,((PetscObject)A)->type_name);
2861: MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);
2862: PetscFree(lens);
2864: PetscMalloc1(n,&cnew);
2865: for (i=0; i<m; i++) {
2866: MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
2867: for (j=0; j<nz; j++) cnew[j] = col[cwork[j]];
2868: MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);
2869: MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
2870: }
2871: PetscFree(cnew);
2873: (*B)->assembled = PETSC_FALSE;
2875: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2876: MatPinToCPU(*B,A->pinnedtocpu);
2877: #endif
2878: MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);
2879: MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);
2880: ISRestoreIndices(irowp,&row);
2881: ISRestoreIndices(icolp,&col);
2882: ISDestroy(&irowp);
2883: ISDestroy(&icolp);
2884: if (rowp == colp) {
2885: if (A->symmetric) {
2886: MatSetOption(*B,MAT_SYMMETRIC,PETSC_TRUE);
2887: }
2888: if (A->hermitian) {
2889: MatSetOption(*B,MAT_HERMITIAN,PETSC_TRUE);
2890: }
2891: }
2892: return(0);
2893: }
2895: PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str)
2896: {
2900: /* If the two matrices have the same copy implementation, use fast copy. */
2901: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
2902: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2903: Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data;
2905: if (a->i[A->rmap->n] != b->i[B->rmap->n]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzeros in two matrices are different");
2906: PetscArraycpy(b->a,a->a,a->i[A->rmap->n]);
2907: PetscObjectStateIncrease((PetscObject)B);
2908: } else {
2909: MatCopy_Basic(A,B,str);
2910: }
2911: return(0);
2912: }
2914: PetscErrorCode MatSetUp_SeqAIJ(Mat A)
2915: {
2919: MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,0);
2920: return(0);
2921: }
2923: PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[])
2924: {
2925: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2928: *array = a->a;
2929: return(0);
2930: }
2932: PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[])
2933: {
2935: *array = NULL;
2936: return(0);
2937: }
2939: /*
2940: Computes the number of nonzeros per row needed for preallocation when X and Y
2941: have different nonzero structure.
2942: */
2943: PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt m,const PetscInt *xi,const PetscInt *xj,const PetscInt *yi,const PetscInt *yj,PetscInt *nnz)
2944: {
2945: PetscInt i,j,k,nzx,nzy;
2948: /* Set the number of nonzeros in the new matrix */
2949: for (i=0; i<m; i++) {
2950: const PetscInt *xjj = xj+xi[i],*yjj = yj+yi[i];
2951: nzx = xi[i+1] - xi[i];
2952: nzy = yi[i+1] - yi[i];
2953: nnz[i] = 0;
2954: for (j=0,k=0; j<nzx; j++) { /* Point in X */
2955: for (; k<nzy && yjj[k]<xjj[j]; k++) nnz[i]++; /* Catch up to X */
2956: if (k<nzy && yjj[k]==xjj[j]) k++; /* Skip duplicate */
2957: nnz[i]++;
2958: }
2959: for (; k<nzy; k++) nnz[i]++;
2960: }
2961: return(0);
2962: }
2964: PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz)
2965: {
2966: PetscInt m = Y->rmap->N;
2967: Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data;
2968: Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data;
2972: /* Set the number of nonzeros in the new matrix */
2973: MatAXPYGetPreallocation_SeqX_private(m,x->i,x->j,y->i,y->j,nnz);
2974: return(0);
2975: }
2977: PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
2978: {
2980: Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data;
2981: PetscBLASInt one=1,bnz;
2984: PetscBLASIntCast(x->nz,&bnz);
2985: if (str == SAME_NONZERO_PATTERN) {
2986: PetscScalar alpha = a;
2987: PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
2988: MatSeqAIJInvalidateDiagonal(Y);
2989: PetscObjectStateIncrease((PetscObject)Y);
2990: /* the MatAXPY_Basic* subroutines calls MatAssembly, so the matrix on the GPU
2991: will be updated */
2992: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
2993: if (Y->offloadmask != PETSC_OFFLOAD_UNALLOCATED) {
2994: Y->offloadmask = PETSC_OFFLOAD_CPU;
2995: }
2996: #endif
2997: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2998: MatAXPY_Basic(Y,a,X,str);
2999: } else {
3000: Mat B;
3001: PetscInt *nnz;
3002: PetscMalloc1(Y->rmap->N,&nnz);
3003: MatCreate(PetscObjectComm((PetscObject)Y),&B);
3004: PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);
3005: MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);
3006: MatSetBlockSizesFromMats(B,Y,Y);
3007: MatSetType(B,(MatType) ((PetscObject)Y)->type_name);
3008: MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);
3009: MatSeqAIJSetPreallocation(B,0,nnz);
3010: MatAXPY_BasicWithPreallocation(B,Y,a,X,str);
3011: MatHeaderReplace(Y,&B);
3012: PetscFree(nnz);
3013: }
3014: return(0);
3015: }
3017: PetscErrorCode MatConjugate_SeqAIJ(Mat mat)
3018: {
3019: #if defined(PETSC_USE_COMPLEX)
3020: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3021: PetscInt i,nz;
3022: PetscScalar *a;
3025: nz = aij->nz;
3026: a = aij->a;
3027: for (i=0; i<nz; i++) a[i] = PetscConj(a[i]);
3028: #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA)
3029: if (mat->offloadmask != PETSC_OFFLOAD_UNALLOCATED) mat->offloadmask = PETSC_OFFLOAD_CPU;
3030: #endif
3031: #else
3033: #endif
3034: return(0);
3035: }
3037: PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3038: {
3039: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3041: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3042: PetscReal atmp;
3043: PetscScalar *x;
3044: MatScalar *aa;
3047: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3048: aa = a->a;
3049: ai = a->i;
3050: aj = a->j;
3052: VecSet(v,0.0);
3053: VecGetArray(v,&x);
3054: VecGetLocalSize(v,&n);
3055: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3056: for (i=0; i<m; i++) {
3057: ncols = ai[1] - ai[0]; ai++;
3058: x[i] = 0.0;
3059: for (j=0; j<ncols; j++) {
3060: atmp = PetscAbsScalar(*aa);
3061: if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;}
3062: aa++; aj++;
3063: }
3064: }
3065: VecRestoreArray(v,&x);
3066: return(0);
3067: }
3069: PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3070: {
3071: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3073: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3074: PetscScalar *x;
3075: MatScalar *aa;
3078: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3079: aa = a->a;
3080: ai = a->i;
3081: aj = a->j;
3083: VecSet(v,0.0);
3084: VecGetArray(v,&x);
3085: VecGetLocalSize(v,&n);
3086: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3087: for (i=0; i<m; i++) {
3088: ncols = ai[1] - ai[0]; ai++;
3089: if (ncols == A->cmap->n) { /* row is dense */
3090: x[i] = *aa; if (idx) idx[i] = 0;
3091: } else { /* row is sparse so already KNOW maximum is 0.0 or higher */
3092: x[i] = 0.0;
3093: if (idx) {
3094: idx[i] = 0; /* in case ncols is zero */
3095: for (j=0;j<ncols;j++) { /* find first implicit 0.0 in the row */
3096: if (aj[j] > j) {
3097: idx[i] = j;
3098: break;
3099: }
3100: }
3101: }
3102: }
3103: for (j=0; j<ncols; j++) {
3104: if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
3105: aa++; aj++;
3106: }
3107: }
3108: VecRestoreArray(v,&x);
3109: return(0);
3110: }
3112: PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3113: {
3114: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3116: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3117: PetscReal atmp;
3118: PetscScalar *x;
3119: MatScalar *aa;
3122: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3123: aa = a->a;
3124: ai = a->i;
3125: aj = a->j;
3127: VecSet(v,0.0);
3128: VecGetArray(v,&x);
3129: VecGetLocalSize(v,&n);
3130: if (n != A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector, %D vs. %D rows", A->rmap->n, n);
3131: for (i=0; i<m; i++) {
3132: ncols = ai[1] - ai[0]; ai++;
3133: if (ncols) {
3134: /* Get first nonzero */
3135: for (j = 0; j < ncols; j++) {
3136: atmp = PetscAbsScalar(aa[j]);
3137: if (atmp > 1.0e-12) {
3138: x[i] = atmp;
3139: if (idx) idx[i] = aj[j];
3140: break;
3141: }
3142: }
3143: if (j == ncols) {x[i] = PetscAbsScalar(*aa); if (idx) idx[i] = *aj;}
3144: } else {
3145: x[i] = 0.0; if (idx) idx[i] = 0;
3146: }
3147: for (j = 0; j < ncols; j++) {
3148: atmp = PetscAbsScalar(*aa);
3149: if (atmp > 1.0e-12 && PetscAbsScalar(x[i]) > atmp) {x[i] = atmp; if (idx) idx[i] = *aj;}
3150: aa++; aj++;
3151: }
3152: }
3153: VecRestoreArray(v,&x);
3154: return(0);
3155: }
3157: PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3158: {
3159: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3160: PetscErrorCode ierr;
3161: PetscInt i,j,m = A->rmap->n,ncols,n;
3162: const PetscInt *ai,*aj;
3163: PetscScalar *x;
3164: const MatScalar *aa;
3167: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3168: aa = a->a;
3169: ai = a->i;
3170: aj = a->j;
3172: VecSet(v,0.0);
3173: VecGetArray(v,&x);
3174: VecGetLocalSize(v,&n);
3175: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3176: for (i=0; i<m; i++) {
3177: ncols = ai[1] - ai[0]; ai++;
3178: if (ncols == A->cmap->n) { /* row is dense */
3179: x[i] = *aa; if (idx) idx[i] = 0;
3180: } else { /* row is sparse so already KNOW minimum is 0.0 or lower */
3181: x[i] = 0.0;
3182: if (idx) { /* find first implicit 0.0 in the row */
3183: idx[i] = 0; /* in case ncols is zero */
3184: for (j=0; j<ncols; j++) {
3185: if (aj[j] > j) {
3186: idx[i] = j;
3187: break;
3188: }
3189: }
3190: }
3191: }
3192: for (j=0; j<ncols; j++) {
3193: if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
3194: aa++; aj++;
3195: }
3196: }
3197: VecRestoreArray(v,&x);
3198: return(0);
3199: }
3201: PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values)
3202: {
3203: Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data;
3204: PetscErrorCode ierr;
3205: PetscInt i,bs = PetscAbs(A->rmap->bs),mbs = A->rmap->n/bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j;
3206: MatScalar *diag,work[25],*v_work;
3207: const PetscReal shift = 0.0;
3208: PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE;
3211: allowzeropivot = PetscNot(A->erroriffailure);
3212: if (a->ibdiagvalid) {
3213: if (values) *values = a->ibdiag;
3214: return(0);
3215: }
3216: MatMarkDiagonal_SeqAIJ(A);
3217: if (!a->ibdiag) {
3218: PetscMalloc1(bs2*mbs,&a->ibdiag);
3219: PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));
3220: }
3221: diag = a->ibdiag;
3222: if (values) *values = a->ibdiag;
3223: /* factor and invert each block */
3224: switch (bs) {
3225: case 1:
3226: for (i=0; i<mbs; i++) {
3227: MatGetValues(A,1,&i,1,&i,diag+i);
3228: if (PetscAbsScalar(diag[i] + shift) < PETSC_MACHINE_EPSILON) {
3229: if (allowzeropivot) {
3230: A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3231: A->factorerror_zeropivot_value = PetscAbsScalar(diag[i]);
3232: A->factorerror_zeropivot_row = i;
3233: PetscInfo3(A,"Zero pivot, row %D pivot %g tolerance %g\n",i,(double)PetscAbsScalar(diag[i]),(double)PETSC_MACHINE_EPSILON);
3234: } else SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot, row %D pivot %g tolerance %g",i,(double)PetscAbsScalar(diag[i]),(double)PETSC_MACHINE_EPSILON);
3235: }
3236: diag[i] = (PetscScalar)1.0 / (diag[i] + shift);
3237: }
3238: break;
3239: case 2:
3240: for (i=0; i<mbs; i++) {
3241: ij[0] = 2*i; ij[1] = 2*i + 1;
3242: MatGetValues(A,2,ij,2,ij,diag);
3243: PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected);
3244: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3245: PetscKernel_A_gets_transpose_A_2(diag);
3246: diag += 4;
3247: }
3248: break;
3249: case 3:
3250: for (i=0; i<mbs; i++) {
3251: ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2;
3252: MatGetValues(A,3,ij,3,ij,diag);
3253: PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected);
3254: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3255: PetscKernel_A_gets_transpose_A_3(diag);
3256: diag += 9;
3257: }
3258: break;
3259: case 4:
3260: for (i=0; i<mbs; i++) {
3261: ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3;
3262: MatGetValues(A,4,ij,4,ij,diag);
3263: PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected);
3264: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3265: PetscKernel_A_gets_transpose_A_4(diag);
3266: diag += 16;
3267: }
3268: break;
3269: case 5:
3270: for (i=0; i<mbs; i++) {
3271: ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4;
3272: MatGetValues(A,5,ij,5,ij,diag);
3273: PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected);
3274: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3275: PetscKernel_A_gets_transpose_A_5(diag);
3276: diag += 25;
3277: }
3278: break;
3279: case 6:
3280: for (i=0; i<mbs; i++) {
3281: ij[0] = 6*i; ij[1] = 6*i + 1; ij[2] = 6*i + 2; ij[3] = 6*i + 3; ij[4] = 6*i + 4; ij[5] = 6*i + 5;
3282: MatGetValues(A,6,ij,6,ij,diag);
3283: PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected);
3284: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3285: PetscKernel_A_gets_transpose_A_6(diag);
3286: diag += 36;
3287: }
3288: break;
3289: case 7:
3290: for (i=0; i<mbs; i++) {
3291: ij[0] = 7*i; ij[1] = 7*i + 1; ij[2] = 7*i + 2; ij[3] = 7*i + 3; ij[4] = 7*i + 4; ij[5] = 7*i + 5; ij[5] = 7*i + 6;
3292: MatGetValues(A,7,ij,7,ij,diag);
3293: PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected);
3294: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3295: PetscKernel_A_gets_transpose_A_7(diag);
3296: diag += 49;
3297: }
3298: break;
3299: default:
3300: PetscMalloc3(bs,&v_work,bs,&v_pivots,bs,&IJ);
3301: for (i=0; i<mbs; i++) {
3302: for (j=0; j<bs; j++) {
3303: IJ[j] = bs*i + j;
3304: }
3305: MatGetValues(A,bs,IJ,bs,IJ,diag);
3306: PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected);
3307: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3308: PetscKernel_A_gets_transpose_A_N(diag,bs);
3309: diag += bs2;
3310: }
3311: PetscFree3(v_work,v_pivots,IJ);
3312: }
3313: a->ibdiagvalid = PETSC_TRUE;
3314: return(0);
3315: }
3317: static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx)
3318: {
3320: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data;
3321: PetscScalar a;
3322: PetscInt m,n,i,j,col;
3325: if (!x->assembled) {
3326: MatGetSize(x,&m,&n);
3327: for (i=0; i<m; i++) {
3328: for (j=0; j<aij->imax[i]; j++) {
3329: PetscRandomGetValue(rctx,&a);
3330: col = (PetscInt)(n*PetscRealPart(a));
3331: MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);
3332: }
3333: }
3334: } else {
3335: for (i=0; i<aij->nz; i++) {PetscRandomGetValue(rctx,aij->a+i);}
3336: }
3337: MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);
3338: MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);
3339: return(0);
3340: }
3342: /* Like MatSetRandom_SeqAIJ, but do not set values on columns in range of [low, high) */
3343: PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat x,PetscInt low,PetscInt high,PetscRandom rctx)
3344: {
3346: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data;
3347: PetscScalar a;
3348: PetscInt m,n,i,j,col,nskip;
3351: nskip = high - low;
3352: MatGetSize(x,&m,&n);
3353: n -= nskip; /* shrink number of columns where nonzeros can be set */
3354: for (i=0; i<m; i++) {
3355: for (j=0; j<aij->imax[i]; j++) {
3356: PetscRandomGetValue(rctx,&a);
3357: col = (PetscInt)(n*PetscRealPart(a));
3358: if (col >= low) col += nskip; /* shift col rightward to skip the hole */
3359: MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);
3360: }
3361: }
3362: MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);
3363: MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);
3364: return(0);
3365: }
3368: /* -------------------------------------------------------------------*/
3369: static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ,
3370: MatGetRow_SeqAIJ,
3371: MatRestoreRow_SeqAIJ,
3372: MatMult_SeqAIJ,
3373: /* 4*/ MatMultAdd_SeqAIJ,
3374: MatMultTranspose_SeqAIJ,
3375: MatMultTransposeAdd_SeqAIJ,
3376: 0,
3377: 0,
3378: 0,
3379: /* 10*/ 0,
3380: MatLUFactor_SeqAIJ,
3381: 0,
3382: MatSOR_SeqAIJ,
3383: MatTranspose_SeqAIJ,
3384: /*1 5*/ MatGetInfo_SeqAIJ,
3385: MatEqual_SeqAIJ,
3386: MatGetDiagonal_SeqAIJ,
3387: MatDiagonalScale_SeqAIJ,
3388: MatNorm_SeqAIJ,
3389: /* 20*/ 0,
3390: MatAssemblyEnd_SeqAIJ,
3391: MatSetOption_SeqAIJ,
3392: MatZeroEntries_SeqAIJ,
3393: /* 24*/ MatZeroRows_SeqAIJ,
3394: 0,
3395: 0,
3396: 0,
3397: 0,
3398: /* 29*/ MatSetUp_SeqAIJ,
3399: 0,
3400: 0,
3401: 0,
3402: 0,
3403: /* 34*/ MatDuplicate_SeqAIJ,
3404: 0,
3405: 0,
3406: MatILUFactor_SeqAIJ,
3407: 0,
3408: /* 39*/ MatAXPY_SeqAIJ,
3409: MatCreateSubMatrices_SeqAIJ,
3410: MatIncreaseOverlap_SeqAIJ,
3411: MatGetValues_SeqAIJ,
3412: MatCopy_SeqAIJ,
3413: /* 44*/ MatGetRowMax_SeqAIJ,
3414: MatScale_SeqAIJ,
3415: MatShift_SeqAIJ,
3416: MatDiagonalSet_SeqAIJ,
3417: MatZeroRowsColumns_SeqAIJ,
3418: /* 49*/ MatSetRandom_SeqAIJ,
3419: MatGetRowIJ_SeqAIJ,
3420: MatRestoreRowIJ_SeqAIJ,
3421: MatGetColumnIJ_SeqAIJ,
3422: MatRestoreColumnIJ_SeqAIJ,
3423: /* 54*/ MatFDColoringCreate_SeqXAIJ,
3424: 0,
3425: 0,
3426: MatPermute_SeqAIJ,
3427: 0,
3428: /* 59*/ 0,
3429: MatDestroy_SeqAIJ,
3430: MatView_SeqAIJ,
3431: 0,
3432: MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ,
3433: /* 64*/ MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ,
3434: MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ,
3435: 0,
3436: 0,
3437: 0,
3438: /* 69*/ MatGetRowMaxAbs_SeqAIJ,
3439: MatGetRowMinAbs_SeqAIJ,
3440: 0,
3441: 0,
3442: 0,
3443: /* 74*/ 0,
3444: MatFDColoringApply_AIJ,
3445: 0,
3446: 0,
3447: 0,
3448: /* 79*/ MatFindZeroDiagonals_SeqAIJ,
3449: 0,
3450: 0,
3451: 0,
3452: MatLoad_SeqAIJ,
3453: /* 84*/ MatIsSymmetric_SeqAIJ,
3454: MatIsHermitian_SeqAIJ,
3455: 0,
3456: 0,
3457: 0,
3458: /* 89*/ MatMatMult_SeqAIJ_SeqAIJ,
3459: MatMatMultSymbolic_SeqAIJ_SeqAIJ,
3460: MatMatMultNumeric_SeqAIJ_SeqAIJ,
3461: MatPtAP_SeqAIJ_SeqAIJ,
3462: MatPtAPSymbolic_SeqAIJ_SeqAIJ_SparseAxpy,
3463: /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy,
3464: MatMatTransposeMult_SeqAIJ_SeqAIJ,
3465: MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ,
3466: MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ,
3467: 0,
3468: /* 99*/ 0,
3469: 0,
3470: 0,
3471: MatConjugate_SeqAIJ,
3472: 0,
3473: /*104*/ MatSetValuesRow_SeqAIJ,
3474: MatRealPart_SeqAIJ,
3475: MatImaginaryPart_SeqAIJ,
3476: 0,
3477: 0,
3478: /*109*/ MatMatSolve_SeqAIJ,
3479: 0,
3480: MatGetRowMin_SeqAIJ,
3481: 0,
3482: MatMissingDiagonal_SeqAIJ,
3483: /*114*/ 0,
3484: 0,
3485: 0,
3486: 0,
3487: 0,
3488: /*119*/ 0,
3489: 0,
3490: 0,
3491: 0,
3492: MatGetMultiProcBlock_SeqAIJ,
3493: /*124*/ MatFindNonzeroRows_SeqAIJ,
3494: MatGetColumnNorms_SeqAIJ,
3495: MatInvertBlockDiagonal_SeqAIJ,
3496: MatInvertVariableBlockDiagonal_SeqAIJ,
3497: 0,
3498: /*129*/ 0,
3499: MatTransposeMatMult_SeqAIJ_SeqAIJ,
3500: MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ,
3501: MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ,
3502: MatTransposeColoringCreate_SeqAIJ,
3503: /*134*/ MatTransColoringApplySpToDen_SeqAIJ,
3504: MatTransColoringApplyDenToSp_SeqAIJ,
3505: MatRARt_SeqAIJ_SeqAIJ,
3506: MatRARtSymbolic_SeqAIJ_SeqAIJ,
3507: MatRARtNumeric_SeqAIJ_SeqAIJ,
3508: /*139*/0,
3509: 0,
3510: 0,
3511: MatFDColoringSetUp_SeqXAIJ,
3512: MatFindOffBlockDiagonalEntries_SeqAIJ,
3513: /*144*/MatCreateMPIMatConcatenateSeqMat_SeqAIJ,
3514: MatDestroySubMatrices_SeqAIJ
3515: };
3517: PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices)
3518: {
3519: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3520: PetscInt i,nz,n;
3523: nz = aij->maxnz;
3524: n = mat->rmap->n;
3525: for (i=0; i<nz; i++) {
3526: aij->j[i] = indices[i];
3527: }
3528: aij->nz = nz;
3529: for (i=0; i<n; i++) {
3530: aij->ilen[i] = aij->imax[i];
3531: }
3532: return(0);
3533: }
3535: /*
3536: * When a sparse matrix has many zero columns, we should compact them out to save the space
3537: * This happens in MatPtAPSymbolic_MPIAIJ_MPIAIJ_scalable()
3538: * */
3539: PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat mat, ISLocalToGlobalMapping *mapping)
3540: {
3541: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3542: PetscTable gid1_lid1;
3543: PetscTablePosition tpos;
3544: PetscInt gid,lid,i,j,ncols,ec;
3545: PetscInt *garray;
3546: PetscErrorCode ierr;
3551: /* use a table */
3552: PetscTableCreate(mat->rmap->n,mat->cmap->N+1,&gid1_lid1);
3553: ec = 0;
3554: for (i=0; i<mat->rmap->n; i++) {
3555: ncols = aij->i[i+1] - aij->i[i];
3556: for (j=0; j<ncols; j++) {
3557: PetscInt data,gid1 = aij->j[aij->i[i] + j] + 1;
3558: PetscTableFind(gid1_lid1,gid1,&data);
3559: if (!data) {
3560: /* one based table */
3561: PetscTableAdd(gid1_lid1,gid1,++ec,INSERT_VALUES);
3562: }
3563: }
3564: }
3565: /* form array of columns we need */
3566: PetscMalloc1(ec+1,&garray);
3567: PetscTableGetHeadPosition(gid1_lid1,&tpos);
3568: while (tpos) {
3569: PetscTableGetNext(gid1_lid1,&tpos,&gid,&lid);
3570: gid--;
3571: lid--;
3572: garray[lid] = gid;
3573: }
3574: PetscSortInt(ec,garray); /* sort, and rebuild */
3575: PetscTableRemoveAll(gid1_lid1);
3576: for (i=0; i<ec; i++) {
3577: PetscTableAdd(gid1_lid1,garray[i]+1,i+1,INSERT_VALUES);
3578: }
3579: /* compact out the extra columns in B */
3580: for (i=0; i<mat->rmap->n; i++) {
3581: ncols = aij->i[i+1] - aij->i[i];
3582: for (j=0; j<ncols; j++) {
3583: PetscInt gid1 = aij->j[aij->i[i] + j] + 1;
3584: PetscTableFind(gid1_lid1,gid1,&lid);
3585: lid--;
3586: aij->j[aij->i[i] + j] = lid;
3587: }
3588: }
3589: PetscLayoutDestroy(&mat->cmap);
3590: PetscLayoutCreateFromSizes(PetscObjectComm((PetscObject)mat),ec,ec,1,&mat->cmap);
3591: PetscTableDestroy(&gid1_lid1);
3592: ISLocalToGlobalMappingCreate(PETSC_COMM_SELF,mat->cmap->bs,mat->cmap->n,garray,PETSC_OWN_POINTER,mapping);
3593: ISLocalToGlobalMappingSetType(*mapping,ISLOCALTOGLOBALMAPPINGHASH);
3594: return(0);
3595: }
3597: /*@
3598: MatSeqAIJSetColumnIndices - Set the column indices for all the rows
3599: in the matrix.
3601: Input Parameters:
3602: + mat - the SeqAIJ matrix
3603: - indices - the column indices
3605: Level: advanced
3607: Notes:
3608: This can be called if you have precomputed the nonzero structure of the
3609: matrix and want to provide it to the matrix object to improve the performance
3610: of the MatSetValues() operation.
3612: You MUST have set the correct numbers of nonzeros per row in the call to
3613: MatCreateSeqAIJ(), and the columns indices MUST be sorted.
3615: MUST be called before any calls to MatSetValues();
3617: The indices should start with zero, not one.
3619: @*/
3620: PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices)
3621: {
3627: PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));
3628: return(0);
3629: }
3631: /* ----------------------------------------------------------------------------------------*/
3633: PetscErrorCode MatStoreValues_SeqAIJ(Mat mat)
3634: {
3635: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3637: size_t nz = aij->i[mat->rmap->n];
3640: if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3642: /* allocate space for values if not already there */
3643: if (!aij->saved_values) {
3644: PetscMalloc1(nz+1,&aij->saved_values);
3645: PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));
3646: }
3648: /* copy values over */
3649: PetscArraycpy(aij->saved_values,aij->a,nz);
3650: return(0);
3651: }
3653: /*@
3654: MatStoreValues - Stashes a copy of the matrix values; this allows, for
3655: example, reuse of the linear part of a Jacobian, while recomputing the
3656: nonlinear portion.
3658: Collect on Mat
3660: Input Parameters:
3661: . mat - the matrix (currently only AIJ matrices support this option)
3663: Level: advanced
3665: Common Usage, with SNESSolve():
3666: $ Create Jacobian matrix
3667: $ Set linear terms into matrix
3668: $ Apply boundary conditions to matrix, at this time matrix must have
3669: $ final nonzero structure (i.e. setting the nonlinear terms and applying
3670: $ boundary conditions again will not change the nonzero structure
3671: $ MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);
3672: $ MatStoreValues(mat);
3673: $ Call SNESSetJacobian() with matrix
3674: $ In your Jacobian routine
3675: $ MatRetrieveValues(mat);
3676: $ Set nonlinear terms in matrix
3678: Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself:
3679: $ // build linear portion of Jacobian
3680: $ MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);
3681: $ MatStoreValues(mat);
3682: $ loop over nonlinear iterations
3683: $ MatRetrieveValues(mat);
3684: $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian
3685: $ // call MatAssemblyBegin/End() on matrix
3686: $ Solve linear system with Jacobian
3687: $ endloop
3689: Notes:
3690: Matrix must already be assemblied before calling this routine
3691: Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before
3692: calling this routine.
3694: When this is called multiple times it overwrites the previous set of stored values
3695: and does not allocated additional space.
3697: .seealso: MatRetrieveValues()
3699: @*/
3700: PetscErrorCode MatStoreValues(Mat mat)
3701: {
3706: if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
3707: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3708: PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));
3709: return(0);
3710: }
3712: PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat)
3713: {
3714: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3716: PetscInt nz = aij->i[mat->rmap->n];
3719: if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3720: if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
3721: /* copy values over */
3722: PetscArraycpy(aij->a,aij->saved_values,nz);
3723: return(0);
3724: }
3726: /*@
3727: MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for
3728: example, reuse of the linear part of a Jacobian, while recomputing the
3729: nonlinear portion.
3731: Collect on Mat
3733: Input Parameters:
3734: . mat - the matrix (currently only AIJ matrices support this option)
3736: Level: advanced
3738: .seealso: MatStoreValues()
3740: @*/
3741: PetscErrorCode MatRetrieveValues(Mat mat)
3742: {
3747: if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
3748: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3749: PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));
3750: return(0);
3751: }
3754: /* --------------------------------------------------------------------------------*/
3755: /*@C
3756: MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format
3757: (the default parallel PETSc format). For good matrix assembly performance
3758: the user should preallocate the matrix storage by setting the parameter nz
3759: (or the array nnz). By setting these parameters accurately, performance
3760: during matrix assembly can be increased by more than a factor of 50.
3762: Collective
3764: Input Parameters:
3765: + comm - MPI communicator, set to PETSC_COMM_SELF
3766: . m - number of rows
3767: . n - number of columns
3768: . nz - number of nonzeros per row (same for all rows)
3769: - nnz - array containing the number of nonzeros in the various rows
3770: (possibly different for each row) or NULL
3772: Output Parameter:
3773: . A - the matrix
3775: It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
3776: MatXXXXSetPreallocation() paradigm instead of this routine directly.
3777: [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
3779: Notes:
3780: If nnz is given then nz is ignored
3782: The AIJ format (also called the Yale sparse matrix format or
3783: compressed row storage), is fully compatible with standard Fortran 77
3784: storage. That is, the stored row and column indices can begin at
3785: either one (as in Fortran) or zero. See the users' manual for details.
3787: Specify the preallocated storage with either nz or nnz (not both).
3788: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3789: allocation. For large problems you MUST preallocate memory or you
3790: will get TERRIBLE performance, see the users' manual chapter on matrices.
3792: By default, this format uses inodes (identical nodes) when possible, to
3793: improve numerical efficiency of matrix-vector products and solves. We
3794: search for consecutive rows with the same nonzero structure, thereby
3795: reusing matrix information to achieve increased efficiency.
3797: Options Database Keys:
3798: + -mat_no_inode - Do not use inodes
3799: - -mat_inode_limit <limit> - Sets inode limit (max limit=5)
3801: Level: intermediate
3803: .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
3805: @*/
3806: PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
3807: {
3811: MatCreate(comm,A);
3812: MatSetSizes(*A,m,n,m,n);
3813: MatSetType(*A,MATSEQAIJ);
3814: MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);
3815: return(0);
3816: }
3818: /*@C
3819: MatSeqAIJSetPreallocation - For good matrix assembly performance
3820: the user should preallocate the matrix storage by setting the parameter nz
3821: (or the array nnz). By setting these parameters accurately, performance
3822: during matrix assembly can be increased by more than a factor of 50.
3824: Collective
3826: Input Parameters:
3827: + B - The matrix
3828: . nz - number of nonzeros per row (same for all rows)
3829: - nnz - array containing the number of nonzeros in the various rows
3830: (possibly different for each row) or NULL
3832: Notes:
3833: If nnz is given then nz is ignored
3835: The AIJ format (also called the Yale sparse matrix format or
3836: compressed row storage), is fully compatible with standard Fortran 77
3837: storage. That is, the stored row and column indices can begin at
3838: either one (as in Fortran) or zero. See the users' manual for details.
3840: Specify the preallocated storage with either nz or nnz (not both).
3841: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3842: allocation. For large problems you MUST preallocate memory or you
3843: will get TERRIBLE performance, see the users' manual chapter on matrices.
3845: You can call MatGetInfo() to get information on how effective the preallocation was;
3846: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3847: You can also run with the option -info and look for messages with the string
3848: malloc in them to see if additional memory allocation was needed.
3850: Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix
3851: entries or columns indices
3853: By default, this format uses inodes (identical nodes) when possible, to
3854: improve numerical efficiency of matrix-vector products and solves. We
3855: search for consecutive rows with the same nonzero structure, thereby
3856: reusing matrix information to achieve increased efficiency.
3858: Options Database Keys:
3859: + -mat_no_inode - Do not use inodes
3860: - -mat_inode_limit <limit> - Sets inode limit (max limit=5)
3862: Level: intermediate
3864: .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo()
3866: @*/
3867: PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[])
3868: {
3874: PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));
3875: return(0);
3876: }
3878: PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz)
3879: {
3880: Mat_SeqAIJ *b;
3881: PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE;
3883: PetscInt i;
3886: if (nz >= 0 || nnz) realalloc = PETSC_TRUE;
3887: if (nz == MAT_SKIP_ALLOCATION) {
3888: skipallocation = PETSC_TRUE;
3889: nz = 0;
3890: }
3891: PetscLayoutSetUp(B->rmap);
3892: PetscLayoutSetUp(B->cmap);
3894: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
3895: if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %D",nz);
3896: #if defined(PETSC_USE_DEBUG)
3897: if (nnz) {
3898: for (i=0; i<B->rmap->n; i++) {
3899: if (nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %D value %D",i,nnz[i]);
3900: if (nnz[i] > B->cmap->n) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than row length: local row %D value %d rowlength %D",i,nnz[i],B->cmap->n);
3901: }
3902: }
3903: #endif
3905: B->preallocated = PETSC_TRUE;
3907: b = (Mat_SeqAIJ*)B->data;
3909: if (!skipallocation) {
3910: if (!b->imax) {
3911: PetscMalloc1(B->rmap->n,&b->imax);
3912: PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt));
3913: }
3914: if (!b->ilen) {
3915: /* b->ilen will count nonzeros in each row so far. */
3916: PetscCalloc1(B->rmap->n,&b->ilen);
3917: PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt));
3918: } else {
3919: PetscMemzero(b->ilen,B->rmap->n*sizeof(PetscInt));
3920: }
3921: if (!b->ipre) {
3922: PetscMalloc1(B->rmap->n,&b->ipre);
3923: PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt));
3924: }
3925: if (!nnz) {
3926: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10;
3927: else if (nz < 0) nz = 1;
3928: nz = PetscMin(nz,B->cmap->n);
3929: for (i=0; i<B->rmap->n; i++) b->imax[i] = nz;
3930: nz = nz*B->rmap->n;
3931: } else {
3932: PetscInt64 nz64 = 0;
3933: for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz64 += nnz[i];}
3934: PetscIntCast(nz64,&nz);
3935: }
3937: /* allocate the matrix space */
3938: /* FIXME: should B's old memory be unlogged? */
3939: MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);
3940: if (B->structure_only) {
3941: PetscMalloc1(nz,&b->j);
3942: PetscMalloc1(B->rmap->n+1,&b->i);
3943: PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*sizeof(PetscInt));
3944: } else {
3945: PetscMalloc3(nz,&b->a,nz,&b->j,B->rmap->n+1,&b->i);
3946: PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));
3947: }
3948: b->i[0] = 0;
3949: for (i=1; i<B->rmap->n+1; i++) {
3950: b->i[i] = b->i[i-1] + b->imax[i-1];
3951: }
3952: if (B->structure_only) {
3953: b->singlemalloc = PETSC_FALSE;
3954: b->free_a = PETSC_FALSE;
3955: } else {
3956: b->singlemalloc = PETSC_TRUE;
3957: b->free_a = PETSC_TRUE;
3958: }
3959: b->free_ij = PETSC_TRUE;
3960: } else {
3961: b->free_a = PETSC_FALSE;
3962: b->free_ij = PETSC_FALSE;
3963: }
3965: if (b->ipre && nnz != b->ipre && b->imax) {
3966: /* reserve user-requested sparsity */
3967: PetscArraycpy(b->ipre,b->imax,B->rmap->n);
3968: }
3971: b->nz = 0;
3972: b->maxnz = nz;
3973: B->info.nz_unneeded = (double)b->maxnz;
3974: if (realalloc) {
3975: MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);
3976: }
3977: B->was_assembled = PETSC_FALSE;
3978: B->assembled = PETSC_FALSE;
3979: return(0);
3980: }
3983: PetscErrorCode MatResetPreallocation_SeqAIJ(Mat A)
3984: {
3985: Mat_SeqAIJ *a;
3986: PetscInt i;
3992: /* Check local size. If zero, then return */
3993: if (!A->rmap->n) return(0);
3995: a = (Mat_SeqAIJ*)A->data;
3996: /* if no saved info, we error out */
3997: if (!a->ipre) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"No saved preallocation info \n");
3999: if (!a->i || !a->j || !a->a || !a->imax || !a->ilen) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"Memory info is incomplete, and can not reset preallocation \n");
4001: PetscArraycpy(a->imax,a->ipre,A->rmap->n);
4002: PetscArrayzero(a->ilen,A->rmap->n);
4003: a->i[0] = 0;
4004: for (i=1; i<A->rmap->n+1; i++) {
4005: a->i[i] = a->i[i-1] + a->imax[i-1];
4006: }
4007: A->preallocated = PETSC_TRUE;
4008: a->nz = 0;
4009: a->maxnz = a->i[A->rmap->n];
4010: A->info.nz_unneeded = (double)a->maxnz;
4011: A->was_assembled = PETSC_FALSE;
4012: A->assembled = PETSC_FALSE;
4013: return(0);
4014: }
4016: /*@
4017: MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format.
4019: Input Parameters:
4020: + B - the matrix
4021: . i - the indices into j for the start of each row (starts with zero)
4022: . j - the column indices for each row (starts with zero) these must be sorted for each row
4023: - v - optional values in the matrix
4025: Level: developer
4027: The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays()
4029: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), MATSEQAIJ
4030: @*/
4031: PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[])
4032: {
4038: PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));
4039: return(0);
4040: }
4042: PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[])
4043: {
4044: PetscInt i;
4045: PetscInt m,n;
4046: PetscInt nz;
4047: PetscInt *nnz, nz_max = 0;
4051: if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]);
4053: PetscLayoutSetUp(B->rmap);
4054: PetscLayoutSetUp(B->cmap);
4056: MatGetSize(B, &m, &n);
4057: PetscMalloc1(m+1, &nnz);
4058: for (i = 0; i < m; i++) {
4059: nz = Ii[i+1]- Ii[i];
4060: nz_max = PetscMax(nz_max, nz);
4061: if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz);
4062: nnz[i] = nz;
4063: }
4064: MatSeqAIJSetPreallocation(B, 0, nnz);
4065: PetscFree(nnz);
4067: for (i = 0; i < m; i++) {
4068: MatSetValues_SeqAIJ(B, 1, &i, Ii[i+1] - Ii[i], J+Ii[i], v ? v + Ii[i] : NULL, INSERT_VALUES);
4069: }
4071: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
4072: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
4074: MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
4075: return(0);
4076: }
4078: #include <../src/mat/impls/dense/seq/dense.h>
4079: #include <petsc/private/kernels/petscaxpy.h>
4081: /*
4082: Computes (B'*A')' since computing B*A directly is untenable
4084: n p p
4085: ( ) ( ) ( )
4086: m ( A ) * n ( B ) = m ( C )
4087: ( ) ( ) ( )
4089: */
4090: PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C)
4091: {
4092: PetscErrorCode ierr;
4093: Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data;
4094: Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data;
4095: Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data;
4096: PetscInt i,n,m,q,p;
4097: const PetscInt *ii,*idx;
4098: const PetscScalar *b,*a,*a_q;
4099: PetscScalar *c,*c_q;
4102: m = A->rmap->n;
4103: n = A->cmap->n;
4104: p = B->cmap->n;
4105: a = sub_a->v;
4106: b = sub_b->a;
4107: c = sub_c->v;
4108: PetscArrayzero(c,m*p);
4110: ii = sub_b->i;
4111: idx = sub_b->j;
4112: for (i=0; i<n; i++) {
4113: q = ii[i+1] - ii[i];
4114: while (q-->0) {
4115: c_q = c + m*(*idx);
4116: a_q = a + m*i;
4117: PetscKernelAXPY(c_q,*b,a_q,m);
4118: idx++;
4119: b++;
4120: }
4121: }
4122: return(0);
4123: }
4125: PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C)
4126: {
4128: PetscInt m=A->rmap->n,n=B->cmap->n;
4129: Mat Cmat;
4132: if (A->cmap->n != B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"A->cmap->n %D != B->rmap->n %D\n",A->cmap->n,B->rmap->n);
4133: MatCreate(PetscObjectComm((PetscObject)A),&Cmat);
4134: MatSetSizes(Cmat,m,n,m,n);
4135: MatSetBlockSizesFromMats(Cmat,A,B);
4136: MatSetType(Cmat,MATSEQDENSE);
4137: MatSeqDenseSetPreallocation(Cmat,NULL);
4139: Cmat->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ;
4141: *C = Cmat;
4142: return(0);
4143: }
4145: /* ----------------------------------------------------------------*/
4146: PETSC_INTERN PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C)
4147: {
4151: if (scall == MAT_INITIAL_MATRIX) {
4152: PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);
4153: MatMatMultSymbolic_SeqDense_SeqAIJ(A,B,fill,C);
4154: PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);
4155: }
4156: PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);
4157: MatMatMultNumeric_SeqDense_SeqAIJ(A,B,*C);
4158: PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);
4159: return(0);
4160: }
4163: /*MC
4164: MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices,
4165: based on compressed sparse row format.
4167: Options Database Keys:
4168: . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions()
4170: Level: beginner
4172: Notes:
4173: MatSetValues() may be called for this matrix type with a NULL argument for the numerical values,
4174: in this case the values associated with the rows and columns one passes in are set to zero
4175: in the matrix
4177: MatSetOptions(,MAT_STRUCTURE_ONLY,PETSC_TRUE) may be called for this matrix type. In this no
4178: space is allocated for the nonzero entries and any entries passed with MatSetValues() are ignored
4180: Developer Notes:
4181: It would be nice if all matrix formats supported passing NULL in for the numerical values
4183: .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType
4184: M*/
4186: /*MC
4187: MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices.
4189: This matrix type is identical to MATSEQAIJ when constructed with a single process communicator,
4190: and MATMPIAIJ otherwise. As a result, for single process communicators,
4191: MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation() is supported
4192: for communicators controlling multiple processes. It is recommended that you call both of
4193: the above preallocation routines for simplicity.
4195: Options Database Keys:
4196: . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions()
4198: Developer Notes:
4199: Subclasses include MATAIJCUSPARSE, MATAIJPERM, MATAIJSELL, MATAIJMKL, MATAIJCRL, and also automatically switches over to use inodes when
4200: enough exist.
4202: Level: beginner
4204: .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ
4205: M*/
4207: /*MC
4208: MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices.
4210: This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator,
4211: and MATMPIAIJCRL otherwise. As a result, for single process communicators,
4212: MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported
4213: for communicators controlling multiple processes. It is recommended that you call both of
4214: the above preallocation routines for simplicity.
4216: Options Database Keys:
4217: . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions()
4219: Level: beginner
4221: .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL
4222: M*/
4224: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*);
4225: #if defined(PETSC_HAVE_ELEMENTAL)
4226: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat,MatType,MatReuse,Mat*);
4227: #endif
4228: #if defined(PETSC_HAVE_HYPRE)
4229: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat A,MatType,MatReuse,Mat*);
4230: PETSC_INTERN PetscErrorCode MatMatMatMult_Transpose_AIJ_AIJ(Mat,Mat,Mat,MatReuse,PetscReal,Mat*);
4231: #endif
4232: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqDense(Mat,MatType,MatReuse,Mat*);
4234: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqSELL(Mat,MatType,MatReuse,Mat*);
4235: PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat,MatType,MatReuse,Mat*);
4236: PETSC_INTERN PetscErrorCode MatPtAP_IS_XAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
4238: /*@C
4239: MatSeqAIJGetArray - gives read/write access to the array where the data for a MATSEQAIJ matrix is stored
4241: Not Collective
4243: Input Parameter:
4244: . mat - a MATSEQAIJ matrix
4246: Output Parameter:
4247: . array - pointer to the data
4249: Level: intermediate
4251: .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90()
4252: @*/
4253: PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array)
4254: {
4258: PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));
4259: return(0);
4260: }
4262: /*@C
4263: MatSeqAIJGetArrayRead - gives read-only access to the array where the data for a MATSEQAIJ matrix is stored
4265: Not Collective
4267: Input Parameter:
4268: . mat - a MATSEQAIJ matrix
4270: Output Parameter:
4271: . array - pointer to the data
4273: Level: intermediate
4275: .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayRead()
4276: @*/
4277: PetscErrorCode MatSeqAIJGetArrayRead(Mat A,const PetscScalar **array)
4278: {
4279: #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_VIENNACL)
4280: PetscOffloadMask oval;
4281: #endif
4285: #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_VIENNACL)
4286: oval = A->offloadmask;
4287: #endif
4288: MatSeqAIJGetArray(A,(PetscScalar**)array);
4289: #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_VIENNACL)
4290: if (oval == PETSC_OFFLOAD_GPU || oval == PETSC_OFFLOAD_BOTH) A->offloadmask = PETSC_OFFLOAD_BOTH;
4291: #endif
4292: return(0);
4293: }
4295: /*@C
4296: MatSeqAIJRestoreArrayRead - restore the read-only access array obtained from MatSeqAIJGetArrayRead
4298: Not Collective
4300: Input Parameter:
4301: . mat - a MATSEQAIJ matrix
4303: Output Parameter:
4304: . array - pointer to the data
4306: Level: intermediate
4308: .seealso: MatSeqAIJGetArray(), MatSeqAIJGetArrayRead()
4309: @*/
4310: PetscErrorCode MatSeqAIJRestoreArrayRead(Mat A,const PetscScalar **array)
4311: {
4312: #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_VIENNACL)
4313: PetscOffloadMask oval;
4314: #endif
4318: #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_VIENNACL)
4319: oval = A->offloadmask;
4320: #endif
4321: MatSeqAIJRestoreArray(A,(PetscScalar**)array);
4322: #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_VIENNACL)
4323: A->offloadmask = oval;
4324: #endif
4325: return(0);
4326: }
4328: /*@C
4329: MatSeqAIJGetMaxRowNonzeros - returns the maximum number of nonzeros in any row
4331: Not Collective
4333: Input Parameter:
4334: . mat - a MATSEQAIJ matrix
4336: Output Parameter:
4337: . nz - the maximum number of nonzeros in any row
4339: Level: intermediate
4341: .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90()
4342: @*/
4343: PetscErrorCode MatSeqAIJGetMaxRowNonzeros(Mat A,PetscInt *nz)
4344: {
4345: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data;
4348: *nz = aij->rmax;
4349: return(0);
4350: }
4352: /*@C
4353: MatSeqAIJRestoreArray - returns access to the array where the data for a MATSEQAIJ matrix is stored obtained by MatSeqAIJGetArray()
4355: Not Collective
4357: Input Parameters:
4358: + mat - a MATSEQAIJ matrix
4359: - array - pointer to the data
4361: Level: intermediate
4363: .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90()
4364: @*/
4365: PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array)
4366: {
4370: PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));
4371: return(0);
4372: }
4374: #if defined(PETSC_HAVE_CUDA)
4375: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCUSPARSE(Mat);
4376: #endif
4378: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B)
4379: {
4380: Mat_SeqAIJ *b;
4382: PetscMPIInt size;
4385: MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);
4386: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1");
4388: PetscNewLog(B,&b);
4390: B->data = (void*)b;
4392: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
4393: if (B->sortedfull) B->ops->setvalues = MatSetValues_SeqAIJ_SortedFull;
4395: b->row = 0;
4396: b->col = 0;
4397: b->icol = 0;
4398: b->reallocs = 0;
4399: b->ignorezeroentries = PETSC_FALSE;
4400: b->roworiented = PETSC_TRUE;
4401: b->nonew = 0;
4402: b->diag = 0;
4403: b->solve_work = 0;
4404: B->spptr = 0;
4405: b->saved_values = 0;
4406: b->idiag = 0;
4407: b->mdiag = 0;
4408: b->ssor_work = 0;
4409: b->omega = 1.0;
4410: b->fshift = 0.0;
4411: b->idiagvalid = PETSC_FALSE;
4412: b->ibdiagvalid = PETSC_FALSE;
4413: b->keepnonzeropattern = PETSC_FALSE;
4415: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
4416: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);
4417: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);
4419: #if defined(PETSC_HAVE_MATLAB_ENGINE)
4420: PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);
4421: PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);
4422: #endif
4424: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);
4425: PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);
4426: PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);
4427: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);
4428: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);
4429: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);
4430: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijsell_C",MatConvert_SeqAIJ_SeqAIJSELL);
4431: #if defined(PETSC_HAVE_MKL_SPARSE)
4432: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijmkl_C",MatConvert_SeqAIJ_SeqAIJMKL);
4433: #endif
4434: #if defined(PETSC_HAVE_CUDA)
4435: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcusparse_C",MatConvert_SeqAIJ_SeqAIJCUSPARSE);
4436: #endif
4437: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);
4438: #if defined(PETSC_HAVE_ELEMENTAL)
4439: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_elemental_C",MatConvert_SeqAIJ_Elemental);
4440: #endif
4441: #if defined(PETSC_HAVE_HYPRE)
4442: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_hypre_C",MatConvert_AIJ_HYPRE);
4443: PetscObjectComposeFunction((PetscObject)B,"MatMatMatMult_transpose_seqaij_seqaij_C",MatMatMatMult_Transpose_AIJ_AIJ);
4444: #endif
4445: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqdense_C",MatConvert_SeqAIJ_SeqDense);
4446: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsell_C",MatConvert_SeqAIJ_SeqSELL);
4447: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_is_C",MatConvert_XAIJ_IS);
4448: PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);
4449: PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);
4450: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);
4451: PetscObjectComposeFunction((PetscObject)B,"MatResetPreallocation_C",MatResetPreallocation_SeqAIJ);
4452: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);
4453: PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);
4454: PetscObjectComposeFunction((PetscObject)B,"MatMatMult_seqdense_seqaij_C",MatMatMult_SeqDense_SeqAIJ);
4455: PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_seqdense_seqaij_C",MatMatMultSymbolic_SeqDense_SeqAIJ);
4456: PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_seqdense_seqaij_C",MatMatMultNumeric_SeqDense_SeqAIJ);
4457: PetscObjectComposeFunction((PetscObject)B,"MatPtAP_is_seqaij_C",MatPtAP_IS_XAIJ);
4458: MatCreate_SeqAIJ_Inode(B);
4459: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
4460: MatSeqAIJSetTypeFromOptions(B); /* this allows changing the matrix subtype to say MATSEQAIJPERM */
4461: return(0);
4462: }
4464: /*
4465: Given a matrix generated with MatGetFactor() duplicates all the information in A into B
4466: */
4467: PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace)
4468: {
4469: Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data;
4471: PetscInt m = A->rmap->n,i;
4474: c = (Mat_SeqAIJ*)C->data;
4476: C->factortype = A->factortype;
4477: c->row = 0;
4478: c->col = 0;
4479: c->icol = 0;
4480: c->reallocs = 0;
4482: C->assembled = PETSC_TRUE;
4484: PetscLayoutReference(A->rmap,&C->rmap);
4485: PetscLayoutReference(A->cmap,&C->cmap);
4487: PetscMalloc1(m,&c->imax);
4488: PetscMemcpy(c->imax,a->imax,m*sizeof(PetscInt));
4489: PetscMalloc1(m,&c->ilen);
4490: PetscMemcpy(c->ilen,a->ilen,m*sizeof(PetscInt));
4491: PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));
4493: /* allocate the matrix space */
4494: if (mallocmatspace) {
4495: PetscMalloc3(a->i[m],&c->a,a->i[m],&c->j,m+1,&c->i);
4496: PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));
4498: c->singlemalloc = PETSC_TRUE;
4500: PetscArraycpy(c->i,a->i,m+1);
4501: if (m > 0) {
4502: PetscArraycpy(c->j,a->j,a->i[m]);
4503: if (cpvalues == MAT_COPY_VALUES) {
4504: PetscArraycpy(c->a,a->a,a->i[m]);
4505: } else {
4506: PetscArrayzero(c->a,a->i[m]);
4507: }
4508: }
4509: }
4511: c->ignorezeroentries = a->ignorezeroentries;
4512: c->roworiented = a->roworiented;
4513: c->nonew = a->nonew;
4514: if (a->diag) {
4515: PetscMalloc1(m+1,&c->diag);
4516: PetscMemcpy(c->diag,a->diag,m*sizeof(PetscInt));
4517: PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));
4518: } else c->diag = NULL;
4520: c->solve_work = 0;
4521: c->saved_values = 0;
4522: c->idiag = 0;
4523: c->ssor_work = 0;
4524: c->keepnonzeropattern = a->keepnonzeropattern;
4525: c->free_a = PETSC_TRUE;
4526: c->free_ij = PETSC_TRUE;
4528: c->rmax = a->rmax;
4529: c->nz = a->nz;
4530: c->maxnz = a->nz; /* Since we allocate exactly the right amount */
4531: C->preallocated = PETSC_TRUE;
4533: c->compressedrow.use = a->compressedrow.use;
4534: c->compressedrow.nrows = a->compressedrow.nrows;
4535: if (a->compressedrow.use) {
4536: i = a->compressedrow.nrows;
4537: PetscMalloc2(i+1,&c->compressedrow.i,i,&c->compressedrow.rindex);
4538: PetscArraycpy(c->compressedrow.i,a->compressedrow.i,i+1);
4539: PetscArraycpy(c->compressedrow.rindex,a->compressedrow.rindex,i);
4540: } else {
4541: c->compressedrow.use = PETSC_FALSE;
4542: c->compressedrow.i = NULL;
4543: c->compressedrow.rindex = NULL;
4544: }
4545: c->nonzerorowcnt = a->nonzerorowcnt;
4546: C->nonzerostate = A->nonzerostate;
4548: MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);
4549: PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);
4550: return(0);
4551: }
4553: PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
4554: {
4558: MatCreate(PetscObjectComm((PetscObject)A),B);
4559: MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);
4560: if (!(A->rmap->n % A->rmap->bs) && !(A->cmap->n % A->cmap->bs)) {
4561: MatSetBlockSizesFromMats(*B,A,A);
4562: }
4563: MatSetType(*B,((PetscObject)A)->type_name);
4564: MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);
4565: return(0);
4566: }
4568: PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer)
4569: {
4570: PetscBool isbinary, ishdf5;
4576: /* force binary viewer to load .info file if it has not yet done so */
4577: PetscViewerSetUp(viewer);
4578: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
4579: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERHDF5, &ishdf5);
4580: if (isbinary) {
4581: MatLoad_SeqAIJ_Binary(newMat,viewer);
4582: } else if (ishdf5) {
4583: #if defined(PETSC_HAVE_HDF5)
4584: MatLoad_AIJ_HDF5(newMat,viewer);
4585: #else
4586: SETERRQ(PetscObjectComm((PetscObject)newMat),PETSC_ERR_SUP,"HDF5 not supported in this build.\nPlease reconfigure using --download-hdf5");
4587: #endif
4588: } else {
4589: 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);
4590: }
4591: return(0);
4592: }
4594: PetscErrorCode MatLoad_SeqAIJ_Binary(Mat newMat, PetscViewer viewer)
4595: {
4596: Mat_SeqAIJ *a;
4598: PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N,rows,cols;
4599: int fd;
4600: PetscMPIInt size;
4601: MPI_Comm comm;
4602: PetscInt bs = newMat->rmap->bs;
4605: PetscObjectGetComm((PetscObject)viewer,&comm);
4606: MPI_Comm_size(comm,&size);
4607: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"view must have one processor");
4609: PetscOptionsBegin(comm,NULL,"Options for loading SEQAIJ matrix","Mat");
4610: PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);
4611: PetscOptionsEnd();
4612: if (bs < 0) bs = 1;
4613: MatSetBlockSize(newMat,bs);
4615: PetscViewerBinaryGetDescriptor(viewer,&fd);
4616: PetscBinaryRead(fd,header,4,NULL,PETSC_INT);
4617: if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file");
4618: M = header[1]; N = header[2]; nz = header[3];
4620: if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ");
4622: /* read in row lengths */
4623: PetscMalloc1(M,&rowlengths);
4624: PetscBinaryRead(fd,rowlengths,M,NULL,PETSC_INT);
4626: /* check if sum of rowlengths is same as nz */
4627: for (i=0,sum=0; i< M; i++) sum +=rowlengths[i];
4628: if (sum != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_READ,"Inconsistant matrix data in file. no-nonzeros = %dD, sum-row-lengths = %D\n",nz,sum);
4630: /* set global size if not set already*/
4631: if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) {
4632: MatSetSizes(newMat,PETSC_DECIDE,PETSC_DECIDE,M,N);
4633: } else {
4634: /* if sizes and type are already set, check if the matrix global sizes are correct */
4635: MatGetSize(newMat,&rows,&cols);
4636: if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */
4637: MatGetLocalSize(newMat,&rows,&cols);
4638: }
4639: if (M != rows || N != cols) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different length (%D, %D) than the input matrix (%D, %D)",M,N,rows,cols);
4640: }
4641: MatSeqAIJSetPreallocation_SeqAIJ(newMat,0,rowlengths);
4642: a = (Mat_SeqAIJ*)newMat->data;
4644: PetscBinaryRead(fd,a->j,nz,NULL,PETSC_INT);
4646: /* read in nonzero values */
4647: PetscBinaryRead(fd,a->a,nz,NULL,PETSC_SCALAR);
4649: /* set matrix "i" values */
4650: a->i[0] = 0;
4651: for (i=1; i<= M; i++) {
4652: a->i[i] = a->i[i-1] + rowlengths[i-1];
4653: a->ilen[i-1] = rowlengths[i-1];
4654: }
4655: PetscFree(rowlengths);
4657: MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);
4658: MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);
4659: return(0);
4660: }
4662: PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg)
4663: {
4664: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data;
4666: #if defined(PETSC_USE_COMPLEX)
4667: PetscInt k;
4668: #endif
4671: /* If the matrix dimensions are not equal,or no of nonzeros */
4672: if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) {
4673: *flg = PETSC_FALSE;
4674: return(0);
4675: }
4677: /* if the a->i are the same */
4678: PetscArraycmp(a->i,b->i,A->rmap->n+1,flg);
4679: if (!*flg) return(0);
4681: /* if a->j are the same */
4682: PetscArraycmp(a->j,b->j,a->nz,flg);
4683: if (!*flg) return(0);
4685: /* if a->a are the same */
4686: #if defined(PETSC_USE_COMPLEX)
4687: for (k=0; k<a->nz; k++) {
4688: if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) {
4689: *flg = PETSC_FALSE;
4690: return(0);
4691: }
4692: }
4693: #else
4694: PetscArraycmp(a->a,b->a,a->nz,flg);
4695: #endif
4696: return(0);
4697: }
4699: /*@
4700: MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format)
4701: provided by the user.
4703: Collective
4705: Input Parameters:
4706: + comm - must be an MPI communicator of size 1
4707: . m - number of rows
4708: . n - number of columns
4709: . i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of elements in that row of the matrix
4710: . j - column indices
4711: - a - matrix values
4713: Output Parameter:
4714: . mat - the matrix
4716: Level: intermediate
4718: Notes:
4719: The i, j, and a arrays are not copied by this routine, the user must free these arrays
4720: once the matrix is destroyed and not before
4722: You cannot set new nonzero locations into this matrix, that will generate an error.
4724: The i and j indices are 0 based
4726: The format which is used for the sparse matrix input, is equivalent to a
4727: row-major ordering.. i.e for the following matrix, the input data expected is
4728: as shown
4730: $ 1 0 0
4731: $ 2 0 3
4732: $ 4 5 6
4733: $
4734: $ i = {0,1,3,6} [size = nrow+1 = 3+1]
4735: $ j = {0,0,2,0,1,2} [size = 6]; values must be sorted for each row
4736: $ v = {1,2,3,4,5,6} [size = 6]
4739: .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
4741: @*/
4742: PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat)
4743: {
4745: PetscInt ii;
4746: Mat_SeqAIJ *aij;
4747: #if defined(PETSC_USE_DEBUG)
4748: PetscInt jj;
4749: #endif
4752: if (m > 0 && i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
4753: MatCreate(comm,mat);
4754: MatSetSizes(*mat,m,n,m,n);
4755: /* MatSetBlockSizes(*mat,,); */
4756: MatSetType(*mat,MATSEQAIJ);
4757: MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);
4758: aij = (Mat_SeqAIJ*)(*mat)->data;
4759: PetscMalloc1(m,&aij->imax);
4760: PetscMalloc1(m,&aij->ilen);
4762: aij->i = i;
4763: aij->j = j;
4764: aij->a = a;
4765: aij->singlemalloc = PETSC_FALSE;
4766: aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
4767: aij->free_a = PETSC_FALSE;
4768: aij->free_ij = PETSC_FALSE;
4770: for (ii=0; ii<m; ii++) {
4771: aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii];
4772: #if defined(PETSC_USE_DEBUG)
4773: if (i[ii+1] - i[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row length in i (row indices) row = %D length = %D",ii,i[ii+1] - i[ii]);
4774: for (jj=i[ii]+1; jj<i[ii+1]; jj++) {
4775: if (j[jj] < j[jj-1]) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %D (actual column %D) in row %D is not sorted",jj-i[ii],j[jj],ii);
4776: if (j[jj] == j[jj-1]) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %D (actual column %D) in row %D is identical to previous entry",jj-i[ii],j[jj],ii);
4777: }
4778: #endif
4779: }
4780: #if defined(PETSC_USE_DEBUG)
4781: for (ii=0; ii<aij->i[m]; ii++) {
4782: if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %D index = %D",ii,j[ii]);
4783: if (j[ii] > n - 1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column index to large at location = %D index = %D",ii,j[ii]);
4784: }
4785: #endif
4787: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
4788: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
4789: return(0);
4790: }
4791: /*@C
4792: MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format)
4793: provided by the user.
4795: Collective
4797: Input Parameters:
4798: + comm - must be an MPI communicator of size 1
4799: . m - number of rows
4800: . n - number of columns
4801: . i - row indices
4802: . j - column indices
4803: . a - matrix values
4804: . nz - number of nonzeros
4805: - idx - 0 or 1 based
4807: Output Parameter:
4808: . mat - the matrix
4810: Level: intermediate
4812: Notes:
4813: The i and j indices are 0 based
4815: The format which is used for the sparse matrix input, is equivalent to a
4816: row-major ordering.. i.e for the following matrix, the input data expected is
4817: as shown:
4819: 1 0 0
4820: 2 0 3
4821: 4 5 6
4823: i = {0,1,1,2,2,2}
4824: j = {0,0,2,0,1,2}
4825: v = {1,2,3,4,5,6}
4828: .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
4830: @*/
4831: PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat,PetscInt nz,PetscBool idx)
4832: {
4834: PetscInt ii, *nnz, one = 1,row,col;
4838: PetscCalloc1(m,&nnz);
4839: for (ii = 0; ii < nz; ii++) {
4840: nnz[i[ii] - !!idx] += 1;
4841: }
4842: MatCreate(comm,mat);
4843: MatSetSizes(*mat,m,n,m,n);
4844: MatSetType(*mat,MATSEQAIJ);
4845: MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);
4846: for (ii = 0; ii < nz; ii++) {
4847: if (idx) {
4848: row = i[ii] - 1;
4849: col = j[ii] - 1;
4850: } else {
4851: row = i[ii];
4852: col = j[ii];
4853: }
4854: MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);
4855: }
4856: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
4857: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
4858: PetscFree(nnz);
4859: return(0);
4860: }
4862: PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A)
4863: {
4864: Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data;
4868: a->idiagvalid = PETSC_FALSE;
4869: a->ibdiagvalid = PETSC_FALSE;
4871: MatSeqAIJInvalidateDiagonal_Inode(A);
4872: return(0);
4873: }
4875: PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat)
4876: {
4878: PetscMPIInt size;
4881: MPI_Comm_size(comm,&size);
4882: if (size == 1) {
4883: if (scall == MAT_INITIAL_MATRIX) {
4884: MatDuplicate(inmat,MAT_COPY_VALUES,outmat);
4885: } else {
4886: MatCopy(inmat,*outmat,SAME_NONZERO_PATTERN);
4887: }
4888: } else {
4889: MatCreateMPIMatConcatenateSeqMat_MPIAIJ(comm,inmat,n,scall,outmat);
4890: }
4891: return(0);
4892: }
4894: /*
4895: Permute A into C's *local* index space using rowemb,colemb.
4896: The embedding are supposed to be injections and the above implies that the range of rowemb is a subset
4897: of [0,m), colemb is in [0,n).
4898: If pattern == DIFFERENT_NONZERO_PATTERN, C is preallocated according to A.
4899: */
4900: PetscErrorCode MatSetSeqMat_SeqAIJ(Mat C,IS rowemb,IS colemb,MatStructure pattern,Mat B)
4901: {
4902: /* If making this function public, change the error returned in this function away from _PLIB. */
4904: Mat_SeqAIJ *Baij;
4905: PetscBool seqaij;
4906: PetscInt m,n,*nz,i,j,count;
4907: PetscScalar v;
4908: const PetscInt *rowindices,*colindices;
4911: if (!B) return(0);
4912: /* Check to make sure the target matrix (and embeddings) are compatible with C and each other. */
4913: PetscObjectBaseTypeCompare((PetscObject)B,MATSEQAIJ,&seqaij);
4914: if (!seqaij) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is of wrong type");
4915: if (rowemb) {
4916: ISGetLocalSize(rowemb,&m);
4917: if (m != B->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Row IS of size %D is incompatible with matrix row size %D",m,B->rmap->n);
4918: } else {
4919: if (C->rmap->n != B->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is row-incompatible with the target matrix");
4920: }
4921: if (colemb) {
4922: ISGetLocalSize(colemb,&n);
4923: if (n != B->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Diag col IS of size %D is incompatible with input matrix col size %D",n,B->cmap->n);
4924: } else {
4925: if (C->cmap->n != B->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is col-incompatible with the target matrix");
4926: }
4928: Baij = (Mat_SeqAIJ*)(B->data);
4929: if (pattern == DIFFERENT_NONZERO_PATTERN) {
4930: PetscMalloc1(B->rmap->n,&nz);
4931: for (i=0; i<B->rmap->n; i++) {
4932: nz[i] = Baij->i[i+1] - Baij->i[i];
4933: }
4934: MatSeqAIJSetPreallocation(C,0,nz);
4935: PetscFree(nz);
4936: }
4937: if (pattern == SUBSET_NONZERO_PATTERN) {
4938: MatZeroEntries(C);
4939: }
4940: count = 0;
4941: rowindices = NULL;
4942: colindices = NULL;
4943: if (rowemb) {
4944: ISGetIndices(rowemb,&rowindices);
4945: }
4946: if (colemb) {
4947: ISGetIndices(colemb,&colindices);
4948: }
4949: for (i=0; i<B->rmap->n; i++) {
4950: PetscInt row;
4951: row = i;
4952: if (rowindices) row = rowindices[i];
4953: for (j=Baij->i[i]; j<Baij->i[i+1]; j++) {
4954: PetscInt col;
4955: col = Baij->j[count];
4956: if (colindices) col = colindices[col];
4957: v = Baij->a[count];
4958: MatSetValues(C,1,&row,1,&col,&v,INSERT_VALUES);
4959: ++count;
4960: }
4961: }
4962: /* FIXME: set C's nonzerostate correctly. */
4963: /* Assembly for C is necessary. */
4964: C->preallocated = PETSC_TRUE;
4965: C->assembled = PETSC_TRUE;
4966: C->was_assembled = PETSC_FALSE;
4967: return(0);
4968: }
4970: PetscFunctionList MatSeqAIJList = NULL;
4972: /*@C
4973: MatSeqAIJSetType - Converts a MATSEQAIJ matrix to a subtype
4975: Collective on Mat
4977: Input Parameters:
4978: + mat - the matrix object
4979: - matype - matrix type
4981: Options Database Key:
4982: . -mat_seqai_type <method> - for example seqaijcrl
4985: Level: intermediate
4987: .seealso: PCSetType(), VecSetType(), MatCreate(), MatType, Mat
4988: @*/
4989: PetscErrorCode MatSeqAIJSetType(Mat mat, MatType matype)
4990: {
4991: PetscErrorCode ierr,(*r)(Mat,MatType,MatReuse,Mat*);
4992: PetscBool sametype;
4996: PetscObjectTypeCompare((PetscObject)mat,matype,&sametype);
4997: if (sametype) return(0);
4999: PetscFunctionListFind(MatSeqAIJList,matype,&r);
5000: if (!r) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_UNKNOWN_TYPE,"Unknown Mat type given: %s",matype);
5001: (*r)(mat,matype,MAT_INPLACE_MATRIX,&mat);
5002: return(0);
5003: }
5006: /*@C
5007: MatSeqAIJRegister - - Adds a new sub-matrix type for sequential AIJ matrices
5009: Not Collective
5011: Input Parameters:
5012: + name - name of a new user-defined matrix type, for example MATSEQAIJCRL
5013: - function - routine to convert to subtype
5015: Notes:
5016: MatSeqAIJRegister() may be called multiple times to add several user-defined solvers.
5019: Then, your matrix can be chosen with the procedural interface at runtime via the option
5020: $ -mat_seqaij_type my_mat
5022: Level: advanced
5024: .seealso: MatSeqAIJRegisterAll()
5027: Level: advanced
5028: @*/
5029: PetscErrorCode MatSeqAIJRegister(const char sname[],PetscErrorCode (*function)(Mat,MatType,MatReuse,Mat *))
5030: {
5034: MatInitializePackage();
5035: PetscFunctionListAdd(&MatSeqAIJList,sname,function);
5036: return(0);
5037: }
5039: PetscBool MatSeqAIJRegisterAllCalled = PETSC_FALSE;
5041: /*@C
5042: MatSeqAIJRegisterAll - Registers all of the matrix subtypes of SeqAIJ
5044: Not Collective
5046: Level: advanced
5048: Developers Note: CUSP and CUSPARSE do not yet support the MatConvert_SeqAIJ..() paradigm and thus cannot be registered here
5050: .seealso: MatRegisterAll(), MatSeqAIJRegister()
5051: @*/
5052: PetscErrorCode MatSeqAIJRegisterAll(void)
5053: {
5057: if (MatSeqAIJRegisterAllCalled) return(0);
5058: MatSeqAIJRegisterAllCalled = PETSC_TRUE;
5060: MatSeqAIJRegister(MATSEQAIJCRL, MatConvert_SeqAIJ_SeqAIJCRL);
5061: MatSeqAIJRegister(MATSEQAIJPERM, MatConvert_SeqAIJ_SeqAIJPERM);
5062: MatSeqAIJRegister(MATSEQAIJSELL, MatConvert_SeqAIJ_SeqAIJSELL);
5063: #if defined(PETSC_HAVE_MKL_SPARSE)
5064: MatSeqAIJRegister(MATSEQAIJMKL, MatConvert_SeqAIJ_SeqAIJMKL);
5065: #endif
5066: #if defined(PETSC_HAVE_VIENNACL) && defined(PETSC_HAVE_VIENNACL_NO_CUDA)
5067: MatSeqAIJRegister(MATMPIAIJVIENNACL, MatConvert_SeqAIJ_SeqAIJViennaCL);
5068: #endif
5069: return(0);
5070: }
5072: /*
5073: Special version for direct calls from Fortran
5074: */
5075: #include <petsc/private/fortranimpl.h>
5076: #if defined(PETSC_HAVE_FORTRAN_CAPS)
5077: #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ
5078: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
5079: #define matsetvaluesseqaij_ matsetvaluesseqaij
5080: #endif
5082: /* Change these macros so can be used in void function */
5083: #undef CHKERRQ
5084: #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr)
5085: #undef SETERRQ2
5086: #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr)
5087: #undef SETERRQ3
5088: #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr)
5090: PETSC_EXTERN void PETSC_STDCALL matsetvaluesseqaij_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[],InsertMode *isis, PetscErrorCode *_ierr)
5091: {
5092: Mat A = *AA;
5093: PetscInt m = *mm, n = *nn;
5094: InsertMode is = *isis;
5095: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
5096: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
5097: PetscInt *imax,*ai,*ailen;
5099: PetscInt *aj,nonew = a->nonew,lastcol = -1;
5100: MatScalar *ap,value,*aa;
5101: PetscBool ignorezeroentries = a->ignorezeroentries;
5102: PetscBool roworiented = a->roworiented;
5105: MatCheckPreallocated(A,1);
5106: imax = a->imax;
5107: ai = a->i;
5108: ailen = a->ilen;
5109: aj = a->j;
5110: aa = a->a;
5112: for (k=0; k<m; k++) { /* loop over added rows */
5113: row = im[k];
5114: if (row < 0) continue;
5115: #if defined(PETSC_USE_DEBUG)
5116: if (row >= A->rmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
5117: #endif
5118: rp = aj + ai[row]; ap = aa + ai[row];
5119: rmax = imax[row]; nrow = ailen[row];
5120: low = 0;
5121: high = nrow;
5122: for (l=0; l<n; l++) { /* loop over added columns */
5123: if (in[l] < 0) continue;
5124: #if defined(PETSC_USE_DEBUG)
5125: if (in[l] >= A->cmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large");
5126: #endif
5127: col = in[l];
5128: if (roworiented) value = v[l + k*n];
5129: else value = v[k + l*m];
5131: if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue;
5133: if (col <= lastcol) low = 0;
5134: else high = nrow;
5135: lastcol = col;
5136: while (high-low > 5) {
5137: t = (low+high)/2;
5138: if (rp[t] > col) high = t;
5139: else low = t;
5140: }
5141: for (i=low; i<high; i++) {
5142: if (rp[i] > col) break;
5143: if (rp[i] == col) {
5144: if (is == ADD_VALUES) ap[i] += value;
5145: else ap[i] = value;
5146: goto noinsert;
5147: }
5148: }
5149: if (value == 0.0 && ignorezeroentries) goto noinsert;
5150: if (nonew == 1) goto noinsert;
5151: if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix");
5152: MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
5153: N = nrow++ - 1; a->nz++; high++;
5154: /* shift up all the later entries in this row */
5155: for (ii=N; ii>=i; ii--) {
5156: rp[ii+1] = rp[ii];
5157: ap[ii+1] = ap[ii];
5158: }
5159: rp[i] = col;
5160: ap[i] = value;
5161: A->nonzerostate++;
5162: noinsert:;
5163: low = i + 1;
5164: }
5165: ailen[row] = nrow;
5166: }
5167: PetscFunctionReturnVoid();
5168: }