Actual source code: aij.c
petsc-master 2020-11-28
1: /*
2: Defines the basic matrix operations for the AIJ (compressed row)
3: matrix storage format.
4: */
7: #include <../src/mat/impls/aij/seq/aij.h>
8: #include <petscblaslapack.h>
9: #include <petscbt.h>
10: #include <petsc/private/kernels/blocktranspose.h>
12: PetscErrorCode MatSeqAIJSetTypeFromOptions(Mat A)
13: {
14: PetscErrorCode ierr;
15: PetscBool flg;
16: char type[256];
19: PetscObjectOptionsBegin((PetscObject)A);
20: PetscOptionsFList("-mat_seqaij_type","Matrix SeqAIJ type","MatSeqAIJSetType",MatSeqAIJList,"seqaij",type,256,&flg);
21: if (flg) {
22: MatSeqAIJSetType(A,type);
23: }
24: PetscOptionsEnd();
25: return(0);
26: }
28: PetscErrorCode MatGetColumnNorms_SeqAIJ(Mat A,NormType type,PetscReal *norms)
29: {
31: PetscInt i,m,n;
32: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data;
35: MatGetSize(A,&m,&n);
36: PetscArrayzero(norms,n);
37: if (type == NORM_2) {
38: for (i=0; i<aij->i[m]; i++) {
39: norms[aij->j[i]] += PetscAbsScalar(aij->a[i]*aij->a[i]);
40: }
41: } else if (type == NORM_1) {
42: for (i=0; i<aij->i[m]; i++) {
43: norms[aij->j[i]] += PetscAbsScalar(aij->a[i]);
44: }
45: } else if (type == NORM_INFINITY) {
46: for (i=0; i<aij->i[m]; i++) {
47: norms[aij->j[i]] = PetscMax(PetscAbsScalar(aij->a[i]),norms[aij->j[i]]);
48: }
49: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown NormType");
51: if (type == NORM_2) {
52: for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]);
53: }
54: return(0);
55: }
57: PetscErrorCode MatFindOffBlockDiagonalEntries_SeqAIJ(Mat A,IS *is)
58: {
59: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
60: PetscInt i,m=A->rmap->n,cnt = 0, bs = A->rmap->bs;
61: const PetscInt *jj = a->j,*ii = a->i;
62: PetscInt *rows;
63: PetscErrorCode ierr;
66: for (i=0; i<m; i++) {
67: if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) {
68: cnt++;
69: }
70: }
71: PetscMalloc1(cnt,&rows);
72: cnt = 0;
73: for (i=0; i<m; i++) {
74: if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) {
75: rows[cnt] = i;
76: cnt++;
77: }
78: }
79: ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,is);
80: return(0);
81: }
83: PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat A,PetscInt *nrows,PetscInt **zrows)
84: {
85: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
86: const MatScalar *aa = a->a;
87: PetscInt i,m=A->rmap->n,cnt = 0;
88: const PetscInt *ii = a->i,*jj = a->j,*diag;
89: PetscInt *rows;
90: PetscErrorCode ierr;
93: MatMarkDiagonal_SeqAIJ(A);
94: diag = a->diag;
95: for (i=0; i<m; i++) {
96: if ((diag[i] >= ii[i+1]) || (jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) {
97: cnt++;
98: }
99: }
100: PetscMalloc1(cnt,&rows);
101: cnt = 0;
102: for (i=0; i<m; i++) {
103: if ((diag[i] >= ii[i+1]) || (jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) {
104: rows[cnt++] = i;
105: }
106: }
107: *nrows = cnt;
108: *zrows = rows;
109: return(0);
110: }
112: PetscErrorCode MatFindZeroDiagonals_SeqAIJ(Mat A,IS *zrows)
113: {
114: PetscInt nrows,*rows;
118: *zrows = NULL;
119: MatFindZeroDiagonals_SeqAIJ_Private(A,&nrows,&rows);
120: ISCreateGeneral(PetscObjectComm((PetscObject)A),nrows,rows,PETSC_OWN_POINTER,zrows);
121: return(0);
122: }
124: PetscErrorCode MatFindNonzeroRows_SeqAIJ(Mat A,IS *keptrows)
125: {
126: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
127: const MatScalar *aa;
128: PetscInt m=A->rmap->n,cnt = 0;
129: const PetscInt *ii;
130: PetscInt n,i,j,*rows;
131: PetscErrorCode ierr;
134: *keptrows = NULL;
135: ii = a->i;
136: for (i=0; i<m; i++) {
137: n = ii[i+1] - ii[i];
138: if (!n) {
139: cnt++;
140: goto ok1;
141: }
142: aa = a->a + ii[i];
143: for (j=0; j<n; j++) {
144: if (aa[j] != 0.0) goto ok1;
145: }
146: cnt++;
147: ok1:;
148: }
149: if (!cnt) return(0);
150: PetscMalloc1(A->rmap->n-cnt,&rows);
151: cnt = 0;
152: for (i=0; i<m; i++) {
153: n = ii[i+1] - ii[i];
154: if (!n) continue;
155: aa = a->a + ii[i];
156: for (j=0; j<n; j++) {
157: if (aa[j] != 0.0) {
158: rows[cnt++] = i;
159: break;
160: }
161: }
162: }
163: ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,keptrows);
164: return(0);
165: }
167: PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is)
168: {
169: PetscErrorCode ierr;
170: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data;
171: PetscInt i,m = Y->rmap->n;
172: const PetscInt *diag;
173: MatScalar *aa = aij->a;
174: const PetscScalar *v;
175: PetscBool missing;
176: #if defined(PETSC_HAVE_DEVICE)
177: PetscBool inserted = PETSC_FALSE;
178: #endif
181: if (Y->assembled) {
182: MatMissingDiagonal_SeqAIJ(Y,&missing,NULL);
183: if (!missing) {
184: diag = aij->diag;
185: VecGetArrayRead(D,&v);
186: if (is == INSERT_VALUES) {
187: #if defined(PETSC_HAVE_DEVICE)
188: inserted = PETSC_TRUE;
189: #endif
190: for (i=0; i<m; i++) {
191: aa[diag[i]] = v[i];
192: }
193: } else {
194: for (i=0; i<m; i++) {
195: #if defined(PETSC_HAVE_DEVICE)
196: if (v[i] != 0.0) inserted = PETSC_TRUE;
197: #endif
198: aa[diag[i]] += v[i];
199: }
200: }
201: #if defined(PETSC_HAVE_DEVICE)
202: if (inserted) Y->offloadmask = PETSC_OFFLOAD_CPU;
203: #endif
204: VecRestoreArrayRead(D,&v);
205: return(0);
206: }
207: MatSeqAIJInvalidateDiagonal(Y);
208: }
209: MatDiagonalSet_Default(Y,D,is);
210: return(0);
211: }
213: PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *m,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
214: {
215: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
217: PetscInt i,ishift;
220: *m = A->rmap->n;
221: if (!ia) return(0);
222: ishift = 0;
223: if (symmetric && !A->structurally_symmetric) {
224: MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,PETSC_TRUE,ishift,oshift,(PetscInt**)ia,(PetscInt**)ja);
225: } else if (oshift == 1) {
226: PetscInt *tia;
227: PetscInt nz = a->i[A->rmap->n];
228: /* malloc space and add 1 to i and j indices */
229: PetscMalloc1(A->rmap->n+1,&tia);
230: for (i=0; i<A->rmap->n+1; i++) tia[i] = a->i[i] + 1;
231: *ia = tia;
232: if (ja) {
233: PetscInt *tja;
234: PetscMalloc1(nz+1,&tja);
235: for (i=0; i<nz; i++) tja[i] = a->j[i] + 1;
236: *ja = tja;
237: }
238: } else {
239: *ia = a->i;
240: if (ja) *ja = a->j;
241: }
242: return(0);
243: }
245: PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
246: {
250: if (!ia) return(0);
251: if ((symmetric && !A->structurally_symmetric) || oshift == 1) {
252: PetscFree(*ia);
253: if (ja) {PetscFree(*ja);}
254: }
255: return(0);
256: }
258: PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
259: {
260: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
262: PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n;
263: PetscInt nz = a->i[m],row,*jj,mr,col;
266: *nn = n;
267: if (!ia) return(0);
268: if (symmetric) {
269: MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,PETSC_TRUE,0,oshift,(PetscInt**)ia,(PetscInt**)ja);
270: } else {
271: PetscCalloc1(n,&collengths);
272: PetscMalloc1(n+1,&cia);
273: PetscMalloc1(nz,&cja);
274: jj = a->j;
275: for (i=0; i<nz; i++) {
276: collengths[jj[i]]++;
277: }
278: cia[0] = oshift;
279: for (i=0; i<n; i++) {
280: cia[i+1] = cia[i] + collengths[i];
281: }
282: PetscArrayzero(collengths,n);
283: jj = a->j;
284: for (row=0; row<m; row++) {
285: mr = a->i[row+1] - a->i[row];
286: for (i=0; i<mr; i++) {
287: col = *jj++;
289: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
290: }
291: }
292: PetscFree(collengths);
293: *ia = cia; *ja = cja;
294: }
295: return(0);
296: }
298: PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
299: {
303: if (!ia) return(0);
305: PetscFree(*ia);
306: PetscFree(*ja);
307: return(0);
308: }
310: /*
311: MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from
312: MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output
313: spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ() and MatFDColoringCreate_SeqXAIJ()
314: */
315: PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
316: {
317: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
319: PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n;
320: PetscInt nz = a->i[m],row,mr,col,tmp;
321: PetscInt *cspidx;
322: const PetscInt *jj;
325: *nn = n;
326: if (!ia) return(0);
328: PetscCalloc1(n,&collengths);
329: PetscMalloc1(n+1,&cia);
330: PetscMalloc1(nz,&cja);
331: PetscMalloc1(nz,&cspidx);
332: jj = a->j;
333: for (i=0; i<nz; i++) {
334: collengths[jj[i]]++;
335: }
336: cia[0] = oshift;
337: for (i=0; i<n; i++) {
338: cia[i+1] = cia[i] + collengths[i];
339: }
340: PetscArrayzero(collengths,n);
341: jj = a->j;
342: for (row=0; row<m; row++) {
343: mr = a->i[row+1] - a->i[row];
344: for (i=0; i<mr; i++) {
345: col = *jj++;
346: tmp = cia[col] + collengths[col]++ - oshift;
347: cspidx[tmp] = a->i[row] + i; /* index of a->j */
348: cja[tmp] = row + oshift;
349: }
350: }
351: PetscFree(collengths);
352: *ia = cia;
353: *ja = cja;
354: *spidx = cspidx;
355: return(0);
356: }
358: PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
359: {
363: MatRestoreColumnIJ_SeqAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done);
364: PetscFree(*spidx);
365: return(0);
366: }
368: PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[])
369: {
370: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
371: PetscInt *ai = a->i;
375: PetscArraycpy(a->a+ai[row],v,ai[row+1]-ai[row]);
376: #if defined(PETSC_HAVE_DEVICE)
377: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && ai[row+1]-ai[row]) A->offloadmask = PETSC_OFFLOAD_CPU;
378: #endif
379: return(0);
380: }
382: /*
383: MatSeqAIJSetValuesLocalFast - An optimized version of MatSetValuesLocal() for SeqAIJ matrices with several assumptions
385: - a single row of values is set with each call
386: - no row or column indices are negative or (in error) larger than the number of rows or columns
387: - the values are always added to the matrix, not set
388: - no new locations are introduced in the nonzero structure of the matrix
390: This does NOT assume the global column indices are sorted
392: */
394: #include <petsc/private/isimpl.h>
395: PetscErrorCode MatSeqAIJSetValuesLocalFast(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
396: {
397: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
398: PetscInt low,high,t,row,nrow,i,col,l;
399: const PetscInt *rp,*ai = a->i,*ailen = a->ilen,*aj = a->j;
400: PetscInt lastcol = -1;
401: MatScalar *ap,value,*aa = a->a;
402: const PetscInt *ridx = A->rmap->mapping->indices,*cidx = A->cmap->mapping->indices;
404: row = ridx[im[0]];
405: rp = aj + ai[row];
406: ap = aa + ai[row];
407: nrow = ailen[row];
408: low = 0;
409: high = nrow;
410: for (l=0; l<n; l++) { /* loop over added columns */
411: col = cidx[in[l]];
412: value = v[l];
414: if (col <= lastcol) low = 0;
415: else high = nrow;
416: lastcol = col;
417: while (high-low > 5) {
418: t = (low+high)/2;
419: if (rp[t] > col) high = t;
420: else low = t;
421: }
422: for (i=low; i<high; i++) {
423: if (rp[i] == col) {
424: ap[i] += value;
425: low = i + 1;
426: break;
427: }
428: }
429: }
430: #if defined(PETSC_HAVE_DEVICE)
431: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && m && n) A->offloadmask = PETSC_OFFLOAD_CPU;
432: #endif
433: return 0;
434: }
436: PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
437: {
438: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
439: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
440: PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen;
442: PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1;
443: MatScalar *ap=NULL,value=0.0,*aa = a->a;
444: PetscBool ignorezeroentries = a->ignorezeroentries;
445: PetscBool roworiented = a->roworiented;
446: #if defined(PETSC_HAVE_DEVICE)
447: PetscBool inserted = PETSC_FALSE;
448: #endif
451: for (k=0; k<m; k++) { /* loop over added rows */
452: row = im[k];
453: if (row < 0) continue;
454: if (PetscUnlikelyDebug(row >= A->rmap->n)) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap->n-1);
455: rp = aj + ai[row];
456: if (!A->structure_only) ap = aa + ai[row];
457: rmax = imax[row]; nrow = ailen[row];
458: low = 0;
459: high = nrow;
460: for (l=0; l<n; l++) { /* loop over added columns */
461: if (in[l] < 0) continue;
462: if (PetscUnlikelyDebug(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);
463: col = in[l];
464: if (v && !A->structure_only) value = roworiented ? v[l + k*n] : v[k + l*m];
465: if (!A->structure_only && value == 0.0 && ignorezeroentries && is == ADD_VALUES && row != col) continue;
467: if (col <= lastcol) low = 0;
468: else high = nrow;
469: lastcol = col;
470: while (high-low > 5) {
471: t = (low+high)/2;
472: if (rp[t] > col) high = t;
473: else low = t;
474: }
475: for (i=low; i<high; i++) {
476: if (rp[i] > col) break;
477: if (rp[i] == col) {
478: if (!A->structure_only) {
479: if (is == ADD_VALUES) {
480: ap[i] += value;
481: (void)PetscLogFlops(1.0);
482: }
483: else ap[i] = value;
484: #if defined(PETSC_HAVE_DEVICE)
485: inserted = PETSC_TRUE;
486: #endif
487: }
488: low = i + 1;
489: goto noinsert;
490: }
491: }
492: if (value == 0.0 && ignorezeroentries && row != col) goto noinsert;
493: if (nonew == 1) goto noinsert;
494: if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%D,%D) in the matrix",row,col);
495: if (A->structure_only) {
496: MatSeqXAIJReallocateAIJ_structure_only(A,A->rmap->n,1,nrow,row,col,rmax,ai,aj,rp,imax,nonew,MatScalar);
497: } else {
498: MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
499: }
500: N = nrow++ - 1; a->nz++; high++;
501: /* shift up all the later entries in this row */
502: PetscArraymove(rp+i+1,rp+i,N-i+1);
503: rp[i] = col;
504: if (!A->structure_only){
505: PetscArraymove(ap+i+1,ap+i,N-i+1);
506: ap[i] = value;
507: }
508: low = i + 1;
509: A->nonzerostate++;
510: #if defined(PETSC_HAVE_DEVICE)
511: inserted = PETSC_TRUE;
512: #endif
513: noinsert:;
514: }
515: ailen[row] = nrow;
516: }
517: #if defined(PETSC_HAVE_DEVICE)
518: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && inserted) A->offloadmask = PETSC_OFFLOAD_CPU;
519: #endif
520: return(0);
521: }
524: PetscErrorCode MatSetValues_SeqAIJ_SortedFullNoPreallocation(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
525: {
526: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
527: PetscInt *rp,k,row;
528: PetscInt *ai = a->i;
530: PetscInt *aj = a->j;
531: MatScalar *aa = a->a,*ap;
534: if (A->was_assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot call on assembled matrix.");
535: if (m*n+a->nz > a->maxnz) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Number of entries in matrix will be larger than maximum nonzeros allocated for %D in MatSeqAIJSetTotalPreallocation()",a->maxnz);
536: for (k=0; k<m; k++) { /* loop over added rows */
537: row = im[k];
538: rp = aj + ai[row];
539: ap = aa + ai[row];
541: PetscMemcpy(rp,in,n*sizeof(PetscInt));
542: if (!A->structure_only) {
543: if (v) {
544: PetscMemcpy(ap,v,n*sizeof(PetscScalar));
545: v += n;
546: } else {
547: PetscMemzero(ap,n*sizeof(PetscScalar));
548: }
549: }
550: a->ilen[row] = n;
551: a->imax[row] = n;
552: a->i[row+1] = a->i[row]+n;
553: a->nz += n;
554: }
555: #if defined(PETSC_HAVE_DEVICE)
556: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && m && n) A->offloadmask = PETSC_OFFLOAD_CPU;
557: #endif
558: return(0);
559: }
561: /*@
562: MatSeqAIJSetTotalPreallocation - Sets an upper bound on the total number of expected nonzeros in the matrix.
564: Input Parameters:
565: + A - the SeqAIJ matrix
566: - nztotal - bound on the number of nonzeros
568: Level: advanced
570: Notes:
571: This can be called if you will be provided the matrix row by row (from row zero) with sorted column indices for each row.
572: Simply call MatSetValues() after this call to provide the matrix entries in the usual manner. This matrix may be used
573: as always with multiple matrix assemblies.
575: .seealso: MatSetOption(), MAT_SORTED_FULL, MatSetValues(), MatSeqAIJSetPreallocation()
576: @*/
578: PetscErrorCode MatSeqAIJSetTotalPreallocation(Mat A,PetscInt nztotal)
579: {
581: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
584: PetscLayoutSetUp(A->rmap);
585: PetscLayoutSetUp(A->cmap);
586: a->maxnz = nztotal;
587: if (!a->imax) {
588: PetscMalloc1(A->rmap->n,&a->imax);
589: PetscLogObjectMemory((PetscObject)A,A->rmap->n*sizeof(PetscInt));
590: }
591: if (!a->ilen) {
592: PetscMalloc1(A->rmap->n,&a->ilen);
593: PetscLogObjectMemory((PetscObject)A,A->rmap->n*sizeof(PetscInt));
594: } else {
595: PetscMemzero(a->ilen,A->rmap->n*sizeof(PetscInt));
596: }
598: /* allocate the matrix space */
599: if (A->structure_only) {
600: PetscMalloc1(nztotal,&a->j);
601: PetscMalloc1(A->rmap->n+1,&a->i);
602: PetscLogObjectMemory((PetscObject)A,(A->rmap->n+1)*sizeof(PetscInt)+nztotal*sizeof(PetscInt));
603: } else {
604: PetscMalloc3(nztotal,&a->a,nztotal,&a->j,A->rmap->n+1,&a->i);
605: PetscLogObjectMemory((PetscObject)A,(A->rmap->n+1)*sizeof(PetscInt)+nztotal*(sizeof(PetscScalar)+sizeof(PetscInt)));
606: }
607: a->i[0] = 0;
608: if (A->structure_only) {
609: a->singlemalloc = PETSC_FALSE;
610: a->free_a = PETSC_FALSE;
611: } else {
612: a->singlemalloc = PETSC_TRUE;
613: a->free_a = PETSC_TRUE;
614: }
615: a->free_ij = PETSC_TRUE;
616: A->ops->setvalues = MatSetValues_SeqAIJ_SortedFullNoPreallocation;
617: A->preallocated = PETSC_TRUE;
618: return(0);
619: }
621: PetscErrorCode MatSetValues_SeqAIJ_SortedFull(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
622: {
623: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
624: PetscInt *rp,k,row;
625: PetscInt *ai = a->i,*ailen = a->ilen;
627: PetscInt *aj = a->j;
628: MatScalar *aa = a->a,*ap;
631: for (k=0; k<m; k++) { /* loop over added rows */
632: row = im[k];
633: if (PetscUnlikelyDebug(n > a->imax[row])) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Preallocation for row %D does not match number of columns provided",n);
634: rp = aj + ai[row];
635: ap = aa + ai[row];
636: if (!A->was_assembled) {
637: PetscMemcpy(rp,in,n*sizeof(PetscInt));
638: }
639: if (!A->structure_only) {
640: if (v) {
641: PetscMemcpy(ap,v,n*sizeof(PetscScalar));
642: v += n;
643: } else {
644: PetscMemzero(ap,n*sizeof(PetscScalar));
645: }
646: }
647: ailen[row] = n;
648: a->nz += n;
649: }
650: #if defined(PETSC_HAVE_DEVICE)
651: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && m && n) A->offloadmask = PETSC_OFFLOAD_CPU;
652: #endif
653: return(0);
654: }
657: PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
658: {
659: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
660: PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
661: PetscInt *ai = a->i,*ailen = a->ilen;
662: MatScalar *ap,*aa = a->a;
665: for (k=0; k<m; k++) { /* loop over rows */
666: row = im[k];
667: if (row < 0) {v += n; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",row); */
668: 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);
669: rp = aj + ai[row]; ap = aa + ai[row];
670: nrow = ailen[row];
671: for (l=0; l<n; l++) { /* loop over columns */
672: if (in[l] < 0) {v++; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",in[l]); */
673: 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);
674: col = in[l];
675: high = nrow; low = 0; /* assume unsorted */
676: while (high-low > 5) {
677: t = (low+high)/2;
678: if (rp[t] > col) high = t;
679: else low = t;
680: }
681: for (i=low; i<high; i++) {
682: if (rp[i] > col) break;
683: if (rp[i] == col) {
684: *v++ = ap[i];
685: goto finished;
686: }
687: }
688: *v++ = 0.0;
689: finished:;
690: }
691: }
692: return(0);
693: }
695: PetscErrorCode MatView_SeqAIJ_Binary(Mat mat,PetscViewer viewer)
696: {
697: Mat_SeqAIJ *A = (Mat_SeqAIJ*)mat->data;
698: const PetscScalar *av;
699: PetscInt header[4],M,N,m,nz,i;
700: PetscInt *rowlens;
701: PetscErrorCode ierr;
704: PetscViewerSetUp(viewer);
706: M = mat->rmap->N;
707: N = mat->cmap->N;
708: m = mat->rmap->n;
709: nz = A->nz;
711: /* write matrix header */
712: header[0] = MAT_FILE_CLASSID;
713: header[1] = M; header[2] = N; header[3] = nz;
714: PetscViewerBinaryWrite(viewer,header,4,PETSC_INT);
716: /* fill in and store row lengths */
717: PetscMalloc1(m,&rowlens);
718: for (i=0; i<m; i++) rowlens[i] = A->i[i+1] - A->i[i];
719: PetscViewerBinaryWrite(viewer,rowlens,m,PETSC_INT);
720: PetscFree(rowlens);
721: /* store column indices */
722: PetscViewerBinaryWrite(viewer,A->j,nz,PETSC_INT);
723: /* store nonzero values */
724: MatSeqAIJGetArrayRead(mat,&av);
725: PetscViewerBinaryWrite(viewer,av,nz,PETSC_SCALAR);
726: MatSeqAIJRestoreArrayRead(mat,&av);
728: /* write block size option to the viewer's .info file */
729: MatView_Binary_BlockSizes(mat,viewer);
730: return(0);
731: }
733: static PetscErrorCode MatView_SeqAIJ_ASCII_structonly(Mat A,PetscViewer viewer)
734: {
736: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
737: PetscInt i,k,m=A->rmap->N;
740: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
741: for (i=0; i<m; i++) {
742: PetscViewerASCIIPrintf(viewer,"row %D:",i);
743: for (k=a->i[i]; k<a->i[i+1]; k++) {
744: PetscViewerASCIIPrintf(viewer," (%D) ",a->j[k]);
745: }
746: PetscViewerASCIIPrintf(viewer,"\n");
747: }
748: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
749: return(0);
750: }
752: extern PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer);
754: PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer)
755: {
756: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
757: const PetscScalar *av;
758: PetscErrorCode ierr;
759: PetscInt i,j,m = A->rmap->n;
760: const char *name;
761: PetscViewerFormat format;
764: if (A->structure_only) {
765: MatView_SeqAIJ_ASCII_structonly(A,viewer);
766: return(0);
767: }
769: PetscViewerGetFormat(viewer,&format);
770: /* trigger copy to CPU if needed */
771: MatSeqAIJGetArrayRead(A,&av);
772: MatSeqAIJRestoreArrayRead(A,&av);
773: if (format == PETSC_VIEWER_ASCII_MATLAB) {
774: PetscInt nofinalvalue = 0;
775: if (m && ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap->n-1))) {
776: /* Need a dummy value to ensure the dimension of the matrix. */
777: nofinalvalue = 1;
778: }
779: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
780: PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",m,A->cmap->n);
781: PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %D \n",a->nz);
782: #if defined(PETSC_USE_COMPLEX)
783: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,4);\n",a->nz+nofinalvalue);
784: #else
785: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,3);\n",a->nz+nofinalvalue);
786: #endif
787: PetscViewerASCIIPrintf(viewer,"zzz = [\n");
789: for (i=0; i<m; i++) {
790: for (j=a->i[i]; j<a->i[i+1]; j++) {
791: #if defined(PETSC_USE_COMPLEX)
792: 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]));
793: #else
794: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",i+1,a->j[j]+1,(double)a->a[j]);
795: #endif
796: }
797: }
798: if (nofinalvalue) {
799: #if defined(PETSC_USE_COMPLEX)
800: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e %18.16e\n",m,A->cmap->n,0.,0.);
801: #else
802: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",m,A->cmap->n,0.0);
803: #endif
804: }
805: PetscObjectGetName((PetscObject)A,&name);
806: PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);
807: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
808: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
809: return(0);
810: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
811: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
812: for (i=0; i<m; i++) {
813: PetscViewerASCIIPrintf(viewer,"row %D:",i);
814: for (j=a->i[i]; j<a->i[i+1]; j++) {
815: #if defined(PETSC_USE_COMPLEX)
816: if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) {
817: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
818: } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) {
819: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
820: } else if (PetscRealPart(a->a[j]) != 0.0) {
821: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
822: }
823: #else
824: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);}
825: #endif
826: }
827: PetscViewerASCIIPrintf(viewer,"\n");
828: }
829: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
830: } else if (format == PETSC_VIEWER_ASCII_SYMMODU) {
831: PetscInt nzd=0,fshift=1,*sptr;
832: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
833: PetscMalloc1(m+1,&sptr);
834: for (i=0; i<m; i++) {
835: sptr[i] = nzd+1;
836: for (j=a->i[i]; j<a->i[i+1]; j++) {
837: if (a->j[j] >= i) {
838: #if defined(PETSC_USE_COMPLEX)
839: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++;
840: #else
841: if (a->a[j] != 0.0) nzd++;
842: #endif
843: }
844: }
845: }
846: sptr[m] = nzd+1;
847: PetscViewerASCIIPrintf(viewer," %D %D\n\n",m,nzd);
848: for (i=0; i<m+1; i+=6) {
849: if (i+4<m) {
850: 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]);
851: } else if (i+3<m) {
852: PetscViewerASCIIPrintf(viewer," %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4]);
853: } else if (i+2<m) {
854: PetscViewerASCIIPrintf(viewer," %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);
855: } else if (i+1<m) {
856: PetscViewerASCIIPrintf(viewer," %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2]);
857: } else if (i<m) {
858: PetscViewerASCIIPrintf(viewer," %D %D\n",sptr[i],sptr[i+1]);
859: } else {
860: PetscViewerASCIIPrintf(viewer," %D\n",sptr[i]);
861: }
862: }
863: PetscViewerASCIIPrintf(viewer,"\n");
864: PetscFree(sptr);
865: for (i=0; i<m; i++) {
866: for (j=a->i[i]; j<a->i[i+1]; j++) {
867: if (a->j[j] >= i) {PetscViewerASCIIPrintf(viewer," %D ",a->j[j]+fshift);}
868: }
869: PetscViewerASCIIPrintf(viewer,"\n");
870: }
871: PetscViewerASCIIPrintf(viewer,"\n");
872: for (i=0; i<m; i++) {
873: for (j=a->i[i]; j<a->i[i+1]; j++) {
874: if (a->j[j] >= i) {
875: #if defined(PETSC_USE_COMPLEX)
876: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) {
877: PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
878: }
879: #else
880: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," %18.16e ",(double)a->a[j]);}
881: #endif
882: }
883: }
884: PetscViewerASCIIPrintf(viewer,"\n");
885: }
886: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
887: } else if (format == PETSC_VIEWER_ASCII_DENSE) {
888: PetscInt cnt = 0,jcnt;
889: PetscScalar value;
890: #if defined(PETSC_USE_COMPLEX)
891: PetscBool realonly = PETSC_TRUE;
893: for (i=0; i<a->i[m]; i++) {
894: if (PetscImaginaryPart(a->a[i]) != 0.0) {
895: realonly = PETSC_FALSE;
896: break;
897: }
898: }
899: #endif
901: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
902: for (i=0; i<m; i++) {
903: jcnt = 0;
904: for (j=0; j<A->cmap->n; j++) {
905: if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) {
906: value = a->a[cnt++];
907: jcnt++;
908: } else {
909: value = 0.0;
910: }
911: #if defined(PETSC_USE_COMPLEX)
912: if (realonly) {
913: PetscViewerASCIIPrintf(viewer," %7.5e ",(double)PetscRealPart(value));
914: } else {
915: PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",(double)PetscRealPart(value),(double)PetscImaginaryPart(value));
916: }
917: #else
918: PetscViewerASCIIPrintf(viewer," %7.5e ",(double)value);
919: #endif
920: }
921: PetscViewerASCIIPrintf(viewer,"\n");
922: }
923: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
924: } else if (format == PETSC_VIEWER_ASCII_MATRIXMARKET) {
925: PetscInt fshift=1;
926: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
927: #if defined(PETSC_USE_COMPLEX)
928: PetscViewerASCIIPrintf(viewer,"%%%%MatrixMarket matrix coordinate complex general\n");
929: #else
930: PetscViewerASCIIPrintf(viewer,"%%%%MatrixMarket matrix coordinate real general\n");
931: #endif
932: PetscViewerASCIIPrintf(viewer,"%D %D %D\n", m, A->cmap->n, a->nz);
933: for (i=0; i<m; i++) {
934: for (j=a->i[i]; j<a->i[i+1]; j++) {
935: #if defined(PETSC_USE_COMPLEX)
936: PetscViewerASCIIPrintf(viewer,"%D %D %g %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
937: #else
938: PetscViewerASCIIPrintf(viewer,"%D %D %g\n", i+fshift, a->j[j]+fshift, (double)a->a[j]);
939: #endif
940: }
941: }
942: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
943: } else {
944: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
945: if (A->factortype) {
946: for (i=0; i<m; i++) {
947: PetscViewerASCIIPrintf(viewer,"row %D:",i);
948: /* L part */
949: for (j=a->i[i]; j<a->i[i+1]; j++) {
950: #if defined(PETSC_USE_COMPLEX)
951: if (PetscImaginaryPart(a->a[j]) > 0.0) {
952: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
953: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
954: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));
955: } else {
956: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
957: }
958: #else
959: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
960: #endif
961: }
962: /* diagonal */
963: j = a->diag[i];
964: #if defined(PETSC_USE_COMPLEX)
965: if (PetscImaginaryPart(a->a[j]) > 0.0) {
966: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)PetscImaginaryPart(1.0/a->a[j]));
967: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
968: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)(-PetscImaginaryPart(1.0/a->a[j])));
969: } else {
970: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(1.0/a->a[j]));
971: }
972: #else
973: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)(1.0/a->a[j]));
974: #endif
976: /* U part */
977: for (j=a->diag[i+1]+1; j<a->diag[i]; j++) {
978: #if defined(PETSC_USE_COMPLEX)
979: if (PetscImaginaryPart(a->a[j]) > 0.0) {
980: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
981: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
982: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));
983: } else {
984: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
985: }
986: #else
987: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
988: #endif
989: }
990: PetscViewerASCIIPrintf(viewer,"\n");
991: }
992: } else {
993: for (i=0; i<m; i++) {
994: PetscViewerASCIIPrintf(viewer,"row %D:",i);
995: for (j=a->i[i]; j<a->i[i+1]; j++) {
996: #if defined(PETSC_USE_COMPLEX)
997: if (PetscImaginaryPart(a->a[j]) > 0.0) {
998: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
999: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
1000: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
1001: } else {
1002: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
1003: }
1004: #else
1005: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
1006: #endif
1007: }
1008: PetscViewerASCIIPrintf(viewer,"\n");
1009: }
1010: }
1011: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
1012: }
1013: PetscViewerFlush(viewer);
1014: return(0);
1015: }
1017: #include <petscdraw.h>
1018: PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
1019: {
1020: Mat A = (Mat) Aa;
1021: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1022: PetscErrorCode ierr;
1023: PetscInt i,j,m = A->rmap->n;
1024: int color;
1025: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r;
1026: PetscViewer viewer;
1027: PetscViewerFormat format;
1030: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
1031: PetscViewerGetFormat(viewer,&format);
1032: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
1034: /* loop over matrix elements drawing boxes */
1036: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
1037: PetscDrawCollectiveBegin(draw);
1038: /* Blue for negative, Cyan for zero and Red for positive */
1039: color = PETSC_DRAW_BLUE;
1040: for (i=0; i<m; i++) {
1041: y_l = m - i - 1.0; y_r = y_l + 1.0;
1042: for (j=a->i[i]; j<a->i[i+1]; j++) {
1043: x_l = a->j[j]; x_r = x_l + 1.0;
1044: if (PetscRealPart(a->a[j]) >= 0.) continue;
1045: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
1046: }
1047: }
1048: color = PETSC_DRAW_CYAN;
1049: for (i=0; i<m; i++) {
1050: y_l = m - i - 1.0; y_r = y_l + 1.0;
1051: for (j=a->i[i]; j<a->i[i+1]; j++) {
1052: x_l = a->j[j]; x_r = x_l + 1.0;
1053: if (a->a[j] != 0.) continue;
1054: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
1055: }
1056: }
1057: color = PETSC_DRAW_RED;
1058: for (i=0; i<m; i++) {
1059: y_l = m - i - 1.0; y_r = y_l + 1.0;
1060: for (j=a->i[i]; j<a->i[i+1]; j++) {
1061: x_l = a->j[j]; x_r = x_l + 1.0;
1062: if (PetscRealPart(a->a[j]) <= 0.) continue;
1063: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
1064: }
1065: }
1066: PetscDrawCollectiveEnd(draw);
1067: } else {
1068: /* use contour shading to indicate magnitude of values */
1069: /* first determine max of all nonzero values */
1070: PetscReal minv = 0.0, maxv = 0.0;
1071: PetscInt nz = a->nz, count = 0;
1072: PetscDraw popup;
1074: for (i=0; i<nz; i++) {
1075: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
1076: }
1077: if (minv >= maxv) maxv = minv + PETSC_SMALL;
1078: PetscDrawGetPopup(draw,&popup);
1079: PetscDrawScalePopup(popup,minv,maxv);
1081: PetscDrawCollectiveBegin(draw);
1082: for (i=0; i<m; i++) {
1083: y_l = m - i - 1.0;
1084: y_r = y_l + 1.0;
1085: for (j=a->i[i]; j<a->i[i+1]; j++) {
1086: x_l = a->j[j];
1087: x_r = x_l + 1.0;
1088: color = PetscDrawRealToColor(PetscAbsScalar(a->a[count]),minv,maxv);
1089: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
1090: count++;
1091: }
1092: }
1093: PetscDrawCollectiveEnd(draw);
1094: }
1095: return(0);
1096: }
1098: #include <petscdraw.h>
1099: PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer)
1100: {
1102: PetscDraw draw;
1103: PetscReal xr,yr,xl,yl,h,w;
1104: PetscBool isnull;
1107: PetscViewerDrawGetDraw(viewer,0,&draw);
1108: PetscDrawIsNull(draw,&isnull);
1109: if (isnull) return(0);
1111: xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0;
1112: xr += w; yr += h; xl = -w; yl = -h;
1113: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
1114: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
1115: PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);
1116: PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL);
1117: PetscDrawSave(draw);
1118: return(0);
1119: }
1121: PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer)
1122: {
1124: PetscBool iascii,isbinary,isdraw;
1127: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
1128: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
1129: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
1130: if (iascii) {
1131: MatView_SeqAIJ_ASCII(A,viewer);
1132: } else if (isbinary) {
1133: MatView_SeqAIJ_Binary(A,viewer);
1134: } else if (isdraw) {
1135: MatView_SeqAIJ_Draw(A,viewer);
1136: }
1137: MatView_SeqAIJ_Inode(A,viewer);
1138: return(0);
1139: }
1141: PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode)
1142: {
1143: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1145: PetscInt fshift = 0,i,*ai = a->i,*aj = a->j,*imax = a->imax;
1146: PetscInt m = A->rmap->n,*ip,N,*ailen = a->ilen,rmax = 0;
1147: MatScalar *aa = a->a,*ap;
1148: PetscReal ratio = 0.6;
1151: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
1152: MatSeqAIJInvalidateDiagonal(A);
1153: if (A->was_assembled && A->ass_nonzerostate == A->nonzerostate) {
1154: /* we need to respect users asking to use or not the inodes routine in between matrix assemblies */
1155: MatAssemblyEnd_SeqAIJ_Inode(A,mode);
1156: return(0);
1157: }
1159: if (m) rmax = ailen[0]; /* determine row with most nonzeros */
1160: for (i=1; i<m; i++) {
1161: /* move each row back by the amount of empty slots (fshift) before it*/
1162: fshift += imax[i-1] - ailen[i-1];
1163: rmax = PetscMax(rmax,ailen[i]);
1164: if (fshift) {
1165: ip = aj + ai[i];
1166: ap = aa + ai[i];
1167: N = ailen[i];
1168: PetscArraymove(ip-fshift,ip,N);
1169: if (!A->structure_only) {
1170: PetscArraymove(ap-fshift,ap,N);
1171: }
1172: }
1173: ai[i] = ai[i-1] + ailen[i-1];
1174: }
1175: if (m) {
1176: fshift += imax[m-1] - ailen[m-1];
1177: ai[m] = ai[m-1] + ailen[m-1];
1178: }
1180: /* reset ilen and imax for each row */
1181: a->nonzerorowcnt = 0;
1182: if (A->structure_only) {
1183: PetscFree(a->imax);
1184: PetscFree(a->ilen);
1185: } else { /* !A->structure_only */
1186: for (i=0; i<m; i++) {
1187: ailen[i] = imax[i] = ai[i+1] - ai[i];
1188: a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0);
1189: }
1190: }
1191: a->nz = ai[m];
1192: 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);
1194: MatMarkDiagonal_SeqAIJ(A);
1195: PetscInfo4(A,"Matrix size: %D X %D; storage space: %D unneeded,%D used\n",m,A->cmap->n,fshift,a->nz);
1196: PetscInfo1(A,"Number of mallocs during MatSetValues() is %D\n",a->reallocs);
1197: PetscInfo1(A,"Maximum nonzeros in any row is %D\n",rmax);
1199: A->info.mallocs += a->reallocs;
1200: a->reallocs = 0;
1201: A->info.nz_unneeded = (PetscReal)fshift;
1202: a->rmax = rmax;
1204: if (!A->structure_only) {
1205: MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,m,ratio);
1206: }
1207: MatAssemblyEnd_SeqAIJ_Inode(A,mode);
1208: return(0);
1209: }
1211: PetscErrorCode MatRealPart_SeqAIJ(Mat A)
1212: {
1213: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1214: PetscInt i,nz = a->nz;
1215: MatScalar *aa = a->a;
1219: for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
1220: MatSeqAIJInvalidateDiagonal(A);
1221: #if defined(PETSC_HAVE_DEVICE)
1222: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
1223: #endif
1224: return(0);
1225: }
1227: PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A)
1228: {
1229: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1230: PetscInt i,nz = a->nz;
1231: MatScalar *aa = a->a;
1235: for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
1236: MatSeqAIJInvalidateDiagonal(A);
1237: #if defined(PETSC_HAVE_DEVICE)
1238: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
1239: #endif
1240: return(0);
1241: }
1243: PetscErrorCode MatZeroEntries_SeqAIJ(Mat A)
1244: {
1245: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1249: PetscArrayzero(a->a,a->i[A->rmap->n]);
1250: MatSeqAIJInvalidateDiagonal(A);
1251: #if defined(PETSC_HAVE_DEVICE)
1252: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
1253: #endif
1254: return(0);
1255: }
1257: PetscErrorCode MatDestroy_SeqAIJ(Mat A)
1258: {
1259: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1263: #if defined(PETSC_USE_LOG)
1264: PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap->n,A->cmap->n,a->nz);
1265: #endif
1266: MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);
1267: ISDestroy(&a->row);
1268: ISDestroy(&a->col);
1269: PetscFree(a->diag);
1270: PetscFree(a->ibdiag);
1271: PetscFree(a->imax);
1272: PetscFree(a->ilen);
1273: PetscFree(a->ipre);
1274: PetscFree3(a->idiag,a->mdiag,a->ssor_work);
1275: PetscFree(a->solve_work);
1276: ISDestroy(&a->icol);
1277: PetscFree(a->saved_values);
1278: PetscFree2(a->compressedrow.i,a->compressedrow.rindex);
1280: MatDestroy_SeqAIJ_Inode(A);
1281: PetscFree(A->data);
1283: /* MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted may allocate this.
1284: That function is so heavily used (sometimes in an hidden way through multnumeric function pointers)
1285: that is hard to properly add this data to the MatProduct data. We free it here to avoid
1286: users reusing the matrix object with different data to incur in obscure segmentation faults
1287: due to different matrix sizes */
1288: PetscObjectCompose((PetscObject)A,"__PETSc__ab_dense",NULL);
1290: PetscObjectChangeTypeName((PetscObject)A,NULL);
1291: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetColumnIndices_C",NULL);
1292: PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL);
1293: PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL);
1294: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsbaij_C",NULL);
1295: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqbaij_C",NULL);
1296: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijperm_C",NULL);
1297: #if defined(PETSC_HAVE_CUDA)
1298: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijcusparse_C",NULL);
1299: PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaijcusparse_seqaij_C",NULL);
1300: #endif
1301: #if defined(PETSC_HAVE_KOKKOS_KERNELS)
1302: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijkokkos_C",NULL);
1303: #endif
1304: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijcrl_C",NULL);
1305: #if defined(PETSC_HAVE_ELEMENTAL)
1306: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_elemental_C",NULL);
1307: #endif
1308: #if defined(PETSC_HAVE_SCALAPACK)
1309: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_scalapack_C",NULL);
1310: #endif
1311: #if defined(PETSC_HAVE_HYPRE)
1312: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_hypre_C",NULL);
1313: PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_transpose_seqaij_seqaij_C",NULL);
1314: #endif
1315: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqdense_C",NULL);
1316: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsell_C",NULL);
1317: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_is_C",NULL);
1318: PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL);
1319: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocation_C",NULL);
1320: PetscObjectComposeFunction((PetscObject)A,"MatResetPreallocation_C",NULL);
1321: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C",NULL);
1322: PetscObjectComposeFunction((PetscObject)A,"MatReorderForNonzeroDiagonal_C",NULL);
1323: PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_is_seqaij_C",NULL);
1324: PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqdense_seqaij_C",NULL);
1325: PetscObjectComposeFunction((PetscObject)A,"MatProductSetFromOptions_seqaij_seqaij_C",NULL);
1326: return(0);
1327: }
1329: PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op,PetscBool flg)
1330: {
1331: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1335: switch (op) {
1336: case MAT_ROW_ORIENTED:
1337: a->roworiented = flg;
1338: break;
1339: case MAT_KEEP_NONZERO_PATTERN:
1340: a->keepnonzeropattern = flg;
1341: break;
1342: case MAT_NEW_NONZERO_LOCATIONS:
1343: a->nonew = (flg ? 0 : 1);
1344: break;
1345: case MAT_NEW_NONZERO_LOCATION_ERR:
1346: a->nonew = (flg ? -1 : 0);
1347: break;
1348: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1349: a->nonew = (flg ? -2 : 0);
1350: break;
1351: case MAT_UNUSED_NONZERO_LOCATION_ERR:
1352: a->nounused = (flg ? -1 : 0);
1353: break;
1354: case MAT_IGNORE_ZERO_ENTRIES:
1355: a->ignorezeroentries = flg;
1356: break;
1357: case MAT_SPD:
1358: case MAT_SYMMETRIC:
1359: case MAT_STRUCTURALLY_SYMMETRIC:
1360: case MAT_HERMITIAN:
1361: case MAT_SYMMETRY_ETERNAL:
1362: case MAT_STRUCTURE_ONLY:
1363: /* These options are handled directly by MatSetOption() */
1364: break;
1365: case MAT_NEW_DIAGONALS:
1366: case MAT_IGNORE_OFF_PROC_ENTRIES:
1367: case MAT_USE_HASH_TABLE:
1368: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1369: break;
1370: case MAT_USE_INODES:
1371: MatSetOption_SeqAIJ_Inode(A,MAT_USE_INODES,flg);
1372: break;
1373: case MAT_SUBMAT_SINGLEIS:
1374: A->submat_singleis = flg;
1375: break;
1376: case MAT_SORTED_FULL:
1377: if (flg) A->ops->setvalues = MatSetValues_SeqAIJ_SortedFull;
1378: else A->ops->setvalues = MatSetValues_SeqAIJ;
1379: break;
1380: default:
1381: SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op);
1382: }
1383: return(0);
1384: }
1386: PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v)
1387: {
1388: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1389: PetscErrorCode ierr;
1390: PetscInt i,j,n,*ai=a->i,*aj=a->j;
1391: PetscScalar *x;
1392: const PetscScalar *aa;
1395: VecGetLocalSize(v,&n);
1396: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1397: MatSeqAIJGetArrayRead(A,&aa);
1398: if (A->factortype == MAT_FACTOR_ILU || A->factortype == MAT_FACTOR_LU) {
1399: PetscInt *diag=a->diag;
1400: VecGetArrayWrite(v,&x);
1401: for (i=0; i<n; i++) x[i] = 1.0/aa[diag[i]];
1402: VecRestoreArrayWrite(v,&x);
1403: MatSeqAIJRestoreArrayRead(A,&aa);
1404: return(0);
1405: }
1407: VecGetArrayWrite(v,&x);
1408: for (i=0; i<n; i++) {
1409: x[i] = 0.0;
1410: for (j=ai[i]; j<ai[i+1]; j++) {
1411: if (aj[j] == i) {
1412: x[i] = aa[j];
1413: break;
1414: }
1415: }
1416: }
1417: VecRestoreArrayWrite(v,&x);
1418: MatSeqAIJRestoreArrayRead(A,&aa);
1419: return(0);
1420: }
1422: #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h>
1423: PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy)
1424: {
1425: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1426: PetscScalar *y;
1427: const PetscScalar *x;
1428: PetscErrorCode ierr;
1429: PetscInt m = A->rmap->n;
1430: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
1431: const MatScalar *v;
1432: PetscScalar alpha;
1433: PetscInt n,i,j;
1434: const PetscInt *idx,*ii,*ridx=NULL;
1435: Mat_CompressedRow cprow = a->compressedrow;
1436: PetscBool usecprow = cprow.use;
1437: #endif
1440: if (zz != yy) {VecCopy(zz,yy);}
1441: VecGetArrayRead(xx,&x);
1442: VecGetArray(yy,&y);
1444: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
1445: fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y);
1446: #else
1447: if (usecprow) {
1448: m = cprow.nrows;
1449: ii = cprow.i;
1450: ridx = cprow.rindex;
1451: } else {
1452: ii = a->i;
1453: }
1454: for (i=0; i<m; i++) {
1455: idx = a->j + ii[i];
1456: v = a->a + ii[i];
1457: n = ii[i+1] - ii[i];
1458: if (usecprow) {
1459: alpha = x[ridx[i]];
1460: } else {
1461: alpha = x[i];
1462: }
1463: for (j=0; j<n; j++) y[idx[j]] += alpha*v[j];
1464: }
1465: #endif
1466: PetscLogFlops(2.0*a->nz);
1467: VecRestoreArrayRead(xx,&x);
1468: VecRestoreArray(yy,&y);
1469: return(0);
1470: }
1472: PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy)
1473: {
1477: VecSet(yy,0.0);
1478: MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);
1479: return(0);
1480: }
1482: #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h>
1484: PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy)
1485: {
1486: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1487: PetscScalar *y;
1488: const PetscScalar *x;
1489: const MatScalar *aa;
1490: PetscErrorCode ierr;
1491: PetscInt m=A->rmap->n;
1492: const PetscInt *aj,*ii,*ridx=NULL;
1493: PetscInt n,i;
1494: PetscScalar sum;
1495: PetscBool usecprow=a->compressedrow.use;
1497: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1498: #pragma disjoint(*x,*y,*aa)
1499: #endif
1502: if (a->inode.use && a->inode.checked) {
1503: MatMult_SeqAIJ_Inode(A,xx,yy);
1504: return(0);
1505: }
1506: VecGetArrayRead(xx,&x);
1507: VecGetArray(yy,&y);
1508: ii = a->i;
1509: if (usecprow) { /* use compressed row format */
1510: PetscArrayzero(y,m);
1511: m = a->compressedrow.nrows;
1512: ii = a->compressedrow.i;
1513: ridx = a->compressedrow.rindex;
1514: for (i=0; i<m; i++) {
1515: n = ii[i+1] - ii[i];
1516: aj = a->j + ii[i];
1517: aa = a->a + ii[i];
1518: sum = 0.0;
1519: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1520: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1521: y[*ridx++] = sum;
1522: }
1523: } else { /* do not use compressed row format */
1524: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
1525: aj = a->j;
1526: aa = a->a;
1527: fortranmultaij_(&m,x,ii,aj,aa,y);
1528: #else
1529: for (i=0; i<m; i++) {
1530: n = ii[i+1] - ii[i];
1531: aj = a->j + ii[i];
1532: aa = a->a + ii[i];
1533: sum = 0.0;
1534: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1535: y[i] = sum;
1536: }
1537: #endif
1538: }
1539: PetscLogFlops(2.0*a->nz - a->nonzerorowcnt);
1540: VecRestoreArrayRead(xx,&x);
1541: VecRestoreArray(yy,&y);
1542: return(0);
1543: }
1545: PetscErrorCode MatMultMax_SeqAIJ(Mat A,Vec xx,Vec yy)
1546: {
1547: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1548: PetscScalar *y;
1549: const PetscScalar *x;
1550: const MatScalar *aa;
1551: PetscErrorCode ierr;
1552: PetscInt m=A->rmap->n;
1553: const PetscInt *aj,*ii,*ridx=NULL;
1554: PetscInt n,i,nonzerorow=0;
1555: PetscScalar sum;
1556: PetscBool usecprow=a->compressedrow.use;
1558: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1559: #pragma disjoint(*x,*y,*aa)
1560: #endif
1563: VecGetArrayRead(xx,&x);
1564: VecGetArray(yy,&y);
1565: if (usecprow) { /* use compressed row format */
1566: m = a->compressedrow.nrows;
1567: ii = a->compressedrow.i;
1568: ridx = a->compressedrow.rindex;
1569: for (i=0; i<m; i++) {
1570: n = ii[i+1] - ii[i];
1571: aj = a->j + ii[i];
1572: aa = a->a + ii[i];
1573: sum = 0.0;
1574: nonzerorow += (n>0);
1575: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1576: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1577: y[*ridx++] = sum;
1578: }
1579: } else { /* do not use compressed row format */
1580: ii = a->i;
1581: for (i=0; i<m; i++) {
1582: n = ii[i+1] - ii[i];
1583: aj = a->j + ii[i];
1584: aa = a->a + ii[i];
1585: sum = 0.0;
1586: nonzerorow += (n>0);
1587: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1588: y[i] = sum;
1589: }
1590: }
1591: PetscLogFlops(2.0*a->nz - nonzerorow);
1592: VecRestoreArrayRead(xx,&x);
1593: VecRestoreArray(yy,&y);
1594: return(0);
1595: }
1597: PetscErrorCode MatMultAddMax_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1598: {
1599: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1600: PetscScalar *y,*z;
1601: const PetscScalar *x;
1602: const MatScalar *aa;
1603: PetscErrorCode ierr;
1604: PetscInt m = A->rmap->n,*aj,*ii;
1605: PetscInt n,i,*ridx=NULL;
1606: PetscScalar sum;
1607: PetscBool usecprow=a->compressedrow.use;
1610: VecGetArrayRead(xx,&x);
1611: VecGetArrayPair(yy,zz,&y,&z);
1612: if (usecprow) { /* use compressed row format */
1613: if (zz != yy) {
1614: PetscArraycpy(z,y,m);
1615: }
1616: m = a->compressedrow.nrows;
1617: ii = a->compressedrow.i;
1618: ridx = a->compressedrow.rindex;
1619: for (i=0; i<m; i++) {
1620: n = ii[i+1] - ii[i];
1621: aj = a->j + ii[i];
1622: aa = a->a + ii[i];
1623: sum = y[*ridx];
1624: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1625: z[*ridx++] = sum;
1626: }
1627: } else { /* do not use compressed row format */
1628: ii = a->i;
1629: for (i=0; i<m; i++) {
1630: n = ii[i+1] - ii[i];
1631: aj = a->j + ii[i];
1632: aa = a->a + ii[i];
1633: sum = y[i];
1634: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1635: z[i] = sum;
1636: }
1637: }
1638: PetscLogFlops(2.0*a->nz);
1639: VecRestoreArrayRead(xx,&x);
1640: VecRestoreArrayPair(yy,zz,&y,&z);
1641: return(0);
1642: }
1644: #include <../src/mat/impls/aij/seq/ftn-kernels/fmultadd.h>
1645: PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1646: {
1647: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1648: PetscScalar *y,*z;
1649: const PetscScalar *x;
1650: const MatScalar *aa;
1651: PetscErrorCode ierr;
1652: const PetscInt *aj,*ii,*ridx=NULL;
1653: PetscInt m = A->rmap->n,n,i;
1654: PetscScalar sum;
1655: PetscBool usecprow=a->compressedrow.use;
1658: if (a->inode.use && a->inode.checked) {
1659: MatMultAdd_SeqAIJ_Inode(A,xx,yy,zz);
1660: return(0);
1661: }
1662: VecGetArrayRead(xx,&x);
1663: VecGetArrayPair(yy,zz,&y,&z);
1664: if (usecprow) { /* use compressed row format */
1665: if (zz != yy) {
1666: PetscArraycpy(z,y,m);
1667: }
1668: m = a->compressedrow.nrows;
1669: ii = a->compressedrow.i;
1670: ridx = a->compressedrow.rindex;
1671: for (i=0; i<m; i++) {
1672: n = ii[i+1] - ii[i];
1673: aj = a->j + ii[i];
1674: aa = a->a + ii[i];
1675: sum = y[*ridx];
1676: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1677: z[*ridx++] = sum;
1678: }
1679: } else { /* do not use compressed row format */
1680: ii = a->i;
1681: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
1682: aj = a->j;
1683: aa = a->a;
1684: fortranmultaddaij_(&m,x,ii,aj,aa,y,z);
1685: #else
1686: for (i=0; i<m; i++) {
1687: n = ii[i+1] - ii[i];
1688: aj = a->j + ii[i];
1689: aa = a->a + ii[i];
1690: sum = y[i];
1691: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1692: z[i] = sum;
1693: }
1694: #endif
1695: }
1696: PetscLogFlops(2.0*a->nz);
1697: VecRestoreArrayRead(xx,&x);
1698: VecRestoreArrayPair(yy,zz,&y,&z);
1699: return(0);
1700: }
1702: /*
1703: Adds diagonal pointers to sparse matrix structure.
1704: */
1705: PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A)
1706: {
1707: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1709: PetscInt i,j,m = A->rmap->n;
1712: if (!a->diag) {
1713: PetscMalloc1(m,&a->diag);
1714: PetscLogObjectMemory((PetscObject)A, m*sizeof(PetscInt));
1715: }
1716: for (i=0; i<A->rmap->n; i++) {
1717: a->diag[i] = a->i[i+1];
1718: for (j=a->i[i]; j<a->i[i+1]; j++) {
1719: if (a->j[j] == i) {
1720: a->diag[i] = j;
1721: break;
1722: }
1723: }
1724: }
1725: return(0);
1726: }
1728: PetscErrorCode MatShift_SeqAIJ(Mat A,PetscScalar v)
1729: {
1730: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1731: const PetscInt *diag = (const PetscInt*)a->diag;
1732: const PetscInt *ii = (const PetscInt*) a->i;
1733: PetscInt i,*mdiag = NULL;
1734: PetscErrorCode ierr;
1735: PetscInt cnt = 0; /* how many diagonals are missing */
1738: if (!A->preallocated || !a->nz) {
1739: MatSeqAIJSetPreallocation(A,1,NULL);
1740: MatShift_Basic(A,v);
1741: return(0);
1742: }
1744: if (a->diagonaldense) {
1745: cnt = 0;
1746: } else {
1747: PetscCalloc1(A->rmap->n,&mdiag);
1748: for (i=0; i<A->rmap->n; i++) {
1749: if (diag[i] >= ii[i+1]) {
1750: cnt++;
1751: mdiag[i] = 1;
1752: }
1753: }
1754: }
1755: if (!cnt) {
1756: MatShift_Basic(A,v);
1757: } else {
1758: PetscScalar *olda = a->a; /* preserve pointers to current matrix nonzeros structure and values */
1759: PetscInt *oldj = a->j, *oldi = a->i;
1760: PetscBool singlemalloc = a->singlemalloc,free_a = a->free_a,free_ij = a->free_ij;
1762: a->a = NULL;
1763: a->j = NULL;
1764: a->i = NULL;
1765: /* increase the values in imax for each row where a diagonal is being inserted then reallocate the matrix data structures */
1766: for (i=0; i<A->rmap->n; i++) {
1767: a->imax[i] += mdiag[i];
1768: a->imax[i] = PetscMin(a->imax[i],A->cmap->n);
1769: }
1770: MatSeqAIJSetPreallocation_SeqAIJ(A,0,a->imax);
1772: /* copy old values into new matrix data structure */
1773: for (i=0; i<A->rmap->n; i++) {
1774: MatSetValues(A,1,&i,a->imax[i] - mdiag[i],&oldj[oldi[i]],&olda[oldi[i]],ADD_VALUES);
1775: if (i < A->cmap->n) {
1776: MatSetValue(A,i,i,v,ADD_VALUES);
1777: }
1778: }
1779: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
1780: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
1781: if (singlemalloc) {
1782: PetscFree3(olda,oldj,oldi);
1783: } else {
1784: if (free_a) {PetscFree(olda);}
1785: if (free_ij) {PetscFree(oldj);}
1786: if (free_ij) {PetscFree(oldi);}
1787: }
1788: }
1789: PetscFree(mdiag);
1790: a->diagonaldense = PETSC_TRUE;
1791: return(0);
1792: }
1794: /*
1795: Checks for missing diagonals
1796: */
1797: PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscBool *missing,PetscInt *d)
1798: {
1799: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1800: PetscInt *diag,*ii = a->i,i;
1804: *missing = PETSC_FALSE;
1805: if (A->rmap->n > 0 && !ii) {
1806: *missing = PETSC_TRUE;
1807: if (d) *d = 0;
1808: PetscInfo(A,"Matrix has no entries therefore is missing diagonal\n");
1809: } else {
1810: PetscInt n;
1811: n = PetscMin(A->rmap->n, A->cmap->n);
1812: diag = a->diag;
1813: for (i=0; i<n; i++) {
1814: if (diag[i] >= ii[i+1]) {
1815: *missing = PETSC_TRUE;
1816: if (d) *d = i;
1817: PetscInfo1(A,"Matrix is missing diagonal number %D\n",i);
1818: break;
1819: }
1820: }
1821: }
1822: return(0);
1823: }
1825: #include <petscblaslapack.h>
1826: #include <petsc/private/kernels/blockinvert.h>
1828: /*
1829: Note that values is allocated externally by the PC and then passed into this routine
1830: */
1831: PetscErrorCode MatInvertVariableBlockDiagonal_SeqAIJ(Mat A,PetscInt nblocks,const PetscInt *bsizes,PetscScalar *diag)
1832: {
1833: PetscErrorCode ierr;
1834: PetscInt n = A->rmap->n, i, ncnt = 0, *indx,j,bsizemax = 0,*v_pivots;
1835: PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE;
1836: const PetscReal shift = 0.0;
1837: PetscInt ipvt[5];
1838: PetscScalar work[25],*v_work;
1841: allowzeropivot = PetscNot(A->erroriffailure);
1842: for (i=0; i<nblocks; i++) ncnt += bsizes[i];
1843: if (ncnt != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Total blocksizes %D doesn't match number matrix rows %D",ncnt,n);
1844: for (i=0; i<nblocks; i++) {
1845: bsizemax = PetscMax(bsizemax,bsizes[i]);
1846: }
1847: PetscMalloc1(bsizemax,&indx);
1848: if (bsizemax > 7) {
1849: PetscMalloc2(bsizemax,&v_work,bsizemax,&v_pivots);
1850: }
1851: ncnt = 0;
1852: for (i=0; i<nblocks; i++) {
1853: for (j=0; j<bsizes[i]; j++) indx[j] = ncnt+j;
1854: MatGetValues(A,bsizes[i],indx,bsizes[i],indx,diag);
1855: switch (bsizes[i]) {
1856: case 1:
1857: *diag = 1.0/(*diag);
1858: break;
1859: case 2:
1860: PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected);
1861: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1862: PetscKernel_A_gets_transpose_A_2(diag);
1863: break;
1864: case 3:
1865: PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected);
1866: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1867: PetscKernel_A_gets_transpose_A_3(diag);
1868: break;
1869: case 4:
1870: PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected);
1871: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1872: PetscKernel_A_gets_transpose_A_4(diag);
1873: break;
1874: case 5:
1875: PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected);
1876: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1877: PetscKernel_A_gets_transpose_A_5(diag);
1878: break;
1879: case 6:
1880: PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected);
1881: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1882: PetscKernel_A_gets_transpose_A_6(diag);
1883: break;
1884: case 7:
1885: PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected);
1886: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1887: PetscKernel_A_gets_transpose_A_7(diag);
1888: break;
1889: default:
1890: PetscKernel_A_gets_inverse_A(bsizes[i],diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected);
1891: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1892: PetscKernel_A_gets_transpose_A_N(diag,bsizes[i]);
1893: }
1894: ncnt += bsizes[i];
1895: diag += bsizes[i]*bsizes[i];
1896: }
1897: if (bsizemax > 7) {
1898: PetscFree2(v_work,v_pivots);
1899: }
1900: PetscFree(indx);
1901: return(0);
1902: }
1904: /*
1905: Negative shift indicates do not generate an error if there is a zero diagonal, just invert it anyways
1906: */
1907: PetscErrorCode MatInvertDiagonal_SeqAIJ(Mat A,PetscScalar omega,PetscScalar fshift)
1908: {
1909: Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data;
1911: PetscInt i,*diag,m = A->rmap->n;
1912: MatScalar *v = a->a;
1913: PetscScalar *idiag,*mdiag;
1916: if (a->idiagvalid) return(0);
1917: MatMarkDiagonal_SeqAIJ(A);
1918: diag = a->diag;
1919: if (!a->idiag) {
1920: PetscMalloc3(m,&a->idiag,m,&a->mdiag,m,&a->ssor_work);
1921: PetscLogObjectMemory((PetscObject)A, 3*m*sizeof(PetscScalar));
1922: v = a->a;
1923: }
1924: mdiag = a->mdiag;
1925: idiag = a->idiag;
1927: if (omega == 1.0 && PetscRealPart(fshift) <= 0.0) {
1928: for (i=0; i<m; i++) {
1929: mdiag[i] = v[diag[i]];
1930: if (!PetscAbsScalar(mdiag[i])) { /* zero diagonal */
1931: if (PetscRealPart(fshift)) {
1932: PetscInfo1(A,"Zero diagonal on row %D\n",i);
1933: A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
1934: A->factorerror_zeropivot_value = 0.0;
1935: A->factorerror_zeropivot_row = i;
1936: } else SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %D",i);
1937: }
1938: idiag[i] = 1.0/v[diag[i]];
1939: }
1940: PetscLogFlops(m);
1941: } else {
1942: for (i=0; i<m; i++) {
1943: mdiag[i] = v[diag[i]];
1944: idiag[i] = omega/(fshift + v[diag[i]]);
1945: }
1946: PetscLogFlops(2.0*m);
1947: }
1948: a->idiagvalid = PETSC_TRUE;
1949: return(0);
1950: }
1952: #include <../src/mat/impls/aij/seq/ftn-kernels/frelax.h>
1953: PetscErrorCode MatSOR_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
1954: {
1955: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1956: PetscScalar *x,d,sum,*t,scale;
1957: const MatScalar *v,*idiag=NULL,*mdiag;
1958: const PetscScalar *b, *bs,*xb, *ts;
1959: PetscErrorCode ierr;
1960: PetscInt n,m = A->rmap->n,i;
1961: const PetscInt *idx,*diag;
1964: if (a->inode.use && a->inode.checked && omega == 1.0 && fshift == 0.0) {
1965: MatSOR_SeqAIJ_Inode(A,bb,omega,flag,fshift,its,lits,xx);
1966: return(0);
1967: }
1968: its = its*lits;
1970: if (fshift != a->fshift || omega != a->omega) a->idiagvalid = PETSC_FALSE; /* must recompute idiag[] */
1971: if (!a->idiagvalid) {MatInvertDiagonal_SeqAIJ(A,omega,fshift);}
1972: a->fshift = fshift;
1973: a->omega = omega;
1975: diag = a->diag;
1976: t = a->ssor_work;
1977: idiag = a->idiag;
1978: mdiag = a->mdiag;
1980: VecGetArray(xx,&x);
1981: VecGetArrayRead(bb,&b);
1982: /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */
1983: if (flag == SOR_APPLY_UPPER) {
1984: /* apply (U + D/omega) to the vector */
1985: bs = b;
1986: for (i=0; i<m; i++) {
1987: d = fshift + mdiag[i];
1988: n = a->i[i+1] - diag[i] - 1;
1989: idx = a->j + diag[i] + 1;
1990: v = a->a + diag[i] + 1;
1991: sum = b[i]*d/omega;
1992: PetscSparseDensePlusDot(sum,bs,v,idx,n);
1993: x[i] = sum;
1994: }
1995: VecRestoreArray(xx,&x);
1996: VecRestoreArrayRead(bb,&b);
1997: PetscLogFlops(a->nz);
1998: return(0);
1999: }
2001: if (flag == SOR_APPLY_LOWER) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented");
2002: else if (flag & SOR_EISENSTAT) {
2003: /* Let A = L + U + D; where L is lower triangular,
2004: U is upper triangular, E = D/omega; This routine applies
2006: (L + E)^{-1} A (U + E)^{-1}
2008: to a vector efficiently using Eisenstat's trick.
2009: */
2010: scale = (2.0/omega) - 1.0;
2012: /* x = (E + U)^{-1} b */
2013: for (i=m-1; i>=0; i--) {
2014: n = a->i[i+1] - diag[i] - 1;
2015: idx = a->j + diag[i] + 1;
2016: v = a->a + diag[i] + 1;
2017: sum = b[i];
2018: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2019: x[i] = sum*idiag[i];
2020: }
2022: /* t = b - (2*E - D)x */
2023: v = a->a;
2024: for (i=0; i<m; i++) t[i] = b[i] - scale*(v[*diag++])*x[i];
2026: /* t = (E + L)^{-1}t */
2027: ts = t;
2028: diag = a->diag;
2029: for (i=0; i<m; i++) {
2030: n = diag[i] - a->i[i];
2031: idx = a->j + a->i[i];
2032: v = a->a + a->i[i];
2033: sum = t[i];
2034: PetscSparseDenseMinusDot(sum,ts,v,idx,n);
2035: t[i] = sum*idiag[i];
2036: /* x = x + t */
2037: x[i] += t[i];
2038: }
2040: PetscLogFlops(6.0*m-1 + 2.0*a->nz);
2041: VecRestoreArray(xx,&x);
2042: VecRestoreArrayRead(bb,&b);
2043: return(0);
2044: }
2045: if (flag & SOR_ZERO_INITIAL_GUESS) {
2046: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
2047: for (i=0; i<m; i++) {
2048: n = diag[i] - a->i[i];
2049: idx = a->j + a->i[i];
2050: v = a->a + a->i[i];
2051: sum = b[i];
2052: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2053: t[i] = sum;
2054: x[i] = sum*idiag[i];
2055: }
2056: xb = t;
2057: PetscLogFlops(a->nz);
2058: } else xb = b;
2059: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
2060: for (i=m-1; i>=0; i--) {
2061: n = a->i[i+1] - diag[i] - 1;
2062: idx = a->j + diag[i] + 1;
2063: v = a->a + diag[i] + 1;
2064: sum = xb[i];
2065: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2066: if (xb == b) {
2067: x[i] = sum*idiag[i];
2068: } else {
2069: x[i] = (1-omega)*x[i] + sum*idiag[i]; /* omega in idiag */
2070: }
2071: }
2072: PetscLogFlops(a->nz); /* assumes 1/2 in upper */
2073: }
2074: its--;
2075: }
2076: while (its--) {
2077: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
2078: for (i=0; i<m; i++) {
2079: /* lower */
2080: n = diag[i] - a->i[i];
2081: idx = a->j + a->i[i];
2082: v = a->a + a->i[i];
2083: sum = b[i];
2084: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2085: t[i] = sum; /* save application of the lower-triangular part */
2086: /* upper */
2087: n = a->i[i+1] - diag[i] - 1;
2088: idx = a->j + diag[i] + 1;
2089: v = a->a + diag[i] + 1;
2090: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2091: x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */
2092: }
2093: xb = t;
2094: PetscLogFlops(2.0*a->nz);
2095: } else xb = b;
2096: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
2097: for (i=m-1; i>=0; i--) {
2098: sum = xb[i];
2099: if (xb == b) {
2100: /* whole matrix (no checkpointing available) */
2101: n = a->i[i+1] - a->i[i];
2102: idx = a->j + a->i[i];
2103: v = a->a + a->i[i];
2104: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2105: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
2106: } else { /* lower-triangular part has been saved, so only apply upper-triangular */
2107: n = a->i[i+1] - diag[i] - 1;
2108: idx = a->j + diag[i] + 1;
2109: v = a->a + diag[i] + 1;
2110: PetscSparseDenseMinusDot(sum,x,v,idx,n);
2111: x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */
2112: }
2113: }
2114: if (xb == b) {
2115: PetscLogFlops(2.0*a->nz);
2116: } else {
2117: PetscLogFlops(a->nz); /* assumes 1/2 in upper */
2118: }
2119: }
2120: }
2121: VecRestoreArray(xx,&x);
2122: VecRestoreArrayRead(bb,&b);
2123: return(0);
2124: }
2127: PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info)
2128: {
2129: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2132: info->block_size = 1.0;
2133: info->nz_allocated = a->maxnz;
2134: info->nz_used = a->nz;
2135: info->nz_unneeded = (a->maxnz - a->nz);
2136: info->assemblies = A->num_ass;
2137: info->mallocs = A->info.mallocs;
2138: info->memory = ((PetscObject)A)->mem;
2139: if (A->factortype) {
2140: info->fill_ratio_given = A->info.fill_ratio_given;
2141: info->fill_ratio_needed = A->info.fill_ratio_needed;
2142: info->factor_mallocs = A->info.factor_mallocs;
2143: } else {
2144: info->fill_ratio_given = 0;
2145: info->fill_ratio_needed = 0;
2146: info->factor_mallocs = 0;
2147: }
2148: return(0);
2149: }
2151: PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
2152: {
2153: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2154: PetscInt i,m = A->rmap->n - 1;
2155: PetscErrorCode ierr;
2156: const PetscScalar *xx;
2157: PetscScalar *bb;
2158: PetscInt d = 0;
2161: if (x && b) {
2162: VecGetArrayRead(x,&xx);
2163: VecGetArray(b,&bb);
2164: for (i=0; i<N; i++) {
2165: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2166: if (rows[i] >= A->cmap->n) continue;
2167: bb[rows[i]] = diag*xx[rows[i]];
2168: }
2169: VecRestoreArrayRead(x,&xx);
2170: VecRestoreArray(b,&bb);
2171: }
2173: if (a->keepnonzeropattern) {
2174: for (i=0; i<N; i++) {
2175: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2176: PetscArrayzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]);
2177: }
2178: if (diag != 0.0) {
2179: for (i=0; i<N; i++) {
2180: d = rows[i];
2181: if (rows[i] >= A->cmap->n) continue;
2182: 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);
2183: }
2184: for (i=0; i<N; i++) {
2185: if (rows[i] >= A->cmap->n) continue;
2186: a->a[a->diag[rows[i]]] = diag;
2187: }
2188: }
2189: } else {
2190: if (diag != 0.0) {
2191: for (i=0; i<N; i++) {
2192: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2193: if (a->ilen[rows[i]] > 0) {
2194: if (rows[i] >= A->cmap->n) {
2195: a->ilen[rows[i]] = 0;
2196: } else {
2197: a->ilen[rows[i]] = 1;
2198: a->a[a->i[rows[i]]] = diag;
2199: a->j[a->i[rows[i]]] = rows[i];
2200: }
2201: } else if (rows[i] < A->cmap->n) { /* in case row was completely empty */
2202: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);
2203: }
2204: }
2205: } else {
2206: for (i=0; i<N; i++) {
2207: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2208: a->ilen[rows[i]] = 0;
2209: }
2210: }
2211: A->nonzerostate++;
2212: }
2213: #if defined(PETSC_HAVE_DEVICE)
2214: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
2215: #endif
2216: (*A->ops->assemblyend)(A,MAT_FINAL_ASSEMBLY);
2217: return(0);
2218: }
2220: PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
2221: {
2222: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2223: PetscInt i,j,m = A->rmap->n - 1,d = 0;
2224: PetscErrorCode ierr;
2225: PetscBool missing,*zeroed,vecs = PETSC_FALSE;
2226: const PetscScalar *xx;
2227: PetscScalar *bb;
2230: if (x && b) {
2231: VecGetArrayRead(x,&xx);
2232: VecGetArray(b,&bb);
2233: vecs = PETSC_TRUE;
2234: }
2235: PetscCalloc1(A->rmap->n,&zeroed);
2236: for (i=0; i<N; i++) {
2237: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2238: PetscArrayzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]);
2240: zeroed[rows[i]] = PETSC_TRUE;
2241: }
2242: for (i=0; i<A->rmap->n; i++) {
2243: if (!zeroed[i]) {
2244: for (j=a->i[i]; j<a->i[i+1]; j++) {
2245: if (a->j[j] < A->rmap->n && zeroed[a->j[j]]) {
2246: if (vecs) bb[i] -= a->a[j]*xx[a->j[j]];
2247: a->a[j] = 0.0;
2248: }
2249: }
2250: } else if (vecs && i < A->cmap->N) bb[i] = diag*xx[i];
2251: }
2252: if (x && b) {
2253: VecRestoreArrayRead(x,&xx);
2254: VecRestoreArray(b,&bb);
2255: }
2256: PetscFree(zeroed);
2257: if (diag != 0.0) {
2258: MatMissingDiagonal_SeqAIJ(A,&missing,&d);
2259: if (missing) {
2260: for (i=0; i<N; i++) {
2261: if (rows[i] >= A->cmap->N) continue;
2262: 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]);
2263: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);
2264: }
2265: } else {
2266: for (i=0; i<N; i++) {
2267: a->a[a->diag[rows[i]]] = diag;
2268: }
2269: }
2270: }
2271: #if defined(PETSC_HAVE_DEVICE)
2272: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
2273: #endif
2274: (*A->ops->assemblyend)(A,MAT_FINAL_ASSEMBLY);
2275: return(0);
2276: }
2278: PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
2279: {
2280: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2281: PetscInt *itmp;
2284: if (row < 0 || row >= A->rmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row);
2286: *nz = a->i[row+1] - a->i[row];
2287: if (v) *v = a->a + a->i[row];
2288: if (idx) {
2289: itmp = a->j + a->i[row];
2290: if (*nz) *idx = itmp;
2291: else *idx = NULL;
2292: }
2293: return(0);
2294: }
2296: /* remove this function? */
2297: PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
2298: {
2300: return(0);
2301: }
2303: PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm)
2304: {
2305: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2306: MatScalar *v = a->a;
2307: PetscReal sum = 0.0;
2309: PetscInt i,j;
2312: if (type == NORM_FROBENIUS) {
2313: #if defined(PETSC_USE_REAL___FP16)
2314: PetscBLASInt one = 1,nz = a->nz;
2315: *nrm = BLASnrm2_(&nz,v,&one);
2316: #else
2317: for (i=0; i<a->nz; i++) {
2318: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
2319: }
2320: *nrm = PetscSqrtReal(sum);
2321: #endif
2322: PetscLogFlops(2.0*a->nz);
2323: } else if (type == NORM_1) {
2324: PetscReal *tmp;
2325: PetscInt *jj = a->j;
2326: PetscCalloc1(A->cmap->n+1,&tmp);
2327: *nrm = 0.0;
2328: for (j=0; j<a->nz; j++) {
2329: tmp[*jj++] += PetscAbsScalar(*v); v++;
2330: }
2331: for (j=0; j<A->cmap->n; j++) {
2332: if (tmp[j] > *nrm) *nrm = tmp[j];
2333: }
2334: PetscFree(tmp);
2335: PetscLogFlops(PetscMax(a->nz-1,0));
2336: } else if (type == NORM_INFINITY) {
2337: *nrm = 0.0;
2338: for (j=0; j<A->rmap->n; j++) {
2339: v = a->a + a->i[j];
2340: sum = 0.0;
2341: for (i=0; i<a->i[j+1]-a->i[j]; i++) {
2342: sum += PetscAbsScalar(*v); v++;
2343: }
2344: if (sum > *nrm) *nrm = sum;
2345: }
2346: PetscLogFlops(PetscMax(a->nz-1,0));
2347: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm");
2348: return(0);
2349: }
2351: /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */
2352: PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B)
2353: {
2355: PetscInt i,j,anzj;
2356: Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b;
2357: PetscInt an=A->cmap->N,am=A->rmap->N;
2358: PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j;
2361: /* Allocate space for symbolic transpose info and work array */
2362: PetscCalloc1(an+1,&ati);
2363: PetscMalloc1(ai[am],&atj);
2364: PetscMalloc1(an,&atfill);
2366: /* Walk through aj and count ## of non-zeros in each row of A^T. */
2367: /* Note: offset by 1 for fast conversion into csr format. */
2368: for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1;
2369: /* Form ati for csr format of A^T. */
2370: for (i=0;i<an;i++) ati[i+1] += ati[i];
2372: /* Copy ati into atfill so we have locations of the next free space in atj */
2373: PetscArraycpy(atfill,ati,an);
2375: /* Walk through A row-wise and mark nonzero entries of A^T. */
2376: for (i=0;i<am;i++) {
2377: anzj = ai[i+1] - ai[i];
2378: for (j=0;j<anzj;j++) {
2379: atj[atfill[*aj]] = i;
2380: atfill[*aj++] += 1;
2381: }
2382: }
2384: /* Clean up temporary space and complete requests. */
2385: PetscFree(atfill);
2386: MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B);
2387: MatSetBlockSizes(*B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));
2388: MatSetType(*B,((PetscObject)A)->type_name);
2390: b = (Mat_SeqAIJ*)((*B)->data);
2391: b->free_a = PETSC_FALSE;
2392: b->free_ij = PETSC_TRUE;
2393: b->nonew = 0;
2394: return(0);
2395: }
2397: PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
2398: {
2399: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) B->data;
2400: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr;
2401: MatScalar *va,*vb;
2403: PetscInt ma,na,mb,nb, i;
2406: MatGetSize(A,&ma,&na);
2407: MatGetSize(B,&mb,&nb);
2408: if (ma!=nb || na!=mb) {
2409: *f = PETSC_FALSE;
2410: return(0);
2411: }
2412: aii = aij->i; bii = bij->i;
2413: adx = aij->j; bdx = bij->j;
2414: va = aij->a; vb = bij->a;
2415: PetscMalloc1(ma,&aptr);
2416: PetscMalloc1(mb,&bptr);
2417: for (i=0; i<ma; i++) aptr[i] = aii[i];
2418: for (i=0; i<mb; i++) bptr[i] = bii[i];
2420: *f = PETSC_TRUE;
2421: for (i=0; i<ma; i++) {
2422: while (aptr[i]<aii[i+1]) {
2423: PetscInt idc,idr;
2424: PetscScalar vc,vr;
2425: /* column/row index/value */
2426: idc = adx[aptr[i]];
2427: idr = bdx[bptr[idc]];
2428: vc = va[aptr[i]];
2429: vr = vb[bptr[idc]];
2430: if (i!=idr || PetscAbsScalar(vc-vr) > tol) {
2431: *f = PETSC_FALSE;
2432: goto done;
2433: } else {
2434: aptr[i]++;
2435: if (B || i!=idc) bptr[idc]++;
2436: }
2437: }
2438: }
2439: done:
2440: PetscFree(aptr);
2441: PetscFree(bptr);
2442: return(0);
2443: }
2445: PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
2446: {
2447: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) B->data;
2448: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr;
2449: MatScalar *va,*vb;
2451: PetscInt ma,na,mb,nb, i;
2454: MatGetSize(A,&ma,&na);
2455: MatGetSize(B,&mb,&nb);
2456: if (ma!=nb || na!=mb) {
2457: *f = PETSC_FALSE;
2458: return(0);
2459: }
2460: aii = aij->i; bii = bij->i;
2461: adx = aij->j; bdx = bij->j;
2462: va = aij->a; vb = bij->a;
2463: PetscMalloc1(ma,&aptr);
2464: PetscMalloc1(mb,&bptr);
2465: for (i=0; i<ma; i++) aptr[i] = aii[i];
2466: for (i=0; i<mb; i++) bptr[i] = bii[i];
2468: *f = PETSC_TRUE;
2469: for (i=0; i<ma; i++) {
2470: while (aptr[i]<aii[i+1]) {
2471: PetscInt idc,idr;
2472: PetscScalar vc,vr;
2473: /* column/row index/value */
2474: idc = adx[aptr[i]];
2475: idr = bdx[bptr[idc]];
2476: vc = va[aptr[i]];
2477: vr = vb[bptr[idc]];
2478: if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) {
2479: *f = PETSC_FALSE;
2480: goto done;
2481: } else {
2482: aptr[i]++;
2483: if (B || i!=idc) bptr[idc]++;
2484: }
2485: }
2486: }
2487: done:
2488: PetscFree(aptr);
2489: PetscFree(bptr);
2490: return(0);
2491: }
2493: PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f)
2494: {
2498: MatIsTranspose_SeqAIJ(A,A,tol,f);
2499: return(0);
2500: }
2502: PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f)
2503: {
2507: MatIsHermitianTranspose_SeqAIJ(A,A,tol,f);
2508: return(0);
2509: }
2511: PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr)
2512: {
2513: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2514: const PetscScalar *l,*r;
2515: PetscScalar x;
2516: MatScalar *v;
2517: PetscErrorCode ierr;
2518: PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz;
2519: const PetscInt *jj;
2522: if (ll) {
2523: /* The local size is used so that VecMPI can be passed to this routine
2524: by MatDiagonalScale_MPIAIJ */
2525: VecGetLocalSize(ll,&m);
2526: if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length");
2527: VecGetArrayRead(ll,&l);
2528: v = a->a;
2529: for (i=0; i<m; i++) {
2530: x = l[i];
2531: M = a->i[i+1] - a->i[i];
2532: for (j=0; j<M; j++) (*v++) *= x;
2533: }
2534: VecRestoreArrayRead(ll,&l);
2535: PetscLogFlops(nz);
2536: }
2537: if (rr) {
2538: VecGetLocalSize(rr,&n);
2539: if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length");
2540: VecGetArrayRead(rr,&r);
2541: v = a->a; jj = a->j;
2542: for (i=0; i<nz; i++) (*v++) *= r[*jj++];
2543: VecRestoreArrayRead(rr,&r);
2544: PetscLogFlops(nz);
2545: }
2546: MatSeqAIJInvalidateDiagonal(A);
2547: #if defined(PETSC_HAVE_DEVICE)
2548: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED) A->offloadmask = PETSC_OFFLOAD_CPU;
2549: #endif
2550: return(0);
2551: }
2553: PetscErrorCode MatCreateSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B)
2554: {
2555: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c;
2557: PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens;
2558: PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi;
2559: const PetscInt *irow,*icol;
2560: PetscInt nrows,ncols;
2561: PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen;
2562: MatScalar *a_new,*mat_a;
2563: Mat C;
2564: PetscBool stride;
2568: ISGetIndices(isrow,&irow);
2569: ISGetLocalSize(isrow,&nrows);
2570: ISGetLocalSize(iscol,&ncols);
2572: PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride);
2573: if (stride) {
2574: ISStrideGetInfo(iscol,&first,&step);
2575: } else {
2576: first = 0;
2577: step = 0;
2578: }
2579: if (stride && step == 1) {
2580: /* special case of contiguous rows */
2581: PetscMalloc2(nrows,&lens,nrows,&starts);
2582: /* loop over new rows determining lens and starting points */
2583: for (i=0; i<nrows; i++) {
2584: kstart = ai[irow[i]];
2585: kend = kstart + ailen[irow[i]];
2586: starts[i] = kstart;
2587: for (k=kstart; k<kend; k++) {
2588: if (aj[k] >= first) {
2589: starts[i] = k;
2590: break;
2591: }
2592: }
2593: sum = 0;
2594: while (k < kend) {
2595: if (aj[k++] >= first+ncols) break;
2596: sum++;
2597: }
2598: lens[i] = sum;
2599: }
2600: /* create submatrix */
2601: if (scall == MAT_REUSE_MATRIX) {
2602: PetscInt n_cols,n_rows;
2603: MatGetSize(*B,&n_rows,&n_cols);
2604: if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size");
2605: MatZeroEntries(*B);
2606: C = *B;
2607: } else {
2608: PetscInt rbs,cbs;
2609: MatCreate(PetscObjectComm((PetscObject)A),&C);
2610: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
2611: ISGetBlockSize(isrow,&rbs);
2612: ISGetBlockSize(iscol,&cbs);
2613: MatSetBlockSizes(C,rbs,cbs);
2614: MatSetType(C,((PetscObject)A)->type_name);
2615: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
2616: }
2617: c = (Mat_SeqAIJ*)C->data;
2619: /* loop over rows inserting into submatrix */
2620: a_new = c->a;
2621: j_new = c->j;
2622: i_new = c->i;
2624: for (i=0; i<nrows; i++) {
2625: ii = starts[i];
2626: lensi = lens[i];
2627: for (k=0; k<lensi; k++) {
2628: *j_new++ = aj[ii+k] - first;
2629: }
2630: PetscArraycpy(a_new,a->a + starts[i],lensi);
2631: a_new += lensi;
2632: i_new[i+1] = i_new[i] + lensi;
2633: c->ilen[i] = lensi;
2634: }
2635: PetscFree2(lens,starts);
2636: } else {
2637: ISGetIndices(iscol,&icol);
2638: PetscCalloc1(oldcols,&smap);
2639: PetscMalloc1(1+nrows,&lens);
2640: for (i=0; i<ncols; i++) {
2641: if (PetscUnlikelyDebug(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);
2642: smap[icol[i]] = i+1;
2643: }
2645: /* determine lens of each row */
2646: for (i=0; i<nrows; i++) {
2647: kstart = ai[irow[i]];
2648: kend = kstart + a->ilen[irow[i]];
2649: lens[i] = 0;
2650: for (k=kstart; k<kend; k++) {
2651: if (smap[aj[k]]) {
2652: lens[i]++;
2653: }
2654: }
2655: }
2656: /* Create and fill new matrix */
2657: if (scall == MAT_REUSE_MATRIX) {
2658: PetscBool equal;
2660: c = (Mat_SeqAIJ*)((*B)->data);
2661: if ((*B)->rmap->n != nrows || (*B)->cmap->n != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size");
2662: PetscArraycmp(c->ilen,lens,(*B)->rmap->n,&equal);
2663: if (!equal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros");
2664: PetscArrayzero(c->ilen,(*B)->rmap->n);
2665: C = *B;
2666: } else {
2667: PetscInt rbs,cbs;
2668: MatCreate(PetscObjectComm((PetscObject)A),&C);
2669: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
2670: ISGetBlockSize(isrow,&rbs);
2671: ISGetBlockSize(iscol,&cbs);
2672: MatSetBlockSizes(C,rbs,cbs);
2673: MatSetType(C,((PetscObject)A)->type_name);
2674: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
2675: }
2676: c = (Mat_SeqAIJ*)(C->data);
2677: for (i=0; i<nrows; i++) {
2678: row = irow[i];
2679: kstart = ai[row];
2680: kend = kstart + a->ilen[row];
2681: mat_i = c->i[i];
2682: mat_j = c->j + mat_i;
2683: mat_a = c->a + mat_i;
2684: mat_ilen = c->ilen + i;
2685: for (k=kstart; k<kend; k++) {
2686: if ((tcol=smap[a->j[k]])) {
2687: *mat_j++ = tcol - 1;
2688: *mat_a++ = a->a[k];
2689: (*mat_ilen)++;
2691: }
2692: }
2693: }
2694: /* Free work space */
2695: ISRestoreIndices(iscol,&icol);
2696: PetscFree(smap);
2697: PetscFree(lens);
2698: /* sort */
2699: for (i = 0; i < nrows; i++) {
2700: PetscInt ilen;
2702: mat_i = c->i[i];
2703: mat_j = c->j + mat_i;
2704: mat_a = c->a + mat_i;
2705: ilen = c->ilen[i];
2706: PetscSortIntWithScalarArray(ilen,mat_j,mat_a);
2707: }
2708: }
2709: #if defined(PETSC_HAVE_DEVICE)
2710: MatBindToCPU(C,A->boundtocpu);
2711: #endif
2712: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
2713: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
2715: ISRestoreIndices(isrow,&irow);
2716: *B = C;
2717: return(0);
2718: }
2720: PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat)
2721: {
2723: Mat B;
2726: if (scall == MAT_INITIAL_MATRIX) {
2727: MatCreate(subComm,&B);
2728: MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n);
2729: MatSetBlockSizesFromMats(B,mat,mat);
2730: MatSetType(B,MATSEQAIJ);
2731: MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE);
2732: *subMat = B;
2733: } else {
2734: MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN);
2735: }
2736: return(0);
2737: }
2739: PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info)
2740: {
2741: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
2743: Mat outA;
2744: PetscBool row_identity,col_identity;
2747: if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu");
2749: ISIdentity(row,&row_identity);
2750: ISIdentity(col,&col_identity);
2752: outA = inA;
2753: outA->factortype = MAT_FACTOR_LU;
2754: PetscFree(inA->solvertype);
2755: PetscStrallocpy(MATSOLVERPETSC,&inA->solvertype);
2757: PetscObjectReference((PetscObject)row);
2758: ISDestroy(&a->row);
2760: a->row = row;
2762: PetscObjectReference((PetscObject)col);
2763: ISDestroy(&a->col);
2765: a->col = col;
2767: /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */
2768: ISDestroy(&a->icol);
2769: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
2770: PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);
2772: if (!a->solve_work) { /* this matrix may have been factored before */
2773: PetscMalloc1(inA->rmap->n+1,&a->solve_work);
2774: PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar));
2775: }
2777: MatMarkDiagonal_SeqAIJ(inA);
2778: if (row_identity && col_identity) {
2779: MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info);
2780: } else {
2781: MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info);
2782: }
2783: return(0);
2784: }
2786: PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha)
2787: {
2788: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
2789: PetscScalar oalpha = alpha;
2791: PetscBLASInt one = 1,bnz;
2794: PetscBLASIntCast(a->nz,&bnz);
2795: PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&oalpha,a->a,&one));
2796: PetscLogFlops(a->nz);
2797: MatSeqAIJInvalidateDiagonal(inA);
2798: #if defined(PETSC_HAVE_DEVICE)
2799: if (inA->offloadmask != PETSC_OFFLOAD_UNALLOCATED) inA->offloadmask = PETSC_OFFLOAD_CPU;
2800: #endif
2801: return(0);
2802: }
2804: PetscErrorCode MatDestroySubMatrix_Private(Mat_SubSppt *submatj)
2805: {
2807: PetscInt i;
2810: if (!submatj->id) { /* delete data that are linked only to submats[id=0] */
2811: PetscFree4(submatj->sbuf1,submatj->ptr,submatj->tmp,submatj->ctr);
2813: for (i=0; i<submatj->nrqr; ++i) {
2814: PetscFree(submatj->sbuf2[i]);
2815: }
2816: PetscFree3(submatj->sbuf2,submatj->req_size,submatj->req_source1);
2818: if (submatj->rbuf1) {
2819: PetscFree(submatj->rbuf1[0]);
2820: PetscFree(submatj->rbuf1);
2821: }
2823: for (i=0; i<submatj->nrqs; ++i) {
2824: PetscFree(submatj->rbuf3[i]);
2825: }
2826: PetscFree3(submatj->req_source2,submatj->rbuf2,submatj->rbuf3);
2827: PetscFree(submatj->pa);
2828: }
2830: #if defined(PETSC_USE_CTABLE)
2831: PetscTableDestroy((PetscTable*)&submatj->rmap);
2832: if (submatj->cmap_loc) {PetscFree(submatj->cmap_loc);}
2833: PetscFree(submatj->rmap_loc);
2834: #else
2835: PetscFree(submatj->rmap);
2836: #endif
2838: if (!submatj->allcolumns) {
2839: #if defined(PETSC_USE_CTABLE)
2840: PetscTableDestroy((PetscTable*)&submatj->cmap);
2841: #else
2842: PetscFree(submatj->cmap);
2843: #endif
2844: }
2845: PetscFree(submatj->row2proc);
2847: PetscFree(submatj);
2848: return(0);
2849: }
2851: PetscErrorCode MatDestroySubMatrix_SeqAIJ(Mat C)
2852: {
2854: Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data;
2855: Mat_SubSppt *submatj = c->submatis1;
2858: (*submatj->destroy)(C);
2859: MatDestroySubMatrix_Private(submatj);
2860: return(0);
2861: }
2863: PetscErrorCode MatDestroySubMatrices_SeqAIJ(PetscInt n,Mat *mat[])
2864: {
2866: PetscInt i;
2867: Mat C;
2868: Mat_SeqAIJ *c;
2869: Mat_SubSppt *submatj;
2872: for (i=0; i<n; i++) {
2873: C = (*mat)[i];
2874: c = (Mat_SeqAIJ*)C->data;
2875: submatj = c->submatis1;
2876: if (submatj) {
2877: if (--((PetscObject)C)->refct <= 0) {
2878: (*submatj->destroy)(C);
2879: MatDestroySubMatrix_Private(submatj);
2880: PetscFree(C->defaultvectype);
2881: PetscLayoutDestroy(&C->rmap);
2882: PetscLayoutDestroy(&C->cmap);
2883: PetscHeaderDestroy(&C);
2884: }
2885: } else {
2886: MatDestroy(&C);
2887: }
2888: }
2890: /* Destroy Dummy submatrices created for reuse */
2891: MatDestroySubMatrices_Dummy(n,mat);
2893: PetscFree(*mat);
2894: return(0);
2895: }
2897: PetscErrorCode MatCreateSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
2898: {
2900: PetscInt i;
2903: if (scall == MAT_INITIAL_MATRIX) {
2904: PetscCalloc1(n+1,B);
2905: }
2907: for (i=0; i<n; i++) {
2908: MatCreateSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);
2909: }
2910: return(0);
2911: }
2913: PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov)
2914: {
2915: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2917: PetscInt row,i,j,k,l,m,n,*nidx,isz,val;
2918: const PetscInt *idx;
2919: PetscInt start,end,*ai,*aj;
2920: PetscBT table;
2923: m = A->rmap->n;
2924: ai = a->i;
2925: aj = a->j;
2927: if (ov < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used");
2929: PetscMalloc1(m+1,&nidx);
2930: PetscBTCreate(m,&table);
2932: for (i=0; i<is_max; i++) {
2933: /* Initialize the two local arrays */
2934: isz = 0;
2935: PetscBTMemzero(m,table);
2937: /* Extract the indices, assume there can be duplicate entries */
2938: ISGetIndices(is[i],&idx);
2939: ISGetLocalSize(is[i],&n);
2941: /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */
2942: for (j=0; j<n; ++j) {
2943: if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j];
2944: }
2945: ISRestoreIndices(is[i],&idx);
2946: ISDestroy(&is[i]);
2948: k = 0;
2949: for (j=0; j<ov; j++) { /* for each overlap */
2950: n = isz;
2951: for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */
2952: row = nidx[k];
2953: start = ai[row];
2954: end = ai[row+1];
2955: for (l = start; l<end; l++) {
2956: val = aj[l];
2957: if (!PetscBTLookupSet(table,val)) nidx[isz++] = val;
2958: }
2959: }
2960: }
2961: ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i));
2962: }
2963: PetscBTDestroy(&table);
2964: PetscFree(nidx);
2965: return(0);
2966: }
2968: /* -------------------------------------------------------------- */
2969: PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B)
2970: {
2971: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2973: PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n;
2974: const PetscInt *row,*col;
2975: PetscInt *cnew,j,*lens;
2976: IS icolp,irowp;
2977: PetscInt *cwork = NULL;
2978: PetscScalar *vwork = NULL;
2981: ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);
2982: ISGetIndices(irowp,&row);
2983: ISInvertPermutation(colp,PETSC_DECIDE,&icolp);
2984: ISGetIndices(icolp,&col);
2986: /* determine lengths of permuted rows */
2987: PetscMalloc1(m+1,&lens);
2988: for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i];
2989: MatCreate(PetscObjectComm((PetscObject)A),B);
2990: MatSetSizes(*B,m,n,m,n);
2991: MatSetBlockSizesFromMats(*B,A,A);
2992: MatSetType(*B,((PetscObject)A)->type_name);
2993: MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);
2994: PetscFree(lens);
2996: PetscMalloc1(n,&cnew);
2997: for (i=0; i<m; i++) {
2998: MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
2999: for (j=0; j<nz; j++) cnew[j] = col[cwork[j]];
3000: MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);
3001: MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
3002: }
3003: PetscFree(cnew);
3005: (*B)->assembled = PETSC_FALSE;
3007: #if defined(PETSC_HAVE_DEVICE)
3008: MatBindToCPU(*B,A->boundtocpu);
3009: #endif
3010: MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);
3011: MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);
3012: ISRestoreIndices(irowp,&row);
3013: ISRestoreIndices(icolp,&col);
3014: ISDestroy(&irowp);
3015: ISDestroy(&icolp);
3016: if (rowp == colp) {
3017: if (A->symmetric) {
3018: MatSetOption(*B,MAT_SYMMETRIC,PETSC_TRUE);
3019: }
3020: if (A->hermitian) {
3021: MatSetOption(*B,MAT_HERMITIAN,PETSC_TRUE);
3022: }
3023: }
3024: return(0);
3025: }
3027: PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str)
3028: {
3032: /* If the two matrices have the same copy implementation, use fast copy. */
3033: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
3034: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3035: Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data;
3037: if (a->i[A->rmap->n] != b->i[B->rmap->n]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzeros in two matrices are different %D != %D",a->i[A->rmap->n],b->i[B->rmap->n]);
3038: PetscArraycpy(b->a,a->a,a->i[A->rmap->n]);
3039: PetscObjectStateIncrease((PetscObject)B);
3040: } else {
3041: MatCopy_Basic(A,B,str);
3042: }
3043: return(0);
3044: }
3046: PetscErrorCode MatSetUp_SeqAIJ(Mat A)
3047: {
3051: MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,NULL);
3052: return(0);
3053: }
3055: PETSC_INTERN PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[])
3056: {
3057: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3060: *array = a->a;
3061: return(0);
3062: }
3064: PETSC_INTERN PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[])
3065: {
3067: *array = NULL;
3068: return(0);
3069: }
3071: /*
3072: Computes the number of nonzeros per row needed for preallocation when X and Y
3073: have different nonzero structure.
3074: */
3075: PetscErrorCode MatAXPYGetPreallocation_SeqX_private(PetscInt m,const PetscInt *xi,const PetscInt *xj,const PetscInt *yi,const PetscInt *yj,PetscInt *nnz)
3076: {
3077: PetscInt i,j,k,nzx,nzy;
3080: /* Set the number of nonzeros in the new matrix */
3081: for (i=0; i<m; i++) {
3082: const PetscInt *xjj = xj+xi[i],*yjj = yj+yi[i];
3083: nzx = xi[i+1] - xi[i];
3084: nzy = yi[i+1] - yi[i];
3085: nnz[i] = 0;
3086: for (j=0,k=0; j<nzx; j++) { /* Point in X */
3087: for (; k<nzy && yjj[k]<xjj[j]; k++) nnz[i]++; /* Catch up to X */
3088: if (k<nzy && yjj[k]==xjj[j]) k++; /* Skip duplicate */
3089: nnz[i]++;
3090: }
3091: for (; k<nzy; k++) nnz[i]++;
3092: }
3093: return(0);
3094: }
3096: PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz)
3097: {
3098: PetscInt m = Y->rmap->N;
3099: Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data;
3100: Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data;
3104: /* Set the number of nonzeros in the new matrix */
3105: MatAXPYGetPreallocation_SeqX_private(m,x->i,x->j,y->i,y->j,nnz);
3106: return(0);
3107: }
3109: PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
3110: {
3112: Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data;
3115: if (str == DIFFERENT_NONZERO_PATTERN) {
3116: if (x->nz == y->nz) {
3117: PetscBool e;
3118: PetscArraycmp(x->i,y->i,Y->rmap->n+1,&e);
3119: if (e) {
3120: PetscArraycmp(x->j,y->j,y->nz,&e);
3121: if (e) {
3122: str = SAME_NONZERO_PATTERN;
3123: }
3124: }
3125: }
3126: }
3127: if (str == SAME_NONZERO_PATTERN) {
3128: PetscScalar alpha = a;
3129: PetscBLASInt one = 1,bnz;
3131: PetscBLASIntCast(x->nz,&bnz);
3132: PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
3133: MatSeqAIJInvalidateDiagonal(Y);
3134: PetscObjectStateIncrease((PetscObject)Y);
3135: /* the MatAXPY_Basic* subroutines calls MatAssembly, so the matrix on the GPU will be updated */
3136: #if defined(PETSC_HAVE_DEVICE)
3137: if (Y->offloadmask != PETSC_OFFLOAD_UNALLOCATED) {
3138: Y->offloadmask = PETSC_OFFLOAD_CPU;
3139: }
3140: #endif
3141: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
3142: MatAXPY_Basic(Y,a,X,str);
3143: } else {
3144: Mat B;
3145: PetscInt *nnz;
3146: PetscMalloc1(Y->rmap->N,&nnz);
3147: MatCreate(PetscObjectComm((PetscObject)Y),&B);
3148: PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);
3149: MatSetLayouts(B,Y->rmap,Y->cmap);
3150: MatSetType(B,(MatType) ((PetscObject)Y)->type_name);
3151: MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);
3152: MatSeqAIJSetPreallocation(B,0,nnz);
3153: MatAXPY_BasicWithPreallocation(B,Y,a,X,str);
3154: MatHeaderReplace(Y,&B);
3155: PetscFree(nnz);
3156: }
3157: return(0);
3158: }
3160: PetscErrorCode MatConjugate_SeqAIJ(Mat mat)
3161: {
3162: #if defined(PETSC_USE_COMPLEX)
3163: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3164: PetscInt i,nz;
3165: PetscScalar *a;
3168: nz = aij->nz;
3169: a = aij->a;
3170: for (i=0; i<nz; i++) a[i] = PetscConj(a[i]);
3171: #if defined(PETSC_HAVE_DEVICE)
3172: if (mat->offloadmask != PETSC_OFFLOAD_UNALLOCATED) mat->offloadmask = PETSC_OFFLOAD_CPU;
3173: #endif
3174: #else
3176: #endif
3177: return(0);
3178: }
3180: PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3181: {
3182: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3184: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3185: PetscReal atmp;
3186: PetscScalar *x;
3187: MatScalar *aa;
3190: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3191: aa = a->a;
3192: ai = a->i;
3193: aj = a->j;
3195: VecSet(v,0.0);
3196: VecGetArrayWrite(v,&x);
3197: VecGetLocalSize(v,&n);
3198: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3199: for (i=0; i<m; i++) {
3200: ncols = ai[1] - ai[0]; ai++;
3201: for (j=0; j<ncols; j++) {
3202: atmp = PetscAbsScalar(*aa);
3203: if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;}
3204: aa++; aj++;
3205: }
3206: }
3207: VecRestoreArrayWrite(v,&x);
3208: return(0);
3209: }
3211: PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3212: {
3213: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3215: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3216: PetscScalar *x;
3217: MatScalar *aa;
3220: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3221: aa = a->a;
3222: ai = a->i;
3223: aj = a->j;
3225: VecSet(v,0.0);
3226: VecGetArrayWrite(v,&x);
3227: VecGetLocalSize(v,&n);
3228: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3229: for (i=0; i<m; i++) {
3230: ncols = ai[1] - ai[0]; ai++;
3231: if (ncols == A->cmap->n) { /* row is dense */
3232: x[i] = *aa; if (idx) idx[i] = 0;
3233: } else { /* row is sparse so already KNOW maximum is 0.0 or higher */
3234: x[i] = 0.0;
3235: if (idx) {
3236: for (j=0; j<ncols; j++) { /* find first implicit 0.0 in the row */
3237: if (aj[j] > j) {
3238: idx[i] = j;
3239: break;
3240: }
3241: }
3242: /* in case first implicit 0.0 in the row occurs at ncols-th column */
3243: if (j==ncols && j < A->cmap->n) idx[i] = j;
3244: }
3245: }
3246: for (j=0; j<ncols; j++) {
3247: if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
3248: aa++; aj++;
3249: }
3250: }
3251: VecRestoreArrayWrite(v,&x);
3252: return(0);
3253: }
3255: PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3256: {
3257: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3259: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3260: PetscScalar *x,*aa;
3263: aa = a->a;
3264: ai = a->i;
3265: aj = a->j;
3267: VecSet(v,0.0);
3268: VecGetArrayWrite(v,&x);
3269: VecGetLocalSize(v,&n);
3270: if (n != m) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector, %D vs. %D rows", m, n);
3271: for (i=0; i<m; i++) {
3272: ncols = ai[1] - ai[0]; ai++;
3273: if (ncols == A->cmap->n) { /* row is dense */
3274: x[i] = *aa; if (idx) idx[i] = 0;
3275: } else { /* row is sparse so already KNOW minimum is 0.0 or higher */
3276: x[i] = 0.0;
3277: if (idx) { /* find first implicit 0.0 in the row */
3278: for (j=0; j<ncols; j++) {
3279: if (aj[j] > j) {
3280: idx[i] = j;
3281: break;
3282: }
3283: }
3284: /* in case first implicit 0.0 in the row occurs at ncols-th column */
3285: if (j==ncols && j < A->cmap->n) idx[i] = j;
3286: }
3287: }
3288: for (j=0; j<ncols; j++) {
3289: if (PetscAbsScalar(x[i]) > PetscAbsScalar(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
3290: aa++; aj++;
3291: }
3292: }
3293: VecRestoreArrayWrite(v,&x);
3294: return(0);
3295: }
3297: PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3298: {
3299: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3300: PetscErrorCode ierr;
3301: PetscInt i,j,m = A->rmap->n,ncols,n;
3302: const PetscInt *ai,*aj;
3303: PetscScalar *x;
3304: const MatScalar *aa;
3307: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3308: aa = a->a;
3309: ai = a->i;
3310: aj = a->j;
3312: VecSet(v,0.0);
3313: VecGetArrayWrite(v,&x);
3314: VecGetLocalSize(v,&n);
3315: if (n != m) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3316: for (i=0; i<m; i++) {
3317: ncols = ai[1] - ai[0]; ai++;
3318: if (ncols == A->cmap->n) { /* row is dense */
3319: x[i] = *aa; if (idx) idx[i] = 0;
3320: } else { /* row is sparse so already KNOW minimum is 0.0 or lower */
3321: x[i] = 0.0;
3322: if (idx) { /* find first implicit 0.0 in the row */
3323: for (j=0; j<ncols; j++) {
3324: if (aj[j] > j) {
3325: idx[i] = j;
3326: break;
3327: }
3328: }
3329: /* in case first implicit 0.0 in the row occurs at ncols-th column */
3330: if (j==ncols && j < A->cmap->n) idx[i] = j;
3331: }
3332: }
3333: for (j=0; j<ncols; j++) {
3334: if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
3335: aa++; aj++;
3336: }
3337: }
3338: VecRestoreArrayWrite(v,&x);
3339: return(0);
3340: }
3342: PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values)
3343: {
3344: Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data;
3345: PetscErrorCode ierr;
3346: PetscInt i,bs = PetscAbs(A->rmap->bs),mbs = A->rmap->n/bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j;
3347: MatScalar *diag,work[25],*v_work;
3348: const PetscReal shift = 0.0;
3349: PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE;
3352: allowzeropivot = PetscNot(A->erroriffailure);
3353: if (a->ibdiagvalid) {
3354: if (values) *values = a->ibdiag;
3355: return(0);
3356: }
3357: MatMarkDiagonal_SeqAIJ(A);
3358: if (!a->ibdiag) {
3359: PetscMalloc1(bs2*mbs,&a->ibdiag);
3360: PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));
3361: }
3362: diag = a->ibdiag;
3363: if (values) *values = a->ibdiag;
3364: /* factor and invert each block */
3365: switch (bs) {
3366: case 1:
3367: for (i=0; i<mbs; i++) {
3368: MatGetValues(A,1,&i,1,&i,diag+i);
3369: if (PetscAbsScalar(diag[i] + shift) < PETSC_MACHINE_EPSILON) {
3370: if (allowzeropivot) {
3371: A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3372: A->factorerror_zeropivot_value = PetscAbsScalar(diag[i]);
3373: A->factorerror_zeropivot_row = i;
3374: PetscInfo3(A,"Zero pivot, row %D pivot %g tolerance %g\n",i,(double)PetscAbsScalar(diag[i]),(double)PETSC_MACHINE_EPSILON);
3375: } 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);
3376: }
3377: diag[i] = (PetscScalar)1.0 / (diag[i] + shift);
3378: }
3379: break;
3380: case 2:
3381: for (i=0; i<mbs; i++) {
3382: ij[0] = 2*i; ij[1] = 2*i + 1;
3383: MatGetValues(A,2,ij,2,ij,diag);
3384: PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected);
3385: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3386: PetscKernel_A_gets_transpose_A_2(diag);
3387: diag += 4;
3388: }
3389: break;
3390: case 3:
3391: for (i=0; i<mbs; i++) {
3392: ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2;
3393: MatGetValues(A,3,ij,3,ij,diag);
3394: PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected);
3395: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3396: PetscKernel_A_gets_transpose_A_3(diag);
3397: diag += 9;
3398: }
3399: break;
3400: case 4:
3401: for (i=0; i<mbs; i++) {
3402: ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3;
3403: MatGetValues(A,4,ij,4,ij,diag);
3404: PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected);
3405: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3406: PetscKernel_A_gets_transpose_A_4(diag);
3407: diag += 16;
3408: }
3409: break;
3410: case 5:
3411: for (i=0; i<mbs; i++) {
3412: ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4;
3413: MatGetValues(A,5,ij,5,ij,diag);
3414: PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected);
3415: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3416: PetscKernel_A_gets_transpose_A_5(diag);
3417: diag += 25;
3418: }
3419: break;
3420: case 6:
3421: for (i=0; i<mbs; i++) {
3422: 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;
3423: MatGetValues(A,6,ij,6,ij,diag);
3424: PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected);
3425: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3426: PetscKernel_A_gets_transpose_A_6(diag);
3427: diag += 36;
3428: }
3429: break;
3430: case 7:
3431: for (i=0; i<mbs; i++) {
3432: 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;
3433: MatGetValues(A,7,ij,7,ij,diag);
3434: PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected);
3435: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3436: PetscKernel_A_gets_transpose_A_7(diag);
3437: diag += 49;
3438: }
3439: break;
3440: default:
3441: PetscMalloc3(bs,&v_work,bs,&v_pivots,bs,&IJ);
3442: for (i=0; i<mbs; i++) {
3443: for (j=0; j<bs; j++) {
3444: IJ[j] = bs*i + j;
3445: }
3446: MatGetValues(A,bs,IJ,bs,IJ,diag);
3447: PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected);
3448: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
3449: PetscKernel_A_gets_transpose_A_N(diag,bs);
3450: diag += bs2;
3451: }
3452: PetscFree3(v_work,v_pivots,IJ);
3453: }
3454: a->ibdiagvalid = PETSC_TRUE;
3455: return(0);
3456: }
3458: static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx)
3459: {
3461: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data;
3462: PetscScalar a;
3463: PetscInt m,n,i,j,col;
3466: if (!x->assembled) {
3467: MatGetSize(x,&m,&n);
3468: for (i=0; i<m; i++) {
3469: for (j=0; j<aij->imax[i]; j++) {
3470: PetscRandomGetValue(rctx,&a);
3471: col = (PetscInt)(n*PetscRealPart(a));
3472: MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);
3473: }
3474: }
3475: } else {
3476: for (i=0; i<aij->nz; i++) {PetscRandomGetValue(rctx,aij->a+i);}
3477: }
3478: MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);
3479: MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);
3480: return(0);
3481: }
3483: /* Like MatSetRandom_SeqAIJ, but do not set values on columns in range of [low, high) */
3484: PetscErrorCode MatSetRandomSkipColumnRange_SeqAIJ_Private(Mat x,PetscInt low,PetscInt high,PetscRandom rctx)
3485: {
3487: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data;
3488: PetscScalar a;
3489: PetscInt m,n,i,j,col,nskip;
3492: nskip = high - low;
3493: MatGetSize(x,&m,&n);
3494: n -= nskip; /* shrink number of columns where nonzeros can be set */
3495: for (i=0; i<m; i++) {
3496: for (j=0; j<aij->imax[i]; j++) {
3497: PetscRandomGetValue(rctx,&a);
3498: col = (PetscInt)(n*PetscRealPart(a));
3499: if (col >= low) col += nskip; /* shift col rightward to skip the hole */
3500: MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);
3501: }
3502: }
3503: MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);
3504: MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);
3505: return(0);
3506: }
3509: /* -------------------------------------------------------------------*/
3510: static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ,
3511: MatGetRow_SeqAIJ,
3512: MatRestoreRow_SeqAIJ,
3513: MatMult_SeqAIJ,
3514: /* 4*/ MatMultAdd_SeqAIJ,
3515: MatMultTranspose_SeqAIJ,
3516: MatMultTransposeAdd_SeqAIJ,
3517: NULL,
3518: NULL,
3519: NULL,
3520: /* 10*/ NULL,
3521: MatLUFactor_SeqAIJ,
3522: NULL,
3523: MatSOR_SeqAIJ,
3524: MatTranspose_SeqAIJ,
3525: /*1 5*/ MatGetInfo_SeqAIJ,
3526: MatEqual_SeqAIJ,
3527: MatGetDiagonal_SeqAIJ,
3528: MatDiagonalScale_SeqAIJ,
3529: MatNorm_SeqAIJ,
3530: /* 20*/ NULL,
3531: MatAssemblyEnd_SeqAIJ,
3532: MatSetOption_SeqAIJ,
3533: MatZeroEntries_SeqAIJ,
3534: /* 24*/ MatZeroRows_SeqAIJ,
3535: NULL,
3536: NULL,
3537: NULL,
3538: NULL,
3539: /* 29*/ MatSetUp_SeqAIJ,
3540: NULL,
3541: NULL,
3542: NULL,
3543: NULL,
3544: /* 34*/ MatDuplicate_SeqAIJ,
3545: NULL,
3546: NULL,
3547: MatILUFactor_SeqAIJ,
3548: NULL,
3549: /* 39*/ MatAXPY_SeqAIJ,
3550: MatCreateSubMatrices_SeqAIJ,
3551: MatIncreaseOverlap_SeqAIJ,
3552: MatGetValues_SeqAIJ,
3553: MatCopy_SeqAIJ,
3554: /* 44*/ MatGetRowMax_SeqAIJ,
3555: MatScale_SeqAIJ,
3556: MatShift_SeqAIJ,
3557: MatDiagonalSet_SeqAIJ,
3558: MatZeroRowsColumns_SeqAIJ,
3559: /* 49*/ MatSetRandom_SeqAIJ,
3560: MatGetRowIJ_SeqAIJ,
3561: MatRestoreRowIJ_SeqAIJ,
3562: MatGetColumnIJ_SeqAIJ,
3563: MatRestoreColumnIJ_SeqAIJ,
3564: /* 54*/ MatFDColoringCreate_SeqXAIJ,
3565: NULL,
3566: NULL,
3567: MatPermute_SeqAIJ,
3568: NULL,
3569: /* 59*/ NULL,
3570: MatDestroy_SeqAIJ,
3571: MatView_SeqAIJ,
3572: NULL,
3573: NULL,
3574: /* 64*/ NULL,
3575: MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ,
3576: NULL,
3577: NULL,
3578: NULL,
3579: /* 69*/ MatGetRowMaxAbs_SeqAIJ,
3580: MatGetRowMinAbs_SeqAIJ,
3581: NULL,
3582: NULL,
3583: NULL,
3584: /* 74*/ NULL,
3585: MatFDColoringApply_AIJ,
3586: NULL,
3587: NULL,
3588: NULL,
3589: /* 79*/ MatFindZeroDiagonals_SeqAIJ,
3590: NULL,
3591: NULL,
3592: NULL,
3593: MatLoad_SeqAIJ,
3594: /* 84*/ MatIsSymmetric_SeqAIJ,
3595: MatIsHermitian_SeqAIJ,
3596: NULL,
3597: NULL,
3598: NULL,
3599: /* 89*/ NULL,
3600: NULL,
3601: MatMatMultNumeric_SeqAIJ_SeqAIJ,
3602: NULL,
3603: NULL,
3604: /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ_SparseAxpy,
3605: NULL,
3606: NULL,
3607: MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ,
3608: NULL,
3609: /* 99*/ MatProductSetFromOptions_SeqAIJ,
3610: NULL,
3611: NULL,
3612: MatConjugate_SeqAIJ,
3613: NULL,
3614: /*104*/ MatSetValuesRow_SeqAIJ,
3615: MatRealPart_SeqAIJ,
3616: MatImaginaryPart_SeqAIJ,
3617: NULL,
3618: NULL,
3619: /*109*/ MatMatSolve_SeqAIJ,
3620: NULL,
3621: MatGetRowMin_SeqAIJ,
3622: NULL,
3623: MatMissingDiagonal_SeqAIJ,
3624: /*114*/ NULL,
3625: NULL,
3626: NULL,
3627: NULL,
3628: NULL,
3629: /*119*/ NULL,
3630: NULL,
3631: NULL,
3632: NULL,
3633: MatGetMultiProcBlock_SeqAIJ,
3634: /*124*/ MatFindNonzeroRows_SeqAIJ,
3635: MatGetColumnNorms_SeqAIJ,
3636: MatInvertBlockDiagonal_SeqAIJ,
3637: MatInvertVariableBlockDiagonal_SeqAIJ,
3638: NULL,
3639: /*129*/ NULL,
3640: NULL,
3641: NULL,
3642: MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ,
3643: MatTransposeColoringCreate_SeqAIJ,
3644: /*134*/ MatTransColoringApplySpToDen_SeqAIJ,
3645: MatTransColoringApplyDenToSp_SeqAIJ,
3646: NULL,
3647: NULL,
3648: MatRARtNumeric_SeqAIJ_SeqAIJ,
3649: /*139*/NULL,
3650: NULL,
3651: NULL,
3652: MatFDColoringSetUp_SeqXAIJ,
3653: MatFindOffBlockDiagonalEntries_SeqAIJ,
3654: MatCreateMPIMatConcatenateSeqMat_SeqAIJ,
3655: /*145*/MatDestroySubMatrices_SeqAIJ,
3656: NULL,
3657: NULL
3658: };
3660: PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices)
3661: {
3662: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3663: PetscInt i,nz,n;
3666: nz = aij->maxnz;
3667: n = mat->rmap->n;
3668: for (i=0; i<nz; i++) {
3669: aij->j[i] = indices[i];
3670: }
3671: aij->nz = nz;
3672: for (i=0; i<n; i++) {
3673: aij->ilen[i] = aij->imax[i];
3674: }
3675: return(0);
3676: }
3678: /*
3679: * When a sparse matrix has many zero columns, we should compact them out to save the space
3680: * This happens in MatPtAPSymbolic_MPIAIJ_MPIAIJ_scalable()
3681: * */
3682: PetscErrorCode MatSeqAIJCompactOutExtraColumns_SeqAIJ(Mat mat, ISLocalToGlobalMapping *mapping)
3683: {
3684: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3685: PetscTable gid1_lid1;
3686: PetscTablePosition tpos;
3687: PetscInt gid,lid,i,j,ncols,ec;
3688: PetscInt *garray;
3689: PetscErrorCode ierr;
3694: /* use a table */
3695: PetscTableCreate(mat->rmap->n,mat->cmap->N+1,&gid1_lid1);
3696: ec = 0;
3697: for (i=0; i<mat->rmap->n; i++) {
3698: ncols = aij->i[i+1] - aij->i[i];
3699: for (j=0; j<ncols; j++) {
3700: PetscInt data,gid1 = aij->j[aij->i[i] + j] + 1;
3701: PetscTableFind(gid1_lid1,gid1,&data);
3702: if (!data) {
3703: /* one based table */
3704: PetscTableAdd(gid1_lid1,gid1,++ec,INSERT_VALUES);
3705: }
3706: }
3707: }
3708: /* form array of columns we need */
3709: PetscMalloc1(ec+1,&garray);
3710: PetscTableGetHeadPosition(gid1_lid1,&tpos);
3711: while (tpos) {
3712: PetscTableGetNext(gid1_lid1,&tpos,&gid,&lid);
3713: gid--;
3714: lid--;
3715: garray[lid] = gid;
3716: }
3717: PetscSortInt(ec,garray); /* sort, and rebuild */
3718: PetscTableRemoveAll(gid1_lid1);
3719: for (i=0; i<ec; i++) {
3720: PetscTableAdd(gid1_lid1,garray[i]+1,i+1,INSERT_VALUES);
3721: }
3722: /* compact out the extra columns in B */
3723: for (i=0; i<mat->rmap->n; i++) {
3724: ncols = aij->i[i+1] - aij->i[i];
3725: for (j=0; j<ncols; j++) {
3726: PetscInt gid1 = aij->j[aij->i[i] + j] + 1;
3727: PetscTableFind(gid1_lid1,gid1,&lid);
3728: lid--;
3729: aij->j[aij->i[i] + j] = lid;
3730: }
3731: }
3732: PetscLayoutDestroy(&mat->cmap);
3733: PetscLayoutCreateFromSizes(PetscObjectComm((PetscObject)mat),ec,ec,1,&mat->cmap);
3734: PetscTableDestroy(&gid1_lid1);
3735: ISLocalToGlobalMappingCreate(PETSC_COMM_SELF,mat->cmap->bs,mat->cmap->n,garray,PETSC_OWN_POINTER,mapping);
3736: ISLocalToGlobalMappingSetType(*mapping,ISLOCALTOGLOBALMAPPINGHASH);
3737: return(0);
3738: }
3740: /*@
3741: MatSeqAIJSetColumnIndices - Set the column indices for all the rows
3742: in the matrix.
3744: Input Parameters:
3745: + mat - the SeqAIJ matrix
3746: - indices - the column indices
3748: Level: advanced
3750: Notes:
3751: This can be called if you have precomputed the nonzero structure of the
3752: matrix and want to provide it to the matrix object to improve the performance
3753: of the MatSetValues() operation.
3755: You MUST have set the correct numbers of nonzeros per row in the call to
3756: MatCreateSeqAIJ(), and the columns indices MUST be sorted.
3758: MUST be called before any calls to MatSetValues();
3760: The indices should start with zero, not one.
3762: @*/
3763: PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices)
3764: {
3770: PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));
3771: return(0);
3772: }
3774: /* ----------------------------------------------------------------------------------------*/
3776: PetscErrorCode MatStoreValues_SeqAIJ(Mat mat)
3777: {
3778: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3780: size_t nz = aij->i[mat->rmap->n];
3783: if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3785: /* allocate space for values if not already there */
3786: if (!aij->saved_values) {
3787: PetscMalloc1(nz+1,&aij->saved_values);
3788: PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));
3789: }
3791: /* copy values over */
3792: PetscArraycpy(aij->saved_values,aij->a,nz);
3793: return(0);
3794: }
3796: /*@
3797: MatStoreValues - Stashes a copy of the matrix values; this allows, for
3798: example, reuse of the linear part of a Jacobian, while recomputing the
3799: nonlinear portion.
3801: Collect on Mat
3803: Input Parameters:
3804: . mat - the matrix (currently only AIJ matrices support this option)
3806: Level: advanced
3808: Common Usage, with SNESSolve():
3809: $ Create Jacobian matrix
3810: $ Set linear terms into matrix
3811: $ Apply boundary conditions to matrix, at this time matrix must have
3812: $ final nonzero structure (i.e. setting the nonlinear terms and applying
3813: $ boundary conditions again will not change the nonzero structure
3814: $ MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);
3815: $ MatStoreValues(mat);
3816: $ Call SNESSetJacobian() with matrix
3817: $ In your Jacobian routine
3818: $ MatRetrieveValues(mat);
3819: $ Set nonlinear terms in matrix
3821: Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself:
3822: $ // build linear portion of Jacobian
3823: $ MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);
3824: $ MatStoreValues(mat);
3825: $ loop over nonlinear iterations
3826: $ MatRetrieveValues(mat);
3827: $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian
3828: $ // call MatAssemblyBegin/End() on matrix
3829: $ Solve linear system with Jacobian
3830: $ endloop
3832: Notes:
3833: Matrix must already be assemblied before calling this routine
3834: Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before
3835: calling this routine.
3837: When this is called multiple times it overwrites the previous set of stored values
3838: and does not allocated additional space.
3840: .seealso: MatRetrieveValues()
3842: @*/
3843: PetscErrorCode MatStoreValues(Mat mat)
3844: {
3849: if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
3850: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3851: PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));
3852: return(0);
3853: }
3855: PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat)
3856: {
3857: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3859: PetscInt nz = aij->i[mat->rmap->n];
3862: if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3863: if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
3864: /* copy values over */
3865: PetscArraycpy(aij->a,aij->saved_values,nz);
3866: return(0);
3867: }
3869: /*@
3870: MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for
3871: example, reuse of the linear part of a Jacobian, while recomputing the
3872: nonlinear portion.
3874: Collect on Mat
3876: Input Parameters:
3877: . mat - the matrix (currently only AIJ matrices support this option)
3879: Level: advanced
3881: .seealso: MatStoreValues()
3883: @*/
3884: PetscErrorCode MatRetrieveValues(Mat mat)
3885: {
3890: if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
3891: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3892: PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));
3893: return(0);
3894: }
3897: /* --------------------------------------------------------------------------------*/
3898: /*@C
3899: MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format
3900: (the default parallel PETSc format). For good matrix assembly performance
3901: the user should preallocate the matrix storage by setting the parameter nz
3902: (or the array nnz). By setting these parameters accurately, performance
3903: during matrix assembly can be increased by more than a factor of 50.
3905: Collective
3907: Input Parameters:
3908: + comm - MPI communicator, set to PETSC_COMM_SELF
3909: . m - number of rows
3910: . n - number of columns
3911: . nz - number of nonzeros per row (same for all rows)
3912: - nnz - array containing the number of nonzeros in the various rows
3913: (possibly different for each row) or NULL
3915: Output Parameter:
3916: . A - the matrix
3918: It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
3919: MatXXXXSetPreallocation() paradigm instead of this routine directly.
3920: [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
3922: Notes:
3923: If nnz is given then nz is ignored
3925: The AIJ format (also called the Yale sparse matrix format or
3926: compressed row storage), is fully compatible with standard Fortran 77
3927: storage. That is, the stored row and column indices can begin at
3928: either one (as in Fortran) or zero. See the users' manual for details.
3930: Specify the preallocated storage with either nz or nnz (not both).
3931: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3932: allocation. For large problems you MUST preallocate memory or you
3933: will get TERRIBLE performance, see the users' manual chapter on matrices.
3935: By default, this format uses inodes (identical nodes) when possible, to
3936: improve numerical efficiency of matrix-vector products and solves. We
3937: search for consecutive rows with the same nonzero structure, thereby
3938: reusing matrix information to achieve increased efficiency.
3940: Options Database Keys:
3941: + -mat_no_inode - Do not use inodes
3942: - -mat_inode_limit <limit> - Sets inode limit (max limit=5)
3944: Level: intermediate
3946: .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
3948: @*/
3949: PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
3950: {
3954: MatCreate(comm,A);
3955: MatSetSizes(*A,m,n,m,n);
3956: MatSetType(*A,MATSEQAIJ);
3957: MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);
3958: return(0);
3959: }
3961: /*@C
3962: MatSeqAIJSetPreallocation - For good matrix assembly performance
3963: the user should preallocate the matrix storage by setting the parameter nz
3964: (or the array nnz). By setting these parameters accurately, performance
3965: during matrix assembly can be increased by more than a factor of 50.
3967: Collective
3969: Input Parameters:
3970: + B - The matrix
3971: . nz - number of nonzeros per row (same for all rows)
3972: - nnz - array containing the number of nonzeros in the various rows
3973: (possibly different for each row) or NULL
3975: Notes:
3976: If nnz is given then nz is ignored
3978: The AIJ format (also called the Yale sparse matrix format or
3979: compressed row storage), is fully compatible with standard Fortran 77
3980: storage. That is, the stored row and column indices can begin at
3981: either one (as in Fortran) or zero. See the users' manual for details.
3983: Specify the preallocated storage with either nz or nnz (not both).
3984: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3985: allocation. For large problems you MUST preallocate memory or you
3986: will get TERRIBLE performance, see the users' manual chapter on matrices.
3988: You can call MatGetInfo() to get information on how effective the preallocation was;
3989: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3990: You can also run with the option -info and look for messages with the string
3991: malloc in them to see if additional memory allocation was needed.
3993: Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix
3994: entries or columns indices
3996: By default, this format uses inodes (identical nodes) when possible, to
3997: improve numerical efficiency of matrix-vector products and solves. We
3998: search for consecutive rows with the same nonzero structure, thereby
3999: reusing matrix information to achieve increased efficiency.
4001: Options Database Keys:
4002: + -mat_no_inode - Do not use inodes
4003: - -mat_inode_limit <limit> - Sets inode limit (max limit=5)
4005: Level: intermediate
4007: .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo(),
4008: MatSeqAIJSetTotalPreallocation()
4010: @*/
4011: PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[])
4012: {
4018: PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));
4019: return(0);
4020: }
4022: PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz)
4023: {
4024: Mat_SeqAIJ *b;
4025: PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE;
4027: PetscInt i;
4030: if (nz >= 0 || nnz) realalloc = PETSC_TRUE;
4031: if (nz == MAT_SKIP_ALLOCATION) {
4032: skipallocation = PETSC_TRUE;
4033: nz = 0;
4034: }
4035: PetscLayoutSetUp(B->rmap);
4036: PetscLayoutSetUp(B->cmap);
4038: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
4039: if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %D",nz);
4040: if (PetscUnlikelyDebug(nnz)) {
4041: for (i=0; i<B->rmap->n; i++) {
4042: 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]);
4043: 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);
4044: }
4045: }
4047: B->preallocated = PETSC_TRUE;
4049: b = (Mat_SeqAIJ*)B->data;
4051: if (!skipallocation) {
4052: if (!b->imax) {
4053: PetscMalloc1(B->rmap->n,&b->imax);
4054: PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt));
4055: }
4056: if (!b->ilen) {
4057: /* b->ilen will count nonzeros in each row so far. */
4058: PetscCalloc1(B->rmap->n,&b->ilen);
4059: PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt));
4060: } else {
4061: PetscMemzero(b->ilen,B->rmap->n*sizeof(PetscInt));
4062: }
4063: if (!b->ipre) {
4064: PetscMalloc1(B->rmap->n,&b->ipre);
4065: PetscLogObjectMemory((PetscObject)B,B->rmap->n*sizeof(PetscInt));
4066: }
4067: if (!nnz) {
4068: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10;
4069: else if (nz < 0) nz = 1;
4070: nz = PetscMin(nz,B->cmap->n);
4071: for (i=0; i<B->rmap->n; i++) b->imax[i] = nz;
4072: nz = nz*B->rmap->n;
4073: } else {
4074: PetscInt64 nz64 = 0;
4075: for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz64 += nnz[i];}
4076: PetscIntCast(nz64,&nz);
4077: }
4079: /* allocate the matrix space */
4080: /* FIXME: should B's old memory be unlogged? */
4081: MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);
4082: if (B->structure_only) {
4083: PetscMalloc1(nz,&b->j);
4084: PetscMalloc1(B->rmap->n+1,&b->i);
4085: PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*sizeof(PetscInt));
4086: } else {
4087: PetscMalloc3(nz,&b->a,nz,&b->j,B->rmap->n+1,&b->i);
4088: PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));
4089: }
4090: b->i[0] = 0;
4091: for (i=1; i<B->rmap->n+1; i++) {
4092: b->i[i] = b->i[i-1] + b->imax[i-1];
4093: }
4094: if (B->structure_only) {
4095: b->singlemalloc = PETSC_FALSE;
4096: b->free_a = PETSC_FALSE;
4097: } else {
4098: b->singlemalloc = PETSC_TRUE;
4099: b->free_a = PETSC_TRUE;
4100: }
4101: b->free_ij = PETSC_TRUE;
4102: } else {
4103: b->free_a = PETSC_FALSE;
4104: b->free_ij = PETSC_FALSE;
4105: }
4107: if (b->ipre && nnz != b->ipre && b->imax) {
4108: /* reserve user-requested sparsity */
4109: PetscArraycpy(b->ipre,b->imax,B->rmap->n);
4110: }
4113: b->nz = 0;
4114: b->maxnz = nz;
4115: B->info.nz_unneeded = (double)b->maxnz;
4116: if (realalloc) {
4117: MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);
4118: }
4119: B->was_assembled = PETSC_FALSE;
4120: B->assembled = PETSC_FALSE;
4121: return(0);
4122: }
4125: PetscErrorCode MatResetPreallocation_SeqAIJ(Mat A)
4126: {
4127: Mat_SeqAIJ *a;
4128: PetscInt i;
4134: /* Check local size. If zero, then return */
4135: if (!A->rmap->n) return(0);
4137: a = (Mat_SeqAIJ*)A->data;
4138: /* if no saved info, we error out */
4139: if (!a->ipre) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"No saved preallocation info \n");
4141: 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");
4143: PetscArraycpy(a->imax,a->ipre,A->rmap->n);
4144: PetscArrayzero(a->ilen,A->rmap->n);
4145: a->i[0] = 0;
4146: for (i=1; i<A->rmap->n+1; i++) {
4147: a->i[i] = a->i[i-1] + a->imax[i-1];
4148: }
4149: A->preallocated = PETSC_TRUE;
4150: a->nz = 0;
4151: a->maxnz = a->i[A->rmap->n];
4152: A->info.nz_unneeded = (double)a->maxnz;
4153: A->was_assembled = PETSC_FALSE;
4154: A->assembled = PETSC_FALSE;
4155: return(0);
4156: }
4158: /*@
4159: MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format.
4161: Input Parameters:
4162: + B - the matrix
4163: . i - the indices into j for the start of each row (starts with zero)
4164: . j - the column indices for each row (starts with zero) these must be sorted for each row
4165: - v - optional values in the matrix
4167: Level: developer
4169: Notes:
4170: The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays()
4172: This routine may be called multiple times with different nonzero patterns (or the same nonzero pattern). The nonzero
4173: structure will be the union of all the previous nonzero structures.
4175: Developer Notes:
4176: An optimization could be added to the implementation where it checks if the i, and j are identical to the current i and j and
4177: then just copies the v values directly with PetscMemcpy().
4179: This routine could also take a PetscCopyMode argument to allow sharing the values instead of always copying them.
4181: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), MATSEQAIJ, MatResetPreallocation()
4182: @*/
4183: PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[])
4184: {
4190: PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));
4191: return(0);
4192: }
4194: PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[])
4195: {
4196: PetscInt i;
4197: PetscInt m,n;
4198: PetscInt nz;
4199: PetscInt *nnz;
4203: if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]);
4205: PetscLayoutSetUp(B->rmap);
4206: PetscLayoutSetUp(B->cmap);
4208: MatGetSize(B, &m, &n);
4209: PetscMalloc1(m+1, &nnz);
4210: for (i = 0; i < m; i++) {
4211: nz = Ii[i+1]- Ii[i];
4212: if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz);
4213: nnz[i] = nz;
4214: }
4215: MatSeqAIJSetPreallocation(B, 0, nnz);
4216: PetscFree(nnz);
4218: for (i = 0; i < m; i++) {
4219: MatSetValues_SeqAIJ(B, 1, &i, Ii[i+1] - Ii[i], J+Ii[i], v ? v + Ii[i] : NULL, INSERT_VALUES);
4220: }
4222: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
4223: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
4225: MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
4226: return(0);
4227: }
4229: #include <../src/mat/impls/dense/seq/dense.h>
4230: #include <petsc/private/kernels/petscaxpy.h>
4232: /*
4233: Computes (B'*A')' since computing B*A directly is untenable
4235: n p p
4236: [ ] [ ] [ ]
4237: m [ A ] * n [ B ] = m [ C ]
4238: [ ] [ ] [ ]
4240: */
4241: PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C)
4242: {
4243: PetscErrorCode ierr;
4244: Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data;
4245: Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data;
4246: Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data;
4247: PetscInt i,j,n,m,q,p;
4248: const PetscInt *ii,*idx;
4249: const PetscScalar *b,*a,*a_q;
4250: PetscScalar *c,*c_q;
4251: PetscInt clda = sub_c->lda;
4252: PetscInt alda = sub_a->lda;
4255: m = A->rmap->n;
4256: n = A->cmap->n;
4257: p = B->cmap->n;
4258: a = sub_a->v;
4259: b = sub_b->a;
4260: c = sub_c->v;
4261: if (clda == m) {
4262: PetscArrayzero(c,m*p);
4263: } else {
4264: for (j=0;j<p;j++)
4265: for (i=0;i<m;i++)
4266: c[j*clda + i] = 0.0;
4267: }
4268: ii = sub_b->i;
4269: idx = sub_b->j;
4270: for (i=0; i<n; i++) {
4271: q = ii[i+1] - ii[i];
4272: while (q-->0) {
4273: c_q = c + clda*(*idx);
4274: a_q = a + alda*i;
4275: PetscKernelAXPY(c_q,*b,a_q,m);
4276: idx++;
4277: b++;
4278: }
4279: }
4280: return(0);
4281: }
4283: PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat C)
4284: {
4286: PetscInt m=A->rmap->n,n=B->cmap->n;
4287: PetscBool cisdense;
4290: 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);
4291: MatSetSizes(C,m,n,m,n);
4292: MatSetBlockSizesFromMats(C,A,B);
4293: PetscObjectTypeCompareAny((PetscObject)C,&cisdense,MATSEQDENSE,MATSEQDENSECUDA,"");
4294: if (!cisdense) {
4295: MatSetType(C,MATDENSE);
4296: }
4297: MatSetUp(C);
4299: C->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ;
4300: return(0);
4301: }
4303: /* ----------------------------------------------------------------*/
4304: /*MC
4305: MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices,
4306: based on compressed sparse row format.
4308: Options Database Keys:
4309: . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions()
4311: Level: beginner
4313: Notes:
4314: MatSetValues() may be called for this matrix type with a NULL argument for the numerical values,
4315: in this case the values associated with the rows and columns one passes in are set to zero
4316: in the matrix
4318: MatSetOptions(,MAT_STRUCTURE_ONLY,PETSC_TRUE) may be called for this matrix type. In this no
4319: space is allocated for the nonzero entries and any entries passed with MatSetValues() are ignored
4321: Developer Notes:
4322: It would be nice if all matrix formats supported passing NULL in for the numerical values
4324: .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType
4325: M*/
4327: /*MC
4328: MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices.
4330: This matrix type is identical to MATSEQAIJ when constructed with a single process communicator,
4331: and MATMPIAIJ otherwise. As a result, for single process communicators,
4332: MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation() is supported
4333: for communicators controlling multiple processes. It is recommended that you call both of
4334: the above preallocation routines for simplicity.
4336: Options Database Keys:
4337: . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions()
4339: Developer Notes:
4340: Subclasses include MATAIJCUSPARSE, MATAIJPERM, MATAIJSELL, MATAIJMKL, MATAIJCRL, and also automatically switches over to use inodes when
4341: enough exist.
4343: Level: beginner
4345: .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ
4346: M*/
4348: /*MC
4349: MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices.
4351: This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator,
4352: and MATMPIAIJCRL otherwise. As a result, for single process communicators,
4353: MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported
4354: for communicators controlling multiple processes. It is recommended that you call both of
4355: the above preallocation routines for simplicity.
4357: Options Database Keys:
4358: . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions()
4360: Level: beginner
4362: .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL
4363: M*/
4365: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*);
4366: #if defined(PETSC_HAVE_ELEMENTAL)
4367: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_Elemental(Mat,MatType,MatReuse,Mat*);
4368: #endif
4369: #if defined(PETSC_HAVE_SCALAPACK)
4370: PETSC_INTERN PetscErrorCode MatConvert_AIJ_ScaLAPACK(Mat,MatType,MatReuse,Mat*);
4371: #endif
4372: #if defined(PETSC_HAVE_HYPRE)
4373: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat A,MatType,MatReuse,Mat*);
4374: #endif
4375: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqDense(Mat,MatType,MatReuse,Mat*);
4377: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqSELL(Mat,MatType,MatReuse,Mat*);
4378: PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat,MatType,MatReuse,Mat*);
4379: PETSC_INTERN PetscErrorCode MatProductSetFromOptions_IS_XAIJ(Mat);
4381: /*@C
4382: MatSeqAIJGetArray - gives read/write access to the array where the data for a MATSEQAIJ matrix is stored
4384: Not Collective
4386: Input Parameter:
4387: . mat - a MATSEQAIJ matrix
4389: Output Parameter:
4390: . array - pointer to the data
4392: Level: intermediate
4394: .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90()
4395: @*/
4396: PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array)
4397: {
4401: PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));
4402: return(0);
4403: }
4405: /*@C
4406: MatSeqAIJGetArrayRead - gives read-only access to the array where the data for a MATSEQAIJ matrix is stored
4408: Not Collective
4410: Input Parameter:
4411: . mat - a MATSEQAIJ matrix
4413: Output Parameter:
4414: . array - pointer to the data
4416: Level: intermediate
4418: .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayRead()
4419: @*/
4420: PetscErrorCode MatSeqAIJGetArrayRead(Mat A,const PetscScalar **array)
4421: {
4422: #if defined(PETSC_HAVE_DEVICE)
4423: PetscOffloadMask oval;
4424: #endif
4428: #if defined(PETSC_HAVE_DEVICE)
4429: oval = A->offloadmask;
4430: #endif
4431: MatSeqAIJGetArray(A,(PetscScalar**)array);
4432: #if defined(PETSC_HAVE_DEVICE)
4433: if (oval == PETSC_OFFLOAD_GPU || oval == PETSC_OFFLOAD_BOTH) A->offloadmask = PETSC_OFFLOAD_BOTH;
4434: #endif
4435: return(0);
4436: }
4438: /*@C
4439: MatSeqAIJRestoreArrayRead - restore the read-only access array obtained from MatSeqAIJGetArrayRead
4441: Not Collective
4443: Input Parameter:
4444: . mat - a MATSEQAIJ matrix
4446: Output Parameter:
4447: . array - pointer to the data
4449: Level: intermediate
4451: .seealso: MatSeqAIJGetArray(), MatSeqAIJGetArrayRead()
4452: @*/
4453: PetscErrorCode MatSeqAIJRestoreArrayRead(Mat A,const PetscScalar **array)
4454: {
4455: #if defined(PETSC_HAVE_DEVICE)
4456: PetscOffloadMask oval;
4457: #endif
4461: #if defined(PETSC_HAVE_DEVICE)
4462: oval = A->offloadmask;
4463: #endif
4464: MatSeqAIJRestoreArray(A,(PetscScalar**)array);
4465: #if defined(PETSC_HAVE_DEVICE)
4466: A->offloadmask = oval;
4467: #endif
4468: return(0);
4469: }
4471: /*@C
4472: MatSeqAIJGetMaxRowNonzeros - returns the maximum number of nonzeros in any row
4474: Not Collective
4476: Input Parameter:
4477: . mat - a MATSEQAIJ matrix
4479: Output Parameter:
4480: . nz - the maximum number of nonzeros in any row
4482: Level: intermediate
4484: .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90()
4485: @*/
4486: PetscErrorCode MatSeqAIJGetMaxRowNonzeros(Mat A,PetscInt *nz)
4487: {
4488: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data;
4491: *nz = aij->rmax;
4492: return(0);
4493: }
4495: /*@C
4496: MatSeqAIJRestoreArray - returns access to the array where the data for a MATSEQAIJ matrix is stored obtained by MatSeqAIJGetArray()
4498: Not Collective
4500: Input Parameters:
4501: + mat - a MATSEQAIJ matrix
4502: - array - pointer to the data
4504: Level: intermediate
4506: .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90()
4507: @*/
4508: PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array)
4509: {
4513: PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));
4514: return(0);
4515: }
4517: #if defined(PETSC_HAVE_CUDA)
4518: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCUSPARSE(Mat);
4519: #endif
4520: #if defined(PETSC_HAVE_KOKKOS_KERNELS)
4521: PETSC_INTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJKokkos(Mat);
4522: #endif
4524: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B)
4525: {
4526: Mat_SeqAIJ *b;
4528: PetscMPIInt size;
4531: MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);
4532: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1");
4534: PetscNewLog(B,&b);
4536: B->data = (void*)b;
4538: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
4539: if (B->sortedfull) B->ops->setvalues = MatSetValues_SeqAIJ_SortedFull;
4541: b->row = NULL;
4542: b->col = NULL;
4543: b->icol = NULL;
4544: b->reallocs = 0;
4545: b->ignorezeroentries = PETSC_FALSE;
4546: b->roworiented = PETSC_TRUE;
4547: b->nonew = 0;
4548: b->diag = NULL;
4549: b->solve_work = NULL;
4550: B->spptr = NULL;
4551: b->saved_values = NULL;
4552: b->idiag = NULL;
4553: b->mdiag = NULL;
4554: b->ssor_work = NULL;
4555: b->omega = 1.0;
4556: b->fshift = 0.0;
4557: b->idiagvalid = PETSC_FALSE;
4558: b->ibdiagvalid = PETSC_FALSE;
4559: b->keepnonzeropattern = PETSC_FALSE;
4561: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
4562: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);
4563: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);
4565: #if defined(PETSC_HAVE_MATLAB_ENGINE)
4566: PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);
4567: PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);
4568: #endif
4570: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);
4571: PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);
4572: PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);
4573: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);
4574: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);
4575: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);
4576: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijsell_C",MatConvert_SeqAIJ_SeqAIJSELL);
4577: #if defined(PETSC_HAVE_MKL_SPARSE)
4578: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijmkl_C",MatConvert_SeqAIJ_SeqAIJMKL);
4579: #endif
4580: #if defined(PETSC_HAVE_CUDA)
4581: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcusparse_C",MatConvert_SeqAIJ_SeqAIJCUSPARSE);
4582: PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqaijcusparse_seqaij_C",MatProductSetFromOptions_SeqAIJ);
4583: #endif
4584: #if defined(PETSC_HAVE_KOKKOS_KERNELS)
4585: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijkokkos_C",MatConvert_SeqAIJ_SeqAIJKokkos);
4586: #endif
4587: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);
4588: #if defined(PETSC_HAVE_ELEMENTAL)
4589: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_elemental_C",MatConvert_SeqAIJ_Elemental);
4590: #endif
4591: #if defined(PETSC_HAVE_SCALAPACK)
4592: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_scalapack_C",MatConvert_AIJ_ScaLAPACK);
4593: #endif
4594: #if defined(PETSC_HAVE_HYPRE)
4595: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_hypre_C",MatConvert_AIJ_HYPRE);
4596: PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_transpose_seqaij_seqaij_C",MatProductSetFromOptions_Transpose_AIJ_AIJ);
4597: #endif
4598: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqdense_C",MatConvert_SeqAIJ_SeqDense);
4599: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsell_C",MatConvert_SeqAIJ_SeqSELL);
4600: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_is_C",MatConvert_XAIJ_IS);
4601: PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);
4602: PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);
4603: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);
4604: PetscObjectComposeFunction((PetscObject)B,"MatResetPreallocation_C",MatResetPreallocation_SeqAIJ);
4605: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);
4606: PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);
4607: PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_is_seqaij_C",MatProductSetFromOptions_IS_XAIJ);
4608: PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqdense_seqaij_C",MatProductSetFromOptions_SeqDense_SeqAIJ);
4609: PetscObjectComposeFunction((PetscObject)B,"MatProductSetFromOptions_seqaij_seqaij_C",MatProductSetFromOptions_SeqAIJ);
4610: MatCreate_SeqAIJ_Inode(B);
4611: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
4612: MatSeqAIJSetTypeFromOptions(B); /* this allows changing the matrix subtype to say MATSEQAIJPERM */
4613: return(0);
4614: }
4616: /*
4617: Given a matrix generated with MatGetFactor() duplicates all the information in A into B
4618: */
4619: PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace)
4620: {
4621: Mat_SeqAIJ *c = (Mat_SeqAIJ*)C->data,*a = (Mat_SeqAIJ*)A->data;
4623: PetscInt m = A->rmap->n,i;
4626: if (!A->assembled && cpvalues!=MAT_DO_NOT_COPY_VALUES) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Cannot duplicate unassembled matrix");
4628: C->factortype = A->factortype;
4629: c->row = NULL;
4630: c->col = NULL;
4631: c->icol = NULL;
4632: c->reallocs = 0;
4634: C->assembled = PETSC_TRUE;
4636: PetscLayoutReference(A->rmap,&C->rmap);
4637: PetscLayoutReference(A->cmap,&C->cmap);
4639: PetscMalloc1(m,&c->imax);
4640: PetscMemcpy(c->imax,a->imax,m*sizeof(PetscInt));
4641: PetscMalloc1(m,&c->ilen);
4642: PetscMemcpy(c->ilen,a->ilen,m*sizeof(PetscInt));
4643: PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));
4645: /* allocate the matrix space */
4646: if (mallocmatspace) {
4647: PetscMalloc3(a->i[m],&c->a,a->i[m],&c->j,m+1,&c->i);
4648: PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));
4650: c->singlemalloc = PETSC_TRUE;
4652: PetscArraycpy(c->i,a->i,m+1);
4653: if (m > 0) {
4654: PetscArraycpy(c->j,a->j,a->i[m]);
4655: if (cpvalues == MAT_COPY_VALUES) {
4656: PetscArraycpy(c->a,a->a,a->i[m]);
4657: } else {
4658: PetscArrayzero(c->a,a->i[m]);
4659: }
4660: }
4661: }
4663: c->ignorezeroentries = a->ignorezeroentries;
4664: c->roworiented = a->roworiented;
4665: c->nonew = a->nonew;
4666: if (a->diag) {
4667: PetscMalloc1(m+1,&c->diag);
4668: PetscMemcpy(c->diag,a->diag,m*sizeof(PetscInt));
4669: PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));
4670: } else c->diag = NULL;
4672: c->solve_work = NULL;
4673: c->saved_values = NULL;
4674: c->idiag = NULL;
4675: c->ssor_work = NULL;
4676: c->keepnonzeropattern = a->keepnonzeropattern;
4677: c->free_a = PETSC_TRUE;
4678: c->free_ij = PETSC_TRUE;
4680: c->rmax = a->rmax;
4681: c->nz = a->nz;
4682: c->maxnz = a->nz; /* Since we allocate exactly the right amount */
4683: C->preallocated = PETSC_TRUE;
4685: c->compressedrow.use = a->compressedrow.use;
4686: c->compressedrow.nrows = a->compressedrow.nrows;
4687: if (a->compressedrow.use) {
4688: i = a->compressedrow.nrows;
4689: PetscMalloc2(i+1,&c->compressedrow.i,i,&c->compressedrow.rindex);
4690: PetscArraycpy(c->compressedrow.i,a->compressedrow.i,i+1);
4691: PetscArraycpy(c->compressedrow.rindex,a->compressedrow.rindex,i);
4692: } else {
4693: c->compressedrow.use = PETSC_FALSE;
4694: c->compressedrow.i = NULL;
4695: c->compressedrow.rindex = NULL;
4696: }
4697: c->nonzerorowcnt = a->nonzerorowcnt;
4698: C->nonzerostate = A->nonzerostate;
4700: MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);
4701: PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);
4702: return(0);
4703: }
4705: PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
4706: {
4710: MatCreate(PetscObjectComm((PetscObject)A),B);
4711: MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);
4712: if (!(A->rmap->n % A->rmap->bs) && !(A->cmap->n % A->cmap->bs)) {
4713: MatSetBlockSizesFromMats(*B,A,A);
4714: }
4715: MatSetType(*B,((PetscObject)A)->type_name);
4716: MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);
4717: return(0);
4718: }
4720: PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer)
4721: {
4722: PetscBool isbinary, ishdf5;
4728: /* force binary viewer to load .info file if it has not yet done so */
4729: PetscViewerSetUp(viewer);
4730: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
4731: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERHDF5, &ishdf5);
4732: if (isbinary) {
4733: MatLoad_SeqAIJ_Binary(newMat,viewer);
4734: } else if (ishdf5) {
4735: #if defined(PETSC_HAVE_HDF5)
4736: MatLoad_AIJ_HDF5(newMat,viewer);
4737: #else
4738: SETERRQ(PetscObjectComm((PetscObject)newMat),PETSC_ERR_SUP,"HDF5 not supported in this build.\nPlease reconfigure using --download-hdf5");
4739: #endif
4740: } else {
4741: 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);
4742: }
4743: return(0);
4744: }
4746: PetscErrorCode MatLoad_SeqAIJ_Binary(Mat mat, PetscViewer viewer)
4747: {
4748: Mat_SeqAIJ *a = (Mat_SeqAIJ*)mat->data;
4750: PetscInt header[4],*rowlens,M,N,nz,sum,rows,cols,i;
4753: PetscViewerSetUp(viewer);
4755: /* read in matrix header */
4756: PetscViewerBinaryRead(viewer,header,4,NULL,PETSC_INT);
4757: if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Not a matrix object in file");
4758: M = header[1]; N = header[2]; nz = header[3];
4759: if (M < 0) SETERRQ1(PetscObjectComm((PetscObject)viewer),PETSC_ERR_FILE_UNEXPECTED,"Matrix row size (%D) in file is negative",M);
4760: if (N < 0) SETERRQ1(PetscObjectComm((PetscObject)viewer),PETSC_ERR_FILE_UNEXPECTED,"Matrix column size (%D) in file is negative",N);
4761: if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk, cannot load as SeqAIJ");
4763: /* set block sizes from the viewer's .info file */
4764: MatLoad_Binary_BlockSizes(mat,viewer);
4765: /* set local and global sizes if not set already */
4766: if (mat->rmap->n < 0) mat->rmap->n = M;
4767: if (mat->cmap->n < 0) mat->cmap->n = N;
4768: if (mat->rmap->N < 0) mat->rmap->N = M;
4769: if (mat->cmap->N < 0) mat->cmap->N = N;
4770: PetscLayoutSetUp(mat->rmap);
4771: PetscLayoutSetUp(mat->cmap);
4773: /* check if the matrix sizes are correct */
4774: MatGetSize(mat,&rows,&cols);
4775: if (M != rows || N != cols) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different sizes (%D, %D) than the input matrix (%D, %D)",M,N,rows,cols);
4777: /* read in row lengths */
4778: PetscMalloc1(M,&rowlens);
4779: PetscViewerBinaryRead(viewer,rowlens,M,NULL,PETSC_INT);
4780: /* check if sum(rowlens) is same as nz */
4781: sum = 0; for (i=0; i<M; i++) sum += rowlens[i];
4782: if (sum != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Inconsistent matrix data in file: nonzeros = %D, sum-row-lengths = %D\n",nz,sum);
4783: /* preallocate and check sizes */
4784: MatSeqAIJSetPreallocation_SeqAIJ(mat,0,rowlens);
4785: MatGetSize(mat,&rows,&cols);
4786: 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);
4787: /* store row lengths */
4788: PetscArraycpy(a->ilen,rowlens,M);
4789: PetscFree(rowlens);
4791: /* fill in "i" row pointers */
4792: a->i[0] = 0; for (i=0; i<M; i++) a->i[i+1] = a->i[i] + a->ilen[i];
4793: /* read in "j" column indices */
4794: PetscViewerBinaryRead(viewer,a->j,nz,NULL,PETSC_INT);
4795: /* read in "a" nonzero values */
4796: PetscViewerBinaryRead(viewer,a->a,nz,NULL,PETSC_SCALAR);
4798: MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY);
4799: MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY);
4800: return(0);
4801: }
4803: PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg)
4804: {
4805: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data;
4807: #if defined(PETSC_USE_COMPLEX)
4808: PetscInt k;
4809: #endif
4812: /* If the matrix dimensions are not equal,or no of nonzeros */
4813: if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) {
4814: *flg = PETSC_FALSE;
4815: return(0);
4816: }
4818: /* if the a->i are the same */
4819: PetscArraycmp(a->i,b->i,A->rmap->n+1,flg);
4820: if (!*flg) return(0);
4822: /* if a->j are the same */
4823: PetscArraycmp(a->j,b->j,a->nz,flg);
4824: if (!*flg) return(0);
4826: /* if a->a are the same */
4827: #if defined(PETSC_USE_COMPLEX)
4828: for (k=0; k<a->nz; k++) {
4829: if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) {
4830: *flg = PETSC_FALSE;
4831: return(0);
4832: }
4833: }
4834: #else
4835: PetscArraycmp(a->a,b->a,a->nz,flg);
4836: #endif
4837: return(0);
4838: }
4840: /*@
4841: MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format)
4842: provided by the user.
4844: Collective
4846: Input Parameters:
4847: + comm - must be an MPI communicator of size 1
4848: . m - number of rows
4849: . n - number of columns
4850: . i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of elements in that row of the matrix
4851: . j - column indices
4852: - a - matrix values
4854: Output Parameter:
4855: . mat - the matrix
4857: Level: intermediate
4859: Notes:
4860: The i, j, and a arrays are not copied by this routine, the user must free these arrays
4861: once the matrix is destroyed and not before
4863: You cannot set new nonzero locations into this matrix, that will generate an error.
4865: The i and j indices are 0 based
4867: The format which is used for the sparse matrix input, is equivalent to a
4868: row-major ordering.. i.e for the following matrix, the input data expected is
4869: as shown
4871: $ 1 0 0
4872: $ 2 0 3
4873: $ 4 5 6
4874: $
4875: $ i = {0,1,3,6} [size = nrow+1 = 3+1]
4876: $ j = {0,0,2,0,1,2} [size = 6]; values must be sorted for each row
4877: $ v = {1,2,3,4,5,6} [size = 6]
4880: .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
4882: @*/
4883: PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat)
4884: {
4886: PetscInt ii;
4887: Mat_SeqAIJ *aij;
4888: PetscInt jj;
4891: if (m > 0 && i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
4892: MatCreate(comm,mat);
4893: MatSetSizes(*mat,m,n,m,n);
4894: /* MatSetBlockSizes(*mat,,); */
4895: MatSetType(*mat,MATSEQAIJ);
4896: MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,NULL);
4897: aij = (Mat_SeqAIJ*)(*mat)->data;
4898: PetscMalloc1(m,&aij->imax);
4899: PetscMalloc1(m,&aij->ilen);
4901: aij->i = i;
4902: aij->j = j;
4903: aij->a = a;
4904: aij->singlemalloc = PETSC_FALSE;
4905: aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
4906: aij->free_a = PETSC_FALSE;
4907: aij->free_ij = PETSC_FALSE;
4909: for (ii=0; ii<m; ii++) {
4910: aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii];
4911: if (PetscDefined(USE_DEBUG)) {
4912: 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]);
4913: for (jj=i[ii]+1; jj<i[ii+1]; jj++) {
4914: 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);
4915: 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);
4916: }
4917: }
4918: }
4919: if (PetscDefined(USE_DEBUG)) {
4920: for (ii=0; ii<aij->i[m]; ii++) {
4921: if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %D index = %D",ii,j[ii]);
4922: 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]);
4923: }
4924: }
4926: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
4927: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
4928: return(0);
4929: }
4930: /*@C
4931: MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format)
4932: provided by the user.
4934: Collective
4936: Input Parameters:
4937: + comm - must be an MPI communicator of size 1
4938: . m - number of rows
4939: . n - number of columns
4940: . i - row indices
4941: . j - column indices
4942: . a - matrix values
4943: . nz - number of nonzeros
4944: - idx - 0 or 1 based
4946: Output Parameter:
4947: . mat - the matrix
4949: Level: intermediate
4951: Notes:
4952: The i and j indices are 0 based
4954: The format which is used for the sparse matrix input, is equivalent to a
4955: row-major ordering.. i.e for the following matrix, the input data expected is
4956: as shown:
4958: 1 0 0
4959: 2 0 3
4960: 4 5 6
4962: i = {0,1,1,2,2,2}
4963: j = {0,0,2,0,1,2}
4964: v = {1,2,3,4,5,6}
4967: .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
4969: @*/
4970: PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat,PetscInt nz,PetscBool idx)
4971: {
4973: PetscInt ii, *nnz, one = 1,row,col;
4977: PetscCalloc1(m,&nnz);
4978: for (ii = 0; ii < nz; ii++) {
4979: nnz[i[ii] - !!idx] += 1;
4980: }
4981: MatCreate(comm,mat);
4982: MatSetSizes(*mat,m,n,m,n);
4983: MatSetType(*mat,MATSEQAIJ);
4984: MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);
4985: for (ii = 0; ii < nz; ii++) {
4986: if (idx) {
4987: row = i[ii] - 1;
4988: col = j[ii] - 1;
4989: } else {
4990: row = i[ii];
4991: col = j[ii];
4992: }
4993: MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);
4994: }
4995: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
4996: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
4997: PetscFree(nnz);
4998: return(0);
4999: }
5001: PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A)
5002: {
5003: Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data;
5007: a->idiagvalid = PETSC_FALSE;
5008: a->ibdiagvalid = PETSC_FALSE;
5010: MatSeqAIJInvalidateDiagonal_Inode(A);
5011: return(0);
5012: }
5014: PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat)
5015: {
5017: PetscMPIInt size;
5020: MPI_Comm_size(comm,&size);
5021: if (size == 1) {
5022: if (scall == MAT_INITIAL_MATRIX) {
5023: MatDuplicate(inmat,MAT_COPY_VALUES,outmat);
5024: } else {
5025: MatCopy(inmat,*outmat,SAME_NONZERO_PATTERN);
5026: }
5027: } else {
5028: MatCreateMPIMatConcatenateSeqMat_MPIAIJ(comm,inmat,n,scall,outmat);
5029: }
5030: return(0);
5031: }
5033: /*
5034: Permute A into C's *local* index space using rowemb,colemb.
5035: The embedding are supposed to be injections and the above implies that the range of rowemb is a subset
5036: of [0,m), colemb is in [0,n).
5037: If pattern == DIFFERENT_NONZERO_PATTERN, C is preallocated according to A.
5038: */
5039: PetscErrorCode MatSetSeqMat_SeqAIJ(Mat C,IS rowemb,IS colemb,MatStructure pattern,Mat B)
5040: {
5041: /* If making this function public, change the error returned in this function away from _PLIB. */
5043: Mat_SeqAIJ *Baij;
5044: PetscBool seqaij;
5045: PetscInt m,n,*nz,i,j,count;
5046: PetscScalar v;
5047: const PetscInt *rowindices,*colindices;
5050: if (!B) return(0);
5051: /* Check to make sure the target matrix (and embeddings) are compatible with C and each other. */
5052: PetscObjectBaseTypeCompare((PetscObject)B,MATSEQAIJ,&seqaij);
5053: if (!seqaij) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is of wrong type");
5054: if (rowemb) {
5055: ISGetLocalSize(rowemb,&m);
5056: 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);
5057: } else {
5058: if (C->rmap->n != B->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is row-incompatible with the target matrix");
5059: }
5060: if (colemb) {
5061: ISGetLocalSize(colemb,&n);
5062: 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);
5063: } else {
5064: if (C->cmap->n != B->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Input matrix is col-incompatible with the target matrix");
5065: }
5067: Baij = (Mat_SeqAIJ*)(B->data);
5068: if (pattern == DIFFERENT_NONZERO_PATTERN) {
5069: PetscMalloc1(B->rmap->n,&nz);
5070: for (i=0; i<B->rmap->n; i++) {
5071: nz[i] = Baij->i[i+1] - Baij->i[i];
5072: }
5073: MatSeqAIJSetPreallocation(C,0,nz);
5074: PetscFree(nz);
5075: }
5076: if (pattern == SUBSET_NONZERO_PATTERN) {
5077: MatZeroEntries(C);
5078: }
5079: count = 0;
5080: rowindices = NULL;
5081: colindices = NULL;
5082: if (rowemb) {
5083: ISGetIndices(rowemb,&rowindices);
5084: }
5085: if (colemb) {
5086: ISGetIndices(colemb,&colindices);
5087: }
5088: for (i=0; i<B->rmap->n; i++) {
5089: PetscInt row;
5090: row = i;
5091: if (rowindices) row = rowindices[i];
5092: for (j=Baij->i[i]; j<Baij->i[i+1]; j++) {
5093: PetscInt col;
5094: col = Baij->j[count];
5095: if (colindices) col = colindices[col];
5096: v = Baij->a[count];
5097: MatSetValues(C,1,&row,1,&col,&v,INSERT_VALUES);
5098: ++count;
5099: }
5100: }
5101: /* FIXME: set C's nonzerostate correctly. */
5102: /* Assembly for C is necessary. */
5103: C->preallocated = PETSC_TRUE;
5104: C->assembled = PETSC_TRUE;
5105: C->was_assembled = PETSC_FALSE;
5106: return(0);
5107: }
5109: PetscFunctionList MatSeqAIJList = NULL;
5111: /*@C
5112: MatSeqAIJSetType - Converts a MATSEQAIJ matrix to a subtype
5114: Collective on Mat
5116: Input Parameters:
5117: + mat - the matrix object
5118: - matype - matrix type
5120: Options Database Key:
5121: . -mat_seqai_type <method> - for example seqaijcrl
5124: Level: intermediate
5126: .seealso: PCSetType(), VecSetType(), MatCreate(), MatType, Mat
5127: @*/
5128: PetscErrorCode MatSeqAIJSetType(Mat mat, MatType matype)
5129: {
5130: PetscErrorCode ierr,(*r)(Mat,MatType,MatReuse,Mat*);
5131: PetscBool sametype;
5135: PetscObjectTypeCompare((PetscObject)mat,matype,&sametype);
5136: if (sametype) return(0);
5138: PetscFunctionListFind(MatSeqAIJList,matype,&r);
5139: if (!r) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_UNKNOWN_TYPE,"Unknown Mat type given: %s",matype);
5140: (*r)(mat,matype,MAT_INPLACE_MATRIX,&mat);
5141: return(0);
5142: }
5145: /*@C
5146: MatSeqAIJRegister - - Adds a new sub-matrix type for sequential AIJ matrices
5148: Not Collective
5150: Input Parameters:
5151: + name - name of a new user-defined matrix type, for example MATSEQAIJCRL
5152: - function - routine to convert to subtype
5154: Notes:
5155: MatSeqAIJRegister() may be called multiple times to add several user-defined solvers.
5158: Then, your matrix can be chosen with the procedural interface at runtime via the option
5159: $ -mat_seqaij_type my_mat
5161: Level: advanced
5163: .seealso: MatSeqAIJRegisterAll()
5166: Level: advanced
5167: @*/
5168: PetscErrorCode MatSeqAIJRegister(const char sname[],PetscErrorCode (*function)(Mat,MatType,MatReuse,Mat *))
5169: {
5173: MatInitializePackage();
5174: PetscFunctionListAdd(&MatSeqAIJList,sname,function);
5175: return(0);
5176: }
5178: PetscBool MatSeqAIJRegisterAllCalled = PETSC_FALSE;
5180: /*@C
5181: MatSeqAIJRegisterAll - Registers all of the matrix subtypes of SeqAIJ
5183: Not Collective
5185: Level: advanced
5187: Developers Note: CUSPARSE does not yet support the MatConvert_SeqAIJ..() paradigm and thus cannot be registered here
5189: .seealso: MatRegisterAll(), MatSeqAIJRegister()
5190: @*/
5191: PetscErrorCode MatSeqAIJRegisterAll(void)
5192: {
5196: if (MatSeqAIJRegisterAllCalled) return(0);
5197: MatSeqAIJRegisterAllCalled = PETSC_TRUE;
5199: MatSeqAIJRegister(MATSEQAIJCRL, MatConvert_SeqAIJ_SeqAIJCRL);
5200: MatSeqAIJRegister(MATSEQAIJPERM, MatConvert_SeqAIJ_SeqAIJPERM);
5201: MatSeqAIJRegister(MATSEQAIJSELL, MatConvert_SeqAIJ_SeqAIJSELL);
5202: #if defined(PETSC_HAVE_MKL_SPARSE)
5203: MatSeqAIJRegister(MATSEQAIJMKL, MatConvert_SeqAIJ_SeqAIJMKL);
5204: #endif
5205: #if defined(PETSC_HAVE_VIENNACL) && defined(PETSC_HAVE_VIENNACL_NO_CUDA)
5206: MatSeqAIJRegister(MATMPIAIJVIENNACL, MatConvert_SeqAIJ_SeqAIJViennaCL);
5207: #endif
5208: return(0);
5209: }
5211: /*
5212: Special version for direct calls from Fortran
5213: */
5214: #include <petsc/private/fortranimpl.h>
5215: #if defined(PETSC_HAVE_FORTRAN_CAPS)
5216: #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ
5217: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
5218: #define matsetvaluesseqaij_ matsetvaluesseqaij
5219: #endif
5221: /* Change these macros so can be used in void function */
5222: #undef CHKERRQ
5223: #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr)
5224: #undef SETERRQ2
5225: #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr)
5226: #undef SETERRQ3
5227: #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr)
5229: PETSC_EXTERN void matsetvaluesseqaij_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[],InsertMode *isis, PetscErrorCode *_ierr)
5230: {
5231: Mat A = *AA;
5232: PetscInt m = *mm, n = *nn;
5233: InsertMode is = *isis;
5234: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
5235: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
5236: PetscInt *imax,*ai,*ailen;
5238: PetscInt *aj,nonew = a->nonew,lastcol = -1;
5239: MatScalar *ap,value,*aa;
5240: PetscBool ignorezeroentries = a->ignorezeroentries;
5241: PetscBool roworiented = a->roworiented;
5244: MatCheckPreallocated(A,1);
5245: imax = a->imax;
5246: ai = a->i;
5247: ailen = a->ilen;
5248: aj = a->j;
5249: aa = a->a;
5251: for (k=0; k<m; k++) { /* loop over added rows */
5252: row = im[k];
5253: if (row < 0) continue;
5254: if (PetscUnlikelyDebug(row >= A->rmap->n)) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
5255: rp = aj + ai[row]; ap = aa + ai[row];
5256: rmax = imax[row]; nrow = ailen[row];
5257: low = 0;
5258: high = nrow;
5259: for (l=0; l<n; l++) { /* loop over added columns */
5260: if (in[l] < 0) continue;
5261: if (PetscUnlikelyDebug(in[l] >= A->cmap->n)) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large");
5262: col = in[l];
5263: if (roworiented) value = v[l + k*n];
5264: else value = v[k + l*m];
5266: if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue;
5268: if (col <= lastcol) low = 0;
5269: else high = nrow;
5270: lastcol = col;
5271: while (high-low > 5) {
5272: t = (low+high)/2;
5273: if (rp[t] > col) high = t;
5274: else low = t;
5275: }
5276: for (i=low; i<high; i++) {
5277: if (rp[i] > col) break;
5278: if (rp[i] == col) {
5279: if (is == ADD_VALUES) ap[i] += value;
5280: else ap[i] = value;
5281: goto noinsert;
5282: }
5283: }
5284: if (value == 0.0 && ignorezeroentries) goto noinsert;
5285: if (nonew == 1) goto noinsert;
5286: if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix");
5287: MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
5288: N = nrow++ - 1; a->nz++; high++;
5289: /* shift up all the later entries in this row */
5290: for (ii=N; ii>=i; ii--) {
5291: rp[ii+1] = rp[ii];
5292: ap[ii+1] = ap[ii];
5293: }
5294: rp[i] = col;
5295: ap[i] = value;
5296: A->nonzerostate++;
5297: noinsert:;
5298: low = i + 1;
5299: }
5300: ailen[row] = nrow;
5301: }
5302: PetscFunctionReturnVoid();
5303: }